Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

default · ~99.6k rows (showing the first 120k)

Split (2)

train · ~72.4k rows (showing the first 93.1k)

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1108.0962	3	1108	2015-02-17T22:52:15	On On_p	[ "math.RA" ]	Generalizing John Conway's construction of the Field On_2, we give the "minimal" definitions of addition and multiplication that turn the ordinals into a Field of characteristic p, for any prime p. We then analyze the structure of the resulting Field, which we will call On_p.	math.RA	math	ON Onp JOSEPH DIMURO Abstract. Generalizing John Conway's construction of the Field On2, we give the "minimal" definitions of addition and multipli- cation that turn the ordinals into a Field of characteristic p, for any prime p. We then analyze the structure of the resulting Field, which we will call On p. In Chapter 6 of [2], John Conway introduces the Field On2: the Class of all ordinals with the appropriate addition and multiplication defined to obtain an algebraically closed Field of characteristic 2. (Follow- ing Conway's convention, we will use capitalized terms like "Group", "Ring", and "Field" when the structure referred to is a proper Class.) The operations are referred to as Nim-addition and Nim-multiplication, due to their connections with the game of Nim. Further descriptions of the structure of this Field are given by Lenstra in [4] and [5]. On pg. 17 of [4], Lenstra gives an addition operation which turns the Class of ordinals into an abelian group of exponent 3. Lenstra then asks if there is an analogous definition of multiplication that produces a Field of characteristic 3, and if other characteristics can similarly be handled. In [3], Fran¸cois Laubie gives an appropriate definition of addition for any prime characteristic p. (The definition is confined to the finite ordinals, but it works just as well for all ordinals.) In this paper, we will provide definitions of addition and multiplication that turn the ordinals into a Field of any prime characteristic p, which we will call Onp. We will also analyze the structure of these Fields Onp, obtaining results analogous to those in all the aforementioned references. In what follows, we will normally use + and × to denote addition and multiplication in Onp. If the standard ordinal operations are needed instead, then the given expression will be enclosed in brackets. For example, [4(4) + 3] = 19, but in On2, 4(4) + 3 = 6 + 3 = 5. We will [42 + 3] = 19, but in sometimes use exponentiation in a similar way: On2, 42 + 3 = 4(4) + 3 = 5. Also, the notation a · b (when not in Date: 12/2/14. 2010 Mathematics Subject Classification. 12F05, 12F20. Key words and phrases. ordinals, fields, field extensions. 1 2 JOSEPH DIMURO brackets) will only be used for repeated addition; in On2, 2(4) = 8, but 2 · 4 = 4 + 4 = 0. 1. Addition and Multiplication in Onp In [2], the definitions of addition and multiplication in On2 are given "genetically", as follows: for α, β ∈ On2, we have and α + β = mex{α′ + β, α + β ′} αβ = mex{α′β + αβ ′ − α′β ′}. Here, α′ ranges over all ordinals less than α, β ′ ranges over all ordi- nals less than β, and "mex" represents the "minimal excludent" of the given set (i.e. the smallest ordinal not in that set). Thus, all sums and products are defined in terms of lexicographically earlier sums and products. In [3], a similar genetic definition is given for addition in Onp. Un- fortunately, finding a genetic definition for multiplication in Onp is apparently much more difficult (reasons for this will be given later). Rather than working with genetic definitions, we will establish the structure of Onp inductively, defining addition and multiplication on progressively larger fields. Following the convention in [2], ordinals in Onp will sometimes be treated as single elements of Onp, and will sometimes be treated as the set of all lesser ordinals. Thus, a single ordinal α ∈ Onp will be called a "group", or "ring", or "field", whenever the set of ordinals β < α forms a group, or ring, or field. For any given ordinal ∆, the "∆-th" field in Onp will be denoted by φ∆. In constructing these fields we must ensure that: (1) For every ordinal ∆, the ordinal φ∆ is a field of characteristic p. (2) If ∆′ < ∆, then φ∆′ < φ∆. (3) The operations on each field extend those of all previous fields. That is, if ∆′ < ∆, and α, β ∈ φ∆′, then α + β and αβ are the same in both φ∆′ and φ∆. We will construct the fields by induction on ∆, as follows: • If ∆ = 0, then φ∆ = φ0 = p, and the operations on p are just ordinary addition and multiplication modulo p. Thus, φ0 is isomorphic to Fp, the finite field of p elements. • If ∆ is a successor ordinal, then we will construct φ∆ from φ[∆−1], using the methods discussed below. ON Onp 3 • If ∆ is a limit ordinal, then let ∆[i] be a fundamental sequence of ∆. Then φ∆ will be the limit ordinal whose fundamental sequence is the following: φ∆[i] = φ∆[i]. (In other words, φ∆ is the supremum of all previous fields.) When ∆ is a limit ordinal, the operations on φ∆ are already deter- mined by induction: if α, β ∈ φ∆, then we have α, β ∈ φ∆[i] for some ordinal ∆[i] in the fundamental sequence of ∆. Then α + β and αβ are defined to be the same in φ∆ as in φ∆[i]. Trivially, since every φ∆[i] is a field of characteristic p, so is φ∆. So the only remaining work is to define φ∆ when ∆ is a successor ordinal. For simplicity in what follows, we will let eφ = φ∆, and we will let φ be the previous field: φ = φ[∆−1]. 1.1. When φ is not algebraically closed. Assume first that the field φ is not algebraically closed. Let φ[x] be the ring of polynomials with coefficients in φ. Let n be the smallest positive integer where not all polynomials in φ[x] of degree n have roots in φ. Let h(x) ∈ φ[x] be the "lexicographically earliest" polynomial such that g(x) = xn − h(x) has no root in φ. In determining which polynomial is lexicographically earliest, we consider the coefficients of the largest power of x first. (For example, 5x3 + 2x2 + 9x + 17 is lexicographically earlier than 5x3 + 3x2 + 1.) Note that g(x) is then irreducible over φ; if g(x) were the product of two polynomials in φ[x] of degree less than n, then each of those polynomials would have a root in φ, contradicting the fact that g(x) has no root in φ. We then define the next field to be eφ = [φn]. The definitions of ad- dition and multiplication on eφ will be chosen so that eφ is the extension of the field φ by a root of g(x); the ordinal φ itself will serve as a root of g(x). Let F be the factor ring φ[x]/hg(x)i. Because g(x) is irreducible over φ, F is a field. Every element of F is of the form f (x) + hg(x)i, where f (x) is a polynomial in φ[x] of degree less than n. That is, every element of F has the form n−1Xi=0 Also, every element of eφ has a similar representation: α ="n−1Xi=0 xiαi!+hg(x)i for some ordinals αi ∈ φ. if α ∈ eφ, then φiαi# for some ordinals αi ∈ φ. We thus have a one-to-one, 4 JOSEPH DIMURO onto map between eφ and F : we can define θ : eφ → F via n−1Xi=0(cid:0)xiαi(cid:1) + hg(x)i. φiαi#! = θ "n−1Xi=0 We will use this map to directly define addition and multiplication φiαi. n−1Xi=0 p extending the operations of φ. Finally, one more thing to note: θ−1(θ(α)+θ(β)) = θ−1((α+hg(x)i)+(β +hg(x)i)) = θ−1(γ1 +hg(x)i) = θ−1(θ(α)θ(β)). We then have θ(α + β) = θ(α) + θ(β) and θ(αβ) = θ(α)θ(β), so θ is an isomorphism. Since F is a field of characteristic p, in eφ: if α, β ∈ eφ, then we let α + β = θ−1(θ(α) + θ(β)), and αβ = so is eφ. Also, note that the given operations on eφ extend those on φ. Given α, β ∈ φ, let γ1 = α + β, γ2 = αβ in φ. Then in eφ, we have α + β = γ1, and similarly αβ = γ2. So as required, eφ is a field of characteristic Lemma 1.1. Given a field φ ∈ Onp, assume that the next field eφ is a degree n extension of φ. If α ="n−1Xi=0 Proof. We have θ "n−1Xi=0 φiαi#! = (θ(φ))iαi = θ n−1Xi=0 n−1Xi=0 n−1Xi=0 to-one, we must have"n−1Xi=0 φiαi# = gebraically closed field. The next field will be eφ = [φφ], and we will define addition and multiplication so that eφ is isomorphic to φ(x), the φiαi# ∈eφ, then α = n−1Xi=0(cid:0)xiαi(cid:1) + hg(x)i φiαi!. Since θ is one- Note: the smallest field in Onp is φ0 = p, and all fields φn for finite n will be algebraic extensions of φ0, since no finite fields are algebraically closed. The next field must be φω = ω. Thereafter, all fields φ∆ will either be a [power] of the preceding field (when ∆ is a successor ordinal) or the supremum of all previous fields (when ∆ is a limit ordinal). Thus, all infinite fields in Onp will be [powers] of ω, and hence, limit ordinals. (In what follows, we will need the fact that all algebraically closed fields are limit ordinals.) 1.2. When φ is algebraically closed. Now assume that φ is an al- field of rational functions with coefficients in φ. = (x + hg(x)i)i αi = φiαi. n−1Xi=0 (cid:3) ON Onp 5 Following the notation on pg. 62 in [2], any rational function f (x) ∈ φ(x) has a partial fraction expansion f (x) =Xi βi (x − αi)ni+1 +Xj xmj δj where αi, βi, δj ∈ φ and ni, mj ∈ ω for every i and j. Also, every element in eφ = [φφ] has the form "Xi φω+ωαi+niβi +Xj φmj δj# where αi, βi, δj ∈ φ and ni, mj ∈ ω for each i and j. We thus have the . xmj δj βi (x − αi)ni+1 +Xj φω+ωαi+niβi +Xj φmj δj#! =Xi θ(β)), and αβ = θ−1(θ(α)θ(β)). We then have θ(α + β) = θ(α) + θ(β) and θ(αβ) = θ(α)θ(β), so θ is an isomorphism. Since φ(x) is a field following one-to-one, onto map: θ : eφ → φ(x), where θ "Xi We define the operations on eφ the same way as in the case where φ is not algebraically closed: if α, β ∈eφ, then we let α + β = θ−1(θ(α) + of characteristic p, so is eφ. And since θ(α) = α for every α ∈ φ, the operations in eφ extend those in φ. So as required, eφ is a field of Lemma 1.2. Given an algebraically closed field φ ∈ Onp, let eφ be the next larger field. If α =hX φδαδi ∈eφ, then α =X(cid:2)φδ(cid:3) αδ. This is essentially the same result as the previous lemma, except that As with the non-algebraically-closed case, we have one more thing characteristic p extending the operations of φ. to note: the exponents on φ can now range over all ordinals δ < φ. Proof. Given any finite collections of ordinals αi, βi, δj ∈ φ and ni, mj ∈ ω, we have θ "Xi φω+ωαi+niβi +Xj θ(cid:0)(cid:2)φω+ωαi+ni(cid:3)(cid:1) βi+Xj =Xi (x − αi)ni+1 +Xj φmj δj#! =Xi θ([φmj ])δj = θ Xi (cid:2)φω+ωαi+ni(cid:3) βi +Xj xmj δj βi [φmj ] δj! 6 JOSEPH DIMURO Thus, since θ is one-to-one, we have: "Xi φω+ωαi+niβi +Xj φmj δj# =Xi (cid:2)φω+ωαi+ni(cid:3) βi +Xj [φmj ] δj (cid:3) We now have defined addition and multiplication operations that turn the ordinals into Onp, a Field of characteristic p. But we still must show that these Fields, as defined, are the correct analogues of Conway's field On2. We must show that these definitions of addition and multiplication are the "minimal" definitions which will turn the ordinals into a Field of characteristic p. 2. The minimality of Onp As stated in the last section, the definitions of addition and multipli- cation in On2 are given in [2] as follows: α + β = mex{α′ + β, α + β ′}, and αβ = mex{α′β + αβ ′ − α′β ′}. These operations can be thought of as the "minimal" operations that turn the ordinals into a Field. That is, let's say we tried to write out an addition table for the ordinals, not filling in any entry until all lexicographically earlier entries are filled. (So we don't determine α + β until everything of the form α′ + β or α + β ′ is determined.) For each entry in the table, we always choose the smallest ordinal which allows the resulting structure to be a Field. Then we do the same with multiplication, determining αβ only after all products of the form α′β, αβ ′, and α′β ′ have been determined. The result would be Conway's definition of On2, for the following reason: Lemma 2.1. If F is a Field consisting of the Class of all ordinals, then regardless of how addition and multiplication are defined, for all ordinals α, β, we have α+β ≥ mex{α′+β, α+β ′}, and αβ ≥ mex{α′β+ αβ ′ − α′β ′}. So, in On2, each sum or product of ordinals yields the smallest or- dinal possible, while still ensuring that On2 is indeed a Field. Proof. Assume that, for some ordinals α and β, we have α + β < mex{α′ + β, α + β ′}. Then either α + β = α′ + β for some α′ < α, or α + β = α′ + β for some β ′ < β. But then either α = α′ or β = β ′, a contradiction. Now assume that, for some ordinals α and β, we have αβ < mex{α′β+ αβ ′ − α′β ′}. Then for some α′ < α and β ′ < β, we have αβ = α′β + αβ ′ − α′β ′. But then (α − α′)(β − β ′) = 0, a contradiction, since fields have no zero divisors. (cid:3) ON Onp 7 Now, in [3], Laubie gives a genetic definition for p-adic addition in the set of finite ordinals (which can be extended to all ordinals). Under his definition, addition can be done by expressing ordinals in base p, then adding in base p without carrying. (For example, in On3, we have 22 + 19 = (9 + 9 + 3 + 1) + (9 + 9 + 1) = 9 + 3 + 1 + 1 = 14.) We will now show that the definitions given in Section 1 produce the same property of addition; thus, our non-genetic definition of addition is the same as Laubie's genetic definition. Theorem 2.2. Given ordinals α, β ∈ Onp, assume α = [P pδaδ] and β = [P pδbδ] (where each δ is an ordinal, and every aδ, bδ ∈ p). Then α + β = [P pδcδ], where each cδ ≡ aδ + bδ (mod p). Proof. There exists some field φ∆ containing both α and β. We will proceed by induction on ∆. The statement is obviously true when ∆ = 0, as then α, β ∈ p, and addition in p is just ordinary addition modulo p. If ∆ is a limit ordinal, then for some ordinal ∆[i] in the fundamental sequence of ∆, both α and β are contained in φ∆[i]. So by induction, the statement is true in that case. p, this is just componentwise addition modulo p. (cid:3) let φ = φ[∆−1] be the previous field. Based on the work in Section 1, we degree n extension of φ, then every δ < n; if φ is algebraically closed, then every δ ∈ φ.) For each δ, let αδ + βδ = γδ; by induction, each of these summations is just componentwise addition modulo p. Then, That leaves the case where ∆ is a successor ordinal; let eφ = φ∆, and have α = [P φδαδ] and β = [P φδβδ], where each αδ, βδ ∈ φ. (If eφ is a from Lemmas 1.1 and 1.2, we have α + β = [P φδαδ] + [P φδβδ] = P[φδ]αδ +P[φδ]βδ =P[φδ]γδ = [P φδγδ]. And since φ is a [power] of So indeed, our definition of addition is as it should be. And this leads to one further consequence: the elements of Onp that are groups are exactly the ordinals of the form [pα], for some ordinal α. We must now show why our definition of multiplication is the correct one. Definition 2.3. Given ordinals α, β ∈ Onp, we will say that the un- ordered pair {α, β} has the "MEX property" if αβ = mex{α′β + αβ ′ − α′β ′}. We will call the set {α′β + αβ ′ − α′β ′} the "MEX set" of the unordered pair {α, β}. Note: from Lemma 2.1, we have αβ ≥ mex{α′β + αβ ′ − α′β ′} for all α, β ∈ Onp. Thus, we can say that {α, β} has the MEX property if αβ ≤ mex{α′β + αβ ′ − α′β ′}. In other words, {α, β} has the MEX property if, for all γ < αβ, we have γ = α′β +αβ ′ −α′β ′ for some choice 8 JOSEPH DIMURO of α′ < α, β ′ < β. We will use this property in the lemmas that follow; to show that a pair {α, β} has the MEX property, we will show that all ordinals less than αβ can be written in the form α′β + αβ ′ − α′β ′. With this definition, we can say that {α, β} has the MEX property for all α, β ∈ On2. It turns out that the same does not apply for Onp when p ≥ 3 (as we will see later). However, we will prove that certain pairs {α, β} do have the MEX property, and that these particular prod- ucts determine the products of any two groups in Onp. (Thus, they determine the entire multiplication table of Onp, since the products of arbitrary elements can be determined from the products of groups via the distributive law.) That should be sufficient evidence that our definition of multiplication is the "correct" one. Lemma 2.4. Let φ ∈ Onp be a field, and let eφ be the next larger field in Onp. Let α, β ∈ Onp be ordinals such that α > 0, [φα] ∈ eφ and β ∈ φ. Then the pair {[φα], β} has the MEX property. Proof. Based on Lemmas 1.1 and 1.2, we have [φα]β = [φαβ], regardless If γ < [φαβ], then γ = of whether φ is algebraically closed or not. [φαβ1 + β2] = [φα]β1 + β2 for some β1 < β and some β2 ∈ [φα]. To show that γ is in the MEX set of {[φα], β}, we must show that γ = α1β + [φα]β ′ − α1β ′ for some ordinals α1 < φα, β ′ < β. Let β ′ = β1. Since β − β1 is a nonzero element of φ, and since φ is a field, there exists some φ′ ∈ φ where φ′(β − β1) = 1. Let α1 = φ′β2 ∈ [φα]. We then have α1β + [φα]β ′ − α1β ′ = [φα]β ′ + α1(β − β ′) = [φα]β1 + φ′β2(β − β1) = [φα]β1 + β2 = γ. So we have rewritten γ as required; {[φα], β} has the MEX property. (cid:3) Lemma 2.5. Let φ ∈ Onp be a field that is not algebraically closed, and assume that eφ ∈ Onp, the next larger field, is an extension of φ of degree n. Let i, j be nonnegative integers such that i + j ≤ n. Then the pair {[φi], [φj]} has the MEX property. Proof. Let h(x) ∈ φ[x] be the lexicographically earliest polynomial such that g(x) = xn − h(x) has no root in φ. If i + j < n, then [φi][φj] = [φi+j]. If γ < [φi+j], then for some polynomial m(x) ∈ φ[x] of degree less than i + j, we have γ = [m(φ)]. Let f (x) = xi+j − m(x); then f (x) is a monic polynomial of degree less than n. So f can be factored into linear factors over φ. Assume f (x) = f1(x)f2(x), where f1(x), f2(x) ∈ φ[x], f1 is monic of degree i, and f2 is monic of degree j. Let m1(x) = xi − f1(x) (which has degree less than i), and let m2(x) = xj − f2(x) (which has degree less than j). Then m(x) = xi+j − f (x) = xi+j − (xi − m1(x))(xj − m2(x)) = xim2(x) + xjm1(x) − ON Onp 9 m1(x)m2(x). So we have γ = [φi][m2(φ)]+[m1(φ)][φj]−[m1(φ)][m2(φ)], and we've rewritten γ in the desired form. So {[φi], [φj]} has the MEX property if i + j < n. Now assume i + j = n. Then [φi][φj] = [h(φ)]. If γ < [h(φ)], then γ = [m(φ)] for some polynomial m(x) ∈ φ[x] that is lexicographically earlier than h(x). Let f (x) = xi+j − m(x); then f (x) can be factored into linear factors over φ. We then proceed as before; let f (x) = f1(x)f2(x), where f1 is monic of degree i, and f2 is monic of degree j. Then if m1(x) = xi − f1(x) (which has degree less than i) and m2(x) = xj − f2(x) (which has degree less than j), then we have γ = [φi][m2(φ)] + [m1(φ)][φj] − [m1(φ)][m2(φ)]. We've rewritten γ in the desired form, so {[φi], [φj]} has the MEX property if i + j = n. (cid:3) This essentially establishes that we have the correct definition of mul- tiplication in eφ, when the preceding field φ is not algebraically closed, and multiplication in φ has already been determined. If eφ = [φn], then all pairs {[φi], [φj]} (where i + j ≤ n) satisfy the MEX property. And by induction, the products of such pairs determine all products [φi][φj] when i+j > n; if [φi][φj] = φkγk is known, and if j < n, then we are n−1Xk=0 φkγk! φ = n−1Xk=0 φk+1γk. Fi- 0 ≤ i, j < n, and a, b ∈ φ are groups. Then αβ is determined by all the aforementioned rules: αβ = [φia][φjb] = [φi][φj](ab) = forced to have [φi][φj+1] = [φi][φj]φ = n−1Xk=0 nally, if α, β ∈ eφ are groups, then α = [φia] and β = [φjb], where φkck#, where ck = γkab (a product in φ). n−1Xk=0 So the products of all groups in eφ are determined by the above rules, and thus, so is the entire multiplication table of eφ (by the distributive φkγk! (ab) = "n−1Xk=0 We still must consider the case where φ is algebraically closed. law). Lemma 2.6. If φ ∈ Onp is an algebraically closed field, then for all i, j ∈ ω, the pair {[φi], [φj]} has the MEX property. The proof is essentially the same as for the i + j < n case in the previous lemma. So, the products [φi][φj] = [φi+j] all satisfy the MEX property, and they establish the fact that [φi] = φi for all i ∈ ω. Lemma 2.7. If φ ∈ Onp is an algebraically closed field, then for all α ∈ φ, the pair {[φω+ωα], φ} has the MEX property. 10 JOSEPH DIMURO Proof. We have [φω+ωα]φ = ( 1 If γ < [φω+ωαα + 1], then either γ = [φω+ωαα] = α φ−α + f (φ), where α′ < α, and f (x) ∈ φ(x) is a rational function whose poles are all less than α. φ−α + 1 = [φω+ωαα + 1]. φ−α, or γ = α′ φ−α)φ = α The typical element of the MEX set of {[φω+ωα], φ} has the form g(φ)φ + 1 φ−α φ′ − g(φ)φ′, where φ′ < φ, and g(x) ∈ φ(x) is a rational function whose poles are all less than α. We may set φ′ = α′, obtaining φ−α α′ − g(φ)α′ = α′ g(φ)φ + 1 φ−α + g(φ)(φ − α′). And this will equal φ−α + f (φ) when g(x) = f (x) α′ x−α′ (which is indeed a rational function whose poles are all less than α). That covers all possible values of γ except γ = α φ−α , which we may obtain by setting φ′ = α and g(x) = 0. So every value of γ is in the MEX set; {[φω+ωα], φ} has the MEX property. (cid:3) So, the products [φω+ωα]φ = [φω+ωαα + 1] all satisfy the MEX prop- erty, and they establish the fact that, for all α ∈ φ, [φω+ωα] = 1 (φ−α) . Lemma 2.8. If φ ∈ Onp is an algebraically closed field, then for all α ∈ φ and all positive integers n, the pair {[φω+ωα+n], φ} has the MEX property. Proof. We have [φω+ωα+n]φ = 1) = ordinal, then γ must take one of the following two forms: (φ−α)n = [φω+ωα+nα + φω+ωα+n−1]. (φ−α)n+1 φ = 1 (φ−α)n+1 + 1 (φ−α)n φ−α = 1 (φ−α)n ( α φ−α + If γ is a smaller α 1 φ 1. γ = α (φ−α)n+1 + f (φ), where all poles of f (x) ∈ φ(x) are at most α, and if α is a pole of f (x), then it has degree at most n − 1. 2. γ = α′ (φ−α)n+1 + f (φ), where α′ < α, all poles of f (x) ∈ φ(x) are at most α, and if α is a pole of f (x), then it has degree at most n. 1 Meanwhile, the typical element of the MEX set of {[φω+ωα+n], φ} (φ−α)n+1 φ′ − g(φ)φ′, where φ′ < φ, all poles of has the form g(φ)φ + g(x) ∈ φ(x) are at most α, and if α is a pole of g(x), then it has degree at most n. This typical element of the MEX set will equal (φ−α)n+1 + f (φ) when φ′ = α and g(x) = f (x) x−α. And this typical element of the MEX set will equal x−α′ . So all values of γ are in the MEX set; {[φω+ωα+n], φ} has the MEX property. (cid:3) (φ−α)n+1 + f (φ) when φ′ = α′ and g(x) = f (x) α′ α So, the products [φω+ωα+n]φ = [φω+ωα+nα + φω+ωα+n−1] all satisfy the MEX property. And they establish the fact that, for all α ∈ φ, [φω+ωα+n] = 1 (φ−α)n+1 . ON Onp 11 These facts are enough to determine the products of all groups in [φφ] (and hence, to determine the product of all elements of [φφ]), as- suming that multiplication in φ has already been determined. Any group in [φφ] has the form [φαβ], where α, β ∈ φ, and β is a group. We have [φαβ] = [φα]β, and based on the above lemmas, [φα] = f (φ) for some f (x) ∈ φ(x). Thus, [φα]β = f (φ)β, and the product of two such rational functions of φ is determined. So multiplication in Onp is completely determined by this collection of products that satisfy the MEX property. However, there are pairs of elements in Onp (for p ≥ 3) that do not satisfy the MEX property. The simplest exception: in On3, we have 3(3) = 2 (as we will see later), so 4(4) = (3+1)(3+1) = 2+3+3+1 = 6. But the minimal excludent of the set {α′β + αβ ′ − α′β ′}, where α′ and β ′ range over all ordinals less than 4, can be shown to be 2. So we certainly cannot use αβ = mex{α′β +αβ ′ −α′β ′} as a genetic definition of multiplication in Onp. It might be suspected that, if not all pairs of elements in Onp satisfy the MEX property, then perhaps all pairs of groups in Onp satisfy the MEX property. We would then be able to extend from products of groups to the full multiplication table, via the distributive property. But this also turns out to be false. In fact, there are pairs of rings that don't satisfy the MEX property. (However, from the previous lemmas, all pairs of fields do satisfy the MEX property.) Theorem 2.9. For any prime p ≥ 3, if φ ∈ Onp is an algebraically closed field, then the pair {[φω·2], [φω]} does not satisfy the MEX prop- erty. Proof. We have [φω·2][φω] = 1 φ = 1 φ−1 φω(p − 1)]. We will show that γ = [φω·2 + φω] = 1 MEX set of {[φω·2], [φω]}. φ−1 − 1 φ = 1 1 φ−1 + [p−1] φ−1 + 1 φ = [φω·2 + φ is not in the φ + 1 An arbitrary element of the MEX set of {[φω·2], [φω]} has the form g1(φ) 1 φ−1g0(φ) − g1(φ)g0(φ), where g0(x), g1(x) ∈ φ(x), g0(x) is a polynomial, and g1(x) has no poles except for possibly 0. We must choose appropriate rational functions g0(x) and g1(x) so that f (x) = g1(x)( 1 x − g0(x)) + 1 x−1g0(x) equals x−1 + 1 x. 1 If g1(x) has a pole at 0, then f (x) would have a pole at 0 of degree at least two. So g1(x) must be a polynomial. Furthermore, if g1(x) = a + xh1(x), and g0(x) = b + (x − 1)h0(x), then we would have f (x) = x−1 + a x + h(x) for some polynomial h(x). So we must have a = b = 1. Plugging those things in and simplifying, we find that f (x) = 1 x−1 + b 12 JOSEPH DIMURO 1 x + h(x), where h(x) = (−x2 + x)h1(x)h0(x) + (−x + 1)h1(x) + (−x + 2)h0(x) − 1. But it is then impossible to have h(x) = 0; if either h1(x) or h0(x) is a nonzero polynomial, then h(x) would be a polynomial of degree at least one. And if h1(x) = h0(x) = 0, then we would have h(x) = −1. So x−1 + 1 regardless of the choice of g1(x) and g0(x), f (x) cannot equal x. Thus, γ is not in the MEX set of {[φω·2], [φω]}; so the pair {[φω·2], [φω]} (cid:3) does not satisfy the MEX property. 1 All this raises the question: is there a purely genetic definition of multiplication in Onp? Given that the property αβ = mex{α′β + αβ ′ − α′β ′} does not even hold for rings, finding a genetic definition of multiplication would seem to be very difficult. So we shall have to rely on the inductive definitions. 3. The algebraic closure of p in Onp We will now focus on the structure of Onp below the first transcen- dental. As we've seen, all ordinals that are fields, but not algebraically closed, will define algebraic extensions of themselves. All ordinals that are algebraically closed fields define transcendental extensions of them- selves; if φ is an algebraically closed field, then since φ is not an element of itself, φ must be transcendental over itself! So we will refer to such elements φ as "transcendentals" in Onp. In [2], Conway describes the general structure on On2 below the first transcendental, [ωωω ]. For example, the first fields in On2 are of the form [22n], and each is a quadratic extension of the previous one: 22 = 3, 42 = 6, 162 = 24, 2562 = 384, and so on. (Each field, when squared, produces the [sesquimultiple] of that field.) The next fields are of the form [ω3n], and each is a cubic extension of the previous one: ω3 = 2, [ω3]3 = ω, [ω9]3 = [ω3], etc. Then we have the quintic extensions, and so on. In this section, we will see that for any prime p, the field Onp has a similar structure below the first transcendental. Borrowing the notation in [5], if r = [un] is a prime [power] (u prime, n a positive integer), and if k is the number of primes less than u (so k = 0 if u = 2, k = 1 if u = 3, etc.), then we write χr = [p(ωk un−1)]. Note that we then have χr = [ω(ωk−1un−1)] if k ≥ 1, so there will be no ambiguity in writing χr without reference to p as long as u ≥ 3. (If u = 2, and if there is a chance of ambiguity, we will write χr,p for [p2n−1].) Note that we have χ[u 2 ] if and only if either u1 < u2, or u1 = u2 and n1 < n2. So, we have χ2 ⊆ χ4 ⊆ χ8 ⊆ · · · ⊆ χ3 ⊆ χ9 ⊆ 1 ] < χ[u n1 n2 ON Onp 13 · · · ⊆ χ5 ⊆ χ25 ⊆ · · · . Also, the supremum of all the elements χr is [pωω ] = [ωωω ]. Our main goal of this section is to prove the following theorem. Theorem 3.1. The following hold for all primes p: (1) The first transcendental in Onp is [ωωω ]. (2) The ordinals in [ωωω ] ∈ Onp that are fields are exactly the χr for prime [powers] r. (3) For each prime u and each integer n ≥ 2, χ[un] is a uth degree extension of χ[un−1]. Also, χ[un] is closed under all extensions of degree u′, for any prime u′ < u. (4) If p 6= u, then we have (χu)u = αu, where αu is the smallest ordinal in χu with no uth root in χu. (5) We have (χ[un+1])u = χ[un] whenever p 6= u and n ≥ 1, except in the case where [un+1] = 4 and p ≡ 3 mod 4. (6) We have (χ4)2 = χ2 + 1 when p ≡ 3 mod 4. (7) Finally, we have the case where p = u: for n ≥ 2, we have (χ[pn])p = χ[pn] + then we have (χp)p = χp + 1. n−1Yk=1 (χ[pk])[p−1] = χ[pn] +(cid:2)(χp)pn−1(cid:3). If n = 1, For example, consider the structure of On3 below the first transcen- dental. The first extensions are all by square roots: we have 32 = 2, 92 = 4 (not 3, since 3 ≡ 3 mod 4), 812 = 9, 65612 = 81, and so on. Then we have the cubic extensions; we have ω3 = ω + 1, [ω3]3 = [ω3 + ω2], [ω9]3 = [ω9 +ω8], [ω27]3 = [ω27 +ω26], and so on. Then come the quintic [ωω]5 = 10, [ωω·5]5 = [ωω], [ωω·25]5 = [ωω·5], etc. And this extensions: pattern continues throughout all extensions. (In Section 4, we will look at methods for determining the values of αu for u 6= p: in On3, we have α2 = 2, α5 = 10, etc.) We will prove Theorem 3.1 by a sequence of lemmas. Lemma 3.2. Every element of Onp below the first transcendental is contained in a finite field within Onp. True by induction: given an element α below the first transcendental, let χ be the smallest ordinal where χ is a field and χ > α. If χ is finite, then χ is the finite field we want; so assume χ is infinite. From the work in Section 1, χ is an algebraic extension of a smaller field χ′, so α is a root of some polynomial f (x) ∈ χ′[x] irreducible over χ′. By induction, each coefficient of f (x) is contained in a finite field, so there is a finite field F containing all the coefficients of f (x). If the degree 14 JOSEPH DIMURO of f (x) is n, then F (α) will be a field in Onp containing α, and it will have order F n, which is finite. So F (α) is the field we want. Lemma 3.3. Let χ ∈ Onp be a field below the first transcendental. Then for any prime u, the following are equivalent: 1. For any f (x) ∈ χ[x] of degree u such that f (x) is irreducible over the field generated by its coefficients, χ contains the roots of f (x). 2. χ contains finite fields of order [pun] for all n ∈ ω. Proof. 1 ⇒ 2: This can be proven by induction on n. Since χ is a field of characteristic p, χ obviously contains a field of order [pu0] = p. Assume that F ⊆ χ is a field of order [puk]. Let f (x) ∈ F [x] be a polynomial of degree u that is irreducible over F ; this polynomial has a root α ∈ χ, but then F (α) ⊆ χ is a field of order [puk+1]. 2 ⇒ 1: Let f (x) ∈ χ[x] be a polynomial of degree u that is irreducible over F ⊆ χ, the finite field generated by the coefficients of f (x). Say F = [pmun], where m is not a multiple of u; then all of the roots of f (x) are in the subfield of Onp of order [pmun+1]. But χ contains both a field of order [pmun] (namely, F ) and a field of order [pun+1] (by assumption), so χ must contain a field of order [pmun+1]. So χ contains the roots of f (x). (cid:3) Lemma 3.4. Let χ ∈ Onp be a field below the first transcendental. Let u1, u2, . . . , um be primes. Then the following are equivalent: 1. For every i, the following holds: for any f (x) ∈ χ[x] of degree ui that is irreducible over the field generated by its coefficients, χ contains the roots of f (x). 2. If all the prime factors of n ∈ N are among the ui, then the following holds: for any f (x) ∈ χ[x] of degree n that is irreducible over the field generated by its coefficients, χ contains the roots of f (x). Proof. 2 ⇒ 1: trivial, as we can just take n = ui for each ui in turn. 1 ⇒ 2: let f (x) ∈ χ[x] be a polynomial of degree n that is irreducible over F ⊆ χ, the finite field generated by its coefficients. Then all the roots of f (x) are in the finite field of order F n. But this field can be built up from F using extensions of prime degree (i.e. of degree ui for some i), and by assumption, the fields resulting from each extension will be contained in χ. So the roots of f (x) will be in χ. (cid:3) Essentially, this proves parts 1 through 3 of Theorem 3.1. Below the first transcendental, the first fields will define quadratic extensions of themselves. If φ is a field, and the next field eφ is a quadratic extension, then we have seen that eφ = [φ2]. So the first fields are p, [p2], [p4], and so on. The supremum of these fields ([pω] = ω) is quadratically closed, ON Onp 15 since it contains fields of order [p2n] for all n. The subsequent fields [ω3], [ω9], etc. will each be cubic extensions of the previous field; there will be no need for further quadratic extensions, since these fields will still contain fields of order [p2n] for all n. Once we have a cubically closed field, then come the quintic extensions (with no need for more cubic extensions), and so on. And the supremum of all these fields is the first transcendental, [ωωω ]. To prove the next parts of Theorem 3.1, we will introduce some notation from [5]. For α ∈ [ωωω ], we will let d(α) be the degree of the minimal polynomial of α over the field p. (So, the smallest field containing α has order [pd(α)].) Also, if α 6= 0, we will let ord(α) be the multiplicative order of α: i.e. the smallest n ∈ N where αn = 1. Lemma 3.5. For every nonzero α ∈ [ωωω ], d(α) is the smallest n ∈ N where ord(α) divides [pn − 1]. Proof. Let n = d(α). Then α is contained in a field of order [pn], but not contained in a field of order [pm] for any m < n. So α is a root in Onp of x[pn] − x (hence of x[pn−1] − 1), but not of x[pm] − x (nor of x[pm−1] − 1) for any m < n. So the multiplicative order of α is a factor of [pn − 1], but not of [pm − 1] for any m < n. (cid:3) From here on, we will say that r ∈ N is a "primitive divisor" of [pn − 1] (for n ∈ N) if r divides [pn − 1], but r does not divide [pm − 1] (Any factor of [p1 − 1] is automatically a for any m ∈ N, m < n. primitive divisor of [p1 − 1].) Thus, we have proven that for every nonzero α ∈ [ωωω ], ord(α) is a primitive divisor of [pd(α) − 1]. Also, if u, m, n ∈ N, we will write [um] k n if [um] divides n, but [um+1] does not divide n. (If u does not divide n, then we will write [u0] k n.) Lemma 3.6. Say F ⊆ Onp is a field of order [pn], α ∈ F , and u is a prime. Assume [um] k [pn − 1]. Then α is a uth power in F (i.e. there exists a β ∈ F where βu = α) if and only if one of the following occurs: 1. m = 0 (so [pn − 1] is not a multiple of u) 2. [um] does not divide ord(α). Proof. If m = 0, then there exists k ∈ N such that [ku] ≡ 1 (mod [pn − 1]). Let β = αk ∈ F ; then βu = α[ku] = α. If m > 0, then assume ord(α) = [uks], where u is not a factor of s. Then k ≤ m, since ord(α) divides [pn − 1], and um is the largest power of u dividing [pn − 1]. Let β ∈ Onp be a uth root of α; then ord(β) [uk+1s]. (We'll actually have ord(β) = [uk+1s] if k ≥ 1; if k = 0, ord(β) may either be s or us.) If k < m (so that [um] does 16 JOSEPH DIMURO not divide ord(α)), then ord(β) [pn − 1], so β ∈ F . If k = m, then since [um] divides ord(α), [um+1] must divide ord(β). So ord(β) does not divide [pn − 1], so β 6∈ F . (cid:3) This shows that, below the first transcendental, no pth degree exten- sions in Onp will be by pth roots. The reason: if α ∈ [ωωω] is contained in a field of order [pn], then α is already a pth power in that finite field, since [pn − 1] is not a multiple of p. But as we will see, if u is a prime other than p, then all extensions of degree u below the first trascendental will be by uth roots. At this point, we will need a number-theoretic lemma before proving parts 4 through 6 of Theorem 3.1. Lemma 3.7. If u is a prime, and s, n are natural numbers (s ≥ 2) such that u [s − 1], then with one exception, ua k [sn − 1] iff ua k [n(s − 1)]. The exception: if u = 2, n is even, and s ≡ 3 (mod 4), then ua k [sn−1] iff ua k [n(s + 1)]. Among other things, this means that if u divides [s − 1] but does not divide n, then the u-part of [s − 1] equals the u-part of [sn − 1]. Proof. We have [sn − 1] = [(s − 1)(sn−1 + sn−2 + · · · + s + 1)]. Since s ≡ 1 (mod u), we have [sn−1 + sn−2 + · · · + s + 1] ≡ n (mod u). So if n is not a multiple of u, then neither will [sn−1 + sn−2 + · · · + s + 1] be, and the u-part of [sn − 1] will equal the u-part of [s − 1]. We will now show that, as long as u > 2, the u-part of [su − 1] equals the u-part of [u(s − 1)]. That will suffice to prove the theorem, except for the case where u = 2; by repeatedly pulling off factors of u, we find that the u-part of [suk −1] equals the u-part of [uk(s−1)] for all positive integers u. Thus, if n = muk (where u does not divide m), then the u-part of [sn − 1] equals the u-part of [uk(sm − 1)], which equals the u-part of [uk(s − 1)], which equals the u-part of [n(s − 1)]. We have [su−1 + su−2 + · · · + s + 1] = [u + (s − 1)[su−2 + 2su−3 + · · · + (u − 2)s + (u − 1)]]. Since s ≡ 1 (mod u), we have [su−2 + 2su−3 + · · · + (u −2)s + (u −1)] ≡ [1 + 2 + · · ·+ (u −1)] = [ u(u−1) ] ≡ 0 (mod u). Thus, [su−1 + su−2 + · · · + s + 1] is of the form [u + u2m] for some positive integer m; the u-part of [su−1 + su−2 + · · · + s + 1] must be exactly u. Thus, the u-part of [su − 1] = [(s − 1)(su−1 + su−2 + · · · + s + 1)] must equal the u-part of [u(s − 1)]. 2 What remains is the case where u = 2. In that case, the 2-part of [s2 − 1] = [(s − 1)(s + 1)] equals the 2-part of [2(s − 1)] if s ≡ 1 (mod 4), but it equals the 2-part of [2(s + 1)] if s ≡ 3 mod 4. Since [sm] ≡ 1 (mod 4) whenever m is even, we can conclude that the 2-part of [s4 − 1] equals the 2-part of [2(s2 − 1)], the 2-part of [s8 − 1] equals ON Onp 17 the 2-part of [4(s2 − 1)], and so on. In general, the 2-part of [sn − 1] (for n even) always equals the 2-part of [ n 2 (s2 − 1)], which equals the 2-part of [n(s − 1)] (for s ≡ 1 (mod 4)) or the 2-part of [n(s + 1)] (for s ≡ 3 (mod 4)). (cid:3) We can now prove part 4 of Theorem 3.1. Lemma 3.8. For each prime u 6= p, there exists an element in χu that has no uth root in χu. Thus, if αu is the smallest such element, then χu u = αu. Proof. Assume that u is a primitive divisor of [pm−1]; assume [un] k [pm− 1]. We have m ≤ u − 1, since u must divide [pu−1 − 1]. Thus, all prime factors of m are less than u. Since χu is closed under extensions of degree less than u, χu must contain a finite field of order [pm], and thus must contain an element of multiplicative order [un]. However, χu cannot contain any elements of multiplicative order [un+1], for the following reason: for any α ∈ χu, let F ⊆ χu be a finite field containing α. Then F = [ps], where all prime factors of s are less than u. Then u [ps − 1] if and only if s is a multiple of m; if that is true, then since u does not divide s, the u-part of [ps − 1] is the same as the u-part of [pm − 1] (namely, [un]). It is thus impossible for [un+1] to divide [ps − 1], so no element of F (including α) can have multiplicative order [un+1]. Thus, if α ∈ χu has multiplicative order [un] (and there must be such an element in χu), then α has no uth root in χu; any uth root of α would have multiplicative order [un+1], and thus cannot be in χu. (cid:3) Now to prove parts 5 and 6 of Theorem 3.1. Lemma 3.9. We have (χ[un+1])u = χ[un] whenever p 6= u and n ≥ 1, with one exception: we have (χ4)2 = χ2 + 1 when p ≡ 3 (mod 4). Proof. Assume that we are not in the exceptional case: either u > 2, or u = 2 and p ≡ 1 (mod 4). By induction, we will assume that (χ[ui+1])u = χ[ui] whenever 1 ≤ i < n, and we will prove that (χ[un+1])u = χ[un]. Note: since (χ[un])[un] = αu, if ord(αu) = a, then ord(χ[un]) = [aun]. And if d(αu) = k, then d(χ[un]) = [kun]. Consider any β < χ[un]; we claim that β is a uth power in χ[un+1]. If β is already a uth power in its minimal field p(β), then we're done. Otherwise, if p(β) = [pb] (then d(β) = b), and if [uc] k [pb − 1], then [uc] k ord(β). (Note that c > 0; if u did not divide [pb − 1], then β would be a uth power in p(β).) But then the finite field p(β, χ[un]) has dimension a multiple of [bu] (say, [bju]). And since [uc+1] [p[bju] − 1], 18 JOSEPH DIMURO but [uc+1] does not divide ord(β), β must be a uth power in p(β, χ[un]). So all ordinals less than χ[un] are uth powers in χ[un+1]. But χ[un] is not a uth power in χ[un+1], for the following reasons: if d(αu) = k, and if [um] k [pk−1], then [um] k ord(αu). So, [um+n] k ord(χ[un]). Consider any finite field F ⊆ χ[un+1] containing χ[un]; it has dimension unt, where t is a multiple of k, and all prime factors of t are less than u. We have [um+n] k [punt − 1]; since [um+n] k ord(χ[un]) also, χ[un] is not a uth power in any finite field in χ[un+1] (hence, it is not a uth power in χ[un+1]). So χ[un] is the smallest element of χ[un+1] that has no uth root in χ[un+1]; we must have (χ[un+1])u = χ[un]. Now assume u = 2 and p ≡ 3 mod 4. The same reasoning as before shows that if β < χ2 = p, then β is a square in χ4 = [p2]. But χ2 will also be a square in χ4: since p ≡ 3 mod 4, we have [21] k [p − 1]. So [21] k ord(α2), and [22] k ord(χ2). But [23] [p2 − 1]. So the largest [power] of 2 dividing [p2 − 1] does not divide the order of χ2, so χ2 is a square in the field of order [p2] (namely, χ4). So if p ≡ 3 mod 4, then χ2 is a square in χ4. If χ2 + 1 is a square in χ4, then for some a, b ∈ p, we have χ2 + 1 = (aχ2 + b)2 = (2 · ab)χ2 + (a2α2 + b2). So we have a2α2 + b2 = 2 · ab (both sides of that equation are equal to 1), and by manipulating that equation, we obtain (b − a)2 = a2(1 − α2). Since (b − a)2 and a2 are squares in χ2, 1 − α2 must also be a square in χ2. But it is not possible for 1 − α2 to be a square in χ2, for the following reasons: since α2 is the smallest non-square in χ2, [α2 − 1] = α2 − 1 is a square in χ2. And since p ≡ 3 (mod 4), −1 is not a square in χ2. So 1 − α2 = (−1)(α2 − 1), the product of a non-square and a square in χ2, cannot be a square in χ2. So there is no element aχ2 + b ∈ χ4 whose square is χ2 + 1; we must have χ2 4 = χ2 + 1. However, the inductive step works thereafter: if [2m′] k ord(χ2 + 1), then [2m′] k [p2 − 1], so [2m′+1] k [p4 − 1]. And since [2m′+1] k ord(χ4), χ4 will not be a square in χ8. So χ2 8 = χ4, and we may proceed by induction as before. So (χ[un+1])u = χ[un] in all cases, except that (χ4)2 = χ2 + 1 when p ≡ 3 mod 4. (cid:3) It remains to prove part 7 of Theorem 3.1. For simplicity, let χ = χp; then χ[pn] = [χpn−1] for all n ∈ N. As discussed, each of these fields is a degree p extension of the previous field, but not by a pth root (since all elements of finite fields of characteristic p already have pth roots in that finite field). So each χ[pn] will be an extension of χ[pn−1] by a root of a polynomial of form xp −x−α, assuming there is such a polynomial with no roots in χ[pn]. And as we will see, there will always be such a ON Onp 19 polynomial; the smallest suitable α will be (χ[pk])[p−1] = [χpn−1]. n−1Yk=1 Let f (x) = xp − x; for any field F ⊆ Onp, f is an additive homomor- phism from F to itself. Let S be the set of all ordinals in χ that are [multiples] of p; that is, all ordinals of the form [pδ] for some ordinal δ. Lemma 3.10. The map f (x) sends the ordinals in χ to exactly the ordinals in S. Proof. First, we'll show that only ordinals in S get mapped to. Since f is an additive homomorphism, we only need to show that the elements of χ that are groups get sent to elements of S. Let α be an arbitrary group in χ; we can assume α > 1, since f (1) = 0 ∈ S. Then α has the form χm r δ, where r = [un] is a [power] of a prime u < p, m is a positive integer less than u, and δ ∈ χr is a group. r δ) = χ[mp] r δp − χm r δ = χ[au+b] r Then f (α) = f (χm r δ, where a, b are nonnegative integers, and b < p. Since [mp] is not a multiple of u, b > 0. We then have χ[au+b] r δ, where δ′ = (χu r δ are in S; thus, so is χb r δ. So all groups in χ (and hence, all elements of χ) are mapped by f into S. r )aδp ∈ χr. Since b, m > 0, both χb r(δ′) − χm r(δ′) and χm r(χu r )aδp −χm r δ = χb δp −χm r δ = χb r(δ′)−χm δp − χm r It remains to show that f maps χ to the entire set S. Choose an arbitrary α ∈ S, and let F = p(α), the smallest finite field containing α. Then f is an additive homomorphism from F to itself, and its kernel contains p elements (namely, the ordinals less than p). So the image of f on F contains [F /p] elements. But the image must contain only elements of F ∩ S, and there are exactly [F /p] elements in F ∩ S. So all elements of F ∩ S, including α itself, must be mapped to by some element of F . So all elements of S are mapped to by some element in χ. (cid:3) This is enough to establish the following: Corollary 3.11. χ is a root of xp − x − 1. The reason: all polynomials in χ[x] of form xp −α already have roots in χ. So does the polynomial xp − x (namely, all the ordinals less than p). But xp − x − 1 has no root in χ; given any α ∈ χ, f (α) = αp − α is a [multiple] of p, hence cannot be 1. So χ must be a root of xp − x − 1. The rest of part 7 of Theorem 3.1 will be established by the following theorem: 20 JOSEPH DIMURO Theorem 3.12. For each field χ[pn+1] = [χpn] (for n ∈ ω), the image [χk]βk : βk ∈ χ, β[pn−1] ∈ S}. The smallest element of [χpn] that is not in Sn is [χpn−1]; thus [χpn] is a root of xp − x − [χpn−1]. of f on [χpn] is the set Sn = {P[pn−1] k=0 For example, in On3, we have χ = χ3 = ω. All elements of χ27 = [ω9] have the form [ω8]β8 + · · · + ωβ1 + β0, where each βi ∈ ω; the elements that are in the image of f on χ27 are exactly those for which β8 ∈ S. The smallest element of χ27 not of that form is [ω8]; thus, χ27 is a root of x3 − x − [ω8]. Proof. We will prove this by induction. We already established the n = 0 case, so we will assume it is true for n − 1, and prove it is true for n. For simplicity, let φ = [χpn−1], and let eφ be the next field, [χpn]. Let γ ∈ eφ; we have γ = φkγk, where each γk ∈ φ. Then f (γ) = f (φkγk) = (φ[kp]γp k − φkγk) [p−1]Xk=0 [p−1]Xk=0 [p−1]Xk=0 ([φ + χpn−1−1]kγp k − φkγk) = φ[p−1](γp [p−1] − γ[p−1]) + δ, = [p−1]Xk=0 where δ ∈ φ[p−1]. By the inductive hypothesis, (γp [p−1] − γ[p−1]) = f (γ[p−1]) = [χk]βk, [pn−1−1]Xk=0 where βk ∈ χ, β[pn−1−1] ∈ S. Thus, f (γ) is equal to [χpn−1]β[pn−1−1] plus a sum of "lesser terms"; f (γ) ∈ Sn. To show that the image of f on [χpn] includes the entire set Sn: let [pn−1]Xk=0 γ ∈ Sn, and let γ = [χk]γk, where each γk ∈ χ. (Then γ[pn−1] ∈ S.) Let F be the smallest finite field containing each γk and each [χk] for all k from 0 to [pn − 1]; then γ ∈ F . The mapping f is an additive homomorphism from F to itself, and its kernel contains p elements (the ordinals less than p). So the image of f over F must contain [F /p] elements. But the image is contained in F ∩ Sn, and there are exactly [F /p] elements in F ∩ Sn. So the image of f over F is exactly the set F ∩ Sn, which contains γ. So all elements γ ∈ Sn are mapped to by some element of [χpn]. (cid:3) ON Onp 21 That completes the proof of part 7 of Theorem 3.1: for each n, [χpn] is a root of xp − x − [χpn−1], since [χpn−1] is the smallest element of [χpn] that is not mapped to by f (x). In summary, we have now determined the full structure of [ωωω ] ∈ Onp, with the exception of the unknown elements αu ∈ Onp for each prime u 6= p. (We will discuss how to find each αu in the next sec- tion.) The structure is similar for all primes p; [ωωω] is always the first transcendental, and if φ ∈ On2 is an infinite field below [ωωω ], then φ is a field in every Onp. I would conjecture that this pattern continues beyond [ωωω ]: Conjecture 3.13. If φ ∈ On2 is an infinite field, then φ is a field in Onp for all primes p. If φ ∈ On2 is a transcendental, then φ is a transcendental in Onp for all primes p. But a proof of this conjecture would seem to be well out of reach. Only one transcendental in On2 is currently known: namely, [ωωω]. The problem of finding the second transcendental in On2 is wide open; the problem would seem to be equally difficult in Onp for other primes p. 4. Effective computation below the first transcendental in Onp We will now discuss methods for finding the elements αu ∈ Onp for each prime u 6= p. In [5], it is shown that the elements αu can be effectively determined in On2; we will show that the same can be done in Onp for every prime p. So multiplication can be done effectively in [ωωω]; division can be done effectively as well, since all elements of [ωωω] have finite multiplicative order. As before, for α ∈ [ωωω ], let d(α) be the degree of the irreducible polynomial of α over the field p. Based on our results from Section 3, we have the following proposition (which is identical to Proposition 1.8 in [5]): Proposition 4.1. We have that χr (for any prime [power] r) is the smallest element of [ωωω ] whose irreducible polynomial has degree di- visible by r. In other words, if r = [un] (for u prime), then χr is the set of all ordinals α ∈ [ωωω ] where d(α) is divisible only by primes ≤ u and where r does not divide d(α). Following [5], we will extend this notation and define χh for all pos- itive integers h; χh is the smallest ordinal α ∈ [ωωω] where d(α) is divisible by h. (We clearly have χ1 = 0.) The following lemma is a generalization of Lemma 2.5 in [5]. 22 JOSEPH DIMURO Lemma 4.2. Let β, γ ∈ Onp be elements of [ωωω ], and let r = [un] be a [power] of a prime u. If r divides d(β) but not d(γ), then r divides both d(β + γ) and d(β − γ). Proof. We have β ∈ p(γ, β + γ), which is an extension of p of degree lcm(d(γ), d(β + γ)). So d(β) divides lcm(d(γ), d(β + γ)). Since r divides d(β), r divides lcm(d(γ), d(β + γ)); since r does not divide d(γ), r must divide d(β + γ). Similar reasoning shows that r divides d(β − γ). (cid:3) Lemma 4.2 will be used in proving the following two results; they are generalizations to Onp of Theorem 2.1 and Corollary 2.2 in [5]. Theorem 4.3. Let h > 1 be a natural number, let u be the smallest prime dividing h, let r be the largest [power] of u dividing h, and let g = [h/r]. Then χh = χg if r divides d(χg); otherwise, χh = χg + χr = [χg + χr]. Proof. The theorem is proven by induction on the number of primes dividing h. If h = r, then g = 1; the theorem holds true in that case, since χh = 0 + χr = χg + χr, and r does not divide d(χg) = d(0) = 1. So assume h has at least two distinct prime divisors. Note that we must have χh ≥ χg, since g divides h. By the inductive hypothesis, χg is a finite sum of terms χr′, where each r′ is a [power] of a larger prime than u. Thus, each χr′ is larger than χr. We can thus conclude that, for all α ≤ χr, we have χg + α = [χg + α]. Specifically, we have χg + χr = [χg + χr]. Assume that r does divide d(χg). By definition, g also divides d(χg); thus, h divides d(χg). But χh is the smallest element of [ωωω ] whose minimal polynomial has degree a multiple of h; thus, χg ≥ χh. We already showed χh ≥ χg, hence they are equal. Now assume that r does not divide d(χg). Since r does divide d(χr), it follows from Lemma 4.2 that r divides d(χg + χr). Now, g and d(χr) are relatively prime, since every prime dividing d(χr) is at most u (based on the work in Section 3), but every prime dividing g is greater than u. On the other hand, g divides d(χg). So applying Lemma 4.2 to each prime [power] factor of g, we get that g must divide d(χg + χr). But if both g and r divide d(χg + χr), then h must divide d(χg + χr). Thus, χh ≤ χg + χr = [χg + χr]. We've now proven that χg ≤ χh ≤ [χg + χr]; that is only possible if χh = [χg+α] for some α ≤ χr. Thus χh = χg+α. But then α = χh−χg, and since r divides d(χh) but not d(χg), r must divide d(α) by Lemma 4.2. Since r divides d(α), we have α ≥ χr; thus, α = χr, and we're done. (cid:3) ON Onp 23 Corollary 4.4. For every natural number h, there is a unique finite set χr. (When there is ambiguity Q(h) of prime [powers] where χh = Xr∈Q(h) about the choice of field Onp, we will write Qp(h) for Q(h).) Every r ∈ Q(h) divides h and is relatively prime to [h/r]. Finally, if h > 1 and u is the largest prime dividing h, then the largest [power] of u dividing h belongs to Q(h). This corollary follows from Theorem 4.3 by induction; we can rewrite χh as either χg or χg + χr, and if g is not a prime [power], then we can rewrite χg in a similar fashion, and so on. Next, a lemma generalizing Lemma 3.4 from [5]. Lemma 4.5. If χ ∈ [ωωω ], then the multiplicative group of the field ω(χ) is generated by the elements χ + m, m ∈ ω. The proof is identical to the proof in [5]; the proof does not depend on the value of p. polynomial in ω[x] with χ as a root. Let β be any nonzero element of Proof. Let f (x) =P xiai ∈ ω[x] (where each ai ∈ ω) be the minimal ω(χ); assume β = P χjbj, where each bj ∈ ω. Let µ be the subfield g(x) = P xjbj and f (x) are both contained in µ[x]. Since f (x) is irreducible in µ[x], and since g(x) (a nonzero polynomial) has smaller degree than f (x), g(x) and f (x) must be relatively prime in µ[x]. of ω generated by all the coefficients ai and bj. Then the polynomials We may then apply Kornblum-Artin's analogue of Dirichlet's theo- rem on primes in arithmetic progressions (which appears on pg. 94 of [1] and on pg. 39 of [6]); if t ∈ ω is sufficiently large, then there exists a monic polynomial h(x) ∈ µ[x] of degree t where h(x) ≡ g(x) mod f (x) (hence, h(χ) = β). If t is chosen to be a [power] of 2, then since ω is quadratically closed, h(x) is a product of linear factors in ω[x]. If (χ + mi). (x + mi) (where each mi ∈ ω), then β = h(χ) =Yi So all nonzero elements of ω(χ) are products of elements of form χ + m (m ∈ ω), and that proves the lemma. (cid:3) h(x) =Yi We will use Lemma 4.5 to prove the next theorem, a generalization to Onp of Theorem 3.1 from [5]. It allows us to fully classify the elements αu up to a finite term. For any prime u 6= p, let ζu be a primitive uth root of unity in [ωωω]. Let f (u) = d(ζu); then u is a primitive divisor of [pf (u) − 1]. (Thus, f (u) is a divisor of [u − 1].) 24 JOSEPH DIMURO Theorem 4.6. For any prime number u 6= p, there exist natural num- bers m and m′ where αu = [χf (u) + m] = χf (u) + m′. As in [5], we will refer to m as the "excess" of αu over χf (u). Proof. By definition, αu is not a uth power in the field χu. Let F = p(αu), and let F ∗ be the multiplicative group of the field F . Consider the additive homomorphism θ1 : F ∗ → F ∗, where θ1(a) = au; αu is not in the image of this map, so it is not surjective. So it is not injective either; there are elements of F (other than 1) whose uth power is 1, and the only such elements are the primitive uth roots of unity. Thus, ζu ∈ F = p(αu). So d(αu) is a multiple of d(ζu) = f (u), so αu ≥ χf (u). On the other hand, since d(χf (u)) is divisible by f (u), ζu must be an element of p(χf (u)). Let G = p(χf (u)), and let G∗ be the multiplicative group of the field G. Consider the additive homomorphism θ2 : G∗ → G∗, where θ2(a) = au; since ζu ∈ G∗, this map cannot be injective. So it is not surjective either; there is some β ∈ G that is not a uth power in G. Now, we have G = p(χf (u)) ⊆ χu, and all extensions of p(χf (u)) contained within χu are of degree less than u. So, β is not a uth power in χu. Note that β ∈ ω(χf (u)); by Lemma 4.5, we can write β as a product of elements of the form χf (u) + m, m ∈ ω. Since β is not a uth power in χu, at least one of the factors χf (u) + m0 is also not a uth power in χu. Thus, αu ≤ χf (u) + m0 (since αu is the smallest element of χu that is not a uth power in χu). Finally, write χf (u) as [λ+m1], where λ is a limit ordinal and m1 ∈ ω. So, αu ≥ [λ + m1]. We have λ + m = [λ + m] for all m ∈ ω, so αu ≤ χf (u) + m0 = λ + m1 + m0 = [λ + m2], where m2 = m1 + m0. So we have [λ + m1] ≤ αu ≤ [λ + m2] for some natural numbers m1 and m2; αu = [λ + m1 + m] for some natural number m. From that, we can conclude the following: αu = [(λ + m1) + m] = [χf (u) + m], and αu = [λ + (m1 + m)] = λ + [m1 + m] = χf (u) − m1 + [m1 + m] = χf (u) + m′, for some natural number m′. That completes the proof. (cid:3) Using these theorems, and with the aid of Mathematica, I was able to assemble Tables 1 through 5, which contain the elements αu ∈ Onp for u ≤ 43, p ≤ 11. (The table for On2 first appeared in [5].) 5. Conclusion: thoughts about On0 To conclude, let's consider how we might turn the ordinals into On0, a Field of characteristic zero. The inductive construction from Section 1 works just as well in the characteristic zero case, so all that needs ON Onp 25 Table 1. αu ∈ On2 u f (u) Q(f (u)) excess 3 5 7 11 13 17 19 23 29 31 37 41 43 {2} {4} {3} {5} {3,4} {8} {9} {11} {7,4} {5} {9,4} {5} {7} 2 4 3 10 12 8 18 11 28 5 36 20 14 0 0 1 1 0 0 4 1 0 1 0 1 1 αu 2 [22] [2ω] + 1 [2ω2] + 1 [2ω] + [22] [24] [2ω· 3] + 4 [2ω4] + 1 [2ω3] + [22] [2ω2] + 1 [2ω· 3] + [22] [2ω2] + 1 [2ω3] + 1 Table 2. αu ∈ On3 u f (u) Q(f (u)) excess 2 5 7 11 13 17 19 23 29 31 37 41 43 ∅ {4} {3,2} {5} {3} {16} {9,2} {11} {7,4} {5,3} {9,2} {8} {7} 1 4 6 5 3 16 18 11 28 30 18 8 42 2 1 0 1 0 1 0 1 0 0 0 1 1 αu 2 [32] + 1 [3ω] + 3 [3ω2] + 1 [3ω] [38] + 1 [3ω· 3] + 3 [3ω4] + 1 [3ω3] + [32] [3ω2] + [3ω] [3ω· 3] + 3 [34] + 1 [3ω3] + 1 to be determined is how to define the smallest field φ0 (which will be isomorphic to Q, the field of rationals). We'll fill in the addition table first, then the multiplication table; for each possible sum or product, we'll choose the smallest ordinal that still allows us to construct a Field of characteristic zero. First of all, we let 0 + 0 = 0; that forces 0 to be the additive identity, so we have 0 + α = α for all ordinals α. Next, we cannot have 1 + 1 = 0 (or else 26 JOSEPH DIMURO Table 3. αu ∈ On5 ∅ {2} {3} {5} {4} {16} {9} u f (u) Q(f (u)) excess 2 3 7 11 13 17 19 23 29 31 37 41 43 1 2 6 5 4 16 9 22 14 3 36 20 42 2 1 1 0 1 1 1 0 1 1 0 0 1 {11,2} {7} {3} {9,4} {5,4} {7} αu 2 5 + 1 [5ω] + 1 [5ω2] [52] + 1 [58] + 1 [5ω· 3] + 1 [5ω4] + 5 [5ω3] + 1 [5ω] + 1 [5ω· 3] + [52] [5ω2] + [52] [5ω3] + 1 Table 4. αu ∈ On7 u f (u) Q(f (u)) excess 2 3 5 11 13 17 19 23 29 31 37 41 43 ∅ ∅ {4} {5} {3,4} {16} {3} {11} {7} {5,3} {9} {5,8} {3,2} 1 1 4 10 12 16 3 22 7 15 9 40 6 3 2 1 1 0 1 1 1 0 0 1 0 3 αu 3 2 [72] + 1 [7ω2] + 1 [7ω] + [72] [78] + 1 [7ω] + 1 [7ω4] + 1 [7ω3] [7ω2] + [7ω] [7ω· 3] + 1 [7ω2] + [74] [7ω] + 7 + 3 we no longer have characteristic zero) or 1 + 1 = 1 (since 1 is not the additive identity), but we can (and must) have 1 + 1 = 2. We then have 1 + 2 = 3, 1 + 3 = 4, and so on; 1 + α = [α + 1] for all α ∈ ω. But that forces the rest of the addition table for all finite ordinals: for α, β ∈ ω, we have α + β = [α + β]. We just have ordinary addition (and hence, ordinary multiplication) within ω. ON Onp 27 Table 5. αu ∈ On11 ∅ {2} ∅ {3} {3,4} {16} {3} u f (u) Q(f (u)) excess 2 3 5 7 13 17 19 23 29 31 37 41 43 {11,2} {7,4} {5,3} {3} {5,8} {7} 1 2 1 3 12 16 3 22 28 30 6 40 7 αu 2 11 + 1 2 [11ω] + 1 [11ω] + [112] [118] + 1 [11ω] + 1 [11ω4] + 11 [11ω3] + [112] [11ω2] + [11ω] [11ω] + 1 [11ω2] + [114] [11ω3] + 1 2 1 2 1 0 1 1 0 0 0 1 0 1 Now, the next sum to determine is the sum of ω and 1. We can have ω + 1 = 0; ω can play the role of −1. Thus, ω + 2 = 1, ω + 3 = 2, etc. The next undetermined sum is then ω + ω; this cannot equal ω or anything in ω, so we must have ω + ω = [ω + 1]. So [ω + 1] = −2, and similarly [ω + 2] = −3, and so on. We thus obtain our first group: [ω · 2] is isomorphic to Z, the ring of integers. Since [ω · 2] is a group, [ω · 2] + α cannot be an element of [ω · 2] for any α ∈ [ω · 2]. So making the simplest choices at every step, we let [ω · 2] + 1 = [ω · 2 + 1], [ω · 2] + 2 = [ω · 2 + 2], and so on; we then have [ω · 2] + α = [ω · 2 + α] for all α ∈ [ω · 2]. The next sum to consider is [ω · 2] + [ω · 2]. This sum cannot be 0, but it can (and thus must) be 1, since [ω · 2] may play the role of 1/2. We thus get [ω · 4] as our next group, consisting of all halves of integers. We similarly get [ω · 4] = 1/4, [ω · 8] = 1/8, and so on; [ω2] is our next ring, the ring of dyadic rationals. We then have [ω2] + α = [ω2 + α] for all α ∈ [ω2], so the next sum to consider is [ω2]+[ω2]. This cannot be anything in [ω2], since all elements of [ω2] already have halves; thus, we must have [ω2] + [ω2] = [ω2 · 2]. However, we can (and must) have [ω2]+[ω2 ·2] = 1, letting [ω2] play the role of 1/3. Similarly, we have [ω2 · 3] = 1/9, [ω2 · 9] = 1/27, etc. Our next ring is then [ω3] = 1/5, and we then have [ω4] = 1/7, [ω5] = 1/11, and so on; for any odd prime p, if p is the kth prime, then [ωk] = 1/p. Our smallest field is thus [ωω], which is isomorphic to Q. 28 JOSEPH DIMURO With the smallest field φ0 thus constructed, we then use the same construction as for Onp to construct all of On0. Unfortunately, further analysis of On0 would seem to be very difficult. We were able to obtain a nearly complete analysis of Onp below the first transcendental, mostly because all elements below [ωωω] are contained in finite fields. But there are no finite fields of characteristic zero. I would imagine that finding the first transcendental in On0 would be as difficult as finding the second transcendental in any Onp. So we won't analyze On0 any further. References [1] E. Artin, Collected Papers, Addison Wesley, Reading, 1965. [2] J.H. Conway, On Numbers and Games 2nd edition, A K Peters, Ltd., Natick, MA, 2001. [3] F. Laubie, A recursive definition of p-ary addition without carry, Journal de Th´eorie des Nombres de Bordeaux, 11 (1999), 307-315 [4] H.W. Lenstra, Jr., Nim multiplication, I.H.E.S., Bures-sur-Yvette. [5] H.W. Lenstra, Jr., On the Algebraic Closure of Two, Proc. Kon. Ned. Akad. Wet. Series A 80, 389-396. [6] M.I. Rosen, Number Theory in Function Fields, Springer-Verlag, New York, 2002. Department of Mathematics and Computer Science, Biola Univer- sity, La Mirada, CA, USA E-mail address: [email protected]
1310.2140	2	1310	2019-06-07T07:28:11	Extending maps to profinite completions in finitely generated quasivarieties	[ "math.RA", "math.LO" ]	We consider the problem of extending maps from algebras to their profinite completions in finitely generated quasivarieties. Our developments are based on the construction of the profinite completion of an algebra as its natural extension. We provide an extension which is a multi-map and we study its continuity properties, and the conditions under which it is a map.	math.RA	math	EXTENDING MAPS TO PROFINITE COMPLETIONS IN FINITELY GENERATED QUASIVARIETIES GEORGES HANSOUL AND BRUNO TEHEUX Abstract. We consider the problem of extending maps from algebras to their profinite completions in finitely generated quasivarieties. Our developments are based on the construction of the profinite completion of an algebra as its natural extension. We provide an extension which is a multi-map and we study its continuity properties, and the conditions under which it is a map. 1. Introduction This paper is a contribution to the study of profinite completions in internally residually finite prevarieties. A class A of algebras is called [7] an internally residu- ally finite prevariety (IRF-prevariety for short) if there is a set M of finite algebras such that A = ISP(M). Every algebra A of an IRF-prevariety A embeds in its A-profinite completion proA(A), which is defined as the inverse limit of the inverse system of the finite quotients of A that belongs to A, with natural homomorphisms as bonding maps (see Section 2 for details). In what follows, we limit ourself to those IRF-prevareties A that are finitely generated quasivarieties, i.e., for which there is a finite set M of finite algebras such that A = ISP(M). Moreover, we assume that A = ISP({M}), but this restriction is a matter of convenience: we claim that our developments admit the obvious generalization to the multi-sorted case where M = {M1, . . . , Mn}. ι It is proved in [7] that proA(A) is isomorphic to the natural extension Aδ of A, that is, the topological closure of eA(A) in MA∗ , where A∗ = A(A, M), where eA : A → MA∗ is the evaluation map defined as eA(a)(φ) = φ(a), and where ι is the discrete topology on M (this representation result actually holds in any IRF- is a discrete structure that yields a natural duality prevariety). Moreover, if M for A, and if A∗ is considered as a (closed) substructure of M A, then Aδ can be f concretely computed as the algebra of structure preserving maps from A∗ to M f [7, Theorem 4.3]. With these results in mind, we adopt the notation Aδ to denote proA(A) for the remaining of the paper. f We consider the following problem: given A, B ∈ A and a map u : A → B, how to define a 'reasonable' extension uδ : Aδ → Bδ of u? Such an extension would allow to study profinite completions of expansions of A-algebras, and preservation of equations through profinite completions. This problem has a well known solu- tion [14] in the particular case where A = DL = ISP(2) is the variety of bounded distributive lattices, in which profinite completions coincide with canonical exten- sions. In this particular case, the theory of canonical extensions provides with a 2010 Mathematics Subject Classification. 03C05, 08C20. Key words and phrases. Profinite completions, natural dualities, natural extensions, canonical extensions. 1 2 GEORGES HANSOUL AND BRUNO TEHEUX lower and an upper extension of any map u : L → L′ to the canonical extensions of L and L′. However, in the more general setting of non lattice-based algebras, no method of extension of maps from algebras to their profinite completions has yet been developed. The paper is organized as follows. In section 2, we recall some results about profinite completions in IRF-prevarieties, and we set up the notations. In section 3, we introduce a new topology δ on Aδ such that A is the largest discrete subspace of Aδ δ. We prove that this topology boils down to the existing one [13, 15] in the specific case A = DL. Finally, we prove that if M yields a logarithmic full duality for A, then the construction of profinite completions (alias, natural extensions) commutes with the one of finite Cartesian products. We generalize this result to Boolean products in the Appendix. f f f Section 4 is the core of the paper. We work under the more restrictive assumption that yields a logarithmic duality for ). Given a map u : A → B, that there is a discrete topological structure M A and that M the topology inherited from M multi-map between Aδ and Bδ rather than a map. Nevertheless, under additional is injective in the dual category IScP(M f we use the topology δ to define an extension eu of u on Aδ. In general, the map eu is ι for every x ∈ Aδ (where ι is not valued on Bδ but eu(x) is a closed subspace of Bδ A(B,M)). It means that eu has to be considered as a continuity assumptions, we show that eu can be turned into a map valued in Bδ, a In Section 5, we study how the construction of eu interacts with function com- positions. We illustrate our developments by considering a sample case, namely, the case where A is the variety of median algebras (a non lattice-based variety). In particular, we exhibit an example of smooth map that is not a homomorphism nor an operation of the type of the algebra. We also prove that median algebras whose profinite completion is Boolean are exactly the Boolean powers of the 2-element median algebra. property that we call smoothness. f In section 6, we consider the special case where M can be equipped with a total order in such a way that every A ∈ A can be considered locally has a (semi)lattice. We also show how the construction of eu shed lights on the existence of an upper and a lower extension of u in the case of distributive lattices. We close the paper by some concluding remarks and topics of further research. 2. Preliminaries We work under the general setting of [6]. Let M = {M1, . . . , Mn} be a finite set of finite algebras of the same type, and let A be the prevariety ISP(M). In what follows we assume for convenience that M = {M}. We denote by M an alter-ego of M, i.e., a topological structure f = hM, G ∪ H ∪ R, ιi, M f where ι is the discrete topology on M , and G, H and R are respectively a set (pos- sibly empty) of algebraic operations, algebraic partial operations (with nonempty domain), and algebraic (nonempty) relations on M, respectively. We use X to de- note the topological prevariety IScP(M ), that is, the class of topological structures f that are isomorphic to a closed substructure of a nonempty power of M . For any X, Y ∈ X we denote by X (X, Y ) the set of the structure preserving continuous maps f : X → Y . We use X∗ to denote X (X, M f ). f EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 3 For any A ∈ A, we denote by A∗ the set A(A, M) of the homomorphisms from A to M. The Preduality Theorem ([6, Theorem 5.2]) states that if A∗ inherits the A, then A∗ ∈ X . Moreover, the evaluation map structure and the topology from M eA : A → (A∗)∗ : a 7→ eA(a) : φ 7→ φ(a) f is an A-embedding. Similarly, for any X ∈ X the map ǫX : X → (X∗)∗ : φ 7→ ǫX (φ) : x 7→ x(φ) is a X -embedding. If τ is a topology on a set X, we denote by hX, τ i or Xτ the corresponding topological space. In particular, we denote by Mι the topological algebra obtained by equipping M with the discrete topology. For any set X, the notation MX ι stands for the power algebra MX equipped with the product topology induced by ι on M . We denote by Aι the category of topological algebras that are isomorphic to a closed subalgebra of a nonempty power of Mι with continuous homomorphisms as arrows. For every A ∈ A, the map eA identifies A with the subspace eA(A) of MA∗ , and we usually consider A up to this identification. ι Definition 2.1 ([7]). The natural extension of A ∈ A, in notation Aδ, is the topological closure of A in MA∗ ι of Aι by considering it as a subspace of MA∗ . The algebra Aδ is turned into an element Aδ . ι ι f When M yields a natural duality for A, i.e., when the map eA : A → (A∗)∗ is an isomorphism for every A ∈ A, then Proposition 2.2 shows how to explicitly construct Aδ from A∗ without relying on any notion of (topological) limit. For any topological structure X and any topological algebra A, we denote respectively by X ♭ and A♭ the structure obtained from X and the algebra obtained from A by dropping the topology. We denote by X ♭ the category whose objects are the X ♭ where X ∈ X with structure preserving maps as arrows. By abuse of notation, we write X ♭(X, Y ) instead of X ♭(X ♭, Y ♭). Note that X ♭(A∗, M ) is a closed subalgebra f of MA∗ for every A ∈ A. ι Proposition 2.2 ([7, Theorems 3.6 and 4.3]). Assume that A ∈ A. (1) The definition of Aδ used to defined M ι is independent of the algebraic structure G ∪ H ∪ R and of the algebra M used to define A. ), ιi. (2) If in addition M f yields a duality for A then Aδ f ι is isomorphic to hX ♭(A∗, M f Recall that for any A ∈ A, the family {A/θ θ ∈ Con(A) and A/θ ∈ A is finite} with the natural bonding maps φθ,θ ′ : x/θ 7→ x/θ′ for every θ ≤ θ′ forms an inverse system, the inverse limit of which is called the A-profinite completion of A (or simply the profinite completion of A) and is denoted by proA(A). Any A ∈ A embeds in proA(A). If in addition A is a variety, then A/θ ∈ A for every congruence of A, and the construction of proA(A) does not rely on A and is commonly denoted by bA. The following result, which follows from [7, Theorem 3.6], states that under our assumption of a finitely generated prevariety A, the A-profinite completion of A ∈ A coincides with its natural extension Aδ. Proposition 2.3. If A ∈ A, then there is an isomorphism between proA(A) and Aδ that fixes A. 4 GEORGES HANSOUL AND BRUNO TEHEUX Informally speaking, Proposition 2.3 shows that natural extension is a tool to compute profinite completions. We close the section by introducing some notation. We write F ⋐ X if F is a finite subset of X. If τ is a topology on X and x ∈ X, then we denote by τx the set of open τ -neighborhoods of x. If b ∈ X Z τ for some Z and if F ⋐ Z, then we denote by [bF ] the basic open set {y ∈ X Z y↾F = b↾F } of X Z τ . If (X, ≤) is an ordered set and x ∈ X, then we denote by x↑ and x↓ the up-set and the down-set generated by x, respectively. 3. The topology δ for profinite completions In the distributive lattice-based setting, it is well known that the topology ι that naturally equips the canonical extension of a DL A can be enriched into a finer topology in which A is definable as the algebra of isolated points. Authors have used various notations for this topology: Gehrke and Jónsson denote it by σ in [14] and Gehrke and Vosmaer denote it by δ in [15]. We aim at defining a similar topology in the more general setting of a finitely generated prevariety A and A-profinite completions. 3.1. A topology to define A in its profinite completion. If X, Y ∈ X we denote by Xp(X, Y ) the set of partial morphisms from X to Y , i.e., the set of the maps f : dom(f ) → Y where dom(f ) is a closed substructure of X and where f ∈ X (dom(f ), Y ). Definition 3.1. If A ∈ A and f ∈ Xp(A∗, M f Of = {x ∈ X ♭(A∗, M f Then, we denote by ∆A, or simply ∆, the family ), we set ) x ⊇ f }. The topology δ is defined as the topology generated by ∆, and we denote by Aδ the topological algebraic structure obtained by equipping A with δ. ∆ = {Of f ∈ Xp(A∗, M f )}. Remark 3.2. (1) If M is injective in X , then ∆ is equal to the family of the sets OK,a := OeA(a)↾K where a ∈ A and K is a closed substructure of A∗. (2) It is not always possible to compare topologies δ and ι. Nevertheless, we have ι ⊆ δ if any finite subset of A∗ generates a finite substructure in A∗. In particular, we have ι ⊆ δ if M is a purely relational structure. f Recall that a strong duality is said to be logarithmic if (finite) coproducts in the dual category (they always exist since they are dual to products) are given by the direct unions, that is, disjoint unions with constants amalgamated (see section 6.3 in [6]). f f Lemma 3.3. If M yields a logarithmic duality for A, then ∆ is a basis of δ. Proof. Let A ∈ A and f, g ∈ Xp(A∗, M ). We prove that Of ∩ Og ∈ ∆ or Of ∩ Og = f ∅. First, we note that dom(f ) ∪ dom(g) is a substructure of A∗. Indeed, if ih is the inclusion map ih : dom(h) → A∗ for h ∈ {f, g}, and if sh is the canonical embedding from dom(h) into dom(f ) ∐ dom(g) for h ∈ {f, g}, then there is a morphism i : dom(f ) ∐ dom(g) → A∗ such that i ◦ sf = if and i ◦ sg = ig. It follows that Im(i) = dom(f ) ∪ dom(g) is a closed substructure of A∗. EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 5 If f ∪ g is not a function, i.e., if f and g do not coincide on dom(f ) ∩ dom(g), ), then or if f ∪ g is a function that does not belong to X ♭(dom(f ) ∪ dom(g), M f Of ∩ Og = ∅. If f ∪g ∈ X ♭(dom(f )∪dom(g), M f by continuity of f and g. It follows that Of ∩ Og = Of ∪g. ), then it also belongs to X (dom(f )∪dom(g), M ) (cid:3) f Lemma 3.4. Assume that A ∈ A. (1) The elements of X (A∗, M f (2) If M is injective in X and if ∆ is a basis of δ, then (A∗)∗ is dense in ) are isolated points in hX ♭(A∗, M f ), δi. ), δi. hX ♭(A∗, M f f f (3) If M is injective in X and yields a duality for A and if ∆ is a basis of δ, then A is a discrete dense subspace of Aδ δ. Proof. (1) If x ∈ (A∗)∗ then Ox = {x} ∈ δ. ). Since M is injective in X , there is an a ∈ (A∗)∗ such (2) Let f ∈ Xp(A∗, M f that a = f on dom(f ). It means by definition of δ that a ∈ Of ∩ (A∗)∗. (3) We know by (1) and (2) that eA(A) = (A∗)∗ is the subspace of isolated (cid:3) δ. The conclusion follows from Proposition 2.2 (2). points of Aδ f By combining Lemmas 3.3 and 3.4, we obtain the following proposition. Proposition 3.5. If M then A is the subspace of isolated points of Aδ f δ for every A ∈ A. is injective in X and yields a logarithmic duality for A, Let DL be the variety of bounded distributive lattices, that is, DL = ISP(2), where 2 = h{0, 1}, ∨, ∧i is the two-element lattice. Recall that 2 := h{0, 1}, ≤ , ιi where 0 ≤ 1 yields a logarithmic natural strong duality for DL, known as Priestley duality. If L ∈ DL, then Lδ coincides with the canonical extension of L, which can be constructed by Proposition 2.2 as the lattice of decreasing subsets of the L∗. In [14], the authors introduce a topology δ′ on Lδ (denoted by σ in [14] and by δ in [15]), and use this topology to extend maps between distributive lattices to their canononical extensions (i.e., their profinite completions). Recall that a basis of δ′ is given by the sets [F, O] where F is a closed element of Lδ (i.e., a closed decreasing subset of L∗), and O is an open element of Lδ (i.e., an open decreasing subset of L∗). In the next proposition, we prove that the topology δ defined in Definition 3.1 coincides with δ′. e Proposition 3.6. If L ∈ DL, then δ(Lδ) = δ′(Lδ). Proof. First, we prove that δ′ ⊆ δ. Let F and O be a closed and an open element of Lδ, respectively, and assume F ⊆ O. Then, G := F ∪ −O is a closed substructure be the map defined by f −1(0) = F . We have f ∈ X (G, 2 of L∗. Let f : G → 2 ) e e and [F, O] = Of . ). Then, f −1(0) is a decreasing clopen subset of dom(f ). Hence, it is a closed subspace of L∗ and F := f −1(0)↓ is a decreasing closed subspace of L∗. Similarly, F ′ = f −1(1)↑ is an increasing closed subspace of L∗. It follows that Conversely, let f ∈ Xp(L∗, M f x ∈ Of ⇔ f −1(0) ⊆ x and f −1(1) ⊆ −x, ⇔ F ⊆ x and x ⊆ −F ′. We conclude that Of = [F, −F ′] ∈ δ′. (cid:3) 6 GEORGES HANSOUL AND BRUNO TEHEUX Note that there is no statement equivalent to Proposition 3.6 for the variety of bounded lattices since it is not an IRF-prevariety. 3.2. Profinite completions and products. We aim to use the topology δ to de- fine extension of maps between algebras of A to their profinite completions. In this view, an important feature is that under our general assumptions the construction of profinite completions commutes with the construction of products, i.e., (A × B)δ ≃ Aδ × Bδ, (1) for every A, B ∈ A. Given a procedure to extend maps from algebras to their profinite completions, property (1) would allow us to extend n-ary operations (n ≥ 2) on A ∈ A to n-ary operations on Aδ. As proved in the next result, this property holds true under rather mild assumptions. Theorem 3.7. Assume that M is of finite type and that M for A. If A, B ∈ A, then f yields a full duality (A × B)δ ≃ Aδ × Bδ ⇐⇒ (A∗ ∐ B∗)♭ ≃ (A∗)♭ ∐ (B∗)♭. In particular, if M logarithmic duality for A, then (1) holds for every A, B ∈ A and we may assume that the isomorphism is also a ι- and δ- homeomorphism. yields a full f Proof. Assume that (A∗ ∐ B∗)♭ ≃ (A∗)♭ ∐ (B∗)♭. It then follows successively that (A × B)δ ≃ X ♭((A × B)∗♭, M f ≃ X ♭((A∗ ∐ B∗)♭, M f ♭) ♭) ♭) ≃ X ♭((A∗)♭ ∐ (B∗)♭, M f ≃ X ♭((A∗)♭, M f ≃ Aδ × Bδ, ♭) × X ♭((B∗)♭, M f (2) (3) (4) ♭) where (2) is obtained because full dualities turn products to coproducts, (4) follows from the fact that (Aι(·, Mι), X ♭(·, M ♭), e, ǫ) is a dual adjunction between Aι and f ISP(M ♭) and hence turns coproducts to products, and (3) holds by assumption. f Moreover, if M yields a logarithmic duality for A, the isomorphism given by the previous piece of argument is easily seen to be a ι- and δ-homeomorphism. Conversely, if (A × B)δ ≃ Aδ × Bδ, it follows successively that f (A∗ ∐ B∗)♭ ≃ (A × B)∗♭ ≃ Aι((A × B)δ, Mι) ≃ Aι(Aδ × Bδ, Mι) ≃ Aι(Aδ, Mι) ∐ Aι(Bδ, Mι) ≃ (A∗)♭ ∐ (B∗)♭ (5) (6) (7) (8) where (6) holds by assumption, and where (5), (7) and (8) follow from the fact that Aι(·, Mι) and X ♭(·, M ♭) (see f [10, Theorem 2.4]). (cid:3) ♭) define a dual equivalence between Aι and ISP(M f In the Appendix, we generalize Theorem 3.7 to Boolean products. EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 7 4. Multi-extension of maps Given a map u : A → B, we consider the problem of defining an extension of u between Aδ and Bδ. Our approach (Definition 4.2) provides with a mutli-extension ι . This multi- eu of u, that is, a map valued in the set of (closed) subsets Γ(Bδ extension eu enjoys some continuity properties (Theorem 4.5). We adopt the following assumption for the remainder of the paper. ι ) of Bδ Assumption 4.1. The structure M injective in X . f yields a logarithmic duality for A and M f is Surprisingly enough, as noted in [6], many known strong dualities are logarithmic and hence, satisfy Assumption 4.1. Definition 4.2. Let A, B ∈ A and u : A → B. The relational extension of u is the relation u defined as the topological closure of u in Aδ ι . The multi-extension of u is the map eu defined on Aδ by setting eu(x) = {y ∈ Bδ (x, y) ∈ u} for every δ × Bδ x ∈ Aδ. Let us recall that if hX, τ i is a topological space, then the set Γ(X) of closed subsets of X is a complete lattice. Moreover, if Y is dense in X, if C is a complete lattice, and if f : Y → C, then lim supτ f is the map defined on X by lim supτ f (x) = V{W f (Y ∩ U ) U ∈ τx}. The next lemma shows that eu can be computed analogously as the upper extension in the setting of bounded distributive lattices (see [14, Definition 2.13]). Lemma 4.3. Let A, B ∈ A and u : A → B. If bu : Aδ → Γ(Bδ as bu(a) = {u(a)}, then eu = lim supδ bu. Proof. For every x ∈ Aδ, we have lim supδbu(x) = \{u(U ∩ A)− U ∈ δx}, ι ) is the map defined where the closure is computed in Bδ and only if (x, y) ∈ u. ι . It follows directly that y ∈ lim supδbu(x) if (cid:3) Combining Proposition 3.5 with Lemma 4.3, and by compactness of Bδ ι we obtain directly the following result. Proposition 4.4. Let A, B ∈ A and u : A → B. (1) If a ∈ A, then eu(a) = {u(a)}. (2) If x ∈ Aδ then eu(x) is nonempty. The following theorem shows that eu enjoys similar continuity properties as the upper extension in the setting of bounded distributive lattices (see [14, Theorem 2.12]). Recall that if hX, τ i is a compact Hausdorff space, then the family of sets (cid:3)U = {F ∈ Γ(X) F ⊆ U }, U ∈ τ, is a basis of a topology σ↓, which is called the co-Scott topology. Theorem 4.5. Let A, B ∈ A and u : A → B. (1) The map eu : Aδ → Γ(Bδ (2) If u′ : Aδ → Γ(Bδ ι ) is (δ, σ↓)-continuous. ι ) is a (δ, σ↓)-continuous map such that u′(a) = {u(a)} for every a ∈ A, then eu(x) ⊆ u′(x) for every x ∈ X. 8 GEORGES HANSOUL AND BRUNO TEHEUX Proof. First, we prove the following claim. Claim. For any x ∈ Aδ and any F ⋐ B∗, it holds eu(x)↾F = T{u(V ∩ A)↾F V ∈ δx}. Proof of the Claim. The inclusion ⊆ is clear. Let us prove inclusion ⊇. Let α ∈ M F be such that α ∈ u(V ∩ A)↾F for every δ-neighborhood V of x. For any finite subset G of B∗ that contains F and any δ-neighborhood V of x, set KG,V := {y ∈ Bδ y↾G ∈ u(V ∩ A)↾G} ∩ [αF ]. We obtain by compactness that HG := T{KG,V V ∈ δx} is a nonempty closed subspace of Bδ ι . It follows again by compactness that the family {HG F ⋐ G ⋐ B∗} has an nonempty intersection H. Any element y of H belongs to eu(x) and satisfies y↾F = α, which proves that α ∈ eu(x)↾F . (1) We prove that eu−1((cid:3)U ) is an open subspace of Aδ ι . By compactness of Bδ δ for any open subspace U of Bδ ι , it suffices to consider the case where U is a finite union of basic open sets [αF ] where F ⋐ B∗ and α ∈ MF . We consider the case where U if the union of two such basic open sets [F1α1] and [F2α2], as the general case can family KF := {u(V ∩ A)↾F V ∈ δx} is a downward directed family of nonempty finite sets, so it has a nonempty intersection. Let W be any δ-neighborhood of x be proved in a similar way. Let x ∈ eu−1(cid:0)(cid:3)([F1α1]∪[F2α2])(cid:1) and F be F1 ∪F2. The such that u(W ∩ A)↾F = T KF . We prove that W ⊆ eu−1(cid:0)(cid:3)([F1α1] ∪ [F2 ∩ α2])(cid:1). Let z ∈ W . We obtain successively eu(z)↾F = \{u(V ∩ A)↾F V ∈ δz} ⊆ u(W ∩ A)↾F = \{u(V ∩ A)↾F V ∈ δx} = eu(x)↾F , (9) (10) (11) (12) where (9) and (12) are obtained by the Claim, where (10) holds because W is a δ-neighborhood of z, and (11) holds by definition of W . We deduce from (12) that (2) By definition of the map eu, it suffices to prove that R := {(x, y) ∈ Aδ eu(z) ⊆ [F1α1] ∪ [F2α2]. δ × Bδ ι y ∈ u′(x)} is a closed relation that contains u. We have u ⊆ R by assumption. ι such that y 6∈ u′(x). Since u′(x) is a closed subspace Now, let (x, y) ∈ Aδ of the Boolean space Bδ ι , there is a clopen subspace U of Bδ ι such that z 6∈ U and u′(x) ⊆ U . It follows by continuity of u′ that u′−1((cid:3)U )×(Bδ ι \U ) is a neighborhood of (x, y) disjoint from R. (cid:3) δ × Bδ Remark 4.6. It follows from the Claim stated in the proof of Theorem 4.5 that if u : A → B then eu(x)↾{φ} = ]φ ◦ u(x) for any x ∈ Aδ and any φ ∈ B∗. 5. Coninuity properties and function compositions Theorem 4.5 characterizes eu as the smallest (δ, σ↓)-continuous extension u′ : Aδ → ι ) of u. In this section, we investigate the properties of eu under additional con- Γ(Bδ tinuity assumptions. EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 9 5.1. Smoothness and strongness. The case where the relational extension u of u : A → B is a function leads us to the following natural definition. Definition 5.1. Let A, B ∈ A and u : A → B. We say that u is is smooth if u is a function, that is, if eu(x) is a one-element set for every x ∈ Aδ. In this case, we denote by uδ the map uδ : Aδ → Bδ defined by uδ(x) ∈ eu(x). Example 5.2. If ι ⊆ δ then any term function is smooth since it is (ι, ι)-continuous. Theorem 4.5 can be rephrased for smooth maps in the following way. Proposition 5.3. Let A, B ∈ A and u : A → B. (1) If u is smooth then uδ : Aδ → Bδ is a (δ, ι)-continuous extension of u. (2) If u admits a (δ, ι)-continuous extension u′ : Aδ → Bδ then u is smooth and uδ = u′. Propositions 3.6 and 5.3 show that the notion of smoothness as defined in Defi- nition 5.1 boils down to the one defined in [13] when it is considered for the variety of bounded distributive lattices. As a corollary of Proposition 5.3(2), we obtain that if u : A → B is not smooth, then it is not even possible to define a continuous extension uδ : Aδ x ∈ Aδ. ι by suitably picking up an element uδ(x) in eu(x) for every δ → Bδ Example 5.4. If ι ⊆ δ, then every element φ ∈ A∗ is smooth. Consider the map φ′ : Aδ → M defined by φ′(x) = x(φ). For any F ⊆ M , we have φ′−1(F ) = Sf ∈F [φ : f ] ∩ Aδ, which proves that φ′ is (ι, ι)-continuous, so it is (δ, ι)-continuous since δ ⊆ ι. The conclusion follows from Proposition 5.3 (2). Proposition 5.5. Let A, B ∈ A and u : A → B. The map u is smooth if and only if φ ◦ u is smooth for every φ ∈ Bδ. Proof. If u is smooth and φ ∈ B∗, then the map φ′ ◦ u where φ′(x) := x(φ) for any x ∈ Bδ is a (δ, ι)-continuous extension of φ ◦ u. It follows that φ ◦ u is smooth by Proposition 5.3(2). Conversely, assume that φ ◦ u is smooth for every φ ∈ B∗. We prove that the map u′ : Aδ ι defined as u′(x) = (φ ◦ u)δ ◦ x is continuous, and the conclusion follows from Proposition 5.3(2). Let F be a finite subset of B∗ and α ∈ M F . We have δ → Bδ u′−1([F α]) = \{(cid:0)(φ ◦ u)δ(cid:1)−1 ({α(φ)}) φ ∈ F }, which proves that u′ is (δ, ι)-continuous by Proposition 5.3(2) and our assumption. (cid:3) Example 5.6. If ι ⊆ δ, then every u ∈ A(A, B) is smooth. This result follows from Example 5.4 and Proposition 5.5. If M is of finite type, it can also be considered as a consequence of [10, Theorem 2.4]. Definition 5.7. Let A, B ∈ A and u : A → B. We say that u is is strong if eu is (ι, σ↓)-continuous. The proof of the following Lemma is straightforward. Lemma 5.8. Assume that ι ⊆ δ, and let u : A → B be a smooth map. Then u is strong if and only if uδ is (ι, ι)-continuous. 10 GEORGES HANSOUL AND BRUNO TEHEUX Example 5.9. If ι ⊆ δ, then every u ∈ A(A, B) is strong. We already know that u is smooth, and we prove as in Example 5.4 that uδ(x)(φ) = x(φ ◦ u) for every x ∈ Aδ and φ ∈ B∗. It follows that uδ is (ι, ι)-continuous, or equivalently, that eu is (ι, σ↓)-continuous by Lemma 5.8. Strongness can be used to obtain the preservation of functional composition through profinite completions, as illustrated in the next proposition. Proposition 5.10. Let A, B, C ∈ A, u : A → B and v : B → C. (1) If v is strong then vu ⊆ v ◦ u. (2) If u is smooth and if v is strong and smooth, then vu is smooth and (vu)δ = vδuδ. Proof. First, we prove the following claim. Claim. For any strong map u : A → B, the map qu : Γ(Aδ qu(K) = Seu(K) is (σ↓, σ↓)-continuous. ι ) → Γ(Bδ ι ) defined by Proof of the Claim. First, we prove that qu(K) is a closed subspace of Bδ ι for every closed subspace K of Aδ ι such that y 6∈ qu(K). For every x ∈ K let Vx and Wx be disjoint ι-neighborhood of eu(x) and y, respectively. By continuity of eu and compactness of K, there is a finite subset F of K such that {eu−1((cid:3)Vx) x ∈ F } covers K. It follows that W := T{Wx x ∈ F } is a ι-neighborhood of y that does ι . Let y ∈ Bδ not meet qu(K). Now, for any open subspace U of Bδ ι , it is not difficult to prove that The continuity of qu follows from the latter identity and strongness of u. qu−1((cid:3)U ) = (cid:3)eu−1((cid:3)U ). obtain (1) by Theorem 4.5. (1) By the Claim, the map qveu is a (ι, σ↓)-continuous extension of vu. Then, we ι that maps every x ∈ Aδ to the only (2) By the Claim, the function w : Aδ (cid:3) element of qveu(x) is continuous. Then, we obtain (2) by Proposition 5.3. δ → Cδ Corollary 5.11. Assume that ι ⊆ δ and let A, B, C ∈ A. (1) Any term function u := tA(sA 1 , . . . , sA ℓ ) is smooth and strong, and uδ = (tA)δ((sA 1 )δ, . . . , (sA ℓ )δ). (2) If u ∈ A(A, B) and v ∈ A(B, C), then vu is smooth and strong and (vu)δ = uδvδ. Proof. (1) The proof is obtained by induction on the construction of the term using Example 5.2, Lemma 5.8 and Proposition 5.10 (2) since any term function is (ι, ι)- continuous. (2) The proof is an application of Lemma 5.8, Example 5.9, and Proposition (cid:3) 5.10 (2). 5.2. A sample case: profinite completions of median algebras. In this sub- section, we illustrate the previous constructions by considering that A is the variety of median algebras, that is, A = ISP(2) where 2 = h{0, 1}, mi is the algebra with a single ternary operation m defined as the majority function on {0, 1}. This variety is of special interest as (i) it is not lattice-based, (ii) it admits a strongly logarithmic duality, and (iii) the dual category is locally finite. Hence, ι(Aδ) ⊆ δ(Aδ) for every A ∈ A. EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 11 5.2.1. A natural duality for median algebras. It is known [6, 20, 25] that the topo- logical structure = h{0, 1}, 0, 1, ≤, •, ιi 2 e with two constants 0 and 1, the natural order ≤, and the unary operation • defined by x• ≡ (x + 1) mod 2, yields a strong logarithmic duality for A. A topological structure X = hX, 0, 1, ≤, •, τ i is a object of the dual category X = IScP(2 ) provided e that hX, ≤, ιi is a Priestley space with bounds 0 and 1, that • is an order reversing homeomorphism that swaps 0 and 1 and that satisfies φ•• = φ, and φ 6≤ φ• for every φ 6= 0. There is an equivalent spectrum-based formulation of this duality that eases computations. A subset φ of a median algebra A is prime convex if for every x, y, z ∈ A, the element m(x, y, z) belongs to φ if and only if at least one of the sets {x, y}, {x, z}, {z, y} is a subset of φ. A subset x of a structure X ∈ X is a disjoint ideal of X if it is a downset set disjoint with x•. If in addition x is a clopen subset of X, then x is called a continuous disjoint ideal. A (continuous) maximal disjoint ideal of X is a (continuous) ideal that contains φ or φ• for every φ ∈ X. It is not difficult to show that the map φ 7→ φ−1(0) is an isomorphism between A∗ and the prime spectrum of A (i.e., the set of prime convex subsets of A) equipped with inclusion order, ∅ and A as bottom and top element respectively, set complementation as map •, and Zariski topology. If X ∈ X , then the dual X∗ of X is isomorphic to the set of continuous maximal disjoint ideals of X equipped with the operation m inherited from the median operation defined on the powerset of X as m(x, y, z) = (x ∩ y) ∪ (x ∩ z) ∪ (y ∩ z). (13) If A ∈ A, then Aδ is isomorphic to the set of the maximal disjoint ideals of A∗ equipped with the operation defined in (13). 5.2.2. Profinite completions of Boolean powers of 2. We can apply Theorem A.1 to compute profinite completions of Boolean powers of the median algebra 2. Proposition 5.12. If A is a median algebra that has a Boolean representation A ֒→ 2X , then Aδ ι is isomorphic (algebraically and topologically) to 2X ι . Proof. The dual of 2 is depicted in Figure 1. Observe that for every nonempty finite sets I and J, every a ∈ 2I and b ∈ 2J , identity 2∗ = [ [ai : 1] ∪ [ [bj : 0] i∈I j∈J holds if and only if Tj∈J [bj : 1] ⊆ Si∈I [ai : 1], that is, if and only if the following condition is satisfied in 2 (for some j0 ∈ J), ^ (bj = bj0) ⇒ _ j∈J i∈I (ai = bj0 ). The latter formula is also equivalent to _ (cid:0)m(ai, bk, bl) = ai(cid:1). k,l∈J; i∈I We conclude the proof by applying Theorem A.1. (cid:3) 12 GEORGES HANSOUL AND BRUNO TEHEUX 2 [0 : 0] [1 : 0] {0} {1} [1 : 1] [0 : 1] ∅ Figure 1. Dual of median algebra 2 Corollary 5.14 is a surprising consequenc of Proposition 5.12. We say that a me- dian algebra A = hA, mi is a Boolean if there is a Boolean algebra hA, ∨, ∧, ¬, 0, 1i such that m(x, y, z) = (x ∧ y) ∨ (x ∧ z) ∨ (y ∧ z) for every x, y, z in A. Recall that an algebra A = hA, m, ·ci of type (3, 1) is a ternary Boolean algebra [17] if hA, mi is a median algebra and the equation m(x, z, xc) = z holds in A. Lemma 5.13 ([17]). A median algebra is Boolean if and only if it is the {m}-reduct of a ternary Boolean algebra. Corollary 5.14. Let A be a median algebra. The following conditions are equiva- lent. (i) Aδ is Boolean. (ii) A is a Boolean power of 2. Proof. (i) =⇒ (ii) Let A be a Boolean median algebra, and let A♭ be a Boolean algebra whose {m}-reduct is A. Then A♭ can be represented as a Boolean power A♭ ֒→ 2X , where X is the Stone dual of A♭. This Boolean representation still holds between the {m}-reducts of A♭ and 2X. (ii) =⇒ (i) We know by Proposition 5.12 that we can identify Aδ with 2X . Denote by ·c the operation defined on 2X by xc(φ) ≡ 1 + x(φ) mod 2, φ ∈ X. Then h2X , m, ·ci is a ternary Boolean algebra, and we conclude the proof by Lemma 5.13 (cid:3) We conclude the section by giving an example of a smooth function which is not a homomorphism. Example 5.15. In the ∧-semillatice hA, ≤i depicted in Fig. 2, any three elements have an upper-bound whenever each pair of them is bounded above, and any prin- cipal ideal is a distributive lattice. Hence, it is a median semilattice [24]. It follows that the operation m defined on A as m(x, y, z) = (x ∧ y) ∨ (x ∧ z) ∨ (y ∧ z) turns A into a median algebra A. This operation can be easily computed explicitly: for every j, k, ℓ ∈ ω m(aj, ak, aℓ) = a(j,k,ℓ) m(aj, bk, bℓ) = a(j,k,ℓ) m(aj, ak, bℓ) = a(j,k,ℓ) m(bj, bk, bℓ) = a(j,k,ℓ), where (j, k, ℓ) denotes the median element of j, k, ℓ ∈ ω. Clearly, the elements of A∗ are Ai = ai↑, A• i = A \ ai↑, Bi = {bi}, B• i = A \ {bi}, i ∈ ω. EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 13 b2 b1 b0 a3 a2 a1 a0 Figure 2. Graph of median algebra A A1 A2 A3 A B• 2 B• 0 B• 1 A• 3 A• 2 A• 1 B1 B2 B0 ∅ Figure 3. The dual A∗ of A. Hence, the dual of A is depicted in Fig. 3. The elements of the bidual of A are easily computed: eA(bn) = An+1↓ ∪ A• n↓ = B• n↓, eA(an) = An+1↓ ∪ A• n↓ ∪ {Bn}, n ∈ ω, n ∈ ω. Then Aδ \ eA(A) = {∞} where ∞ = [{{A• n, Bn} n ∈ ω}. A simple computation shows that, up to identification of A with eA(A) m(∞, am, bn) = m(∞, am, an) = m(∞, bm, bn) = am∨n, m, n ∈ ω. Let us illustrate the inclusion ι ⊆ δ. For any φ ∈ A∗, the subasis clopen subsets {x φ ∈ x} and {x φ 6∈ x} of Aδ ι are respectively equal to Of and Og where f = {φ} and g = {φ•} correspond to morphisms defined on the closed substructure {φ, φ•} of A∗. Now, let u : A → 2 be the map defined by u(bi) = 1 and u(ai) = 0 for any i ∈ ω. Clearly, the map u is not a median homomorphism (neither a ∧-homomorphism). Let us denote by u′ the extension of u on Aδ that satisfies u′(∞) = 0. We prove that u′ is (δ, ι)-continuous which implies that u is smooth by Proposition 5.3. We have to prove that u′−1(0) = {∞, a0, a1, . . .} is a δ-open subset of Aδ. Consider K = {∅, B0, B1, B2, . . .} = Ti∈ω eA(ai). It follows from the continuity of • that K ∪ K • is a closed substructure of A∗. Hence, the map f : K ∪ K • → 2 defined by f (x) = 0 if and only if x ∈ K is a partial morphism on A∗. It is easily seen that ∞ ∈ Of ⊆ u′−1(0). 14 GEORGES HANSOUL AND BRUNO TEHEUX 6. Extensions of functions in ordered setting δ × Bδ Given a map u : A → B, Definition 4.2 provides a relation (or a multi-map) ¯u ⊆ Aδ ι that extends u. In the case of bounded (distributive) lattices A and B, the classical technique [11, 14] adopted to extend u to the canonical extensions (i.e., profinite completions) of A and B provides two functions: the lower extension uσ and the upper extension uπ. In this section, we reconcile these two approaches and prove that in the context of bounded distributive lattices, the multi-extension eu enables us to recover uσ and uπ, but not conversely. Our approach leads to more general results about varieties of algebras that are ι-locally semilattices (Definition 6.2). Notation 6.1. Let ≤ be a fixed total order on M . We denote by ι↑, respectively ι↓, the topologies formed by the upsets, respectively the downsets, of (M, ≤). We can use the total order ≤ defined on M to construct an upper and a lower extension of any map u : A → B. Definition 6.2. Let A, B ∈ A and u : A → B. We define the maps u△ : Aδ → M B∗ and u▽ : Aδ → M B∗ by u▽(x) := ^eu(x), u△(x) := _eu(x), for every x ∈ Aδ. We call u△ the upper extension of u, and u▽ the the lower extension of u. Lemma 6.3. If A, B ∈ A and u : A → B, then u▽(x)(φ) = V ^(φ ◦ u)(x) and u△(x)(φ) = W ^(φ ◦ u)(x). Proof. The proof follows from Remark 4.6. (cid:3) Theorem 6.5 gives sufficient conditions for u△ and u▽ to be valued in Bδ. Definition 6.4. An algebra A ∈ A is a local meet-semilattice if for every b, c ∈ A and every F ⋐ A∗, it holds (b ∧ c)↾F ∈ A↾F . Local join-semilattices are defined dually. A local lattice is an algebra of A that is both a local meet-semilattice and a local join-semilattice. Theorem 6.5. Let A, B ∈ A and u : A → B. (1) The map u▽ : Aδ → M B∗ (2) The map u△ : Aδ → M B∗ (3) If B is a local meet-semilattice, then the map u▽ is valued in Bδ. (4) If B is a local join-semilattice, then the map u△ is valued in Bδ. (5) If B is a local lattice, then u▽ and u△ are valued in Bδ. is a (δ, ι↑)-continuous extension of u. is a (δ, ι↓)-continuous extension of u. Proof. (1) Let F ⋐ B∗ and α ∈ M F . We obtain by Lemma 6.3 that (u▽)−1([F ≥ α]) = \{]φ ◦ u −1 ((cid:3)Uφ) φ ∈ F }, where Uφ := α(φ)↑ for every φ ∈ F . Then, the continuity of u▽ follows from Theorem 4.5. The fact that u▽ is an extension of u follows from Proposition 4.4. (2) is obtained from (1) by duality. (3) Let x be an element of Aδ and let us prove that u▽(x) is in the closure of B in MB∗ . We proceed as in the proof of Theorem 4.5 and for any F ⋐ B∗ we choose a δ-neighborhood W of x such that eu(x)↾F = u(W ∩ A)↾F . The family u(W ∩ A)↾F ι EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 15 is finite, and since B is a local meet-semilattice, there is some c ∈ B such that u▽(x)↾F = V u(W ∩ A)↾F = c↾F . We have proved that any δ-neighborhood of x meets B. (4) is obtained from (3) by duality, and (5) follows from (3) and (4) . (cid:3) Given a map u : L → L′ between two bounded distributive lattices L and L′, the theory of canonical extension [14] provides with the upper extension uπ : Lδ → L′δ and the lower extension uσ : Lδ → L′δ. The following corollary proves that they can be recovered from the multi-extension eu of u. Corollary 6.6. If L and L′ are two bounded distributive lattices and u : L → L′, then for every x ∈ L it holds uσ = u▽ and uπ = u△. Proof. The proof follows from the application of Theorem 6.5 to the variety A of bounded distributive lattices with M = {0, 1} ordered in the natural way. (cid:3) Example 6.7. Let L be the bounded distributive lattice made of ω and the finite subsets of ω with inclusion order. The Priestley dual L∗ = ω ∪ {∞} is the one point Alexandroff compactification of the antichain ω, with ∞ as top element. Hence, Lδ = 2ω ∪ {⊤} is the power set of L∗ \ {∞} with an additional top element ⊤ = L∗. (1) We easily build functions u : L → 2 that are smooth without being homo- morphisms. Indeed let u be the non trivial permutation of 2 and φ ∈ L∗. Then u ◦ φ : L → 2 is a smooth function that does not belong to L∗. Other examples are given by the maps uA : L → 2 (for A ⊆ ω) that are defined by uA(x) = 0 if and only if x ⊆ A. If A is infinite and co-infinite then uA is smooth but not strong. (2) The function u : L → 2 defined by u(X) = X mod 2 if X 6= ω and u(ω) = 1 is not smooth. Indeed, if X is an infinite proper subset of ω then eu(X) = {0, 1} = [u▽(x), u△(x)]. (3) The function u : L → 22 defined by u(X) = (X mod 2, (X + 1) mod 2), X 6= ω, u(ω) = (1, 1), is not smooth. Moreover, contrary to example (2), the set eu(x) is not determined by u▽(x) and u△(x). Indeed, if X is an infinite proper subset of ω then eu(X) = {(0, 1), (1, 0)} while u▽(x) = (0, 0) and u△(x) = (1, 1). (4) For k ≥ 2 let uk : L → L be the function defined by uk(X) = (1 + X mod k) × X for any X 6= ω and uk(ω) = ω. Then uk is not smooth. Indeed, if X is a proper infinite subset of ω then fuk(X) = {X, 2×X, . . . , k× X}. Moreover, we have ^ul ◦ uk = eul ◦fuk if and only if l and k are coprime. 7. Concluding remarks and further research In this paper, we have considered the question of extending functions between algebras to their profinite completions in the setting of finitely generated quasiva- rieties. Our answer is only partly satisfactory as we provide an extension which is a multi-map rather than a function. This multi-extension has strong continuity properties and there are interesting cases in which it turns out to be a function. Moreover, the construction of the multi-extension shed lights (Corollary 6.6) on the existence of two canonical extensions in the bounded distributive lattice setting. 16 GEORGES HANSOUL AND BRUNO TEHEUX We now identify some topics of further research. (1) Topology δ (Definition 3.1) is one of the possible topologies in which A can be defined as the algebra of isolated points of X b(A∗, M ) and is duality f dependent. A general study of the topologies that enjoy this property would lead to other multi-extensions which could be 'closer' to a function than the relation eu considered in this work. (2) Sufficient conditions for eu to be smooth are needed. (3) Canonical extensions have proved to be a useful tool to look for Kripke complete modal logics. Fields of applications of the techniques developed in this paper should be found outside the lattice-based setting (4) Median algebras and median semilattices are equivalent. Natural extensions of median algebras and canonical extensions of their median semilattices [16] should be compared. This constitutes a topic of current investigation. References [1] S. P. Avann, Metric ternary distributive semi-lattices. Proceedings of the American Mathe- matical Society,12:407 -- 414, 1961. [2] H. J. Bandelt and J. Hedlíková. Median algebras. Discrete mathematics, 45:1 -- 30, 1983. [3] G. Birkhoff and S. A. Kiss. A ternary operation in distributive lattices. Bulletin of the Amer- ican Mathematical Society, 53:749 -- 752, 1947. [4] Rukiye Cavus. Dualities induced by canonical extensions, June 2012. http://people.maths.ox.ac.uk/hap/CavusAbs.pdf. [5] B. M. Davey and H. Werner. Dualities and equivalences for varieties of algebras. In Contribu- tions to lattice theory (Szeged, 1980), volume 33 of Colloquia Mathematica Societatis János Bolyai, pages 101 -- 275. North-Holland, Amsterdam, 1983. [6] David M. Clark and Brian A. Davey. Natural dualities for the working algebraist, volume 57 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 1998. [7] B. A. Davey, M. J. Gouveia, M. Haviar, and H. A. Priestley. Natural extensions and profinite completions of algebras. Algebra Universalis, 66(3):205 -- 241, 2011. [8] B. A. Davey and H. A. Priestley. Canonical extensions and discrete dualities for finitely generated varieties of lattice-based algebras. Studia Logica, 100(1-2):137 -- 161, 2012. [9] B. A. Davey and H. A. Priestley. Canonical extensions and discrete dualities for finitely generated varieties of lattice-based algebras. Studia Logica, 100(1-2):137 -- 161, 2012. [10] Brian A. Davey, Miroslav Haviar, and Hilary A. Priestley. Natural dualities in partnership. Applied Categorical Structures, 20:583 -- 602, 2012. [11] Mai Gehrke and John Harding. Bounded lattice expansions. J. Algebra, 238(1):345 -- 371, 2001. [12] Mai Gehrke and Bjarni Jónsson. Bounded distributive lattices with operators. Math. Japon., 40(2):207 -- 215, 1994. [13] Mai Gehrke and Bjarni Jónsson. Monotone bounded distributive lattice expansions. Math. Japon., 52(2):197 -- 213, 2000. [14] Mai Gehrke and Bjarni Jónsson. Bounded distributive lattice expansions. Math. Scand., 94(1):13 -- 45, 2004. [15] Mai Gehrke and Jacob Vosmaer. A view of canonical extension. In Logic, language, and com- putation, volume 6618 of Lecture Notes in Comput. Sci., pages 77 -- 100. Springer, Heidelberg, 2011. [16] M.J. Gouveia and H.A. Priestley. Canonical extensions and profinite completions of semilat- tices and lattices. Order, 31(2):189 -- 216, 2014. [17] A. A. Grau. Ternary Boolean algebra. Bull. Amer. Math. Soc., 53:567 -- 572, 1947. [18] G. Hansoul and L. Vrancken-Mawet. Décompositions Booléennes de lattis distributifs bornés. Bull. Soc. R. Sci. Liège, 53:88 -- 92, 1984. [19] John Harding. On profinite completions and canonical extensions. Algebra Universalis, 55(2- 3):293 -- 296, 2006. Special issue dedicated to Walter Taylor. [20] John R. Isbell. Median algebra. Trans. Amer. Math. Soc., 260(2):319 -- 362, 1980. EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 17 [21] Bjarni Jónsson. On the canonicity of Sahlqvist identities. Studia Logica, 53(4):473 -- 491, 1994. [22] Bjarni Jónsson and Alfred Tarski. Boolean algebras with operators. I. Amer. J. Math., 73:891 -- 939, 1951. [23] Bjarni Jónsson and Alfred Tarski. Boolean algebras with operators. II. Amer. J. Math., 74:127 -- 162, 1952. [24] M Sholander. Medians, lattices, and trees. Proceedings of the American Mathematical Society, 5(5):808 -- 812, 1954. [25] H. Werner. A duality for weakly associative lattices. In Finite algebra and multiple-valued logic (Szeged, 1979), volume 28 of Colloq. Math. Soc. János Bolyai, pages 781 -- 808. North- Holland, Amsterdam, 1981. Appendix A. Profinite completions of Boolean products The generalization of Theorem 3.7 to Boolean products depends on the possibil- ity to express emptiness in the dual space in terms of formulas in the algebra, as seen in the next result. Recall the following notation: if a ∈ A and m ∈ M we denote by [a : m] the set {ψ ∈ A(A, M) ψ(a) = m}. The family {[a : m] a ∈ A, m ∈ M} is a basis of clopen subsets of A∗. The following theorem generalizes the developments in [18] about Boolean prod- ucts of bounded distributive lattices. Theorem A.1. Assume that M yields a logarithmic duality for A and that M is of finite type. Let A be a Boolean product of the family (Ai)i∈I of algebras of A. If for every n ∈ N and every m1, . . . , mn ∈ M there is an open formula φ(x1, . . . , xn) in the language of M such that for every i ∈ I and every a1, . . . , an ∈ Ai, it holds f A∗ i = n[ [aλ : mλ] ⇐⇒ Ai = φ(a1, . . . , an), λ=1 then Aδ is Aι-isomorphic to Qi∈I Aδ i . Proof. Let f : A ֒→ Qi∈I Ai be a Boolean representation of the family (Ai)i∈I of i ֒→ ∐{A∗ algebras of A. For every i ∈ I we denote by ρi the embedding (πi)∗ : A∗ i i ∈ I} where πi denotes the projection map from Qi∈I Ai onto its i-th factor Ai, i.e., ρi is the map defined by ρi(ψ) = ψ ◦ πi. Let X be the set S{ρi(A∗ i ) i ∈ I}. Since i ) i ∈ J} M f is isomorphic to ∐{A∗ i i ∈ J} for every finite subset J of I. It follows that X is a (not necessarily closed) substructure of ∐{A∗ i i ∈ I} (such a verification involves only finitely many terms ρi(A∗ yields a logarithmic duality for A, it is not difficult to see that S{ρi(A∗ i )). In particular, X can be seen as X = ∐{(A∗ i )♭ i ∈ I}. (14) We are going to prove that X can be equipped with a Boolean topology τ to obtain a topological structure that is isomorphic to A∗ and that is embeddable into ∐{A∗ i i ∈ I}. We define the topology τ on X as the topology generated by the sets [a : m] = [{[πi(f (a)) : m] i ∈ I}, a ∈ A, m ∈ M. The topology τ is clearly finer than the topology induced on X by ∐{A∗ i i ∈ I}. Let us show that hX, τ i is Boolean. It suffices to prove that it is compact. Assume that X = S{[aλ : mλ] λ ∈ L} for some aλ ∈ A and mλ ∈ M. For every i ∈ I, the family {[πi(f (aλ)) : mλ] λ ∈ L} is an open covering of ρi(A∗ i ) and there is a finite subset Li of L such that ρi(A∗ i ) = [{[πi(f (aλ)) : mλ] λ ∈ Li}. (15) 18 GEORGES HANSOUL AND BRUNO TEHEUX By hypothesis, for every i ∈ I there is an open formula formula φini with ni variables (where ni denotes Li) such that identity (15) is equivalent to Ai = φini ((πi(aλ))λ∈Li ). (16) Now, for every i ∈ I let Ωi be defined by Ωi = {j ∈ I Aj = φin((πj (f (aλ)))λ∈Li )}. The family {Ωi i ∈ I} is an open covering of I. By compactness, there is a finite subset J of I such that I = [{Ωj j ∈ J}. (17) By combining (16) and (17), we obtain, X = [{[{[aλ : mλ] λ ∈ Lj} j ∈ J}, which is a finite open covering of X extracted from {[aλ : mλ] λ ∈ L}. Let us denote by g the restriction of f ∗ to X. Hence, for any ρi(ψ) ∈ ρi(Ai), we have g(ρi(ψ)) = ψ ◦ πi ◦ f . We aim to prove that g is an X -isomorphism between hX, τ i and A∗. First we prove that g is a X ♭-embedding. We have to prove that if r represents and if an n-ary relation or the graph of a (partial) operation in the language of M ψ1, . . . , ψn ∈ X, we have the following equivalence f (ψ1, . . . , ψn) ∈ rX ⇔ (g(ψ1), . . . , g(ψn)) ∈ rA∗ . (18) Let J be a finite subset of I such that {ψ1, . . . , ψn} ⊆ S{ρj(A∗ j ) j ∈ J}. Let us denote by Y the latter set. We have already noted that Y , considered as a substructure of ∐{A∗ j j ∈ J}. Since f : A ֒→ i Qi∈I Ai is a Boolean representation of A, the map fJ : A → Qj∈J Aj : a 7→ (πj (a))j∈J is onto. Hence, the dual map f ∗ J : Y → A∗ is an embedding and is clearly equal to the restriction of g to Y . Then, it follows successively i ∈ I} is isomorphic to ∐{A∗ (ψ1, . . . , ψn) ∈ rX ⇔ (ψ1, . . . , ψn) ∈ rY J (ψ1), . . . , f ∗ ⇔ (f ∗ ⇔ (g(ψ1), . . . , g(ψn)) ∈ rA∗ J (ψn)) ∈ rA∗ , which establishes equivalence (18), as required. Finally, since g is the restriction on X of a continuous map, it is a continuous map for the induced topology on X. From the fact that τ is finer than the induced topology we eventually conclude that g : hX, τ i → A∗ is an X -embedding. We deduce that hX, τ i ∈ X . For the last part of the proof, we show that the evaluation map h : A → X (X, M ) : a 7→ h(a) : ρi(ψ) 7→ ψ(πi(f (a))) f is an isomorphism. It is clearly a homomorphism. Moreover, if a, b ∈ A and a 6= b then there is an i ∈ I such that πi(f (a)) 6= πi(f (b)), i.e., such that eAi(πi(f (a))) 6= eAi(πi(f (b))). Let ψ ∈ A∗ i with eAi(πi(f (a)))(ψ) 6= eAi (πi(f (b)))(ψ). It means that ψ(πi(f (a))) 6= ψ(πi(f (b))) which proves that h is one-to-one. Moreover, since h∗ = g and since g is an embedding, we deduce that h is onto and so, is an isomorphism. Hence, it follows successively that Aδ ≃ X ♭(A∗, M f ) ≃ X ♭(X, M f ) ≃ X ♭(∐i∈I (A∗ i )♭, M f ), EXT. MAPS TO PROFINITE COMPL. IN FINIT. GEN. QUASIVARIETIES 19 where we have used (14) to obtain the latter isomorphism. Then, we obtain X ♭(∐i∈I (A∗ ) ≃ Y i∈I i )♭, M f ) ≃ Y Aδ i i∈I X ♭(A∗ i , M f where the first isomorphism is obtained by partnership duality [10, Theorem 2.4]) and is also an Aι-isomorphism. (cid:3) Département de Mathématiques, Université de Liège, Grande Traverse, 12, 4000 Liège, Belgium E-mail address: [email protected] Mathematics Research Unit, FSTC, University of Luxembourg, 6, Rue Coudenhove- Kalergi, L-1359 Luxembourg, Luxembourg E-mail address: [email protected]
1304.4225	1	1304	2013-04-15T18:00:42	The group of automorphisms of the Lie algebra of derivations of a field of rational functions	[ "math.RA", "math.AG" ]	We prove that the group of automorphisms of the Lie algebra $\Der_K (Q_n)$ of derivations of the field of rational functions $Q_n=K(x_1,..., x_n)$ over a field of characteristic zero is canonically isomorphic to the group of automorphisms of the $K$-algebra $Q_n$.	math.RA	math	The group of automorphisms of the Lie algebra of derivations of a field of rational functions V. V. Bavula Abstract We prove that the group of automorphisms of the Lie algebra DerK (Qn) of derivations of the field of rational functions Qn = K(x1, . . . , xn) over a field of characteristic zero is canonically isomorphic to the group of automorphisms of the K-algebra Qn. Key Words: Group of automorphisms, monomorphism, Lie algebra, automorphism, locally nilpotent derivation, the field of rational functions in n variables. Mathematics subject classification 2010: 17B40, 17B20, 17B66, 17B65, 17B30. 1 Introduction In this paper, module means a left module, K is a field of characteristic zero and K ∗ is its group of units, and the following notation is fixed: • Pn := K[x1, . . . , xn] = Lα∈Nn Kxα is a polynomial algebra over K where xα := xα1 1 · · · xαn n and Qn := K(x1, . . . , xn) is the field of rational functions, • Gn := AutK−alg(Pn) and Qn := AutK−alg(Qn); are the partial derivatives (K-linear derivations) of Pn, i=1 Pn∂i ⊆ En := DerK(Qn) = Ln i=1 Qn∂i are the Lie algebras of • ∂1 := ∂ , . . . , ∂n := ∂ ∂x1 ∂xn • Dn := DerK(Pn) = Ln K-derivations of Pn and Qn respectively where [∂, δ] := ∂δ − δ∂, • Gn := AutLie(Dn) and En := AutLie(En), • δ1 := ad(∂1), . . . , δn := ad(∂n) are the inner derivations of the Lie algebras Dn and En where ad(a)(b) := [a, b], • Dn := Ln • Hn := Ln i=1 K∂i, i=1 KHi where H1 := x1∂1, . . . , Hn := xn∂n, • for each natural number n ≥ 2, un := K∂1 + P1∂2 + · · · + Pn−1∂n is the Lie algebra of triangular polynomial derivations (it is a Lie subalgebra of Dn) and AutLie(un) is its group of automorphisms. Theorem 1.1 [4] Gn = Gn. The aim of the paper is to prove the following theorem. Theorem 1.2 En = Qn. 1 Structure of the proof. (i) Qn ⊆ En via the group monomorphism (Lemma 2.3 and (3)) Qn → En, σ 7→ σ : ∂ 7→ σ(∂) := σ∂σ−1. n := σ(∂n) are commuting derivations of Qn such that (ii) Let σ ∈ En. Then ∂′ 1 := σ(∂1), . . . , ∂′ i=1 Qn∂′ i (Lemma 2.12.(2)) and En = Ln (iii) Tn (iv)(crux) There exist elements x′ i=1 kerQn (∂′ i) = K (Lemma 2.12.(1)). 1, . . . , x′ n ∈ Qn such that ∂′ i(x′ j ) = δij for i, j = 1, . . . , n (Lemma 2.12.(3)). (v) σ(xα∂i) = x′α∂′ i for all α ∈ Nn and i = 1, . . . , n (Lemma 2.12.(6)). (vi) The K-algebra homomorphism σ′ : Qn → Qn, xi 7→ x′ i, i = 1, . . . , n is an automorphism such that σ′(q∂i) = σ′(q)∂′ i for all q ∈ Qn and i = 1, . . . , n. (vii) FixEn(∂1, . . . , ∂n, H1, . . . , Hn) = {e} (Proposition 2.9.(1)). Hence, σ = σ′ ∈ Qn, by (v) and (vi), i.e. En = Qn. (cid:3) The groups of automorphisms of the Lie algebras Dn and un. n × En)) where Tn is an Theorem 1.3 (Theorem 5.3, [3]) AutLie(un) ≃ Tn ⋉ (UAutK(Pn)n ⋊ (F′ algebraic n-dimensional torus, UAutK(Pn)n is an explicit factor group of the group UAutK(Pn) of unitriangular polynomial automorphisms, F′ n and En are explicit groups that are isomorphic respectively to the groups I and Jn−2 where I := (1 + t2K[[t]], ·) ≃ K N and J := (tK[[t]], +) ≃ K N. Comparing the groups Gn, En and AutLie(un) we see that the group UAutK(Pn)n of polynomial automorphisms is a tiny part of the group AutLie(un) but in contrast Gn = Gn and En = Qn. Theorem 1.4 [1] Every monomorphism of the Lie algebra un is an automorphism. Not every epimorphism of the Lie algebra un is an automorphism. Moreover, there are count- ably many distinct ideals {Iiωn−1 i ≥ 0} such that I0 = {0} ⊂ Iωn−1 ⊂ I2ωn−1 ⊂ · · · ⊂ Iiωn−1 ⊂ · · · and the Lie algebras un/Iiωn−1 and un are isomorphic (Theorem 5.1.(1), [2]). Conjecture, [4]. Every homomorphism of the Lie algebra Dn is an automorphism. The groups of automorphisms of the Witt Wn (n ≥ 2) and the Virasoro Vir Lie algebras were found in [5]. 2 Proof of Theorem 1.2 This section can be seen as a proof of Theorem 1.2. The proof is split into several statements that reflect 'Structure of the proof of Theorem 1.2' given in the Introduction. Let G be a Lie algebra and H be its Lie subalgebra. The centralizer CG(H) := {x ∈ G [x, H] = 0} of H in G is a Lie subalgebra of G. In particular, Z(G) := CG(G) is the centre of the Lie algebra G. The normalizer NG(H) := {x ∈ G [x, H] ⊆ H} of H in G is a Lie subalgebra of G, it is the largest Lie subalgebra of G that contains H as an ideal. Let V be a vector space over K. A K-linear map δ : V → V is called a locally nilpotent map if V = Si≥1 ker(δi) or, equivalently, for every v ∈ V , δi(v) = 0 for all i ≫ 1. When δ is a locally nilpotent map in V we also say that δ acts locally nilpotently on V . Every nilpotent linear map 2 δ, that is δn = 0 for some n ≥ 1, is a locally nilpotent map but not vice versa, in general. Let G be a Lie algebra. Each element a ∈ G determines the derivation of the Lie algebra G by the rule ad(a) : G → G, b 7→ [a, b], which is called the inner derivation associated with a. The set Inn(G) of all the inner derivations of the Lie algebra G is a Lie subalgebra of the Lie algebra (EndK(G), [·, ·]) where [f, g] := f g − gf . There is the short exact sequence of Lie algebras 0 → Z(G) → G ad→ Inn(G) → 0, that is Inn(G) ≃ G/Z(G) where Z(G) is the centre of the Lie algebra G and ad([a, b]) = [ad(a), ad(b)] for all elements a, b ∈ G. An element a ∈ G is called a locally nilpotent element (respectively, a nilpotent element) if so is the inner derivation ad(a) of the Lie algebra G. The Lie algebra En. Since En = nM i=1 Qn∂i = nM i=1 QnHi (1) every element ∂ ∈ En is a unique sum ∂ = Pn field Qn is the union S06=f ∈Pn algebra Qn is a localization of Pn,f . Hence Dn,f := DerK(Pn,f ) = Ln i=1 ai∂i = Pn i=1 biHi where ai = xibi ∈ Qn. The Pn,f where Pn,f is the localization of Pn at the powers of f . The i=1 Pn,f ∂i ⊆ En and En = [ Dn,f . 06=f ∈Pn Qn is an En-module. The field Qn is a (left) En-module: En × Qn → Qn, (∂, q) 7→ ∂ ∗ q. In more detail, if ∂ = Pn i=1 ai∂i where ai ∈ Qn then nX ∂ ∗ q = ai ∂q ∂xi . i=1 The En-module Qn is not a simple module since K is an En-submodule of Qn, and n\ i=1 kerQn (∂i) = K. (2) Lemma 2.1 The En-module Qn/K is simple with EndEn (Qn/K) = Kid where id is the identity map. Proof. We have to show that for each non-scalar rational function, say pq−1 ∈ Qn, the En- submodule M of Qn/K it generates coincides with the En-module Qn/K. By (2), ai = ∂i ∗ (pq−1) 6= 0 for some i. Then for all elements u ∈ Qn, ua−1 i ∂i ∗ (pq−1 + K) = u + K. So, Qn/K is a simple En-module. Let f ∈ EndEn (Qn/K). Then applying f to the equalities ∂i ∗ (x1 + K) = δi1 for i = 1, . . . , n, we obtain the equalities ∂i ∗ f (x1 + K) = δi1 for i = 1, . . . , n. Hence, f (x1 + K) ∈ Tn So, f (x1 + K) = λ(x1 + K) and so f = λid, by the simplicity of the En-module Qn/K. (cid:3) i=2 kerQn/K(∂i) ∩ kerQn/K(∂2 i ) = (K(x1)/K) ∩ kerQn/K (∂2 i ) = K(x1 + K). The Cartan subalgebra Hn of En. A nilpotent Lie subalgebra C of a Lie algebra G is called a Cartan subalgebra of G if it coincides with its normalizer. We use often the following obvious observation: An abelian Lie subalgebra that coincides with its centralizer is a maximal abelian Lie subalgebra. Lemma 2.2 1. Hn is a Cartan subalgebra of En. 3 2. Hn = CEn(Hn) is a maximal abelian Lie subalgebra of En. Proof. 2. Clearly, Hn ⊆ CEn (Hn). Let ∂ = Pn i=1 aiHi ∈ CEn (Hn) where ai ∈ Qn. Then all i=1kerQn (∂i) = K, by (2), and so ∂ ∈ Hn. Therefore, Hn = CEn (Hn) is i=1kerQn (Hi) = ∩n ai ∈ ∩n a maximal abelian Lie subalgebra of En. 1. By statement 2, we have to show that Hn = N := NEn(Hn). Let ∂ = Pn i=1 aiHi ∈ N , we i=1 Hj(ai)Hi, and so Hj(ai) ∈ K for all i and j. This condition holds if all ai ∈ K, i.e. ∂ ∈ Hn. Suppose that have to show that all ai ∈ K. By statement 2, for all j = 1, . . . , n, Hn ∋ [Hj, ∂] = Pn ai 6∈ K for some i, we seek a contradiction. Then necessarily, ai 6∈ K(x1, . . . ,bxj, . . . , xn) for some j. Since Qn = K(x1, . . . ,bxj , . . . , xn)(xj ), the result follows from the following claim. K((x)) := {Pi>−∞ λixi λi ∈ K}. Since H(Pi>−∞ λixi) = Pi>−∞ iλixi, the Claim is obvious. Claim: If a ∈ K(x)\K then H(a) 6∈ K. The field K(x) is a subfield of the series field Then, by the Claim, Hj(ai) 6∈ K, a contradiction. (cid:3) Lemma 2.3 [5] Let R be a commutative ring such that there exists a derivation ∂ ∈ Der(R) such that r∂ 6= 0 for all nonzero elements r ∈ R (eg, R = Pn, Qn and δ = ∂1). Then the group homomorphism Aut(R) → AutLie(Der(R)), σ 7→ σ : δ 7→ σ(δ) := σδσ−1, is a monomorphism. The Qn-module En. The Lie algebra En is a Qn-module, Qn × En → En, (σ, ∂) 7→ σ(∂) := σ∂σ−1. By Lemma 2.3, the Qn-module En is faithful and the map Qn → En, σ 7→ σ : ∂ 7→ σ(∂) = σ∂σ−1, (3) is a group monomorphism. We identify the group Qn with its image in En, Qn ⊆ En. Every automorphism σ ∈ Qn is uniquely determined by the elements x′ 1 := σ(x1), . . . , x′ n := σ(xn). Let Mn(Qn) be the algebra of n × n matrices over Qn. The matrix J(σ) := (J(σ)ij ) ∈ Mn(Qn), where J(σ)ij = , is called the Jacobian matrix of the automorphism (endomorphism) σ and its determinant J (σ) := det J(σ) is called the Jacobian of σ. So, the j'th column of J(σ) is the gradient grad x′ )T of the polynomial x′ j. Then the derivations j := ( ∂x′ j ∂xi , . . . , ∂x′ j ∂x1 ∂x′ j ∂xn are the partial derivatives of Qn with respect to the variables x′ 1, . . . , x′ n, 1 := σ∂1σ−1, . . . , ∂′ ∂′ n := σ∂nσ−1 , . . . , ∂′ Every derivation ∂ ∈ En is a unique sum ∂ = Pn (∂1, . . . , ∂n)T and ∂′ := (∂′ 1, . . . , ∂′ ∂′ 1 = ∂ ∂x′ 1 n = ∂ ∂x′ n . (4) n)T where T stands for the transposition. Then i=1 ai∂i where ai = ∂ ∗ xi ∈ Qn. Let ∂ := ∂′ = J(σ)−1∂, i.e. ∂′ i = nX j=1 (J(σ)−1)ij∂j for i = 1, . . . , n. (5) In more detail, if ∂′ = A∂ where A = (aij) ∈ Mn(Qn), i.e. ∂i = Pn j=1 aij ∂j. Then for all i, j = 1, . . . , n, δij = ∂′ i ∗ x′ j = nX k=1 aik ∂x′ j ∂xk 4 where δij is the Kronecker delta function. The equalities above can be written in the matrix form as AJ(σ) = 1 where 1 is the identity matrix. Therefore, A = J(σ)−1. The maximal abelian Lie subalgebra Dn of En. Suppose that a group G acts on a set S. For a nonempty subset T of S, StG(T ) := {g ∈ G gT = T } is the stabilizer of the set T in G and FixG(T ) := {g ∈ G gt = t for all t ∈ T } is the fixator of the set T in G. Clearly, FixG(T ) is a normal subgroup of StG(T ). Lemma 2.4 1. CEn (Dn) = Dn and so Dn is a maximal abelian Lie subalgebra of En. 2. FixQn (Dn) = FixQn (∂1, . . . , ∂n) = Shn. 3. FixQn = (∂1, . . . , ∂n, H1, . . . , Hn) = {e}. 4. CenEn(Dn + Hn) = 0. Proof. 1. Statement 1 follows from (2): Clearly, Dn ⊆ CEn (Dn). Let ∂ = P ai∂i ∈ CEn (Dn) i=1 kerQn ∂i = K, by (2), and so ∂ ∈ Dn. So, CEn (Dn) = where ai ∈ Qn. Then all elements ai ∈ Tn Dn and as a result Dn is a maximal abelian Lie subalgebra of En. 2. Let σ ∈ FixQn (Dn) and J(σ) = (Jij ). By (5), ∂ = J(σ)∂, and so, for all i, j = 1, . . . , n, δij = ∂i ∗ xj = Jij , i.e. J(σ) = 1, or equivalently, by (2), x′ 1 = x1 + λ1, . . . , x′ n = xn + λn for some scalars λi ∈ K, and so σ ∈ Shn (since x′ j=1 kerQn (∂j) = K for i = 1, . . . , n). i − xi ∈ Tn 3. Let σ ∈ FixQn = (∂1, . . . , ∂n, H1, . . . , Hn). Then σ ∈ FixQn (∂1, . . . , ∂n) = Shn, by statement 2. So, σ(x1) = x1 + λ1, . . . , σ(xn) = xn + λn where λi ∈ K. Then xi∂i = σ(xi∂i) = (xi + λi)∂i for i = 1, . . . , n, and so λ1 = · · · = λn = 0. This means that σ = e. So, FixQn = (∂1, . . . , ∂n, H1, . . . , Hn) = {e}. 4. Statement 4 follows from statement 1 and Lemma 2.2. (cid:3) Lemma 2.5 Let A be a K-algebra, DerK(A) be the Lie algebra of K-derivations of A and D(A) be the ring of differential operators on A. If the algebra D(A) is simple and generated by A and DerK(A) then the D(A)-module A is simple. Proof. Let a be a nonzero D(A)-submodule of A. So, a is an ideal of A such that ∂(a) ⊆ a for all ∂ ∈ DerK(A). The algebra D := D(A) is generated by A and D. So, Da ⊆ aD and aD ⊆ Da, i.e. Da = aD is a nonzero ideal of the simple algebra D. Hence, 1 ∈ Da and so 1 = Pi aidi for some elements di ∈ D and ai ∈ a ⊆ D. Then 1 = 1 ∗ 1 = X i aidi ∗ 1 ∈ a, hence a = A, i.e. A is a simple D(A)-module. (cid:3) Theorem 2.6 1. En is a simple Lie algebra. 2. Z(En) = {0}. 3. [En, En] = En. Proof. 1. (i) n = 1, i.e. E1 = K(x)∂ is a simple Lie algebra: We split the proof into several steps. (a) D1 := K[x]∂ and W1 := K[x, x−1]∂ are simple Lie subalgebras of E1 (easy). (b) For all λ ∈ K, W1(λ) := K[x, (x − λ)−1] is a simple Lie subalgebra of E1, by applying the K-automorphism sλ : x 7→ x − λ of the K-algebra Q1 to W1, i.e. sλ(W1) = W1(λ). (c) For any nonempty subset I ⊂ K, W1(I) := W1(I)K := K[x, (x − λ)−1 λ ∈ I]∂ is a simple Lie subalgebra of E1: Let a be a nonzero ideal of W1(I) and 0 6= a∂ ∈ a. Then either a∂ ∈ D1 or 5 0 6= [p∂, a∂] ∈ D1 ∩ a for some p ∈ P1. Since D1 ⊆ W1(λ) for all λ ∈ I and W1(λ) are simple Lie algebra, a ∩ W1(λ) = W1(λ). Hence a = W1(I) since W1(I) = [ W1(λ), λ∈I i.e. W1(I) is a simple Lie algebra. (d) If K is an algebraically closed field then E1 is a simple Lie algebra since E1 = W1(K). The algebra E1 is the union S06=f ∈P1 W1[f −1] of the Lie algebras W1[f −1] := P1,f ∂ where P1,f is the localization of P1 at the powers of the element f . Let a be the ideal of E1 generated by a nonzero element a = pq−1∂ for some pq−1 ∈ Q1. Clearly, a ∈ W1[(f q)−1] for all nonzero elements f ∈ P1 and E1 = S06=f ∈P1 W1[(f g)−1]. So, to finish the proof of (i) it suffices to show that all the algebras W1[f −1] are simple. (e) A := W1[f −1] is a simple Lie algebra for all 0 6= f ∈ P1: Let K ′ := K(ν1, . . . , νs) be the subfield of the algebraic closure K of K generated by the roots ν1, . . . , νs of the polynomial f and G = Gal(K ′/K) be the Galois group of the finite Galois field extension K ′/K (since char(K) = 0). Let K ′ = ⊕d i=1Kθi for some elements θi ∈ K ′ and θ1 = 1. By (c), A′ := K ′[x, f −1]∂ = W1(ν1, . . . , νs)K ′ is a simple Lie K ′-algebra. Let a ∈ A\{0}, a and d a′ be the ideals in A and A′ respectively that are generated by the element a. Then a′ = A′, by (c). Notice that A′ = Pd i=1 θiA and for a′ = Pd i=1 θiθj[ai, bj]. Moreover, every element in A′ = a′ is a linear combination of several commutators in A′ (where c = Pd i=1 θkck ∈ A′ i=1 θibi ∈ A′ where ai, bi ∈ A, [a′, b] = Pd i=1 θiai, b = Pd [a, [a′, . . . [b, c] . . .] = X θi · · · θjθk[a, [ai, . . . [bj, ck] . . .]. (6) and ck ∈ A), The symmetrization map Sym : K ′ → K, λ 7→ G−1 Pg∈G g(λ), is a surjection such that Sym(µ) = µ for all µ ∈ K. Clearly, K ′(x)/K(x) is a Galois field extension with the Galois group G where the elements of G act trivially on the element x. So, the symmetrization map Sym can be extended to the surjection K ′(x) → K(x) by the same rule, and then to the surjection A′ → A, f ∂ 7→ Sym(f )∂. Each element e ∈ A ⊆ A′, can be expressed as a finite sum of elements in (6). Then applying Sym, we see that e is a linear combination of elements (commutators) from a, i.e. A is a simple Lie algebra. (ii) En is a simple Lie algebra for n ≥ 2: Let a ∈ En\{0} and a = (a) be the ideal in En generated by the element a = Pn i=1 ai∂i where ai ∈ Qn. (a) a ∩ Dn 6= 0: If a ∈ Dn then there is nothing to prove. Suppose that a 6∈ Dn. (a1) Suppose that ai ∈ K(xi) for all i. Then ai 6∈ K[xi] for some i (since a 6∈ Dn), and so a ∋ [Hi, a] = Hi(ai)∂i ∈ K(xi)∂i\{0}. By (i), ∂1 ∈ a ∩ Dn. (a2) Suppose that ai 6∈ K(xi) for some i. Then ∂j(ai) 6= 0 for some j 6= i. Let q ∈ Pn be the common denominator of the fractions a1, . . . , an, that is a1 = p1q−1, . . . , an = pnq−1 for some elements pi ∈ Pn. For all n ≥ 2, Dn ∩ a ∋ [qn∂j, a] = qn∂j(ai)∂i + X (. . .)∂k 6= 0. k6=i (b) a = Dn since Dn is a simple Lie algebra, [4]. (c) a ⊇ K(xi)∂i for i = 1, . . . , n: In view of symmetry it suffices to prove that a ⊇ K(x1)∂1. Notice that for all u ∈ Qn and i = 2, . . . , n, a ∋ [u∂1, x1∂i] = u∂i − x1∂i(u)∂1. 6 Therefore, a + Qn∂1 = En. The field of rational functions Qn = Qn(K) can be seen as the field of rational functions Qn(K) = Qn−1(K ′) where K ′ = K(x1). Then E′ n−1 := DerK ′ (Qn−1(K ′)) = nM i=2 Qn−1(K ′)∂i = nM i=2 Qn∂i. n−1-module Q′ n−1/K ′ = Qn/K(x1) is simple. The Lie algebra E′ By Lemma 2.5, the E′ Lie subalgebra of En, and En can be seen as a left E′ action. The ideal a of En is an E′ and a ∩ K(x1)∂1 is a nonzero ideal of it (by (b)). Therefore, K(x1)∂1 ⊆ a. The E′ En/a = (a + Qn∂1)/a ≃ Qn∂1/a ∩ Qn∂1 is an epimorphic image of the simple E′ Qn/K(x1) via n−1 is a n−1-module with respect to the adjoint n−1-submodule of En. The Lie algebra K(x1)∂1 is simple n−1-module n−1-module ϕ : Qn/K(x1) → Qn∂1/a ∩ Qn∂1, u + K(x1) 7→ u∂1 + a ∩ Qn∂1, with 0 6= (Pn +K(x1))/K(x1) ⊆ ker(ϕ). Therefore, Qn∂1 = a∩Qn∂1 ⊆ a, and so En = a+Qn∂1 = a. So, En is a simple Lie algebra. 2 and 3. Statements 2 and 3 follow from statement 1. (cid:3) Lemma 2.7 For all nonzero elements q ∈ Qn and i = 1, . . . , n, CEn (qPn∂i) = {0}. Proof. Let c ∈ CEn (qPn∂i). Then for all elements p ∈ Pn, 0 = [c, qp∂i] = c(p) · q∂i + p[c, q∂i] = c(p) · q∂i. Then c(p) = 0 for all p ∈ Pn, and so c = 0. (cid:3) Proposition 2.8 [4] FixGn (∂1, . . . , ∂n, H1, . . . , Hn) = {e}. Let d1, . . . , dn be a commuting linear maps acting in a vector space E. Let NilE(d1, . . . , dn) := i e = 0 for all i = 1, . . . , n and some j = j(e)}. Let NilEn (Dn) := NilEn (δ1, . . . , δn). {e ∈ E dj Clearly, NilEn (Dn) = Dn is a Lie subalgebra of En. Proposition 2.9 1. FixEn (∂1, . . . , ∂n, H1, . . . , Hn) = {e}. 2. FixEn(∂1, . . . , ∂n) = Shn. Proof. 1. Let σ ∈ F := FixEn(∂1, . . . , ∂n, H1, . . . , Hn). We have to show that σ = e. Then σ−1 ∈ F and σ±1(NilEn (Dn)) ⊆ NilEn(Dn), i.e. σ(Dn) = Dn since NilEn (Dn) = Dn. So, σDn ∈ FixGn (∂1, . . . , ∂n, H1, . . . , Hn) = {e} (Proposition 2.8), i.e. σ(∂) = ∂ for all ∂ ∈ Dn. Let 0 6= δ ∈ En. Then δ = q−1∂ for some 0 6= q ∈ Pn and ∂ ∈ Dn. Now, [q2p∂i, δ] = ∂′ ∈ Dn for all p ∈ Pn. Applying σ to the equality yields the equality [q2p∂i, σ(δ)] = ∂′. By taking the difference, we obtain σ(δ) − δ ∈ CEn(q2Pn∂i) = {0}, by Lemma 2.7, hence σ = e. 2. Clearly, Shn ⊆ F := FixEn (∂1, . . . , ∂n). Let σ ∈ F and H ′ n := σ(Hn). Applying the automorphism σ to the commutation relations [∂i, Hj] = δij∂i gives the relations [∂i, H ′ j − Hj] = 0 for all i and j. Therefore, i = Hi + di for some elements di ∈ CEn (Dn) = Dn (Lemma 2.4.(1)), and so di = Pn H ′ j=1 λij ∂j for some elements λij ∈ K. The elements H ′ j] = δij ∂i. By taking the difference, we see that [∂i, H ′ i := σ(Hi), . . . , H ′ 1, . . . , H ′ n commute, hence or equivalently, [Hj, di] = [Hi, dj] for all i, j, λij ∂j = λji∂i for all i, j. This means that λij = 0 for all i 6= j, i.e. H ′ i = Hi + λii∂i = (xi + λii)∂i = sλ(Hi) 7 where sλ ∈ Shn, sλ(xi) = xi + λii for all i. Then s−1 (statement 1), and so σ = sλ ∈ Shn. (cid:3) λ σ ∈ FixEn(∂1, . . . , ∂n, H1, . . . , Hn) = {e} The automorphism ν. Let ν be the K-automorphism of Qn given by the rule ν(xi) = x−1 i for i = 1, . . . , n. Then ν(∂i) = −xiHi, ν(Hi) = −Hi, ν(xiHi) = −∂i, i = 1, . . . , n. (7) By (7), the elements X1 := x1H1, . . . , Xn := xnHn commute and the next lemma follows from Lemma 2.4 and Proposition 2.9 since Xn := ν(Dn) = Ln i=1 KXi. Lemma 2.10 1. CEn (Xn) = Xn is a maximal abelian Lie subalgebra of En. 2. FixQn (X1, . . . , Xn) = FixEn(X1, . . . , Xn) = Shn. 3. FixQn (X1, . . . , Xn, H1, . . . , Hn) = FixEn (X1, . . . , Xn, H1, . . . , Hn) = {e}. The following lemma is well-known and it is easy to prove. Lemma 2.11 Let ∂ be a locally nilpotent derivation of a commutative K-algebra A such that ∂(x) = 1 for some element x ∈ A. Then A = A∂ [x] is a polynomial algebra over the ring A∂ := ker(∂) of constants of the derivation ∂ in the variable x. The next lemma is the core of the proof of Theorem 1.2. Lemma 2.12 Let σ ∈ En, ∂′ 1 := σ(∂1), . . . , ∂′ n := σ(∂n) and δ′ n are commuting derivations of Qn such that Tn i=1 Qn∂′ i. 1. ∂′ 1, . . . , ∂′ 2. En = Ln i for some elements x′ i∂′ 3. For each i = 1, . . . , n, σ(xi∂i) = x′ i(x′ are algebraically independent and ∂′ j ) = δij for i, j = 1, . . . , n. 1 := ad(∂′ 1), . . . , δ′ n := ad(∂′ n). Then i=1 kerQn (∂′ i) = K. i ∈ Qn. The elements x′ 1, . . . , x′ n 4. NilQn (∂′ 1, . . . , ∂′ 5. NilEn (δ′ 1, . . . , δ′ 6. σ(xα∂i) = x′α∂′ 1, . . . , x′ n := K[x′ n where P ′ n) = P ′ n) = Ln i for all α ∈ Nn and i = 1, . . . , n. i=1 P ′ n∂′ i. n]. 7. σ′ : Qn → Qn, xi 7→ x′ σ′(a∂i) = σ′(a)σ(∂i). i, i = 1, . . . , n is a K-algebra homomorphism (statement 3) such that 8. The K-algebra homomorphism σ′ is an automorphism. Proof. 1. The elements ∂1, . . . , ∂n are commuting derivations, hence so are ∂′ 1, . . . , ∂′ n. Let λ ∈ Tn i=1 kerQn (∂′ i). Then λ∂′ 1 ∈ CEn (∂′ 1, . . . , ∂′ n) = σ(CEn (∂1, . . . , ∂n)) = σ(CEn (Dn)) = σ(Dn) = σ( nM i=1 K∂i) = nM i=1 K∂′ i, 2. since CEn (Dn) = Dn, Lemma 2.4.(1). Then λ ∈ K since otherwise the infinite dimensional space Li≥0 Kλi∂′ 1 would be a subspace of the finite dimensional space σ(Dn). It suffices to show that the elements ∂′ n of the n-dimensional (left) vector space En over the field Qn are Qn-linearly independent (the key reason for that is statement 1). Let i. Suppose that m := dimQn (V ) < n, we seek a contradiction. Up to order, let i for some elements ai ∈ Qn. By applying j) = K, by statement 1. i=1 Qn∂′ m be a Qn-basis of V . Then ∂m+1 = Pm V = Pn ∂′ 1, . . . , ∂′ j (j = 1, . . . , n), we see that 0 = Pm δ′ 1, . . . , ∂′ This means that the elements ∂′ m are K-linearly dependent, a contradiction. i, and so ai ∈ Tn i=1 kerQn (∂′ 1, . . . , ∂′ i=1 ai∂′ j(a)∂′ i=1 ∂′ 8 3. Let H ′ i := σ(xi∂i) for i = 1, . . . , n. By statement 2, H ′ i=1 aij∂′ j for some elements aij ∈ Qn. Applying the automorphism σ to the relations δij∂j = [∂j, Hi] yields the relations i = Pn δij∂′ i = nX i=1 j (aik)∂′ ∂′ k. i := aii. Then ∂′ j(x′ i) = δji and ∂′ j(aik) = 0 for all k 6= i. By statement 1, aik ∈ K for all Let x′ i 6= k. Now, H ′ i := x′ i + X i∂′ aij∂′ j. j6=i 1, . . . , H ′ The elements H ′ aij = 0. Therefore, H ′ i(x′ The equalities ∂′ n commute, hence for all i 6= j, 0 = [H ′ i, H ′ j] = −aji∂′ i + aij∂′ j, and so i∂′ i. i = x′ j) = δij imply that the elements x′ n ∈ Qn are algebraically inde- pendent over K: Suppose that f (x′ n) = 0 for some nonzero polynomial f (t1, . . . , xn) ∈ K[t1, . . . , xn]. We can assume that the (total) degree deg(f ) is the least possible. Clearly, f 6∈ K, hence ∂f n)) = ∂xi ∂i(0) = 0, a contradiction. i=1 K∂′ 6= 0 for some i and deg( ∂f ∂xi n = Pn n). By statement 3 and Lemma 2.11, ) < deg(f ), but ∂f ∂xi i and N = NilQn (D′ n) = ∂i(f (x′ 4. Let D′ 1, . . . , x′ 1, . . . , x′ 1, . . . , x′ (x′ 1, . . . , x′ N = N D ′ n[x′ 1, . . . , x′ n] = K[x′ 1, . . . , x′ n] since K ⊆ N D ′ 5. Let ∂ = Pn n ⊆ QD ′ n n = K (by statement 1). i=1 ai∂′ i ∈ N := NilEn (δ′ n) where ai ∈ Qn (statement 2). For all α ∈ Nn, 1, . . . , δ′ nX ∂′α(ai)∂′ i δ′α(∂) = i=1 where δ′α := Qn i=1 δ′αi i , δ′ i = ad(∂′ i) and ∂′α := Qn i=1 ∂′αi i i = 1, . . . , n (statement 2). Now, statement 5 follows from statement 4. . So, δ′α(ai) = 0 iff ∂′α(ai) = 0 for 6. First, let us show that, by induction on α, that σ(xα∂i) − x′α∂′ i ∈ CenEn(D′ n) = D′ n (Lemma 2.4.(1)). The initial case when α = 0 is obvious. So, let α > 0. Then [∂′ j, σ(xα∂i) − x′α∂′ i] = σ([∂j , xα∂i]) − αjx′α−ej ∂′ i − αjx′α−ej ∂′ = αjx′α−ej ∂′ i = σ(αj xα−ej ∂i) − αjx′α−ej ∂′ i i = 0. Therefore, σ(xα∂i) = x′α∂′ j for some scalars λij = λij (α) ∈ K. Notice that i + P λij ∂′ σ(Hi) = σ(xi∂i) = x′ i∂′ i := H ′ i, by the definition of the elements x′ σ to the equalities (αj − δij)xα∂i = [Hj, xα∂i] we have (we may assume that xα∂i 6= Hi) nX k) = σ((αj − δij )xα∂i) = σ([Hj, xα∂i]) = [H ′ i. Since α > 0, αj 6= 0 for some j. Applying the automorphism (αj − δij )(x′α∂′ j, x′α∂′ λik∂′ k] nX λik∂′ i + i + k=1 k=1 = (αj − δij)x′α∂′ i − λij ∂′ j, and so (αj − δij + 1)λij = 0 and (αj − δij)λik = 0 for all k 6= j. This means that all λis = 0. 7. By statement 3, σ′ is a K-algebra homomorphism such that im(σ′) = Q′ n := K(x′ 1, . . . , x′ n). By statement 3, for all elements a ∈ Qn, since ∂′ i acts as ∂ ∂x′ i on Q′ n. ∂′ iσ′(a) = σ′∂i(a) 9 Let a = pq−1 6= 0 where p, q ∈ Pn. Then, for all r ∈ q2Pn, [a∂i, r∂i] = (a∂i(r) − ∂i(a)r)∂i ∈ Pn∂i. By applying σ, we have the equality [σ(a∂i), σ′(r)∂′ i] = σ′(a∂i(r) − ∂i(a)r)∂′ i. On the other hand, [σ′(a)∂′ i, σ′(r)∂′ i] = (σ′(a)∂′ iσ′(r) − ∂′ iσ′(a)σ′(r))∂′ i = (σ′(a)σ′∂i(r) − σ′∂i(a)σ′(r))∂′ i = σ′(a∂i(r) − ∂i(a)r)∂′ i. Hence, σ(a∂i) − σ′(a)∂′ i ∈ CEn (σ′(q2Pn)∂′ i) = CEn (σ(q2Pn∂i)) = σ(CEn (q2Pn∂i)) = σ(CEn (q2Pn∂i)) = 0, by Lemma 2.7. Therefore, σ(a∂i) = σ′(a)σ(∂i). 8. Since σ(Qn∂i) = σ′(Qn)∂′ i for all i = 1, . . . , n (statement 7), we must have σ′(Qn) = Qn, by statement 2, and so σ′ ∈ Qn. (cid:3) Proof of Theorem 1.2. Let σ ∈ En. By Corollary 2.12.(8), we have the automorphism σ′ ∈ Qn such that, by Lemma 2.12.(3,6), σ′−1σ ∈ FixEn (∂1, . . . , ∂n, H1, . . . , Hn) = {e} (Proposition 2.9). Therefore, σ = σ′ and so En = Qn. (cid:3) The work is partly supported by the Royal Society and EPSRC. Acknowledgements References [1] V. V. Bavula, Every monomorphism of the Lie algebra of triangular polynomial derivations is an auto- morphism, C. R. Acad. Sci. Paris, Ser. I, 350 (2012) no. 11-12, 553 -- 556. (Arxiv:math.AG:1205.0797). [2] V. V. Bavula, Lie algebras of triangular polynomial derivations and an isomorphism criterion for their Lie factor algebras, Izvestiya: Mathematics, (2013), in print. (Arxiv:math.RA:1204.4908). [3] V. V. Bavula, The groups of automorphisms of the Lie algebras of triangular polynomial derivations, Arxiv:math.AG/1204.4910. [4] V. V. Bavula, The group of automorphisms of the Lie algebra of derivations of a polynomial algebra. Arxiv:math.RA:1304.6524. [5] V. V. Bavula, The groups of automorphisms of the Witt Wn and Virasoro Lie algebras. Arxiv:math.RA:1304.6578. Department of Pure Mathematics University of Sheffield Hicks Building Sheffield S3 7RH UK email: [email protected] 10
1509.03458	1	1509	2015-09-11T11:09:31	Block representations of generalized inverses of matrices	[ "math.RA", "math.OA" ]	In this paper will be considered standard forms of generalized inverses for matrices in the shape of block representations {1, 2, 3, 4, 5, 5^k}-inverse. Especially will be considered Moore-Penrose inverse and the group inverse. Results from Rhode's technique are used and methods for calculating some inverse are shown on examples.	math.RA	math	The Symposium on Matematics and Applications, Faculty of Mathematics, University of Belgarde 2014, Vol. V(1) Block representations of generalized inverses of matrices School of Electrical Engineering, University of Belgrade, 73 Bulevar kralja Aleksandra Vera Miler Jerković e-mail: [email protected] Branko Malešević School of Electrical Engineering, University of Belgrade, 73 Bulevar kralja Aleksandra e-mail: [email protected] Abstract. In this paper will be considered standard forms of generalized inverses for matrices in the shape of block representations {1, 2, 3, 4, 5, 5k}-inverse. Especially will be considered Moore-Penrose inverse and the group inverse. Results from Rhode’s technique are used and methods for calculating some inverse are shown on examples. Index Terms. Generalized inverse, Moore-Penrose inverse, Group inverse, Drazin inverse 1. Introduction The concept of generalized inverse for matrices was considered by E.H. Moore (1920.) and R. Penrose (1955.) [1], [2] and [3]. The generalized form of {1, 2, 3, 4, 5, 5k}-inverse as well as some combination (the Moore-Penrose inverse, the group inverse, the Drazin inverse) play major role in solving problems in various areas of sciences, such as fuzzy mathematics, linear regression,... The combinations {1,3} and {1,4} of generalized inverses and also Moore-Penrose inverse have minimax properties and it can be applied in solving linear systems. Also, combination {1,2,3} of generalized inverses can be used for finding least squared solution of linear system. The group inverse has many applications in singular differential equations; Markov chains iterate methods and so on. The generalized inverses have application in linear statistical modeling, especially in solving singularity of covariance matrix. Application of {2}-inverse is using in Newton methods for solving systems of nonlinear equations. The Drazin inverse is using for solving singular linear difference equations. The overview of applications can be found in textbooks [2] and [3]. Presentations of various forms of generalized inverse are given according to method of C.A. Rhode [4]. This method is also presented in [5] and [6]. The Rhode’s method is applied in many fields. It can be used for finding solution of matrix equation (cid:1827)(cid:1850)(cid:1828)(cid:3404)(cid:1829) where solution is described in the terms of the Rhode’s general form of the equation (cid:1827)(cid:1850)(cid:1828)(cid:3404)(cid:1829) using Penrose’s general solution [12]. The aim of this paper is to describe {1}-inverse [6], [7], [8], [9], [10]; see also [11]. This method is also applicable for solving matrix generalized form of {1, 2, 3, 4, 5, 5k}-inverse using Rhode’s method. Consequently, we obtain forms of the Moore-Penrose inverse, the group inverse and the Drazin inverse. Using block representations of generalized inverse of matrix, we got detailed structure of generalized inverses. In this way, we can approach parts of generalized inverse and modify it to do better, which will our future plan. All relevant theorems and methods for pseudoinverses in this paper are presented by appropriate examples. Solutions of matrix equations (1), (2), (3), (4), (5) and (6) we defined as {1}, {2}, {3}, {4}, {5} and {5k} generalized inverse of the matrix A. Definition 1. The index k of matrix A is smallest non-negative number such that the equality (cid:1827)(cid:1850)(cid:1827)(cid:3404)(cid:1827) (1) (cid:1850)(cid:1827)(cid:1850)(cid:3404)(cid:1850) (2) (cid:4666)(cid:1827)(cid:1850)(cid:4667)∗(cid:3404)(cid:1827)(cid:1850) (3) (cid:4666)(cid:1850)(cid:1827)(cid:4667)∗(cid:3404)(cid:1850)(cid:1827) (4) (cid:1827)(cid:1850)(cid:3404)(cid:1850)(cid:1827) (5) (cid:1827)(cid:3038)(cid:1850)(cid:1827)(cid:3404)(cid:1827)(cid:3038) (6) For the matrix (cid:1827)∈(cid:2159)(cid:3040)(cid:3400)(cid:3041) system of four Penrose’s equations is consider: where matrix (cid:1850)∈(cid:2159)(cid:3041)(cid:3400)(cid:3040) is unknown. For square matrix, we add matrix equations: where (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)(cid:1863) is index of the matrix A. r(cid:1853)(cid:1866)(cid:1863)(cid:3435)(cid:1827)(cid:3038)(cid:3439)(cid:3404) (cid:1870)(cid:1853)(cid:1866)(cid:1863) (cid:4666)(cid:1827)(cid:3038)(cid:2878)(cid:2869)(cid:4667) is true. Let (cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041) be a set of all matrices over set of complex numbers of order (cid:1865)(cid:3400)(cid:1866) with a rank r. For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041) we can make expanded matrix (cid:3428)(cid:1827) (cid:1835)(cid:3040) 0(cid:3432) which can be transformed, by (cid:1835)(cid:3041) (cid:3428)(cid:1827) (cid:1835)(cid:3040) 0(cid:3432)~…~(cid:4674)(cid:1831)(cid:3045) (cid:1843)(cid:1842) 0(cid:4675) (7) (cid:1835)(cid:3041) where (cid:1831)(cid:3045)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041) is matrix with r ones on first r places of the main diagonal and zeros on the all other places. The matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) are regular and the following equality is true: (cid:1843)(cid:1827)(cid:1842)(cid:3404)(cid:1831)(cid:3045)(cid:3404)(cid:4674)(cid:1835)(cid:3045) 0 0(cid:4675) . (8) 0 Definition 2. The generalized inverses of the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041) , which satisfy some of matrix equations (1) (cid:3398) (4), and also (5)(cid:3398) (6) in the case of square matrix, can be defined as block (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1850)(cid:2868) (cid:1850)(cid:2869) (cid:1850)(cid:2870) (cid:1850)(cid:2871)(cid:3432)∙(cid:1843) (9) where (cid:1850)(cid:2868)∈(cid:2159)(cid:3045)(cid:3400)(cid:3045), (cid:1850)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667), (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045), (cid:1850)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) are appropriate submatrices. elementary transformations on the columns and rows, in equivalent matrix: 2. Block representations representations: In the next section are presented some essential theorems for block representations. Concretely, specifying theorems we describe block representations of {1, 2, 3, 4, 5, 5k} inverses. Likewise, using corollaries we describe combinations of these inverses, which are unique. The details of proofs of these theorems are shown in [2], [3], [4], [5] and [6]. Theorem 1. (Generalized {1} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8). The matrix X of the shape (9) satisfies matrix equation (1) if and only if: (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1835)(cid:3045) (cid:1850)(cid:2869) (cid:1850)(cid:2870) (cid:1850)(cid:2871)(cid:3432)∙(cid:1843) (10) where (cid:1850)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667), (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045), and (cid:1850)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) are arbitrary submatrices. In the case when m=r, then the matrix A is matrix of full rank by rows, hence submatrices (cid:1850)(cid:2869) and (cid:1850)(cid:2871) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1835)(cid:3045)(cid:1850)(cid:2870)(cid:3432)∙(cid:1843) (11) dissapear by dimension. Therefore, the matrix X from (10) has the shape satisfies matrix equations (1) and (2) if and only if: dimension. Therefore, the matrix X from (14) has the shape In the case when n=r, then the matrix A is matrix of full rank by columns, hence submatrices (cid:1850)(cid:2870) and (cid:1850)(cid:2871) dissapear by dimension. Therefore, the matrix X from (10) has the shape (cid:1850)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:1850)(cid:2869)(cid:4671)∙(cid:1843) (12) Theorem 2. (Generalized {2} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8). The matrix X of the shape (9) satisfies matrix equation (2) if and only if submatrices (cid:1850)(cid:2868)∈(cid:2159)(cid:3045)(cid:3400)(cid:3045) , (cid:1850)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) , (cid:1850)(cid:2870)∈ (cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045) (cid:1853)(cid:1866)(cid:1856) (cid:1850)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) satisfy matrix equations: (cid:1850)(cid:2868)(cid:2870)(cid:3404)(cid:1850)(cid:2868) ,(cid:1850)(cid:2868) (cid:1850)(cid:2869)(cid:3404)(cid:1850)(cid:2869) ,(cid:1850)(cid:2870) (cid:1850)(cid:2868)(cid:3404)(cid:1850)(cid:2870) ,(cid:1850)(cid:2870) (cid:1850)(cid:2869)(cid:3404)(cid:1850)(cid:2871) . (13) Corollary 1. (Generalized {1, 2} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8). The matrix X of the shape (9) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1835)(cid:3045) (cid:1850)(cid:2869) (cid:1850)(cid:2870) (cid:1850)(cid:2870)(cid:1850)(cid:2869)(cid:3432)∙(cid:1843) (14) where (cid:1850)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) and (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045) are arbitrary submatrices. In the case of m=r the matrix A is matrix of full rank by rows, hence submatrix (cid:1850)(cid:2869) dissapears by (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1835)(cid:3045)(cid:1850)(cid:2870)(cid:3432)∙(cid:1843) (15) In the case of n=r the matrix A is matrix of full rank by columns, hence submatrix (cid:1850)(cid:2870) dissapears by (cid:1850)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:1850)(cid:2869)(cid:4671)∙(cid:1843) (16) (cid:1843)∙(cid:1843)∗(cid:3404)(cid:3428)(cid:1845)(cid:2869) (cid:1845)(cid:2870) (cid:1845)(cid:2871) (cid:1845)(cid:2872)(cid:3432) and (cid:1842)∗∙(cid:1842)(cid:3404)(cid:3428)(cid:1846)(cid:2869) (cid:1846)(cid:2870) (cid:1846)(cid:2871) (cid:1846)(cid:2872)(cid:3432) (17) with appropriate submatrices (cid:1845)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:3045),(cid:1845)(cid:2870)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667), (cid:1845)(cid:2871)∈(cid:2159)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045), (cid:1845)(cid:2872)∈(cid:2159)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) and(cid:1846)(cid:2869)∈ (cid:2159)(cid:3045)(cid:3400)(cid:3045) , (cid:1846)(cid:2870)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667),(cid:1846)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045),(cid:1846)(cid:2872)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667). Matrices (cid:1843)∙(cid:1843)∗∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∗∙(cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) (cid:1845)(cid:2869)∗(cid:3404)(cid:1845)(cid:2869),(cid:1845)(cid:2870)∗(cid:3404)(cid:1845)(cid:2871), (cid:1845)(cid:2872)∗(cid:3404)(cid:1845)(cid:2872),(cid:1846)(cid:2869)∗(cid:3404)(cid:1846)(cid:2869),(cid:1846)(cid:2870)∗(cid:3404)(cid:1846)(cid:2871), (cid:1846)(cid:2872)∗(cid:3404)(cid:1846)(cid:2872) (18) and also square matrices (cid:1845)(cid:2872) and (cid:1846)(cid:2872) are invertible. Definition 3. The matrix matrix (cid:1827)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) is Hermitian if (cid:1827)(cid:3404)(cid:1827)∗. Theorem 3. (Generalized {3} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8) and let square matrix (cid:1843)∙(cid:1843)∗ be submatrices (cid:1850)(cid:2868)∈(cid:2159)(cid:3045)(cid:3400)(cid:3045), and (cid:1850)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) satisfy: (cid:4666)(cid:1845)(cid:2869)(cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:1845)(cid:2870)∗(cid:4667)(cid:1850)(cid:2868)∗(cid:3404)(cid:1850)(cid:2868)(cid:4666)(cid:1845)(cid:2869)(cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:1845)(cid:2870)∗(cid:4667) and (cid:1850)(cid:2869)(cid:3404)(cid:3398)(cid:1850)(cid:2868)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869) (19) where (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045) (cid:1853)(cid:1866)(cid:1856) (cid:1850)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) are arbitrary submatrices . in the shape (17). The matrix X of the shape (9) satisfies matrix equation (3) if and only if To detect {3} and {4} inverses of matrix A in the shape of block, it is necessary to make square dimension. Therefore, the matrix X from (14) has the shape are Hermitian. According to [4] hold: block matrices: dissapear by dimension. Therefore, the matrix X from (20) has the shape in the shape (17). Matrix X of the shape (9) satisfies matrix equations (1) and (3) if and only if: in the shape (17). The matrix X of the shape (9) satisfies matrix equations (1), (2) and (3) if and only if: Corollary 2. (Generalized {1, 3} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8) and let square matrix (cid:1843)∙(cid:1843)∗ be (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1835)(cid:3045) (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869) (cid:3432)∙(cid:1843) (20) (cid:1850)(cid:2871) (cid:1850)(cid:2870) where (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045) (cid:1853)(cid:1866)(cid:1856) (cid:1850)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) are arbitrary submatrices. In the case of m=r the matrix A is matrix of full rank by rows, hence submatrices (cid:1845)(cid:2870), (cid:1845)(cid:2872) and (cid:1850)(cid:2871) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1835)(cid:3045)(cid:1850)(cid:2870)(cid:3432)∙(cid:1843) (21) In the case of n=r, then the matrix A is matrix of full rank by columns, hence submatrices (cid:1850)(cid:2870) and (cid:1850)(cid:2871) dissapear by dimension. Therefore, the matrix X from (20) has the shape (cid:1850)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:4671)∙(cid:1843) (22) Corollary 3. (Generalized {1, 2, 3} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8) and let square matrix (cid:1843)∙(cid:1843)∗ be (cid:1850)(cid:3404)(cid:1842)∙(cid:4680)(cid:1835)(cid:3045) (cid:1850)(cid:2870) (cid:1850)(cid:2870)∙(cid:4666)(cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:4667)(cid:4681)∙(cid:1843) (23) where (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045) is arbitrary submatrix. In the case of m=r the matrix A is matrix of full rank by rows, hence submatrices (cid:1845)(cid:2870) and (cid:1845)(cid:2872) dissapear (cid:1850)(cid:3404)(cid:1842)∙(cid:3428)(cid:1835)(cid:3045)(cid:1850)(cid:2870)(cid:3432)∙(cid:1843) (24) In the case of n=r the matrix A is matrix of full rank by columns, hence submatrix (cid:1850)(cid:2870) dissapears by (cid:1850)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:4671)∙(cid:1843) (25) Theorem 4. (Generalized {4} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8) and let square matrix (cid:1842)∗∙(cid:1842) be (cid:1850)(cid:2868)∈(cid:2159)(cid:3045)(cid:3400)(cid:3045) and (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045) satisfy: (cid:1850)(cid:2868)∗(cid:4666)(cid:1846)(cid:2869)(cid:3398)(cid:1846)(cid:2870)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2870)∗(cid:4667)(cid:3404)(cid:4666)(cid:1846)(cid:2869)(cid:3398)(cid:1846)(cid:2870)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2870)∗(cid:4667)(cid:1850)(cid:2868) and (cid:1850)(cid:2870)(cid:3404)(cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:1850)(cid:2868) (26) where (cid:1850)(cid:2868)∈(cid:2159)(cid:3045)(cid:3400)(cid:3045) and (cid:1850)(cid:2870)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:3045) are arbitrary submatrices. Corollary 4. (Generalized {1, 4} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8) and let square matrix (cid:1842)∗∙(cid:1842) be (cid:1835)(cid:3045) (cid:1850)(cid:2869) (cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871) (cid:1850)(cid:2871)(cid:3432)∙(cid:1843) (27) where (cid:1850)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) (cid:1853)(cid:1866)(cid:1856) (cid:1850)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) are arbitrary submatrices. in the shape (17). Matrix X of the shape (9) satisfies matrix equation (4) if and only if submatrices in the shape (17). The matrix X of the shape (9) satisfies matrix equations (1) and (4) if and only if: by dimension. Therefore, the matrix X from (23) has the shape dimension. Therefore, the matrix X from (23) has the shape (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428) dimension. Therefore, the matrix X from (30) has the shape by dimension. Therefore, the matrix X from (27) has the shape dissapear by dimension. Therefore, the matrix X from (27) has the shape the shape (17). The matrix X of the shape (9) satisfies matrix equations (1), (2) and (4) if and only if: In the case of m=r the matrix A is matrix of full rank by rows, hence submatrices (cid:1850)(cid:2869) and (cid:1850)(cid:2871) dissapear (cid:1835)(cid:3045)(cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:3432)∙(cid:1843) (28) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428) In the case of n=r the matrix A is matrix of full rank by columns, hence submatrices (cid:1846)(cid:2871),(cid:1846)(cid:2872) and (cid:1850)(cid:2871) (cid:1850)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:1850)(cid:2869)(cid:4671)∙(cid:1843) (29) Corollary 5. (Generalized {1, 2, 4} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such as to satisfy (8) and let square matrix (cid:1842)∗∙(cid:1842) be in (cid:1835)(cid:3045) (cid:1850)(cid:2869) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428) (cid:4666)(cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:4667)∙(cid:1850)(cid:2869)(cid:3432)∙(cid:1843) (30) (cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871) where (cid:1850)(cid:2869)∈(cid:2159)(cid:3045)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) is arbitrary submatrix. In the case of m=r the matrix A is matrix of full rank by rows, hence submatrix (cid:1850)(cid:2869) dissapears by (cid:1835)(cid:3045)(cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:3432)∙(cid:1843) (31) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428) In the case of n=r the matrix A is matrix of full rank by columns, hence submatrices (cid:1846)(cid:2871) and (cid:1846)(cid:2872) (cid:1850)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:1850)(cid:2869)(cid:4671)∙(cid:1843) (32) Corollary 6. (Generalized {1, 3, 4} – inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8) and let square matrices (cid:1843)∙ (cid:1843)∗ (cid:1853)(cid:1866)(cid:1856) (cid:1842)∗∙(cid:1842) have the shape (17). The matrix X of the shape (9) satisfies matrix equations (1), (3) (cid:1835)(cid:3045) (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869) (cid:1850)(cid:3404)(cid:1842)∙(cid:4680) (cid:4681)∙(cid:1843) (33) (cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871) (cid:1850)(cid:2871) where (cid:1850)(cid:2871)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3040)(cid:2879)(cid:3045)(cid:4667) is arbitrary submatrix. In the case of m=r the matrix A is matrix of full rank by rows, hence submatrices (cid:1845)(cid:2870), (cid:1845)(cid:2872) and (cid:1850)(cid:2871) (cid:1835)(cid:3045)(cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:3432)∙(cid:1843) (34) (cid:1850)(cid:3404)(cid:1842)∙(cid:3428) In the case when n=r, then the matrix A is matrix of full rank by columns, hence submatrices (cid:1846)(cid:2871),(cid:1846)(cid:2872) and (cid:1850)(cid:2871) dissapear by dimension. Therefore, the matrix X from (33) has the shape (cid:1850)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:4671)∙(cid:1843) (35) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041) system of matrix equations (1), (2), (3) and (4) has unique solution that is denoted by (cid:1827)(cid:2993) and known as Moore-Penrose inverse of matrix. From above theorems we presente dissapear by dimension. Therefore, the matrix X from (30) has the shape dissapear by dimension. Therefore, the matrix X from (33) has the shape and (4) if and only if: the block representation of Moore-Penrose inverse: (cid:1827)(cid:2993)(cid:3404)(cid:1842)∙(cid:4680) presented. In next example, determination of the Moore-penrose inverse of matrix in a shape of block is Theorem 5. (Moore-Penrose inverse) For the matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3040)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), let there be given regular matrices (cid:1843)∈(cid:2159)(cid:3040)(cid:3400)(cid:3040) and (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8) and let square matrices (cid:1843)∙(cid:1843)∗ (cid:1853)(cid:1866)(cid:1856) (cid:1842)∗∙(cid:1842) have the shape (17). Unique solution of matrix equations (1), (2), (3) and (4) is given with (cid:1835)(cid:3045) (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869) (cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871) (cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:4681)∙(cid:1843) (36) In the case of m=r the matrix A is matrix of full rank by rows, hence submatrices (cid:1845)(cid:2870) and (cid:1845)(cid:2872) dissapear by dimension. Therefore, the matrix (cid:1827)(cid:2993) from (36) has the shape (cid:1835)(cid:3045)(cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:3432)∙(cid:1843) (37) (cid:1827)(cid:2993)(cid:3404)(cid:1842)∙(cid:3428) In the case of n=r the matrix A is matrix of full rank by columns, hence submatrices (cid:1846)(cid:2871) and (cid:1846)(cid:2872) dissapear by dimension. Therefore, the matrix (cid:1827)(cid:2993) from (36) has the shape (cid:1827)(cid:2993)(cid:3404)(cid:1842)∙(cid:4670)(cid:1835)(cid:3045) (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:4671)∙(cid:1843) (38) In the case that the matrix A is regular and square, m=n=r, the Moore-Penrose inverse (cid:1827)(cid:2993) of the matrix (cid:1827) is equal with (cid:1827)(cid:2879)(cid:2869). In that case, the matrix (cid:1827)(cid:2993) from (36) has the shape (cid:1827)(cid:2993)(cid:3404)(cid:1827)(cid:2879)(cid:2869)(cid:3404)(cid:1842)∙(cid:1843) (39) Example 1. For the matrix (cid:1827)(cid:3404)(cid:3429)1 2 3 7 8 9(cid:3433) calculate the Moore-Penrose inverse . 4 5 6 Solution. Using elementary transformations by columns and rows of expanded matrix (cid:3428)(cid:1827) (cid:1835)(cid:2871) 0(cid:3432) we can (cid:1835)(cid:2871) 0 1(cid:3433) and (cid:1843)(cid:3404)(cid:3430)‐(cid:2873)(cid:2871) 1 (cid:1842)(cid:3404)(cid:3429)1 0 (cid:2870)(cid:2871) ‐2 1(cid:3434) 0 1 ‐2 ‐(cid:2869)(cid:2871) 0 0 0 (cid:2872)(cid:2871) 1 such that (cid:1843)(cid:1827)(cid:1842)(cid:3404)(cid:1831)(cid:2870). By making product of matrices ‐(cid:2870)(cid:2870)(cid:2877) ‐3 0 6(cid:3434) and (cid:1842)∗∙(cid:1842)(cid:3404)(cid:3429)1 (cid:1843)∙(cid:1843)∗(cid:3404)(cid:3430)(cid:2870)(cid:2877)(cid:2877) 0 1 2 ‐(cid:2870)(cid:2870)(cid:2877) 1 ‐2 (cid:2869)(cid:2875)(cid:2877) ‐3 2 we obtain submatrices (cid:1845)(cid:2870)(cid:3404)(cid:4674)‐32(cid:4675) , (cid:1845)(cid:2872)(cid:3404)(cid:4670)6(cid:4671) , (cid:1846)(cid:2871)(cid:3404)(cid:4670)1 ‐2(cid:4671) , (cid:1846)(cid:2872)(cid:3404)(cid:4670)6(cid:4671) based on which we can determine submatrices (cid:3398)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:3404)(cid:4686)(cid:2869)(cid:2870)‐(cid:2869)(cid:2871)(cid:4687), (cid:3398)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:3404)(cid:4674)‐(cid:2869)(cid:2874) (cid:2869)(cid:2871)(cid:4675), (cid:4666)(cid:1846)(cid:2872)(cid:2879)(cid:2869)(cid:1846)(cid:2871)(cid:4667)(cid:4666)(cid:1845)(cid:2870)(cid:1845)(cid:2872)(cid:2879)(cid:2869)(cid:4667)(cid:3404)(cid:4674)‐(cid:2875)(cid:2871)(cid:2874)(cid:4675). Therefore, the (cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1741)‐2336 ‐16 1136 12 ‐736(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1744) ‐16 13 ‐736(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1744)∙(cid:1843)(cid:3404) (cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1741)1 0 ‐13 118 ‐118 0 1 0 16 1936 (cid:4574) Moore-Penrose inverse of matrix A, according to theorem 5, is given by 1 6(cid:3433) ‐2 determine regular matrices P and Q: (cid:1827)(cid:2993)(cid:3404)(cid:1842)∙ Using to minimal polynomial we can form q–polynomial: of matrix equations (1), (2) and (5) given by q–polynomial is Block representations of generalized inverses of squared matrices Relation between minimal polynomial (40) and q – polynomial (41) can be defined by: In this part of paper shall be considered block representations of generalized inverses of the Next theorem is shown and proved in [13]. According to block representation of the group inverse of square matrix can be obtained without determining q–polynomial. square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041). Let the matrix A be square matrix with index (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)(cid:1863), and let given minimal polynomial be (cid:2020)(cid:4666)(cid:1876)(cid:4667). Then: (cid:2020)(cid:4666)(cid:1876)(cid:4667)(cid:3404)(cid:1876)(cid:3040)(cid:3397)(cid:1855)(cid:3040)(cid:2879)(cid:2869)(cid:1876)(cid:3040)(cid:2879)(cid:2869)(cid:3397)⋯(cid:3397)(cid:1855)(cid:3038)(cid:1876)(cid:3038), (cid:1855)(cid:3038)(cid:3405)0. (40) (cid:1869)(cid:4666)(cid:1876)(cid:4667)(cid:3404)(cid:3398)(cid:2869)(cid:3030)(cid:3286)∙((cid:1876)(cid:3040)(cid:2879)(cid:3038)(cid:2879)(cid:2869)(cid:3397)(cid:1855)(cid:3040)(cid:2879)(cid:2869)(cid:1876)(cid:3040)(cid:2879)(cid:3038)(cid:2879)(cid:2870)(cid:3397)⋯(cid:3397)(cid:1855)(cid:3038)(cid:2878)(cid:2869)(cid:4667). (41) (cid:2020)(cid:4666)(cid:1876)(cid:4667)(cid:3404)(cid:1855)(cid:3038)(cid:1876)(cid:3038)(cid:4666)1(cid:3398)(cid:1876)(cid:1869)(cid:4666)(cid:1876)(cid:4667)(cid:4667) (42) Definition 4. Let is matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) with index (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3409)1. Then system of matrix equations (1), (2) and (5) has unique solution denoted by (cid:1827)# and known as the group inverse. If the matrix A has (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)0, then it is regular and hold (cid:1827)#(cid:3404)(cid:1827)(cid:2879)(cid:2869). Theorem 6. (Group inverse) For the square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) with index (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)1 and adequate q–polynomial (cid:1869)(cid:4666)(cid:1876)(cid:4667), the matrix q(A) represents one {1}-inverse of matrix A. Unique solution of system (cid:1827)#(cid:3404)(cid:1827)(cid:4666)(cid:1869)(cid:4666)(cid:1827)(cid:4667)(cid:4667)(cid:2870) . (43) Theorem 7. (Group inverse) For square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041), (cid:1870)(cid:3407)(cid:1865)(cid:1861)(cid:1866)(cid:4668)(cid:1865),(cid:1866)(cid:4669), with index (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404) 1 let there be given regular matrices (cid:1843),(cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) such that satisfy (8). Let block decomposition (cid:1843)∙(cid:1842)(cid:3404)(cid:3428)(cid:1848)(cid:2869) (cid:1848)(cid:2870) (cid:1848)(cid:2871) (cid:1848)(cid:2872)(cid:3432) (44) is true under assumptation that the submatrix (cid:1848)(cid:2872)∈(cid:2159)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667)(cid:3400)(cid:4666)(cid:3041)(cid:2879)(cid:3045)(cid:4667) is regular. Unique solution of system (cid:1835)(cid:3045) (cid:1827)#(cid:3404)(cid:1842)∙(cid:4680) (cid:3398)(cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:1848)(cid:2871) (cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:1848)(cid:2871)(cid:1848)(cid:2870)(cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:4681)∙(cid:1843) (45) Example 2. For the matrix (cid:1827)(cid:3404)(cid:3429)1 2 3 7 8 9(cid:3433) calculate the group inverse. 4 5 6 Solution. Minimal polynomial and q–polynomial of the matrix A are (cid:2020)(cid:4666)(cid:1876)(cid:4667)(cid:3404)(cid:1876)(cid:2871)(cid:3398)15(cid:1876)(cid:2870)(cid:3398)18(cid:1876) and (cid:1869)(cid:4666)(cid:1876)(cid:4667)(cid:3404) (cid:3051)(cid:2869)(cid:2876)(cid:3398)(cid:2873)(cid:2874) respecivly. The index of matrix A is (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)1. The group inverse of the matrix A, (cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1741)‐2336 ‐16 1136 ‐736(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1744) 118 ‐118 0 16 1936 The theorems 6 and 7 are illustrated by next example. (cid:1827)#(cid:3404)(cid:1827)(cid:4666)(cid:1869)(cid:4666)(cid:1827)(cid:4667)(cid:4667)(cid:2870)(cid:3404) of matrix equations (1), (2) and (5) is given by block representation (cid:3398)(cid:1848)(cid:2870)(cid:1848)(cid:2872)(cid:2879)(cid:2869) according to theorem 6, is 0 (cid:2870)(cid:2871) ‐2 1(cid:3434) ‐(cid:2869)(cid:2871) 0 Above result can be reached using theorem 7. By use of the elementary transformation on columns and rows of expanded matrix (cid:3428)(cid:1827) (cid:1835)(cid:2871) 0(cid:3432) we can determine regular matrices P and Q: (cid:1835)(cid:2871) 1(cid:3433) , (cid:1843)(cid:3404)(cid:3430)‐(cid:2873)(cid:2871) (cid:1842)(cid:3404)(cid:3429)1 0 1 0 1 ‐2 0 0 (cid:2872)(cid:2871) 1 such that (cid:1843)(cid:1827)(cid:1842)(cid:3404)(cid:1831)(cid:2870). From product of matrices (cid:1843)∙(cid:1842)(cid:3404)(cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1741)‐53 23 6(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1744) ‐3 ‐13 43 2 1 ‐2 we can determine submatrices: (cid:1848)(cid:2869)(cid:3404)(cid:4686)‐(cid:2873)(cid:2871) ‐(cid:2869)(cid:2871)(cid:4687), (cid:1848)(cid:2870)(cid:3404)(cid:4674)‐32(cid:4675), (cid:1848)(cid:2871)(cid:3404)(cid:4670)1 ‐2(cid:4671), (cid:1848)(cid:2872)(cid:3404)(cid:4670)6(cid:4671). Next, according (cid:2870)(cid:2871) (cid:2872)(cid:2871) to (cid:1848)(cid:2872)(cid:3404)6(cid:3405)0, we get ‐(cid:2875)(cid:2871)(cid:2874)(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1744)∙(cid:1843)(cid:3404)(cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1741)‐(cid:2870)(cid:2871)(cid:2871)(cid:2874) ‐(cid:2869)(cid:2874) 0 (cid:3398)(cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:1848)(cid:2871) (cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:1848)(cid:2871)(cid:1848)(cid:2870)(cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:4681)∙(cid:1843) (cid:3404)(cid:1842)∙(cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1741)1 ‐(cid:2875)(cid:2871)(cid:2874)(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1744) (cid:2869)(cid:2869)(cid:2871)(cid:2874) (cid:2869)(cid:2870) (cid:1835)(cid:3045) (cid:3398)(cid:1848)(cid:2870)(cid:1848)(cid:2872)(cid:2879)(cid:2869) ‐(cid:2869)(cid:2869)(cid:2876) ‐(cid:2869)(cid:2871) 0 0 1 (cid:1827)#(cid:3404)(cid:1842)∙(cid:4680) (cid:2869)(cid:2869)(cid:2876) ‐(cid:2869)(cid:2874) (cid:2869)(cid:2871) (cid:2869)(cid:2877)(cid:2871)(cid:2874) (cid:2869)(cid:2874) (cid:4574) Definition 5. The square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) with index (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)(cid:1863), system of matrix equations (2), (5) and (6) has unique solution denoted by (cid:1827)(cid:3005) and known as the Drazin inverse of matrix. Theorem 8. (Drazin inverse) For the square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) with index (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)(cid:1863) and adequate (cid:1827)(cid:3005)(cid:3404)(cid:1827)(cid:3038)(cid:4666)(cid:1869)(cid:4666)(cid:1827)(cid:4667)(cid:4667)(cid:3038)(cid:2878)(cid:2869) (46) If (cid:1863)(cid:3404) (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3409)1, then (cid:1827)(cid:3005) = (cid:1827)#. Theorem 9. [3] For the square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) holds (cid:1827)(cid:1827)(cid:3005)(cid:1827)(cid:3404)(cid:1827) if and only if (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3409)1. Definition 6. The square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) for which holds true that (cid:1827)(cid:2993)(cid:3404)(cid:1827)(cid:3005) is called EP–matrix. According to theorem 8, for EP–matrices holds (cid:1827)(cid:2993)(cid:3404)(cid:1827)(cid:3005)(cid:3404) (cid:1827)#. At the end of this section, next Theorem 10. [13] Square matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) is EP–matrix if and only if one of equivalent condition 1. (cid:1840)(cid:4666)(cid:1827)(cid:4667)(cid:3404)(cid:1840)(cid:4666)(cid:1827)∗(cid:4667). 2. (cid:1844)(cid:4666)(cid:1827)(cid:4667)(cid:3404)(cid:1844)(cid:4666)(cid:1827)∗(cid:4667). 3. (cid:2159)(cid:3041)(cid:3404)(cid:1840)(cid:4666)(cid:1827)(cid:4667)⊕(cid:1844)(cid:4666)(cid:1827)(cid:4667). Given solutions according to theorem 6 and theorem 7 are the same. theorem determines one characterisation of EP–matrices: q–polynomial, unique solution of system of matrix equations (2), (5) and (6) is given with satisfied: 4. There exist regular matrices (cid:1842)∈(cid:2159)(cid:3041)(cid:3400)(cid:3041) and (cid:1827)(cid:3045)∈(cid:2159)(cid:3045)(cid:3400)(cid:3045) such that: Corollary 6. Every square singular symmetrical matrix (cid:1827)∈(cid:2159)(cid:3045)(cid:3041)(cid:3400)(cid:3041) is EP–matrix. (cid:1827)(cid:3404)(cid:1842)⋅(cid:4674)(cid:1827)(cid:3045) 0 0 0(cid:4675)⋅(cid:1842)(cid:2879)(cid:2869) (47) Based on previous two theorems here we give one example of computing Moore-Penrose inverse of EP– matrix. Example 3. Calculate the Moore-Penrose inverse of matrix: Solution. Let us note that the matrix A is symmetrical and singular ((cid:1827)(cid:3404)0(cid:4667). It is meaning that matrix A is one EP–matrix with index (cid:1861)(cid:1866)(cid:1856)(cid:4666)(cid:1827)(cid:4667)(cid:3404)1. So, (cid:1827)(cid:2993)(cid:3404)(cid:1827)(cid:3005)(cid:3404) (cid:1827)#, and we can applied theorem 7. By applying elementary transformations on columns and rows of expanded matrix (cid:3428)(cid:1827) (cid:1835)(cid:2873) 0(cid:3432) we can (cid:1835)(cid:2873) 0 0 (cid:1827)(cid:3404)(cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1741)1 1 1 1(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1744) 1 1 1 2 0 ‐1 ‐1 1 0 2 1 1 0 1 ‐1 1 0 1 ‐1 ‐1 0 0 0 and (cid:1843)(cid:3404)(cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1741)2 (cid:1842)(cid:3404)(cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1741)1 0 ‐2 1 1 1(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1744) ‐1 0 0 1(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1744) 1 0 0 0 ‐1 0 1 1 ‐1 ‐1 1 1 0 0 0 0 1 0 0 ‐2 ‐1 0 1 0 1 0 0 0 1 0 0 0 0 0 1 such that (cid:1843)(cid:1827)(cid:1842)(cid:3404)(cid:1831)(cid:2870). From the product of matrices 3 ‐1 ‐5 3 (cid:1843)∙(cid:1842)(cid:3404)(cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1741)2 3(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1744) ‐1 ‐2 ‐2 1 3 1 6 ‐3 ‐3 ‐2 ‐1 ‐3 3 1 2 ‐1 ‐3 2 1 ‐2 ‐2(cid:4675) , (cid:1848)(cid:2871)(cid:3404)(cid:3429)‐2 1(cid:4675) , (cid:1848)(cid:2870)(cid:3404)(cid:4674)‐5 we can determine the submatrices (cid:1848)(cid:2869)(cid:3404)(cid:4674)2 ‐1 3 3 1 3 ‐1 1 (cid:1848)(cid:2872)(cid:3404)(cid:3429)6 ‐3 ‐3 3(cid:3433). Then, according to (cid:1848)(cid:2872)(cid:3404)12(cid:3405)0, we get 3 2 ‐3 2 ‐3 ‐(cid:2869)(cid:2872) ‐(cid:2869)(cid:2872) 0 (cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1741)1 (cid:2869)(cid:2869)(cid:2874)(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1744) (cid:2875)(cid:2869)(cid:2870) 1 0 ‐(cid:2869)(cid:2872) (cid:2869)(cid:2872) (cid:2869)(cid:2872) (cid:1835)(cid:3045) ‐(cid:2869)(cid:2869)(cid:2874) ‐(cid:2869)(cid:2869)(cid:2874) (cid:1827)(cid:2993)(cid:3404) (cid:1827)#(cid:3404)(cid:1842)∙(cid:4680) (cid:3398)(cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:1848)(cid:2871) (cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:1848)(cid:2871)(cid:1848)(cid:2870)(cid:1848)(cid:2872)(cid:2879)(cid:2869)(cid:4681)∙(cid:1843) (cid:3404)(cid:1842)∙ (cid:2869)(cid:2869)(cid:2870) (cid:2869)(cid:2871) (cid:2870)(cid:2873)(cid:2869)(cid:2872)(cid:2872) ‐(cid:2869)(cid:2869)(cid:2874) 0 (cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2872) (cid:2869)(cid:2869)(cid:2874) ‐(cid:2869)(cid:2869)(cid:2874) 0 (cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2872) 1 ‐1(cid:3433) and ‐1 (cid:3398)(cid:1848)(cid:2870)(cid:1848)(cid:2872)(cid:2879)(cid:2869) ∙(cid:1843), obtain regular matrices P and Q: and (cid:1827)(cid:2993)(cid:3404) (cid:1743)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1742)(cid:1741)(cid:2869)(cid:2877) (cid:2869)(cid:2877) (cid:2869)(cid:2877) 0 0 (cid:2869)(cid:2877) (cid:2870)(cid:2873)(cid:2869)(cid:2872)(cid:2872) (cid:2875)(cid:2869)(cid:2872)(cid:2872) (cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2877) (cid:2875)(cid:2869)(cid:2872)(cid:2872) (cid:2870)(cid:2873)(cid:2869)(cid:2872)(cid:2872) ‐(cid:2869)(cid:2869)(cid:2874) ‐(cid:2869)(cid:2869)(cid:2874) 0 (cid:2869)(cid:2869)(cid:2874) ‐(cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2869)(cid:2874) 0 (cid:2869)(cid:2869)(cid:2874) ‐(cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2869)(cid:2874) (cid:2869)(cid:2869)(cid:2874)(cid:1746)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1745)(cid:1744) (cid:4574) 3. Conclusion In this paper were presented theorems for obtaining Moore-Penrose inverse of matrices based on block representations. In addition, one part of this paper was dedicate to EP matrices. Clases {1,3} and {1,4} of generalized inverses and Moore-Penrose inverse have some minimax properties which can be applied on linear systems. Described procedure for detecting block representations of generalized inverses of matrices can be used in image reconstruction [14]. In the future, this will be subject of the further research. Acknowledgment. This work was supported in part by the Serbian Ministry of Education, Science and Technological Development under Grant No. 175016 and Grants No. 174032 and 44006. References [1] R. Penrose. A generalized inverses for matrices. Mathematical Proceedings of the Cambridge Philosophical Society 1955, 51, 406-413 [2] A. Ben-Israel ,T.N.E. Greville. Generalized Inverses, Theory and Applications. Springer, New York, 2003. [3] S.L. Cambell, C.D. Meyer. Inverses of Linear Transformations. Siam, Philadelphia, 2009. [4] C.A. Rhode. Contribution to the theory, computation and application of generalized inverses (PhD dissertation). University of North Carolina at Releigh, 1964. [5] V. Perić. Generalizirana reciproka matrice, Stručno-metodički časopis Matematika, Zagreb, 1982, 11 (1), 40-57. [6] B. Malešević. Grupna funkcionalna jednačina (magistarski rad). Matematički fakultet, Univerzitet u Beogradu, 1998. [7] I. Jovović, B. Malešević. A note on Solutions of the Matrix Equations AXB=C. Scientific publications of the state university of Novi Pazar, Ser.A: Appl. Math. Inform. And Mech. 2014, 6 (1), 45-55 [8] B. Malešević, B. Radičić. Reproductive and non-reproductive solutions of matrix equation AXB = C. Proceedings of the second symposium on Matematics and Applications Faculty of Mathematics, 2011, 157-163 [9] B. Malešević, B. Radičić. Non-reproductive and reproductive solutions of some matrix equations. Proceedings of the Conference on Mathematical and Informational Technologies MIT, 2011, 246-251 [10] B. Malešević, B. Radičić. Some considerations of matrix equations using the concept of reproductivity. Kragujevac Journal of Mathematics, 2012, 36 (1), 151-161 [11] B. Malešević, I. Jovović, M. Makragić, B. Radičić. A Note on Solutions of Linear Systems. ISRN Algebra, vol. 2013 Article ID 142124, 6 pages,2013.doi:10.1155/2013/142124 [12] B. Radičić, B. Malešević. Some considerations in Relation to the Matrix Equation AXB=C. The Mediterranean Journal of Mathematics 2014, 11, 841-856 [13] N. Matzakos, D. Pappas. EP matrices: Computation of the Moore-Penrose inverse via factorizations. Journal of Applied Mathematics and Computing, 2010, 34, 113-127 [14] B. Malešević, B. Banjac, M. Makragić, R. Obradović. Application of polynomial texture mapping in process of digitalization of cultural heritage, arXiv:1312.6935 [cs.GR]
1704.05171	1	1704	2017-04-18T02:09:17	A note on a separating system of rational invariants for finite dimensional generic algebras	[ "math.RA" ]	The paper deals with a construction of a separating system of rational invariants for finite dimensional generic algebras. In the process of dealing an approach to a rough classification of finite dimensional algebras is offered by attaching them some quadratic forms.	math.RA	math	In the name of Allah, the Beneficent, the Merciful. A note on a separating system of rational invariants for finite dimensional generic algebras U. Bekbaev Deparment of Science in Engineering, KOE, IIUM, Malaysia. [email protected] Abstract. The paper deals with a construction of a separating system of rational invariants for finite dimensional generic algebras. In the process of dealing an approach to a rough classification of finite dimensional algebras is offered by attaching them some quadratic forms. INTRODUCTION In [1] we have offered an approach to classification problem of finite dimensional algebras with respect to basis changes. It has been shown that if one has a special map with some properties then he is able to classify, to list canonical representations, all algebras who's set of structural constants , with respect to a fixed basis, do not nullify some polynomial. In this case he is also able to provide a separating system of rational invariants for those algebras. It was successfully applied in [2] to get a complete classification of all 2-dimensional algebras over algebraically closed fields. Unfortunately, so far we have no example of such a map in 3-dimensional case. Therefore in the current paper we deal with a weaker problem, namely with a construction of separating system of rational invariants for finite dimensional generic algebras. The theoretical existence of such system of invariants is known [3]. By generic algebras we mean the set of all algebras who's system of structural constants does not nullify a fixed nonzero polynomial in structural variables, over the basic field F. In process of dealing with the problem we show a way for a rough classification of finite dimensional algebras by attaching them some quadratic forms. The next section contains the main results. Further whenever A = (ai j) ∈ Mat(p × q, F), B ∈ Mat(p′ × q′ , F) we use A ⊗ B for the matrix MAIN RESULTS   a11B a12B ... a21B a22B ... . . . a1qB a2qB . a p1B a p2B ... a pqB   , where F − is a field of characteristic not 2. Let us consider any m-dimensional algebra A with multiplication · given by a bilinear map (u, v) 7→ u · v. If e = (e1 , e2 , ..., em) is a basis for A then one can represent the bilinear map by a matrix Ae = (Ai e jk)i, j ,k=1,2,...,m ∈ Mat(m × m2; F), where e j · ek = e1A1 e jk + e2A2 e jk + ... + emAm e jk, j, k = 1, 2, ..., m, such that for any u = eu, v = ev, where u = (u1 , u2 , ..., um), v = (v1 , v2 , ..., vm) are column vectors. So the algebra A (binary operation, bilinear map, tensor) is presented by the matrix Ae ∈ Mat(m × m2; F)-the matrix of structure constants (MSC) of A with respect to the basis e. u · v = eAe(u ⊗ v) If e′ = (e′ 1 , e′ 2 , ..., e′ m) is also a basis for A, g ∈ GL(m, F), e′g = e then it is well known that is valid. Further a basis e is fixed and therefore instead of Ae we use A, we do not make difference between A and its matrix A. Let X = (X i jk)i, j ,k=1,2,...,m stand for a variable matrix and Tr1(X), Tr2(X) stand for the row vectors Ae′ = gAe(g−1)⊗2 m ∑ ( i=1 X i i1 , m ∑ i=1 X i i2 , ..., m ∑ i=1 X i im), ( m ∑ i=1 X i 1i , m ∑ i=1 X i 2i , ..., m ∑ i=1 X i mi), respectively. We use τ for the representation of GL(m, F) on the n = m3 dimensional vector space V = Mat(m × m2; F) defined by τ : (g, A) 7→ B = gA(g−1 ⊗ g−1). For simplicity instead of "τ-equivalent", "τ-invariant" we use "equivalent" and "invariant". We represent each MSC A as a row vector with entries from Mat(m, F) by parting it consequently into elements of Mat(m, F): A = (A1 , A2 , ..., Am), A1 , A2 , ..., Am ∈ Mat(m, F). If C is a block matrix with blocks from Mat(m, F) we use notation C∗, where ∗ is the tensor product or transpose operation, to mean that the operation ∗ with C is done "over Mat(m, F)" (not over F), for example for the above presented matrix A At =   A1 A2 ... Am   − column vector over Mat(m, F), A⊗2 = (A2 1 , A1A2 , ..., A1Am , A2A1 , A2 2 , ..., A2Am , ..., AmA1 , AmA2 , ..., A2 m). One can see that the equality B = gA(g−1 ⊗ g−1) can be presented as B = (B1 , B2 , ..., Bm) = gA(g−1)⊗2 = (gA1g−1 , gA2g−1 , ..., gAmg−1)(g−1 ⊗ I), where I stands for m × m size identity matrix. Moreover for any matrices C and D the equality holds true. Therefore the following equalities hold true. (C ⊗ I)⊗(D ⊗ I) = (C ⊗ D) ⊗ I (B1 , B2 , ..., Bm)⊗k = (gA1g−1 , gA2g−1 , ..., gAmg−1)⊗k((g−1)⊗k ⊗ I), ⊗k   B1 B2 ... Bm   (B1 , B2 , ..., Bm)⊗k =   B2 1 B2B1 ... B1B2 B2 2 ... BmB1 BmB2   B2 1 B2B1 ... B1B2 B2 2 ... BmB1 BmB2 · · · B1Bm · · · B2Bm · · · · · · ... B2 m ⊗k   ⊗k ,   · · · B1Bm · · · B2Bm ... B2 m · · · · · · = (((gt )−1)⊗k ⊗ I)   gA2 1g−1 gA2A1g−1 ... gAmA1g−1 gA1A2g−1 gA2 2g−1 ... gAmA2g−1 · · · · · · · · · · · · gA1Amg−1 gA2Amg−1 ... gA2 mg−1 ⊗k   ((g−1)⊗k ⊗ I). Component-wise application of trace to this equality, which is denoted by Tr results in  Tr(  B2 1 B2B1 ... B1B2 B2 2 ... BmB1 BmB2 · · · B1Bm · · · B2Bm · · · · · · ... B2 m ⊗k   ) =  ((gt )−1)⊗k Tr(  gA2 1g−1 gA2A1g−1 ... gAmA1g−1 gA1A2g−1 gA2 2g−1 ... gAmA2g−1 · · · · · · · · · · · · gA1Amg−1 gA2Amg−1 ... gA2 mg−1 ⊗k   )(g−1)⊗k =  ((g−1)⊗k)t Tr(  A2 1 A2A1 ... A1A2 A2 2 ... AmA1 AmA2 · · · A1Am · · · A2Am · · · · · · ... A2 m ⊗k   )(g−1)⊗k , as far as for any matrices C, D and E, where D is a block matrix with blocks from Mat(m, F) and (C ⊗ I)D(E ⊗ I) has meaning, the equality Tr((C ⊗ I)D(E ⊗ I)) = C Tr(D)E is valid. One can represent the above obtained matrix equality in the following compact form Tr((Bt B)⊗k) = ((g−1)⊗k)t Tr((At A)⊗k)(g−1)⊗k . Note that Tr((At A)⊗k) is a symmetric matrix. The obtained equality allows formulation of the following theorem. Theorem 1. Invariants of the quadratic forms given by the matrix Tr((X t X)⊗k) are invariants of the m-dimensional algebras. This result can be used for a rough classification of finite dimensional algebras: Two m-dimensional algebras A, B are rough equivalent if the quadratic forms given by matrices Tr(At A) = Tr(Bt B) =     Tr(A2 1) Tr(A2A1) ... Tr(A1A2) Tr(A2 2) ... Tr(AmA1) Tr(AmA2) · · · Tr(A1Am) · · · Tr(A2Am) · · · · · · ... Tr(A2 m) Tr(B2 1) Tr(B2B1) ... Tr(B1B2) Tr(B2 2) ... Tr(BmB1) Tr(BmB2) · · · Tr(B1Bm) · · · Tr(B2Bm) · · · · · · ... Tr(B2 m)   ,   are equivalent. It is clear that entries of Tr(X t X) are polynomials in components of X and there exists nonsingular matrix Q(X t X) with rational entries in X such that the matrix t Tr(X X) = (Q(X t X)−1)t Tr(X t X)Q(X t X)−1 = D(X) is a diagonal matrix and Q(g) = I whenever g is a nonsingular diagonal matrix, where X = τ(Q(X t X), X). In algebraically closed field F case it means that one can define a nonempty invariant open subset V0 ⊂ V such that t Tr(A A) = D(A) and D(A) is nonsingular whenever A ∈ V0. Theorem 2. Two algebras A, B ∈ V0 are equivalent(isomorphic) if and only if B = τ(g0 , A) for some g0 ∈ GL(m, F) for which gt 0D(B)g0 = D(A). Proof. If B = τ(g, A) then B = τ(Q(Bt B), B) = τ(Q(Bt B), τ(g, A)) = τ(Q(Bt B)g, A) = τ(Q(Bt B)gQ(At A)−1 , τ(Q(At A), A) = τ(Q(Bt B)gQ(At A)−1 , A), and for g0 = Q(Bt B)gQ(At A)−1 one has 0D(B)g0 = (Q(Bt B)gQ(At A)−1)t D(B)Q(Bt B)gQ(At A)−1 = gt (Q(At A)−1)t (gt(Q(Bt B)t D(B)Q(Bt B))g)Q(At A)−1 = (Q(At A)−1)t (gt ( Tr(Bt B))g)Q(At A)−1 = (Q(At A)−1)t Tr(At A)Q(At A)−1 = D(A). Visa versa if B = τ(g0 , A) for some g0 for which gt 0D(B)g0 = D(A) then for g = Q(Bt B)−1g0Q(At A) one has τ(g, A) = τ(Q(Bt B)−1g0Q(At A), A) = τ(Q(Bt B)−1g0 , τ(Q(At A), A) = τ(Q(Bt B)−1g0 , A) = τ(Q(Bt B)−1 , τ(g0 , A)) = τ(Q(Bt B)−1 , B) = B. Assume that there exists matrix P(X), with rational entries with respect to components of X , such that P(A) is nonsingular for any A ∈ V0 and the equality P(τ(g, A)) = P(A)g−1 holds true whenever gt D(τ(g, A))g = D(A). (1) Theorem 3. For A,B ∈ V0 there exists g0 ∈ GL(m, F) such that gt 0D(B)g0 = D(A) and B = τ(g0 , A) if and only if τ(P(B), B) = τ(P(A), A), (P(B)−1)t D(B)P(B)−1 = (P(A)−1)t D(A)P(A)−1 . Proof. If B = τ(g0 , A) and gt 0D(B)g0 = D(A) then τ(P(B), B) = τ(P(τ(g0 , A)), τ(g0 , A)) = τ(P(A)g−1 0 , τ(g0 , A)) = τ(P(A), A) and (P(B)−1)t D(B)P(B)−1 = ((P(A)g−1 0 )−1)t D(B)(P(A)g−1 0 )−1 = ((P(A)−1)t gt 0D(B)g0P(A)−1 = (P(A)−1)t D(A)P(A)−1 . Visa versa, if equalities τ(P(B), B) = τ(P(A), A), (P(B)−1)t D(B)P(B)−1 = (P(A)−1)t D(A)P(A)−1 are valid then for g0 = P(B)−1P(A) one has gt 0D(B)g0 = D(A) and τ(g0 , A) = τ(P(B)−1P(A), A) = τ(P(B)−1 , τ(P(A), A)) = τ(P(B)−1 , τ(P(B), B)) = B. So Theorems 2 and 3 imply that the system of entries of matrices τ(P(X), X), (P(X)−1)t Tr(X t X)P(X)−1 is a separating system of rational invariants for algebras from V0. The above presented results show importance of construction of matrix P(X) with properties (1). Further we discuss a construction of such matrix by the use of rows r(A) for which the equality r(τ(g, A)) = r(A)g−1 is valid, whenever gt D(τ(g, A))g = D(A). To construct such rows one can use the following approach. Assume that the equalities B = gA(g−1)⊗2 , C = gCgt are true, where Ct = C and C is a nonsingular matrix. In this case C⊗2 = g⊗2C⊗2(g⊗2)t , B C⊗2 = gAC⊗2(g⊗2)t , C⊗2B t t = g⊗2C⊗2A gt and B C⊗2B t = gAC⊗2A t gt . On induction it is easy to see that for any natural k the equality ⊗20 ⊗21 B B ...B ⊗2k−1 C⊗2k (B ⊗20 ⊗21 B ⊗2k−1 ...B ⊗20 )t = gA ⊗21 A ⊗2k−1 ...A C⊗2k (A ⊗20 ⊗21 A ...A ⊗2k−1 )t gt holds true. Therefore due to the equalities ⊗20 ⊗21 B B ...B ⊗2k−1 C⊗2k (B ⊗20 ⊗21 B ...B ⊗2k−1 )t C−1 = ⊗20 gA ⊗21 A ...A ⊗2k−1 C⊗2k (A ⊗20 ⊗21 A ⊗2k−1 ...A )tC−1g−1 , ⊗20 ⊗21 B B ⊗2k−1 ...B C⊗2k (B ⊗20 ⊗21 B ⊗2k−1 ...B )t C−1B = ⊗20 gA ⊗21 A ...A ⊗2k−1 C⊗2k (A ⊗20 ⊗21 A ...A ⊗2k−1 )tC−1A(g−1)⊗2 one has Tri(B ⊗20 ⊗21 B ⊗2k−1 ...B C⊗2k (B ⊗20 ⊗21 B ...B ⊗2k−1 )t C−1B) = Tri(A ⊗20 ⊗21 A ...A ⊗2k−1 C⊗2k (A ⊗20 ⊗21 A ...A ⊗2k−1 )tC−1A)g−1 , i = 1, 2. The last equality shows that in our algebra case on can try to construct the needed matrix P(X) by the use of rows Tri(X ⊗20 X ⊗2 ...X ⊗2(k−1)(( Tr(X t X))−1)⊗2k (X ⊗20 X ⊗2 ...X ⊗2(k−1))t Tr(X t X)X), where i = 1, 2, k = 0, 1, 2, .... What is left unjustified here is that one should justify existence, in general, of a linear independent system consisting of m such rows. Remark. After a rough classification one can classify further each case of the rough classification with respect to the corresponding stabilizer. The author acknowledges MOHE for a support by grant FRGS14-153-0394. ACKNOWLEDGMENTS 1. U. Bekbaev, (IOP Conf. Series: Journal of Physics: Conf. Series, 819), 2017, pp. 2-9. 2. H. Ahmed, U. Bekbaev, I. Rakhimov,(arXiv 1702.08616), 2017, pp. 1-11. 3. V. Popov, (arXiv: 1411.6570v2[math.AG]), 2014, pp. 1-20. REFERENCES
1905.09613	1	1905	2019-05-23T12:19:00	Gerstenhaber brackets for skew group algebras in positive characteristic	[ "math.RA", "math.RT" ]	The deformation theory of an algebra is controlled by the Gerstenhaber bracket, a Lie bracket on Hochschild cohomology. We develop techniques for evaluating Gerstenhaber brackets of semidirect product algebras recording actions of finite groups over fields of positive characteristic. The Hochschild cohomology and Gerstenhaber bracket of these skew group algebras can be complicated when the characteristic of the underlying field divides the group order. We show how to investigate Gerstenhaber brackets using twisted product resolutions, which are often smaller and more convenient than the cumbersome bar resolution typically used. These resolutions provide a concrete description of the Gerstenhaber bracket suitable for exploring questions in deformation theory. We demonstrate with the prototypical example of a graded Hecke algebra (rational Cherednik algebra) in positive characteristic.	math.RA	math	GERSTENHABER BRACKETS FOR SKEW GROUP ALGEBRAS IN POSITIVE CHARACTERISTIC A.V. SHEPLER AND S. WITHERSPOON Abstract. The deformation theory of an algebra is controlled by the Gerstenhaber bracket, a Lie bracket on Hochschild cohomology. We develop techniques for evaluating Gerstenhaber brackets of semidirect product algebras recording actions of finite groups over fields of positive characteristic. The Hochschild cohomology and Gerstenhaber bracket of these skew group algebras can be complicated when the characteristic of the underlying field divides the group order. We show how to investigate Gerstenhaber brackets using twisted product resolutions, which are often smaller and more conve- nient than the cumbersome bar resolution typically used. These resolutions provide a concrete description of the Gerstenhaber bracket suitable for exploring questions in deformation theory. We demonstrate with the prototypical example of a graded Hecke algebra (rational Cherednik algebra) in positive characteristic. 9 1 0 2 y a M 3 2 ] . A R h t a m [ 1 v 3 1 6 9 0 . 5 0 9 1 : v i X r a 1. Introduction The Hochschild cohomology space of an associative algebra is a Gerstenhaber algebra under two binary operations, the cup product and the Gerstenhaber bracket. The Gerstenhaber bracket is a Lie bracket controlling the deformation theory of the algebra. Historically, it has been more difficult to compute than the cup product: The bracket is defined in terms of the cumbersome bar resolution and notoriously resists transfer to more convenient resolutions. In general, we lack user-friendly formulas giving the Gerstenhaber bracket explicitly. We consider the Hochschild cohomology of a skew group algebra (semidirect product algebra) arising from the action of a finite group G on an algebra S. We work in the modular setting, i.e., over a field k of positive characteristic that may divide the group order G. In this setting, the Hochschild cohomology of S ⋊ G is complicated by the potentially onerous cohomology of kG, in contrast to the characteristic zero case where it is always trivial. Computations of the Gerstenhaber bracket on S ⋊ G directly using the bar resolution often yield little useful information -- the bar resolution itself is too large and unwieldy. It can be a struggle even to describe adequately the Hochschild cohomology using the bar resolution. Thus one seeks a description of the Gerstenhaber bracket in terms of smaller Date: May 22, 2019. Key words and phrases. Hochschild cohomology, Gerstenhaber brackets, skew group algebras. The first author was partially supported by Simons grant 429539. The second author was partially supported by NSF grant DMS-1665286. Corresponding author: Anne Shepler. 1 2 A.V. SHEPLER AND S. WITHERSPOON resolutions used to compute Hochschild cohomology, a description that is concrete and straightforward to apply in specific examples. In this note, we consider the flexible twisted product resolution of a skew group algebra: one chooses a convenient resolution for S and another for G and then combines them to create a resolution of S ⋊ G. We show how to apply new techniques from [4] on Gersten- haber brackets to twisted product resolutions for skew group algebras from [8, 9]. This approach provides advantages over employing the often unmanageable but traditional bar resolution. We produce an explicit description of the Gerstenhaber bracket that should prove user-friendly and we illustrate with an example from deformation theory. This quintessential example using a small transvection group captures the difference between the modular and nonmodular settings, both in the theory of reflection groups and in the theory of graded Hecke algebras (and rational Cherednik algebras, see [3]). In Section 2, we recall the twisted product resolution from [8, 9] obtained by twisting a resolution of S with one for G. We recall methods of [4] analyzing Gerstenhaber brack- ets in Section 3 and show how they apply to twisted product resolutions for skew group algebras. We illustrate these techniques by showing how to compute some Gersten- haber brackets concretely for a small transvection group example from [8] in Section 5. Throughout, k is a field of arbitrary characteristic and ⊗ = ⊗k. 2. Twisted product resolutions We recall the twisted product resolution from [8, 9]. Consider a finite group G acting on a k-algebra S by automorphisms. Let A = S ⋊ G be the corresponding skew group algebra: As a vector space, S ⋊ G = S ⊗ kG, and we abbreviate the element s ⊗ g by sg (s ∈ S, g ∈ G) when no confusion can arise. Multiplication is defined by (sg) · (s′g′) = s(gs′) gg′ for all s, s′ ∈ S and g, g′ ∈ G . The action of g on s′ here is denoted by gs′. We use the enveloping algebra Se = S ⊗ Sop of any algebra S to express bimodule actions as left actions. The twisted product resolution. We consider projective resolutions (i) C : . . . → C2 → C1 → C0 → 0 of kG as a kG-bimodule, and (ii) D : . . . → D2 → D1 → D0 → 0 of S as an S-bimodule. We assume the resolution C is G-graded, with compatible group action: (2.1) g1(cid:0)(Ci)g2(cid:1)g3 = (Ci)g1g2g3 for all g1, g2, g3 ∈ G and all degrees i. We also assume D q carries a compatible action of G: Each Di is left kG-module with (2.2) g · (s · d) = gs · (g · d) for all g ∈ G, s ∈ S, d ∈ D and the differentials are kG-module homomorphisms. This ensures D q is compatible with the twisting map g ⊗ s 7→ gs ⊗ g given by the group action (see [9, Definition 2.17]). This is the setting, for example, when C q is the bar or reduced bar resolution of kG and when D q is the Koszul resolution of a Koszul algebra S (see [9, Prop 2.20(ii)]). GERSTENHABER BRACKETS 3 The twisted product resolution X = C ⊗G D of the algebra S ⋊ G is the total complex of the double complex C q ⊗ D q, X = C ⊗G D where Xn = M i+j=n Ci ⊗ Dj with each Xn suffused with the additional structure of a (S ⋊ G)-bimodule defined by s′g′ · (c ⊗ d) · sg = g′cg ⊗ ((g′hg)−1 The differential on X is ∂n = Pi+j=n(∂i ⊗ 1) + (−1)i(1 ⊗ ∂j) as usual. of A-bimodules (see [9] or [7, §4]): With this action, X is a resolution of A = S ⋊ G, i.e., X provides an exact sequence s′)(g−1 (ds)) for c ∈ Ch, d ∈ D, g, g′, h ∈ G, s, s′ ∈ S . . . . → X2 → X1 → X0 → A → 0 . When the A-bimodules Xn are all projective as Ae-modules, X is also a projective resolution of A. This occurs, for example, when D is a Koszul resolution of a Koszul algebra and C is the bar resolution of kG. (See [9, Proposition 2.20(ii)].) 3. Gerstenhaber brackets on differential graded coalgebras In this section, we summarize some results of [4] and develop additional techniques for computing Gerstenhaber brackets in the modular setting. Contrast with [5, 6], where the characteristic of the underlying field was 0. Resolutions as differential graded coalgebras. Consider a k-algebra A and a pro- jective resolution P of A as an A-bimodule: . . . → P2 → P1 → P0 → 0 . The resolution P is a differential graded coalgebra when P = ⊕iPi has a coalgebra structure compatible with its differential ∂P . This means there is a (degree 0) chain ∼ map ∆P : P → P ⊗A P lifting the canonical isomorphism A −→ A ⊗A A, called a diagonal map, that is required to be coassociative, i.e., (∆P ⊗ 1)∆P = (1 ⊗ ∆P )∆P as maps P → P ⊗A P ⊗A P, and counital, i.e., (µP ⊗ 1P )∆P = 1P = (1P ⊗ µP )∆P as maps P → P, where µP : P0 → A is augmentation of the complex (with µP zero on Pi for i ≥ 1). Throughout, we define µP ⊗1P : P ⊗P → P as the map p⊗p′ 7→ µP (p)·p′ (and similarly for 1P ⊗ µP ). Recall that the differential on Pn ⊗A Pm is just ∂P ⊗ 1P + (−1)n1P ⊗ ∂P . Homotopy from right to left. We may map the complex P ⊗A P to the complex P using either µP ⊗ 1P or 1P ⊗ µP . When P is a differential graded coalgebra, these mappings are chain homotopic by [4, Lemma 3.2.1]. (The hypotheses there are slightly stronger, but the same proof works under our hypotheses here.) Thus there exists a chain homotopy from µP ⊗ 1P to 1P ⊗ µP , i.e., a map φP : P ⊗A P → P with Pm ⊗A Pn → Pm+n+1 satisfying (3.1) ∂P φP + φP ∂P ⊗AP = µP ⊗ 1P − 1P ⊗ µP . 4 A.V. SHEPLER AND S. WITHERSPOON Example 3.2. The bar resolution B of the algebra A is a differential graded coalgebra. Indeed, for Bn = A ⊗ A⊗n ⊗ A, a diagonal map ∆B : B → B ⊗A B is defined by n (3.3) ∆B(a0 ⊗ · · · ⊗ an+1) = (a0 ⊗ · · · ⊗ aj ⊗ 1) ⊗A (1 ⊗ aj+1 ⊗ · · · ⊗ an+1) X j=0 for a0, . . . , an+1 in A. This map is coassociative and counital. One choice of homotopy φB : B ⊗A B → B from µB ⊗ 1B to 1B ⊗ µB is defined by (3.4) φB(cid:0)(a0 ⊗ · · · ⊗ ap−1 ⊗ ap) ⊗A (a′ p ⊗ ap+1 ⊗ · · · ⊗ an+1)(cid:1) = (−1)p−1 a0 ⊗ · · · ⊗ ap−1 ⊗ apa′ p ⊗ ap+1 ⊗ · · · ⊗ an+1 for all ai, a′ p ∈ A . Koszul resolutions of Koszul algebras are also differential graded coalgebras [1]. The Koszul resolution P of a Koszul algebra embeds into the bar resolution, however the above map φB does not preserve the image. Instead, a homotopy φP may be found directly in this case; see [4, §4], [5, §3.2], or [2, §4] for some examples. Definition of the Gerstenhaber bracket. The Gerstenhaber bracket for A is defined on cochains on the bar resolution B of A. Identify each space of cochains Hom Ae(Bn, A) with Hom k(A⊗n, A) via the canonical isomorphism. Then the Gerstenhaber bracket [ , ] : Hom k(A⊗n, A) × Hom k(A⊗m, A) → Hom k(A⊗(n+m−1), A) on cochains is defined by where, for ai in A, the circle product (f ◦ f ′)(a1 ⊗ · · · ⊗ an+m−1) is [f, f ′] = f ◦ f ′ − (−1)(n−1)(m−1)f ′ ◦ f n X i=1 (−1)(m−1)(i−1) f(cid:16)a1 ⊗ · · · ⊗ ai−1 ⊗ f ′(ai ⊗ · · · ⊗ ai+m−1) ⊗ ai+m ⊗ · · · ⊗ an+m−1(cid:17) . Gerstenhaber brackets on differential graded coalgebras. Although the Ger- stenhaber bracket is defined using the bar resolution, we seek descriptions in terms of more convenient resolutions used to compute Hochschild cohomology. Suppose P is a projective resolution of A with a differential graded coalgebra structure. The Gersten- haber bracket can be defined directly at the chain level on P using [4, Theorem 3.2.5]; we recall how a homotopy φP (see (3.1)) gives the bracket explicitly. Extend any cochain f ∈ Hom Ae(Pn, A) to all of P by defining f ≡ 0 on Pm with m 6= n. For f ∈ Hom Ae(Pn, A) and f ′ ∈ Hom Ae(Pm, A), define [f, f ′]P = f ◦P f ′ − (−1)(n−1)(m−1)f ′ ◦P f (3.5) where f ◦P f ′ (similarly f ′ ◦P f ) is the composition f ◦P f ′ : P ∆(2) P−−−−→ P ⊗A P ⊗A P 1P ⊗f ′⊗1P −−−−−→ P ⊗A P φP−→ P f −→ A. (3.6) Here, ∆(2) P = (1P ⊗ ∆P )∆P = (∆P ⊗ 1P )∆P and 1P ⊗ f ′ ⊗ 1P has signs attached so that (3.7) (1P ⊗ f ′ ⊗ 1P )(x ⊗ y ⊗ z) = (−1)imx ⊗ f ′(y) ⊗ z for x ∈ Pi, y, z ∈ P . Then [4, Theorem 3.2.5] implies that the Gerstenhaber bracket [ , ] of any elements in cohomology is given at the cochain level on P by the map [ , ]P on GERSTENHABER BRACKETS 5 cocycles. (Note that [4, Theorem 3.2.5] has slightly stronger hypotheses, but the proof indeed holds for any resolution P with the structure of a differential graded coalgebra.) 4. Twisted product resolution as a differential graded coalgebra We show in this section that a twisted product resolution X of S ⋊ G constructed from two differential graded coalgebras C and D is again a differential graded coalgebra. We then give the Gerstenhaber bracket for X in terms of the maps describing the Gerstenhaber brackets of C and D individually. Throughout this section, we fix • a differential graded coalgebra bimodule resolution (C, ∆C , µC) of G and • a differential graded coalgebra bimodule resolution (D, ∆D, µD) of S, producing • a twisted product resolution X = C ⊗G D of A = S ⋊ G. We assume that C is G-graded (as in (2.1)) with ∆C, µC preserving the grading and also that D carries a G-action (as in (2.2)) with ∆D, µD both kG-module homomorphisms. This is the case, for example, if C is the bar (or reduced bar) resolution of kG and D is the Koszul resolution of a Koszul algebra (see [9, Proposition 2.20(ii)]). Twisted comultiplication. In the next lemmas, we use diagonal maps for C and D to produce a diagonal map ∆X : X → X ⊗A X. Lemma 4.1. Define a twisting map τ : C ⊗ D → D ⊗ C by (4.2) τi,j(c ⊗ d) = (−1)ij (gd) ⊗ c for all c ∈ (Ci)g and d ∈ Dj . Then τ extends to a well-defined chain map 1 ⊗ τ ⊗ 1 : (C ⊗kG C) ⊗ (D ⊗S D) −→ (C ⊗G D) ⊗S⋊G (C ⊗G D) . Proof. Consider the map C ⊗ C ⊗ D ⊗ D 1⊗τ ⊗1 −−−−→ C ⊗ D ⊗ C ⊗ D −→ (C ⊗G D) ⊗S⋊G (C ⊗G D), where the latter map is the canonical surjection. Calculations show that the composition of these two maps is kG-middle linear in the first two arguments and S-middle linear in the last two arguments, and so it induces a well-defined map as claimed. A calculation shows that it is a chain map. (cid:3) Lemma 4.3. Let X = C ⊗G D be a twisted product resolution of S ⋊ G for differential graded coalgebras C and D resolving kG and S, respectively, as above. Then X is a differential graded coalgebra as well with comultiplication ∆X : X → X ⊗A X given by ∆X = (1 ⊗ τ ⊗ 1)(∆C ⊗ ∆D) . 6 A.V. SHEPLER AND S. WITHERSPOON Proof. We first check that ∆X is coassociative using the fact that ∆C and ∆D are each coassociative. We use the G-grading on C and the compatible G-action on D: (∆X ⊗ 1X )∆X = (cid:0)(1 ⊗ τ ⊗ 1)(∆C ⊗ ∆D) ⊗ 1 ⊗ 1(cid:1)(1 ⊗ τ ⊗ 1)(∆C ⊗ ∆D) = (1 ⊗ τ ⊗ 1 ⊗ 1 ⊗ 1)(∆C ⊗ ∆D ⊗ 1 ⊗ 1)(1 ⊗ τ ⊗ 1)(∆C ⊗ ∆D) = (1 ⊗ τ ⊗ 1 ⊗ 1 ⊗ 1)(1 ⊗ 1 ⊗ 1 ⊗ τ ⊗ 1)(1 ⊗ 1 ⊗ τ ⊗ 1 ⊗ 1)(∆C ⊗ 1 ⊗ ∆D ⊗ 1)(∆C ⊗ ∆D) = (1 ⊗ τ ⊗ 1 ⊗ 1 ⊗ 1)(1 ⊗ 1 ⊗ 1 ⊗ τ ⊗ 1)(1 ⊗ 1 ⊗ τ ⊗ 1 ⊗ 1)(1 ⊗ ∆C ⊗ 1 ⊗ ∆D)(∆C ⊗ ∆D) = (1 ⊗ 1 ⊗ 1 ⊗ τ ⊗ 1)(1 ⊗ τ ⊗ 1 ⊗ 1 ⊗ 1)(1 ⊗ 1 ⊗ τ ⊗ 1 ⊗ 1)(1 ⊗ ∆C ⊗ 1 ⊗ ∆D)(∆C ⊗ ∆D) = (1 ⊗ 1 ⊗ 1 ⊗ τ ⊗ 1)(1 ⊗ 1 ⊗ ∆C ⊗ ∆D)(1 ⊗ τ ⊗ 1)(∆C ⊗ ∆D) = (1X ⊗ ∆X )∆X . We next verify that ∆X is counital using the fact that ∆C and ∆D are each counital. We use the extra assumption that µC preserves the G-grading and µD is a kG-module homomorphism as well as the definition of the S ⋊ G-bimodule structure on C ⊗G D: (µX ⊗ 1X)∆X = (µC ⊗ µD ⊗ 1 ⊗ 1)(1 ⊗ τ ⊗ 1)(∆C ⊗ ∆D) = (µC ⊗ 1 ⊗ µD ⊗ 1)(∆C ⊗ ∆D) = ((µC ⊗ 1)∆C ) ⊗ ((µD ⊗ 1)∆D) = 1 ⊗ 1 = 1X , and, similarly, (1X ⊗ µX)∆X = 1X . We now need only check that ∆X is a chain map, i.e., ∆X ∂ = (∂ ⊗ 1+ 1⊗ ∂)∆X , for ∂ the differential on X. This follows from the fact that τ, ∆C, ∆D are all chain maps. (cid:3) Remark 4.4. One may check that the map 1 ⊗ τ ⊗ 1 of (4.2) interpolates between the maps of the form µ ⊗ 1 − 1 ⊗ µ for the various complexes, that is, (4.5) µX ⊗ 1X − 1X ⊗ µX = (µC ⊗ 1C ⊗ µD ⊗ 1D − 1C ⊗ µC ⊗ 1D ⊗ µD)(1C ⊗ τ −1 ⊗ 1D). We now give a theorem describing a homotopy from µX ⊗ 1X to 1X ⊗ µX concretely in terms of homotopies from µC ⊗ 1C to 1C ⊗ µC and from µD ⊗ 1D to 1D ⊗ µD by adapting [2, Lemmas 3.3, 3.4, and 3.5] to our setting. Theorem 4.6. Let X = C ⊗G D as above with homotopies φC from µC ⊗ 1C to 1C ⊗ µC and φD from µD ⊗ 1D to 1D ⊗ µD. Define φX : X ⊗A X → X by φX = (cid:0)φC ⊗ µD ⊗ 1D + ǫC(1C ⊗ µC) ⊗ φD(cid:1)(1C ⊗ τ −1 ⊗ 1D) for ǫC : C → C defined by c 7→ (−1)c for homogeneous c. Then φX is a homotopy from µX ⊗ 1X to 1X ⊗ µX. Proof. Let φ′ so that φX = φ′ X : C ⊗ C ⊗ D ⊗ D → C ⊗ D be the map φC ⊗ µD ⊗ 1 + ǫC(1 ⊗ µC) ⊗ φD X(1 ⊗ τ −1 ⊗ 1). Then on (C ⊗ C) ⊗ (D ⊗ D), ∂X φX (1 ⊗ τ ⊗ 1) = ∂X φ′ X = (∂C ⊗ 1 + ǫC ⊗ ∂D)(cid:0)φC ⊗ µD ⊗ 1 + ǫC(1 ⊗ µC) ⊗ φD(cid:1) = ∂CφC ⊗ µD ⊗ 1 − φC (ǫC ⊗ ǫC) ⊗ ∂D(µD ⊗ 1) (4.7) − ǫC∂C (1 ⊗ µC) ⊗ φD + 1 ⊗ µC ⊗ ∂DφD , GERSTENHABER BRACKETS 7 and, since 1 ⊗ τ ⊗ 1 is a chain map from (C ⊗ C) ⊗ (D ⊗ D) to X ⊗A X, φX ∂X⊗X (1 ⊗ τ ⊗ 1) = φX (1 ⊗ τ ⊗ 1) ∂(C⊗C)⊗(D⊗D) = φ′ X ∂(C⊗C)⊗(D⊗D) (4.8) X (cid:0)∂C⊗C ⊗ 1D⊗D + (ǫC ⊗ ǫC) ⊗ ∂D⊗D(cid:1) = φ′ = φC∂C⊗C ⊗ µD ⊗ 1 + φC(ǫC ⊗ ǫC) ⊗ (µD ⊗ 1)∂D⊗D + ǫC(1 ⊗ µC)∂C⊗C ⊗ φD + (1 ⊗ µC) ⊗ φD∂D⊗D. Here we used the fact that ǫCφC = −φC(ǫC ⊗ ǫC), ǫC(1 ⊗ µC)(ǫC ⊗ ǫC) = 1 ⊗ µC, and ∂C ǫC = −ǫC∂C. The second term of (4.7) cancels with the second term of (4.8) as µD ⊗1 is a chain map; likewise, the third terms cancel as µC ⊗ 1 is a chain map. Hence (∂X φX +φX ∂X⊗X )(1 ⊗ τ ⊗ 1) = (∂φC + φC ∂) ⊗ µD ⊗ 1 + 1 ⊗ µC ⊗ (∂φD + φD∂) = (µC ⊗ 1 − 1 ⊗ µC) ⊗ µD ⊗ 1 + 1 ⊗ µC ⊗ (µD ⊗ 1 − 1 ⊗ µD) = µC ⊗ 1 ⊗ µD ⊗ 1 − 1 ⊗ µC ⊗ 1 ⊗ µD, and, by equation (4.5), ∂φX + φX∂ = (µC ⊗ 1 ⊗ µD ⊗ 1 − 1 ⊗ µC ⊗ 1 ⊗ µD)(1 ⊗ τ −1 ⊗ 1) = µX ⊗ 1 − 1 ⊗ µX. (cid:3) Gerstenhaber bracket for skew group algebras. The next theorem gives the Ger- stenhaber bracket on a twisted product resolution X. Note that the twisting map τ in the theorem is from Lemma 4.1, the map 1X ⊗ f ′ ⊗ 1X has signs attached as in (3.7), and ǫC merely adjusts signs, c 7→ (−1)c for homogeneous c in C. Theorem 4.9. Let X = C ⊗G D be a twisted product resolution of S ⋊ G for differential graded coalgebras (C, ∆C , µC) and (D, ∆D, µD) resolving kG and S, respectively, as above. The Gerstenhaber bracket of elements of Hochschild cohomology represented by cocycles f ∈ Hom Ae(Xn, A) and f ′ ∈ Hom Ae(Xm, A) is represented by the cocycle (4.10) [f, f ′] = f ◦X f ′ − (−1)(n−1)(m−1)f ′ ◦X f , where f ◦X f ′ (similarly f ′ ◦X f ) is the composition (4.11) X (1X ⊗∆X )(∆X ) −−−−−−−−−−−−→ X ⊗A X ⊗A X 1X ⊗f ′⊗1X −−−−−−−−→ X ⊗A X φX−−−→ X f −−→ A with ∆X = (1C ⊗ τ ⊗ 1D)(∆C ⊗ ∆D) , and φX = (cid:0)φC ⊗ µD ⊗ 1D + (1 ⊗ µC)(ǫC ⊗ 1) ⊗ φD(cid:1)(1 ⊗ τ −1 ⊗ 1) . Proof. We combine Lemmas 4.1, Lemma 4.3, and Theorem 4.6 with (3.6) and (3.5). (cid:3) Example 4.12. In case S = S(V ) ∼= k[x1, . . . , xn], the symmetric algebra on a finite dimensional vector space V , we take D to be the Koszul resolution for which a choice of φD has been made in [4, §4] (see also [5, §3.2]). We may take C to be the bar or reduced bar resolution of kG for some applications, with homotopy φC as defined by equation (3.4). 8 A.V. SHEPLER AND S. WITHERSPOON 5. A small transvection group example We end by demonstrating how to use a twisted product resolution to compute Ger- stenhaber brackets explicitly via Theorem 4.9. We also see how computation of explicit brackets can shed light on questions in deformation theory (see [8]). We illustrate with the prototype example of a graded Hecke algebra (or rational Cherednik algebra) in positive characteristic (see [3] and [8]). In the nonmodular setting, these algebras have parameters supported only on the identity group element and on bireflections; in the modular setting, parameters can also be supported on reflections. All reflections in a finite linear group G acting in the modular setting are either diagonalizable or act as in this example. We include some explicit details to illustrate how to evaluate the maps in Theorem 4.9 concretely. We find both a nonzero and a zero Gerstenhaber bracket. 5.1. Group action and twisted product resolution. Say char(k) = p > 0 and consider the cyclic group G ≃ Z/pZ acting on V = k2 with basis v, w generated by g = ( 1 1 0 1 ) , so that gv = v and gw = v + w. We work in the twisted product resolution X = C ⊗G D of S(V ) ⋊ G obtained from twisting the reduced bar resolution C of kG with the Koszul resolution D of S(V ): Xn = M i+j=n Xi,j for Xi,j = kG ⊗ (kG)⊗i ⊗ kG ⊗ S(V ) ⊗ Vj V ⊗ S(V ). Here, Cn = kG ⊗ (kG)⊗n ⊗ kG with kG = kG/k1G and Dn = S(V ) ⊗ Vn V ⊗ S(V ). Then C and D satisfy the conditions specified in Section 3, and Theorem 4.9 applies. 5.2. Cochains. Consider cochains on the resolution X: κ ∈ Hom (S(V )⋊G)e(X0,2, S(V ) ⋊ G), λ ∈ Hom (S(V )⋊G)e(X1,1, S(V ) ⋊ G), and δ ∈ Hom (S(V )⋊G)e (X0,1, S(V ) ⋊ G) defined by (with subscripts on the tensor signs suppressed for brevity) λ(cid:0)(1G ⊗ gi ⊗ 1G) ⊗ (1S ⊗ v ⊗ 1S)(cid:1) = 0, λ(cid:0)(1G ⊗ gi ⊗ 1G) ⊗ (1S ⊗ w ⊗ 1S)(cid:1) = igi−1, κ(cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ v ∧ w ⊗ 1S)(cid:1) = g, δ(cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ v ⊗ 1S)(cid:1) = v, δ(cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ w ⊗ 1S)(cid:1) = 0 for 0 ≤ i ≤ p − 1, with all other values determined by these. One can check directly that λ and κ are 2-cocycles and that δ is a 1-cocycle for X. We will show that [δ, κ] 6= 0 and [λ, λ] = [λ, κ] = 0 . GERSTENHABER BRACKETS 9 The diagonal maps. We give some values of the diagonal maps at play in finding the Gerstenhaber brackets. The diagonal map ∆C on the reduced bar resolution of kG is deduced from (3.3). For example, after identifying gi with its image in kG, ∆C (1G ⊗ gi ⊗ 1G) = (1G ⊗ 1G) ⊗kG (1G ⊗ gi ⊗ 1G) + (1G ⊗ gi ⊗ 1G) ⊗kG (1G ⊗ 1G), and ∆C (1G ⊗ 1G) = (1G ⊗ 1G) ⊗kG (1G ⊗ 1G) . The diagonal map ∆D is found from embedding the Koszul into the bar resolution and then using (3.3). For example, we identify v ∧ w with v ⊗ w − w ⊗ v and observe that ∆D(1S ⊗ v ∧ w ⊗ 1S) = (1S ⊗ 1S) ⊗S (1S ⊗ v ∧ w ⊗ 1S) +(1S ⊗ v ⊗ 1S) ⊗S (1 ⊗ w ⊗ 1) − (1S ⊗ w ⊗ 1S) ⊗S (1S ⊗ v ⊗ 1S) +(1S ⊗ v ∧ w ⊗ 1S) ⊗S (1S ⊗ 1S) , and ∆D(1S ⊗ v ⊗ 1S) = (1S ⊗ 1S) ⊗S (1S ⊗ v ⊗ 1S) + (1S ⊗ v ⊗ 1S) ⊗S (1S ⊗ 1S) . Homotopies. Let φC : C ⊗kGC → C be the homotopy from µC ⊗1 to 1⊗µC from (3.4). We choose the homotopy φD : D ⊗S D → D from µD ⊗ 1 to 1 ⊗ µD given in [4, Definition 4.1.3] and record a few values here for later use: φD((1 ⊗ w ⊗ 1) ⊗S (v ⊗ 1)) = 1 ⊗ v ∧ w ⊗ 1, φD((1 ⊗ v) ⊗S (1 ⊗ w ⊗ 1)) = 0, φD((1 ⊗ 1) ⊗S (1 ⊗ v ⊗ 1)) = 0, φD((1 ⊗ v ⊗ 1) ⊗S (1 ⊗ 1)) = 0 . Nonzero bracket. We use Theorem 4.9 to show explicitly that [δ, κ] = κ. First note that [δ, κ] is zero on all components of X except possibly X0,2. We consider the com- position (4.11) with f ′ = δ and f = κ to find κ ◦X δ. As a first step, we apply the map (∆X ⊗ 1X )∆X to the element (1G ⊗ 1G) ⊗ (1S ⊗ v ∧ w ⊗ 1S) of X0,2, where, recall ∆X = (1C ⊗ τ ⊗ 1D)(∆C ⊗ ∆D) . Direct calculation confirms that (1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) (∆X ⊗1)∆X 7−−−−−−−−−→ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) −(1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) −(1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) −(1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) as an element of X ⊗A X ⊗A X. We have suppressed all subscripts for brevity; for example, the second summand may be written (cid:0)(1G ⊗kG 1G)⊗(1S ⊗S 1S)(cid:1)⊗A(cid:0)(1G ⊗kG 1G)⊗(1S ⊗S v⊗S 1S)(cid:1)⊗A(cid:0)(1G ⊗kG 1G)⊗(1S ⊗S w⊗S 1S)(cid:1) . We next apply the map 1X ⊗ δ ⊗ 1X ; it is nonzero on exactly two summands, the second and the penultimate, and we obtain (with the tensor products over A indicated here) (cid:0)(1G⊗1G)⊗(1S⊗v)(cid:1)⊗A(cid:0)(1G⊗1G)⊗(1S⊗w⊗1S)(cid:1)−(cid:0)(1G⊗1G)⊗(1S⊗w⊗v)(cid:1)⊗A(cid:0)(1G⊗1G)⊗(1S⊗1S)(cid:1). 10 A.V. SHEPLER AND S. WITHERSPOON To apply φX next, we first rearrange terms with 1G ⊗ τ −1 ⊗ 1S, producing (1G ⊗ 1G) ⊗ (1G ⊗ 1G) ⊗ (1S ⊗ v) ⊗ (1S ⊗ w ⊗ 1S) − (1G ⊗ 1G) ⊗ (1G ⊗ 1G) ⊗ (1S ⊗ w ⊗ v) ⊗ (1S ⊗ 1S), and then apply the map φC ⊗ µD ⊗ 1D + 1C ⊗ µC ⊗ φD to obtain (1G ⊗ 1G ⊗ 1G) ⊗ (v ⊗ w ⊗ 1S) − (1G ⊗ 1G) ⊗ (1S ⊗ v ∧ w ⊗ 1S). Lastly, we apply κ as the last step of (4.11) and obtain 0 from the first term and −g from the second. Thus (κ ◦ δ)(cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ v ∧ w ⊗ 1S)(cid:1) = −g = κ(cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ v ∧ w ⊗ 1S)(cid:1) and κ ◦X δ = −κ. We inspect the above calculation with an eye toward switching the order of κ and δ and deduce that δ ◦X κ = 0. We conclude, as claimed, [δ, κ] = δ ◦X κ − κ ◦X δ = κ . Zero brackets. We now use Theorem 4.9 to show that [λ, f ] = 0 when f is λ or κ. We evaluate composition (4.11) on X1,2 with f ′ = λ. Other calculations are similar. We first apply ∆X = (1G ⊗ τ ⊗ 1S)(∆C ⊗ ∆D) to sample input in X1,2, noting that gi w = iv + w (with subscripts suppressed again): (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) ∆C ⊗∆D 7−−−−−−→ (1 ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) −(1 ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) ⊗ (1 ⊗ 1) +(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) +(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) −(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) +(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) ⊗ (1 ⊗ 1) 1⊗τ ⊗1 7−−−−→ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) −(1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ (iv + w) ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) +(1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) ⊗ (1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ 1) +(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) +(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) −(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ v ⊗ 1) +(1 ⊗ gi ⊗ 1) ⊗ (1 ⊗ v ∧ w ⊗ 1) ⊗ (1 ⊗ 1) ⊗ (1 ⊗ 1) , an element of X ⊗A X. Next we apply ∆X ⊗ 1X : Evaluating ∆C ⊗ ∆D ⊗ 1X on the last expression yields 27 summands; the map (1G ⊗ τ ⊗ 1S) ⊗ 1X transforms these to 27 summands in X ⊗A X ⊗A X. A quick check verifies that 1X ⊗ λ ⊗ 1X vanishes on all but two summands, namely − (cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ 1S)(cid:1) ⊗A (cid:0)(1G ⊗ gi ⊗ 1G) ⊗ (1S ⊗ w ⊗ 1S)(cid:1) ⊗A (cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ v ⊗ 1S)(cid:1), − (cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ v ⊗ 1S)(cid:1) ⊗A (cid:0)1G ⊗ gi ⊗ 1G) ⊗ (1S ⊗ w ⊗ 1S)(cid:1) ⊗A (cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ 1S)(cid:1), GERSTENHABER BRACKETS 11 and we obtain −(cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ 1S)(cid:1) ⊗A (cid:0)(igi−1 ⊗ 1G) ⊗ (1S ⊗ v ⊗ 1S)(cid:1) −(cid:0)(1G ⊗ 1G) ⊗ (1S ⊗ v ⊗ 1S)(cid:1) ⊗A (cid:0)(igi−1 ⊗ 1G) ⊗ (1S ⊗ 1S)(cid:1). Applying φX followed by f = λ or f = κ gives 0 as w does not appear in the input. Remark 5.1. The cocycles λ and κ above were not chosen randomly. These cocycles define a PBW deformation of S ⋊ G, and the zero brackets calculated above predict the PBW property. Indeed, in [8], we considered PBW deformations of S ⋊ G given by analogs of Lusztig's graded Hecke algebras and symplectic reflection algebras over fields of positive characteristic. These algebras Hλ,κ depend on two parameters λ and κ with λ : kG ⊗ V → kG and κ : V ⊗ V → kG. The Hochschild 2-cocycles above of the same name λ and κ are these parameters converted into cocycles on the resolution X; see [8, Example 2.2] and also [10, Section 5]. A necessary condition for the parameters λ and κ to define a PBW deformation is that [λ, λ] = 0 and [λ, κ] = 0 when the cochains κ and λ they define are cocycles. (More generally, we require that λ is a cocycle, [λ, λ] = 0, and [λ, λ] = 2∂∗κ.) Thus knowing explicit values for brackets is helpful for finding new deformations. The cocycle δ above is included merely for illustration purposes; it provides an example of a nonzero Gerstenhaber bracket. References [1] R.-O. Buchweitz, E. L. Green, N. Snashall, and Ø. Solberg, "Multiplicative structures for Koszul algebras," Q. J. Math. 59 (2008), 441 -- 454. [2] L. Grimley, V. C. Nguyen, and S. Witherspoon, "Gerstenhaber brackets on Hochschild cohomology of twisted tensor products," J. Noncommutative Geometry 11 (2017), no. 4, 1351 -- 1379. [3] E. Norton, "Symplectic reflection algebras in positive charactersitic as Ore extensions," arXiv:1302.5411. [4] C. Negron and S. Witherspoon, "An alternate approach to the Lie bracket on Hochschild coho- mology," Homology, Homotopy and Applications 18 (1) (2016), 265 -- 285. [5] C. Negron and S. Witherspoon, "The Gerstenhaber bracket as a Schouten bracket for polynomial rings extended by finite groups," Proc. London Math. Soc. (3) 115 (2017), no. 6, 1149 -- 1169. [6] A.V. Shepler and S. Witherspoon, "Group actions on algebras and the graded Lie structure of Hochschild cohomology," J. Algebra 351 (2012), 350 -- 381. [7] A.V. Shepler and S. Witherspoon, "A Poincar´e-Birkhoff-Witt Theorem for quadratic algebras with group actions," Trans. Amer. Math. Soc. 366 (2014), no. 12, 6483 -- 6506. [8] A.V. Shepler and S. Witherspoon, "PBW deformations of skew group algebras in positive charac- teristic," Algebras and Representation Theory 18 (2015), no. 1, 257 -- 280. [9] A.V. Shepler and S. Witherspoon, "Resolutions for twisted tensor products," Pacific J. Math. 298 (2019), no. 2, 445 -- 469. [10] A.V. Shepler and S. Witherspoon, "Group twisted Alexander-Witney and Eilenberg-Zilber maps," submitted, http://arxiv.org/abs/1905.06892. Department of Mathematics, University of North Texas, Denton, Texas 76203, USA E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, Texas 77843, USA E-mail address: [email protected]
1005.3192	1	1005	2010-05-18T13:14:47	Associative Geometries. I: Torsors, linear relations and Grassmannians	[ "math.RA" ]	We define and investigate a geometric object, called an associative geometry, corresponding to an associative algebra (and, more generally, to an associative pair). Associative geometries combine aspects of Lie groups and of generalized projective geometries, where the former correspond to the Lie product of an associative algebra and the latter to its Jordan product. A further development of the theory encompassing involutive associative algebras will be given in subsequent work.	math.RA	math	ASSOCIATIVE GEOMETRIES. I: TORSORS, LINEAR RELATIONS AND GRASSMANNIANS WOLFGANG BERTRAM AND MICHAEL KINYON Abstract. We define and investigate a geometric object, called an associative geometry, corresponding to an associative algebra (and, more generally, to an associative pair). Associative geometries combine aspects of Lie groups and of generalized projective geometries, where the former correspond to the Lie product of an associative algebra and the latter to its Jordan product. A further develop- ment of the theory encompassing involutive associative algebras will be given in subsequent work [BeKi09]. 0 1 0 2 y a M 8 1 ] . A R h t a m [ 1 v 2 9 1 3 . 5 0 0 1 : v i X r a Introduction What is the geometric object corresponding to an associative algebra? The ques- tion may come as a bit of a surprise: the philosophy of Noncommutative Geometry teaches us that, as soon as an algebra becomes noncommutative, we should stop looking for associated point-spaces, such as manifolds or varieties. Nevertheless, we raise this question, but aim at something different than Noncommutative Geome- try: we do not try to generalize the relation between, say, commutative associative algebras and algebraic varieties, but rather look for an analog of the one between Lie algebras and Lie groups. Namely, every associative algebra A gives rise to a Lie algebra A− with commutator bracket [x, y] = xy−yx, and thus can be seen as a "Lie algebra with some additional structure". Since the geometric object corresponding to a Lie algebra should be a Lie group (the unit group A×, in this case), the object corresponding to the associative algebra, called an "associative geometry", should be some kind of "Lie group with additional structure". To get an idea of what this additional structure might be, consider the decomposition xy = xy + yx 2 + xy − yx 2 =: x • y + 1 2 [x, y] of the associative product into its symmetric and skew-symmetric parts. The sym- metric part is a Jordan algebra, and the additional structure will be related to the geometric object corresponding to the Jordan part. As shown in [Be02], the "geo- metric Jordan object" is a generalized projective geometry. Therefore, we expect an associative geometry to be some sort of mixture of projective geometry and Lie groups. Another hint is given by the notion of homotopy in associative algebras: 2000 Mathematics Subject Classification. 20N10, 17C37, 16W10. Key words and phrases. associative algebras and pairs, torsor (heap, groud, principal homo- geneous space), semitorsor, linear relations, homotopy and isotopy, Grassmannian, generalized projective geometry. 1 2 WOLFGANG BERTRAM AND MICHAEL KINYON an associative product xy really gives rise to a family of associative products x ·a y := xay for any fixed element a, called the a-homotopes. Therefore we should rather expect to deal with a whole family of Lie groups, instead of looking just at one group corresponding to the choice a = 1. 0.1. Grassmannian torsors. The following example gives a good idea of the kind of geometries we have in mind. Let W be a vector space or module over a com- mutative field or ring K, and for a subspace E ⊂ W , let C E denote the set of all subspaces of W complementary to E. It is known that C E is, in a natural way, an affine space over K. We prove that a similar statement is true for arbitrary inter- sections C E ∩ C F (Theorem 1.2): they are either empty, or they carry a natural "affine" group structure. By this we mean that, after fixing an arbitrary element Y ∈ C E ∩ C F , there is a natural (in general noncommutative) group structure on C E ∩ C F with unit element Y . The construction of the group law is very simple: for X, Z ∈ C E ∩ C F , we let X · Z := (P E F )(Y ), where, for any complementary pair (U, V ), P U V is the projector onto V with kernel U. Since X · Z indeed depends on X, E, Y, F, Z, we write it also in quintary form X − P Z (0.1) Γ(X, E, Y, F, Z) := (P E X − P Z F )(Y ). The reader is invited to prove the group axioms by direct calculations. The proofs are elementary, however, the associativity of the product, for example, is not obvious at a first glance. Some special cases, however, are relatively clear. If E = F , and if we then identify a subspace U with the projection P E U , then it is straightforward to show that the expression Γ(X, E, Y, E, Z) in C E is equivalent to the expression P E Y + P E Z in the space of projectors with kernel E, and we recover the classical affine space structure on C E (see Theorem 1.2). On the other hand, if E and F happen to be mutually complementary, then any common complement of E and F may be identified with the graph of a bijective linear map E → F , and hence C E ∩ C F is identified with the set Iso(E, F ) of linear isomorphisms between E and F . Fixing an origin Y in this set fixes an identification of E and F , and thus identifies C E ∩C F with the general linear group GlK(E). X − P E Summing up, the collection of groups C E ∩ C F , where (E, F ) runs through Gras(W ) × Gras(W ), the direct product of the Grassmannian of W with itself, can be seen as some kind of interpolation, or deformation between general linear groups and vector groups, encoded in Γ. The quintary map Γ has remarkable properties that will lead us to the axiomatic definition of associative geometries. 0.2. Torsors and semitorsors. To eliminate the dependence of the group struc- tures C E ∩ C F on the choice of unit element Y , we now recall the "affine" or "base point free" version of the concept of group. There are several equivalent versions, going under different names such as heap, groud, flock, herd, principal homogeneous space, abstract coset, pregroup or others. We use what seems to be the most cur- rently fashionable term, namely torsor. The idea is quite simple (see Appendix A ASSOCIATIVE GEOMETRIES. I 3 for details): if, for a given group G with unit element e, we want to "forget the unit element", we consider G with the ternary product G × G × G → G; (x, y, z) 7→ (xyz) := xy−1z. As is easily checked, this map has the following properties: for all x, y, z, u, v ∈ G, (G1) (G2) (xy(zuv)) = ((xyz)uv) , (xxy) = y = (yxx) . Conversely, given a set G with a ternary composition having these properties, for any element x ∈ G we get a group law on G with unit x by letting a ·x b := (axb) (the inverse of a is then (xax)) and such that (abc) = ab−1c in this group. (This observation is stated explicitly by Certaine in [Cer43], based on earlier work Prufer, Baer, and others.) Thus the affine concept of the group G is a set G with a ternary map satisfying (G1) and (G2); this is precisely the structure we call a torsor. One advantage of the torsor concept, compared to other, equivalent notions men- tioned above, is that it admits two natural and important extensions. On the one hand, a direct check shows that in any torsor the relation (G3) (xy(zuv)) = (x(uzy)v) = ((xyz)uv) , called the para-associative law, holds (note the reversal of arguments in the middle term). Just as groups are generalized by semigroups, torsors are generalized by semitorsors which are simply sets with a ternary map satisfying (G3). It is already known that this concept has important applications in geometry and algebra. The idea can be traced back at least as far work of V.V. Vagner, e.g. [Va66],. On the other hand, restriction to the diagonal in a torsor gives rise to an inter- esting product m(x, y) := (xyx). The map σx : y 7→ m(x, y) is just inversion in the group (G, x). If G is a Lie torsor (defined in the obvious way), then (G, m) is a symmetric space in the sense of Loos [Lo69]. 0.3. Grassmannian semitorsors. One of the remarkable properties of the quin- tary map Γ defined above is that it admits an "algebraic continuation" from the subset D(Γ) ⊂ X 5 of 5-tuples from the Grassmannian X = Gras(W ) where it was initially defined to all of X 5. The definition given above requires that the pairs (E, X) and (F, Z) are complementary. On the other hand, fixing an arbitrary com- plementary pair (E, F ), there is another natural ternary product: with respect to the decomposition W = E ⊕ F , subspaces X, Y, Z, . . . of W can be considered as linear relations between E and F , and can be composed as such: ZY −1X is again a linear relation between E and F . Since ZY −1X depends on E and on F , we get another map Γ(X, E, Y, F, Z) := XY −1Z. Looking more closely at the definition of this map, one realizes that there is a natural extension of its domain for all pairs (E, F ), and that on D(Γ) this new definition of Γ coincides with the earlier one given by (0.1) (Theorem 2.3). Moreover, for any fixed pair (E, F ), the ternary product (XY Z) := Γ(X, E, Y, F, Z) 4 WOLFGANG BERTRAM AND MICHAEL KINYON turns the Grassmannian X into a semitorsor. The list of remarkable properties of Γ does not end here -- we also have symmetry properties with respect to the Klein 4-group acting on the variables (X, E, F, Z), certain interesting diagonal values re- lating the map Γ to lattice theoretic properties of the Grassmannian (Theorem 2.4) as well as self-distributivity of the product, reflecting the fact that all partial maps of Γ are structural, i.e., compatible with the whole structure (Theorem 2.7). To- gether, these properties can be used to give an axiomatic definition of an associative geometry (Chapter 3). 0.4. Correspondence with associative algebras and pairs. Taking the Lie functor for Lie groups as model, we wish to define a multilinear tangent object attached to an associative geometry at a given base point. A base point in X is a fixed complementary (we say also transversal ) pair (o+, o−). The pair of abelian , C o+) then plays the role of a pair of "tangent spaces", groups (A+, A−) := (C o− and the role of the Lie bracket is taken by the following pair of maps: (x, y, z) 7→ Γ(x, o+, y, o−, z). f ± : A± × A∓ × A± → A±; One proves that f ± are trilinear (Theorem 3.4). Since the maps f ± come from a semitorsor, they form an associative pair, i.e., they satisfy the para-associative law (see Appendix B). Conversely, one can construct, for every associative pair, an asso- ciative geometry having the given pair as tangent object (Theorem 3.5). The proto- type of an associative pair are operator spaces, (A+, A−) = (Hom(E, F ), Hom(F, E)), with trilinear products f +(X, Y, Z) = XY Z, f −(X, Y, Z) = ZY X. They corre- spond precisely to Grassmannian geometries X = Gras(E ⊕ F ) with base point (o+, o−) = (E, F ). Associative unital algebras are associative pairs of the form (A, A); in the exam- ple just mentioned, this corresponds to the special case E = F . In this example, the unit element e of A corresponds to the diagonal ∆ ⊂ E ⊕ E, and the subspaces (E, ∆, F ) are mutually complementary. On the geometric level, this translates to the existence of a transversal triple (o+, e, o−). Thus the correspondence between associative geometries and associative pairs contains as a special case the one be- tween associative geometries with transversal triples and unital associative algebras (Theorem 3.7). 0.5. Further topics. Since associative algebras play an important role in modern mathematics, the present work is related to a great variety of topics and leads to many new problems located at the interface of geometry and algebra. We mention some of them in the final chapter of this work, without attempting to be exhaus- tive. In particular, in part II of this work ([BeKi09]) we will extend the theory to involutive associative algebras (topic (2) mentioned in Chapter 4). Acknowledgment. We would like to thank Boris Schein for enlightening us as to the history of the torsor/groud concept. The second named author would like to thank the Institut ´Elie Cartan Nancy for hospitality when part of this work was carried out. Notation. Throughout this work, K denotes a commutative unital ring and B an associative unital K-algebra, and we will consider right B-modules V, W, . . .. We ASSOCIATIVE GEOMETRIES. I 5 think of B as "base ring", and the letter A will be reserved for other associative K-algebras such as EndB(W ). For a first reading, one may assume that B = K; only in Theorem 3.7 the possibility to work over non-commutative base rings becomes crucial. When viewing submodules as elements of a Grassmannian, we will frequently use lower case letters to denote them, since this matches our later notation for abstract associative geometries. However, we will also sometimes switch back to the upper case notation we have already used whenever it adds clarity. 1. Grassmannian torsors The Grassmannian of a right B-module W is the set X = Gras(W ) = GrasB(W ) of all B-submodules of W . If x ∈ X and a ∈ X are complementary (W = x ⊕ a), we will write x⊤a and call the pair (x, a) transversal. We write Ca := a⊤ := {x ∈ X x⊤a} for the set of all complements of a and Cab := a⊤ ∩ b⊤ for the set of common complements of a, b ∈ X . We think of a⊤ and Cab (which may or may not be empty) as "open chart domains" in X . The following discussion makes this more precise. 1.1. Connected components and base points. Connectedness. We define an equivalence relation in X : x ∼ y if there is a finite sequence of "charts joining x and y", i.e.: ∃a0, a1, . . . , ak such that a0 = x, ak = y and ∀i = 0, . . . , k − 1 : Cai,ai+1 6= ∅. The equivalence classes of this relation are called connected components of X . We say that x ∈ X is isolated if its connected component is a singleton. If B = K and K is a field, then connected components are never reduced to a point (unless x = 0 or x = W ). For instance, the connected components of Gras(Kn) are the Grass- mannians Grasp(Kn) of subspaces of a fixed dimension p (indeed, two subspaces of the same dimension p in Kn always admit a common complement, hence sequences of length 1 always suffice in the above condition. Base points and pair geometries. A base pair or base point in X is a fixed transversal pair, often denoted by (o+, o−). If (o+, o−) is a base point, then in general o+ and o− belong to different connected components, which we denote by X + and X −. For instance, in the Grassmann geometry Gras(Kn) over a field K, if o+ is of dimension p, then o− has to be of dimension q = n − p, and hence they belong to different components unless p = q = n 2 . More generally, we may consider certain subgeometries of X , namely pairs (X +, X −) of sets X ± ⊂ X such that, for every x ∈ X ±, the set x⊤ is a nonempty subset of X ∓. We refer to (X +, X −) as a pair geometry. For instance, if W = B, then X is the space of right ideals in B. Fix an idempotent e ∈ B and let o+ := eB, o− = (1 − e)B and X ± the set of all right ideals in B that are isomorphic to o± and have a complement isomorphic to o∓. Then (X +, X −) is a pair geometry. 6 WOLFGANG BERTRAM AND MICHAEL KINYON Transversal triples and spaces of the first kind. We say that X is of the first kind if there exists a triple (a, b, c) of mutually transversal elements, and of the second kind else. Clearly, a, b, c then all belong to the same connected component of X ; taking (a, c) as base point (o+, o−), we thus have X + = X −. Note that W = a ⊕ c with a ∼= b ∼= c, so W is "of even dimension". For instance, the Grassmann geometry Gras(Kn) over a field K is of the first kind if and only if n is even, and the preceding example of a pair geometry of right ideals is of the first kind if and only if o+ and o− are isomorphic as B-modules. In other words, B is a direct sum of two copies of some other algebra, and X + = X − is the projective line over this algebra, cf. [BeNe05]. 1.2. Basic operators and the product map Γ. If x and a are two complemen- tary B-submodules, let P a x : W → W be the projector onto x with kernel a. Since a and x are B-modules, this map is B-linear. The relations P a x ◦ P a z = P a x , P a x ◦ P b x = P b x, P z b ◦ P a z = 0 will be constantly used in the sequel. For a B-linear map f : W → W , we denote by [f ] := f mod K× be its projective class with respect to invertible scalars from K. By 1 we denote the (class of) the identity operator on W . We define the following operators : if a⊤x and z⊤b, define the middle multiplication operator (motivation for this terminology will be given below) and if a⊤x and y⊤b, define the left multiplication operator Mxabz := [P a x − P z b ], Lxayb := [1 − P x a P b y ] and if a⊤y and z⊤b, define the right multiplication operator Raybz := Lzbya = [1 − P z b P a y ]. For a scalar s ∈ K and a transversal pair (x, a), the dilation operator is defined by δ(s) xa := [sP x a + P a x ] = [1 − (1 − s)P x a ] = [s1 + (1 − s)P a x ]. Note that the dilation operator for the scalar −1 is also a middle multiplication operator: δ(−1) xa = [−P x a + P a x ] = Mxaax, and it is induced by a reflection with respect to a subspace. Also, δ(1) δ(0) xa = [P a x ]. xa = 1 and Proposition 1.1. i ) (Symmetry) Mxabz is invariant under permutations of indices by the Klein 4-group: ii ) (Fundamental Relation) Whenever u, x⊤a and v, z⊤b, Mxabz = Maxzb = Mbzxa = Mzbax. RaubzLxavb = MxabzMuabv = LxavbRaubz. ASSOCIATIVE GEOMETRIES. I 7 iii ) (Diagonal values) If x ∈ Cab, and, for all u ∈ Cab and z⊤b, Lxaxb = 1 = Raxbx, Muabz(u) = z = Raubz(u). iv ) (Compatibility) If x⊤a, y⊤b, z⊤b and Cab is not empty, then and if x⊤a, z⊤b, y⊤a and Cab is not empty, then Lxayb(z) = Mxabz(y), Mxabz(y) = Raybz(x). v ) (Invertibility) Let (x, a, y, b, z) ∈ X 5 such that x, y, z ∈ Cab. Then the operators are invertible, with inverse operators, respectively, Lxayb, Mxabz, Raybz Lyaxb, Mzabx, Razby. Proof. (i): Mxabz = [P a x − P z b ] = [P z b − P a x ] = Mbzxa. Since this is the only place where we really use that [f ] = [−f ], for simplicity of notation, we henceforth omit the brackets [ a = (1 − P z v P z a P b a P b b ) − (1 − P a b P a u = 0: v )(1 − P z v − P z b P a (ii): Using in the second line that P x Mzbax = P b x ) = Mxabz z − P x b P a u ) ]. u a P b LxavbRaubz = (1 − P x = 1 − P x = 1 − (1 − P a = P a = (P a = MxabzMuabv. x + P v x − P z x )(1 − P v b ) − P z b P a u b − P a b )(P a b − P z x P v u − P v b ) b P a u The relation RaubzLxavb = MxabzMuabv now follows from (i). (iii): Lxaxb = 1 = Raxbx is clear. Fix an element u ∈ Cab. Then, for all z⊤b, Muabz(u) = Mzbau(u) = (P b z − P u a )(u) = P b z (u) = z since both u and z are complements of b. Similarly, u )(u) = (1 − P z Raubz(u) = (1 − P z b P a b )(u) = P b z (u) = z. (iv): By (ii), MxabzMuaby = LxaybRaubz. Apply this operator to u ∈ Cab and use that, by (iii), Muaby(u) = y and Raubz(u) = z. One gets Mxabz(y) = MxabzMuaby(u) = LxaybRaubz(u) = Lxayb(z). Via the symmetry relation (i), the second equality can also be written Mzbax(y) = Lzbya(x) and hence is equivalent to the first one. (v): Since Lxaxb = 1 = Raxbx, the fundamental relation (ii) implies MxabzMzaby = Lxayb and MxabzMzabx = Lxaxb = 1, 8 WOLFGANG BERTRAM AND MICHAEL KINYON hence Mxabz is invertible with inverse Mzabx. The other relations are proved simi- larly. (cid:3) Remark. We will prove in Chapter 2 by different methods that the assumption Cab 6= ∅ in (iv) is unnecessary. Definition (of the product map Γ). We define a map Γ : D(Γ) → X on the following domain of definition: let DL := {(x, a, y, b, z) ∈ X 5 x⊤a and y⊤b} DR := {(x, a, y, b, z) ∈ X 5 y⊤a and z⊤b} DM := {(x, a, y, b, z) ∈ X 5 x⊤a, z⊤b and Cab 6= ∅} D(Γ) := DL ∪ DR ∪ DM , and define Γ : D(Γ) → X by Γ(x, a, y, b, z) :=  Lxayb(z) Raybz(x) Maxbz(y) if if if (x, a, y, b, z) ∈ DL (x, a, y, b, z) ∈ DR (x, a, y, b, z) ∈ DM . This is well-defined: empty and the preceding proposition implies that if (x, a, y, b, z) ∈ DL ∩ DR, then y ∈ Cab, hence Cab is not Lxayb(z) = Mxabz(y) = Raybz(x). Similar remarks apply to the cases (x, a, y, b, z) ∈ DL ∩ DM or (x, a, y, b, z) ∈ DR ∩ DM . The quintary map Γ explains our terminology and notation: Lxayb is the left multiplication operator, acting on the last argument z, and similarly R and M denote right and middle multiplication operators. From the definition it follows easily that the symmetry relation holds for all (x, a, y, b, z) ∈ D(Γ). On the other hand, the relation Γ(x, a, y, b, z) = Γ(z, b, y, a, x) Γ(x, a, y, b, z) = Γ(a, x, y, z, b) holds if (x, a, y, z, b) ∈ DM ; but at present it is somewhat complicated to show that this relation is valid on all of D(Γ) (this will follow from the results of Chapter 2). As to the "diagonal values", for x ∈ Cab we have Γ(x, a, x, b, z) = z = Γ(z, b, x, a, x) . If we assume just a⊤x and b⊤z, then we can only say in general that Γ(x, a, x, b, z) = (1 − P z b P a x )(x) = P b z (x) ⊂ z . If a, b⊤x and b⊤z, then, thanks to the symmetry relation Mxabz = Maxzb, (1.1) Mxabz(a) = Γ(x, a, a, b, z) = Γ(a, x, a, z, b) = b . Definition (of the dilation map Πs). Fix s ∈ K. Let D(Πs) := {(x, a, z) ∈ X 3 x⊤a or z⊤a} ASSOCIATIVE GEOMETRIES. I 9 and define a ternary map Πs : D(Πs) → X by Πs(x, a, z) :=(δ(s) xa (z) δ(1−s) za (x) if x⊤a z⊤a. if As above, this map is well-defined. The symmetry relation Πs(x, a, y) = Π1−s(y, a, x) follows easily from the definition. Note that, if s is invertible in K and x⊤a, then the dilation operator δ(s) xa is invertible with inverse δ(s−1) xa . 1.3. Grassmannian torsors and their actions. Recall from §0.2 and Appendix A the definition and elementary properties of torsors. Theorem 1.2. The Grassmannian geometry (X ; Γ, Πr) defined in the preceding subsection has the following properties: i ) For a, b ∈ X fixed, Cab with product (xyz) := Γ(x, a, y, b, z) is a torsor (which will be denoted by Uab). In particular, for a triple (a, y, b) with y ∈ Cab, Cab is a group with unit y and multiplication xz = Γ(x, a, y, b, z). ii ) The map Γ is symmetric under the permutation (15)(24) (reversal of argu- ments): Γ(x, a, y, b, z) = Γ(z, b, y, a, x) In other words, Uab is the opposite torsor of Uba (same set with reversed prod- uct). In particular, the torsor Ua := Uaa is commutative. iii ) The commutative torsor Ua is the underlying additive torsor of an affine space: for any a ∈ X , Ua is an affine space over K, with additive structure given by (sum of x and z with respect to the origin y), and action of scalars given by x +y z = Γ(x, a, y, a, z), (multiplication of y by s with respect to the origin x). sy + (1 − s)x = Πs(x, y) Proof. (i) Let us show first that Cab is stable under the ternary map (xyz). Let x, y, z ∈ Cab and consider the bijective linear map g := Mxabz. We show that g(y) ∈ Cab. By equation (1.1), we have the "diagonal values" Mxabz(a) = b and Mxabz(b) = a. Thus, if y is complementary to a and b, g(y) is complementary both to g(a) = b and to g(b) = a, which means that g(y) ∈ Cab. The associativity follows immediately from the "fundamental relation" (Proposi- tion 1.1(ii)): (xv(yuz)) = LxavbRaubz(y) = RaubzLxavb(y) = ((xvy)uz), and the idempotent laws from (xxy) = Lxaxb(y) = 1(y) = y, (yxx) = Raxbx(y) = 1(y) = y. Thus Cab is a torsor. (ii) This has already been shown in the preceding section. 10 WOLFGANG BERTRAM AND MICHAEL KINYON (iii) The set Ca is the space of complements of a. It is well-known that this is an affine space over K. Let us recall how this affine structure is defined (see, e.g., [Be04]): elements v ∈ Ca are in one-to-one correspondence with projectors of the form P a v . Then, for u, v, w ∈ Ua, the structure map (u, v, w) 7→ u +v w in the affine space Ca is given by associating to (u, v, w) the point corresponding to the projector P a w, and the structure map (v, w) 7→ r ·v w = (1 − r)v + rw by associating to (v, w) the point corresponding to the projector rP a u + (1 − r)P a v . Now write y = P a y (W ); then we have u − P a v + P a Γ(x, a, y, a, z) = (P a = (P a = (P a x − P z a )(y) x − 1 + P a x − P a y + P a z )P a y (W ) z )(W ), and x + s1)(z) = ((1 − s)P a proving that (iii) describes the usual affine structure of Ca. Πs(x, a, y) = ((1 − s)P a x + sP a z )(W ), (cid:3) Homomorphisms. We think of the maps Γ : D(Γ) → X and Πr : D(Πr) → X as quintary, resp. ternary "product maps" defined on (parts of) direct products X 5, resp. X 3. Thus we have basic categorical notions just as for groups, rings, modules etc.: homomorphisms are maps g : X → Y preserving transversality (x⊤y implies g(x)⊤g(y)) and such that, for all 5-tuples in D(Γ), resp. triples in D(Πr), g (Γ(u, c, v, d, w)) = Γ (g(u), g(c), g(v), g(d), g(w)) , g (Πr(u, c, v)) = Πr (g(u), g(c), g(v)) . Essentially, this means that all restrictions of g, Uab → Ug(a),g(b), Ua → Ug(a), are usual homomorphisms (of torsors, resp. of affine spaces). We may summa- rize this by saying that g is "locally linear" and "compatible with all local group structures". Theorem 1.3. Assume x, y, z ∈ Uab. Then the operators Mxabz : X → X , Lxayb : X → X , Raybz : X → X are automorphisms of the geometry (X , Γ, Πr), and the groups (Uab, y) act on X by automorphisms both from the left and from the right via (Uab, y) × X → X , (x, z) 7→ Lxayb(z) = Γ(x, a, y, b, z), respectively X × (Uab, y) → X , (x, z) 7→ Raybz(x) = Γ(x, a, y, b, z). For fixed (a, y, b), the left and right actions commute. Proof. The construction of the product map Γ is "natural" in the sense that all elements of GlB(W ) (acting from the left on W , commuting with the right B-module structure) act by automorphisms of (X , Γ), just by ordinary push-forward of sets. This follows immediately from the relation g ◦ P a g(x) ◦ g. In particular, the invertible linear operators Mxabz, Lxayb and Raybz induce automorphisms of (X , Γ). x = P g(a) ASSOCIATIVE GEOMETRIES. I 11 Now fix y ∈ Uab and consider it as the unit in the group (Uab, y). The claim on the left action amounts to the identities Lyayb = id (which we already know) and, for all x, x′ ∈ Uab and all z ∈ X , Γ(x, a, y, b, Γ(x′, a, y, b, z)) = Γ(Γ(x, a, y, b, x′), a, y, b, z). First, note that, if z is "sufficiently nice", i.e., such that the fundamental relation (Proposition 1.1(ii)) applies, then this holds indeed. We will show in Chapter 2 that the identity in question holds very generally, and this will prove our claim. Therefore we leave it as a (slighly lengthy) exercise to the interested reader to prove the claim in the present framework. The claims concerning the right action are proved in the same way, and the fact that both actions commute is precisely the content of the fundamental relation (Proposition 1.1(ii)) (cid:3) Inner automorphisms. We call automorphisms of the geometry defined by the pre- ceding theorem inner automorphisms, and the group generated by them the inner automorphism group. Note that middle multiplications Mxabz are honest automor- phisms of the geometry (X , Γ), although they are anti -automorphisms of the torsor Uab; this is due to the fact that they exchange a and b. On the other hand, Lxayb and Raybz are automorphisms of the whole geometry and of Uab. Note also that the action of the groups Uab is of course very far from being regular on its orbits, except on Uab itself. For instance, a and b are fixed points of these actions, since Γ(x, a, y, b, b) = b and Γ(x, a, y, b, a) = a. Finally, the statements of the preceding two theorems amount to certain algebraic identities for the multiplication map Γ. This will be taken up in Chapter 2, where we will not have to worry about domains of definition. 1.4. Affine picture of the torsor Uab. It is useful to have "explicit formulas" for our map Γ. Such formulas can be obtained by introducing "coordinates" on X in the following way (see [Be04]). First of all, choose a base point (o+, o−) and consider the pair geometry (X +, X −), where X ± is the space of all submodules isomorphic to o± and having a complement isomorphic to o∓. We identify X + with injections x : o+ → W of B right-modules, modulo equivalence under the action of the group G := Gl(o+) (x ∼= x ◦ g, where g acts on o+ on the left), and X − with B-linear surjections a : W → o+ (modulo equivalence a ∼= g ◦ a for g ∈ G). Equivalence classes are denoted by [x], resp. [a]. Proposition 1.4. The following explicit formulae hold for x, y, z ∈ X +, a, b ∈ X −. i ) if x⊤a and z⊤b (middle multiplication), then Γ(cid:16)[x], [a], [y], [b], [z](cid:17) =hx(ax)−1ay − y + z(bz)−1byi , ii ) if a⊤x and b⊤y (left multiplication), then (1.2) (1.3) (1.4) iii ) if a⊤y and b⊤z (right multiplication), then Γ(cid:16)[x], [a], [y], [b], [z](cid:17) =hx(ax)−1ay(by)−1(bz) − y(by)−1(bz) + zi , Γ(cid:16)[x], [a], [y], [b], [z](cid:17) =hx − y(ay)−1ax + z(bz)−1za(ay)−1axi . 12 WOLFGANG BERTRAM AND MICHAEL KINYON Proof. The right hand side of (1.2) is a well-defined element of X , as is seen be replacing x by x ◦ g, resp. y by y ◦ g, z by z ◦ g and a by g ◦ a, b by g ◦ b. Note that [x] and [a] are transversal if and only if ax : o+ → o+ is invertible. Now, the operator x(ax)−1a : W → o+ → W has kernel a and image x and is idempotent, therefore it is P a see that z(bz)−1b is P b P a by the linear operator x . Similarly, we z , and hence the right hand side is induced by the operator z = Mxabz. Similarly, we see that the right hand side of (1.3) is induced x − 1 + P b P a x P b y − P b y + 1 = (P a x − 1)P b y + 1 = 1 − P x a P b y = Laxby and the one of (1.4) by 1 − P a y = Rzbya. (cid:3) y + P b z P a y = 1 + (P b z − 1)P a y = 1 − P z b P a As usual in projective geometry, the projective formulas from the preceding result may be affinely re-written: if y⊤b, we may affinize by taking ([y], [b]) as base point (o+, o−): we write W = o− ⊕ o+; then injections x : o+ → W , z : o+ → W that are transversal to the first factor can be identified with column vectors (by normalizing the second component to be the identity operator on o+) x =(cid:18)X 1(cid:19) , z =(cid:18)Z 1(cid:19) (columns with X, Z ∈ Hom(o+, o−)). In other terms, x and z are graphs of linear operators X, Z : o+ → o−. Surjections a : W → o+ that are transversal to the second factor correspond to row vectors (A, 1) (row with A ∈ Hom(o−, o+)). Note, however, that the kernel of (A, 1) is determined by the condition Au + v = 0, i.e., v = −Au, and hence a is the graph of −A : o− → o+. Therefore we write a = (−A, 1). The base point y = o+ is the column (0, 1)t, and the base point b = o− is the row (0, 1). Since ax = (−A, 1)(X, 1)t = 1 − AX, a and x are transversal iff 1 − AX : o+ → o+ is an invertible operator (in Jordan theoretic language: the pair (X, A) is quasi-invertible, cf. [Lo75]). Using this, any of the three formulas from the preceding proposition leads to the "affine picture": Γ(x, a, y, b, z) =(cid:20)(cid:18)X 1(cid:19) (1 − AX)−1 −(cid:18)0 1(cid:19) +(cid:18)Z 1(cid:19)(cid:21) =(cid:20)(cid:18)−ZAX + X + Z 1 (cid:19)(cid:21) . Finally, identifying x with X, y with Y and so on, we may write Γ(X, A, O+, O−, Z) = X − ZAX + Z . This formula is interesting in many respects: it is affine in all three variables, and the product ZAX from the associative pair (cid:0)Hom(o+, o−), Hom(o−, o+)(cid:1) shows up. We will give conceptual explanations of these facts later on. Also, it is an easy exercise to check directly that (X, Z) 7→ X − ZAX + Z defines an associative product on Hom(o+, o−) and induces a group structure on the set of elements X such that 1 − AX is invertible. ASSOCIATIVE GEOMETRIES. I 13 Other "rational" formulas. More generally, having fixed (o+, o−), we may write a, b as row-, and x, y, z as column vectors, and then we get the general formula Γ(X, A, Y, B, Z) = (cid:20)(cid:18)X 1(cid:19) (1 − AX)−1(1 − AY ) −(cid:18)Y 1(cid:19) +(cid:18)Z 1(cid:19) (1 − BZ)−1(1 − BY )(cid:21) , which is (the class of) a vector with second component ("denominator") D := (1 − AX)−1(1 − AY ) − 1 + (1 − BZ)−1(1 − BY ), and first component ("numerator") N := X(1 − AX)−1(1 − AY ) − Y + Z(1 − BZ)−1(1 − BY ), so that the affine formula is Γ(X, A, Y, B, Z) = N D−1. Besides the above choice (Y = O+, B = O−), another reasonable choice is just B = O−, leading to Γ(X, A, Y, O−, Z) = X − (Y − Z)(1 − AY )−1(1 − AX) . Similarly, for Y = O+ we get formulas that, in case A = B, correspond to well- known Jordan theoretic formulas for the quasi-inverse. Such formulas show that, if we work in finite dimension over a field, Γ is a rational map in the sense of algebraic geometry, and if we work in a topological setting over topological fields or rings, then Γ will have smoothness properties similar to the ones described in [BeNe05]. Case of a geometry of the first kind. Assume there is a transversal triple, say, (o+, e, o−). We may assume that e is the diagonal in W = o− ⊕ o+. Take, in the formulas given above, a = 0 = (0, 1), b = ∞ = (1, 0), y = (1, 1)t, ax = (0, 1)(X, 1)t = 1, bz = (1, 0)(Z, 1)t = Z, ay = 1, by = 1, so we get Γ(X, 0, e, ∞, Z) =(cid:20)(cid:18)X 1(cid:19) −(cid:18)1 1(cid:19) +(cid:18)Z 1(cid:19) Z −1(cid:21) =(cid:20)(cid:18) X Z −1(cid:19)(cid:21) =(cid:20)(cid:18)XZ 1 (cid:19)(cid:21) , and hence the affine picture is the algebra EndB(o+) with its usual product. Taking a = ∞, b = 0 gives the opposite of the usual product. Replacing e by y = {(v, Y v) Y : o+ → o−} (graph of an invertible linear map Y ), we get the affine picture Γ(X, 0, Y, ∞, Z) =(cid:20)(cid:18)XY −1Z 1 (cid:19)(cid:21) . 1.5. Affinization: the transversal case. If a and b are arbitrary, then in general the torsor Uab will be empty. Therefore we look at the pair (Ua, Ub). Theorem 1.5. For all a, b ∈ X , we have Γ(Ua, a, Ub, b, Ua) ⊂ Ua, Γ(Ub, a, Ua, b, Ub) ⊂ Ub . In other words, the maps Ua × Ub × Ua → Ua; Ub × Ua × Ub → Ub; (x, y, z) 7→ (xyz)+ := Lxayb(z) = Γ(x, a, y, b, z) , (x, y, z) 7→ (xyz)− := Raybz(x) = Γ(x, a, y, b, z) 14 WOLFGANG BERTRAM AND MICHAEL KINYON are well-defined. If, moreover, a⊤b, then both maps are trilinear, and they form an associative pair, i.e., they satisfy the para-associative law (cf. Appendix B) (xy(uvw)±)± = ((xyu)±vw)± = (x(vuy)∓w)±. Proof. Assume that x⊤a and y⊤b. By a direct calculation, we will show that Lxayb(Ua) ⊂ Ua. Let us write Lxayb in matrix form with respect to the decomposition W = a ⊕ x. The projectors P x y can be written a and P b P x a =(cid:18)1 0 0 0(cid:19) , P b y =(cid:18)α β γ δ(cid:19) whith α ∈ End(a), β ∈ Hom(x, a), etc. Thus Lxayb = 1 −(cid:18)1 0 0 0(cid:19)(cid:18)α β γ δ(cid:19) =(cid:18)1 − α −β 1 (cid:19) . 0 Let z ∈ Ua; it can be written as the graph {(Zv, v) v ∈ x} of a linear operator Z : x → a. Since Lxayb(cid:18)Zv v (cid:19) =(cid:18)1 − α −β 1 (cid:19)(cid:18)Zv v (cid:19) =(cid:18)(1 − α)Zv − βv 0 v (cid:19) , Lxayb(z) is the graph of the linear operator (1 − α)Z − β : x → a, and hence is )− again transversal to a, so ( is well-defined. Moreover, the calculation shows that z 7→ (xyz)+ is affine (we will see later that this map is actually affine with respect to all three variables, see Corollary 2.9). )+ is well-defined. By symmetry, it follows that ( Now assume that a⊤b, and write Lxayb in matrix form with respect to the de- composition W = a ⊕ b. The projectors P a x and P b y can be written P a x =(cid:18)0 X 1(cid:19) , P b y =(cid:18) 1 Y 0(cid:19) 0 0 where X ∈ Hom(b, a) and Y ∈ Hom(a, b). We get Lxayb = 1 −(cid:18)1 −(cid:18)0 X 1(cid:19)(cid:19)(cid:18) 1 Y 0(cid:19) = 1 −(cid:18)1 − XY 0 0(cid:19) =(cid:18)XY 0 1(cid:19) 0 0 0 0 and, writing z ∈ Ua as a graph {(Zv, v)v ∈ b}, we get Lxayb(cid:18)Zv v (cid:19) =(cid:18)XY 0 1(cid:19)(cid:18)Zv v (cid:19) =(cid:18)XY Zv v (cid:19) , 0 hence Lxayb(z) is the graph of XY Z : b → a. Thus, with V + = Ua V − = Ub ∼= Hom(a, b), the first ternary map is given by ∼= Hom(b, a), V + × V − × V + → V +, (X, Y, Z) 7→ XY Z. Similarly, one shows that the second ternary map is given by V − × V + × V − → V −, (X, Y, Z) 7→ ZY X. This pair of maps is the prototype of an associative pair (see Appendix B). (cid:3) ASSOCIATIVE GEOMETRIES. I 15 At this stage, the appearance of the trilinear expression ZY X, resp. ZAX, both in the affine pictures of the map from the preceding theorem and in the preceding section, related by the identity (1.5) X − (X − ZAX + Z) + Z = ZAX, looks like a pure coincidence. A conceptual explanation will be given in Chapter 3 (Lemma 3.2). 2. Grassmannian semitorsors In this chapter we extend the definition of the product map Γ onto all of X 5, and we show that the most important algebraic identities extend also. We use notation and general notions explained in the first section of the preceding chapter. 2.1. Composition of relations. Recall that, if A, B, C, . . . are any sets, we can compose relations: for subsets x ⊂ A × B, y ⊂ B × C, y ◦ x := yx := {(u, w) ∈ A × C ∃v ∈ B : (u, v) ∈ x, (v, w) ∈ y} . Composition is associative: both (z ◦ y) ◦ x and z ◦ (y ◦ x) are equal to (2.1) z ◦ y ◦ x = {(u, w) ∈ A × D ∃(v1, v2) ∈ y : (u, v1) ∈ x, (v2, w) ∈ z} . If x and y are graphs of maps X, resp. Y (v = Xu, w = Y v) then y ◦ x is the graph of Y X (w = Y v = Y Xu). The reverse relation of x is x−1 := {(w, v) ∈ B × A (v, w) ∈ x}. We have (yx)−1 = x−1y−1, and if x is the graph of a bijective map, then x−1 is the graph of its inverse map. For x, y, z ⊂ A × B, we get another relation between A and B by zy−1x. Obviously, this ternary composition satisfies the para-associative law, and hence relations between A and B form a semitorsor. Letting W := A × B, we have the explicit formula zy−1x =(cid:26)ω = (α′, β ′) ∈ W (cid:12)(cid:12)(cid:12) =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) ∃η = (α′′, β ′′) ∈ y : (α′, β ′′) ∈ x, (α′′, β ′) ∈ z (cid:27) ∃α′, α′′ ∈ A, ∃β ′, β ′′ ∈ B, ∃η ∈ y, ∃ξ ∈ x, ∃ζ ∈ z : ω = (α′, β ′), η = (α′′, β ′′), ξ = (α′, β ′′), ζ = (α′′, β ′) (cid:27) 2.2. Composition of linear relations. Now assume that A, B, C, . . . are linear spaces over B (i.e., right modules) and that all relations are linear relations (i.e., submodules of A ⊕ B, etc.). Then zy−1x is again a linear relation. Identifying A with the first and B with the second factor in W := A ⊕ B, the description of zy−1x given above can be rewritten, by introducing the new variables α := α′ − α′′, β := β ′ − β ′′, zy−1x =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) ∃α′, α′′ ∈ a, ∃β ′, β ′′ ∈ b, ∃η ∈ y, ∃ξ ∈ x, ∃ζ ∈ z : ω = α′ + β ′, η = α′′ + β ′′, ξ = α′ + β ′′, ζ = α′′ + β ′ (cid:27) ω = α′ + β ′, η = ω − α − β, ξ = ω − β, ζ = ω − α (cid:27) . ∃α′, α ∈ a, ∃β ′, β ∈ b, ∃η ∈ y, ∃ξ ∈ x, ∃ζ ∈ z : In order to stress that the product xy−1z depends also on A and B, we will hence- forth use lowercase letters a and b and write W = a ⊕ b. 16 WOLFGANG BERTRAM AND MICHAEL KINYON Lemma 2.1. Assume W = a ⊕ b and let x, y, z ∈ GrasB(W ). Then zy−1x =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) ∃ξ ∈ x, ∃α ∈ a, ∃η ∈ y, ∃β ∈ b, ∃ζ ∈ z : ω = ζ + α = α + η + β = ξ + β (cid:27) . Proof. Since W = a ⊕ b, the first condition (∃α′ ∈ a, β ′ ∈ b: ω = α′ + β ′) in the preceding description is always satisfied and can hence be omitted in the description of zy−1x. Replacing α by −α and β by −β, the claim follows. (cid:3) 2.3. The extended product map. Motivated by the considerations from the preceding section, we now define the product map Γ : X 5 → X for all 5-tuples of the Grassmannian X = GrasB(W ) by Γ(x, a, y, b, z) :=(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) ∃ξ ∈ x, ∃α ∈ a, ∃η ∈ y, ∃β ∈ b, ∃ζ ∈ z : ω = ζ + α = α + η + β = ξ + β (cid:27) . We will show, among other things, that this notation is in keeping with the one introduced in the preceding chapter. Firstly, however, we collect various equivalent formulas for Γ. The three conditions (2.2) ω = η + α + β ω = ω = β + ξ α + ζ can be re-written in various ways. For instance, subtracting the last two equations from the first one we get the equivalent conditions (2.3) ω = −η + ξ + ζ, ω = β + ξ, ω = α + ζ and hence, replacing η by −η, we get Γ(x, a, y, b, z) =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) ∃ξ ∈ x, ∃α ∈ a, ∃η ∈ y, ∃β ∈ b, ∃ζ ∈ z : ω = ζ + α = ξ + η + ζ = ξ + β (cid:27) . Next, letting α′ = −α and β ′ = −β, conditions (2.2) are equivalent to (2.4) η = ω + α′ + β ′, ζ = ω + α′, ξ = ω + β ′, and hence we get Γ(x, a, y, b, z) =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) ∃ξ ∈ x, ∃α ∈ a, ∃η ∈ y, ∃β ∈ b, ∃ζ ∈ z : η = ω + α + β, ζ = ω + α, ξ = ω + β (cid:27) ∃β ∈ b, ∃α ∈ a : ω + α ∈ z, ω + α + β ∈ y, ω + β ∈ xo . The following lemma now follows by straightforward changes of variables: ASSOCIATIVE GEOMETRIES. I 17 Lemma 2.2. For all x, a, y, b, z ∈ X , Γ(x, a, y, b, z) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) ∃ξ ∈ x, ∃ζ ∈ z : ζ + ω ∈ a, ζ + ω + ξ ∈ y, ω + ξ ∈ bo ∃ξ ∈ x, ∃α ∈ a : ω − α ∈ z, ξ − α ∈ y, ω − ξ ∈ bo ∃β ∈ b, ∃ζ ∈ z : ζ − ω ∈ a, ζ − β ∈ y, ω − β ∈ xo ∃η ∈ y, ∃β ∈ b : ω − η − β ∈ a, β + η ∈ z, ω − β ∈ xo ∃η ∈ y, ∃ζ ∈ z : ω + ζ ∈ a, ζ + η ∈ b, ω + ζ + η ∈ xo We refer to the descriptions of the lemma as the "(x, z)-", "(x, a)-description", and so on. The (a, b)-description is particularly useful for the proof of the theorem below. One may note that the only pairs of variables that cannot be used for such a description are (a, z) and (x, b), and that the signs in the terms appearing in these descriptions can be chosen positive if the pair is "homogeneous" (a subpair of (x, y, z) or of (a, b)), whereas for "mixed" pairs we cannot get rid of signs. Theorem 2.3. The map Γ : X 5 → X extends the product map defined in the preceding chapter, and has the following properties: (1 ) It is symmetric under the Klein 4-group: (a) (b) Γ(x, a, y, b, z) = Γ(z, b, y, a, x) , Γ(x, a, y, b, z) = Γ(a, x, y, z, b) . (2 ) For any pair (a, b) ∈ X 2, the product (xyz) := Γ(x, a, y, b, z) on X 3 satisfies the properties of a semitorsor, that is, Γ(cid:0)x, a, u, b, Γ(y, a, v, b, z)(cid:1) = Γ(cid:0)x, a, Γ(v, a, y, b, u), b, z(cid:1) = Γ(cid:0)Γ(x, a, u, b, y), a, v, b, z(cid:1) . We will write Xab for X equipped with this semitorsor structure. Then the semi- torsor Xba is the opposite semitorsor of Xab; in particular, Xaa is a commutative semitorsor, for any a. Proof. (1) The symmetry relation (a) is obvious from the definition of Γ. Exchang- ing x and a amounts in the (x, a)-description to exchanging simultaneously z and b, hence the symmetry relation (b) follows. For (2), we use the (a, b)-description: on the one hand, Γ(cid:0)x, a, u, b, Γ(y, a, v, b, z)(cid:1) = =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) = ω ∈ W (cid:12)(cid:12)(cid:12)  On the other hand, ω + α ∈ Γ(y, a, v, b, z), ω + α + β ∈ u, ω + β ∈ x (cid:27) ∃α ∈ a, ∃β ∈ b : ∃α ∈ a, ∃β ∈ b, ∃α′ ∈ a, ∃β ′ ∈ b : ω + α + β ∈ u, ω + β ∈ x, ω + α + α′ ∈ z, ω + α + α′ + β ′ ∈ v, ω + α + β ′ ∈ y   18 WOLFGANG BERTRAM AND MICHAEL KINYON Γ(cid:0)x, a, Γ(u, b, y, a, v), b, z(cid:1) = =(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) = ω ∈ W (cid:12)(cid:12)(cid:12)  ∃α′′ ∈ a, ∃β ′′ ∈ b : ∃α′′ ∈ a, ∃β ′′ ∈ b, ∃α′′′ ∈ a, ∃β ′′′ ∈ b : ω + α′′ ∈ z, ω + α′′ + β ′′ ∈ Γ(u, b, y, a, v), ω + β ′′ ∈ x (cid:27) ω + α′′ + β ′′ + β ′′′ ∈ v, ω + α′′ + β ′′ + α′′′ + β ′′′ ∈ y   ω + α′′ ∈ z, ω + β ′′ ∈ x, ω + α′′ + β ′′ + α′′′ ∈ u, Via the change of variables α′′ = α + α′, α′′′ = α′, β ′′ = β, β ′′′ = β, we see that these two subspaces of W are the same. The remaining equality is equivalent to the one just proved via the symmetry relation (a). Next, we show that the new map Γ coincides with the old one on D(Γ). Let us assume that (x, a, y, b, z) ∈ DL, so x⊤a and y⊤b. We use the (y, b)-description and let ζ := η + β, whence η = P b y ζ and β = P y b ζ. We get Γ(x, a, y, b, z) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) ∃ζ ∈ z : ω − P y ∃η ∈ y, ∃β ∈ b : ω − η − β ∈ a, β + η ∈ z, ω − β ∈ xo b (ζ)o b (ζ) ∈ x, ω − ζ ∈ ao b (ζ)) = ω − P y x (ω − ζ) = 0, P a x (ω − P y ∃ζ ∈ z : P a x ζ = P a x ω, ω = P y b ζ + P a x ω − P a x P y ∃ζ ∈ z : P a b ζo and a straightforward calculation shows that ∃ζ ∈ z : ω = (P y b + P a x − P a x P y b )ζo P y b + P a x − P a x P y b = 1 − P x a P b y = Lxayb so that Γ(x, a, y, b, z) = Lxayb(z). This proves that the old and new definitions of Γ coincide on DL, and hence also on DR by the symmetry relation. Now we show that the new map Γ coincides with the old one on DM : assume a⊤x and b⊤z and use the (x, z)-description; let η := ζ − ω + ξ and observe that P a x ξ = ξ (since ζ − ω ∈ a), and similarly P b x )η, and thus z η = ζ, whence ω = ζ − η + ξ = (P b z − 1 + P a x η = P a Γ(x, a, y, b, z) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) ∃ξ ∈ x, ∃ζ ∈ z : ζ − ω ∈ a, ζ − ω + ξ ∈ y, ω − ξ ∈ bo ∃η ∈ y : ω = (P b z − 1 + P a x )ηo , that is, ω = −Mxabzη, and hence Γ(x, a, y, b, z) = Mxabz(y). (cid:3) 2.4. Diagonal values. We call diagonal values the values taken by Γ on the subset of X 5 where at least two of the five variables x, a, y, b, z take the same value. There are two different kinds of behavior on such diagonals: for the diagonal a = b (or, equivalently, x = z), we still have a rich algebraic theory which is equivalent to the Jordan part of our associative products; this topic is left for subsequent work (cf. Chapter 4). The three remaining diagonals (x = y, resp. a = z, resp. b = z) have an entirely different behavior: the algebraic operation Γ restricts in these cases to lattice theoretic operations, that is, can be expressed by intersections and sums of ASSOCIATIVE GEOMETRIES. I 19 subspaces. We will use the lattice theoretic notation x ∧ y = x ∩ y and x ∨ y = x + y. It is remarkable that two important aspects of projective geometry (the lattice theoretic and the Jordan theoretic) arise as a sort of "contraction" of the full map Γ, or, put differently, that they have a common "deformation", given by Γ. Theorem 2.4. The map Γ : X 5 → X takes the following diagonal values: (1 ) values on the "diagonal x = y": for all (x, a, b, z) ∈ X 4, Γ(x, a, x, b, z) = (z ∨ (x ∧ a)) ∧ (b ∨ x). In particular, we get the following "subdiagonal values": for all x, a, y, b, z, (i ) subdiagonal x = y = z: Γ(x, a, x, b, x) = x (law (xxx) = x in Xab), (ii ) subdiagonal x = y = a: Γ(x, x, x, b, z) = (z ∨ x) ∧ (b ∨ x) (iii ) subdiagonal x = y = a and b = z: Γ(x, x, x, z, z) = z ∨ x (iv ) subdiagonal x = y = b: Γ(x, a, x, x, z) = (z ∨ (x ∧ a)) ∧ x. (v ) subdiagonal x = y = b and a = z: Γ(x, a, x, x, a) = a ∧ x (vi ) subdiagonal x = y, a = z: Γ(x, a, x, b, a) = a ∧ (b ∨ x) (vii ) subdiagonal x = y, a = b: Γ(x, a, x, a, z) = (z ∨ (x ∧ a)) ∧ (x ∨ a) (viii ) subdiagonal x = y, z = b: Γ(x, a, x, z, z) = z ∨ (x ∧ a) (2 ) diagonal a = z: for all (x, a, y, b) ∈ X 4, Γ(x, a, y, b, a) = a ∧ (b ∨ (x ∧ (y ∨ a))) In particular, on the subdiagonal x = z = b, we have, for all x, a, y ∈ X , (3 ) diagonal b = z: for all (x, a, y, b) ∈ X 4, Γ(x, a, y, x, a) = a ∧ x. Γ(x, a, y, b, b) = b ∨ (a ∧ (x ∨ (y ∧ b))) In particular, on the subdiagonal b = z, x = a, we have, for all a, y, b ∈ X , and on a = b = az: for all x, a, y, Γ(x, a, y, a, a) = a. Γ(a, a, y, b, b) = b ∨ a, Proof. In the following proof, in order to avoid unnecessary repetitions, it is always understood that α ∈ a, ξ ∈ x, β ∈ b, η ∈ y, ζ ∈ z. In all three items, the determination of the "subdiagonal values" is a straightforward consequence, using the absorption laws u ∨ (u ∧ v) = u, u ∧ (u ∨ v) = u. Now we prove (1) (diagonal x = y). Let ω ∈ Γ(x, a, x, b, z), then ω = ξ + β, hence ω ∈ (x ∨ b), and ω = η + ξ + ζ with v := ω − ζ = η + ξ ∈ x (since x = y). On the other hand, v = ω − ζ = α ∈ a, whence ω = v + ζ with v ∈ (x ∧ a), proving one inclusion. Conversely, let ω ∈ (z ∨(x∧a))∧(b∨x). Then ω = β +ξ = α +ζ with α ∈ (x∧a). Let η := ξ − α. Then η ∈ x, and ω = ξ + β = η + α + β, hence ω ∈ Γ(x, a, x, b, z). Next we prove (2) (diagonal a = z). Let ω ∈ Γ(x, a, y, b, a), then ω = ζ + α with ζ, α ∈ z = a, whence ω ∈ a. Moreover, ω = ξ + β = η + α + β, with η + α = ξ ∈ x and η + α ∈ y ∨ a, whence ω ∈ b ∨ (x ∧ (y ∨ a)). Conversely, let ω ∈ a ∧ (b ∨ (x ∧ (y ∨ a))). Then ω ∈ b ∨ (x ∧ (y ∨ a)), that is, ω = β + (η + α) with ξ := η + α ∈ x. Letting ζ := ω − α ∈ a (here we use that ω ∈ a), we have ω = ζ + α, proving that ω ∈ Γ(x, a, y, b, a). 20 WOLFGANG BERTRAM AND MICHAEL KINYON The proof for (3) (diagonal z = b) is "dual" to the preceding one and will be left (cid:3) to the reader. Remark. By arguments of the same kind as above, one can show that the diagonal value for x = y (part (1)) admits also another, kind of "dual", expression: (2.5) Γ(x, a, x, b, z) = (z ∧ (x ∨ b)) ∨ (a ∧ x). The equality of these two expressions is equivalent to the modular law (2.6) Γ(x, a, x, x, z) = (z ∧ x) ∨ (a ∧ x) = ((z ∧ x) ∨ a) ∧ x. It is known [PR09] that any (finitely based) variety of lattices can be axiomatized by a single quaternary operation q(·, ·, ·, ·) given in terms of the lattice operations by q(x, b, z, a) = (z ∧ (x ∨ b)) ∨ (a ∧ x). That is, one may start with a quarternary operation q satisfying certain identities (which we omit), define x ∨ y = q(y, x, x, y) and x ∧ y = q(y, y, x, x), and the resulting structure will be a lattice. From the preceding paragraph, we see that in our setting, q(x, b, z, a) = Γ(x, a, x, b, z). Thus the quaternary approach to lattices emerges from the present theory in a completely natural way. Corollary 2.5. (1 ) If b ∨ x = W and a ∧ x = 0, then Γ(x, a, x, b, z) = z. (2 ) If a ∨ y = W and b ∨ x = W , then Γ(x, a, y, b, a) = a. (3 ) If x ∧ a = 0 and y ∧ b = 0, then Γ(x, a, y, b, b) = b. Proof. Straightforward consequences of the theorem, again using the absorption laws. (cid:3) 2.5. Structural transformations and self-distributivity. Homomorphisms be- tween sets with quintary product maps Γ, Γ′ are defined in the usual way, and may serve to define the category of Grassmannian geometries with their product maps Γ. We call this the "usual" category. There is another and often more useful way to turn them into a category which we call "structural": Definition. Let W, W ′ be two right B-modules and (X , Γ), (X ′, Γ′) their Grass- mannian geometries. A structural or adjoint pair of transformations between X and X ′ is a pair of maps f : X → X ′, g : X ′ → X such that, for all x, a, y, b, z ∈ X , x′, a′, y′, b′, z′ ∈ X ′, f(cid:0)Γ(x, g(a′), y, g(b′), z)(cid:1) = Γ′(f (x), a′, f (y), b′, f (z)), g(cid:0)Γ′(x′, f (a), y′, f (b), z′)(cid:1) = Γ(g(x′), a, g(y′), b, g(z′)) . In other words, for fixed a, b, resp. a′, b′, the restrictions f : Xg(a′),g(b′) → X ′ a′,b′ , g : X ′ f (a),f (b) → Xa,b are homomorphisms of semitorsors. We will sometimes write (f, f t) for a structural pair (although g need not be uniquely determined by f ). It is easily checked that the composition of structural pairs gives again a struc- tural pair, and Grassmannian geometries with structural pairs as morphisms form a category. Isomorphisms, and, in particular, the automorphism group of (X , Γ), are essentially the same in the usual and in the structural categories, but the en- domorphism semigroups may be very different. Roughly speaking, Grassmannian ASSOCIATIVE GEOMETRIES. I 21 geometries tend to be "simple objects" in the usual category (hence morphisms tend to be either trivial or injective), whereas they are far from being simple in the structural category, so there are many morphisms. One way of constructing such morphisms is via ordinary B-linear maps f : W → W ′, which induce maps between the corresponding Grassmannians X = Gras(W ) and X ′ = Gras(W ′): f ∗ : X ′ → X ; y 7→ f −1(y). f∗ : X → X ′; x 7→ f (x), Note that, in general, these maps do not restrict to maps between connected com- ponents (for instance, f∗ and f ∗ do not restrict to everywhere defined maps between projective spaces PW and PW ′ if f is not injective). We will show that (f∗, f ∗) is an adjoint pair, as a special case of the following result: Theorem 2.6. Given a linear relation r ⊂ W ⊕ W ′, let r∗ : X → X ′; x 7→ r(x) := {ω′ ∈ W ′ ∃ξ ∈ x : (ξ, ω′) ∈ r} , y 7→ r−1(y) := {ω ∈ W ∃η ∈ y : (ω, η) ∈ r} . r∗ : X ′ → X ; Then (r∗, r∗) is a structural pair of transformations between X and X ′. Proof. Using the (a, b)-description, on the one hand, r∗Γ(x, r∗a′, y, r∗b′, z) = ∃ω ∈ W, ∃α ∈ r∗a′, ∃β ∈ r∗b′ : , The subspaces of W determined by these two conditions are the same, as is seen by the change of variables ζ = ω + α, ξ = ω + β, η = ω + α + β, α′′ = α′, β ′′ = β ′ in one direction, and ω = η − ζ − ξ, α′′ = α′, β ′′ = β ′, α = ζ − ω = η − ξ, β = ξ − ω = η − ζ in the other, and using that r is a linear subspace. (cid:3) Remark. The proof shows that the same result would hold if we had formulated the structurality property with respect to another "admissible" pair of variables instead of (a, b), for instance (y, b) or (x, z), by using the corresponding description. However, we prefer to distinguish the pair formed by the second and fourth variable =nω′ ∈ W ′(cid:12)(cid:12)(cid:12) =(cid:26)ω′ ∈ W ′(cid:12)(cid:12)(cid:12) = ω′ ∈ W ′(cid:12)(cid:12)(cid:12)  =(cid:26)ω′ ∈ W ′(cid:12)(cid:12)(cid:12) =(cid:26)ω′ ∈ W ′(cid:12)(cid:12)(cid:12) and on the other hand, Γ(r∗x, a′, r∗y, b′, r∗z) = ∃ω ∈ W, ∃α′ ∈ a′, ∃α ∈ W, ∃β ′ ∈ b′, ∃β ∈ W : (ω, ω′) ∈ r, (α, α′) ∈ r, (β, β ′) ∈ r, ω + α ∈ z, ω + β ∈ x, ω + α + β ∈ y ∃ω ∈ Γ(x, r∗a′, y, r∗b′, z) : (ω, ω′) ∈ ro (ω, ω′) ∈ r, ω + α ∈ z, ω + β ∈ x, ω + α + β ∈ y (cid:27)   ω′ + α′′ ∈ r∗z, ω′ + β ′′ ∈ r∗x, ω′ + α′′ + β ′′ ∈ r∗y (cid:27) (ζ, ω′ + α′′) ∈ r, (ξ, ω′ + β ′′) ∈ r, (η, ω′ + α′′ + β ′′) ∈ r (cid:27) . ∃α′′ ∈ a′, ∃β ′′ ∈ b′, ∃ζ ∈ z, ∃ξ ∈ x, ∃η ∈ y : ∃α′′ ∈ a′, ∃β ′′ ∈ b′ : 22 WOLFGANG BERTRAM AND MICHAEL KINYON in order to have the interpretation of structural transformations in terms of torsor homomorphisms, for fixed (a, b). Remark. The construction from the theorem is functorial. In particular, the semi- group of linear relations on W × W (to be more precise: a quotient with respect to scalars) acts by structural pairs on X . Theorem 2.7. We define operators of left-, middle- and right multiplication on X by Lxayb(z) := Raybz(x) := Mxabz(y) := Γ(x, a, y, b, z). Then, for all x, a, y, b, z ∈ X , the pairs (Lxayb, Lyaxb), (Mxabz, Mzabx), (Raybz, Razby) are structural transformations of the Grassmannian geometry X . Proof. Let lx,a,y,b ⊂ W ⊕ W be the linear relation defined by lx,a,y,b := {(ζ, ω) ∈ W ⊕ W ∃ξ ∈ x : ω + ζ ∈ a, ω + ζ + ξ ∈ y, ω + ξ ∈ b}. Then it follows immediately by using the (x, z)-description that (lx,a,y,b)∗(z) = {ω ∈ W ∃ζ ∈ z : (ζ, ω) ∈ lx,a,y,b} = Γ(x, a, y, b, z) = Lxayb(z). On the other hand, (lx,a,y,b)∗(z) = {ω ∈ W ∃ζ ∈ z : (ω, ζ) ∈ lx,a,y,b} ∃ζ ∈ z, ∃ξ ∈ x : ω + ζ ∈ a, ω + ζ + ξ ∈ y, ζ + ξ ∈ bo = Γ(y, a, x, b, z) = Lyaxb(z) , =nω ∈ W (cid:12)(cid:12)(cid:12) where the third equality follows by using the (y, z)-description with permuted vari- ables. This proves that (Lxayb, Lyaxb) is a structural pair; the claim for right multipli- cations is just an equivalent version of this, and the claim for middle multiplications is proved in the same way as above. (cid:3) Remark. We have proved that, in terms of inverses of linear relations, (2.7) (lx,a,y,b)−1 = ly,a,x,b. If x⊤a and y⊤b, then lxaby is the graph of the linear operator Lxayb ∈ End(W ); for x, y ∈ Uab, this operator is invertible and the preceding formula holds in the sense of an operator equation. Corollary 2.8. The multiplication map satisfies the following "self-distributivity" identities: Γ(cid:16)x, a, Γ(cid:0)u, Γ(a, z, c, x, b), v, Γ(a, z, d, x, b), w(cid:1), b, z(cid:17) = Γ(cid:16)Γ(x, a, u, b, z), c, Γ(x, a, v, b, z), d, Γ(x, a, w, b, z)(cid:17) Γ(cid:16)x, a, y, b, Γ(cid:0)u, Γ(y, a, x, b, c), v, Γ(y, a, x, b, d), w(cid:1)(cid:17) = Γ(cid:16)Γ(x, a, y, b, u), c, Γ(x, a, y, b, v), d, Γ(x, a, y, b, w)(cid:17) ASSOCIATIVE GEOMETRIES. I 23 Proof. The first identity follows by applying the adjoint pair (f, f t) = (Mxabz, Mzabx) to Γ(u, c, v, d, w) (and using the symmetry property), and similarly the second by using the pair (f, f t) = (Lxayb, Lyaxb). (cid:3) )+ : Ua × Ub × Ua → Ua and )− : Ub × Ua × Ub → Ub defined in Theorem 1.5 are tri-affine (i.e., affine in all Corollary 2.9. For all a, b ∈ X , the maps ( ( three variables) and satisfy the para-associative law (xy(uvw)±)± = ((xyu)±vw)± = (x(vuy)∓w)±. Proof. Let us show that Mxabz induces an affine map Ub → Ua, y 7→ (xyz)+, for fixed x, z ∈ Ua. We know already that this map is well-defined (Theorem 1.5). Since (f, g) = (Mxabz, Mzabx) is structural, the map f : Ug(a) → Ua is affine, where (according to Corollary 2.5, (1)), g(a) = Mzabx(a) = Γ(z, a, a, b, x) = Γ(a, z, a, x, b) = b. By the same kind of argument, using Corollary 2.5, (2) and (3), wee see that the )− follow by other partial maps are affine. The corresponding statements for ( symmetry, and the para-associative law follows by restriction of the para-associative law in the semitorsor Xab. (cid:3) Remark. For a = b, we get the additive torsor Ua, and if Uab 6= ∅, then we get a sort of "triaffine extension" of the torsor Uab. If a⊤b, then we have base points a in Ub and b in Ua, and obtain a trilinear product (Theorem 1.5). 2.6. The extended dilation map. Next we (re-)define, for r ∈ K, the dilation map Πr : X × X × X → X by the following equivalent expressions Πr(x, a, z) :=nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) =nω ∈ W (cid:12)(cid:12)(cid:12) ∃α ∈ a, ∃ζ ∈ z, ∃ξ ∈ x : ω − rα = ξ = ζ − αo ∃α ∈ a, ∃ζ ∈ z, ∃ξ ∈ x : ω + (1 − r)α = ζ = α + ξo ∃α ∈ a, ∃ζ ∈ z, ∃ξ ∈ x : ω = (1 − r)ξ + rζ, ζ − ξ = αo We refer to the last expression as the "(x, z)-description", and we define partial maps X → X by λr xa(z) := ρr az(x) := µr xz(a) := Πr(x, a, z) (where λ reminds us of "left", ρ "right" and µ "middle"). Theorem 2.10. The map Πr : X 3 → X extends the ternary map defined in the preceding chapter (and denoted by the same symbol there), and it has the following properties: (1 ) Symmetry: µr zx , that is, λr xz = µ1−r xa = ρ1−r ax or (2 ) Multiplicativity: if x⊤a and r, s ∈ K, Πr(x, a, z) = Π1−r(z, a, x). Πr(x, a, Πs(x, a, y)) = Πrs(x, a, y), (3 ) Diagonal values: Πr(x, a, x) = x, Π0(x, a, z) = Π1(z, a, x) = x ∧ (z ∨ a) = Γ(a, x, x, a, z). 24 WOLFGANG BERTRAM AND MICHAEL KINYON (4 ) Structurality: if r(1 − r) ∈ K×, then, for all x, a, z ∈ X , the pairs (λr xa, λr ax), (µr xz, µr zx) are structural transformations of (X , Γ). Proof. The symmetry relation (1) follows directly from the (x, z)-description. Next we show that Πr coincides with the dilation map from the preceding chapter. Assume first that x⊤a. We show that Πr(x, a, z) = (rP x a + P a x )(z): (rP x a + P a ∃ζ ∈ z : e = rP x a (ζ) + P a x (ζ)o x )(z) =ne ∈ W (cid:12)(cid:12)(cid:12) =ne ∈ W (cid:12)(cid:12)(cid:12) a (ζ) + P a writing ζ = P x symmetry relation (1). ∃α ∈ a, ∃ζ ∈ z, ∃ξ ∈ x : e − rα = ζ − α = ξo = Πr(x, a, z) x (ζ) = α + ξ. For z⊤a, the claim follows now from the (3) With ω = (1 − r)ξ + rζ, it follows for x = z that Πr(x, a, x) ⊂ x. Conversely, we get x ⊂ Πr(x, a, x) by letting α = 0 and ζ = ξ, given ξ ∈ x. The other relations are proved similarly. (2) Under the assumption x⊤a, the claim amounts to the operator identity (rP x a + P a x )(sP x a + P a x ) = (rsP x a + P a x ) which is easily checked. (4) Fix x, a ∈ X , r ∈ K and define the linear subspace r ⊂ W ⊕ W by r := rxa :=(cid:8)(ζ, ω) ∈ W ⊕ W ∃α ∈ a, ∃ξ ∈ x : ω = ζ − (1 − r)α, ζ − α = x(cid:9) Then r∗(z) = {ω ∈ W ∃ζ ∈ z : (ζ, ω) ∈ r} = Πr(x, a, z). On the other hand, by a straightforward change of variables (which is bijective since r is assumed to be invertible), one checks that r∗(z) = {ω ∈ W ∃ζ ∈ z : (ω, ζ) ∈ r} = Πr(a, x, z). Hence (λr xa, λr tions is similar. ax) = (r∗, r∗) is structural. The calculation for the middle multiplica- (cid:3) If r is invertible, then Πr(a, x, z) = Πr−1(x, a, z). Combining with Remarks. 1. (1), we see that Π has the same behaviour under permutations as for the classical cross-ratio. 2. If x⊤a and r ∈ K an arbitrary scalar, we still have structurality in (4). The situation is less clear if x, a, r are all arbitrary. 3. One can define structurality with respect to Πr in the same way as for Γ, by conditions of the form f(cid:0)Πr(x, g(a′), z(cid:1) = Π′ r(f (x), a′, f (z)), Then partial maps of Γ are structural for Πr, and partial maps of Πs are structural for Πr (this property has been used in [Be02] to characterize generalized projective geometries). The proofs are similar to the ones given above. g(cid:0)Π′ r(x′, f (a), z′(cid:1) = Πr(g(x′), a, g(z′)). ASSOCIATIVE GEOMETRIES. I 25 3. Associative geometries In this chapter we give an axiomatic definition of associative geometry, and we show that, at a base point, the corresponding "tangent object" is an associative pair. Conversely, given an associative pair, one can reconstruct an associative ge- ometry. The question whether these constructions can be refined to give a suitable equivalence of categories will be left for future work. 3.1. Axiomatics. Definition. An associative geometry over a commutative unital ring K is given by a set X which carries the following structures: X is a complete lattice (with join denoted by x ∨ y and meet denoted by x ∧ y), and maps (where s ∈ K) such that the following holds. We use the notation Γ : X 5 → X , Πs : X 3 → X , Lxaby(z) := Mxabz(y) := Raybz(x) := Γ(x, a, y, b, z) for the partial maps of Γ, and call x and y transversal, denoted by x⊤y, if x ∧ y = 0 and x ∨ y = 1, and we let Ca := a⊤ := {x ∈ X x⊤a}, Cab := Ca ∩ Cb for sets of elements transversal to a, resp. to a and b. ( 1) The semitorsor property: for all x, y, z, u, v, a, b ∈ X : Γ(Γ(x, a, y, b, z), a, u, b, v) = Γ(x, a, Γ(u, a, z, b, y), b, v) = Γ(x, a, y, b, Γ(z, a, u, b, v)). In other words, for fixed a, b, the product (xyz) := Γ(x, a, y, b, z) turns X into a semitorsor, which will be denoted by Xab. ( 2) Invariance of Γ under the Klein 4-group in (x, a, b, z): for all x, a, y, b, z ∈ X , ( i) Γ(x, a, y, b, z) = Γ(z, b, y, a, x) ( ii) Γ(x, a, y, b, z) = Γ(a, x, y, z, b) In particular, Xba is the opposite semitorsor of Xab. ( 3) Structurality of partial maps: for all x, a, y, b, z ∈ X , the pairs (Lxayb, Lyaxb), (Mxabz, Mzabx), (Raybz, Razby) are structural transformations (see definition below). ( 4) Diagonal values: ( i) for all a, b, y ∈ X , Γ(a, a, y, b, b) = a ∨ b, ( ii) for all a, b, y ∈ X , Γ(a, b, y, a, b) = a ∧ b, ( iii) if x ∈ Cab, then Γ(x, a, x, b, z) = z = Γ(z, b, x, a, x), ( iv) if a⊤x and y⊤b, then Γ(x, a, y, b, b) = b, ( v) if a⊤y and b⊤x, then Γ(x, a, y, b, a) = a. ( 5) The affine space property: for all a ∈ X and r ∈ K, Ca is stable under the dilation map Πr, and Ca becomes an affine space with additive torsor structure and scalar action given for x, y ∈ Ca by x − y + x = Γ(x, a, y, a, z) r ·x y = (1 − r)x + ry = Πr(x, a, y). 26 WOLFGANG BERTRAM AND MICHAEL KINYON ( 6) The semitorsored pairs: for all a, b ∈ X , Γ(Ua, a, Ub, b, Ua) ⊂ Ua, Γ(Ub, a, Ua, b, Ub) ⊂ Ub. Definition. The opposite geometry of an associative geometry (X , ⊤, Γ, Π), de- noted by X op, is X with the same dilation map Π, the opposite quintary product map Γop(x, a, y, b, z) := Γ(z, a, y, b, x) , (which by (4) induces the dual lattice structure) and transversality relation ⊤ de- termined by the lattice structure. A base point in X is a fixed transversal pair (o+, o−), and the dual base point in X is then (o−, o+). Definition. Homomorphisms of associative geometries are maps φ : X → Y such that φ(cid:0)Γ(x, a, y, b, z)(cid:1) = Γ(φx, φa, φy, φb, φz) φ(cid:0)Πr(x, a, y)) = Πr(φx, φa, φy)(cid:1) It is clear that associative geometries over K with their homomorphisms form a category. Antihomomorphisms are homomorphisms into the opposite geometry. Note that, by (4), homomorphisms are in particular lattice homomorphisms, and antihomomorphisms are in particular lattice antihomomorphisms. Involutions are antiautomorphims of order two; they play an important role which will be discussed in subsequent work ( [BeKi09]). For a fixed base point (o+, o−), we define the structure group as the group of automorphisms of X that preserve (o+, o−). An adjoint or structural pair of transformations is a pair g : X → Y, h : Y → X such that g(cid:0)Γ(x, h(u), y, h(v), z)(cid:1) = Γ(g(x), u, g(y), v, g(z)) g(cid:0)Πr(x, h(u), y(cid:1) = Πr(g(x), u, g(y)) and vice versa. Clearly, this also defines a category. 3.2. Consequences. We are going to derive some easy consequences of the axioms. Let us rewrite the semitorsor property in operator form: RaubvLxayb LxaybLzaub = MxabvMuaby = LxaybRaubv = Lx,a,Lybza(u),b = LLxayb(z),a,y,b MΓ(x,a,y,b,z),a,b,v = MxabvLybza = LxaybMzabv Assume that x, y ∈ Uab and z ∈ X . Then, according to (4), Lxaxb = idX = Lybyb, whence LxabbLyaxb = Lx,a,Lybyb(x),b = Lxaxb = idX , and Lxayb : X → X is invertible with inverse (Lxayb)−1 = Lyaxb. By (2), this is equivalent to (Raybx)−1 = Raxby, and in the same way one shows that Mxaby is invertible with inverse (Mxaby)−1 = Mxbay. It follows that Lxayb, Raybx and Mxaby are automorphisms of the geometry. particular, Mxaay and Mxabx are of automorphisms of order two. In ASSOCIATIVE GEOMETRIES. I 27 Proposition 3.1. For all a, b ∈ X , Cab is stable under the ternary map (x, y, z) 7→ Γ(x, a, y, b, z), which turns it into a torsor denoted by Uab. For any y ∈ Uab, the group (Uab, y) acts on X from the left and from the right by the formulas given in Theorem 1.3, and both actions commute. Proof. As remarked above, Lxayb is an automorphism of the geometry. It stabilizes a and b and hence also Ca and Cb. Thus Cab is stable under the ternary map, and the para-associative law and the idempotent law hold by (1) and (4) (iii). The remaining statements follow easily from (1). (cid:3) In part II ([BeKi09]) we will also describe the "Lie algebra" of Uab, thus giving a relatively simple description of the group structure of Uab. -- Next we give the promised conceptual interpretation of Equation (1.5). Lemma 3.2. For all z ∈ Ub, x ∈ Uab, and all y ∈ X , Proof. Using that Rxaxb = idX for a, b ∈ Ux, we have, for all x, z ∈ Ub, Γ(cid:0)x, b, Γ(x, a, y, b, z), b, z(cid:1) = Γ(z, b, a, y, x). Γ(cid:0)x, b, Γ(x, a, y, b, z), b, z(cid:1) = MxbbzMxabz(y) = MbzxbMbzxa(y) = LbzaxRzbxb(y) = Lbzax(y) = Γ(b, z, a, x, y) = Γ(z, b, a, y, x). (cid:3) Since the operator Mxbbz is invertible with inverse Mzbbx, we have, equivalently, Γ(x, a, y, b, z) = Γ(z, b, Γ(z, b, a, y, x), b, x). If a and b are transversal, we may rewrite the lemma in the form (1.5): with b = o−, y = o+: for all x, z ∈ V +, Γ(z, o−, a, o+, x) = Γ(x, o−, Γ(x, a, o+, o−, z), o−, z) = x − Γ(x, a, o+, o−, z) + z. We will see in the following result that Γ(z, o−, a, o+, x) is trilinear in (z, a, x), and hence Γ(x, a, o+, o−, z) is tri-affine in (x, a, z), and both expressions can be considered as geometric interpretations of the associative pair attached to (o+, o−) (see §0.4). More generally, the lemma implies the following analog of Axiom (6): for all b, y ∈ X (transversal or not), the map Ub × Uy × Ub → Ub, (x, a, z) 7→ Γ(x, a, y, b, z) is well-defined and affine in all three variables. 3.3. From geometries to associative pairs. See Appendix B for the notion of associative pair. Theorem 3.3. Let (X , ⊤, Γ, Πr) be an associative geometry over K. i ) Assume X admits a transversal pair, which we take as base point (o+, o−). Then, letting A+ := Uo− and A− := Uo+, the pair of linear spaces (A+, A−) with origins o+, resp. o−, becomes an associative pair when equipped with hxbzi+ := Γ(x, o−, b, o+, z), hayci− := Γ(a, o−, y, o+, c). This construction is functorial (in the "usual" category). 28 WOLFGANG BERTRAM AND MICHAEL KINYON ii ) Assume X admits a transversal triple (a, b, c). Then, letting B := Uc, the K- module B with origin o+ := a becomes an associative unital algebra with unit u := b and product map A × A → A, (x, z) 7→ xz := Γ(x, a, u, c, z). This construction is functorial (in the "usual" category) . Proof. (i) By the "semi-torsored pair axiom" (6), the maps A± × A∓ × A± → A± are well-defined. By restriction from Xo+,o−, they satisfy the para-associative law. They are tri-affine: the proof is exactly the same as the one of Corollary 2.9. Thus it only remains to be shown that they are trilinear, with respect to the origins o± ∈ A±. Let x, z ∈ A+ and b ∈ A−. Then hxbo+i+ = Γ(x, o−, b, o+, o+) = o+, ho+bzi+ = Γ(o+, o−, b, o+, z) = o+ hxo−zi+ = Γ(x, o−, o−, o+, z) = Γ(o−, x, o−, z, o+) = o+. by the Diagonal Value Axiom (4). If φ : X → Y is a base-point preserving ho- momorphism, then restriction of φ yields, by definition of a homomorphism, a pair of K-linear maps A± → (A′)±, which commutes with the product maps Γ, Γ′ and hence is a homomorphism of associative pairs. (ii) With notation from (i), we have xz = hxuzi+, and hence the product is well- defined, bilinear and associative A × A → A. We only have to show that u is a unit element: but this is immediate from xu = Γ(x, a, u, b, u) = x = Γ(u, a, u, b, x) = ux. (cid:3) Example. For any B-module W , the Grassmannian geometry X is an associative geometry, by the results of Chapter 2. For a decomposition W = o+ ⊕ o−, the corre- sponding associative pair is (A+, A−) = (HomB(o+, o−), HomB(o−, o+)), by Theorem 1.7. In case W is a topological module over a topological ring K, we may also work with subgeometries of the whole Grassmannian, such as Grassmannians of closed subspaces with closed complement. For K = R or C, if W is, e.g., a Banach space, the associated associative pair is a pair of spaces of bounded linear operators. Remark. There is a natural definition of structural transformations of associative pairs. They are induced by structural pairs (f, g) satisfying f (o+) = o+, g(o−) = o− and f (A+) ⊂ A+, g(A−) ⊂ A−. With respect to such pairs, the construction obtained from the theory is still functorial. 3.4. From associative pairs to geometries. Theorem 3.4. i ) For every associative pair (A+, A−) there exists an associative geometry X with base point (o+, o−) having (A+, A−) as associated pair. ii ) For every unital associative algebra (A, 1) there exists an associative geometry X with transversal triple (o+, ∆, o−) having (A, 1) as associated algebra. Proof. (ii) Let W = A ⊕ A, o+ the first and o− the second factor and ∆ the diagonal. Then (o+, ∆, o−) is a transversal triple in the Grassmannian geometry X = GrasA(W ), and its associated algebra is A ∼= HomA(A, A) (see the preceding example, with o+ ∼= o− ∼= A). Note that the connected component of o+ can be interpreted as the projective line over A, cf. [BeNe05], [Be08]. ASSOCIATIVE GEOMETRIES. I 29 (i) Consider any algebra imbedding ( A, e) of the pair (A+, A−), for instance, its standard imbedding (see Appendix B). Let e denote the idempotent giving the grading of A and set f = 1 − e. Then A = A00 ⊕ A01 ⊕ A10 ⊕ A11 where A00 = f Af , A01 = A− = f Ae, A10 = A+ = e Af and A11 = e Ae. Let X = GrasA( A) be the Grassmannian of all right ideals in A. As base point in X we choose o+ := e A = A11 ⊕ A10, o− := f A = A00 ⊕ A01 . The associative pair corresponding to (X ; o+, o−) is (see example at the end of the last section) (Hom A(o−, o+), HomA(o+, o−)). But this pair is naturally isomorphic to (A+, A−). Indeed, Hom A(e A, f A) → A01 = f Ae, f 7→ f (e) is K-linear, well-defined (since f (e)e = f (ee) = f (e), f being right A-linear) and has as inverse mapping c 7→ (x 7→ cx), hence is a K-isomorphism. Identifying both pairs of K-modules in this way, a direct check shows that the triple products also coincide, thus establishing the desired isomorphism of associative pairs. (cid:3) Remark. It is of course also possible to see (ii) as a special case of (i). In this case we may work with the algebra imbedding of (A, A) into the matrix algebra A = M(2, 2; A), cf. Appendix B. Remark (Functoriality). Is the construction from the preceding theorem functo- rial, or can it be modified such that it becomes functorial? In the present form, the construction depends on the chosen algebra imbedding and hence is not func- torial (even if we always chose the standard imbedding the construction would not become functoriel, see [Ca04]). However, motivated by corresponding results from Jordan theory ([Be02]), we conjecture that the geometry generated by the connected component depends functorially on the associative pair, thus leading to an equiva- lence of categories between associative pairs and certain associative geometries with base point (whose algebraic properties reflect "connectedness and simple connect- edness"). 4. Further topics (1) Jordan geometries revisited. The present work sheds new light on geometries associated to Jordan algebraic structures: in the same way as associative pairs give rise to Jordan pairs by restricting to the diagonal (Q(x)y = hxyxi; see Appendix B), associative geometries give rise to "Jordan geometries". The new feature is that we get two diagonal restrictions Γ(x, a, y, b, x) and Γ(x, a, y, a, z) which are equivalent. They can be used to give a new axiomatic foundation of "Jordan geometries". Unlike the theory developed in [Be02], this new foundation will be valid also in case of characteristic 2 and hence corresponds to general quadratic Jordan pairs. In this theory, the torsors from the associative theory will be replaced by symmetric spaces (the diagonal (xyx)). 30 WOLFGANG BERTRAM AND MICHAEL KINYON (2) Involutions, Jordan-Lie algebras, classical groups. From a Lie theoretic point of view, the present work deals with classical groups of type An (the "general linear" family). The other classical series (orthogonal, unitary and symplectic families) can be dealt with by adding an involution to an associative geometry. This will be discussed in detail in [BeKi09]. From a more algebraic point of view, this amounts to looking at Jordan-Lie or Lie-Jordan algebras instead of associative pairs (and hence is closely related to (1)), and asking for the geometric counterpart. In [Be08], it is advocated that this might also be interesting in relation with foundational issues of quantum mechanics. (3) Tensor Products. In the associative and in the Jordan-Lie categories, tensor products exist (cf. [Be08] for historical remarks on this subject in relation with foundations of Quantum Mechanics). What is the geometric interpretation of this remarkable fact? (4) Alternative Geometries. The geometric object corresponding to alternative pairs (see [Lo75]) should be a collection of Moufang loops, interacting among each other in a similar way as the torsors Uab do in an associative geometry. (5) Classical projective geometry revisted. The torsors Uab show already up in or- dinary projective spaces, and their alternative analogs will show up in octonion projective planes. It should be interesting to review classical approaches from this point of view. (6) Invariant Theory. The problem of classifying the torsors Uab in a given ge- ometry X is very close to classifying orbits in X × X under the automorphism group. Invariants of torsors ("rank") give rise to invariants of pairs. Similarly, in- variants of groups (Uab, y) give rise to invariants of triples ("rank and signature"), and invariants (conjugacy class) of projective endomorphisms Lxayb to invarinats of quadrupels ("cross-ratio"). (7) Structure theory: ideals and intrinsic subspaces. We ask to translate features of the structure theory of associative pairs and algebras to the level of associative geometries: what are the geometric notions corresponding to left-, right- and inner ideals? See [BeL08] for the Jordan case. (8) Positivity and convexity: case of C ∗-algebras. C ∗-algebras and related triple systems ("ternary rings of operators", see [BM04]) are distinguished among general ones by properties involving "positivity" and "convexity". What is their geometric counterpart on the level of associative geometries? Note that these properties really belong to the involution ∗, so these questions can be seen to fall in the realm of topic (2). Appendix A: torsors and semitorsors Definition. A torsor (G, (·, ·, ·)) is a set G together with a ternary operation G3 → G; (x, y, z) 7→ (xyz) satisfying the identities (G1) and (G2) discussed in the Introduction (§0.2). ASSOCIATIVE GEOMETRIES. I 31 An early term for this notion, due to Prufer, was Schar. This was translated by Suschkewitsch into Russian as grud. This was later somewhat unfortunately translated into English as "heap". Other terms that have been used are "flock" and "herd". B. Schein, in various publications (e.g., [Sch62]) and in private communi- cation, suggested adapting the Russian term directly into English as "groud" (this rhymes with "rude", not "crowd"). In earlier drafts of this paper, we followed his suggestion. However, we found that in talks on this subject, the general audience reaction to the term was negative, and the referee noted that it is hard to pronounce in an English sentence. The other terms noted above are not appropriate, either. In the follow-up [BeKi09] to this paper, we wish to write of "classical" objects just as we speak of classical groups, and "classical heap" or "classical herd", for instance, do not seem suitable. In the end, we decided to go with torsor. This term is usually used in a geo- metric sense to mean principal homogeneous space, and is now generally accepted, largely due to the popularizing efforts of Baez [Ba09]. Using the same term for the equivalent algebraic notion seemed to us a quite reasonable step. For more on the history of the concept, as well as of what we call semitorsors defined below, we refer the reader to the work of Schein, e.g., [Sch62]. In a torsor (G, (·, ·, ·), introduce left-, right- and middle multiplications by (xyz) =: ℓx,y(z) =: ry,z(x) =: mx,z(y) . Then the axioms of a torsor can be rephrased as follows: (G1') (G2') or, in yet another way, (G1") (G2") ℓx,y ◦ ru,v = ru,v ◦ ℓx,y ℓx,x = rx,x = id ℓx,y ◦ ℓz,u = ℓℓx,y(z),u ℓx,y(y) = ry,x(y) = x . Taking y = z in (G1"), and using (G2"), we get what one might call "Chasle's relation" for left translations ℓx,y ◦ ℓy,u = ℓx,u which for u = x shows that the inverse of ℓx,y is ℓy,x. Similarly, we have a Chasle's relation for right translations, and the inverse of rx,y is ry,x. Unusual, compared to group theory, is the role of the middle multiplications. Namely, fixing for the moment a unit e, we have (x(uyw)z) = x(uy−1w)−1z = xw−1yu−1z = ((xwy)uz) = (xw(yuz)) (the para-associative law, cf. relation (G3), Introduction), i.e., (G3') mx,z ◦ mu,w = ℓx,w ◦ ru,z = ru,z ◦ ℓx,w. Taking x = w, resp. u = z, we see that all left and right multiplications can be expressed via middle multiplications: ru,z = mx,z ◦ mu,x, ℓx,w = mx,z ◦ mz,w. 32 WOLFGANG BERTRAM AND MICHAEL KINYON Taking u = z, resp. x = w, we see that mx,z ◦mz,x = id, hence middle multiplications are invertible. In particular m2 x,x = id, which reflects the fact that mx,x is inversion in the group (G, x). Also, (G3') implies that mx,e ◦ mx,e = rx,e ◦ ℓx,e = ℓx,e ◦ (re,x)−1, which means that conjugation by x in the group with unit e is equal to (mx,e)2. Since a torsor can be viewed as an equational class in the sense of universal algebra (G, (·, ·, ·)), all of the usual notions apply. For instance, a homomorphism of torsors is a map φ : G → H such that φ(cid:0)(xyz)(cid:1) = (φ(x)φ(y)φ(z)), and an anti-homomorphism of torsors is a homomorphism to the opposite torsor (same set with product (x, y, z) 7→ (zyx)). Homomorphisms enjoy similar properties as usual affine maps. It is easily proved that left and right multiplications are automorphisms (called inner ), whereas middle multiplications are inner anti-automorphisms. Other notions, such as subtorsors, products, congruences and quotients follow standard patterns. Definition. A semitorsor (G, (·, ·, ·)) is a set G with a ternary operation G3 → G; (x, y, z) 7→ (xyz) satisfying the para-associative law (G3) from the Introduction. The basic example is the symmetric semitorsor on sets A and B, the set of all relations between A and B with (rst) = r ◦ s−1 ◦ t, where ◦ is the composition of relations. Clearly, fixing the middle element in a semitorsor gives rise to a semigroup; but, in contrast to the case of groups, not all semigroups are obtained in this way. For more on semitorsors, see, e.g. [Sch62] and the references therein. Appendix B: Associative pairs Definition. An associative pair (over K) is a pair (A+, A−) of K-modules together with two trilinear maps h·, ·, ·i± : A± × A∓ × A∓ → A± such that hxyhzuvi±i± = hhxyzi±uvi± = hxhuzyi∓vi±. Note that we follow here the convention of Loos [Lo75]. Other authors (e.g. [MMG]) use a modified identity, replacing the last term by hxhyzui∓vi±. But both versions i− by the trilinear map (x, y, z) 7→ hz, y, xi−. are equivalent: it suffices to replace h We prefer the definition given by Loos since it takes the same form as the para- associative law in a semitorsor. We should mention, however, that for associative triple systems, i.e., K-modules A with a trilinear map A3 → A, (x, y, z) 7→ hxyzi these two versions of the defining identity have to be distinguished, leading to two different kinds of associative triple systems ("ternary rings", cf. [Li71], and asso- ciative triple systems [Lo72]; all this is best discussed in the context of associative pairs, resp. geometries, with involution, see topic (2) in Chapter 4 and [BeKi09].) In any case, for fixed a ∈ A−, A+ with is an associative algebra, called the a-homotope and denoted by A+ a . x ·a y := hxayi ASSOCIATIVE GEOMETRIES. I 33 Examples of associative pairs. (1) Every associative algebra A gives rise to an associative pair A+ = A− = A via hxyzi+ = xyz, hxyzi− = zyx. (2) For K-modules E and F , let A+ = Hom(E, F ), A− = Hom(F, E), hXY Zi+ = X ◦ Y ◦ Z hXY Zi− = Z ◦ Y ◦ X. (3) Let A be an associative algebra with unit 1 and idempotent e and f := 1 − e. Let A = f Af ⊕ f Ae ⊕ e Ae ⊕ e Af = A00 ⊕ A01 ⊕ A11 ⊕ A10 with Aij = {x ∈ A ex = ix, xe = jx} the associated Peirce decomposition. Then (A+, A−) := (A01, A10), hxyzi+ := xyz, hxyzi− := zyx is an associative pair. The standard imbedding. It is not difficult to show that every associative pair arises from an associative algebra A with idempotent e in the way just described (see [Lo75], Notes to Chapter II). We call this an algebra imbedding for (A+, A−). There are several such imbeddings (see [Ca04] for a comparison of some of them). Among these is a minimal choice called the standard imbedding of the associative pair. For instance, in Example (2) we may take A = End(E ⊕ F ) with e the projector onto E along F (but this choice will in general not be minimal). In Example (1), take A := EndA(A ⊕ A) = M(2, 2; A) and e the projector onto the first factor. The associated Jordan pair. Formally, associative pairs give rise to Jordan pairs in exactly the same way as torsors give rise to symmetric spaces: the Jordan pair is (V +, V −) := (A+, A−) with the quadratic map Q±(x)y := hxyxi± and its polarized version T ±(x, y, z) := Q±(x + z)y − Q±(x)y − Q±(z)y = hxyzi± + hzyxi±. Associative pairs with invertible elements. We call x ∈ A± invertible if Q(x) : A∓ → A±, y 7→ hxyxi is an invertible operator. As shown in [Lo75], associative pairs with invertible elements correspond to unital associative algebras: namely, x is invertible if and only if the algebra Ax has a unit (which is then x−1 := Q(x)−1x). References J. Baez, Torsors made easy, http://math.ucr.edu/home/baez/torsors.html. [Ba09] [Be02] W. Bertram, Generalized projective geometries: general theory and equivalence with Jordan structures, Adv. Geom. 2 (2002), 329 -- 369 (electronic version: preprint 90 at http://homepage.uibk.ac.at/~c70202/jordan/index.html). [Be03] W. Bertram, The geometry of null systems, Jordan algebras and von Staudt's Theorem, Ann. Inst. Fourier 53 (2003) fasc. 1, 193 -- 225. [Be04] W. Bertram, From linear algebra via affine algebra to projective algebra, Linear Algebra and its Applications 378 (2004), 109 -- 134. [Be08b] W. Bertram, Homotopes and conformal deformations of symmetric spaces. J. Lie Theory 18 (2008), 301 -- 333; arXiv: math.RA/0606449. 34 WOLFGANG BERTRAM AND MICHAEL KINYON [Be08] W. Bertram, On the Hermitian projective line as a home for the geometry of quantum theory. In Proceedings XXVII Workshop on Geometrical Methods in Physics, Bia lowieza 2008 ; arXiv: math-ph/0809.0561. [Be10] W. Bertram, Jordan structures and non-associative geometry. In Trends and Develop- ments in Infinite Dimensional Lie Theory, Progress in Math., Birkhaeuser, to appear 2010; arXiv: math.RA/0706.1406. [BeKi09] W. Bertram and M. Kinyon, Associative Geometries. II: Involutions, the classical tor- sors, and their homotopes, J. Lie Theory, to appear; arXiv: 0909.4438. [BeL08] W. Bertram and H. Loewe, Inner ideals and intrinsic subspaces, Adv. in Geometry 8 (2008), 53 -- 85; arXiv: math.RA/0606448. [BeNe05] W. Bertram and K.-H. Neeb, Projective completions of Jordan pairs. II: Manifold structures and symmetric spaces, Geom. Dedicata 112 (2005), 73 -- 113; arXiv: math.GR/0401236. [BM04] D. P. Blecher and C. Le Merdy, Operator algebras and their modules -- an operator space [Cer43] [Ca04] approach, Clarendon Press, Oxford, 2004. J. Certaine, The ternary operation (abc) = ab−1c of a group, Bull. Amer. Math. Soc. 49 (1943), 869 -- 877. I. de las Penas Cabrera, A note on the envelopes of an associative pair, Comm. Algebra 32 (2004), 4133 -- 4140. [Li71] W. G. Lister, Ternary rings, Trans. Amer. Math. Soc. 154 (1971), 37 -- 55. [Lo69] [Lo72] [Lo75] [MMG] O. Loos, Symmetric Spaces I, Benjamin, New York, 1969. O. Loos, Assoziative Tripelsysteme, Manuscripta Math. 7 (1972), 103 -- 112. O. Loos, Jordan Pairs, Lecture Notes in Math. 460, Springer, New York, 1975. J. A. Cuenca Mira, A. Garc`ıa Mart`ın and C. Mart`ın Gonz´alez, Jacobson density for associative pairs and its applications, Comm. Algebra 17 (1989), 2595 -- 2610. [PR09] R. Padmanabhan and S. Rudeanu, Axioms for Lattices and Boolean Algebras, World Scientific, 2009. [Sch62] B. Schein, On the theory of inverse semigroups and generalized grouds, in Twelve papers [Va66] in logic and algebra, AMS Translations, Ser. 2, 113, 1979, pp. 89 -- 123. V. V. Vagner, On the algebraic theory of coordinate atlases. (Russian) Trudy Sem. Vektor. Tenzor. Anal. 13 (1966), 510 -- 563. Institut ´Elie Cartan Nancy, Nancy-Universit´e, CNRS, INRIA,, Boulevard des Aiguillettes, B.P. 239,, F-54506 Vandoeuvre-l`es-Nancy, France E-mail address: [email protected] Department of Mathematics, University of Denver, 2360 S Gaylord St, Denver, Colorado 80208 USA E-mail address: [email protected]
1612.00252	2	1612	2017-06-21T12:48:14	Disjoint-union partial algebras	[ "math.RA", "cs.LO", "math.LO" ]	Disjoint union is a partial binary operation returning the union of two sets if they are disjoint and undefined otherwise. A disjoint-union partial algebra of sets is a collection of sets closed under disjoint unions, whenever they are defined. We provide a recursive first-order axiomatisation of the class of partial algebras isomorphic to a disjoint-union partial algebra of sets but prove that no finite axiomatisation exists. We do the same for other signatures including one or both of disjoint union and subset complement, another partial binary operation we define. Domain-disjoint union is a partial binary operation on partial functions, returning the union if the arguments have disjoint domains and undefined otherwise. For each signature including one or both of domain-disjoint union and subset complement and optionally including composition, we consider the class of partial algebras isomorphic to a collection of partial functions closed under the operations. Again the classes prove to be axiomatisable, but not finitely axiomatisable, in first-order logic. We define the notion of pairwise combinability. For each of the previously considered signatures, we examine the class isomorphic to a partial algebra of sets/partial functions under an isomorphism mapping arbitrary suprema of pairwise combinable sets to the corresponding disjoint unions. We prove that for each case the class is not closed under elementary equivalence. However, when intersection is added to any of the signatures considered, the isomorphism class of the partial algebras of sets is finitely axiomatisable and in each case we give such an axiomatisation.	math.RA	math	Logical Methods in Computer Science Vol. 13(2:10)2017, pp. 1–31 https://lmcs.episciences.org/ Submitted Dec. 07, 2016 Published Jun. 22, 2017 DISJOINT-UNION PARTIAL ALGEBRAS ROBIN HIRSCH AND BRETT MCLEAN Department of Computer Science, University College London, Gower Street, London WC1E 6BT e-mail address: {r.hirsch,b.mclean}@ucl.ac.uk Abstract. Disjoint union is a partial binary operation returning the union of two sets if they are disjoint and undefined otherwise. A disjoint-union partial algebra of sets is a collection of sets closed under disjoint unions, whenever they are defined. We provide a recursive first-order axiomatisation of the class of partial algebras isomorphic to a disjoint- union partial algebra of sets but prove that no finite axiomatisation exists. We do the same for other signatures including one or both of disjoint union and subset complement, another partial binary operation we define. Domain-disjoint union is a partial binary operation on partial functions, returning the union if the arguments have disjoint domains and undefined otherwise. For each signature including one or both of domain-disjoint union and subset complement and optionally including composition, we consider the class of partial algebras isomorphic to a collection of partial functions closed under the operations. Again the classes prove to be axiomatisable, but not finitely axiomatisable, in first-order logic. We define the notion of pairwise combinability. For each of the previously considered signatures, we examine the class isomorphic to a partial algebra of sets/partial functions under an isomorphism mapping arbitrary suprema of pairwise combinable sets to the corresponding disjoint unions. We prove that for each case the class is not closed under elementary equivalence. However, when intersection is added to any of the signatures considered, the isomorphism class of the partial algebras of sets is finitely axiomatisable and in each case we give such an axiomatisation. 1. Introduction Sets and functions are perhaps the two most fundamental and important types of object in all mathematics. Consequently, investigations into the first-order properties of collections of such objects have a long history. Boole, in 1847, was the first to focus attention directly on the algebraic properties of sets [2]. The outstanding result in this area is the Birkhoff- Stone representation theorem, completed in 1934, showing that boolean algebra provides a first-order axiomatisation of the class of isomorphs of fields of sets [18]. For functions, the story starts around the same period, as we can view Cayley's theorem of 1854 as proof that the group axioms are in fact an axiomatisation of the isomorphism class of collections of bijective functions, closed under composition and inverse [5]. Schein's survey article of 1970 contains a summary of the many similar results about algebras of partial functions that were known by the time of its writing [17]. l IN COMPUTER SCIENCE LOGICAL METHODS DOI:10.23638/LMCS-13(2:10)2017 c(cid:13) Robin Hirsch and Brett McLean CC(cid:13) Creative Commons 2 ROBIN HIRSCH AND BRETT MCLEAN The past fifteen years have seen a revival of interest in algebras of partial functions, with results finding that such algebras are logically and computationally well behaved [10, 13, 11, 12, 9, 14]. In particular, algebras of partial functions with composition, intersection, domain and range have the finite representation property [15]. Separation logic is a formalism for reasoning about the state of dynamically-allocated computer memory [16]. In the standard 'stack-and-heap' semantics, dynamic memory states are modelled by (finite) partial functions. Thus statements in separation logic are statements about partial functions. The logical connective common to all flavours of separation logic is the separating conjunction ∗. In the stack-and-heap semantics, the formulas are evaluated at a given heap (a partial function, h) and stack (a variable assignment, s). In this semantics h, s = ϕ ∗ ψ if and only if there exist h1, h2 with disjoint domains, such that h = h1 ∪ h2 and h1, s = ϕ and h2, s = ψ. So lying behind the semantics of the separating conjunction is a partial operation on partial functions we call the domain-disjoint union, which returns the union when its arguments have disjoint domains and is undefined otherwise. Another logical connective that is often employed in separation logic is the separating implication and again a partial operation on partial functions lies behind its semantics. Separation logic has enjoyed and continues to enjoy great practical successes [1, 4]. However Brotherston and Kanovich have shown that, for propositional separation logic, the validity problem is undecidable for a variety of different semantics, including the stack-and- heap semantics [3]. The contrast between the aforementioned positive results concerning algebras of partial functions and the undecidability of a propositional logic whose semantics are based on partial algebras of partial functions, suggests a more detailed investigation into the computational and logical behaviour of collections of partial functions equipped with the partial operations arising from separation logic. In this paper we examine, from a first-order perspective, partial algebras of partial functions over separation logic signatures-signatures containing one or more of the partial operations underlying the semantics of separation logic. Specifically, we study, for each signature, the isomorphic closure of the class of partial algebras of partial functions. Because these partial operations have not previously been studied in a first-order context we also include an investigation into partial algebras of sets over these signatures. In Section 2 we give the definitions needed to precisely define these classes of partial algebras. In Section 3 we show that each of our classes is first-order axiomatisable and in Section 4 we give a method to form recursive axiomatisations that are easily understandable as statements about certain two-player games. In Section 5 we show that though our classes are axiomatisable, finite axiomatisations do not exist. In Section 6 we show that when ordinary intersection is added to the previously examined signatures, the classes of partial algebras become finitely axiomatisable. In Section 7 we examine decidability and complexity questions and then conclude with some open problems. 2. Disjoint-union Partial Algebras In this section we give the fundamental definitions that are needed in order to state the results contained in this paper. We first define the partial operations that we use. DISJOINT-UNION PARTIAL ALGEBRAS 3 Definition 2.1. Given two sets S and T the disjoint union S S ∩ T = ∅, else it is undefined. The subset complement S it is undefined. •∪ T equals S ∪ T if •\ T equals S \ T if T ⊆ S, else •∪ T = U if and only if U •\ S = T . Observe that S The next definition involves partial functions. We take the set-theoretic view of a function as being a functional set of ordered pairs, rather than requiring a domain and codomain to be explicitly specified also. In this sense there is no notion of a function being 'partial'. But using the word partial serves to indicate that when we have a set of such functions they are not required to share a common domain (of definition)-they are 'partial functions' on (any superset of) the union of these domains. Definition 2.2. Given two partial functions f and g the domain-disjoint union f • (cid:94) g equals f ∪ g if the domains of f and g are disjoint, else it is undefined. The symbol denotes the total operation of composition. Observe that if the domains of two partial functions are disjoint then their union is a partial function. So domain-disjoint union is a partial operation on partial functions. If f and g are partial functions with g ⊆ f then f \ g is also a partial function. Hence subset complement gives another partial operation on partial functions. of separation logic, which we now detail precisely. The reason for our interest in these partial operations is their appearance in the semantics The separating conjunction ∗ is a binary logical connective present in all forms of separation logic. As mentioned in the introduction, in the stack-and-heap semantics the formulas are evaluated at a given heap (a partial function, h) and stack (variable assignment, s). In this semantics h, s = ϕ ∗ ψ if and only if there exist h1, h2 such that h = h1 • (cid:94) h2 and both h1, s = ϕ and h2, s = ψ. h, s = emp if and only if h = ∅. The constant emp also appears in all varieties of separation logic. The semantics is The separating implication −∗ is another binary logical connective common in separ- •\ h1 ation logic. The semantics is h, s = ϕ −∗ ψ if and only if for all h1, h2 such that h = h2 we have h1, s = ϕ implies h2, s = ψ. Because we are working with partial operations, the classes of structures we will examine are classes of partial algebras. Definition 2.3. A partial algebra A = (A, (Ωi)i<β) consists of a domain, A, together with a sequence Ω0, Ω1, . . . of partial operations on A, each of some finite arity α(i) that should be clear from the context. Two partial algebras A = (A, (Ωi)i<β) and B = (B, (Πi)i<β) are similar if for all i < β the arities of Ωi and Πi are equal. (So in particular A and B must have the same ordinal indexing their partial operations.) We use the word 'signature' flexibly. Depending on context it either means a sequence of symbols, each with a prescribed arity and each designated to be a function symbol, a partial function symbol or a relation symbol. Or, it means a sequence of actual operations/partial operations/relations. Definition 2.4. Given two similar partial algebras A = (A, (Ωi)i<β) and B = (B, (Πi)i<β), a map θ : A → B is a partial-algebra homomorphism from A to B if for all i < β and 4 ROBIN HIRSCH AND BRETT MCLEAN all a1, . . . , aα(i) ∈ A the value Ωi(a1, . . . , aα(i)) is defined if and only if Πi(θ(a1), . . . , θ(aα(i))) is defined, and in the case where they are defined we have θ(Ωi(a1, . . . , aα(i))) = Πi(θ(a1), . . . , θ(aα(i))) . If θ is surjective then we say B is a partial-algebra homomorphic image of A. A partial-algebra embedding is an injective partial-algebra homomorphism. An isomorph- ism is a bijective partial-algebra homomorphism. We are careful never to drop the words 'partial-algebra' when referring to the notions defined in Definition 2.4, since a bald 'homomorphism' is an ambiguous usage when speaking of partial algebras-at least three differing definitions have been given in the literature. What we call a partial-algebra homomorphism, Gratzer calls a strong homomorphism [7, Chapter 2]. Given a partial algebra A, when we write a ∈ A or say that a is an element of A, we mean that a is an element of the domain of A. While total algebras are by convention nonempty, we make the choice, for reasons of convenience, to allow partial algebras to be empty. When we want to refer to a signature consisting of a single symbol we will often abuse notation by using that symbol to denote the signature. We write ℘(X) for the power set of a set X. •\,∅}. A partial Definition 2.5. Let σ be a signature whose symbols are members of { •∪, σ-algebra of sets, A, with domain A, consists of a subset A ⊆ ℘(X) (for some base set X), closed under the partial operations in σ, wherever they are defined, and containing the •∪) is called a disjoint-union empty set if ∅ is in the signature. The particular case of σ = ( •∪,∅) is a disjoint-union partial algebra of sets partial algebra of sets and the case σ = ( with zero. •\,,∅}. A partial Definition 2.6. Let σ be a signature whose symbols are members of { • (cid:94), σ-algebra of partial functions, A consists of a set of partial functions closed under the partial and total operations in σ, wherever they are defined, and containing the empty set if ∅ is in the signature. The base of A is the union of the domains and codomains of all the partial functions in A. •\,∅}. A σ- Definition 2.7. Let σ be a signature whose symbols are members of { •∪, representation by sets of a partial algebra is an isomorphism from that partial algebra •∪) is called a disjoint-union to a partial σ-algebra of sets. The particular case of σ = ( representation (by sets). •\,,∅}. A σ- Definition 2.8. Let σ be a signature whose symbols are members of { • (cid:94), representation by partial functions of a partial algebra is an isomorphism from that partial algebra to a partial σ-algebra of partial functions. For a partial algebra A and an element a ∈ A, we write aθ for the image of a under a representation θ of A. We will be consistent about the symbols we use for abstract (partial) operations-those in the partial algebras being represented-employing them according to DISJOINT-UNION PARTIAL ALGEBRAS 5 the correspondence indicated below. • (cid:94) •(cid:116) (cid:32) •∪ or •− (cid:32) •\ ; (cid:32) 0 (cid:32) ∅ For each notion of representability we are interested in the associated representation class-the class of all partial algebras having such a representation. It is usually clear whether we are talking about a representation by sets or a representation by partial functions. For example if the signature contains (cid:94) we must be talking of partial functions. However, as part (1) of the next proposition shows, for the partial operations we are considering, representability by sets and representability by partial functions are the same thing. •∪ we must be talking of sets and if it contains • Proposition 2.9. (1) Let σ be a signature whose symbols are a subset of { •∪, formed by replacing sets if and only if it is σ(cid:48)-representable by partial functions. •∪ (if present) by • •(cid:116), ;, 0)-algebra. If the ( •\,∅} and let σ(cid:48) be the signature (cid:94) in σ. A partial algebra is σ-representable by •∪,∅)-representable and •(cid:116), 0)-reduct of A is ( (2) Let A be a partial ( A validates a ; b = 0, then A is ( • (cid:94),,∅)-representable. Proof. For part (1), let σ be one of the signatures in question and let A be a partial algebra. Suppose θ is a σ-representation of A by sets over base X. Then the map ρ defined by aρ = {(x, x) x ∈ aθ} is easily seen to be a σ(cid:48)-representation of A by partial functions. Conversely, suppose ρ is a σ(cid:48)-representation of A by partial functions over base X. Let Y be a disjoint set of the same cardinality as X and let f : X → Y be any bijection. Define θ by aθ = aρ ∪ {(f (x), f (x)) x ∈ dom(aρ)}. Then it is easy to see that θ is another σ(cid:48)- representation of A by partial functions. By construction, θ has the property that any aθ and bθ have disjoint domains if and only if they are disjoint. Hence θ is also a σ-representation of A by sets. •(cid:116), 0)-reduct of A over base set X. For part (2), let θ be a ( Let Y be a disjoint set of the same cardinality as X and let f : X → Y be any bijection. (cid:94),,∅)-representation of A. The map ρ defined by aρ = f(cid:22) Remark 2.10. In each of the following cases let the signature σ∅ be formed by the addition of ∅ to σ. •∪. A partial algebra A is σ-representable if and only its • Let σ be a signature containing • Let σ be a signature containing • reduct to the signature without 0 is σ∅-representable and A satisfies 0 •∪,∅)-representation of the ( aθ is easily seen to be a ( • reduct to the signature without 0 is σ∅-representable and A satisfies 0 Hence axiomatisations of representation classes for signatures without ∅ would immediately yield axiomatisations for the case including ∅ also. reduct to the signature without 0 is σ∅-representable and A satisfies 0 (cid:94). A partial algebra A is σ-representable if and only its •\. A partial algebra A is σ-representable if and only its • Let σ be a signature containing •(cid:116) 0 = 0. •(cid:116) 0 = 0. •− 0 = 0. 6 ROBIN HIRSCH AND BRETT MCLEAN •(cid:116) c is defined and a We now define a version of complete representability. For a partial ( •(cid:116), . . .)-algebra A, define a relation (cid:46) over A by letting a (cid:46) b if and only if either a = b or there is c ∈ A such •∪, . . .)-representable, •(cid:116) c = b. By definition, (cid:46) is reflexive. If A is ( that a •(cid:116) c is defined then by elementary properties of sets, it is necessarily the case that if (a •(cid:116) c) is also defined and equal to it, which is precisely what is required to see then a that (cid:46) is transitive. Antisymmetry of (cid:46) also follows by elementary properties of sets. Hence (cid:46) is a partial order. •(cid:116) (b •(cid:116) b) •(cid:116), . . .)-algebra A is pairwise combinable if for Definition 2.11. A subset S of a partial ( •∪, . . .)-representation θ of A is (cid:46)-complete if •(cid:116) t is defined. A ( all s (cid:54)= t ∈ S the value s for any pairwise-combinable subset S of A with a supremum a (with respect to the order (cid:46)) we have aθ =(cid:83) s∈S sθ. •(cid:116), . . .)-algebra then every ( •∪, . . .)-representation Proposition 2.12. If A is a finite partial ( of A is (cid:46)-complete. •∪ tθ is defined for all s (cid:54)= t ∈ S. Then by the definition of •∪, . . .)-representation of A and let S ⊆ A be pairwise combinable Proof. Let θ be a ( •∪, . . .)-representation, we with supremum a. As S is pairwise combinable and θ is a ( •∪, the set have that sθ {sθ s ∈ S} is pairwise disjoint. As A is finite, S must be too, so S = {s1, . . . , sn}, say. By •(cid:116) . . . i=1 sθ induction, for each k we have that s1 i . •(cid:116) . . . It is clear that for any b1, b2 ∈ A the implication s∈S sθ. Hence (s1 b1 (cid:46) b2 =⇒ bθ i and must be a subset •(cid:116) sn is clearly an upper bound for S so we of bθ for any upper bound b of S. But s1 conclude that aθ = (s1 2 holds. Therefore aθ must be a superset of each sθ •(cid:116) sn)θ = (cid:83) •(cid:116) sk)θ =(cid:83)k •(cid:116) . . . s∈S sθ as required. •(cid:116) sn)θ =(cid:83) •(cid:116) sk is defined and (s1 1 ⊆ bθ •(cid:116) . . . •(cid:116) . . . Finally, a word about logic. In our meta-language, that is, English, we can talk in terms of partial operations and partial algebras, which is what we have been doing so far. However, the traditional presentation of first-order logic does not include partial function symbols. Hence, in order to examine the first-order logic of our partial algebras we must view them formally as relational structures. •(cid:116) over A we •(cid:116) b is defined and •(cid:116)) be a partial algebra. From the partial binary operation may define a ternary relation J over A by letting J(a, b, c) if and only if a equal to c. Since a partial operation is (at most) single valued, we have Let A = (A, J(a, b, c) ∧ J(a, b, d) → c = d. (2.1) •(cid:116) over A by letting a Conversely, given any ternary relation J over A satisfying (2.1), we may define a partial •(cid:116) b be defined if and only if there exists c such that J(a, b, c) operation •(cid:116) b = c. The definition of J from holds (unique, by (2.1)) and when this is the case we let a •− is in the signature we •(cid:116) and the definition of can define a corresponding ternary relation K in the same way. •(cid:116) from J are clearly inverses. Similarly, if •(cid:116) nor To remain in the context of classical first-order logic we adopt languages that feature •− but have ternary relation symbols J and/or K as appropriate (as well as •(cid:116) b as an abbreviation of the neither equality). In the relational language L(J), we may write ∃a formula ∃cJ(a, b, c) and write a •(cid:116) b = c in place of J(a, b, c). Similarly for •− and K. DISJOINT-UNION PARTIAL ALGEBRAS 7 3. Axiomatisability In this section we show there exists a first-order L(J)-theory that axiomatises the class J of •∪-representations. Hence J, viewed as a class of L(J)-structures, is partial •\-representations (as •(cid:116)-algebras with elementary. We do the same for the class K of partial •−)-algebras with ( •(cid:116), •−-algebras with •\)-representations. •∪, sets) and the class L of partial ( Definition 3.1. If A1 ⊆ A2 are similar partial algebras and the inclusion map is a partial- algebra embedding then we say that A1 is a partial-subalgebra of A2. Let Ai = (Ai, Ω0, . . .) be partial algebras, for i ∈ I, and let U be an ultrafilter over I. The ultraproduct Πi∈I Ai/U •(cid:116) [(bi)i∈I ] (where ai, bi ∈ Ai is defined in the normal way, noting that, for example, [(ai)i∈I ] •(cid:116) bi is defined in Ai} ∈ for all i ∈ I) is defined in the ultraproduct if and only if {i ∈ I ai U . Ultrapowers and ultraroots also have their normal definitions: an ultrapower is an ultraproduct of identical partial algebras and A is an ultraroot of B if B is an ultrapower of A. It is almost trivial that the class of It is clear that a partial-subalgebra of A is always a substructure of A, as relational structures, and also that any substructure of A is a partial algebra, that is, satisfies (2.1). However, in order for a relational substructure of A to be a partial-subalgebra it is necessary that it be closed under the partial operations, wherever they are defined in A. •∪-representable partial algebras is closed under partial-subalgebras. This class is not however closed under substructures. Indeed it is •(cid:116)-algebra A with a disjoint-union representation but where an easy to construct a partial L(J)-substructure of A has no disjoint-union representation. We give an example now. Example 3.2. The collection ℘{1, 2, 3} of sets forms a disjoint-union partial algebra of sets and so is trivially a •(cid:116)-algebra, if we identify •∪-representable partial The substructure with domain ℘{1, 2, 3} \ {1, 2, 3} is not •∪-representable, because {1} •(cid:116) {2},{2} •(cid:116) {3} and {3} •(cid:116) {1} all exist, so {1},{2},{3} would have to be represented by pairwise disjoint sets. But then {1, 2} •(cid:116) {3} would have to exist, which is not the case. •(cid:116) with •∪. We obtain the following corollary. Corollary 3.3. The isomorphic closure of the class of disjoint-union partial algebras of sets is not axiomatisable by a universal first-order L(J)-theory. Returning to our objective of proving that the classes J, K and L are elementary, this could be achieved by showing that they are closed under ultraproducts and ultraroots. However this is not entirely straightforward, since many of the relevant model-theoretic results are known for total operations only. Instead, to apply these known results, we first describe a way to view an arbitrary partial algebra as a total algebra. Then, having established elementarity of the resulting class of total algebras we describe how to convert back to an axiomatisation of the partial algebras. Definition 3.4. Let A = (A, (Ωi)i<β) be a partial algebra. The totalisation of A is the algebra A+ = (A ∪ {∞},∞, (Ωi)i<β), where ∞ (cid:54)∈ A and for each i the interpretation of Ωi in A+ agrees with the interpretation in A whenever the latter is defined, and in all other cases returns ∞. The totalisation of a class C of similar partial algebras is the class C+ = {A+ A ∈ C} of total algebras. 8 ROBIN HIRSCH AND BRETT MCLEAN Inversely to totalisation, suppose we have a total algebra B = (B,∞, (Ωi)i<β) where for each i < β, if any element of the α(i)-tuple ¯b is ∞ then Ωi(¯b) = ∞. Then we may define a partial algebra B− = (B \ {∞}, (Ωi)i<β) where each Ωi(b1, . . . , bn) is defined in B− if and only if Ωi(b1, . . . , bn) (cid:54)= ∞ in B, in which case it has the same value as in B. Clearly for any partial algebra A we have (A+)− = A and for any total algebra B with a suitable ∞ ∈ B we have (B−)+ = B. In the following, we show that each of the classes J+, K+ and L+ is closed under both •−) •−) respectively. We then give a translation from the universal formulas defining ultraproducts and subalgebras and hence is universally axiomatisable in L(∞, and L(∞, J+ to a set of L(J)-formulas that defines J. Similarly for the other two cases. •(cid:116)), L(∞, •(cid:116), copies of all symbols of L, We will be using the notion of pseudoelementarity, and since there are various possible equivalent definitions of this, we state the one we wish to use. It can be found, for example, as [8, Definition 9.1]. Definition 3.5. Given an unsorted first-order language L, a class C of L-structures is pseudoelementary if there exist • a two-sorted first-order language L(cid:48), with sorts algebra and base, containing algebra-sorted • an L(cid:48)-theory T , such that C = {M algebra(cid:22)L M = T}. Lemma 3.6. The class J+ is universally axiomatisable in L(∞, axiomatisable in L(∞, Proof. We start with J+. By definition, J+ is closed under isomorphism. We first show that J+ is pseudoelementary, hence also closed under ultraproducts. •−) and the class L+ is universally axiomatisable in L(∞, •(cid:116)), the class K+ is universally •−). •(cid:116), Consider a two-sorted language, with an algebra sort and a base sort. The signature •(cid:116) on the algebra sort, an algebra-sorted constant ∞ and a consists of a binary operation binary predicate ∈, written infix, of type base × algebra. Consider the formulas •(cid:116) ∞ = ∞ •(cid:116) a = ∞ a (a (cid:54)= b) ∧ (a (cid:54)= ∞) ∧ (b (cid:54)= ∞) → ∃x((x ∈ a ∧ x (cid:54)∈ b) ∨ (x (cid:54)∈ a ∧ x ∈ b)) •(cid:116) b = ∞) ↔ ∃x(x ∈ a ∧ x ∈ b)) (a (cid:54)= ∞) ∧ (b (cid:54)= ∞) → ((a •(cid:116) b (cid:54)= ∞) → ((x ∈ a •(cid:116) b) ↔ (x ∈ a ∨ x ∈ b)) (a where a, b, c are algebra-sorted variables and x is a base-sorted variable. These formulas merely state that the base-sorted elements form the base of a repres- entation of the non-∞ elements of algebra sort and that ∞ behaves as it should for an •(cid:116))-reducts of restrictions of models of the algebra in J+. Hence J+ is the class of (∞, formulas to algebra-sorted elements, that is, J+ is pseudoelementary. Hence J+ is closed under ultraproducts. •(cid:116)) and there is no quantification of algebra-sorted variables, J+ is closed under substructures. A consequence of this is that J+ is closed under ultraroots, by the simple observation that the diagonal map embeds any ultraroot into its ultrapower. •(cid:116), in our defining formulas is already in L(∞, Since the only function symbol, We now know that J+ is closed under isomorphism, ultraproducts and ultraroots. This is a well-known algebraic characterisation of elementarity (for example see [6, Theorem 6.1.16]). DISJOINT-UNION PARTIAL ALGEBRAS 9 Then as J+ is elementary and closed under substructures it is universally axiomatisable, by the (cid:32)Lo´s-Tarski preservation theorem. For K+ and L+ the same line of reasoning applies. Each is by definition closed under isomorphism. For K+ we show pseudoelementarity and closure under substructures using the formulas •− ∞ = ∞ •− a = ∞ a (a (cid:54)= b) ∧ (a (cid:54)= ∞) ∧ (b (cid:54)= ∞) → ∃x((x ∈ a ∧ x (cid:54)∈ b) ∨ (x (cid:54)∈ a ∧ x ∈ b)) •− b = ∞) ↔ ∃x(x (cid:54)∈ a ∧ x ∈ b)) (a (cid:54)= ∞) ∧ (b (cid:54)= ∞) → ((a •− b (cid:54)= ∞) → ((x ∈ a •− b) ↔ (x ∈ a ∧ x (cid:54)∈ b)) (a and for L+ we do the same using the union of the formulas for J+ and the formulas for K+. Proposition 3.7. Let C be a class of partial algebras of the signature (Ωi)i<β, which we view as relational structures over the signature (Ri)i<β, where for each i the arity of Ri is one greater than that of Ωi. Suppose C+ is universally axiomatisable in the language L(∞, (Ωi)i<β). Then C is axiomatisable in the language L((Ri)i<β). Proof. Let Σ+ be a universal axiomatisation of C+ in the language L(∞, (Ωi)i<β). Since it is the validity of all the formulas in Σ+ that defines C+ we may assume that each axiom in Σ+ is quantifier free. We define a translation − from L(∞, (Ωi)i<β) to L((Ri)i<β) such that (3.1) A+ = ψ ⇐⇒ A = ψ− for any nonempty partial algebra A of the signature (Ωi)i<β and any quantifier-free L(∞, (Ωi)i<β)-formula ψ. Let V (ψ) be the finite set of variables occurring in ψ and let S(ψ) be the set of subterms of ψ. We may also write V (t) and S(t) to denote the set of all variables and subterms of the term t. For any assignment ρ : V (ψ) → A+ and t ∈ S(ψ) let [t]ρ denote the evaluation of t under ρ in A+. Let v be any injective mapping from S(ψ) to our set of first-order variables, mapping the term s ∈ S(ψ) to the variable vs and satisfying vu = u for all u ∈ V (ψ). Let V (ψ)∗ = {vt t ∈ S(ψ)}, so V (ψ)∗ ⊇ V (ψ). A grounded subset D ⊆ V (ψ)∗ satisfies • v∞ (cid:54)∈ D, • vΩi(t1,...,tn) ∈ D =⇒ vt1, . . . , vtn ∈ D, for any Ωi(t1, . . . , tn) ∈ S(ψ). Informally, each grounded D determines a partition of the subterms into 'defined' terms t (when vt ∈ D) and 'undefined' terms s (when vs ∈ V (ψ)∗ \ D), in a way that is consistent with the structure of the terms. For any subset D ⊆ V (ψ)∗ define ϕ(D) = Ri(vt1, . . . , vtn, vΩi(t1,...,tn)) ∀w¬Ri(vt1, . . . , vtn, w) (cid:94) vt1 ,...,vtn∈D, vΩi(t1,...,tn)∈D where w is a new variable. For any equation s = t occurring in ψ define ∧ (cid:94) vt1 ,...,vtn∈D, vΩi(t1,...,tn)(cid:54)∈D if vs, vt ∈ D if vs, vt (cid:54)∈ D otherwise  vs = vt (cid:62) ⊥ (s = t)D by 10 ROBIN HIRSCH AND BRETT MCLEAN and then let ψD be obtained from ψ by replacing each equation s = t by (s = t)D. Translate ψ to the L(J)-formula (cid:94) ψ− = (ϕ(D) → ψD). D ⊆ V (ψ)∗ D grounded We must prove (3.1). Suppose ψ is not valid in A+, say ρ : V (ψ) → A+ is an assignment such that A+, ρ (cid:54)= ψ. Let ρ∗ : V (ψ)∗ → A satisfy ρ∗(vt) = [t]ρ for any t ∈ S(ψ) provided [t]ρ (cid:54)= ∞ (else ρ∗(vt) ∈ A is arbitrary-possible since A is nonempty) and let D = {vt ∈ V (ψ)∗ [t]ρ (cid:54)= ∞}. Then D is grounded, A, ρ∗ = ϕ(D) and by formula induction we get A+, ρ = χ ⇐⇒ A, ρ∗ = χD for any subformula χ of ψ. So A, ρ∗ (cid:54)= ϕ(D) → ψD and therefore ψ− is not valid in A. Conversely, suppose ψ− is not valid in A, so there is a grounded subset D ⊆ V (ψ)∗ and a variable assignment µ : V (ψ−) → A such that A, µ = ϕ(D) ∧ ¬ψD. As A is nonempty we may extend µ to an assignment λ : V (ψ)∗ → A and we will have A, λ = ϕ(D) ∧ ¬ψD. Define λ+ : V (ψ) → A+ by Since A, λ = ϕ(D) and λ+ agrees with λ over D ∩ V (ψ) we have λ+(u) = (cid:26) λ(u) u ∈ D (cid:26) λ(vt) vt ∈ D ∞ u ∈ V (ψ) \ D. ∞ vt (cid:54)∈ D. [t]λ+ = (3.2) A+, λ+ = χ ⇐⇒ A, λ = χD We claim that (3.3) for any subformula χ of ψ. For the base case let χ be an equation (s = t). If vs, vt ∈ D then χD is vs = vt and (3.3) holds, by (3.2). If vs ∈ D but vt (cid:54)∈ D then χD is ⊥ and ∞ = [t]λ+ (cid:54)= [s]λ+ ∈ A so both sides of (3.3) are false. The case where vt ∈ D but vs (cid:54)∈ D is similar. Finally, if vs, vt (cid:54)∈ D then χD is (cid:62) and [s]λ+ = [t]λ+ = ∞ so both sides of (3.3) are true. Now (3.3) follows for all subformulas χ of ψ, by a simple structural induction. Since A, λ (cid:54)= ψD we deduce that A+, λ+ (cid:54)= ψ, so ψ is not valid in A+. This completes the proof of (3.1). If A is nonempty, we have A ∈ C ⇐⇒ A+ ∈ C+ ⇐⇒ A+ = Σ+ ⇐⇒ A = {ψ− ψ ∈ Σ+}. So if the empty partial algebra is in C then {∀xψ− ψ ∈ Σ+} is an axiomatisation of C. If the empty partial algebra is not in C then {∃x(cid:62) ∧ ψ− ψ ∈ Σ+} is an axiomatisation of C. Theorem 3.8. Let σ be any one of the signatures ( algebras σ-representable as sets, viewed as a class of relational structures, is elementary. •∪), ( •\) or ( •∪, •\). The class of partial Proof. The classes in question are J, K and L. Lemma 3.6 tells us that each class satisfies the condition for Proposition 3.7 to apply. Hence each class is axiomatisable in the appropriate relational language. DISJOINT-UNION PARTIAL ALGEBRAS 11 We can now easily establish elementarity in all cases without composition. Corollary 3.9. Let σ be any signature whose symbols are a subset of { •∪, of partial algebras that are σ-representable by sets is elementary. •\,∅}. The class •∪,∅), ( •\). Proof. The previous theorem gives us the result for the three signatures ( Then as we noted in Remark 2.10, axiomatisations for these signatures yield axiomatisations •\,∅) with the addition of a single extra axiom, either for the signatures ( J(0, 0, 0) or K(0, 0, 0). The remaining cases, the empty signature and the signature (∅), trivially are axiomatised by the empty theory. Corollary 3.10. Let σ be any signature whose symbols are a subset of { • (cid:94), of partial algebras that are σ-representable by partial functions is elementary. •\,∅}. The class •\,∅) and ( •∪, •∪), ( •\) and ( •∪, Proof. By Proposition 2.9(1) these representation classes are the same as those in Corol- lary 3.9. 4. A Recursive Axiomatisation via Games •∪-representable partial In this section we describe a recursive axiomatisation of the class of algebras. This axiomatisation can be understood quite simply, as a sequence of statements about a particular two-player game. The efficacy of this approach using games relies on our prior knowledge, obtained in the previous section, that the class in question is elementary. The reader should note that everything in this section can be adapted quite easily to •\-representability by sets and ( •∪, •\)-representability by sets. Fix some partial •(cid:116)-algebra A. The following definition and lemma are the motivation behind our two-player game. Definition 4.1. We call a subset U of A • •(cid:116)-prime if a • bi-closed if the two conditions a ∈ U or b ∈ U and a •(cid:116) b ∈ U implies either a ∈ U or b ∈ U , •(cid:116) b defined, together imply •(cid:116) b ∈ U , a •(cid:116) b is undefined. • pairwise incombinable if a, b ∈ U implies a •(cid:116)-prime, bi-closed, pairwise-incombinable subsets Lemma 4.2. Let F(A) be the set of all of A. Then A has a disjoint-union representation if and only if there is a B ⊆ F(A) such that (i) for all a (cid:54)= b ∈ A there is U ∈ B such that either a ∈ U and b (cid:54)∈ U or b ∈ U and a (cid:54)∈ U , (ii) for all a, b ∈ A if a Proof. For the left-to-right implication, if θ is a disjoint-union representation of A on a base set X then for each x ∈ X let U (x) = {a ∈ A x ∈ aθ} and let B = {U (x) x ∈ X}. It is •(cid:116)-prime, bi-closed, pairwise-incombinable set, for all x ∈ X, and easy to see that U (x) is a that B includes all elements required by (i) and (ii) of this lemma. Conversely, assuming that B ⊆ F(A) has the required elements we can define a rep- resentation θ of A by aθ = {U ∈ B a ∈ U}. Condition (i) ensures that θ is faithful, that is, distinguishes distinct elements of A. Condition (ii) ensures aθ and bθ are disjoint only •(cid:116) b is undefined then there is U ∈ B such that a, b ∈ U . 12 ROBIN HIRSCH AND BRETT MCLEAN •(cid:116) b is defined. The pairwise incombinable condition on each U ∈ B ensures a •(cid:116) b is •(cid:116)-prime and bi-closed conditions on elements of if a defined only if aθ and bθ are disjoint. The B ensure that when a •(cid:116) b is defined, (a •(cid:116) b)θ = aθ ∪ bθ. We define a two player game Γn over A with n ≤ ω rounds, played by players ∀ and ∃. A position (Y, N ) consists of two finite subsets Y and N of A. It might help to think of Y as a finite set of sets such that some given point belongs to each of them and N is a finite set of sets such that the same point belongs to none of them. In the initial round (round 0) ∀ either (i) picks a (cid:54)= b ∈ A, or (ii) picks a, b ∈ A such that a In the former case ∃ responds with an initial position, either ({a},{b}) or ({b},{a}), at her choice. In the latter case she must respond with the initial position ({a, b},∅). •(cid:116) b is undefined. In all later rounds, if the position is (Y, N ) then ∀ either •(cid:116) b is defined, or •(cid:116) b is defined. •(cid:116) b is defined and belongs to Y , or (a) picks a, b ∈ A such that a (b) picks a ∈ Y and b ∈ A such that a (c) picks a ∈ A and b ∈ Y such that a In case (a) player ∃ responds with either (Y ∪ {a}, N ) or (Y ∪ {b}, N ), in cases (b) and (c) •(cid:116) b}, N ). Observe that N never changes as the she must respond with the position (Y ∪ {a game proceeds, it is either a singleton or empty. A position (Y, N ) is a win for ∀ if either (1) Y ∩ N (cid:54)= ∅, or (2) there are a, b ∈ Y such that a Player ∀ wins a play of Γn if he wins in some round 0 ≤ i < n, else ∃ wins the play of the game. •(cid:116) b is defined. The game Γn(Y, N ) is similar (where Y, N are finite subsets of A), but the initial round is omitted and play begins from the position (Y, N ). Lemma 4.3. If A is representable then ∃ has a winning strategy for Γω. If A is countable and ∃ has a winning strategy for Γω then A has a representation on a base of size at most 2A2. Proof. First suppose A has a representation, θ say. By Lemma 4.2 there is a set B of •(cid:116)-prime, bi-closed, pairwise-incombinable subsets of A such that (i) for all a (cid:54)= b ∈ A there •(cid:116) b is undefined is U ∈ B such that either a ∈ U, b (cid:54)∈ U or b ∈ U, a (cid:54)∈ U and (ii) whenever a there is U ∈ B with a, b ∈ U . We describe a winning strategy for ∃. In response to any initial ∀-move she will select a suitable U ∈ B and play an initial position (Y, N ) such that (4.1) Y ⊆ U and N ∩ U = ∅. and the remainder of her strategy will be to preserve this condition throughout the play. In the initial round there are two possibilities. (i) If ∀ plays a (cid:54)= b ∈ A then there is a U ∈ B with either a ∈ U, b (cid:54)∈ U or b ∈ U, a (cid:54)∈ U . In the former case ∃ plays an initial position ({a},{b}) and in the latter case she plays ({b},{a}). DISJOINT-UNION PARTIAL ALGEBRAS 13 In each case, (4.1) holds. (ii) If ∀ plays (a, b) where a •(cid:116) b is undefined, there is U ∈ B where a, b ∈ U and ∃ selects •(cid:116) b ∈ U so ∃ plays (Y ∪ {a •(cid:116) b is defined (or b ∈ Y and a •(cid:116) b ∈ Y is defined then since U is type (i) move (a, b) initially (so N is a singleton). For each a, b ∈ A where a a (cid:54)= b ∈ A let Sa,b =(cid:83) let Ta,b =(cid:83) such a U and plays ({a, b},∅). In a subsequent round, if the current position (Y, N ) satisfies (4.1) and ∀ plays a, b •(cid:116)-prime either Y ∪ {a} ⊆ U or Y ∪ {b} ⊆ U , so ∃ where a may play either (Y ∪ {a}, N ) or (Y ∪ {b}, N ), as appropriate, preserving (4.1). Similarly, if •(cid:116) b is defined), then since U is ∀ plays a, b where a ∈ Y and a •(cid:116) b}, N ), preserving condition (4.1). This bi-closed we have a condition suffices to prove that ∃ does not lose in any round of the play. Conversely, suppose A is countable and ∃ has a winning strategy for Γω. Then for each i<ω Yi, where (Y0, N ), (Y1, N ), . . . is a play of Γω in which ∀ plays the •(cid:116) b is undefined i<ω Yi be the limit of a play in which ∀ plays the type (ii) move (a, b) initially (so N is empty). In each case we suppose-here is where we use the hypothesis that A is countable-that ∀ plays all possible moves subsequently. We also suppose that ∃ uses her winning strategy. •(cid:116)-prime, bi-closed and pairwise incombinable, since ∀ plays all possible moves in a play and ∃ never loses. Hence •(cid:116) b is undefined} satisfies the conditions of Lemma 4.2. B = {Sa,b a (cid:54)= b ∈ A} ∪ {Ta,b a Clearly the size of the base set B is at most 2A2. Lemma 4.4. For each n < ω there is a first-order L(J)-formula ρn such that A = ρn if and only if ∃ has a winning strategy in Γn. Proof. Let V and W be disjoint finite sets of variables. For each n < ω we define formulas •(cid:116)-algebra A and any variable assignment µn(V, W ) in such a way that for any partial λ : vars → A we have Each set Sa,b (where a (cid:54)= b) or Ta,b (where a •(cid:116) b is undefined) is A, λ = µn(V, W ) ⇐⇒ ∃ has a winning strategy in Γn(λ[V ], λ[W ]). (4.2) Let (cid:94) ¬∃cJ(v, v(cid:48), c) ∧ (cid:94) v (cid:54)= w v∈V, w∈W µ0(V, W ) = v,v(cid:48)∈V where c is a fresh variable. So (4.2) is clear when n = 0. For the recursive step let (J(a, b, v) → µn(V ∪ {a}, W ) ∨ µn(V ∪ {b}, W )) µn+1(V, W ) = ∀a, b (cid:18) (cid:94) ∧ (cid:94) ∧ (cid:94) v∈V v∈V v∈V (J(a, v, b) → µn(V ∪ {b}, W )) (J(v, a, b) → µn(V ∪ {b}, W )) (cid:19) where a and b are fresh variables. By a simple induction on n we see that (4.2) holds, for all n. Finally, (let ρ0 = (cid:62) and) let ρn+1 = ∀a, b (a = b ∨ µn({a},{b}) ∨ µn({b},{a})) ∧ (∃cJ(a, b, c) ∨ µn({a, b},∅)) (cid:19) (cid:18) 14 ROBIN HIRSCH AND BRETT MCLEAN where again a, b and c are fresh variables. Observe that each formula µn(V, W ) is equivalent to a universal formula and therefore ρn, but for the clause ∃cJ(a, b, c), is universal. Theorem 4.5. The isomorphic closure of the class of disjoint-union partial algebras of sets is axiomatised by {ρn n < ω}. Proof. We will use Lemma 4.3 and Lemma 4.4, but we must be slightly careful, because we chose to present the lemmas with the assumption that the L(J)-structure in question is a partial algebra. Hence we must check that (2.1) holds before appealing to either lemma. If an L(J)-structure A is isomorphic to a disjoint-union partial algebra of sets then certainly it satisfies (2.1). Then by Lemma 4.3, player ∃ has a winning strategy in the game of length n for each n < ω. So A = ρn by Lemma 4.4. Conversely, if A = {ρn n < ω} let B be any countable elementary substructure of A. Then B = {ρn n < ω}. The validity of ρ3 tells us that (2.1) holds, as we now explain. For if J(a, b, c) and J(a, b, d), with c (cid:54)= d, then from ρ3 we know that either µ2({c},{d}) holds or µ2({d},{c}) holds. Without loss of generality, we assume the former. From µ2({c},{d}), assigning c to v in the first conjunct, we deduce µ1({c, a},{d}) or µ1({c, b},{d}) and again we may assume the former. From the second conjunct in µ1({c, a},{d}) (assigning b to the variable v and d to the variable b) we deduce µ0({c, a, d},{d}), which is contradicted by the final inequality v (cid:54)= w, when v and w are both assigned d. Hence we can use Lemma 4.4 and conclude that ∃ has a winning strategy in game Γn for each n < ω. Then since ∃ has only finitely many choices open to her in each round (actually, at most two choices), by Konig's tree lemma she also has a winning strategy in Γω. So by Lemma 4.3 the partial algebra B is isomorphic to a disjoint-union partial algebra of sets. Since A is elementarily equivalent to B, we deduce A is also isomorphic to a disjoint-union partial algebra of sets, by Proposition 3.7. 5. Non-axiomatisability •∪), ( •\,∅) or ( •∪, •\), ( •\), ( •∪,∅), ( •∪, •∪ is replaced by • •\,∅) In this section we show that for any of the signatures ( the class of partial algebras representable by sets is not finitely axiomatisable. Hence the same is true for representability by partial functions, when (cid:94). For partial functions, we also show the same holds when we add composition to these signatures. Our strategy is to describe a set of non-representable partial algebras that has a representable ultraproduct. Let m and n be sets of cardinality greater than two. We will call a subset of m × n axial if it has the form {i} × J (for some i ∈ m, J ⊆ n) or the form I × {j} (for some I ⊆ m, j ∈ n). Observe that ∅ × {j} = {i} × ∅ = ∅ for any i ∈ m, j ∈ n. •(cid:116), 0)-algebra X(m, n). It has a domain consisting of all axial •(cid:116) T is defined and Now for any i (cid:54)= i(cid:48) ∈ m and j (cid:54)= j(cid:48) ∈ n the set I(i, j, i(cid:48), j(cid:48)) = {a ∈ X(m, n) •(cid:116)-prime, bi-closed, pairwise-incombinable set. The collection (i, j) ∈ a or (i(cid:48), j(cid:48)) ∈ a} is a {I(i, j, i(cid:48), j(cid:48)) i (cid:54)= i(cid:48) ∈ m, j (cid:54)= j(cid:48) ∈ n} of such sets satisfies conditions (i) and (ii) of subsets of m × n. The constant 0 is interpreted as the empty set and S equal to S ∪ T if S is disjoint from T and S ∪ T is axial, else it is undefined. Next we define a partial ( DISJOINT-UNION PARTIAL ALGEBRAS 15 Lemma 4.2 and hence the also satisfied, so X(m, n) is ( •(cid:116)-reduct of X(m, n) is •∪,∅)-representable. •∪-representable. The formula 0 •(cid:116) 0 = 0 is Definition 5.1. Given a partial algebra A = (A, (Ωi)i<β), a partial-algebra congru- ence on A is an equivalence relation ∼ with the property that for each i and every a1, . . . , aα(i), b1, . . . , bα(i) ∈ A, if a1 ∼ b1, . . . , aα(i) ∼ bα(i) then Ωi(a1, . . . , aα(i)) is defined if and only if Ωi(b1, . . . , bα(i)) is defined and when these are defined Ωi(a1, . . . , aα(i)) ∼ Ωi(b1, . . . , bα(i)). Note our condition for being a partial-algebra congruence is strictly stronger than that obtained by viewing a partial algebra as a relational structure and then using the familiar definition of a congruence-for signatures with no function symbols a congruence relation is merely an equivalence relation. Our definition of a partial-algebra congruence takes the 'algebraic' rather than 'relational' view of the structure. If a congruence ∼ on a partial algebra A is a partial-algebra congruence, that is sufficient for the quotient A/∼ (with its usual meaning on the relational view of A) to itself be a partial algebra. Definition 5.2. Given a partial algebra A = (A, (Ωi)i<β) and a partial-algebra congruence ∼ on A, the partial-algebra quotient of A by ∼, written A/∼, is the partial algebra of the signature (Ωi)i<β with domain the set of ∼-equivalence classes and well-defined partial operations given by: Ωi([a1], . . . , [aα(i)])= [Ωi(a1, . . . , aα(i))] if Ωi(a1, . . . , aα(i)) is defined, else Ωi([a1], . . . , [aα(i)]) is undefined. Given a partial-algebra congruence ∼ on A, the partial-algebra quotient by ∼ is the same structure as the quotient by ∼, which is why we reuse the notation A/∼. All the expected relationships between partial-algebra homomorphisms, partial-algebra congruences and partial-algebra quotients hold. Returning to our task, we define a binary relation ∼ over X(m, n) as the smallest equivalence relation such that {i} × n ∼ m × {j} {i} × (n \ {j}) ∼ (m \ {i}) × {j} •(cid:116) T is defined if and only if T for all i ∈ m, j ∈ n. The equivalence class of {i} × n (for any choice of i ∈ m) is denoted 1 and the equivalence class of {i} × (n \ {j}) is denoted (i, j), for each i ∈ m, j ∈ n. All other equivalence classes are singletons, either {{i} × J} for some i ∈ m, J (cid:40) n or {I × {j}} for •(cid:116) is some I (cid:40) m, j ∈ n. We show next that ∼ is a partial-algebra congruence. Clearly •(cid:116) S is defined and then they commutative in the sense that S are equal. Hence it suffices to show, for any S ∼ S(cid:48), that S ∩ T = ∅ and S ∪ T is axial if and only if S(cid:48) ∩ T = ∅ and S(cid:48) ∪ T is axial, and if these statements are true then S ∪ T ∼ S(cid:48) ∪ T . Further, by symmetry, it suffices to prove only one direction of this biconditional. Suppose then that S ∼ S(cid:48), that S ∩ T = ∅ and that S ∪ T is axial. We may assume S (cid:54)= S(cid:48), so without loss of generality there are two cases to consider: the case S = {i} × n and the case S = {i} × (n \ {j}) and S(cid:48) = (m \ {i}) × {j}. In the first case, since S ∪ T is axial and n > 1 we know T must be a subset of S. But T is also disjoint from S, hence T is empty. Then it is clear that S(cid:48) ∩ T = ∅ and S(cid:48) ∪ T is axial and that S ∪ T ∼ S(cid:48) ∪ T . In the second case, since S ∪ T is axial and n > 2 we know T must be a subset of {i} × n. But T is also disjoint from S and so T is either {(i, j)} or ∅. Either way, it is clear that S(cid:48) ∩ T = ∅ and S(cid:48) ∪ T is axial and that S ∪ T ∼ S(cid:48) ∪ T . 16 ROBIN HIRSCH AND BRETT MCLEAN •(cid:116), 0)-algebra A(m, n) as ( Now define a partial the partial-algebra quotient X(m, n)/∼. Since the elements of A(m, n) are ∼-equivalence classes and these are typ- ically singletons, we will suppress the [· ] notation and let the axial set S denote the equivalence class of S, taking care to identify ∼-equivalent axial sets. For the following lemma, recall the notion of (cid:46)-complete representability given in Definition 2.11. •∪, 0)-representable if and only if m = n. Lemma 5.3. For any sets m and n of cardinality greater than two, the partial algebra A(m, n) is (cid:46)-completely ( Proof. For the left-to-right implication let θ be a (cid:46)-complete representation of A(m, n) over •(cid:116) 1 is undefined. Fix some x ∈ 1θ and the base X. The set 1θ must be nonempty, because 1 define a subset R of m × n by letting (i, j) ∈ R ⇐⇒ x ∈ {(i, j)}θ for i ∈ m, j ∈ n. For each i ∈ m, since 1 is the supremum of {{(i, j)} j ∈ n} and θ is (cid:46)-complete, there is j ∈ n such that x ∈ {(i, j)}θ and hence (i, j) ∈ R. Dually, for any j ∈ n, since 1 is the supremum of {{(i, j)} i ∈ m} there is i ∈ m such that (i, j) ∈ R. We cannot have (i, j), (i(cid:48), j) ∈ R, for distinct i, i(cid:48) ∈ m since θ is a representation and {(i, j)} •(cid:116) {(i(cid:48), j)} is defined. Similarly, for distinct j, j(cid:48) ∈ n we cannot have (i, j), (i, j(cid:48)) ∈ R. Hence R is a bijection from m onto n. We deduce that m = n. For the right-to-left implication suppose m = n. It suffices to describe a (cid:46)-complete We now show that θ is a ( representation of A(n, n). The base of the representation is the set Pn of all permutations on n. If S is any axial set it has the form {i} × J for some i ∈ n, J ⊆ n or the form I × {j} for some I ⊆ n, j ∈ n. Define a representation θ over Pn by letting ({i} × J)θ be the set of all permutations σ ∈ Pn such that σ(i) ∈ J and (I ×{j})θ be the set of all permutations σ ∈ Pn such that σ−1(j) ∈ I. Observe this is well defined, since firstly if an axial set is both of the form {i} × J and of the form I × {j} then the definitions agree, and secondly it is easily seen that ∼-equivalent axial sets are assigned the same set of permutations. •∪, 0)-representation. To see that θ is faithful we show that ∼-inequivalent axial sets are represented as distinct sets of permutations. We may assume the axial sets are not in the equivalence class 1, since 1θ = Pn and all axial sets not in 1 are clearly assigned proper subsets of Pn. Similarly, we may assume the axial sets are not the empty set. First suppose we have two inequivalent vertical sets {i} × J and {i(cid:48)} × J(cid:48). If i = i(cid:48) there must be a j in the symmetric difference of J and J(cid:48). Then any permutation with i (cid:55)→ j witnesses the distinction between ({i}× J)θ and ({i(cid:48)}× J(cid:48))θ. Otherwise i (cid:54)= i(cid:48), and if we can choose j (cid:54)= j(cid:48) with j ∈ J and j(cid:48) (cid:54)∈ J(cid:48) then any permutation with i (cid:55)→ j and i(cid:48) (cid:55)→ j(cid:48) belongs to ({i} × J)θ \ ({i(cid:48)} × J(cid:48))θ. Since we assumed our axial sets are neither ∅ nor in 1 we can do this unless J and n \ J(cid:48) are the same singleton set, {j0} say. But then for any distinct j, j(cid:48) ∈ n \ {j0} we have j (cid:54)∈ J and j(cid:48) ∈ J(cid:48) so any permutation with i (cid:55)→ j, i(cid:48) (cid:55)→ j(cid:48) belongs to ({i(cid:48)} × J(cid:48))θ \ ({i} × J)θ. Hence θ always distinguishes inequivalent vertical sets. If we have two inequivalent horizontal sets I × {j} and I(cid:48) × {j}(cid:48) then the argument is similar. Lastly, suppose we have inequivalent sets {i} × J and I × {j}. If we can choose a k ∈ J not equal to j and an l (cid:54)∈ I not equal to i then there exist permutations with i (cid:55)→ k and l (cid:55)→ j and any such permutation belongs to ({i} × J)θ \ (I × {j})θ. We can do this unless either J = {j}, in which case we have two horizontal sets, which we have already considered, or I = n\{i}. By a symmetrical argument, we can witness the distinction unless J = n\{j}. DISJOINT-UNION PARTIAL ALGEBRAS 17 •∪. If S •(cid:116) correctly as •∪ T θ is defined and that (S It is clear that θ correctly represents 0 as ∅. Now to see that θ is a ( Hence ({i} × J)θ (cid:54)= (I × {j})θ unless {i} × J = {i} × (n \ {j}) and I × {j} = (n \ {i}) × {j}, contradicting the assumed inequivalence of {i}× J and I ×{j}. This completes the argument that θ is faithful. •∪,∅)-representation •(cid:116) T is defined then we may it remains to show that θ represents assume S = {i} × J1 and T = {i} × J2 for some disjoint J1 and J2, since the case where •(cid:116) T is a horizontal set is similar. Then it is clear from the definition of θ that Sθ and T θ S •(cid:116) T is undefined are disjoint and so Sθ then either there is some (i, j) ∈ S ∩ T , in which case Sθ and T θ clearly are non-disjoint, or S ∪ T is not axial, in which case there are i (cid:54)= i(cid:48) and j (cid:54)= j(cid:48) with (i, j) ∈ S and (i(cid:48), j(cid:48)) ∈ T . In the second case, any permutation with i (cid:55)→ j and i(cid:48) (cid:55)→ j(cid:48) witnesses that Sθ and T θ are •∪ T θ is undefined. This completes the non-disjoint. Hence when S proof that θ is a ( Finally we show that θ is (cid:46)-complete. Let γ be a pairwise-combinable subset of A(n, n). If γ has supremum {i}× J for some J with n\ J ≥ 2 then for all S ∈ γ since the supremum is an upper bound and by the definition of (cid:46), either S = {i} × J or there is T such that •(cid:116) T ∼ {i} × J. It follows that each S ∈ γ has the form {i} × JS for some JS ⊆ J and S S∈γ JS. Then for any σ ∈ Pn we since the {i} × J is the least upper bound we have J =(cid:83) •(cid:116) T is undefined, Sθ •∪,∅)-representation. •(cid:116) T )θ = Sθ •∪ T θ. If S have σ ∈ ({i} × J)θ ⇐⇒ σ(i) ∈ J ⇐⇒ σ(i) ∈ JS for some S ∈ γ ⇐⇒ σ ∈ ({i} × JS)θ for some S ∈ γ Sθ. ⇐⇒ σ ∈ (cid:91) ({i} × JS)θ = (cid:91) S∈γ Similarly if the supremum of γ is I×{j} for some I with m\I ≥ 2, then (I×{j})θ =(cid:83) S∈γ Sθ. If the supremum of γ is (i, j) then either γ = {(i, j)}, so the proof of the required equality is trivial, or, because γ is pairwise combinable, each S ∈ γ has the form {i} × JS or each S ∈ γ has the form IS × {j} in which cases the proof is similar to above. If the supremum of γ is 1, then either γ = {1} or γ = {{(i, j)}, (i, j)} for some i, j, or each S ∈ γ has the form {i} × JS, or each S ∈ γ has the form IS × {j}. In every case the required equality is seen to hold. So θ is a (cid:46)-complete representation. S∈γ •∪,∅)-representation, but, by Lemma 5.3 Remark 5.4. We have seen that X(3, 4) has a ( and Proposition 2.12, the partial algebra A(3, 4) = X(3, 4)/∼ does not. Since the latter is a •∪,∅)-representable partial-algebra homomorphic image of the former we see that the class of ( partial algebras is not closed under partial-algebra homomorphic images, in contrast to the corresponding result for algebras representable as fields of sets, that is, boolean algebras. We now have a source of non-representable partial algebras with which to prove our first non-axiomatisability result. Theorem 5.5. The class of ( •∪,∅)-representable partial algebras is not finitely axiomatisable. 18 ROBIN HIRSCH AND BRETT MCLEAN Proof. Write ν for ω \{0, 1, 2} and let m ∈ ν. By Lemma 5.3 the partial algebra A(m, m + 1) •∪,∅)-representation. Since this partial algebra is finite, it follows, by has no (cid:46)-complete ( Proposition 2.12, that it has no ( •∪,∅)-representation. Let U be a non-principal ultrafilter over ν. We claim that the ultraproduct Πm∈νA(m, m+ 1)/U is isomorphic to a partial-subalgebra of A(Πm∈νm/U, Πm∈ν(m + 1)/U ). Note that every element of Πm∈νA(m, m + 1)/U is the equivalence class of a sequence of vertical sets [({im} × Jm)m∈ν] where im ∈ m and Jm ⊆ m + 1 for each m ∈ ν, or the equivalence class of a sequence of horizontal sets [(Im × {jm})m∈ν] where Im ⊆ m and j ∈ m + 1 for each m ∈ ν. The partial-algebra embedding θ maps [({im}× Jm)m∈ν] to {[(im)m∈ν]}×{[(jm)m∈ν] {m ∈ ν jm ∈ Jm} ∈ U}, and it maps [(Im × {jm})m∈ν] to {[(im)m∈ν] {m ∈ ν im ∈ Im} ∈ U} × {[(jm)m∈ν]}. We prove the contrapositive. Suppose a •(cid:116) bθ is defined in A(Πm∈νm/U, Πm∈ν(m+1)/U ) then a It is easy to check that θ is a well-defined partial-algebra embedding. We limit ourselves •(cid:116) b is defined in to showing that if aθ Πm∈νA(m, m + 1)/U , since it is this condition that distinguishes partial-algebra embeddings from embeddings of relational structures. •(cid:116) b is undefined and let [(am)m∈ν] = a and [(bm)m∈ν] = b. Then we can find S ∈ U such that one of the following two possibilities holds. One, for each m ∈ S there exists (im, jm) belonging to both (a representative of) am and (a representative of) bm. Or two, for each m ∈ S there exists im (cid:54)= i(cid:48) m such m, j(cid:48) that (im, jm) belongs to (a representative of) am and (i(cid:48) m) belongs to (a representative of) bm. Extend (im)m∈S, (jm)m∈S and, if appropriate, (i(cid:48) m)m∈S and (j(cid:48) m)m∈S to ν-sequences arbitrarily. If the first alternative holds then ([(im)m∈ν], [(jm)m∈ν]) belongs to (representat- •(cid:116) bθ is undefined since the representatives are non-disjoint. ives of) both aθ and bθ. So aθ If the second alternative holds then [(im)m∈ν] (cid:54)= [(i(cid:48) m)m∈ν] and ([(im)m∈ν], [(jm)m∈ν]) belongs to (a representative of) aθ and ([(i(cid:48) m)m∈ν]) belongs •(cid:116) bθ is undefined since the union of the representatives is to (a representative of) bθ. So aθ not axial. We now argue that A(Πm∈νm/U, Πm∈ν(m + 1)/U ) is representable, by showing that the cardinalities of its two parameters are equal. The map f : Πm∈νm/U → Πm∈ν(m + 1)/U defined by f ([(im)m∈ν]) = [(im + 1)m∈ν] is injective and its range is all of Πm∈ν(m + 1)/U except [(0, 0, . . .)]. Since these are infinite sets it follows that the cardinality of Πm∈νm/U equals the cardinality of Πm∈ν(m + 1)/U . It follows by Lemma 5.3 that A(Πm∈νm/U, Πm∈ν(m + 1)/U ) is ( m)m∈ν], [(jm)m∈ν] (cid:54)= [(j(cid:48) •∪,∅)-representable. m and jm (cid:54)= j(cid:48) m)m∈ν], [(j(cid:48) Since the partial algebra Πm∈νA(m, m + 1)/U has a partial algebra embedding into a representable partial algebra and the class of representable partial algebras is closed under partial subalgebras, we conclude that Πm∈νA(m, m + 1)/U is itself representable. Hence we have an ultraproduct of unrepresentable partial algebras that is itself representable. It •∪,∅)-representable partial algebras cannot be follows by (cid:32)Lo´s's theorem that the class of ( defined by finitely many axioms. (cid:94),,∅). Corollary 5.6. Let σ be any one of the signatures ( The class of σ-representable partial algebras is not finitely axiomatisable in L(J), L(J, 0), L(J, ;) or L(J, ;, 0), as appropriate. (cid:94),) or ( • (cid:94),∅), ( • •∪,∅), ( • •∪), ( • (cid:94)), ( Proof. The case σ = ( which tells us that the representation classes for ( •∪,∅) is Theorem 5.5. The case σ = ( • (cid:94),∅) follows by Proposition 2.9(1), •∪,∅) and ( • (cid:94),∅) coincide. DISJOINT-UNION PARTIAL ALGEBRAS 19 •(cid:116), 0) has no ( • For the case σ = ( • (cid:94),,∅), for any sets m, n of cardinality greater than two, expand •(cid:116), ;, 0)-algebra B(m, n) by defining a ; b = 0 for all a, b. As in the proof A(m, n) to a partial ( of Theorem 5.5, write ν for ω \ {0, 1, 2} and let U be a non-principal ultrafilter over ν. Then (cid:94),,∅)-representation, as its reduct for every m ∈ ν the partial algebra B(m, m + 1) has no ( • (cid:94),∅)-representation. However, as we saw in the proof of Theorem 5.5, the to ( (cid:94),∅)-representation and moreover, reduct of Πm∈νB(m, m + 1)/U to ( by (cid:32)Lo´s's theorem, it validates a ; b = 0. By Proposition 2.9, these conditions ensure (cid:94),,∅)-representation. Once again we have an ultraproduct of Πm∈νB(m, m + 1)/U has a ( • unrepresentable partial algebras that is itself representable. Hence the representation class is not finitely axiomatisable. For each of the signatures containing ∅ the result follows from the result for the corresponding signature without ∅, by Remark 2.10. Because if the representation class for the signature without ∅ were finitely axiomatisable we could finitely axiomatise the case with ∅ by the addition of the single extra axiom J(0, 0, 0). •(cid:116), 0) does have a ( • We can prove a stronger negative result about (cid:46)-complete representability. Theorem 5.7. The class of (cid:46)-completely ( under elementary equivalence. •∪,∅)-representable partial algebras is not closed •(cid:116),∅)-algebras A1 = A(ω1, ω) and A0 = A(ω, ω), where Proof. Consider the two partial ( ω1 denotes the first uncountable ordinal. By Lemma 5.3 the former is not (cid:46)-completely •∪,∅)-representable while the latter is. We prove these two partial algebras are elementarily ( equivalent by showing that the second player has a winning strategy in the Ehrenfeucht- Fraiss´e game of length ω played over A1 and A0. Although elements of A1 or A0 are formally equivalence classes of axial sets, we may take {0} × ω as the representative of 1 and {i} × (ω \ {j}) as the representative of (i, j), in either partial algebra. Since all elements are axial, each nonzero a ∈ Ai uniquely determines (given this choice of representatives) sets hi(a) and vi(a) such that a = hi(a) × vi(a), for i = 0, 1. For example h1({i} × J) = {i}, v1({i} × J) = J, h1(1) = {0} and v1(1) = ω. We will view 0 as ∅ × ∅, in that hi(0) = vi(0) = ∅. For any sets X, Y we write X ≈ Y if either • both X and Y contain 0 • or neither contain 0 and either X = Y or both sets are infinite. Observe, for any X, Y and U ⊆ X, that X ≈ Y ⇐⇒ there is V ⊆ Y with U ≈ V and X \ U ≈ Y \ V. (5.1) Initially there are no pebbles in play. After k rounds there will be k pebbles on 0. For each S ⊆ k 1 and k matching pebbles on ¯a = (a0, . . . , ak−1) ∈ Ak ¯b = (b0, . . . , bk−1) ∈ Ak let (cid:92) (cid:92) i∈S i∈S h1(bi) ∩ (cid:92) v1(bi) ∩ (cid:92) i∈k\S i∈k\S (ω1 \ h1(bi)), (ω \ v1(bi)), h1(¯b, S) = v1(¯b, S) = 20 ROBIN HIRSCH AND BRETT MCLEAN with similar definitions for h0(¯a, S) and v0(¯a, S). Observe that {h1(¯b, S) S ⊆ k} \ {∅} is a finite partition of ω1 and each of {v1(¯b, S) S ⊆ k} \ {∅}, {h0(¯a, S) S ⊆ k} \ {∅} and {v0(¯a, S) S ⊆ k} \ {∅} is a finite partition of ω. As an induction hypothesis we assume, for each S ⊆ k, that h1(¯b, S) ≈ h0(¯a, S) and v1(¯b, S) ≈ v0(¯a, S). Initially, when k = 0, the only subset of k is ∅ and we have h1(( ),∅) = ω1 ≈ ω = h0(( ),∅) and v1(( ),∅) = ω = v0(( ),∅). In round k, suppose ∀ picks bk ∈ A1. The subsets of k + 1 are {S ∪ {k} S ⊆ k} ∪ {S S ⊆ k}. For any S ⊆ k, since h0((a0, . . . , ak−1), S) ≈ h1((b0, . . . , bk−1), S), by (5.1) there is XS ⊆ h0((a0, . . . , ak−1), S) such that h0((a0, . . . , ak−1), S) \ XS ≈ h1((b0, . . . , bk), S). XS ≈ h1((b0, . . . , bk), S ∪ {k}), S⊆k XS)× ((cid:83) S⊆k XS)× ((cid:83) S⊆k XS and v0(ak) =(cid:83) ted by ((cid:83) bk is the representative of its equivalence class, ((cid:83) ative of its equivalence class, so h0(ak) =(cid:83) (5.2) Similarly there is YS ⊆ v0((a0, . . . , ak−1), S) such that YS ≈ v1((b0, . . . , bk), S ∪ {k}) and v0((a0, . . . , ak−1), S)\YS ≈ v1((b0, . . . , bk), S). Player ∃ lets ak be the element of A0 represen- S⊆k YS), which is an axial set since bk is. In fact more is true: because S⊆k YS) will be the represent- S⊆k YS. Then it follows that h0((a0, . . . , ak), S ∪ {k}) = XS and h0((a0, . . . , ak), S) = h0((a0, . . . , ak−1), S) \ XS and similar identities hold for the vertical components. Hence, by (5.2), the induction hypothesis is maintained. Similarly if ∀ picks ak ∈ A0, we know ∃ can find bk ∈ A1 so as to maintain the induction hypothesis. We claim the induction hypothesis ensures ∃ will not lose the play. To prove that ∃ does not lose, we must prove that {(ai, bi) i < k} is a partial isomorphism from A1 to A0 for every k. That is, we must prove for any i, j, l < k that (1) bi = 0 ⇐⇒ ai = 0, (2) bi = bj ⇐⇒ ai = aj, (3) J(bi, bj, bl) ⇐⇒ J(ai, aj, al). Conditions (1) and (2) follow immediately from the induction hypothesis. Given that (1) and (2) hold, it follows that (3) also holds whenever 0 ∈ {bi, bj}. To prove (3) for the remaining cases, we assume J(bi, bj, bl) holds, where 0 (cid:54)∈ {bi, bj} and distinguish three cases: bl = 1, bl = (i(cid:48), j(cid:48)) (for some i(cid:48) ∈ ω1, j(cid:48) ∈ ω) and bl (cid:54)∈ {1} ∪ {(i(cid:48), j(cid:48)) i(cid:48) ∈ ω1, j(cid:48) ∈ ω}. For bl = 1 we have h1(bl) = {0}, v1(bl) = ω and either h1(bi) = h1(bj) is a singleton and •∪ v1(bj) = ω, or v1(bi) = v1(bj) is a singleton and h1(bi) •∪ h1(bj) = ω1. The induction v1(bi) hypothesis shows that a similar condition holds for the vertical and horizontal components of ai, aj, al, hence J(ai, aj, al) also holds. For bl = (i(cid:48), j(cid:48)) we have h1(bl) = {i(cid:48)}, v1(bl) = ω \ {j(cid:48)} and either h1(bi) = h1(bj) = {i(cid:48)} •∪ h1(bj) = ω1 \ {i(cid:48)}. and v1(bi) Again, the induction hypothesis implies that a similar condition holds for the vertical and horizontal components of ai, aj, al, hence J(ai, aj, al) holds. (cid:54)∈ {1} ∪ {(i(cid:48), j(cid:48)) i(cid:48) < ω1, j(cid:48) < ω} (still with 0 (cid:54)∈ {bi, bj}) then either •∪ v1(bj) = v1(bl), or a similar case, with h1(bi) = h1(bj) = h1(bl) is a singleton and v1(bi) h1 and v1 swapped. As before, an equivalent property holds on ai, aj, al and J(ai, aj, al) follows. This completes the argument that the implication J(bi, bj, bl) =⇒ J(ai, aj, al) is valid. The implication J(ai, aj, al) =⇒ J(bi, bj, bl) is similar. •∪ v1(bj) = ω \ {j(cid:48)}, or v1(bi) = v1(bj) = {j(cid:48)} and h1(bi) When bl DISJOINT-UNION PARTIAL ALGEBRAS 21 As ∃ can win all ω rounds of the play, the two structures A1 and A0 are elementarily •∪,∅)-representable partial algebras are not closed under equivalent. Hence the (cid:46)-completely ( elementary equivalence. (cid:94),,∅). Corollary 5.8. Let σ be any one of the signatures ( The class of (cid:46)-completely σ-representable partial algebras is not closed under elementary equivalence. (cid:94),) or ( • (cid:94),∅), ( • •∪,∅), ( • •∪), ( • (cid:94)), ( (cid:94),∅) coincide. For the case σ = ( • •∪,∅) is Theorem 5.7. For the case σ = ( • (cid:94),∅), note that the proof used Proof. The case σ = ( in Proposition 2.9(1) of the equivalence of representability by sets and by partial functions extends to (cid:46)-complete representability. Hence the (cid:46)-complete representation classes for •∪,∅) and ( • ( (cid:94),,∅), let A1, A0 be as defined in Theorem 5.7. Expand A1 and A0 by adding a binary operation ; defined by a ; b = 0. It is clear that the two expansions are still elementarily equivalent since we have given the same first-order definition of ; for (cid:94),,∅)-representation as A1 itself both. The expansion of A1 does not have a (cid:46)-complete ( • (cid:94),,∅)- is not completely representable. The expansion of A0 does have a (cid:46)-complete ( • representation, which we can easily see via the same method employed in the proof of Proposition 2.9(2). The results for signatures not including ∅ again follow straightforwardly from those for the corresponding signatures with ∅. For a signature with ∅, take any elementarily equivalent A1, A2 with A1 (cid:46)-completely representable and A2 not. Let B1, B2 be the reducts of A1, A2 to the signature without 0. Then B1 is (cid:46)-completely representable since A1 is. As A1 is representable, it satisfies J(0, 0, 0), so A2 does too, by elementary equivalence. Now note that the content of Remark 2.10 applies to (cid:46)-complete representability just as it does to representability. Hence if B2 were (cid:46)-completely representable then A2 would have to be-a contradiction. Hence B2 is not (cid:46)-completely representable. So for the signature without ∅ we have elementarily equivalent B1, B2 with the first (cid:46)-completely representable and the second not. Finally we prove that all the negative results concerning representability for signatures •\. First note that if a partial algebra containing •(cid:116), A = (A, •∪ carry over to signatures containing •−) has a ( •\)-representation then it satisfies •(cid:116) c = a. •− b = c ⇐⇒ b •∪, a whose •(cid:116)-reduct is •(cid:116), However as we see in the following example there exist partial ( •−-reduct has no •∪-representable but whose •(cid:116), •(cid:116)-reduct has no •−)-algebras satisfying (5.3), whose •∪-representation. there exist partial ( whose (5.3) •−)-algebras satisfying (5.3), •\-representation. Similarly •\-representable but •−-reduct is Example 5.9. Our first partial algebra can be quite simple: a partial algebra consisting It satisfies (5.3) and is of a single element a, with a •\-representable. Moreover, we give an example of a partial algebra •∪-representable but not •∪ and then containing a zero element. The domain is ℘{1, 2, 3} \ {3} and we define •(cid:116)-reduct (in fact, a define •− using (5.3). The identity map is a •∪-representation of the •− a both undefined. •(cid:116) a and a •(cid:116) as 22 ROBIN HIRSCH AND BRETT MCLEAN •\-representation of the •(cid:116), 0)-reduct). Suppose θ is a Similarly, if we take a partial algebra with domain ℘{1, 2, 3} \ {1, 2, 3}, define •−-reduct. •∪,∅)-representation of the ( ( We show that {1}θ ⊆ {1, 3}θ, which is a contradiction as {1, 3} •− {1} is undefined. Let x ∈ {1}θ. Then x ∈ {1, 2, 3}θ since {1, 2, 3} •− {1} is defined. As {1, 2} •− {1} = {2} and x ∈ {1}θ we cannot have x ∈ {2}. From {1, 2, 3} •− {2} = {1, 3} we deduce that x ∈ {1, 3}θ. •\ •− as •(cid:116)-reduct of the and define •∪-representation. To see this, note that since {1, 3} = {1} •(cid:116) {3}, partial algebra has no •∪-representation {1} and {3} would have to be represented by disjoint sets. By in any similar arguments, {1},{2} and {3} would have to be represented by pairwise disjoint sets, contradicting the fact that {1} •(cid:116) {2} •(cid:116) {3} is undefined. •(cid:116) using (5.3), the identity map represents the •−-reduct, but the Notwithstanding Example 5.9 there is a simple condition that ensures a is always a •\-representation, and vice versa. •∪-representation Definition 5.10. A partial algebra A = (A, and there is a unique 1 ∈ A such that 1 •(cid:116), •−, . . .) is complemented if it satisfies (5.3) •− a. •− a is defined for all a ∈ A. We write a for 1 Observe by (5.3) that •(cid:116) a = 1. a Hence in a complemented partial algebra A, if σ is a signature containing either θ is a σ-representation of A then where Y = 1θ •\ Y for any Y ⊆ 1θ. aθ = aθ, (5.4) •\ and •∪ or (5.5) •\-analogue of (cid:46)-completeness. In any partial (. . . , Before we articulate the consequences of a partial algebra being complemented, we •−, . . .)-algebra A, define a •\, . . .)- •−) structures satisfying (5.3), describe a relation (cid:46)(cid:48) by letting a (cid:46)(cid:48) b if and only if a = b or b representable then it is clear that (cid:46)(cid:48) is a partial order. For ( observe that (cid:46)(cid:48)=(cid:46). •− a is defined. If A is (. . . , •(cid:116), Definition 5.11. A subset S of a partial (. . . , if for all distinct s, t ∈ S there exists u ∈ A such that u we may define a (. . . , •−-pairwise-combinable sets to (necessarily disjoint) unions. •−-pairwise-combinable •− s = t. As in Definition 2.11 •\, . . .)-representation to be (cid:46)(cid:48)-complete if it maps (cid:46)(cid:48)-suprema of •−, . . .)-algebra A is Lemma 5.12. Let A = (A, subset of ℘(X) (for some X). Then θ is a •\-representation (of the is a and only if it is (cid:46)(cid:48)-complete. •(cid:116), •−, . . .) be complemented and let θ be a map from A to a •(cid:116)-reduct) if and only if it •−-reduct). Moreover, if θ is a representation it is (cid:46)-complete if •∪-representation (of the Proof. Suppose A is complemented and let θ be a is defined then by (5.3) we know that (a •(cid:116) b) •\-representation. For any a, b ∈ A, if a •(cid:116) b •− a = b, which, by our hypothesis about θ, DISJOINT-UNION PARTIAL ALGEBRAS 23 •∪ bθ is defined then a implies that aθ is disjoint from bθ, so aθ •(cid:116) b is defined and (a aθ both sides are defined and equal, assuming aθ •∪ bθ is defined. We now show that, conversely, if •∪ bθ. Using equations to mean •(cid:116) b)θ = aθ •∪ bθ is defined, we have a aθ ∩ bθ = ∅ aθ ⊇ bθ •− b is defined •− b is defined a •\ bθ •− b) •∪ bθ θ (a = aθ = aθ •− a = b •(cid:116) b = a a •(cid:116) b)θ = aθ •− b a (a •− b •∪ bθ •∪ by the definition of as aθ = aθ and bθ ⊆ 1θ as θ is a •\-representation as A is complemented as θ is a •\-representation and by (5.5) by elementary set theory •\-representation as θ is a by (5.3) by the calculation of (a θ •− b) above and hence θ represents Dually, if θ is a •(cid:116) correctly as •∪-representation and a •∪. •− b) = a, implying aθ •(cid:116) (a b both are defined they are equal, assume aθ •− b is defined then we know by (5.3) that •\ bθ is defined. For the converse and for showing that when •\ bθ is defined, so •\ by the definition of as aθ = aθ •∪-representation •(cid:116) a as θ is a by (5.4) cancelling the a's, as θ is a •∪-representation by (5.3) as θ is a •\-representation and by (5.5) by elementary set theory •(cid:116) b) (a b aθ ⊇ bθ aθ ∩ bθ = ∅ •(cid:116) b is defined a •(cid:116) b = 1 = a •(cid:116) a •(cid:116) b = a •(cid:116) a •− b)θ = a = aθ θ •(cid:116) b •(cid:116) bθ •\ bθ = aθ (a and so •− is correctly represented as •\. For the final sentence of this lemma we do not need A to be complemented, only that •−-pairwise combinable' it validates (5.3). Then the concepts 'pairwise combinable' and ' coincide and the relations (cid:46) and (cid:46)(cid:48) are equal. Hence the concepts '(cid:46)-complete' and '(cid:46)(cid:48)- complete' coincide. •\,∅}. The Theorem 5.13. Suppose class of partial algebras σ-representable by sets is not finitely axiomatisable. The class of partial algebras (cid:46)(cid:48)-completely σ-representable by sets is not closed under elementary equivalence. •\ is included in σ and all symbols in σ are from { •∪, 24 ROBIN HIRSCH AND BRETT MCLEAN •(cid:116), •(cid:116), •−, 0) where Proof. For m, n of cardinality greater than two, let A(cid:48)(m, n) be the expansion of A(m, n) to •− is defined by (5.3). Observe that A(cid:48)(m, n) is complemented. Let Aσ(m, n) •(cid:116), ( be the reduct of A(cid:48)(m, n) to the abstract analogue of σ. By Lemma 5.12 (and the fact that •(cid:116) 0 = 0) we see that Aσ(m, n) is (cid:46)(cid:48)-completely σ-representable if and A(cid:48)(m, n) satisfies 0 •\,∅)-representable, which is true if and only if A(m, n) only if A(cid:48)(m, n) is (cid:46)(cid:48)-completely ( •(cid:116),∅)-representable. By Lemma 5.3 this is the case precisely when m = n. is (cid:46)-completely ( So Aσ(m, m + 1) is not σ-representable for 2 < m < ω. As before, write ν for ω\{0, 1, 2} and let U be any non-principle ultrafilter over ν. We will argue that Πm∈νAσ(m, m + 1)/U is σ-representable. From Πm∈νAσ(m, m + 1)/U , form the •−, 0) using (5.3) and defining 0 in the obvious way, if partial algebra B(cid:48) by expanding to ( •(cid:116), 0)-reduct of B(cid:48). We can easily see that, B(cid:48) is complemented necessary. Then let B be the ( and in particular it validates (5.3). Hence Πm∈νAσ(m, m + 1)/U is σ-representable if and •(cid:116), 0)-representable. It is easy to check that B = Πm∈νA(m, m + 1)/U , which only if B is ( •(cid:116), 0)-representable. Hence the ultraproduct we know, by the proof of Theorem 5.5, is ( Πm∈νAσ(m, m + 1)/U of non-σ-representable partial algebras is itself σ-representable and so the class of σ-representable partial algebras is not finitely axiomatisable. For the second half of the theorem, we know, from the proof of Theorem 5.7, that A(ω1, ω) ≡ A(ω, ω), where ≡ denotes elementary equivalence. Hence A(cid:48)(ω1, ω) ≡ A(cid:48)(ω, ω) •−. The elementary equivalence since both expansions use the same first-order definition of of the reducts Aσ(ω1, ω) and Aσ(ω, ω) follows. We established earlier in this proof that Aσ(ω1, ω) is not (cid:46)(cid:48)-completely σ-representable, while Aσ(ω, ω) is. Hence the class of (cid:46)(cid:48)- completely σ-representable partial algebras is not closed under elementary equivalence. •\ is included in σ and all symbols in σ are from { • (cid:94), •\,,∅}. The Corollary 5.14. Suppose class of partial algebras σ-representable by partial functions is not finitely axiomatisable. The class of partial algebras (cid:46)(cid:48)-completely σ-representable by partial functions is not closed under elementary equivalence. •\,∅}, representability Proof. Proposition 2.9(1) tells us that when all symbols are from { • (cid:94), •∪). The by partial functions is the same as representability by sets (with • (cid:94) in place of proof of Proposition 2.9(1) extends to equality of (cid:46)(cid:48)-complete representability by partial functions and by sets. Hence for these signatures the results are immediate corollaries of Theorem 5.13. For signatures σ including both and ∅ we use the same methods as in the proofs of Corollary 5.6 and Corollary 5.8. Let σ− be the signature formed by removing from σ. First we expand the partial algebras Aσ−(m, m + 1) described in the proof of Theorem 5.13 to a signature including ; by defining a ; b = 0 for all a, b. The expanded partial algebras are not representable since the Aσ−(m, m + 1)'s are not. The ultraproduct of the expanded partial algebras validates a ; b = 0, by (cid:32)Lo´s's theorem and so is representable, by the same method as in the proof of Proposition 2.9(2). This refutes finite axiomatisability. For (cid:46)(cid:48)-complete representability, again define ; by a ; b = 0, to expand both of the two elementarily equivalent partial algebras Aσ−(ω1, ω) and Aσ−(ω, ω). The expansions B1 and B0 remain elementarily equivalent and the first is not (cid:46)(cid:48)-completely representable whilst the second is, by the same method as in the proof of Proposition 2.9(2). DISJOINT-UNION PARTIAL ALGEBRAS 25 The remaining cases are signatures including but not ∅, that is ( •\,). For these the results follow from the corresponding signatures that include ∅, by the now-familiar arguments involving Remark 2.10 and its generalisation to (cid:46)(cid:48)-complete representability. •\,) and ( • (cid:94), 6. Signatures Including Intersection We start with the signatures ( •∪,∩,∅) and ( In this section we consider signatures including a total operation · to be represented as intersection. In contrast to the results of the previous section, the classes of partial algebras representable by sets are finitely axiomatisable. This is true for all signatures containing •\,∅}. In order to control the size of intersection and with other operations members of { •∪, this paper we do not consider representability by partial functions, only noting that the proofs in this section are not immediately adaptable to that setting. •∪,∩). Consider the following finite set Ax(J,·, 0) of L(J,·, 0) axioms. •(cid:116) is single valued: J(a, b, c) ∧ J(a, b, c(cid:48)) → c = c(cid:48) •(cid:116) is commutative: J(a, b, c) → J(b, a, c) ·-semilattice: · is commutative, associative and idempotent · distributes over 0 is identity for domain of Let Ax(J,·) be obtained from Ax(J,·, 0) by replacing the axioms concerning 0 (the '0 is •(cid:116)' axioms) by the following axiom stating that either there identity for exists an element z that acts like 0, or else the partial operation •(cid:116): J(b, c, d) → J(a · b, a · c, a · d) •(cid:116): J(a, 0, a) •(cid:116): ∃cJ(a, b, c) ↔ a · b = 0 •(cid:116) is nowhere defined. •(cid:116)' and 'domain of ∃z(∀aJ(a, z, a) ∧ ∀a, b(a · b = z ↔ ∃cJ(a, b, c)) ∨ ∀a, b, c ¬J(a, b, c) (6.1) •∪,∩,∅)-representable by sets is •∪,∩)-representable by Theorem 6.1. The class of (J,·, 0)-structures that are ( axiomatised by Ax(J,·, 0). The class of (J,·)-structures that are ( sets is axiomatised by Ax(J,·). Proof. We first give a quick justification for the axioms being sound in both cases. It suffices to argue that the axioms are sound for disjoint-union partial algebras of sets, with or without zero respectively. Ax(J,·, 0) in turn. •∪ b is defined and is equal to •(cid:116) is single valued: if J(a, b, c) and J(a, b, c(cid:48)) hold then a both c and c(cid:48). Hence c = c(cid:48). •(cid:116) is commutative: J(a, b, c) holds if and only if a∩ b = ∅ and a∪ b = c. By commutativity of intersection and union this is equivalent to the conjunction b ∩ a = ∅ and b ∪ a = c, which holds if and only if J(b, a, c) holds. ·-semilattice: the easily verifiable facts that intersection is commutative, associative and Let A be a disjoint-union partial algebra of sets with zero. We attend to each axiom of idempotent are well known. 26 ROBIN HIRSCH AND BRETT MCLEAN domain of 0 is identity for · distributes over •(cid:116): if J(b, c, d) then b ∩ c = ∅, so certainly (a ∩ b) ∩ (a ∩ c) = ∅. The other condition necessary for J(a · b, a · c, a · d) to hold is that (a ∩ b) ∪ (a ∩ c) = a ∩ d. The left-hand side equals a ∩ (b ∪ c) and by our hypothesis b ∪ c = d, so we are done. •(cid:116): for any set a we have a ∩ ∅ = ∅ and a ∪ ∅ = a, which are the two •(cid:116): for any sets a and b there exists a set c such that J(a, b, c) if and only if conditions needed to establish J(a, 0, a). •∪ b is defined, which is true if and only if a ∩ b = ∅. a Now let A be a disjoint-union partial algebra of sets without zero. It is clear that for all the axioms not concerning 0 the above soundness arguments still hold. To see that axiom (6.1) holds, note that if ∅ ∈ A then ∅ is an element z that acts like 0, as the first clause of (6.1) asks for. Alternatively, if ∅ /∈ A, then for any sets a and b the intersection a ∩ b, which •∪ b is undefined and so for any c we have is an element of A, must be nonempty. Hence a ¬J(a, b, c), meaning the second clause of (6.1) holds. The sufficiency of the axioms is proved for (J,·, 0)-structures by a modification of the proof of Birkhoff's representation theorem for distributive lattices. Assume that Ax(J,·, 0) •(cid:116) is single valued' axiom we can view A as a is valid on a (J,·, 0)-structure A. By the ' •(cid:116),·, 0)-algebra. A filter F is a nonempty subset of A such that a · b ∈ F ⇐⇒ (a ∈ partial ( F and b ∈ F ). For any nonempty subset S of A let (cid:104)S(cid:105) be the filter generated by S, that is, {a ∈ A ∃s1, s2, . . . , sn ∈ S (some finite n), a ≥ s1 · s2 · . . . · sn}, where ≤ is the partial ordering given by the ·-semilattice.1 A filter is proper if it is a proper subset of A. Recall that a set F is •(cid:116) b ∈ F implies either a ∈ F or b ∈ F . •(cid:116)-prime if a •(cid:116)-prime filters of A. Define a map θ from A to ℘(Φ) by Let Φ be the set of all proper letting aθ = {F ∈ Φ a ∈ F}. We will show that θ is a representation of A. The requirement that (a · b)θ = aθ ∩ bθ follows directly from the filter condition a · b ∈ F ⇐⇒ (a ∈ F and b ∈ F ). It follows easily from the axioms concerning 0 that 0 is the minimal element with respect to ≤. Hence a filter is proper if and only if it does not contain 0. Then the requirement that 0θ = ∅ follows directly from the condition that the filters in Φ be proper. We next show that θ is faithful. For this we show that if a (cid:54)≤ b then there is a proper •(cid:116)-prime filter F such that a ∈ F but b (cid:54)∈ F . The filters containing a but not b, ordered by inclusion, form an inductive poset, that is, a poset in which every chain has an upper bound. (The empty chain has an upper bound since the up-set of a is an example of a filter containing a but not b.) Hence, by Zorn's lemma, there exists a maximal such filter, F say. We claim that F is proper and •(cid:116) d is defined and belongs to F but that neither c ∈ F nor d ∈ F . By maximality of F we have b ∈ (cid:104)F ∪ {c}(cid:105) and b ∈ (cid:104)F ∪ {d}(cid:105). Then there is an f ∈ F such that f · c ≤ b and f · d ≤ b. Then by the definition of ≤ and the distributive •(cid:116) d) axiom, b· f · (c •(cid:116) d are in F we get that b should be too-a contradiction. Thus and since both f and c •(cid:116)-prime condition. Clearly F is either c ∈ F or d ∈ F . We conclude that F satisfies the proper, as b (cid:54)∈ F . Hence F is a proper and •(cid:116)-prime filter and so θ is faithful. •(cid:116) d). Hence b ≥ f · (c •(cid:116) (b· f · d) = (f · c) Suppose, for contradiction, that c •(cid:116) (f · d) = f · (c •(cid:116) d) = (b· f · c) •(cid:116)-prime. 1There might be no 'filter generated by the empty set', that is, no smallest filter, as the intersection of two or more filters can be empty. ⇐⇒ (a · b)θ = 0θ ⇐⇒ aθ ∩ bθ = ∅ ⇐⇒ aθ •(cid:116) b and aθ •∪ bθ is defined by the domain of •(cid:116) axiom as θ is faithful as 0 and · are represented correctly by the definition of •∪. DISJOINT-UNION PARTIAL ALGEBRAS 27 To complete the proof that θ is a representation we show that •∪. That is, a •(cid:116) b)θ = aθ •(cid:116) b is defined if and only if aθ •∪ bθ. We have that •(cid:116) is correctly represented •∪ bθ is defined, and when they are defined as (a •(cid:116) b is defined ⇐⇒ a · b = 0 a •∪ bθ ⊆ (a •(cid:116) b)θ ⊆ aθ •(cid:116) b)θ. By the •(cid:116) b)θ, giving us aθ •(cid:116) b)θ, and similarly bθ ⊆ (a •(cid:116) b). So if a is in a filter then by the filter condition a •∪ bθ are defined it follows easily from Ax(J,·,∅) that Further, when both a •(cid:116) b is too. Hence a = a · (a •(cid:116)-prime aθ ⊆ (a •∪ bθ. condition on filters we get the reverse inclusion (a For a (J,·)-structure A, if Ax(J,·) is valid in A then (6.1) holds. If the first alternative of (6.1) holds then we may form an expansion of A to a (J,·, 0)-structure, interpreting 0 as the z given by this clause. Then by the above proof for (J,·, 0)-structures we can •∪,∩,∅)-representation of the expansion. By ignoring the constant 0 we obtain a find a ( •∪,∩)-representation of A. ( Otherwise, the second alternative in (6.1) is true and J(a, b, c) never holds, so we may define a representation θ of A by letting aθ = {b ∈ A b ≤ a}. Clearly θ represents · as ∩ •(cid:116) b is never correctly, by the ·-semilattice axioms. Since a · b ∈ aθ ∩ bθ for any a, b ∈ A and a defined, θ also represents •∪ bθ. Hence (a •∪ correctly. •(cid:116) b)θ = aθ •(cid:116) as •−. From the previous theorem we can easily obtain finite axiomatisations for the signatures •\,∩,∅) and ( •\,∩). Recall that we use the ternary relation K to make first-order •∪, •∪, ( statements about the partial binary operation Corollary 6.2. The class of (J, K,·, 0)-structures that are ( is finitely axiomatisable. The class of (J, K,·)-structures that are ( sets is finitely axiomatisable. Proof. To Ax(J,·, 0) and Ax(J,·) add the formulas a · b = b → ∃cK(a, b, c) and the relational form of (5.3) (that is, K(a, b, c) ↔ J(b, c, a)), which are valid on the representable partial algebras. Then when these axiomatisations hold, the representations in the proof of •\,∩,∅)-representable by sets •\,∩)-representable by •∪, •∪, Theorem 6.1 will correctly represent •− as •\. We claimed finite representability for all signatures containing intersection and with other •\,∅}. For the signatures (∩) and (∩,∅) finite axiomatisability is operations coming from { •∪, •\,∩).2 easy and well known. So the signatures remaining to be examined are ( •∪,∩)-no new ideas are needed-but Our treatment is very similar to the cases ( we provide the details anyway. Consider the following finite set Ax(K,·, 0) of L(K,·, 0) axioms. •\,∩,∅) and ( •∪,∩,∅) and ( 2As an aside, note these are signatures for which representability by sets and by partial functions are easily seen to be the same thing. 28 ROBIN HIRSCH AND BRETT MCLEAN •− is single valued: K(a, b, c) ∧ K(a, b, c(cid:48)) → c = c(cid:48) •− is left injective: K(a, b, c) ∧ K(a(cid:48), b, c) → a = a(cid:48) •− is subtractive: K(a, b, c) ↔ K(a, c, b) ·-semilattice: · is commutative, associative and idempotent · distributes over 0 is identity for domain of Let Ax(K,·) be obtained from Ax(K,·, 0) by replacing the '0 is identity for the axiom •−: K(b, c, d) → K(a · b, a · c, a · d) •−: K(a, 0, a) •−: ∃cK(a, b, c) ↔ a · b = b ¬∃a ∨ ∃z∀aK(a, z, a) •−' axiom by (6.2) •\,∩,∅)-representable by sets is •\,∩)-representable by stating that, provided A is nonempty, there exists an element z that acts like 0. Theorem 6.3. The class of (K,·, 0)-structures that are ( axiomatised by Ax(K,·, 0). The class of (K,·)-structures that are ( sets is axiomatised by Ax(K,·). Proof. Again we give a quick justification for the soundness of the axioms. It suffices to argue •\,∩)-algebras •\,∩,∅)-algebra of sets. We attend to each axiom of Ax(K,·, 0) in turn. •\ b is defined and is equal to •− is single valued: if K(a, b, c) and K(a, b, c(cid:48)) hold then a •− is left injective: re-write the axiom with the predicate J, using (5.3), then it becomes that the axioms are sound for partial ( of sets respectively. •\,∩,∅)-algebras of sets and for partial ( both c and c(cid:48). Hence c = c(cid:48). Let A be a partial ( •(cid:116) is single valued', which we verified in Theorem 6.1. ' •− is subtractive: re-write with J, then it becomes ' ·-semilattice: as in proof of Theorem 6.1. · distributes over 0 is identity for domain of •−: re-write with J, then it becomes '· distributes over •−: clear. •(cid:116) is commutative'. •−: for any sets a and b there exists a set c such that K(a, b, c) if and only if •(cid:116)'. a •\ b is defined, which is true if and only if b ⊆ a, true if and only if a ∩ b = b. Now let A be a partial ( •\,∩)-algebra. It is clear that for all the axioms not concerning 0 the above soundness arguments still hold. To see that axiom (6.2) holds, note that either •\ a is defined and hence its the clause ¬∃a holds or we can take any a ∈ A and find that a value, ∅, is a member of A and witnesses the existence of a z such that ∀aK(a, z, a). To prove the sufficiency of the axioms for (K,·, 0)-structures we use the same method employed in the proof of Theorem 6.1. The definitions of the ordering ≤, of filters and of •−-prime if a ∈ F proper filters remain the same. This time however, we define a filter to be and ∃a •− b together imply either b ∈ F or a •−-prime filters of A and our representation will be the map θ from A to ℘(Φ) defined by aθ = {F ∈ Φ a ∈ F}. That · is correctly Similarly to before, Φ is the set of all proper •− b ∈ F . DISJOINT-UNION PARTIAL ALGEBRAS 29 represented as intersection is again immediate from the (unchanged) definition of a filter. It •−' axioms that once again a filter is follows from the '0 is identity for proper if and only if it does not contain 0. Hence 0 is represented correctly as the empty set. To show that θ is faithful, given a (cid:54)≤ b, as before, we can find a maximal filter F •−' and 'domain of containing a but not b and we show F is proper and Suppose, for contradiction, that c ∈ F and c •− d is defined but that neither d ∈ F nor •− d}(cid:105). So there •− d ∈ F . By maximality of F we have b ∈ (cid:104)F ∪ {d}(cid:105) and b ∈ (cid:104)F ∪ {c c •− d) ≤ b. Then by the definition of ≤ and the is an f ∈ F such that f · d ≤ b and f · (c •− f · d = b· f · c •− f · d. •− b· f · d = b· f · (c distributive axiom, b· f · c •− we obtain b · f · c = f · c, that is, b ≥ f · c. Since both f and c Then by left-injectivity of •− d ∈ F . We are in F we see that b should be too-a contradiction. Thus either d ∈ F or c •−-prime condition. Clearly F is proper, as b (cid:54)∈ F . Hence F is conclude that F satisfies the a proper •−-prime filter and so θ is faithful. •− d) = f · (c •− d) = f · c •−-prime. Finally, we show that •− is correctly represented as •− b is defined ⇐⇒ a · b = b a •\. We have that by the domain of •− axiom as θ is faithful as · is represented correctly by the definition of •\. ⇐⇒ (a · b)θ = bθ ⇐⇒ bθ ⊆ aθ ⇐⇒ aθ •− b and aθ •− b). So if a filter then b is not. Hence (a Further, when both a •− b = a · (a a •− b)θ ⊆ aθ. Similarly, it is easy to show that (a (a •\ bθ is defined •\ bθ are defined it follows easily from Ax(K,·,∅) that •− b is in a filter then by the filter condition a is too. Hence •− b is in a proper •−-prime •\ bθ. For a (K,·)-structure A, if Ax(K,·) is valid in A then (6.2) holds. If the first alternative of (6.2) holds then A is empty and so the empty function forms a representation. Otherwise, the second alternative in (6.2) is true. Then we may form an expansion of A to a (K,·, 0)- structure, interpreting 0 as the z given by this clause. Then by the above proof for •\,∩,∅)-representation of the expansion. By ignoring the (K,·, 0)-structures we can find a ( constant 0 we obtain a ( •− b)θ ∩ bθ = ∅, giving us (a •\,∩) representation of A. condition on filters we get the reverse inclusion (a •− b) · b = 0, so if a •\ bθ. Hence (a •\ bθ. By the •− b)θ ⊆ aθ •− b)θ ⊇ aθ •− b)θ = aθ 7. Decidability and Complexity We finish with a discussion of the decidability and complexity of problems of representability and validity. We also highlight some still-open questions. Theorem 7.1. The problem of determining whether a finite partial union representation is in NP. •(cid:116)-algebra has a disjoint- •(cid:116)), a non-deterministic polynomial-time Proof. Given a finite partial algorithm based on the proof of Lemma 4.3 runs as follows. For each distinct pair a (cid:54)= b •(cid:116)-algebra A = (A, 30 ROBIN HIRSCH AND BRETT MCLEAN •(cid:116) b is undefined it creates a set Ta,b (all it creates a set Sa,b and for each pair a, b where a these sets are initially empty). Then for each c ∈ A, each set Sa,b and each set Ta,b it guesses whether c ∈ Sa,b and whether c ∈ Ta,b. Once this is done, the algorithm then verifies that exactly one of a and b belongs to Sa,b, that both a and b belong to Ta,b and that each of •(cid:116)-prime, bi-closed, pairwise incombinable set (to verify this for any single set these sets is a •(cid:116))). This takes quartic time. By takes quadratic time, in terms of the size of the input (A, Lemma 4.3 this non-deterministic algorithm solves the problem. Problem 7.2. Is the problem of determining whether a finite partial disjoint-union representation NP-complete? •(cid:116)-algebra has a Now turning our attention to validity, let s(¯a), t(¯a) be terms built from variables in •(cid:116). We take the view that the equation s(¯a) = t(¯a) is valid if ¯a and the constant 0, using for every disjoint-union partial algebra of sets with zero, A, and every assignment of the variables in ¯a to sets in A, either both s(¯a) and t(¯a) are undefined or they are both defined and are equal. The following result is rather trivial but worth noting. It contrasts with Theorem 5.5 by showing that the equational fragment of the first-order theory of partial •∪,∅)-algebras is an extremely simple object. ( Theorem 7.3. The validity problem for ( •(cid:116), 0)-equations can be solved in linear time. •∪ b) •∪ c = a •∪ (b •(cid:116). Now •(cid:116), 0)-term is formed from variables and 0, using •∪ is associative in Proof. A ( •∪ c) are defined and equal, or the sense that either both sides of (a •∪ is also commutative. Hence, in representable partial neither is defined. In the same sense, •(cid:116), 0)-algebras, the bracketing and order of variables in a term does not affect whether a ( term is defined, under a given variable assignment, or the value it denotes when it is defined. Similarly, any zeros occurring in a term may be deleted from the term without altering its denotation. If a variable a occurs more than once in a term then the term can only be defined if a is assigned the value 0. Hence an equation s(¯a) = t(¯a) is valid if and only if (a) the set of variables occurring in s(¯a) is the same as the set of variables occurring in t(¯a), (b) the set of variables occurring more than once in s(¯a) is the same as the set of variables occurring more than once in t(¯a). This can be tested in linear time. Problem 7.4. Consider the set Σ of all first-order L(J)-formulas satisfiable over some disjoint-union partial algebras of sets. Is this language decidable and if so, what is its complexity? We have seen that the class of partial algebras with σ-representations by sets is not •\,∅}, finitely axiomatisable, provided either and the same negative result holds for representations by partial functions (with • (cid:94) in place •∪). However when intersection is added to these signatures the representation classes of are finitely axiomatisable by sets. This leaves some cases in question, with regards to finite axiomatisability. •\ is in σ and all symbols in σ are from { •∪, •∪ or Problem 7.5. Determine whether the class of partial algebras σ-representable by partial •\, where functions is finitely axiomatisable for signatures σ containing ∩ and either • symbols in σ are from { • (cid:94), •\,∩,,∅}. (cid:94) or DISJOINT-UNION PARTIAL ALGEBRAS 31 References [1] Josh Berdine, Cristiano Calcagno, and Peter W. O'Hearn, Smallfoot: Modular automatic assertion checking with separation logic, International Symposium on Formal Methods for Components and Objects, Springer, 2005, pp. 115–137. [2] George Boole, The mathematical analysis of logic, being an essay towards a calculus of deductive reasoning, MacMillan, Barclay, & MacMillan, Cambridge and George Bell, London, 1947. [3] James Brotherston and Max Kanovich, Undecidability of propositional separation logic and its neighbours, 25th Annual IEEE Symposium on Logic in Computer Science, 2010, pp. 137–146. [4] Cristiano Calcagno, Dino Distefano, J´er´emy Dubreil, Dominik Gabi, Pieter Hooimeijer, Martino Luca, Peter O'Hearn, Irene Papakonstantinou, Jim Purbrick, and Dulma Rodriguez, Moving fast with software verification, NASA Formal Methods Symposium, Springer, 2015, pp. 3–11. [5] Arthur Cayley, On the theory of groups, as depending on the symbolic equation θn = 1, Philosophical Magazine 7 (1854), no. 42, 40–47. [6] C. C. Chang and H. Jerome Keisler, Model theory, 3rd ed., North-Holland, Amsterdam, 1990. [7] George Gratzer, Universal algebra, 2nd ed., Springer-Verlag, New York, 1979. [8] Robin Hirsch and Ian Hodkinson, Relation algebras by games, Studies in Logic and the Foundations of Mathematics, North-Holland, 2002. [9] Robin Hirsch, Marcel Jackson, and Szabolcs Mikul´as, The algebra of functions with antidomain and range, Journal of Pure and Applied Algebra 220 (2016), no. 6, 2214–2239. [10] Marcel Jackson and Tim Stokes, An invitation to C-semigroups, Semigroup Forum 62 (2001), no. 2, 279–310. [11] Marcel Jackson and Tim Stokes, Partial maps with domain and range: extending Schein's representation, Communications in Algebra 37 (2009), no. 8, 2845–2870. [12] Marcel Jackson and Tim Stokes, Modal restriction semigroups: towards an algebra of functions, Interna- tional Journal of Algebra and Computation 21 (2011), no. 7, 1053–1095. [13] Marcel Jackson and Tim Stokes, Agreeable semigroups, Journal of Algebra 266 (2003), no. 2, 393–417. [14] Brett McLean, Complete representation by partial functions for composition, intersection and antidomain, Journal of Logic and Computation 27 (2017), no. 4, 1143–1156. [15] Brett McLean and Szabolcs Mikul´as, The finite representation property for composition, intersection, domain and range, International Journal of Algebra and Computation 26 (2016), no. 6, 1199–1216. [16] John C. Reynolds, Separation logic: A logic for shared mutable data structures, 17th Annual IEEE Symposium on Logic in Computer Science, 2002, pp. 55–74. [17] Boris M. Schein, Relation algebras and function semigroups, Semigroup Forum 1 (1970), no. 1, 1–62. [18] Marshall H. Stone, Boolean algebras and their applications to topology, Proceedings of the National Academy of Sciences of the United States of America 20 (1934), no. 3, 197–202. This work is licensed under the Creative Commons Attribution-NoDerivs License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nd/2.0/ or send a letter to Creative Commons, 171 Second St, Suite 300, San Francisco, CA 94105, USA, or Eisenacher Strasse 2, 10777 Berlin, Germany
1504.01194	3	1504	2015-09-23T01:23:27	On classification of finite dimensional algebras	[ "math.RA" ]	Classification and invariants, with respect to basis changes, of finite dimensional algebras are considered. An invariant open, dense (in the Zariscki topology) subset of the space of structural constants is defined. The algebras with structural constants from this set are classified and a basis to the field of invariant rational functions of structural constants is provided.	math.RA	math	In the name of Allah, the Beneficent, the Merciful. On classification of finite dimensional algebras Ural Bekbaev Department of Science in Engineering, Faculty of Engineering, IIUM, Kuala Lumpur, Malaysia; [email protected] Abstract Classification and invariants, with respect to basis changes, of finite dimensional algebras are considered. An invariant open, dense (in the Zariski topology) subset of the space of structural constants is defined. The algebras with structural constants from this set are classified and a basis to the field of invariant rational functions of structural constants is provided. Keyword: binary operation, bilinear map, algebra, structural constants, invariant. MSC2010: 15A21, 15A63, 15A69, 17A45. 1 Introduction The classification of finite dimensional algebras is an important problem in Algebra. For example, the classification of finite dimensional simple and semi-simple associative algebras by Wedderburn, the classification of finite dimensional simple and semi-simple Lie algebras by Cartan are considered as key results in the theory of corresponding algebras. In these both and many other cases the used approach to the classification problem is structural (basis free, invariant). Unlike the structural approach in this paper we are going to deal with all algebras by the use of their structural constants. For similar approach in small dimensional case one can consider [1-3]. In general case similar approach is considered in [4]. We also consider the equivalence problem of algebras in general case. Though there are some intersecting results in this paper with [4] our approach is more simple and more constructive than of [4]. For the given dimension we show how to construct an invariant, open, dense subset of the space of structural constants and classify all algebras who's system of structural constants are in this set. We provide a basis for the field of invariant rational functions of structural constants as well. The paper is organized in the following way. The key results which are used to obtain the classification and invariants of algebras are presented in Section 2. Section 3 is a realization of Section 2 results in the case of representation of general linear group in the space structural constants. 2 Preliminaries In this section we consider a linear representation of a subgroup of the general linear group and under an assumption prove some general results about the equivalence and invariance problems with respect to this subgroup. Let n, m be any natural numbers, τ : (G, V ) → V be a fixed linear algebraic representation of an algebraic subgroup G of GL(m, F ) over V , where F is any field and V is n-dimensional vector space over F . Further we consider this representation under the following assumption: 1 Assumption.There exists a nonempty G-invariant subset V0 of V and an algebraic map P : V0 → G such that P (τ (g, v)) = P (v)g−1 whenever v ∈ V0 and g ∈ G. Theorem 2.1. Elements u, v ∈ V0 are G-equivalent, that is u = τ (g, v) for some g ∈ G, if and only if τ (P (u), u) = τ (P (v), v). Proof. If u = τ (g, v) then τ (P (u), u) = τ (P (τ (g, v)), τ (g, v)) = τ (P (v)g−1, τ (g, v)) = τ (P (v), τ (g−1, τ (g, v))) = τ (P (v), v). Visa versa, if τ (P (u), u) = τ (P (v), v) then τ (P (u)−1P (v), v) = τ ((P (u))−1, τ (P (v), v)) = τ ((P (u))−1, τ (P (u), u)) = u that is u = τ (g, v), where g = P (u)−1P (v). This proposition shows that the system of components of τ (P (x), x) is a separating system of invariants for the G-orbits in V0, where x = (x1, x2, ..., xn) is an algebraic independent system of variables over F . Further in this paper it is assumed that F is algebraically closed field, V0 in the above Assumption is an open, dense (in Zariski topology) G-invariant subset of V . In such case for any u, v ∈ V0 one has P (τ (P (u), v)) = P (v)P (u)−1 and due to density of V0 in V one has P (τ (P (y), x)) = P (x)P (y)−1, where y = (y1, y2, ..., yn) is also an algebraic independent system of variables over F . Theorem 2.2. The field of G-invariant rational functions F (x)G is generated over F by the system of components of τ (P (x), x). Proof. It is evident that all components of τ (P (x), x) are in F (x)G. If f (x) = f (τ (g, x)) for all g ∈ G then, in particular, f (x) = f (τ (P (u), x)) whenever u ∈ V0. It implies, as far as V0 is dense in V , that for the variable vector y the equality f (x) = f (τ (P (y), x)) holds true. In y = x case one gets that f (x) = f (τ (P (x), x)). Remark 2.1. In the previous variant of this paper the pure transcendence of the extension F ⊂ F (x)GL(m,F ) is stated with some justification. It turns out that the justification is not valid and therefore it is still not clear if the assumption alone is sufficient to state the pure transcendence of the extension F ⊂ F (x)G. Let V00 stand for {v ∈ V0 : P (v) = Im}, where Im stands for the m-order identity matrix. If w = τ (g, v) and w, v ∈ V00 then Im = P (w) = P (τ (g, v)) = P (v)g−1 that is g = Im. It implies that each G-orbit from V0 has only one common element with V00. Each G-invariant function on V0 is uniquely defined by its values on V00. Note that the sets G × V00 and V0 are bi-rationally isomorph. Indeed the map Q : G × V00 → V0, where Q(g, v) = τ (g, v) is invertible with Q−1 : V0 → G × V00, where Q−1(v) = (P (v)−1, τ (P (v), v)). Corollary 2.1. The field F (x) is generated over F (x)G by the system of components of P (x). Proof. Indeed F (x)G(P (x)) = F (τ (P (x), x))(P (x)) = F (τ (P (x), x), P (x)) and τ (P (x)−1, τ (P (x), x)) = x and therefore F (x)G(P (x)) = F (x). Proposition 2.1. The equality trdegF (P (x))/F = dimG holds true. 2 Proof. To prove the equality it is enough to show equality of the vanishing ideals of P (x) and G. If polynomial p vanished on P (x), that is p[P (x)] = 0, then p[P (τ (g, x, ))] = p[P (x)g−1] = 0. In particular, p[g] = 0 for any g ∈ G that is p vanishes on G as well. If p[g] = 0 for any g ∈ G then, in particular, p[P (u)] = 0 for any u ∈ V0. Due to density of V0 in V one has p[P (x)] = 0. Theorem 2.3. The equality trdegF (x)G/F = n − dimG holds true. Proof. A proof of the theorem can be deduced from Remark 1. Here is another more detailed proof. Let ]P (x) stand for any system of entries of P (x) which is a transcendence basis for the field F (P (x)) over F . We show that the system ]P (x) is algebraic independent over F (x)G as well. Indeed let p[ ^(yij)i,j=1,2,...,m] be any polynomial over F (x)G for which p[]P (x)] = 0 that is pv[]P (v)] = 0 for all v ∈ V1, where V1 is a G-invariant nonempty open subset of V0, where pv[ ^(yij)i,j=1,2,...,m] stands for the polynomial obtained from p[ ^(yij)i,j=1,2,...,m] by substitution v for x. The equality 0 = pv[]P (v)] = pτ (g,v)[ ^P (τ (g, v))] = pv[ ^P (v)g−1] implies that pv[eg] = 0 for any g ∈ G. Therefore pv[]P (x)] = 0, that is pv[ ^(yij)i,j=1,2,...,m] is zero polynomial for any v ∈ V1. It means that p[ ^(yij)i,j=1,2,...,m] is zero polynomial itself. Now due to F ⊂ F (x)G ⊂ F (x), tr.deg.F (x)/F = n and Corollary 2.1 one has the required result. Corollary 2.2 The transcendence degree of F (x)GL(m,F ) over F equals to n − m2 and the field extension F (x)GL(m,F ) ⊂ F (x) is a pure transcendental extension. For a transcendental basis one can take the system of components of P (x). Question 2.1. Under the assumption for G = GL(m, F ) is it true that F ⊂ F (x)GL(m,F ) is also a pure transcendental extension? Question 2.2. Is it a typical situation for any representation of G = GL(m, F ) that if one of the extensions F ⊂ F (x)G, F (x)G ⊂ F (x) is a pure transcendental extension then the second one also has the same property? 3 Classification of algebras 3.1 General case In this paper we use the standard notation (the Einstein notation) for tensors as well as the matrix representation for tensors which is more convenient in dealing with equivalence and invariance problems of tensors with respect to basis changes. The use of matrix representation for tensors makes the descriptions more transparent as well. Let us consider any m dimensional algebra W with multiplication · given by a bilinear map (u, v) 7→ u · v. If e = (e1, e2, ..., em) is a basis for W then one can represent the bilinear map by a matrix A ∈ Mat(m × m2; F ) such that u · v = eA(u ⊗ v) for any u = eu, v = ev, where u = (u1, u2, ..., um), v = (v1, v2, ..., vm) are column vectors. So the binary operation (bilinear map, tensor) is presented by the matrix A ∈ Mat(m × m2; F ) with respect to the basis e. Further we deal only with such matrices of rank m. If e′ = (e′1, e′2, ..., e′m) is also a basis for W , g ∈ G = GL(m, F ), e′g = e and u · v = e′B(u′ ⊗ v′), where u = e′u′, v = e′v′, then u · v = eA(u ⊗ v) = e′B(u′ ⊗ v′) = eg−1B(gu ⊗ 3 gv) = eg−1B(g ⊗ g)(u ⊗ v) as far as u = eu = e′u′ = eg−1u′, v = ev = e′v′ = eg−1v′. Therefore the equality B = gA(g−1)⊗2 (1) is valid. Now let τ stand for the representation of G = GL(m, F ) on the n = m3 dimensional vector space V = Mat(m × m2; F ) defined by τ : (g, A) 7→ B = gA(g−1 ⊗ g−1). To have Theorems 2.1-2.3 for this case we will construct a map P : V0 → GL(m, F ) with property (1) in the following way. For any natural number k due to (2) one has B⊗k = g⊗kA⊗k(g−1)⊗2k (2) Let us consider all its possible contractions with respect to k upper and k lower indices. It is clear that the result of each of such contraction will be f (B) = f (A)(g−1)⊗k type equality, where f (A) is a row vector with mk entries. In k = 1 case one gets the following 211! = 2 different row equalities: Tr1(B) = Tr1(A)g−1, Tr2(B) = Tr2(A)g−1, where Tr1(A) stands for the row vector with entries Aj j,i- the contraction on the first upper and lower indices and Tr2(A) stands for the row vector with entries Aj i,j- the contraction on the first upper and second lower indices. j,i =Pn i,j = Pn j=1 Aj j=1 Aj In k = 2 case one gets the following 222! + 211! = 10 different row equalities: Tri(B) ⊗ Trj(B) = (Tri(A) ⊗ Trj(A))(g−1)⊗2, Tri(B)B = Tri(A)A(g−1)⊗2, where i, j = 1, 2, and (Bi j,pBj i,q) = (Ai j,pAj i,q)(g−1)⊗2, (Bi j,pBj q,i) = (Ai j,pAj q,i)(g−1)⊗2, (Bi p,jBj i,q) = (Ai p,jAj i,q)(g−1)⊗2, (Bi p,jBj q,i) = (Ai p,jAj q,i)(g−1)⊗2. In any k case only the number of contractions of A⊗k when all k different upper indices are contracted with lower indices of different A is (2k) × (2(k − 1)) × (2(k − 2)) × ... × 2 = 2kk!. In general it is nearly clear that the corresponding resulting system of 2kk! rows depending on the variable matrix A := X = (X i j,k)i,j,k=1,2,...,m is linear independent over F . But for big enough k the inequality 2kk! ≥ mk holds true as well. Therefore in general for big enough k it is possible to choose mk contractions (rows) among the all contractions of X ⊗k for which the matrix Q(X) consisting of these mk rows is a nonsingular matrix. For the matrix Q(X) one has equality Q(Y ) = Q(X)(g−1)⊗k whenever g ∈ G, Y = gX(g−1)⊗2. Now note that for any A ∈ {X : det(Q(X)) 6= 0} and g ∈ G one has, for example, (B ⊗ (Tr1(B))⊗k−2)Q(B)−1 = g(A ⊗ (Tr1(A))⊗k−2)(g−1)⊗k(Q(A)(g−1)⊗k)−1 = g(A ⊗ (Tr1(A))⊗k−2)Q(A)−1. Therefore if P (A)−1 stands for arbitrary nonsingular m × m size sub-matrix of (A ⊗ (Tr1(A))⊗k−2)Q(A)−1 then one has the equality P (B)−1 = gP (A)−1, where g ∈ G, B = 4 gA(g−1)⊗2. It implies that whenever A ∈ V0 = {A : det(P (A)) det(Q(A)) 6= 0} the equality P (B) = P (A)g−1 holds true for any g ∈ G and B = gA(g−1)⊗2. Note that V0 = {A : det(P (A)) det(Q(A)) 6= 0} is a G-invariant, open and dense subset of V . Therefore we have the following results. Theorem 3.1. Two algebras with matrices of structural constants A, B ∈ V0 are same (isomorph) algebras if and only if P (A)A(P (A)−1 ⊗ P (A)−1) = P (B)B(P (B)−1 ⊗ P (B)−1). Theorem 3.2. The field of G-invariant rational functions F (x)G of structural constants j,k)i,j,k=1,2,...,m is generated by the system of entries of defined by variable matrix x = (xi P (x)x(P (x)−1 ⊗ P (x)−1) over F , that is the equality F (x)G = F (P (x)x(P (x)−1 ⊗ P (x)−1)) holds true. Theorem 3.3. The transcendence degree of F (x)G over F equals to m3 − m2 and the field extension F (x)G ⊂ F (x) is a pure transcendental extension. Remark 3.1. One of the main results (Theorem 1) of [4] states that the field extension F ⊂ F (x)GL(m<F ) is a pure transcendental extension, which we could not get by our approach. Theorem 3.3 can be considered as a complementary result to that Theorem 1. Now let us consider two and three dimensional algebra cases. Example 3.l. Two dimensional (m = 2) case. Let A =(cid:18) A1 A2 1,1 A1 1,1 A2 1,2 A1 1,2 A2 2,1 A1 2,2 2,1 A2 2,2 (cid:19) be the matrix of structural constants with respect to a basis. In this case at k = 1 already 211! = m1 and therefore for the rows of P (A) on can take Tr1(A) = (A1 1,1 + A2 2,1, A1 1,2 + A2 2,2) and Tr2(A) = (A1 1,1 + A2 1,2, A1 2,1 + A2 2,2) and V0 consists of all A for which det P (A) = (A1 1,1 + A2 2,1)(A1 2,1 + A2 2,2) − (A1 1,2 + A2 2,2)(A1 1,1 + A2 1,2) 6= 0. To see the corresponding system of generators one can evaluate P (x)x(P (x)−1 ⊗ P (x)−1), where x =(cid:18) x1 1,1 x1 x2 1,1 x2 1,2 x1 1,2 x2 2,1 x1 2,2 2,1 x2 2,2 (cid:19) . On classification problem of two dimensional algebras one can see [1,2]. Example 3.2. Three dimensional (m = 3) case. Let A =  A1 A2 A31 1,1 A1 1,1 A2 1,1 A3 1,2 A1 1,2 A2 1,2 A3 1,3 A1 1,3 A2 1,3 A3 2,1 A1 2,1 A2 2,1 A3 2,2 A1 2,2 A2 2,2 A3 2,3 A1 2,3 A2 2,3 A3 3,1 A1 3,1 A2 3,1 A3 3,2 A1 3,3 3,2 A2 3,3 3,2 A3 3,3   be the matrix of the structural constants with respect to a basis. In this case at k = 1 one has 211! < 31. At k = 2 already 222! + 211! = 10 > 32 and the following 10 equalities 5 Tri(B) ⊗ Trj(B) = Tri(A) ⊗ Trj(A)(g−1)⊗2, Tri(B)B = Tri(A)A(g−1)⊗2, where i, j = 1, 2, (Bi j,pBj i,q) = (Ai j,pAj i,q)(g−1)⊗2, (Bi j,pBj q,i) = (Ai j,pAj q,i)(g−1)⊗2, (Bi p,jBj i,q) = (Ai p,jAj i,q)(g−1)⊗2, (Bi p,jBj q,i) = (Ai p,jAj q,i)(g−1)⊗2 hold true. Therefore, for example, for Q(A) one can take the following matrix Q(A) =   2,j j,2 Ai Ai Ai Ai Ai Ai Ai Ai Ai i,1Aj i,1Aj 1,iAj 1,iAj i,jAj j,iAj j,1Aj j,1Aj 1,jAj j,1 Ai 1,j Ai j,1 Ai 1,j Ai 1,1 Ai 1,1 Ai i,1 Ai 1,i Ai i,1 Ai i,1Aj i,1Aj 1,iAj 1,iAj i,jAj j,iAj j,1Aj j,1Aj 1,jAj j,2 Ai 2,j Ai j,2 Ai 2,j Ai 1,2 Ai 1,2 Ai i,2 Ai 2,i Ai i,2 Ai i,1Aj i,1Aj 1,iAj 1,iAj i,jAj j,iAj j,1Aj j,1Aj 1,jAj j,3 Ai 3,j Ai j,3 Ai 3,j Ai 1,3 Ai 1,3 Ai i,3 Ai 3,i Ai i,3 Ai i,2Aj i,2Aj 2,iAj 2,iAj i,jAj j,iAj j,2Aj j,2Aj 2,jAj j,1 Ai 1,j Ai j,1 Ai 1,j Ai 2,1 Ai 2,1 Ai i,1 Ai 1,i Ai i,1 Ai i,2Aj i,2Aj 2,iAj 2,iAj 2,j i,jAj 2,2 j,iAj 2,2 j,2Aj i,2 j,2Aj 2,jAj j,2 2,i i,2 3,j j,3 Ai Ai Ai Ai Ai Ai Ai Ai Ai i,2Aj i,2Aj 2,iAj 2,iAj i,jAj j,iAj j,2Aj j,2Aj 2,jAj j,3 Ai 3,j Ai j,3 Ai 3,j Ai 2,3 Ai 2,3 Ai i,3 Ai 3,i Ai i,3 Ai i,3Aj i,3Aj 3,iAj 3,iAj i,jAj j,iAj j,3Aj j,3Aj 3,jAj j,1 Ai 1,j Ai j,1 Ai 1,j Ai 3,1 Ai 3,1 Ai i,1 Ai 1,i Ai i,1 Ai i,3Aj i,3Aj 3,iAj 3,iAj i,jAj j,iAj j,3Aj j,3Aj 3,jAj j,2 Ai 2,j Ai j,2 Ai 2,j Ai 3,2 Ai 3,2 Ai i,2 Ai 2,i Ai i,2 Ai i,3Aj i,3Aj 3,iAj 3,iAj 3,j i,jAj 3,3 j,iAj 3,3 j,3Aj j,3Aj 3,jAj j,1 i,3 i,3 3,i   . For P (A)−1 one can take any 3 × 3 size nonsingular sub-matrix of (A ⊗ Tr1(A))Q(A)−1, where (A ⊗ Tr1(A)) =   A1 A2 A3 1,1Ai 1,1Ai 1,1Ai i,1 A1 i,1 A2 i,1 A3 1,1Ai 1,1Ai 1,1Ai i,2 A1 i,2 A2 i,2 A3 1,1Ai 1,1Ai 1,1Ai i,3 A1 i,3 A2 i,3 A3 1,2Ai 1,2Ai 1,2Ai i,1 A1 i,1 A2 i,1 A3 1,2Ai i,2 1,2Ai i,2 1,2Ai i,2 A1 A2 A3 1,2Ai 1,2Ai 1,3Ai i,3 A1 i,3 A2 i,3 A3 1,3Ai 1,3Ai 1,3Ai i,1 A1 i,1 A2 i,1 A3 1,3Ai 1,3Ai 1,3Ai i,2 A1 i,2 A2 i,2 A3 1,3Ai i,3 1,3Ai i,3 1,3Ai i,3   . 3.2 Commutative and anti-commutative algebra cases For the classification purpose instead of all m dimensional algebras one can consider only such commutative or anti-commutative algebras. The commutativity (anti-commutativity) 6 of the binary operation in terms of the corresponding matrix A means Ai (respectively, Ai algebra case for the V we consider V = k,j) for all i, j, k = 1, 2, ..., m. So in commutative (anti-commutative) j,k = −Ai j,k = Ai k,j {A ∈ Mat(m × m2; F ) : Ai j,k = Ai k,j(resp. Ai j,k = −Ai k,j) for all i, j, k = 1, 2, ..., m.} Note that in commutative (anti-commutative) case the dimension of V is m(m+1) 2 (respectively, m(m−1) ). 2 To have Theorems 2.1-2.3 for these cases one can construct a map P : V0 → GL(m, F ) with property (1) in a similar way as in the general algebra case. Consider once again equality (3) and all its possible contractions with respect to k upper and k lower indices. In commutative (anti-commutative) case at k = 1 one gets the following 1! = 1 k,j (respectively, row equality: Tr1(B) = Tr1(A)g−1 = Tr2(A)g−1 as far as Ai Tr1(B) = Tr1(A)g−1 = −Tr2(A)g−1 as far as Ai k,j) for all i, j, k = 1, 2, ..., m. j,k = −Ai j,k = Ai In k = 2 case one gets the following 2! + 1! = 3 different row equalities: Tr1(B) ⊗ Tr1(B) = (Tr1(A) ⊗ Tr1(A))(g−1)⊗2, Tr1(B)B = Tr1(A)A(g−1)⊗2, (Bi j,pBj i,q) = (Ai j,pAj i,q)(g−1)⊗2. In any k case only the number of contractions of A⊗k when all k different upper indices are contracted with lower indices of different A is k!. Once again in general it is nearly clear that the corresponding resulting system of k! rows depending on variable matrix A := x = (xi k,j) for all i, j, k = 1, 2, ..., m, is linear independent over F . But for big enough k the inequality k! ≥ mk holds true as well. Therefore in general for big enough k it is possible to choose mk contractions (rows) among the all contractions of x⊗k for which the matrix Q(x) consisting of these mk rows is nonsingular. For the matrix Q(x) one has equality Q(y) = Q(x)(g−1)⊗k whenever g ∈ G, y = gx(g−1)⊗2. j,k)i,j,k=1,2,...,m, where xi k,j (respectively, xi j,k = −xi j,k = xi Therefore if P (A)−1 stands for arbitrary m × m-size nonsingular sub-matrix of (A ⊗ (Tr1(A))⊗k−2)Q(A)−1 then one has the equality P (B)−1 = gP (A)−1, where g ∈ G, B = gA(g−1)⊗2. It implies that whenever A ∈ V0 = {A ∈ V : det(P (A)) det(Q(A)) 6= 0} the equality P (B) = P (A)g−1 holds true for any g ∈ G, where B = gA(g−1)⊗2. Note that V0 = {A ∈ V : det(P (A)) det(Q(A)) 6= 0} is a G-invariant, open and dense subset of V . Therefore we have the following results. Theorem 3.1'. Two commutative (anti-commutative) algebras with the matrices of structural constants A, B ∈ V0 are the same algebras if and only if P (A)A(P (A)−1 ⊗ P (A)−1) = P (B)B(P (B)−1 ⊗ P (B)−1). Theorem 3.2'. The field of G-invariant rational functions F (x)G of the structural constants presented by the matrix x = ((xi j,k)i,j,k=1,2,...,m of the variable commutative (respectively, anti-commutative) algebras, where xi k,j) for all i, j, k = 1, 2, ..., m, is generated by the system of entries of P (x)x(P (x)−1 ⊗P (x)−1) over F , that is the equality k,j (respectively, xi j,k = −xi j,k = xi F (x)G = F (P (x)x(P (x)−1 ⊗ P (x)−1)) 7 holds true. Theorem 3.3'. In commutative (anti-commutative) algebra case the transcendence , m ≥ 3) and the field degree of F (x)G over F equals to m2(m−1) extension F (x)G ⊂ F (x) is a pure transcendental extension. (respectively, m2(m−3) 2 2 Now let us consider two dimensional commutative algebra case. Example 3.1'. Let A =(cid:18) A1 A21 1,1 A1 1,1 A2 1,2 A1 1,2 A2 2,1 A1 2,2 2,1 A2 2,2 (cid:19) 1,2 = Ai , where Ai 2,1 at i = 1, 2, be the matrix of structural constants of a commutative algebra with respect to a basis. Consider B⊗3 = g⊗3A(g−1)⊗6 and its all contractions on 3 upper and 3 lower indices. Among them in particular one gets the following 6 equalities: (Bi σ(i),pBj σ(j),qBk σ(k),r) = (Ai σ(i),pAj σ(j),qAk σ(k),r)(g−1)⊗3 , where σ ∈ S3- the symmetric group of permutations of symbols i, j, k, and (Bi p,qBj r,kBk i,j) = (Ai p,qAj r,jAk i,j)(g−1)⊗3, (Bi p,qBj r,iBk k,j) = (Ai p,qAj r,iAk k,j)(g−1)⊗3. So for Q(A) one can take the matrix consisting of the following 8 rows (Ai σ(i),pAj σ(j),qAk σ(k),r)σ∈S3, (Ai p,qAj r,kAk i,j), (Ai p,qAj r,iAk k,j) and for the P (A)−1 any nonsingular 2×2 -size sub-matrix of (A⊗T r1(A))Q(A)−1 provided that det(Q(A)) 6= 0. On classification of three dimensional anti-commutative algebras one can see [3]. Remark 3.2. One can consider classification of finite dimensional (general, commutative, anti-commutative) algebras with respect to subgroups of the general linear group as well. By the use of Section 3 results such classifications can be obtained for all classical subgroups of the general linear group. Acknowledgement. After the publication of the first variant of this paper the author was kindly informed by the author of [4] about his work on the same topic [4], which I have failed to notice before. It helped me to understand better the problems related to the classification of finite dimensional algebras. References [1] Michel Goze and Elisabeth Remm, 2-dimensional algebras, African Journal of Mathematical Physics, v. 10(2011),81-91. [2] R. Dur´an D´ıaz et al, Classifying quadratic maps from plane to plane, Linear Algebra and its Applications 364 (2003),pp. 1-12. [3] L. Hern´aandez Encinas, et al, Non-degenerate bilinear alternating maps f : V ×V → V, dim(V ) = 3, over an algebraically closed field. Linear Algebra and its Applications 387 (2004),pp. 689-82. [4] V. Popov, Generic Algebras: Rational Parametrization and Normal Forms, arXiv: 1411.6570v2[math.AG]. 8
1710.07141	1	1710	2017-10-19T13:45:43	Finite generation of some cohomology rings via twisted tensor product and Anick resolutions	[ "math.RA", "math.RT" ]	Over a field of prime characteristic $p>2$, we prove that the cohomology rings of some pointed Hopf algebras of dimension $p^3$ are finitely generated. These are Hopf algebras arising in the ongoing classification of finite dimensional Hopf algebras in positive characteristic, and include bosonizations of Nichols algebras of Jordan type in a general setting as well as their liftings when $p=3$. Our techniques are applications of twisted tensor product resolutions and Anick resolutions in combination with May spectral sequences.	math.RA	math	FINITE GENERATION OF SOME COHOMOLOGY RINGS VIA TWISTED TENSOR PRODUCT AND ANICK RESOLUTIONS VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON Abstract. Over a field of prime characteristic p > 2, we prove that the cohomology rings of some pointed Hopf algebras of dimension p3 are finitely generated. These are Hopf algebras arising in the ongoing classification of finite dimensional Hopf algebras in positive characteristic, and include bosonizations of Nichols algebras of Jordan type in a general setting as well as their liftings when p = 3. Our techniques are applications of twisted tensor product resolutions and Anick resolutions in combination with May spectral sequences. Contents Introduction 1. 2. Settings 2.1. Pointed Hopf algebras 2.2. Our setting: Rank two Nichols algebra and its bosonization 2.3. Our setting: A class of 27-dimensional pointed Hopf algebras 3. Twisted tensor product resolutions 3.1. The resolution construction 3.2. Resolution over the Nichols algebra R 3.3. Resolution over the bosonization R# kG 4. Anick resolutions 4.1. The resolution construction 4.2. A truncated polynomial ring 5. Finite generation of some cohomology rings 5.1. Cohomology of the Nichols algebra and its bosonization 5.2. Cohomology of some pointed Hopf algebras of dimension 27 References 1 2 3 3 4 4 4 5 11 18 18 22 25 25 27 29 1. Introduction The cohomology ring of a finite dimensional Hopf algebra is conjectured to be finitely gen- erated. Friedlander and Suslin [8] proved this for cocommutative Hopf algebras, generalizing earlier results of Evens [6], Golod [10], and Venkov [24] for finite group algebras and of Fried- lander and Parshall [7] for restricted Lie algebras. There are many finite generation results as well for various types of noncocommutative Hopf algebras (see, e.g., [3, 9, 11, 14, 22]). Date: October 18, 2017. 2010 Mathematics Subject Classification. 16E05, 16E40, 16T05. Key words and phrases. cohomology, positive characteristic, (pointed) Hopf algebras, Anick resolutions, twisted tensor products. The third author was partially supported by NSF grant DMS-1401016. 1 2 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON Most of these results are in characteristic 0. In this paper, we prove finite generation for classes of noncocommutative Hopf algebras in prime characteristic p > 2. These are some of the pointed Hopf algebras arising in classification work of the first two authors. Our main result is the following combination of Theorems 5.1.2 and 5.2.1 below: Theorem. Let k be an algebraically closed field of prime characteristic p > 2. Consider the following Hopf algebras over k: (1) the p2q-dimensional bosonization R# kG of a rank two Nichols algebra R of Jordan type over a cyclic group G of order q, where q is divisible by p; and (2) a lifting H of R# kG when p = q = 3. Then the cohomology rings of R# kG and of H are finitely generated. Our theorem is exclusively an odd characteristic result since the Nichols algebra of Jordan type does not appear in characteristic 2. Instead there is another related Nichols algebra [4] that will require different techniques. Part (2) of our main theorem above is only stated for characteristic 3; this is because we use the classification of such Hopf algebras from [18], and liftings are only known completely in this case. We expect our homological techniques will be able to handle liftings when p > 3 once more is known about their structure. More specifically, we let R# kG be a p2q-dimensional Hopf algebra given by the bosoniza- tion of a Jordan plane as introduced in [4] (see Section 2.2 below). We prove in Theo- rem 5.1.2 that the cohomology ring of such a Hopf algebra R#kG, that is H∗(R#kG, k) := Ext∗ R#kG(k, k), is finitely generated. We also consider liftings H of R#kG in the special case p = q = 3 (see Section 2.3 below). We prove in Theorem 5.2.1 that the cohomology ring H∗(H, k) is finitely generated. Our techniques for proving the main theorem above rely on the May spectral sequence for the cohomology of a filtered algebra. In either setting, (1) or (2), we may choose a filtration for which the associated graded algebra is a truncated polynomial ring whose cohomology is straightforward. The hard work is in finding some permanent cycles as required to use a spectral sequence lemma that goes back to Friedlander and Suslin [8]. This we do in two ways, by constructing two types of resolutions which should be of independent interest. The general definitions of the resolutions are not new, but we provide here some nontrivial examples, with our main theorem as an application. The first resolution is the twisted tensor product resolution of [20], recalled in Section 3.1, used in Section 3.2, and iterated in Section 3.3, to obtain a resolution over the bosonization R#kG of the Nichols algebra R. This is used in Section 5.1 to prove finite generation of its cohomology. The second resolution is the Anick resolution [1], explained in Section 4.1, with a general result for the Anick resolution of a truncated polynomial ring in Section 4.2. This Anick resolution is used in Section 5.2 to prove finite generation of the cohomology of the liftings H. We could have chosen to work with just one type of resolution, either Anick or twisted tensor product, for proofs of both parts (1) and (2) of our main theorem above. We instead chose to work with both resolutions to illustrate a wider variety of techniques available, each having its own advantages. 2. Settings Throughout, let k be a field. For our main results we will assume that k is algebraically closed field of prime characteristic p > 2 since we work with Hopf algebras found in classifica- tion work under this assumption. The tensor product ⊗ is ⊗k unless specified otherwise. In COHOMOLOGY OF POINTED HOPF ALGEBRAS 3 this section we will define the Nichols algebras and pointed Hopf algebras that are featured in this paper, and summarize some structural results that will be needed. 2.1. Pointed Hopf algebras. Let H be any finite dimensional pointed Hopf algebra over k. The coradical (the sum of all simple subcoalgebras) of H is H0 = kG, a Hopf subalgebra of H generated by the grouplike elements G := {g ∈ H ∆(g) = g ⊗ g}, where ∆ is the coproduct on H. Let H0 ⊆ H1 ⊆ H2 ⊆ · · · ⊆ H be the coradical filtration of H, where Hn = ∆−1(H ⊗ Hn−1 + H0 ⊗ H) inductively, see [17, Chapter 5]. Consider the associated graded Hopf algebra gr H = Ln≥0 Hn/Hn−1, with the convention H−1 = 0. Note that the zero term of gr H equals its coradical, i.e., (gr H)0 = H0. There is a projection π : gr H → H0 and an inclusion ι : H0 → gr H such that πι = 1 (the identity map on H0). Let R be the algebra of coinvariants of π: R := (gr H)co π = {h ∈ gr H : (1H ⊗ π)∆(h) = h ⊗ 1}, where 1H denotes the identity map on H. By results of Radford [23] and Majid [15], R is a Hopf algebra in the braided category G GYD of left Yetter-Drinfeld modules over H0 = kG. Moreover, gr H is the bosonization (or Radford biproduct) of R and H0 so that gr H ∼= R#H0 with the Hopf structure given in [17, Theorem 10.6.5]. As an algebra, it is simply the smash product of H0 with R, analogous to a semidirect product of groups. 2.2. Our setting: Rank two Nichols algebra and its bosonization. For Sections 3 and 5.1, we use the following setup. Let k be a field of prime characteristic p > 2. Let G := hgi ∼= Z/qZ be a cyclic group whose order q is divisible by p. Consider R := khx, yi.(cid:18)xp, yp, yx − xy − 1 2 x2(cid:19) , which is a p2-dimensional algebra as described in [4, Theorem 3.5], with a vector space basis {xiyj 0 ≤ i, j ≤ p − 1}. Observe that R is the Nichols algebra of a rank two Yetter-Drinfeld module V = kx + ky over G, where the G-action on V is given by gx = x and gy = x + y, (here the left superscript indicates group action), and the G-gradings of x and y are both given by g. Let R# kG be the bosonization of R and kG. It is the corresponding p2q-dimensional pointed Hopf algebra as studied in [4, Corollary 3.14], [18, §3 (Case B)], and [19, §4]. Its Hopf structure is given by: ∆(x) = x ⊗ 1 + g ⊗ x, ∆(y) = y ⊗ 1 + g ⊗ y, ∆(g) = g ⊗ g, ε(x) = 0, S(x) = −g−1x, ε(y) = 0, S(y) = −g−1y, ε(g) = 1, S(g) = g−1, where ∆ is the coproduct, ε is the counit, and S is the antipode map of R# kG. Remark 2.2.1. We remark the following: (1) These R, G, R# kG appear in [4, 18, 19] for various purposes. In their settings, G is a cyclic group of order p. Here, we consider a more general setting with G being cyclic of order q divisible by p. 4 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON (2) In [18], the first two authors classified p3-dimensional pointed Hopf algebra over prime characteristic p. In their classification work, this p2-dimensional Nichols algebra R of Jordan type is unique, up to isomorphism, and only occurs when p > 2. (3) The authors of [4] used right modules in their settings, whereas left modules were used in the classification work in [18], thus inducing a sign difference in the relation yx − xy − 1 2x2 of R there. Here, we adopt the relation described in [4]. In Section 5.2, when we study the cohomology of pointed Hopf algebras H lifted from the associated graded algebras gr H ∼= R# kG, we modify the relations of the lifting structure given in [18] accordingly, see the next Section 2.3. 2.3. Our setting: A class of 27-dimensional pointed Hopf algebras. Let k be a field of characteristic p = 3 and consider Hopf algebras H(ǫ, µ, τ ) defined from three scalar pa- rameters ǫ, µ, τ as in [18]. These are pointed Hopf algebras of dimension 27 whose associated graded algebra is gr H(ǫ, µ, τ ) ∼= R# kG, the Hopf algebra described in Section 2.2, in this case p = q = 3. As an algebra, H(ǫ, µ, τ ) is generated by g, x, y with relations g3 = 1, x3 = ǫx, yg − gy = xg + µ(g − g2), y3 = −ǫy2 − (µǫ − τ − µ2)y, xg − gx = −ǫ(g − g2), yx − xy = −x2 + (µ + ǫ)x + ǫy + τ (1 − g2), where ǫ ∈ {0, 1} and τ, µ ∈ k are arbitrary scalars. The coalgebra structure is the same as that of R#kG described in Section 2.2. By setting w = g − 1, we get a new presentation in which the generators are w, x, y and the relations are: w3 = 0, x3 = ǫx, y3 = −ǫy2 − (µǫ − τ − µ2)y, yw − wy = wx + x − (µ − ǫ)(w2 + w), xw − wx = ǫ(w2 + w), yx − xy = −x2 + (µ + ǫ)x + ǫy − τ (w2 − w). This choice of generating set will be convenient for our homological arguments later. As shown in [18], H(ǫ, µ, τ ) has dimension 27, with vector space basis {wixjyk 0 ≤ i, j, k ≤ 2}. 3. Twisted tensor product resolutions In this section, we apply the construction of twisted tensor product resolutions introduced in [20] to our Nichols algebra R and its bosonization R# kG that was defined in Section 2.2. 3.1. The resolution construction. Let A and B be associative algebras over k with mul- tiplication maps mA : A ⊗ A → A and mB : B ⊗ B → B, and multiplicative identities 1A and 1B, respectively. We write 1 for the identity map on any set. A twisting map τ : B ⊗A → A⊗B is a bijective k-linear map for which τ (1B ⊗a) = a⊗1B and τ (b ⊗ 1A) = 1A ⊗ b, for all a ∈ A and b ∈ B, and (3.1.1) τ ◦ (mB ⊗ mA) = (mA ⊗ mB) ◦ (1 ⊗ τ ⊗ 1) ◦ (τ ⊗ τ ) ◦ (1 ⊗ τ ⊗ 1) as maps B ⊗ B ⊗ A ⊗ A → A ⊗ B. The twisted tensor product algebra A ⊗τ B is the vector space A ⊗ B together with multiplication mτ given by such a twisting map τ , that is, mτ : (A ⊗ B) ⊗ (A ⊗ B) → A ⊗ B is given by mτ = (mA ⊗ mB) ◦ (1 ⊗ τ ⊗ 1). COHOMOLOGY OF POINTED HOPF ALGEBRAS 5 Definition 3.1.2. [20, Definition 5.1] Let M be an A-module with module structure map ρA,M : A ⊗ M → M. We say M is compatible with the twisting map τ if there is a bijective k-linear map τB,M : B ⊗ M → M ⊗ B such that (3.1.3) (3.1.4) τB,M ◦ (mB ⊗ 1) = (1 ⊗ mB) ◦ (τB,M ⊗ 1) ◦ (1 ⊗ τB,M ), and τB,M ◦ (1 ⊗ ρA,M ) = (ρA,M ⊗ 1) ◦ (1 ⊗ τB,M ) ◦ (τ ⊗ 1) as maps on B ⊗ B ⊗ M and on B ⊗ A ⊗ M, respectively. Let M be an A-module that is compatible with τ . We say a projective A-module resolution P q(M) of M is compatible with the twisting map τ if each module Pi(M) is compatible with τ via maps τB,i for which τB, q : B ⊗ P q(M) → P q(M) ⊗ B is a k-linear chain map lifting τB,M : B ⊗ M → M ⊗ B. Let N be a B-module and let P q(N) be a projective resolution of N over B. We put an A ⊗τ B-module structure on the bicomplex P q(M) ⊗ P q(N) by using maps τB, q. Under such compatibility conditions, the twisted tensor product resolutions for left mod- ules over A ⊗τ B were constructed in [20] satisfying the following theorem. Theorem 3.1.5. [20, Theorem 5.12] Let A and B be k-algebras with twisting map τ : B ⊗ A → A ⊗ B. Let P q(M) be an A-projective resolution of M and P q(N) be a B-projective resolution of N. Assume (a) M and P q(M) are compatible with τ , and (b) Yi,j = Pi(M) ⊗ Pj(N) is a projective A ⊗τ B-module, for all i, j. Then the twisted tensor product complex Y q = Tot(P q(M) ⊗ P q(N)) with is a projective resolution of M ⊗ N as a module over the twisted tensor product A ⊗τ B. Yn = Mi+j=n Pi(M) ⊗ Pj(N) By Tot(P q(M) ⊗ P q(N)) we mean the total complex of the bicomplex P q(M) ⊗ P q(N), that is the complex whose nth component is Yn = ⊕i+j=n(Pi(M) ⊗ Pj(N)) and differential is dn =Pi+j=n dij where dij = di ⊗ 1 + (−1)i1 ⊗ dj. In the statement of [20, Theorem 5.12], we have replaced one of the hypotheses by our hypothesis (b) above, and in this case the proof is given by [20, Lemmas 5.8 and 5.9]. In some contexts, such as ours here, the hypothesis (b) can be checked directly, in which case this version of the theorem is sufficient. For the rest of this section, let R, G be as defined in Section 2.2. Let k also denote the trivial (left) R- or (R# kG)-module, that is, the field k with action given by the augmentation ε(x) = 0, ε(y) = 0, ε(g) = 1. We will construct a resolution of k as an R-module via a twisted tensor product, and as an (R# kG)-module by iterating the twisted product construction. 3.2. Resolution over the Nichols algebra R. Let k be a field of characteristic p > 2, and let R be the Nichols algebra described in Section 2.2. Let k be the R-module on which x and y both act as 0. We start with a construction of a resolution of k as an R-module. Using the relation (3.9) in [4, Lemma 3.8], we can view R as the twisted tensor product R = khx, yi.(cid:18)xp, yp, yx − xy − 1 2 x2(cid:19) ∼= A ⊗τ B, 6 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON where A := k[x]/(xp), B := k[y]/(yp). The twisting map τ : B ⊗ A → A ⊗ B is defined by τ (yr ⊗ xℓ) = where we use the convention: r Xt=0 (cid:18)r 2(cid:19)t t(cid:19)(cid:18)1 [ℓ][t] xℓ+t ⊗ yr−t, [ℓ][t] = ℓ(ℓ + 1)(ℓ + 2) · · · (ℓ + t − 1), with [ℓ][0] = 1, for any ℓ. Consider the following free resolutions of k as A-module and as B-module, respectively: P q(A) : · · · xp−1· / A x· / A xp−1· / A x· / A ε / k P q(B) : yp−1· / B y· / B · · · yp−1· / B y· / B ε / k / 0 / 0. The map ε on A (respectively, on B) takes x to 0 (respectively, y to 0). Consider the tensor product P q(A) ⊗ P q(B) as a graded vector space. The total complex is a complex of vector spaces with differential in degree n given by (di ⊗ 1 + (−1)i ⊗ dj). dn = Xi+j=n We will put the structure of an R-module on each Pi(A) ⊗ Pj(B) so that this tensor product complex is an R-projective resolution of k as an R-module. This will follow from [20, Lemmas 5.8 and 5.9] once we define a chain map as in Definition 3.1.2 and check that the resulting R-modules are indeed projective. That is, as in Definition 3.1.2, there are bijective k-linear maps τB,i that we abbreviate here as τi, for which the diagram · · · · · · 1⊗xp−1· / B ⊗ A 1⊗x· / B ⊗ A 1⊗xp−1· / B ⊗ A 1⊗x· / B ⊗ A 1⊗ε / B ⊗ k τ3 τ2 τ1 τ0 ∼= xp−1·⊗1 / A ⊗ B x·⊗1 / A ⊗ B xp−1·⊗1 / A ⊗ B x·⊗1 / A ⊗ B ε⊗1 / k ⊗ B 0 / 0 commutes and conditions (3.1.3) and (3.1.4) hold. We claim that the following maps τi satisfy the above conditions. Lemma 3.2.1. For any integer i ≥ 0, let τi : B ⊗ A → A ⊗ B be defined as follows: τi(yr ⊗ xℓ) =  τ (yr ⊗ xℓ), Pr t=0(cid:0)r 2(cid:1)t [ℓ + 1][t] xℓ+t ⊗ yr−t, t(cid:1)(cid:0) 1 i is even i is odd. Then (a) τi is a bijective k-linear map whose inverse is τ −1 i (xℓ ⊗ yr) =  Pr t=0(cid:0)r Pr t=0(cid:0)r t(cid:1)(cid:0)− 1 t(cid:1)(cid:0)− 1 2(cid:1)t [ℓ][t] yr−t ⊗ xℓ+t, 2(cid:1)t [ℓ + 1][t] yr−t ⊗ xℓ+t, i is even i is odd. / / / / / / / / / / / / / / / / / / / / / / / / / COHOMOLOGY OF POINTED HOPF ALGEBRAS 7 (b) τi satisfies conditions (3.1.3) and (3.1.4). In particular, τi ◦ (mB ⊗ 1) = (1 ⊗ mB) ◦ (τi ⊗ 1) ◦ (1 ⊗ τi) and τi ◦ (1 ⊗ mA) = (mA ⊗ 1) ◦ (1 ⊗ τi) ◦ (τ ⊗ 1), as maps on B ⊗ B ⊗ A and on B ⊗ A ⊗ A, respectively. (c) Each square in the above diagram commutes. Consequently, k and its resolution P q(A) are compatible with τ . Proof. Let τi be defined as in the lemma. (a) Bijection: To show τi is bijective with the given inverse, it suffices to show that τi ◦τ −1 1A⊗B and τ −1 the remaining case is similar. i = i = 1A⊗B for the case when i is odd; ◦ τi = 1B⊗A. Here, we will check τi ◦ τ −1 i τi ◦ τ −1 i 1 1 r−k [ℓ + 1][k] 2(cid:19)k [ℓ + 1][k] yr−k ⊗ xℓ+k! k(cid:19)(cid:18)− 2(cid:19)k 2(cid:19)s s (cid:19)(cid:18)1 Xs=0(cid:18)r − k 2(cid:19)k(cid:18)1 2(cid:19)s s (cid:19)(cid:18)− k(cid:19)(cid:18)r − k 2(cid:19)k+s s (cid:19)(−1)k(cid:18)1 k(cid:19)(cid:18)r − k 1 r r = = r−k (xℓ ⊗ yr) = τi r Xk=0(cid:18)r k(cid:19)(cid:18)− Xk=0(cid:18)r Xs=0(cid:18)r Xk=0 Xt=0 Xk+s=t(cid:18)r s (cid:19)(−1)k(cid:18)1 k(cid:19)(cid:18)r − k Xk+s=t(cid:18)r = r It suffices to show [ℓ + k + 1][s] xℓ+k+s ⊗ yr−k−s [ℓ + 1][k][ℓ + k + 1][s] xℓ+k+s ⊗ yr−k−s [ℓ + 1][k][ℓ + k + 1][s]! xℓ+t ⊗ yr−t. 2(cid:19)k+s [ℓ + 1][k][ℓ + k + 1][s] =(1, 0, if t = 0 if t 6= 0. The case t = 0 is clear, as k = s = 0, all coefficients become 1. For t > 0, t Xk=0(cid:18)r = [ℓ + 1][k][ℓ + k + 1][t−k] (ℓ + t)! 2(cid:19)t t − k(cid:19)(−1)k(cid:18)1 k(cid:19)(cid:18)r − k 2(cid:19)t t (−1)k(cid:18)r ℓ! (cid:18) 1 Xk=0 2(cid:19)t t (−1)k(cid:18)r ℓ! (cid:18) 1 Xk=0 t(cid:19) t 2(cid:19)t(cid:18)r ℓ! (cid:18) 1 Xk=0 (ℓ + t)! (ℓ + t)! = = t − k(cid:19)! k(cid:19)(cid:18)r − k k(cid:19)! t(cid:19)(cid:18)t k(cid:19)! = 0. (−1)k(cid:18)t Therefore, τi ◦ τ −1 i = 1A⊗B. Similarly, τ −1 i ◦ τi = 1B⊗A and hence τi is bijective. (b) Compatible conditions: To show that the maps τi satisfy conditions (3.1.3) and (3.1.4), observe that if i is 0 or even, then both conditions hold from the definition of τ , as τi = τ in this case and satisfies (3.1.1). It remains to check the case when i is odd. 8 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON Let us verify (3.1.3) first. On B ⊗ B ⊗ A, the left hand side is τi ◦ (mB ⊗ 1)(yr1 ⊗ yr2 ⊗ xℓ) = τi(yr1+r2 ⊗ xℓ) = while the right hand side is (1 ⊗ mB) ◦ (τi ⊗ 1) ◦ (1 ⊗ τi)(yr1 ⊗ yr2 ⊗ xℓ) r1+r2 Xt=0 (cid:18)r1 + r2 2(cid:19)t t (cid:19)(cid:18)1 [ℓ + 1][t] xℓ+t ⊗ yr1+r2−t, r2 = (1 ⊗ mB) ◦ (τi ⊗ 1) yr1 ⊗ 2(cid:19)k k(cid:19)(cid:18) 1 Xk=0(cid:18)r2 [ℓ + 1][k] r1 = (1 ⊗ mB)" r2 2(cid:19)k k(cid:19)(cid:18)1 Xs=0(cid:18)r1 Xk=0(cid:18)r2 Xt=0 Xk+s=t(cid:18)r2 [ℓ + 1][k] [ℓ + k + 1][s]! xℓ+t ⊗ yr1+r2−t. 2(cid:19)t s(cid:19)(cid:18) 1 k(cid:19)(cid:18)r1 [ℓ + 1][k] xℓ+k ⊗ yr2−k! 2(cid:19)s s(cid:19)(cid:18)1 r1+r2 = [ℓ + k + 1][s] xℓ+k+s ⊗ yr1−s! ⊗ yr2−k# It is straightforward to check that [ℓ + 1][k] [ℓ + k + 1][s] = [ℓ + 1][k+s], and t Xk=0(cid:18)r2 k(cid:19)(cid:18) r1 t (cid:19). t − k(cid:19) =(cid:18)r1 + r2 This gives us the equality (3.1.3) as desired. For (3.1.4), on B ⊗ A ⊗ A, the left hand side is τi ◦ (1 ⊗ mA)(yr ⊗ xℓ1 ⊗ xℓ2) = τi(yr ⊗ xℓ1+ℓ2) = while the right hand side is (mA ⊗ 1) ◦ (1 ⊗ τi) ◦ (τ ⊗ 1)(yr ⊗ xℓ1 ⊗ xℓ2) r Xt=0 (cid:18)r 2(cid:19)t t(cid:19)(cid:18)1 [ℓ1 + ℓ2 + 1][t] xℓ1+ℓ2+t ⊗ yr−t, [ℓ1][k] xℓ1+k ⊗ yr−k! ⊗ xℓ2# Xs=0(cid:18)r − k 2(cid:19)s s (cid:19)(cid:18) 1 = (mA ⊗ 1) ◦ (1 ⊗ τi)" r 2(cid:19)k k(cid:19)(cid:18)1 Xk=0(cid:18)r = (mA ⊗ 1)" r [ℓ1][k] xℓ1+k ⊗ r−k 2(cid:19)k k(cid:19)(cid:18) 1 Xk=0(cid:18)r 2(cid:19)k+s s (cid:19)(cid:18) 1 k(cid:19)(cid:18)r − k Xs=0(cid:18)r Xk=0 Xt=0 Xk+s=t(cid:18)r 2(cid:19)t s (cid:19)(cid:18)1 k(cid:19)(cid:18)r − k r−k = = r r [ℓ2 + 1][s] xℓ2+s ⊗ yr−k−s!# [ℓ1][k] [ℓ2 + 1][s] x(ℓ1+ℓ2)+(k+s) ⊗ yr−(k+s) [ℓ1][k] [ℓ2 + 1][s]! x(ℓ1+ℓ2)+t ⊗ yr−t. To show the left hand side of (3.1.4) is equal to its right hand side, it suffices to show t(cid:19)[ℓ1 + ℓ2 + 1][t] = (cid:18)r t Xk=0(cid:18)r k(cid:19)(cid:18)r − k t − k(cid:19)[ℓ1][k] [ℓ2 + 1][t−k]. COHOMOLOGY OF POINTED HOPF ALGEBRAS 9 Using binomial identity in [4, p. 4038], [ℓ1 + ℓ2 + 1][t] =Pt k=0(cid:0) t t k(cid:1)[ℓ1][k] [ℓ2 + 1][t−k], we have t(cid:19)[ℓ1 + ℓ2 + 1][t] = (cid:18)r Xk=0(cid:18)r Xk=0(cid:18)r t t(cid:19)(cid:18)t k(cid:19)(cid:18)r − k k(cid:19)[ℓ1][k] [ℓ2 + 1][t−k] t − k(cid:19)[ℓ1][k] [ℓ2 + 1][t−k], where the last equality is due to = t(cid:19)(cid:18)t (cid:18)r k(cid:19) = = r! (r − t)! t! · t! (t − k)! k! = r! (r − t)! (t − k)! k! r! k! (r − k)! · (r − k)! (t − k)! (r − t)! =(cid:18)r t − k(cid:19). k(cid:19)(cid:18)r − k This gives us the equality (3.1.4) as desired. (c) Commutativity of diagram: We need to check that the following diagrams commute: B ⊗ A 1⊗x· / B ⊗ A τodd τeven and A ⊗ B x·⊗1 / A ⊗ B B ⊗ A 1⊗xp−1· / B ⊗ A τeven τodd A ⊗ B xp−1·⊗1 / A ⊗ B. For the first diagram, we have: τeven ◦ (1 ⊗ x·)(yr ⊗ xℓ) = τ (yr ⊗ xℓ+1) r = Xt=0 (cid:18)r 2(cid:19)t t(cid:19)(cid:18)1 (x · ⊗1) ◦ τodd(yr ⊗ xℓ) = (x · ⊗1) r Xt=0 (cid:18)r 2(cid:19)t t(cid:19)(cid:18)1 Xt=0 (cid:18)r = r [ℓ + 1][t] xℓ+1+t ⊗ yr−t, 2(cid:19)t t(cid:19)(cid:18)1 [ℓ + 1][t] xℓ+t ⊗ yr−t! [ℓ + 1][t] xℓ+1+t ⊗ yr−t. Similarly, for the second diagram: τodd ◦ (1 ⊗ xp−1·)(yr ⊗ xℓ) = τodd(yr ⊗ xℓ+p−1) if ℓ > 0 if ℓ = 0 τodd(yr ⊗ xp−1) =(0 =(0 Pr t=0(cid:0)r =(0 xp−1 ⊗ yr 2(cid:1)t t(cid:1)(cid:0) 1 if ℓ > 0 if ℓ = 0 [p][t] xp−1+t ⊗ yr−t if ℓ > 0 if ℓ = 0 (xp−1 · ⊗1) ◦ τeven(yr ⊗ xℓ) = (xp−1 · ⊗1) r Xt=0 (cid:18)r 2(cid:19)t t(cid:19)(cid:18) 1 [ℓ][t] xℓ+t ⊗ yr−t! / / / / 10 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON 2(cid:19)t t(cid:19)(cid:18)1 r = Xt=0 (cid:18)r =(0 xp−1 ⊗ yr [ℓ][t] xℓ+t+p−1 ⊗ yr−t if ℓ > 0 if ℓ = 0. (cid:3) For each i, the map τi is used to give Pi(A) ⊗ Pj(B) = A ⊗ B the structure of a left (A ⊗τ B)-module. In the case when i is even, this is the usual (A ⊗τ B)-module structure. In the case when i is odd, the (A ⊗τ B)-module structure via τi is given by (A ⊗τ B) ⊗ (A ⊗ B) 1⊗τi⊗1 / A ⊗ A ⊗ B ⊗ B mA⊗mB / / A ⊗ B. Lemma 3.2.2. Retaining the above module structure, Pi(A) ⊗ Pj(B) = A ⊗ B is a free (A ⊗τ B)-module of rank one, generated by 1 ⊗ 1, via the (A ⊗τ B)-module isomorphism ϕ : A ⊗τ B → A ⊗ B given by whose inverse is given by ϕ(xℓ ⊗ yr) =(xℓ ⊗ yr, Pr t=0(cid:0)r t(cid:1) t! ϕ−1(xℓ ⊗ yr) =(xℓ ⊗ yr, xℓ ⊗ yr − r 2t xℓ+t ⊗ yr−t, 2 xℓ+1 ⊗ yr−1, i is even i is odd, i is even i is odd. Proof. When i is even, ϕ is clearly bijective. We check that ϕ is a bijection with the given inverse when i is odd: r 2 r = r−1 r 2 xℓ+1 ⊗ yr−1(cid:17) ϕ ◦ ϕ−1(xℓ ⊗ yr) = ϕ(cid:16)xℓ ⊗ yr − k (cid:19) k! Xk=0(cid:18)r − 1 t(cid:19) t! Xt=0 (cid:18)r 2t xℓ+t ⊗ yr−t − t(cid:19) t! Xt=1 (cid:18)r Xt=1 2t xℓ+t ⊗ yr−t − t(cid:19)t − r(cid:18)r − 1 Xt=1 (cid:18)(cid:18)r Xt=1 t − 1(cid:19)(cid:19) (t − 1)! = xℓ ⊗ yr + = xℓ ⊗ yr + = xℓ ⊗ yr + (t − 1)! (0) r r r 2t 2t xℓ+t ⊗ yr−t = xℓ ⊗ yr. r 2k xℓ+1+k ⊗ yr−1−k t − 1(cid:19)(t − 1)! r(cid:18)r − 1 2t xℓ+t ⊗ yr−t xℓ+t ⊗ yr−t By (3.1.4), ϕ is a module isomorphism. Therefore A ⊗ B is free as an (A ⊗τ B)-module. (cid:3) In particular, the following is useful for our computations later: (3.2.3) ϕ−1(1 ⊗ y) =(1 ⊗ y, ϕ−1(1 ⊗ yp−1) =(1 ⊗ yp−1, 1 ⊗ y − 1 i is even i is odd, 2x ⊗ 1, 1 ⊗ yp−1 + 1 2x ⊗ yp−2, i is even i is odd. / COHOMOLOGY OF POINTED HOPF ALGEBRAS 11 By Theorem 3.1.5, the total complex K q := Tot(P q(A) ⊗ P q(B)) is a free resolution of k as A ⊗τ B-module. For each i, j ≥ 0, let φi,j denote the free generator 1 ⊗ 1 of Pi(A) ⊗ Pj(B) as an A ⊗τ B-module. Then as an R-module: Kn = Mi+j=n R φi,j. Recall that the differentials of this total complex are dn = Pi+j=n(di ⊗ 1 + (−1)i ⊗ dj). As Pi(A) ⊗ Pj(B) is free as an A ⊗τ B-module with generator φi,j, we can write the image of φi,j under the differential map as the action of A ⊗τ B on φi,j, using values of the inverse map ϕ−1 defined in (3.2.3) where needed. We express the differential on elements via this notation: xp−1φi−1,j + yp−1φi,j−1 xp−1φi−1,j + yφi,j−1 xφi−1,j − (yp−1 + 1 xφi−1,j − (y − 1 2x)φi,j−1 2 xyp−2)φi,j−1 if i, j are even, if i is even and j is odd, if i is odd and j is even, if i, j are odd. d(φi,j) =   We interpret φi,j to be 0 if either i or j is negative. 3.3. Resolution over the bosonization R# kG. Again we take k to be a field of charac- teristic p > 2 and R, G as described in Section 2.2. For the group G = hgi ∼= Z/qZ, where q is divisible by p, we define an action of G on the R-complex K q constructed in Section 3.2, for the purpose of forming a twisted tensor product resolution of K q with a resolution of k as kG-module. The group action will give us a twisting map on the complex K q analogous to the twisting map defining a skew group algebra. The resulting complex will give us a resolution of k over the bosonization R# kG. Let R = A ⊗τ B be the twisted tensor product with A := k[x]/(xp), B := k[y]/(yp) and τ : B ⊗ A → A ⊗ B as defined before. Under our setting in characteristic p > 2 and relation yx = xy + 1 2x2 in R, we obtain the following relations in R (here, we drop the tensor symbols and write xy in place of x ⊗ y in R): Proposition 3.3.1. The following relations hold in R: (1) For any integer ℓ ≥ 0, yxℓ = xℓy + xℓ+1. (2) For any integer n ≥ 1, (x + y)n = ℓ 2 n Xi=0 (cid:18)n i(cid:19) (i + 1)! 2i xiyn−i. Proof. (1) This was shown right after [4, Lemma 3.8]; we provide the proof here for com- pleteness. We proceed by induction on ℓ. Case ℓ = 0 is trivial. Case ℓ = 1 is from the relation yx = xy + 1 2x2. Now assume the statement holds up to ℓ − 1. Then by the induction hypothesis, we have: yxℓ = (yxℓ−1)x = (xℓ−1y + xℓ)x = xℓ−1yx + = xℓ−1(xy + 1 2 x2) + xℓ+1 = xℓy + 1 2 xℓ+1 + xℓ+1 = xℓy + ℓ 2 xℓ+1. 2 ℓ − 1 2 ℓ − 1 2 ℓ − 1 xℓ+1 2 ℓ − 1 12 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON (2) We proceed by induction on n. Case n = 1 is trivial. Assume the assertion is true for up to n − 1. Then by part (1), we have: (x + y)n = (x + y)(x + y)n−1 = (x + y) n−1 xiyn−1−i! (yxi)yn−1−i (xiy + i 2 xi+1)yn−1−i xiyn−i xi+1yn−1−i + xi+1yn−1−i + xi+1yn−1−i + + n−1 Xi=0 (cid:18)n − 1 xi+1yn−1−i + 2i 2i 2i n−1 n−1 n−1 i (cid:19) (i + 1)! i (cid:19) (i + 1)! i (cid:19) (i + 1)! Xi=0 (cid:18)n − 1 i (cid:19) (i + 1)! Xi=0 (cid:18)n − 1 Xi=0 (cid:18)n − 1 Xi=0 (cid:18)n − 1 i (cid:19)(i + 1)! Xi=0 (cid:18)n − 1 i (cid:19)(i + 1)! Xi=1 (cid:18)n − 1 i (cid:19) (i + 1)! i 2 n−1 n−1 2i 2i 2i 2i xi+1yn−1−i xiyn−i + xiyn−i + yn xiyn−i = = = = = = n−1 2i n−1 Xi=0 (cid:18)n − 1 Xi=0 (cid:18)n − 1 Xi=0 (cid:18)n − 1 i (cid:19)(i + 1)! i (cid:19)(i + 1)! i (cid:19)(i + 1)! n−1 2i 2i n−1 n n 2i 2i+1 i (cid:19)(i + 2)! i − 1(cid:19)(i + 1)! i(cid:19)(i + 1)! Xi=0 (cid:18)n − 1 Xi=1 (cid:18)n − 1 Xi=0 (cid:18)n i(cid:19) =(cid:18)n − 1 (cid:18)n 2i xiyn−i, where the last equality is due to the recursive formula of the binomial coefficients i − 1(cid:19) +(cid:18)n − 1 i (cid:19), for all 1 ≤ i ≤ n − 1. Lemma 3.3.2. Set an element α := −yp−2 + Then α satisfies the following: (−1)i+1 (i + 1)! 2i+1 xiyp−2−i ∈ R. p−2 Xi=1 (cid:3) 2x[(x + y)p−2 − yp−2] , (a) xα = (x + y)p−1 − yp−1 + 1 (b) (x + y)α = −yp−1 − 1 (c) αx = (x + y)p−1 − yp−1 , (d) α(y − 1 2x) = −(x + y)p−1. 2 xyp−2 , Proof. (a): By Proposition 3.3.1 (2), we have (x + y)p−1 − yp−1 + 1 2 x[(x + y)p−2 − yp−2] COHOMOLOGY OF POINTED HOPF ALGEBRAS 13 xiyp−2−i xi+1yp−2−i xiyp−1−i + 1 2 xi+1yp−2−i + p−2 x 2i p−2 2i+1 i (cid:19) (i + 1)! i (cid:19)(i + 1)! Xi=1 (cid:18)p − 2 Xi=1 (cid:18)p − 2 2i+1 +(cid:18)p − 2 (i + 2)! 2i+1 + p−2 Xi=1 (cid:18)(cid:18)p − 1 i + 1(cid:19)(i + 2)! (p − 1)! (i + 1)!(p − i − 2)! i (cid:19)(i + 1)! 2i+1 (cid:19) xi+1yp−2−i (p − 2)! i!(p − i − 2)! (i + 1)! 2i+1 (cid:19) xiyp−2−i# [(i + 2)(p − i − 1) · · · (p − 1) + (i + 1)(p − i − 1) · · · (p − 2)] 1 2i+1 xiyp−2−i# 1 2i+1 xiyp−2−i# p−1 2i = = p−2 p−2 2i+1 i (cid:19)(i + 1)! Xi=1 (cid:18)p − 1 i + 1(cid:19)(i + 2)! Xi=0 (cid:18)p − 1 1 (cid:19) xyp−2 + =(cid:18)p − 1 = x"−yp−2 + Xi=1 (cid:18) = x"−yp−2 + Xi=1 = x"−yp−2 + Xi=1 = x"−yp−2 + Xi=1 p−2 p−2 p−2 [(−1)i+11 · 2 · · · (i + 2) + (−1)i1 · 2 · · · (i + 1)(i + 1)] (−1)i+1(i + 1)! 2i+1 (i + 2 − i − 1) xiyp−2−i# = xα. (b): By Proposition 3.3.1 (2), we have (x + y)α = (x + y) −yp−2 + (−1)i+1(i + 1)! 2i+1 xiyp−2−i! (−1)i+1(i + 1)! xi+1yp−2−i − yp−1 p−2 Xi=1 2i+1 = −xyp−2 + + = −xyp−2 + + p−2 p−2 p−1 Xi=1 Xi=1 Xi=2 Xi=2 p−1 (−1)i+1(i + 1)! 2i+1 (xiy + i 2 xi+1)yp−2−i (−1)i i! 2i xiyp−1−i − yp−1 + (−1)i i! i − 1 2i 2 xiyp−1−i (−1)i+1(i + 1)! 2i+1 xiyp−1−i p−2 Xi=1 = −xyp−2 − yp−1 + = −xyp−2 − yp−1 + (−1)i+1(i + 1)! p−2 Xi=1 2! 22 xyp−2 + p−1 2i+1 p−2 xiyp−1−i + Xi=2 Xi=2 (cid:18) (−1)i+1(i + 1)! 2i+1 + (−1)i i! i + 1 2i 2 xiyp−1−i (−1)i(i + 1)! 2i+1 (cid:19) xiyp−1−i + (−1)p−1 p! 2p xpyp−1−(p−1) 14 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON = −yp−1 − 1 2 xyp−2. We are going to check for identities (c) and (d) in the domain khx, yi/(yx − xy − 1 2x2). (c): Since (x + y) is a regular element in this domain, to check (c) it suffices to show (x + y)αx = (x + y)[(x + y)p−1 − yp−1]. As identity (b) holds in the domain and by the relation (3.9) in [4, Lemma 3.8], we have: (x + y)αx = (−yp−1 − 1 2 xyp−2)x = −yp−1x − 1 2 xyp−2x = − = − = − p−1 p−1 2(cid:19)t t (cid:19)(cid:18) 1 Xt=0 (cid:18)p − 1 2(cid:19)t t (cid:19)(cid:18) 1 Xt=0 (cid:18)p − 1 2(cid:19)i−1 i − 1(cid:19)(cid:18) 1 Xi=1 (cid:18)p − 1 Xi=2 (cid:18) p p = −xyp−1 − = −xyp−1, t! x1+typ−1−t − (p − 1)! (i − 1)!(p − i)! (i − 1)! + [1][t] x1+typ−1−t − [1][s] x1+syp−2−s s! x2+syp−2−s p−2 x p−2 1 2 2(cid:19)s s (cid:19)(cid:18)1 Xs=0(cid:18)p − 2 2(cid:19)s+1 s (cid:19)(cid:18) 1 Xs=0(cid:18)p − 2 2(cid:19)i−1 i − 2(cid:19)(cid:18) 1 Xi=2 (cid:18)p − 2 (i − 2)!(cid:19)(cid:18) 1 (i − 2)!(p − i)! (p − 2)! p 2(cid:19)i−1 xiyp−i (i − 1)! xiyp−i − (i − 2)! xiyp−i and (x + y)[(x + y)p−1 − yp−1] = (x + y)p − xyp−1 − yp = p Xi=0 (cid:18)p i(cid:19)(i + 1)! 2i xiyp−i − xyp−1 − yp = yp − xyp−1 − yp = −xyp−1. Thus, as (x + y)αx = (x + y)[(x + y)p−1 − yp−1], we have identity (c) as claimed. (d): By a similar technique as in (c), to check (d) it suffices to show (x + y)α(y − 1 2 x) = (x + y)[−(x + y)p−1]. As we showed identities (b) and (x + y)αx = −xyp−1 hold above, we have (x + y)α(y − 1 2 x) =(cid:18)−yp−1 − 1 2 xyp−2(cid:19) y − 1 2 (x + y)αx = −yp − xyp−1 − (−xyp−1) = −yp = −(x + y)p = (x + y)[−(x + y)p−1]. 1 2 1 2 Thus, identity (d) holds. (cid:3) COHOMOLOGY OF POINTED HOPF ALGEBRAS 15 Recall that G = hgi ∼= Z/qZ acts on R by gx = x and gy = x + y. We define an action of G on the complex K q, constructed in Section 3.2, as follows: φi,j, φi,j + φi+1,j−1, φi,j + α φi+1,j−1, if i is odd if i is even and j is odd if i, j are even, gφi,j =  gs φi,j =  φi,j =  g−s and in general for all 0 ≤ s ≤ q − 1 φi,j, φi,j + s φi+1,j−1, φi,j + ((gs−1+···+g+1)α) φi+1,j−1, if i is odd if i is even and j is odd if i, j are even. Moreover, it is not hard to figure out the following for all 1 ≤ s ≤ q (3.3.3) φi,j, φi,j − s φi+1,j−1, φi,j − ((g−s+···+g−1)α) φi+1,j−1, if i is odd if i is even and j is odd if i, j are even. Lemma 3.3.4. With the above G-action, the complex K q is G-equivariant. Proof. We first show that this G-action is well-defined, that is, gq φi,j = φi,j. It is clear that when i is odd, or when i is even and j is odd that gqφi,j = φi,j, as q is divisible by the characteristic p of the field k. When i, j are both even, we need to show gqφi,j = φi,j + ((gq−1+···+g+1)α) φi+1,j−1 = φi,j, that is, we need to show (gq−1+···+g+1)α = 0. From Lemma 3.3.2 (c), we have αx = (x + y)p−1 − yp−1. Now apply gs-action on both sides, we have: gs α gs gs x = (gs x + gs y)p−1 − (gs y)p−1 α x = (x + y + sx)p−1 − (y + sx)p−1. Thus, summing over all 0 ≤ s ≤ q − 1: q−1 q−1 Xs=0 gs α! x = Xs=0(cid:0)[y + (s + 1)x]p−1 − (y + sx)p−1(cid:1) = (y + qx)p−1 − yp−1 = 0. gsα = 0 in the domain khx, yi/(yx − xy − 1 2x2) and hence is also 0 in R. Therefore, in all cases, we have gqφi,j = φi,j and the above G-action is well-defined. s=0 So Pq−1 To check that complex K q is G-equivariant, we need to check such G-action is compatible with the differential maps in each degree, that is, d( gφi,j) = gd(φi,j), for all i, j ≥ 0. When i, j are even: d( gφi,j) = d(φi,j + αφi+1,j−1) = xp−1φi−1,j + yp−1φi,j−1 + α(cid:18)xφi,j−1 − (y − gd(φi,j) = gxp−1 gφi−1,j + gyp−1 gφi,j−1 1 2 x)φi+1,j−2(cid:19) , = xp−1φi−1,j + (x + y)p−1(φi,j−1 + φi+1,j−2). 16 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON By comparing the coefficients, we see that coefficients for the terms φi,j−1 and φi+1,j−2 are exactly identities (c) and (d) in Lemma 3.3.2, respectively. When i is even and j is odd: d( gφi,j) = d(φi,j + φi+1,j−1) = xp−1φi−1,j + yφi,j−1 + xφi,j−1 − (yp−1 + gd(φi,j) = gxp−1 gφi−1,j + gy gφi,j−1 = xp−1φi−1,j + (x + y)(φi,j−1 + αφi+1,j−2). 1 2 xyp−2)φi+1,j−2, The coefficient for the term φi+1,j−2 is exactly the identity (b) in Lemma 3.3.2. When i is odd and j is even: d( gφi,j) = d(φi,j) = xφi−1,j − (yp−1 + 1 2 xyp−2)φi,j−1, gd(φi,j) = gx gφi−1,j − ( gyp−1 + 1 2 gx gyp−2) gφi,j−1 = x[φi−1,j + αφi,j−1] − [(x + y)p−1 + 1 2 x(x + y)p−2]φi,j−1. The coefficient for the term φi,j−1 is exactly the identity (a) in Lemma 3.3.2. When i, j are odd: d( gφi,j) = d(φi,j) = xφi−1,j − (y − 1 2 x)φi,j−1, gd(φi,j) = gx gφi−1,j − ( gy − 1 2 gx) gφi,j−1 = x(φi−1,j + φi,j−1) − (x + y − 1 2 x)φi,j−1. In all cases, we have d( gφi,j) = gd(φi,j). Thus, complex K q is G-equivariant. (cid:3) We use this G-action next to form a twisted tensor product resolution of k as an R# kG- n : kG ⊗Kn → Kn ⊗ kG be given by the action of G on Kn, module. Let the twisting map τ ′ so that τ ′ i+j(g ⊗ φi,j) =  φi,j ⊗ g (φi,j + φi+1,j−1) ⊗ g (φi,j + α φi+1,j−1) ⊗ g if i is odd if i is even and j is odd if i, j are even. Then K q is compatible with τ ′ (giving the twisting that governs the smash product construc- tion) via the maps τ ′ n. Let P q(kG) be the following free resolution of the trivial kG-module k: P q(kG) : · · · where ε takes g to 1. (Pq−1 s=0 gs)· / kG (g−1)· / kG (Pq−1 s=0 gs)· / kG (g−1)· / kG ε / k / 0, Let Y q := Tot(K q ⊗ P q(kG)). By [20, Lemma 5.9], Y q → k is exact. The modules are free (R# kG)-modules by a similar argument to what we used earlier: In each degree we have a direct sum of modules of the form R ⊗ kG. Each such is freely generated by some φi,j ⊗ φk, where φk denotes the free generator for Pk(kG) = kG. So by Theorem 3.1.5, Y q is a free resolution of the (R# kG)-module k. / / / / / / COHOMOLOGY OF POINTED HOPF ALGEBRAS 17 For each i, j, k ≥ 0, let φi,j,k denote the free generator φi,j ⊗ φk of Ki+j ⊗ Pk(kG) as an (R# kG)-module. We set φi,j,k = 0 if one of i, j, k is negative. Then, for all n ≥ 0, as an (R# kG)-module, Yn = Mi+j+k=n (R# kG) φi,j,k. We express the differentials on elements via this notation. d(φi,j,k) = d(φi,j) ⊗ φk + (−1)i+j   (g − 1) φi,j,k−1, if i, k are odd (g − 1) φi,j,k−1 − g φi+1,j−1,k−1, if i is even and j, k are odd (g − 1) φi,j,k−1 − αg φi+1,j−1,k−1, if i, j are even and k is odd (cid:0)Pq−1 s=0 gs(cid:1) φi,j,k−1, (cid:0)Pq−1 s=0 gs(cid:1) φi,j,k−1 −(cid:0)Pq−1 (cid:0)Pq−1 s=0 gs(cid:1) φi,j,k−1 −Pq−1 s=1((gs−1+···+g+1)α) gsφi+1,j−1,k−1, s=0 sgs(cid:1) φi+1,j−1,k−1, if i is odd and k is even if i, k are even and j is odd if i, j, k are even. We will give partial verification to the above differentials in view of (3.3.3). The case for i, k are even and j is odd. d(φi,j,k) = d(φi,j) ⊗ φk + (−1)i+jφi,j ⊗ d(φk) = d(φi,j) ⊗ φk + (−1)i+jφi,j q−1 Xs=0 gs! ⊗ φk−1 q−1 gs(cid:16)g−s φi,j(cid:17) ⊗ φk−1 gs (φi,j − sφi+1,j−1) ⊗ φk−1 q−1 = d(φi,j) ⊗ φk + (−1)i+j = d(φi,j) ⊗ φk + (−1)i+j Xs=0 Xs=0 = d(φi,j) ⊗ φk + (−1)i+j" q−1 Xs=0 gs! φi,j,k−1 − q−1 Xs=0 sgs! φi+1,j−1,k−1# The case for i, j, k are even. d(φi,j,k) = d(φi,j) ⊗ φk + (−1)i+jφi,j ⊗ d(φk) = d(φi,j) ⊗ φk + (−1)i+jφi,j q−1 Xs=0 gs! ⊗ φk−1 q−1 = d(φi,j) ⊗ φk + (−1)i+j gs(cid:16)g−s φi,j(cid:17) ⊗ φk−1 gs(φi,j −(g−1+···+g−s)α φi+1,j−1)! ⊗ φk−1# = d(φi,j) ⊗ φk + (−1)i+j" φi,j + Xs=1 Xs=0 q−1 18 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON = d(φi,j) ⊗ φk + (−1)i+j" q−1 Xs=0 = d(φi,j) ⊗ φk + (−1)i+j" q−1 Xs=0 gs! φi,j,k−1 − q−1 Xs=1 gs! φi,j,k−1 − q−1 Xs=1 gs ((g−1+···+g−s)α)! φi+1,j−1,k−1# (gs−1+···+g+1)α gs! φi+1,j−1,k−1# . s=0 gs(cid:1) = 0 = ε(cid:0)Pq−1 s=0 sgs(cid:1) Note that in the above formulas for the differentials, we have ε(cid:0)Pq−1 in characteristic p, since p divides q. Among these differentials of free R# kG-module basis elements, the only terms in the outcomes d(φi,j,k) that do not have coefficients in the aug- mentation ideal Ker(ε) are the terms −gφi+1,j−1,k−1, occurring when i is even and j, k are odd. Consequently, letting n = i + j + k and φ∗ i,j,k be the dual basis vector to φi,j,k in Homk(cid:16)Li′+j ′+k′=n kφi′,j ′,k′, k(cid:17) ∼= HomR# kG(Yn, k), we have i−1,j+1,k+1, d∗(φ∗ i,j,k) =(cid:26) −φ∗ 0, if i is odd and j, k are even otherwise. The cocycles are thus all the φ∗ coboundaries are the φ∗ a vector space, i,j,k except those for which i is odd and j, k are even. The i,j,k for which i is even and j, k are odd. Therefore, for all n ≥ 0, as Hn(R# kG, k) ∼=  Spank{φ∗ − Spank{φ∗ Spank{φ∗ − Spank{φ∗ i,j,k i + j + k = n} i,j,k i is even and j, k are odd}, i,j,k i + j + k = n} i,j,k i is odd and j, k are even}, if n is even if n is odd. 4. Anick resolutions In this section, we recall the Anick resolution and make some additional observations about it in our setting. We will use the Anick resolution in Section 5.2. 4.1. The resolution construction. We generally construct the Anick resolution [1] as envisioned by Cojocaru and Ufnarovski [5], adapted here to left modules under some condi- tions. An algorithmic description using Grobner bases is given by Green and Solberg [12]. The construction of the resolution also serves as a proof of exactness, since the differentials are defined recursively in each degree, making use of a contracting homotopy in the previous degree that is constructed recursively as well. See Theorem 4.1.2 below, due to Anick. We include a proof in our setting because we will use the construction in Sections 4.2 and 5.2. Let A = T (V )/(I) where V is a finite dimensional vector space, T (V ) = Tk(V ) is the tensor algebra on V over k, and I is a set of relations generating an ideal (I). We denote the image of an element v of V in A also by v when it will cause no confusion. We assume that A is augmented by an algebra homomorphism ε : A → k with ε(v) = 0 for all v ∈ V . Fix a totally ordered basis v1, . . . , vn of V (say v1 < · · · < vn) and consider the degree lexicographic ordering on words in v1, . . . , vn. That is, we give each vi the degree 1, and monomials (words) are ordered first according to total degree, then monomials having the same degree (i.e. word length) are ordered as in a dictionary. A normal word (called an element of an order ideal of monomials, or o.i.m. in [1]) is a monomial (considered as an element of T (V )) that cannot be written as a linear combination COHOMOLOGY OF POINTED HOPF ALGEBRAS 19 of smaller words in A. As a vector space, A has a basis in one-to-one correspondence with the set of normal words. A tip (called an obstruction in [1]) is a word (considered as an element in T (V )) that is not normal but for which any proper subword is normal. It follows that the tips correspond to the relations: Let u be a tip and write the image of u in A as a linear combination correspondence with a Grobner basis of (I) [12]. u =P aiti, where each ti is a normal word and ai is a scalar. Then, viewed as an element of T (V ), u −P aiti is in the ideal of relations, (I). It also follows that the tips are in one-to-one We construct the Anick resolution from the chosen sets of generators and relations in A as follows. We reindex in comparison to [1] so that indices for spaces correspond to homological degrees, and indices for functions correspond to homological degrees of their domains. The Anick resolution is a free resolution of k considered to be an A-module under the augmentation ε. We will first describe a free basis Cn in each homological degree n of the resolution. We will write the resolution as: d3 · · · / A ⊗ kC2 d2 / A ⊗ kC1 d1 / A ε / k / 0, where kCn denotes the vector space with basis Cn. We adapt the degree lexicographic ordering on monomials in T (V ) to each A-module A ⊗ kCn by giving an element s ⊗ t, where s is a normal word and t ∈ Cn, the degree of st viewed as an element of T (V ). Let C1 = {v1, . . . , vn}, that is, C1 is the chosen set of generators. Let C2 be the set of tips (or obstructions). The remaining sets Cn will be defined as sets of paths of length n in a directed graph (or quiver) associated to the generators and tips as follows [5]. The graph will have at most one directed arrow joining two vertices, and paths will be denoted by the product of their vertices in T (V ), written from right to left, for example, if f, g are vertices and there is an arrow from f to g, we denote the arrow by gf , and if there is a further arrow from g to h, then hgf denotes the path f → g → h starting at f , passing through g, and ending at h. Let B = {v1, . . . , vn} be a basis of V , equipped with the ordering v1 < · · · < vn as above, so that we may identify B with C1. Let T be the set of tips. Let R be the set of all proper prefixes (that is, left factors) of the tips considered as elements of T (V ). (Note that B ⊂ R.) Let Q = Q(B, T ) be the following quiver. The vertex set is {1} ∪ R. The arrows are all 1 → vi for vi ∈ B and all f → g for which the word gf (viewed as an element of T (V )) uniquely contains a tip, and that tip is a prefix (possibly coinciding with gf ). The set Cn consists of all paths of length n starting from 1 in the quiver. In this context, the path 1 → f → g is identified with the product gf . (Note that f → g does not occur on its own as an element of any Ci if f 6= 1, so for our purposes there will be no confusion in denoting paths this way.) For use in constructing the chains Cn, we observe that we only use the vertices that are in the connected component of Q containing 1. Let Q = Q(B, T ) be the connected component of 1 in Q, called the reduced quiver of B and T . / / / / / 20 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON The differentials d are defined recursively, with a simultaneous recursive definition of a k-linear contracting homotopy s: (4.1.1) · · · d3 s2 / A ⊗ kC2 d2 s1 / A ⊗ kC1 d1 s0 / A ε η / k / 0, where η is the unit map (taking the multiplicative identity of k to the multiplicative identity of A). We give these definitions next in our setting, simultaneously proving the following theorem. Examples are given in [1, 5] and below in Sections 4.2 and 5.2. Theorem 4.1.2. [1, Theorem 1.4] There are maps dn, sn for which (A ⊗ kC q, d q) is a free resolution of k as an A-module and s q is a contracting homotopy. Proof. We first define the maps dn, sn−1 for n = 1, 2 to illustrate the general method. We then use induction on n. Degree 1: We take n = 1 and let d1(1 ⊗ vi) = vi for all vi in B and extend d1 so that it is a left A-module homomorphism. To define the k-linear map s0 : A → A ⊗ kC1, first write elements of A as k-linear combinations of normal words (which form the chosen vector space basis of A). Define s0 on A via its values on all normal words, which are as follows. Set s0(1) = 0 and s0(uvi) = u ⊗ vi for all normal words of the form uvi for some word u and vi in B. Extend s0 so that it is a k-linear map on A, and note that it will not be an A-module homomorphism in general. We now see that by construction, a = (d1s0 + ηε)(a) for all a ∈ A. It also follows that A ⊗ kC1 = Ker(d1) ⊕ Im(s0). To see this, let b ∈ A ⊗ kC1 and write b = (b − s0d1(b)) + s0d1(b). One checks that b − s0d1(b) ∈ Ker(d1); by definition, s0d1(b) ∈ Im(s0). The intersection of these two spaces is 0 by the above equation and definitions: If b ∈ Ker(d1) ∩ Im(s0), write b = s0(c). Then which implies c ∈ k so that b = s0(c) = 0. c = (d1s0 + ηε)(c) = ηε(c), Degree 2: We take n = 2 and define d2(1 ⊗ u) for u in C2 as follows. By definition of C2, we may write u = rvi uniquely in T (V ) for a word r in R and vi ∈ C1. Consider r ⊗ vi as an element of A ⊗ kC1, and further take its image under the A-module homomorphism d1: d1(r ⊗ vi) = rd1(1 ⊗ vi) = rvi. Define d2(1 ⊗ u) = r ⊗ vi − s0(d1(r ⊗ vi)) = r ⊗ vi − s0(rvi), and extend d2 so that it is an A-module homomorphism on A ⊗ kC2. By its definition, rvi when considered as an element of T (V ) is a tip (not a normal word), and considered here as an element of A, it must be rewritten as a k-linear combination of normal words before applying s0 (since s0 is a k-linear map but not an A-module homomorphism). Now the definitions of d1 and s0 imply d1s0Ker(ε) = 1Ker(ε), the identity map on Ker(ε). It also follows that d1d2 = 0. We wish to define s1 so that d2s1Ker(d1) = 1Ker(d1) and more generally so that d2s1 + s0d1 = 1A⊗kC1. / o o / o o / o o / o o / u′ +Pℓ Set α = −Pℓ k=1 ai1bkv′t′ j=2 aij rij ⊗ vij and k ⊗ vik +Pm s1 m Xj=1 aij rij ⊗ vij! = ai1v′ ⊗ u′ + s1(α). COHOMOLOGY OF POINTED HOPF ALGEBRAS 21 First define s1 on elements in Ker(d1) by induction on their degrees, starting with those that are least in the ordering, which are elements of A ⊗ kC1 corresponding to relations: Ordering the elements of C2 as u1, . . . , uℓ, with u1 least, we define s1(d2(1 ⊗ u1)) = 1 ⊗ u1. Recall that we have chosen a total order on a vector space basis of A ⊗ kC1, given by elements r ⊗ vi where r is a normal word and vi ∈ B, to coincide with the order on the corresponding words rvi in T (V ). Assume s1 has been defined on elements of Ker(d1) with highest term j=1 aij rij ⊗ vij ∈ Ker(d1) for some nonzero aij ∈ k, rij ∈ A, with vi1, . . . , vim distinct elements of B, and terms ordered so of degree (i.e. position in the total order) less than n. Let Pm that ri1 ⊗ vi1 is greatest and has degree n. Since d1(P aij rij ⊗ vij ) = 0 in A by assumption, and ri1 ⊗ vi1 is greatest, the monomial ri1vi1 in T (V ) must contain a tip. Since ri1 is a nonzero normal word, the tip must be a suffix (that is, right factor) of ri1vi1, say ri1vi1 = v′u′ in T (V ) with u′ a tip. Since u′ is a tip, there is an element in the ideal (I) of the form kvik for words t′ k. k=1 bktk for some normal words tk and scalars bk. Write each tk = t′ Note that α consists of terms of lower degree than ri1 ⊗ vi1, and α ∈ Ker(d1) by construction, so we may now apply the induction hypothesis to define s1(α). Recall that A ⊗ kC1 = Ker(d1) ⊕ Im(s0). We define s1 on Im(s0) to be 0. We claim that by these definitions, d2s1 + s0d1 = 1A⊗kC1. To see this, we check separately for elements of Ker(d1) and of Im(s0). If x =Pm vij ∈ Ker(d1) as above, then by the inductive definition of s1, we have j=1 aij rij ⊗ (d2s1 + s0d1)(x) = d2s1(x) = x. If x ∈ Im(s0) then (d2s1 + s0d1)(x) = s0d1(x) = x since d1s0 + ηε = 1A. If b ∈ A ⊗ kC2 then b = (b − s1d2(b)) + s1d2(b), with b − s1d2(b) in Ker(d2) and s1d2(b) in Im(s1). If b ∈ Ker(d2) ∩ Im(s1), write b = s1(c), and we have c = (d2s1 + s0d1)(c) = s0d1(c). Then b = s1(c) = s1s0d1(c) = 0 since s1s0 = 0 by definition of s1. It follows that A ⊗ kC2 = Ker(d2) ⊕ Im(s1): Degree at least 3: We take n ≥ 3 and assume that A-module homomorphisms d1, . . . , dn−1 and k-linear maps s0, . . . , sn−2 have been defined so that di−1di = 0, si−1si−2 = 0, and disi−1 + si−2di−1 = 1A⊗kCi−1 for 1 ≤ i ≤ n − 1. It follows by an argument similar to the above that A ⊗ kCi = Ker(di) ⊕ Im(si−1) for all 1 ≤ i ≤ n − 1. In particular, A ⊗ kCn−1 = Ker(dn−1) ⊕ Im(sn−2). We will define dn and sn−1. The map dn is defined first as follows. Let u ∈ Cn. We may write uniquely u = ru′ for u′ ∈ Cn−1 and r in R by construction of the quiver Q. Let (4.1.3) dn(1 ⊗ u) = r ⊗ u′ − sn−2dn−1(r ⊗ u′). Now dn−1(r ⊗ u′) = rdn−1(1 ⊗ u′) since dn−1 is an A-module homomorphism, and in order to apply sn−2 to this element of A ⊗ kCn−2, any elements of A will need to be rewritten as linear combinations of normal words before applying sn−2 (since sn−2 is k-linear but not an 22 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON A-module homomorphism in general). It follows directly from the definition of dn and the induction hypothesis that dn−1dn = 0. We wish to define sn−1 so that dnsn−1Ker(dn−1) = 1Ker(dn−1) and more generally so that dnsn−1 + sn−2dn−1 = 1A⊗kCn−1. The map sn−1 is defined inductively as follows. Let Pm i=1 airi ⊗ ui ∈ Ker(dn−1) for some ai ∈ k, normal words ri ∈ A, and ui ∈ Cn−1. Recall that we have chosen a total order on a vector space basis of A ⊗ kCn−1, given by elements r ⊗ u where r is a normal word and u ∈ Cn−1, to coincide with the order on the corresponding words ru in T (V ). We may assume r1 ⊗ u1 is the highest term among all ri ⊗ ui. Write u1 = u′u′′, uniquely, where u′′ ∈ Cn−2. Then by definition of dn−1 (replacing n by n − 1 in equation (4.1.3)), we have 0 = dn−1 m Xi=1 airi ⊗ ui! = a1r1u′ ⊗ u′′ + β, where β = −sn−3dn−2(a1r1u′⊗u′′)+dn−1(Pm i=2 airi⊗ui), and when the term dn−1(P airi⊗ui) is expanded, due to cancellation, the resulting expression for β consists of terms lower in the order than r1 ⊗ u1. Since 0 = a1r1u′ ⊗ u′′ + β, considering r1u′ as a word in T (V ), there is a tip v′′ that is a factor of r1u′ in T (V ). To make a unique choice of such a tip, write r1 = vj1 · · · vjℓ as a word in the letters in B. Now u′ is not a tip, but r1u′ contains a tip, and so there is a largest k (k ≤ ℓ) for which vjk · · · vjℓu′ (uniquely) contains a tip, and by construction this tip will then be a prefix. Thus we may write, uniquely, r1u′ = v′tu′ where t ∈ R and tu′ uniquely contains a tip that is a prefix. So there is an arrow u′ → t in the reduced quiver Q by definition. Therefore tu1 = tu′u′′ ∈ Cn. We may thus set airi ⊗ ui! = a1v′ ⊗ tu1 + sn−1(γ), airi ⊗ ui − dn(a1v′ ⊗ tu1) (4.1.4) where sn−1 m Xi=1 Xi=1 γ = m has highest term that is lower in the order than r1 ⊗ u1. (To obtain the above expression, in the argument P airi ⊗ ui of sn−1, we have added and subtracted dn(a1v′ ⊗ tu1).) Now continue in the same fashion to obtain sn−1(γ) in terms of elements lower in the total order, and so on. Since the chosen basis of A ⊗ kCn−1 is well-ordered, we will eventually reach an expression involving sn−1(0) = 0. As before, we define sn−1 on Im(sn−2) to be 0, so that sn−1sn−2 = 0. A calculation as before now shows that dnsn−1 + sn−2dn−1 = 1A⊗kCn−1. Now by its definition and the above arguments, s q is a contracting homotopy for the com- plex (4.1.1), implying that the complex is exact, and therefore (A⊗kC q, d q) is a free resolution of k as an A-module. (cid:3) 4.2. A truncated polynomial ring. In this section, we let k be any field and m1, m2, m3 be positive integers at least 2. We look closely at the Anick resolution for the algebra A = k[w, x, y]/(wm1, xm2, ym3) which will be used in the next section. For this purpose, choose generating set B = {w, x, y} and relations (4.2.1) I = {wm1, xm2, ym3, wx − xw, wy − yw, xy − yx}. COHOMOLOGY OF POINTED HOPF ALGEBRAS 23 Then A has basis {wixjyk 0 ≤ i ≤ m1 − 1, 0 ≤ j ≤ m2 − 1, 0 ≤ k ≤ m3 − 1}. We choose the ordering w < x < y. So, for example, the degree lex ordering on basis elements in degrees 0, 1, 2 is 1 < w < x < y < w2 < wx < wy < x2 < xy < y2. Note that xy is a normal word, while yx is a tip (or obstruction). Generally, the normal words correspond to the PBW basis of A, and the tips are The proper prefixes of the tips are T := {wm1, xm2, ym3, xw, yw, yx}. R = {wi, xj, yk 0 ≤ i ≤ m1 − 1, 0 ≤ j ≤ m2 − 1, 0 ≤ k ≤ m3 − 1}. The corresponding reduced quiver Q, as defined in Section 4.1, is as follows. (The nonre- duced quiver Q contains additional vertices and arrows if m > 3, but we will not need this here.) If m1 = 2, the vertices w and wm1−1 are identified, and there is a loop at that vertex. Similarly for m2, m3. 1 w (❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ v❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧ 3❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢ 6♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ 6♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ x y wm1−1 xm2−1 ym3−1 We have C1 = {w, x, y}, C2 = {wm1, xm2, ym3, xw, yw, yx}, C3 = {wm1+1, xm2+1, ym3+1, xwm1, ywm1, yxm2, xm2w, ym3w, ym3x, yxw}, and similarly we may find Cn for n > 3. A free basis of Cn is all 1 ⊗ u where u is a path of length n starting at 1 in the above reduced quiver Q. Fix such a free basis element 1 ⊗ u. Suppose n = i + j + k and the first i vertices in the path u are in the set {w, wm1−1}, the second j vertices of u are in the set {x, xm2−1}, and the third k vertices are in the set {y, ym3−1}. Write u = ui,j,k and note that the triple of indices i, j, k uniquely determines the path. For convenience, we set C0 = {1} and u000 = 1, identifying A with A ⊗ kC0. We set uijk = 0 if i, j, or k is negative. Lemma 4.2.2. Let A = k[w, x, y]/(wm1, xm2, ym3), and let P q denote the Anick resolution of A with respect to the chosen generators w, x, y and relations (4.2.1). Then dn(1 ⊗ uijk) = yσ3(k) ⊗ ui,j,k−1 + (−1)kxσ2(j) ⊗ ui,j−1,k + (−1)j+kwσ1(i) ⊗ ui−1,j,k where n = i + j + k, and σa(ℓ) =(1, ma − 1, if ℓ is odd, if ℓ is even. Proof. We will prove the formula for dn by induction on n. By definition, d1(1 ⊗ u100) = w ⊗ u000, d1(1 ⊗ u010) = y ⊗ u000, and d1(1 ⊗ u001) = x ⊗ u000, and these values agree with the claimed formula for d1. v ( / / + + / / I I 6 3 H H 6 J J 24 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON Assume the formula holds for dn−1. We first consider the case j = k = 0 and i > 0: dn(1 ⊗ ui00) = wσ1(i) ⊗ ui−1,0,0 − sn−2dn−1(wσ1(i) ⊗ ui−1,0,0) = wσ1(i) ⊗ ui−1,0,0 − sn−2(wσ1(i)(wσ1(i−1) ⊗ ui−2,0,0)) = wσ1(i) ⊗ ui−1,0,0 − sn−2(0) = wσ1(i) ⊗ ui−1,0,0, since wσ1(i)wσ1(i−1) = 0 in the algebra A. This outcome agrees with the stated formula for dn. Next consider the case k = 0 and j > 0, applying the construction of sn−2 described in the proof of Theorem 4.1.2: The vertex xσ2(j) is last in the path uij0, so by induction, dn(1 ⊗ uij0) = xσ2(j) ⊗ ui,j−1,0 − sn−2dn−1(xσ2(j) ⊗ ui,j−1,0) = xσ2(j) ⊗ ui,j−1,0 − sn−2((−1)j−1wσ1(i)xσ2(j) ⊗ ui−1,j−1,0) = xσ2(j) ⊗ ui,j−1,0 + (−1)jwσ1(i) ⊗ ui−1,j,0, since xσ2(j)xσ2(j−1) = 0 in A. This agrees with the stated formula for dn. In case k > 0, since the vertex labeled yσ3(k) is the last in the path uijk, by induction, since yσ3(k)yσ3(k−1) = 0, dn(1 ⊗ uijk) = yσ3(k) ⊗ ui,j,k−1 − sn−2dn−1(yσ3(k) ⊗ ui,j,k−1) = yσ3(k) ⊗ ui,j,k−1 − sn−2(yσ3(k)((−1)k−1xσ2(j) ⊗ ui,j−1,k−1 + (−1)j+k−1wσ1(i) ⊗ ui−1,j,k−1)) = yσ3(k) ⊗ ui,j,k−1 − sn−2((−1)k−1xσ2(j)yσ3(k) ⊗ ui,j−1,k−1 + (−1)j+k−1wσ1(i)yσ3(k) ⊗ ui−1,j,k−1). Compare the two terms comprising the argument of sn−2; they are xσ2(j)yσ3(k) ⊗ ui,j−1,k−1 and wσ1(i)yσ3(k) ⊗ ui−1,j,k−1, up to sign. These terms have the same total degree, since each arises via an application to ui,j,k of some differential maps (that do not change total degree). Thus we must compare them lexicographically, and we see that xσ2(j)yσ3(k) ⊗ ui,j−1,k−1 is the higher of the two terms. The first step of applying sn−2 thus involves the term xσ2(j) ⊗ui,j−1,k corresponding to this expression as the first term on the right side of equation (4.1.4). Continuing by working inductively, with appropriate signs, we obtain yσ3(k) ⊗ ui,j,k−1 + (−1)kxσ2(j) ⊗ ui,j−1,k + (−1)j+kwσ1(i) ⊗ ui−1,j,k, as desired. (cid:3) Next we show that the Anick resolution is isomorphic to a twisted tensor product resolution for this small example. Lemma 4.2.3. Let A = k[w, x, y]/(wm1, xm2, ym3). The Anick resolution P q for A is equiv- alent to the total complex K q of the tensor product of the minimal resolutions of Aw = k[w]/(wm1), Ax = k[x]/(xm2), and Ay = k[y]/(ym3), that is, for each n there is an A-module isomorphism ψn : Pn → Kn and ψ q is a chain map lifting the identity map on k. Proof. Let P (Aw) be the following free resolution of k as an A-module: P (Aw) : · · · (wm1−1)· / Aw w· / Aw (wm1−1)· / Aw w· / Aw ε / k / 0. Let P (Ax) and P (Ay) be similar free resolutions of k as an Ax-module and as an Ay-module, respectively. Let K q = Tot(P (Ay) ⊗ P (Ax) ⊗ P (Aw)), be the total complex of the tensor product of these three complexes. Let δ denote the differential on P q. / / / / / / COHOMOLOGY OF POINTED HOPF ALGEBRAS 25 We will show that Pn ∼= Kn as an A-module for each n and that such isomorphisms may be chosen so as to constitute a chain map between P q and K q. We will prove this by induction on n, beginning with n = 0 and n = 1. For n = 0, note that P0 = A ∼= Ay ⊗ Ax ⊗ Aw = K0 and each maps onto k via ε. We take ψ0 to be this isomorphism. For n = 1, note that P1 = A ⊗ k{w, x, y}, while K1 is equal to (P (Ay)1 ⊗ P (Ax)0 ⊗ P (Aw)0) ⊕ (P (Ay)0 ⊗ P (Ax)1 ⊗ P (Aw)0) ⊕ (P (Ay)0 ⊗ P (Ax)0 ⊗ P (Aw)1). To keep track of degrees, let φ100 denote 1 ⊗ 1 ⊗ 1 in P (Ay)1 ⊗ P (Ax)0 ⊗ P (Aw)0 and similarly φ010, φ001. Let ψ1 : P1 → K1 be defined by ψ1(1 ⊗ w) = φ100, ψ1(1 ⊗ x) = φ010 and ψ1(1 ⊗ y) = φ001. More generally let φijk denote 1 ⊗1 ⊗1 in P (Ay)k ⊗P (Ax)j ⊗P (Aw)i. Recall similar notation uijk for free basis elements of Pn described above. Define ψn : Pn → Kn as follows: Extend ψn to an A-module isomorphism. The differential on K q may be written as ψn(1 ⊗ uijk) = φijk. dn(φijk) = yσ3(k)φi,j,k−1 + (−1)kxσ2(j)φi,j−1,k + (−1)j+kwσ1(i)φi−1,j,k. Comparing with Lemma 4.2.3, we see that ψ q is a chain map. (cid:3) 5. Finite generation of some cohomology rings We now apply the constructions of twisted tensor product and Anick resolutions discussed in Sections 3 and 4 to prove that the cohomology rings of the Hopf algebras in our settings (see Sections 2.2 and 2.3) are finitely generated. 5.1. Cohomology of the Nichols algebra and its bosonization. Let R, G be defined as in Section 2.2. Recall that we have used a twisted tensor product to construct a resolution K q for k as an R-module in Section 3.2 and further to construct a resolution Y q for k as a module over the bosonization R# kG in Section 3.3. By examining the expression for the cohomology H2(R# kG, k) given at the end of Sec- tion 3.3, we see that it includes nonzero elements represented by the 2-cocycles φ∗ 0,2,0, φ∗ 0,0,2. We find their cup products, which will be used in the proof of Theorem 5.1.2 below. To simplify notation, let 2,0,0, φ∗ ξx = φ∗ 2,0,0, ξy = φ∗ 0,2,0, ξg = φ∗ 0,0,2. Using the projectivity of the resolution Y q, one can show that these functions may be extended to chain maps on Y q as follows: ξx(φi,j,k) = φi−2,j,k, ξy(φi,j,k) = φi,j−2,k, ξg(φi,j,k) = φi,j,k−2, for all i, j, k, where we set φi′,j ′,k′ = 0 if any one of i′, j ′, k′ is negative. Consequently, ξx, ξy, ξg are generators of a polynomial subalgebra k[ξx, ξy, ξg] of H∗(R# kG, k) in even degrees. For example, the above formulas can be used to show that (φ∗ 2,0,0)2 = φ∗ 4,0,0 and 2,0,0 ⌣ φ∗ φ∗ 0,2,0 = φ∗ 2,2,0 = φ∗ 0,2,0 ⌣ φ∗ 2,0,0 and generally if i, j, k, i′, j ′, k′ are all even, then i,j,k ⌣ φ∗ φ∗ i′,j ′,k′ = φ∗ i+i′,j+j ′,k+k′. 26 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON We will also need the following lemma, which is [14, Lemma 2.5] as adapted from [8, Lemma 1.6]. Lemma 5.1.1. Let Ep,q ∞ be a multiplicative spectral sequence of k-algebras 1 concentrated in the half plane p + q ≥ 0, and let B∗,∗ be a bigraded commutative k-algebra concentrated in even (total) degrees. Assume that there exists a bigraded map of algebras from B∗,∗ to E ∗,∗ is a ∞ is a noetherian module over Tot(B∗,∗). noetherian module over the image of B∗,∗. Then E ∗ such that the image of B∗,∗ consists of permanent cycles, and E ∗,∗ =⇒ Ep+q 1 1 We are now ready to prove our first main theorem. Theorem 5.1.2. Let R := khx, yi/(xp, yp, yx − xy − 1 2x2) be the Nichols algebra defined over a field k of prime characteristic p > 2, and G := hgi be a cyclic group of order q divisible by p, acting on R by automorphisms with gx = x and gy = x + y. Then the cohomology ring of the bosonization, H∗(R# kG, k), is finitely generated. Proof. Without loss of generality we may assume that q = pa for some a. To see this, note that G ∼= Z/paZ × Z/ℓZ for some ℓ coprime to p and some a ≥ 1. Elements of the subgroup of G that is isomorphic to Z/ℓZ act trivially on R since their orders are coprime to p, and so R#kG ∼= (R#kZ/paZ) ⊗ (kZ/ℓZ) as an algebra. Thus the cohomology of R#kG is the graded tensor product of the cohomology of R#kZ/paZ and of kZ/ℓZ. The cohomology of kZ/ℓZ is concentrated in degree 0, where it is simply k, since ℓ is coprime to p. Now assume that q = pa. Let w = g − 1 and note that since the order of g is q, R# kG ∼= khw, x, yi/(wq, xp, yp, yx − xy − 1 2 x2, xw − wx, yw − wy − wx − x). Assign the degree lexicographic order on monomials in w, x, y, with w < x < y. This gives rise to an N-filtration on R#kG for which the associated graded algebra gr(R# kG) ∼= k[w, x, y]/(wq, xp, yp). (See, e.g., [2, Theorem 4.6.5].) We will apply the May spectral sequence [16] in our context, for which: ∼= gr H∗(R# kG, k). ∼= H∗(gr(R# kG), k) =⇒ E ∗,∗ E ∗,∗ ∞ 1 The algebra gr(R# kG) ∼= k[w, x, y]/(wq, xp, yp) has a resolution given by a tensor prod- uct as in Lemma 4.2.3, equivalently by repeating the twisted tensor product construction in Section 3.3 but with trivial twisting. We find that there are elements in degree 2 of H∗(gr(R# kG), k) corresponding to ξw, ξx, ξy ∈ H2(R# kG, k) (here, we identify ξw = ξg), and we use the same notation for them, by abuse of notation. These elements are permanent cycles in the May spectral sequence: We have already seen that these cocycles exist for the filtered algebra R# kG, as constructed in Section 3.3. They are permanent cocycles as we may identify their images with the corresponding elements of H∗(R# kG, k). Specifically, let B∗,∗ = k[ξw, ξx, ξy]. By identifying H∗(gr(R# kG), k) with group coho- ∼= H∗(gr(R# kG), k) is a mology, or by arguments in [14, Section 4], we see that E ∗,∗ noetherian B∗,∗-module, (it is generated over B∗,∗ by some elements ηw, ηx, ηy in degree 1). ∼= gr H∗(R# kG, k) is a noetherian module over k[ξw, ξx, ξy]. By an By Lemma 5.1.1, E ∗,∗ ∞ appropriate Zariskian filtration [13, Chapter 2], one can lift information from the associ- ated graded ring to the filtered ring; thus, H∗(R# kG, k) is noetherian over k[ξw, ξx, ξy]. Therefore, by [6, Proposition 2.4], H∗(R# kG, k) is finitely generated as an algebra. (cid:3) 1 Remark 5.1.3. There is a different proof of Theorem 5.1.2 that is closer to Evens' original proof of finite generation of group cohomology. See [19, Theorem 3.1, Section 5, and Erratum] for details. COHOMOLOGY OF POINTED HOPF ALGEBRAS 27 5.2. Cohomology of some pointed Hopf algebras of dimension 27. In this section, we let k be a field of characteristic p = 3 and consider the Hopf algebras H(ǫ, µ, τ ) defined in Section 2.3. Consider k to be the H(ǫ, µ, τ )-module on which w, x, y each act as 0. Theorem 5.2.1. Let H(ǫ, µ, τ ) be a Hopf algebra of dimension 27 over an algebraically closed field k of characteristic p = 3, as defined in Section 2.3. The cohomology H∗(H(ǫ, µ, τ ), k) is finitely generated. Proof. Choose the ordering w < x < y as before, and the corresponding degree lexicographic ordering on monomials. Due to the form of the relations, this gives rise to an N-filtration on H(ǫ, µ, τ ) for which the associated graded algebra is gr H(ǫ, µ, τ ) ∼= k[w, x, y]/(w3, x3, y3). (See, e.g., [2, Theorem 4.6.5].) We consider the Anick resolution P q for H(ǫ, µ, τ ), filtered correspondingly, and the resulting May spectral sequence of the complex HomH(ǫ,µ,τ )(P q, k). This is a multiplicative spectral sequence under the product induced by a diagonal map P q → P q ⊗ P q lifting the identity map on k. Note that the associated graded resolution to P q is gr P q, which we may identify with the Anick resolution for gr H(ǫ, µ, τ ), described in Lemma 4.2.3. The Anick resolution for A = H(ǫ, µ, τ ) has the same free basis sets Cn as that for gr H(ǫ, µ, τ ) described in Section 4.2. Direct calculations show that it has the following differentials in degrees 2 and 3 (recall that the parameter ǫ only takes the values 0 or 1): By the proof of Theorem 4.1.2, values of d2 on tips correspond to the relations, specifically, d2(1 ⊗ w3) = w2 ⊗ w, d2(1 ⊗ x3) = x2 ⊗ x − ǫ ⊗ x, d2(1 ⊗ y3) = y2 ⊗ y + ǫy ⊗ y + (µǫ − τ − µ2) ⊗ y, d2(1 ⊗ xw) = x ⊗ w − w ⊗ x − ǫw ⊗ w − ǫ ⊗ w, d2(1 ⊗ yw) = y ⊗ w − w ⊗ y − w ⊗ x − 1 ⊗ x + (µ − ǫ)w ⊗ w + (µ − ǫ) ⊗ w, d2(1 ⊗ yx) = y ⊗ x − x ⊗ y + x ⊗ x − (µ + ǫ) ⊗ x − ǫ ⊗ y + τ w ⊗ w − τ ⊗ w. Values of d3 require some computation, using the algorithm outlined as part of the proof of Theorem 4.1.2, and on free basis elements they are: d3(1 ⊗ x4) = x ⊗ x3, d3(1 ⊗ w4) = w ⊗ w3, d3(1 ⊗ xw3) = x ⊗ w3 − w2 ⊗ xw, d3(1 ⊗ x3w) = x2 ⊗ xw + w ⊗ x3 + ǫwx ⊗ xw + ǫx ⊗ xw + ǫw ⊗ xw, d3(1 ⊗ yw3) = y ⊗ w3 − w2 ⊗ yw + w2 ⊗ xw + w ⊗ xw, d3(1 ⊗ yxw) = y ⊗ xw − x ⊗ yw + w ⊗ yx + ǫw ⊗ yw + x ⊗ xw + (µ + ǫ)w ⊗ xw, d3(1 ⊗ y3w) = y2 ⊗ yw + w ⊗ y3 + wy ⊗ yx + wx ⊗ yx + (ǫ − µ)wy ⊗ yw d3(1 ⊗ y4) = y ⊗ y3, +(µ − ǫ)wx ⊗ yw − τ w2 ⊗ yw + y ⊗ yx − (ǫ + µ)y ⊗ yw +τ w2 ⊗ xw + x ⊗ yx + (µ − ǫ)x ⊗ yw + (µ2 − ǫµ)w ⊗ yw + τ w ⊗ xw, d3(1 ⊗ yx3) = y ⊗ x3 − x2 ⊗ yx + τ wx ⊗ xw + ǫx ⊗ yx − τ x ⊗ xw + ǫτ w ⊗ xw, d3(1 ⊗ y3x) = y2 ⊗ yx + x ⊗ y3 − xy ⊗ yx − τ wx ⊗ yw − τ wy ⊗ yw +τ w2 ⊗ yx + τ wx ⊗ xw + ǫτ w2 ⊗ yw + (ǫτ + µτ )w2 ⊗ xw +µy ⊗ yx + τ y ⊗ yw − µx ⊗ yx + τ x ⊗ xw +τ w ⊗ yx + (ǫτ + µτ )w ⊗ yw + ǫτ w ⊗ xw. 28 VAN C. NGUYEN, XINGTING WANG, AND SARAH WITHERSPOON For example, to find d3(1 ⊗ yw3), we first compute d3(1 ⊗ yw3) = y ⊗ w3 − s1d2(y ⊗ w3) = y ⊗ w3 − s1(yw2 ⊗ w). Now, using the relations, rewrite yw2 as w2y − w2x − wx + (µ + ǫ)w2 + ǫw so that the above expression is = y ⊗ w3 − s1(w2y ⊗ w − w2x ⊗ w − wx ⊗ w + (µ + ǫ)w2 ⊗ w + ǫw ⊗ w). Now d2(w2 ⊗ yw) = w2y ⊗ w − w2 ⊗ x + (µ − ǫ)w2 ⊗ w and, adding and subtracting −w2 ⊗ x + (µ − ǫ)w2 ⊗ w, the above may be rewritten as = y ⊗ w3 − s1(w2y ⊗ w − w2 ⊗ x + (µ − ǫ)w2 ⊗ w + w2 ⊗ x − (µ − ǫ)w2 ⊗ w − w2x ⊗ w − wx ⊗ w + (µ + ǫ)w2 ⊗ w + ǫw ⊗ w) = y ⊗ w3 − w2 ⊗ yw − s1(w2 ⊗ x − w2x ⊗ w − wx ⊗ w − ǫw2 ⊗ w + ǫw ⊗ w)). For the next two steps, we note that d2(w2 ⊗ xw) = w2x ⊗ w − ǫw2 ⊗ w and d2(w ⊗ xw) = wx ⊗ w − w2 ⊗ x − ǫw2 ⊗ w − ǫw ⊗ w and so the above may be rewritten as = y ⊗ w3 − w2 ⊗ yw − s1(w2 ⊗ x − w2x ⊗ w + ǫw2 ⊗ w + ǫw2 ⊗ w − wx ⊗ w + ǫw ⊗ w) = y ⊗ w3 − w2 ⊗ yw + w2 ⊗ xw − s1(w2 ⊗ x − wx ⊗ w + ǫw2 ⊗ w + ǫw ⊗ w). We recognize the argument of s1 above as d2(w⊗xw) and so we obtain the value of d3(1⊗yw3) as claimed. Looking at the values of d3 given above, note that the coefficients in the factor A = H(ǫ, µ, τ ) of A ⊗ kC2 in the image of each of these free basis elements (under d3) are in the augmentation ideal. (A similar statement does not apply to d2.) In particular, letting (w3)∗, . . . denote elements in the dual basis in Homk(kC2, k) ∼= HomA(A ⊗ kC2, k), where A = H(ǫ, µ, τ ), to the tips w3, . . . of kC2, it follows that 3((w3)∗) = 0, d∗ 3((x3)∗) = 0, d∗ 3((y3)∗) = 0. d∗ Setting ξw = (w3)∗, ξx = (x3)∗, ξy = (y3)∗, we see that these functions are cocycles in Homk(kC2, k) ∼= HomA(A ⊗ kC2, k). It follows from the above observations that ξw, ξx, ξy are permanent cocycles in the May spectral sequence, and we may use them in an application of Lemma 5.1.1: On the E1-page, ξw, ξx, ξy correspond to analogous 2-cocycles on gr H(ǫ, µ, τ ) that generate a polynomial subalgebra of its cohomology ring by a similar analysis to that in earlier sections. That is, by Lemma 4.2.3, the Anick resolution is essentially the same as the (twisted) tensor product resolution used earlier. Now let B = k[ξw, ξx, ξy]. Let ηw = (w)∗, ηx = (x)∗, ηy = (y)∗ in Homk(kC1, k) ∼= HomA(A ⊗ kC1, k). The cohomology of gr H(ǫ, µ, τ ) is finitely generated as a module over B, by ηw, ηx, ηy and their products (note η2 y = 0, so these products constitute a finite set). By Lemma 5.1.1, using an appropriate Zariskian filtration [13, Chapter 2], the cohomology H∗(H(ǫ, µ, τ ), k) is noetherian over k[ξw, ξx, ξy]. By [6, Propostion 2.4], it is finitely as an algebra. (cid:3) w = 0, η2 x = 0, η2 Remark 5.2.2. An alternative proof of our earlier Theorem 5.1.2 would proceed just as our above proof of Theorem 5.2.1: One could compute the differentials on the Anick resolution of the algebra there to show existence of the needed elements ξw, ξx, ξy. We chose instead to use the twisted tensor product construction, for which we were able to give formulas for the differentials in all degrees, yielding a more explicit, if not shorter, presentation. COHOMOLOGY OF POINTED HOPF ALGEBRAS 29 References [1] D. J. Anick, On the homology of associative algebras, Trans. Amer. Math. Soc. 296 (1986), no. 2, 641 -- 659. [2] J. L. Bueso, J. G´omez-Torecillas, and A. Verschoren, Algorithmic Methods in Non-Commutative Algebra, Kluwer Academic Publishers, 2003. [3] C. Bendel, D. K. Nakano, B. J. Parshall, and C. Pillen, Cohomology for quantum groups via the geometry of the nullcone, Mem. Amer. Math. Soc. 229 (2014), no. 1077. [4] C. Cibils, A. Lauve, and S. Witherspoon, Hopf quivers and Nichols algebras in positive characteristic, Proc. Amer. Math. Soc. 137 (2009), no. 12, 4029 -- 4041. [5] S. Cojocaru and V. Ufnarovski, BERGMAN under MS-DOS and Anick's resolution, Discrete Math. Theoretical Comp. Sci. 1 (1997), 139 -- 147. [6] L. Evens, The cohomology ring of a finite group, Trans. Amer. Math. Soc. 101 (1961), 224 -- 239. [7] E. Friedlander and B. Parshall, Cohomology of infinitesimal and discrete groups, Math. Ann. 273 (1986), 353 -- 374. [8] E. Friedlander and A. Suslin, Cohomology of finite group schemes over a field, Invent. Math. 127 (1997), no. 2, 209 -- 270. [9] V. Ginzburg and S. Kumar, Cohomology of quantum groups at roots of unity, Duke Math. J. 69 (1993), 179 -- 198. [10] E. Golod, The cohomology ring of a finite p-group, (Russian) Dokl. Akad. Nauk SSSR 235 (1959), 703 -- 706. [11] I. G. Gordon, Cohomology of quantized function algebras at roots of unity, Proc. London Math. Soc. (3) 80 (2000), no. 2, 337 -- 359. [12] E. L. Green and Ø. Solberg, An algorithmic approach to resolutions, J. Symbolic Comput. 42 (2007), no. 11 -- 12, 1012 -- 1033. [13] H.-S. Li and F. Van Oystaeyen, "Zariskian filtrations", K-Monographs in Mathematics, 2, Kluwer Academic Publishers, Dordrecht, 1996. [14] M. Mastnak, J. Pevtsova, P. Schauenburg, and S. Witherspoon, Cohomology of finite dimensional pointed Hopf algebras, Proc. London Math. Soc. (3) 100 (2010), no. 2, 377 -- 404. [15] S. Majid, Crossed products by braided groups and bosonization, J. Algebra 163 (1994), 165 -- 190. [16] J. P. May, The cohomology of restricted Lie algebras and of Hopf algebras, J. Algebra 3 (1966), 123 -- 146. [17] S. Montgomery, "Hopf Algebras and Their Actions on Rings", CBMS Regional Conference Series in Mathematics, 82, Amer. Math. Soc., Providence, RI, 1993. [18] V. C. Nguyen and X.-T. Wang, Pointed p3-dimensional Hopf algebras in positive characteristic, arXiv:1609.03952. [19] V. C. Nguyen and S. Witherspoon, Finite generation of the cohomology of some skew group algebras, Algebra and Number Theory 8 (2014), no. 7, 1647 -- 1657; Erratum, in preparation. [20] A. V. Shepler and S. Witherspoon, Resolutions for twisted tensor products, arXiv:1610.00583. [21] P. Shroff, Finite generation of the cohomology of quotients of PBW algebras, J. Algebra 390 (2013), 44 -- 55. [22] D. S¸tefan and C. Vay, The cohomology ring of the 12-dimensional Fomin-Kirillov algebra, Adv. Math. 291 (2016), 584 -- 620. [23] D. E. Radford, The structure of Hopf algebras with a projection, J. Algebra 92 (1985), 322 -- 347. [24] B. B. Venkov, Cohomology algebras for some classifying spaces, Dokl. Akad. Nauk. SSR 127 (1959), 943 -- 944. Department of Mathematics, Hood College, Frederick, MD 21701 E-mail address: [email protected] Department of Mathematics, Temple University, Philadelphia, PA 19122 E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, TX 77843 E-mail address: [email protected]
1203.0729	1	1203	2012-03-04T11:07:50	Correspondences of coclosed submodules	[ "math.RA" ]	We establish an order-preserving bijective correspondence between the sets of coclosed elements of some bounded lattices related by suitable Galois connections. As an application, we deduce that if $M$ is a finitely generated quasi-projective left $R$-module with $S=End_R(M)$ and $N$ is an $M$-generated left $R$-module, then there exists an order-preserving bijective correspondence between the sets of coclosed left $R$-submodules of $N$ and coclosed left $S$-submodules of $Hom_R(M,N)$.	math.RA	math	CORRESPONDENCES OF COCLOSED SUBMODULES SEPTIMIU CRIVEI, HATICE INANKIL, M. TAMER KOS¸AN, AND GABRIELA OLTEANU Abstract. We establish an order-preserving bijective correspondence between the sets of co- closed elements of some bounded lattices related by suitable Galois connections. As an ap- plication, we deduce that if M is a finitely generated quasi-projective left R-module with S = EndR(M ) and N is an M -generated left R-module, then there exists an order-preserving bijective correspondence between the sets of coclosed left R-submodules of N and coclosed left S-submodules of HomR(M, N ). 1. Introduction An important general problem in module theory is to relate properties of a module with prop- erties of its endomorphism ring; or more generally, for a left R-module M with S = EndR(M ), to relate properties of a left R-module N with properties of the left S-module HomR(M, N ). Within the vast literature dealing with this problem, we point out the work of J. Zelmanowitz [12], which is closely related to our topic and motivates our study. He showed that if M is a left R-module with S = EndR(M ) and N is an M -faithful left R-module, then there exists an order-preserving bijective correspondence between the sets of closed left R-submodules of N and closed left S-submodules of HomR(M, N ) [12, Theorem 1.2]. The aim of the present paper is to establish a result dual to that of J. Zelmanowitz, in terms of coclosed submodules. While closed submodules coincide with complement submodules and every module has a complement, in general coclosed submodules are different of supplement submodules and not every submodule has a supplement (see [2]). These are some of the main obstacles in dualizing results on topics related to coclosed submodules; for instance compare the theories of lifting modules [2] and extending modules [5]. In order to overcome these problems, we need to find a suitable setting for our results and to use a more general approach. We shall first consider the context of bounded lattices and we shall make use of Galois connections between them. The reason for doing that is twofold: first, the notions involved are lattice-theoretic, and secondly, our approach clarifies the exposition and gives a more natural explanation for certain conditions we have to impose on our modules. Our main theorem shows that if (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) are two bounded lattices, and (α, β) is a special Galois connection (whose properties will be detailed later on), then α and β induce mutually inverse bijections between the set of coclosed elements a ∈ A such that α(a) is coclosed in B and the set of coclosed elements b ∈ B such that β(b) has a unique coclosure in A. In particular, under certain conditions, we obtain order-preserving mutually inverse bijections between the sets of coclosed elements in A and coclosed elements in B, which dualize and generalize the main theorem of J. Zelmanowitz [12, Theorem 1.2]. We apply our result to a particular Galois connection for modules, previously pointed out by T. Albu and C. Nastasescu [1], in order to deduce the following consequence: if M is a finitely generated quasi-projective left R-module with S = EndR(M ) and N is an M - generated left R-module, then there exists an order-preserving bijective correspondence between the sets of coclosed left R-submodules of N and coclosed left S-submodules of HomR(M, N ). We also relate the dual Goldie dimensions as well as the supplemented and the lifting properties of the left R-module N and the left S-module HomR(M, N ). 2000 Mathematics Subject Classification. 16D10, 06A15. Key words and phrases. Galois connection, lattice, coclosed submodule, quasi-projective module, lifting module. The first author acknowledges the support of the grant PN-II-RU-TE-2011-3-0065. The fourth author acknowl- edges the support of the grant PN-II-RU-TE-2009-1 project ID 303. Part of the paper was carried out when the first author was visiting Gebze Institute of Technology in November 2010. He gratefully acknowledges the support of TUBITAK and the kind hospitality of the host university. 1 2 S. CRIVEI, H. INANKIL, M.T. KOS¸AN, AND G. OLTEANU 2. Cosmall Galois connections We shall make use of the concept of (monotone) Galois connection (e.g., see [6]). This is defined on arbitrary partially ordered sets, but we recall its definition on lattices, this being the setting in which we shall employ it. Definition 2.1. Let (A, ≤) and (B, ≤) be lattices. A Galois connection between them consists of a pair (α, β) of two order-preserving functions α : A → B and β : B → A such that for all a ∈ A and b ∈ B, we have α(a) ≤ b if and only if a ≤ β(b). Equivalently, (α, β) is a Galois connection if and only if for all a ∈ A, a ≤ βα(a) and for all b ∈ B, αβ(b) ≤ b. An element a ∈ A (respectively b ∈ B) is called a Galois element if βα(a) = a (respectively αβ(b) = b). As usual, one may view any lattice (A, ≤) as a triple (A, ∧, ∨), where ∧ and ∨ denote the infimum and the supremum of elements in A. Recall that the lattice A is bounded if it has a least element, denoted by 0, and a greatest element, denoted by 1. If A is bounded, then we tacitly assume that 0 6= 1, and we denote it as (A, ∧, ∨, 0, 1). For a, a′ ∈ A, we also denote [a, a′] = {x ∈ A a ≤ x ≤ a′}. We gather in the following lemma some well-known results (e.g., see [1, Proposition 3.3], [6]), which shall be used throughout the paper without further reference. Lemma 2.2. Let (A, ≤) and (B, ≤) be lattices, and (α, β) a Galois connection, where α : A → B and β : B → A. Then: (i) αβα = α and βαβ = β. (ii) α preserves all suprema in A and β preserves all infima in B. (iii) If A and B are bounded, then α(0) = 0 and β(1) = 1. (iv) The restrictions of α and β to the corresponding sets of Galois elements are mutually inverse bijections. The module-theoretic concepts of direct summand, cosmall inclusion, coclosed submodule and coclosure of a submodule (e.g., see [2]) have natural lattice counterparts. Definition 2.3. Let (A, ∧, ∨, 0, 1) be a bounded lattice. (1) An element a ∈ A is called a complement in A if there exists a′ ∈ A such that a ∧ a′ = 0 and a ∨ a′ = 1. (2) Let a, a′ ∈ A be such that a ≤ a′. Then a′ is called cosmall in [a, 1] if for any x ∈ A, 1 = a′ ∨ x implies 1 = a ∨ x. (3) An element a′ ∈ A is called coclosed in A if for any a ∈ A, a′ cosmall in [a, 1] implies a = a′. (4) An element a′ ∈ A is called a coclosure of a ∈ A in A if a is cosmall in [a′, 1] and a′ is coclosed in A. The following lemma is well-known for submodule lattices, and its proof is straightforward. Lemma 2.4. Let (A, ∧, ∨, 0, 1) be a bounded lattice. (i) Let a, b, c ∈ A such that a ≤ b ≤ c. Then c is cosmall in [a, 1] if and only if b is cosmall in [a, 1] and c is cosmall in [b, 1]. (ii) If A is modular, then every complement in A is coclosed. Note that a coclosure of a given element might not exist. For instance, the subgroup 2Z of the abelian group Z has no coclosure in the subgroup lattice of Z ([2, 3.10]). We continue with an easy, but useful lemma. Lemma 2.5. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded lattices, and (α, β) a Galois con- nection, where α : A → B, α(1) = 1 and β : B → A preserves finite suprema. Then: (i) For all b ∈ B, b is cosmall in [αβ(b), 1]. (ii) Every coclosed element of B is Galois. Proof. (i) Let b ∈ B and let 1 = b ∨ b′ for some b′ ∈ B. Then 1 = αβ(1) = αβ(b) ∨ αβ(b′) ≤ αβ(b) ∨ b′, and so 1 = αβ(b) ∨ b′. Hence b is cosmall in [αβ(b), 1]. (ii) Clear by (i). (cid:3) CORRESPONDENCES OF COCLOSED SUBMODULES 3 Note that if (α, β) is a Galois connection between two bounded lattices and α, β are lattice homomorphisms, then α(1) = 1 and β preserves finite suprema. We introduce two special types of Galois connection, which will be useful to us. Definition 2.6. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded lattices, and (α, β) a Galois connection, where α : A → B and β : B → A. We say that (α, β) is cosmall if for all a ∈ A, βα(a) is cosmall in [a, 1]. We say that a cosmall Galois connection (α, β) is UCC if for every coclosed element a ∈ A, a is the unique coclosure of βα(a) in A. We present some examples to illustrate the above theory. Example 2.7. Consider the abelian groups G = Zp × Z q2 for some primes p and q with p 6= q, and G′ = Z2 × Z4, where Zn denotes the cyclic group of order n ∈ N. Their subgroup lattices S(G) and S(G′) have the following forms respectively: ⑥⑥⑥⑥⑥⑥⑥⑥ H4 H2 ✵ ✵ ✵ ✵✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ G ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ H3 ✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺ H1 ⑥⑥⑥⑥⑥⑥⑥⑥ 0 ❆❆❆❆❆❆❆❆ ⑥⑥⑥⑥⑥⑥⑥⑥ H ′ 6 G′ ⑥⑥⑥⑥⑥⑥⑥⑥ ❆❆❆❆❆❆❆❆ ⑥⑥⑥⑥⑥⑥⑥⑥ H ′ 5 H ′ 3 H ′ 1 H ′ 4 ⑥⑥⑥⑥⑥⑥⑥⑥ ❆❆❆❆❆❆❆❆ H ′ 2 0′ It is easy to check that H1 is cosmall in [0, G], H4 is cosmall in [H2, G], H1 and H4 are the only subgroups of G which are not coclosed, 0 is a coclosure of H1 in G, and H2 is a coclosure 3 and H ′ of H4 in G. Also, H ′ 4 1, H ′ are the only subgroups of G′ which are not coclosed, 0′ is a coclosure of H ′ 2 are coclosures of H ′ 4 in G′. For properties of subgroups with unique (co)closure and cosmall subgroups of abelian groups the reader is referred to [3] and [4]. 3 is cosmall in [0′, G′], H ′ 3 in G′, and H ′ 4 is cosmall in [H ′ 1, G′] and [H ′ 2, G′], H ′ (1) Consider the functions α : S(G) → S(G) defined by α(0) = α(H1) = 0, α(H2) = α(H4) = H3, α(H3) = H2 and α(G) = G, β : S(G) → S(G) defined by β(0) = β(H1) = H1, β(H2) = β(H4) = H3, β(H3) = H4, β(G) = G. Then (α, β) is a Galois connection from the lattice (S(G), ⊆) to itself. For every H ∈ S(G) \ {H1, H4} we have αβ(H) = H. Hence 0, H2, H3 and G are Galois elements in the codomain B = S(G) of α. Also, H1 is cosmall in [αβ(H1), G] = [0, G] and H4 is cosmall in [αβ(H4), G] = [H2, G]. On the other hand, for every H ∈ S(G) \ {0, H2} we have βα(H) = H. Hence H1, H3, H4 and G are Galois elements in the domain A = S(G) of α. Also, βα(0) = H1 is cosmall in [0, G] and βα(H2) = H4 is cosmall in [H2, G]. Moreover, for every coclosed H ∈ S(G), H is the unique coclosure of βα(H) in S(G). Hence (α, β) is a UCC cosmall Galois connection. Note that H2 is a coclosed element, but not a Galois element in the domain A = S(G) of α. Hence not every coclosed element of A is Galois under the hypotheses of Lemma 2.5. (2) Consider the functions α : S(G) → S(G) defined by α(H2) = H4, α(H3) = G, and α(H) = H for every H ∈ S(G) \ {H2, H3}, and β : S(G) → S(G) defined by β(H2) = 0, β(H3) = H1 and β(H) = H for every H ∈ S(G) \ {H2, H3}. Then (α, β) is a Galois connection from the lattice (S(G), ⊆) to itself. But (α, β) is not a cosmall Galois connection, because for instance βα(H3) = G is not cosmall in [H3, G]. Using the same setting, let us also show that the hypothesis on β to preserve finite suprema in Lemma 2.5 cannot be removed. Note that we have β(H3 + H4) = G 6= H4 = β(H3) + β(H4). Then H2 is not cosmall in [αβ(H2), G] = [0, G]. Also, H4 is a Galois element in the codomain B = S(G) of α, but not a coclosed element. 4 S. CRIVEI, H. INANKIL, M.T. KOS¸AN, AND G. OLTEANU 1) = H ′ 4 and β(H ′) = H ′ for every H ′ ∈ S(G′)\{0′, H ′ (3) Consider the functions α′ : S(G′) → S(G′) defined by α′(H ′ α(H ′) = H ′ for every H ′ ∈ S(G′) \ {H ′ β ′(H ′ from the lattice (S(G′), ⊆) to itself. For every H ′ ∈ S(G′) \ {H ′ Also, H ′ the other hand, for every H ′ ∈ S(G′) \ {0′, H ′ is cosmall in [0′, G′] and β ′α′(H ′ Galois connection. But it is not UCC, because for the coclosed subgroup H ′ coclosures in G′, namely H ′ 1 and 4}, and β ′ : S(G′) → S(G′) defined by β ′(0′) = H ′ 3, 1}. Then (α′, β ′) is a Galois connection 4} we have αβ(H ′) = H ′. 1, G′]. On 1} we have β ′α′(H ′) = H ′. Also, β ′α′(0′) = H ′ 3 1, G′]. Hence (α′, β ′) is a cosmall 1) has two 3), G′] = [0′, G′] and H ′ 4 is cosmall in [α′β ′(H ′ 4), G′] = [H ′ 3 is cosmall in [α′β ′(H ′ 3) = 0, α(H ′ 4) = H ′ 1) = H ′ 4 is cosmall in [H ′ 1 and H ′ 2. 1, β ′α′(H ′ 3, H ′ 3, H ′ We end this section with some results which show that the cosmall property and the dual Goldie dimension may be transferred through cosmall Galois connections. Lemma 2.8. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded lattices, and (α, β) a cosmall Galois connection, where α : A → B, α(1) = 1 and β : B → A preserves finite suprema. (i) Let a, a′ ∈ A. Then a′ is cosmall in [a, 1] if and only if α(a′) is cosmall in [α(a), 1]. (ii) Let b, b′ ∈ B. Then b′ is cosmall in [b, 1] if and only if β(b′) is cosmall in [β(b), 1]. Proof. (i) Assume that a′ is cosmall in [a, 1]. Let 1 = α(a′) ∨ b′ for some b′ ∈ B. Then 1 = β(1) = βα(a′)∨β(b′). Since (α, β) is a cosmall Galois connection, it follows that 1 = a′∨β(b′). Now by hypothesis, we must have 1 = a ∨ β(b′). Then 1 = α(1) = α(a) ∨ αβ(b′) ≤ α(a) ∨ b′, and so 1 = α(a) ∨ b′. Hence α(a′) is cosmall in [α(a), 1]. Conversely, assume that α(a′) is cosmall in [α(a), 1]. Let 1 = a′ ∨ a′′ for some a′′ ∈ A. Then 1 = α(1) = α(a′) ∨ α(a′′), whence 1 = α(a) ∨ α(a′′) by hypothesis. Then 1 = β(1) = βα(a) ∨ βα(a′′). Since (α, β) is a cosmall Galois connection, it follows that 1 = a ∨ a′′. Hence a′ is cosmall in [a, 1]. (ii) Assume that b′ is cosmall in [b, 1]. Let 1 = β(b′) ∨ a′ for some a′ ∈ A. Then 1 = α(1) = αβ(b′) ∨ α(a′) ≤ b′ ∨ α(a′), hence 1 = b′ ∨ α(a′), and by hypothesis 1 = b ∨ α(a′). Then 1 = β(1) = β(b)∨βα(a′). Since (α, β) is a cosmall Galois connection, it follows that 1 = β(b)∨a′. Hence β(b′) is cosmall in [β(b), 1]. Conversely, assume that β(b′) is cosmall in [β(b), 1]. Let 1 = b′ ∨ b′′ for some b′′ ∈ B. Then 1 = β(1) = β(b′) ∨ β(b′′), whence 1 = β(b) ∨ β(b′′) by hypothesis. Then 1 = α(1) = αβ(b) ∨ αβ(b′′) ≤ b ∨ b′′, and so 1 = b ∨ b′′. Hence b′ is cosmall in [b, 1]. (cid:3) Let (X, ∧, ∨, 0, 1) be a bounded modular lattice. Recall that a subset Y of X \ {1} is called meet-independent if (y1 ∧ . . . ∧ yn) ∨ x = 1 for every finite subset {y1, . . . , yn} of Y and every x ∈ Y \ {y1, . . . , yn}. If there is a finite supremum d of all numbers k such that X has a meet- independent subset with k elements, then X has dual Goldie dimension (or hollow dimension) d; otherwise X has infinite dual Goldie dimension (see [9, Theorem 9]). We denote the dual Goldie dimension of X by hdim(X). Corollary 2.9. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded modular lattices, and (α, β) a cosmall Galois connection, where α : A → B, α(1) = 1 and β : B → A preserves finite suprema. Then: (i) hdim(A) ≤ hdim(B). (ii) If every element of A is Galois, then hdim(A) = hdim(B). Proof. (i) We prove that if A has a finite meet-independent subset with m elements, then so has B. This will imply that hdim(A) ≤ hdim(B) in both finite and infinite cases for hdim(A). To this end, let {a1, . . . , am} be a meet-independent subset of A. We claim that Y = {α(a1), . . . , α(am)} is a meet-independent subset of B with m elements. First, we point out that α(ai) 6= 1 for every i ∈ {1, . . . , m}. Indeed, if there is i ∈ {1, . . . , m} such that α(ai) = 1, then βα(ai) = 1. Since βα(ai) is cosmall in [ai, 1], we must have ai = 1, contradiction. Secondly, suppose that α(ai) = α(aj) for some i, j ∈ {1, . . . , m} with i 6= j. Since ai ∨ aj = 1, we have α(ai) ∨ α(aj) = α(ai ∨ aj) = α(1) = 1. We conclude that α(ai) = α(aj) = 1, which contradicts the previous point. Hence Y has m elements. CORRESPONDENCES OF COCLOSED SUBMODULES 5 Finally, the meet-independence of {a1, . . . , am} is known to be equivalent to the condition (a1∧ . . .∧ak−1)∨ak = 1 for every k ∈ {2, . . . , m}. Let k ∈ {2, . . . , m}. Then α(a1∧. . .∧ak−1)∨α(ak) = 1. Since α(a1∧. . .∧ak−1) ≤ α(a1)∧. . .∧α(ak−1), it follows that (α(a1)∧. . .∧α(ak−1))∨α(ak) = 1. This shows that Y is meet-independent. (ii) Suppose that every element of A is Galois. When A has infinite dual Goldie dimension, the conclusion is clear by (i). Assume that hdim(A) = m. By [9, Theorem 9], A has a meet- independent subset {a1, . . . , am} such that a1 ∧ . . . ∧ am is cosmall in [0, 1]. Now assume that there is a meet-independent subset Y of B with more than m elements, possibly infinite. Then there is a meet-independent subset {b1, . . . , bm+1} ⊆ Y of B. By dualizing an argument from the proof of [9, Theorem 5], {α(a1), . . . , α(am), bm+1} is also a meet-independent subset of B. Then we have (α(a1) ∧ . . . ∧ α(am)) ∨ bm+1 = 1. Since every element of A is Galois, it follows that (a1 ∧ . . . ∧ am) ∨ β(bm+1) = (βα(a1) ∧ . . . ∧ βα(am)) ∨ β(bm+1) = β((α(a1) ∧ . . . ∧ α(am)) ∨ bm+1) = β(1) = 1. Since a1 ∧ . . . ∧ am is cosmall in [0, 1], we deduce that β(bm+1) = 1. Then 1 = αβ(bm+1) ≤ bm+1, and so bm+1 = 1, which is a contradiction. Therefore, by (i) and the above, hdim(B) = m = hdim(A). (cid:3) 3. Correspondences Let (A, ≤) and (B, ≤) be two lattices, and let (α, β) be a Galois connection, where α : A → B and β : B → A. We have seen in Lemma 2.5 and Example 2.7 (2) that the set of coclosed elements in B is in general strictly included in the set of Galois elements in B. Also, we have already seen that the restrictions of α and β to the corresponding sets of Galois elements are mutually inverse bijections. We shall show that, under certain conditions, these bijections restrict to ones between the corresponding sets of complement Galois elements, or between the corresponding sets of coclosed Galois elements. Lemma 3.1. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded lattices, and (α, β) a Galois con- nection, where α : A → B, α(1) = 1 and β : B → A preserves finite suprema. Assume that all complements in A are Galois elements. Then: (i) If β is injective, then α preserves complement Galois elements. (ii) β preserves complement Galois elements. (iii) If β is injective, then α and β restrict to order-preserving mutually inverse bijections between the sets of complement Galois elements in A and B. Proof. First note that, since 0 ∈ A is a Galois element, we have βα(0) = 0, and so β(0) = 0. (i) Assume that β is injective. Let a be a complement (Galois) element in A. Then there is a′ ∈ A such that a ∧ a′ = 0 and a ∨ a′ = 1. Hence α(a) ∨ α(a′) = α(1) = 1. Also, we have 0 = a ∧ a′ = βα(a) ∧ βα(a′) = β(α(a) ∧ α(a′)), whence α(a) ∧ α(a′) = 0 by the injectivity of β. Thus α(a) is a complement in B. By Lemma 2.2, α(a) is a Galois element in B. (ii) Let b be a complement Galois element in B. Then there is b′ ∈ B such that b ∧ b′ = 0 and b ∨ b′ = 1. Then β(b) ∧ β(b′) = β(0) = 0 and β(b) ∨ β(b′) = β(1) = 1. Thus β(b) is a complement in A. By Lemma 2.2, β(b) is a Galois element in A. (iii) Clear by (i), (ii) and Lemma 2.2. (cid:3) Next we need to recall the following notions, which are the lattice-theoretic versions of the corresponding ones for modules (e.g., see [2]). Definition 3.2. Let (A, ∧, ∨, 0, 1) be a bounded lattice. (1) An element a ∈ A is called a supplement if it is a supplement of some a′ ∈ A, that is, a is minimal in A with the property that 1 = a ∨ a′. (2) A is called supplemented if every a ∈ A has a supplement in A. (3) A is called amply supplemented if for every a ∈ A there exists a supplement x ∈ A such that a is cosmall in [x, 1]. (4) A is called UCC if every a ∈ A has a unique coclosure in A, which will be denoted by ¯a. (5) A is called lifting if for every a ∈ A there exists a complement x ∈ A such that a is cosmall in [x, 1]. 6 S. CRIVEI, H. INANKIL, M.T. KOS¸AN, AND G. OLTEANU The following lemma has a similar proof as for modules (see [7, Corollary 3.8], [10, Lemma 1.1 and Proposition 1.5] and [11, Chapter 41]). Lemma 3.3. Let (A, ∧, ∨, 0, 1) be a bounded modular lattice. (i) If a ∈ A is a supplement, then a is coclosed. (ii) If A is supplemented and a ∈ A is coclosed, then a is a supplement. (iii) If A is amply supplemented, then every element a ∈ A has a coclosure in A. (iv) If A is amply supplemented UCC and a1, a2 ∈ A are such that a1 ≤ a2, then their coclosures satisfy a1 ≤ a2. Now we may relate supplemented and lifting type properties of some bounded modular lattices. Theorem 3.4. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded modular lattices and (α, β) a Galois connection, where α : A → B, α(1) = 1 and β : B → A preserves finite suprema. Assume that every element of A is Galois. Then: (i) A is supplemented if and only if B is supplemented. (ii) If B is lifting, then A is lifting. Conversely, if β is injective and A is lifting, then B is lifting. Proof. (i) First assume that A is supplemented. Let b ∈ B. Then β(b) has a supplement x in A. Hence 1 = β(b) ∨ x, which implies that 1 = α(1) = αβ(b) ∨ α(x) ≤ b ∨ α(x), and so 1 = b ∨ α(x). Now let 1 = b ∨ y′ for some y′ ∈ B with y′ ≤ α(x). Then 1 = β(1) = β(b) ∨ β(y′). Since x is Galois, we have β(y′) ≤ βα(x) = x, whence β(y′) = x by the minimality of x. It follows that α(x) = αβ(y′) ≤ y′, and so y′ = α(x). This shows the minimality of the supplement α(x) of b in B. Thus B is supplemented. Conversely, assume that B is supplemented, and let a ∈ A. Then α(a) has a supplement y in B, and so 1 = α(a)∨y. By hypothesis it follows that 1 = β(1) = βα(a)∨β(y) = a∨β(y). Now let 1 = a ∨ x′ for some x′ ∈ A with x′ ≤ β(y). Then 1 = α(1) = α(a) ∨ α(x′) and α(x′) ≤ αβ(y) ≤ y. By the minimality of y, α(x′) = y. Since x′ is Galois, we have x′ = βα(x′) = β(y). This shows the minimality of the supplement β(y) of a in A. Thus A is supplemented. (ii) Assume that B is lifting. Let a ∈ A. Then there exists a complement b′ ∈ B such that α(a) is cosmall in [b′, 1]. Since a is Galois, a = βα(a) is cosmall in [β(b′), 1] by Lemma 2.8. By Lemmas 2.4 and 2.5, b′ is a Galois element in B. Then by Lemma 3.1, β(b′) is a complement in A. It follows that A is lifting. Now assume that β is injective and A is lifting. Let b ∈ B. Then there exists a complement a′ ∈ A such that β(b) is cosmall in [a′, 1]. By Lemma 2.8, αβ(b) is cosmall in [α(a′), 1]. By Lemma 2.5, b is cosmall in [αβ(b), 1]. Hence by Lemma 2.4, b is cosmall in [α(a′), 1]. By Lemma 3.1, α(a′) is a complement in B. It follows that B is lifting. (cid:3) Next we establish our general theorem on a bijective correspondence between sets of coclosed elements induced by some special Galois connections. In order to obtain it, it is natural to try to define some maps by means of unique coclosures of elements, when they do exist. We see in the following theorem that the Galois connection and the condition that these maps are well-defined create a slightly asymmetric situation. Theorem 3.5. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded lattices and (α, β) a UCC cosmall Galois connection, where α : A → B, α(1) = 1 and β : B → A preserves finite suprema. For x ∈ A ∪ B denote by ¯x the unique coclosure of x, when it does exist. (i) Denote A = {a ∈ A a is coclosed in A and α(a) has a unique coclosure in B}, B = {b ∈ B b is coclosed in B and β(b) has a unique coclosure in A}. Consider ϕ : A → B defined by ϕ(a) = α(a) for every a ∈ A, and ψ : B → A defined by ψ(b) = β(b) for every b ∈ B. Then ϕ is a well-defined map if and only if A coincides with the set of coclosed elements a ∈ A such that α(a) is coclosed in B. Also, ψ is a well-defined map. CORRESPONDENCES OF COCLOSED SUBMODULES 7 (ii) Denote A = {a ∈ A a is coclosed in A and α(a) is coclosed in B}, B = {b ∈ B b is coclosed in B and β(b) has a unique coclosure in A}. Then the maps ϕ : A → B defined by ϕ(a) = α(a) for every a ∈ A, and ψ : B → A defined by ψ(b) = β(b) for every b ∈ B, are mutually inverse bijections. Proof. (i) First assume that ϕ is well-defined. Let a ∈ A. Then ϕ(a) = α(a) ∈ B, and so β(α(a)) has a unique coclosure, say a0. Then β(α(a)) is cosmall in [a0, 1]. Since α(a) is cosmall in [α(a), 1], βα(a) is cosmall in [β(α(a)), 1] by Lemma 2.8. Then by Lemma 2.4 βα(a) is cosmall in [a0, 1]. Since (α, β) is UCC cosmall, a is the unique coclosure of βα(a), hence a0 = a. Then β(α(a)) is cosmall in [a, 1]. By Lemmas 2.5 and 2.8, α(a) = αβ(α(a)) is cosmall in [α(a), 1]. Hence α(a) = α(a), and so α(a) is coclosed in B. Now assume that A coincides with the set of coclosed elements a ∈ A such that α(a) is coclosed in B. Then ϕ(a) = α(a) = α(a) for every a ∈ A. Let a ∈ A. Then α(a) is coclosed in B by hypothesis. Since (α, β) is UCC cosmall, βα(a) has a unique coclosure in A. Hence ϕ(a) = α(a) ∈ B, and so ϕ is well-defined. In order to show that ψ is well-defined, let b ∈ B. Then β(b) is coclosed in A. Since β(b) is cosmall in [β(b), 1], αβ(b) is cosmall in [α(β(b)), 1] by Lemma 2.8. Since b ∈ B is coclosed, we have αβ(b) = b by Lemma 2.5. Then b is cosmall in [α(β(b)), 1]. It follows that α(β(b)) = b, because b ∈ B is coclosed. Hence α(β(b)) is coclosed in B. Thus ψ(b) = β(b) ∈ A. Note that ψ is also well-defined if the codomain is the set of coclosed elements a ∈ A such that α(a) is coclosed in B. (ii) The maps ϕ and ψ are well-defined by (i). If a ∈ A, then a is the unique coclosure of βα(a) in A because (α, β) is UCC cosmall, and we have ψϕ(a) = βα(a) = a. If b ∈ B, then we have ϕψ(b) = α(β(b)) = b as above. Therefore, ϕ and ψ are mutually inverse bijections. (cid:3) We apply Theorem 3.5 in two relevant situations as follows. The first one, when every element in A is Galois, will be particularly considered in the last section of this paper. The second one, when A is amply supplemented modular and (α, β) is UCC cosmall, may be applied to any cosmall Galois connection (α, β) between finite UCC abelian groups (e.g., see [3], [4]). Theorem 3.6. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded lattices and (α, β) a Galois con- nection, where α : A → B, α(1) = 1 and β : B → A preserves finite suprema. Then there are mutually inverse bijections between the sets CA of coclosed elements in A and CB of coclosed elements in B provided one of the following conditions holds: (i) Every element in A is Galois. (ii) A is amply supplemented modular and (α, β) is UCC cosmall. If either every element in A is Galois, or A is amply supplemented modular UCC and (α, β) is cosmall, then the above bijections are order-preserving. Proof. In both cases we use the notation from Theorem 3.5, and prove that A = CA and B = CB. Then by Theorem 3.5 the required mutually inverse bijections will be given by ϕ and ψ. Clearly, A ⊆ CA and B ⊆ CB. (i) Assume that every element in A is Galois. Then (α, β) is clearly UCC cosmall. Let a ∈ CA. By hypothesis, we have a = βα(a). We claim that α(a) is coclosed in B. To this end, let b′ ∈ B be such that α(a) is cosmall in [b′, 1]. By Lemma 2.8, a = βα(a) is cosmall in [β(b′), 1]. Since a is coclosed in A, we must have β(b′) = a. It follows that α(a) = αβ(b′) ≤ b′, whence b′ = α(a). Thus α(a) is coclosed in B, and so a ∈ A. Now let b ∈ CB. We claim that β(b) is coclosed in A. To this end, let a′ ∈ A be such that β(b) is cosmall in [a′, 1]. By Lemma 2.8, αβ(b) is cosmall in [α(a′), 1]. By Lemma 2.5, every coclosed element in B is a Galois element, hence b = αβ(b), and so b is cosmall in [α(a′), 1]. Since b is coclosed in B, it follows that α(a′) = b. Then by hypothesis we have a′ = βα(a′) = β(b). Thus β(b) is coclosed in A, and so b ∈ B. It follows that A = CA and B = CB. By Theorem 3.5, the maps ϕ = α and ψ = β are mutually inverse bijections between CA and CB. 8 S. CRIVEI, H. INANKIL, M.T. KOS¸AN, AND G. OLTEANU (ii) Assume that A is amply supplemented modular and (α, β) is UCC cosmall. Let a ∈ CA. We claim that α(a) is coclosed in B. To this end, let b′ ∈ B be such that α(a) is cosmall in [b′, 1]. By Lemma 2.8, βα(a) is cosmall in [β(b′), 1]. Since A is amply supplemented modular, Lemma 3.3 yields a coclosure of β(b′) in A, say a0. Then β(b′) is cosmall in [a0, 1], whence βα(a) is cosmall in [a0, 1] by Lemma 2.4. Hence a0 is a coclosure of βα(a) in A. Since (α, β) is UCC cosmall, a is the unique coclosure of βα(a) in A, hence a0 = a. Then a ≤ β(b′), which implies α(a) ≤ b′ because (α, β) is a Galois connection. Hence α(a) = b′, and so α(a) is coclosed in B. Thus a ∈ A. Now let b ∈ CB. We claim that β(b) has a unique coclosure in A. The existence of a coclosure of β(b) in A follows by Lemma 3.3, because A is amply supplemented modular. Now assume that a1 and a2 are two coclosures of β(b) in A. Then β(b) is coclosed both in [a1, 1] and in [a2, 1]. By Lemma 2.8, αβ(b) is cosmall both in [α(a1), 1] and in [α(a2), 1]. Since b is coclosed, we have αβ(b) = b by Lemma 2.5. Hence b is cosmall both in [α(a1), 1] and in [α(a2), 1]. But b is coclosed in B, hence we must have α(a1) = α(a2) = b. Since (α, β) is a cosmall Galois connection, βα(a1) = βα(a2) is cosmall in [a1, 1]. Then a1 is a coclosure of βα(a2) in A. Since (α, β) is UCC cosmall, a2 is the unique coclosure of βα(a2) in A, hence a1 = a2. Then β(b) has a unique coclosure in A. Thus b ∈ B. It follows that A = CA and B = CB. By Theorem 3.5, the maps ϕ and ψ are mutually inverse bijections between CA and CB. Finally, if every element in A is Galois, then ϕ and ψ are order-preserving, because so are α and β. Also, if A is amply supplemented modular UCC and (α, β) is cosmall, then ϕ and ψ are order-preserving by hypothesis and Lemma 3.3. (cid:3) Corollary 3.7. Let (A, ∧, ∨, 0, 1) and (B, ∧, ∨, 0, 1) be bounded lattices and (α, β) a UCC cosmall Galois connection, where α : A → B, α(1) = 1, β : B → A preserves finite suprema, and A is amply supplemented modular. Then the following are equivalent: (i) The mutually inverse bijections from Theorem 3.6 are the restrictions of α and β to the sets CA of coclosed elements in A and CB of coclosed elements in B respectively. (ii) Every coclosed element in A is Galois. Proof. Note that (i) is equivalent to the condition: β(b) = β(b) for every coclosed element b ∈ B. First assume (i). Let a ∈ A be a coclosed element. By Theorems 3.5 and 3.6 (ii) we have a = βα(a). By hypothesis we have βα(a) = βα(a). Hence a = βα(a), and so a is Galois. Conversely, assume (ii). Let b ∈ B be a coclosed element. By hypothesis the coclosed element β(b) ∈ A is Galois, hence β(b) = βα(β(b)). By Theorems 3.5 and 3.6 (ii) we have α(β(b)) = b. Hence β(b) = β(b). (cid:3) Remark 3.8. (1) Under the hypotheses of Theorem 3.6, every coclosed element in B is Galois by Lemma 2.5. In case every element in A is Galois, Theorem 3.6 establishes in fact a bijection between the sets of coclosed Galois elements in A and B. We point out that this is not true in general. Indeed, let (α, β) be the UCC cosmall Galois connection from Example 2.7 (1). Then α(G) = G and β preserves finite suprema. The coclosed Galois elements in the domain A = S(G) of α are H3 and G, whereas the coclosed Galois elements in the codomain B = S(G) of α are 0, H2, H3 and G. Hence there is no bijection between the two sets of coclosed Galois elements. Note that there are elements in A = S(G) which are not Galois, for instance H2. (2) Consider again the UCC cosmall Galois connection from Example 2.7 (1). Then α(G) = G and β preserves finite suprema. Also, A = S(G) is amply supplemented modular, as any subgroup lattice of a finite group. But condition (i) in Corollary 3.7 does not hold, because for instance H3 is coclosed in S(G), but β(H3) = H4 is not coclosed in S(G). Obviously, neither condition (ii) in Corollary 3.7 holds, because for instance H2 is a coclosed element in A = S(G) which is not Galois. 4. Applications In this section we apply the above results to submodule lattices of suitable modules, which are clearly bounded modular lattices. Let us first identify some Galois connection between submodule lattices. CORRESPONDENCES OF COCLOSED SUBMODULES 9 For modules M and X, we denote by Gen(M ) the set of M -generated modules, and by TrM (X) = M HomR(M, X) the greatest submodule of X which is M -generated. Let M and N be left R-modules, and let S = EndR(M ). Consider the following functions between the complete modular lattices SR(N ) and SS(HomR(M, N )) of left R-submodules of N and left S-submodules of HomR(M, N ) respectively: α : SS(HomR(M, N )) → SR(N ), α(Y ) = M Y, β : SR(N ) → SS(HomR(M, N )), β(X) = HomR(M, X). Then (α, β) is a Galois connection [1, Proposition 3.4]. The coclosed elements in SR(N ) are exactly the coclosed submodules of N . A submodule X of N is a Galois element in SR(N ) if and only if X ∈ Gen(M ) [1, Proposition 3.5]. Now assume that M is finitely generated quasi-projective and N ∈ Gen(M ). Then α(HomR(M, N )) = N , β preserves finite suprema and Y = HomR(M, M Y ) = βα(Y ) for every submodule Y of HomR(M, N ) by [1, Proposition 4.9] or [11, 18.4]. Hence every submodule of HomR(M, N ) is Galois. Let σ[M ] be the full subcategory of the category of left R-modules consisting of all submodules of M -generated modules. Recall that N ∈ σ[M ] is called M -faithful if for every 0 6= g ∈ HomR(X, N ) with X ∈ σ[M ], HomR(M, X)g 6= 0. Note that if N is M -faithful, then the above function β is clearly injective. When applied to the bounded modular submodule lattice of a module, Definition 3.2 yields the notions of supplemented and lifting modules. Now Corollary 2.9 and Theorem 3.4 give the following corollary, whose first two properties are given in [8, Theorem 4.2] and [11, 43.7] respectively. Corollary 4.1. Let M be a finitely generated quasi-projective left R-module with S = EndR(M ) and N ∈ Gen(M ). Then: (i) The left R-module N and the left S-module HomR(M, N ) have the same dual Goldie dimension. (ii) N is supplemented if and only if HomR(M, N ) is supplemented. (iii) If N is lifting, then HomR(M, N ) is lifting. If N is M -faithful and HomR(M, N ) is lifting, then N is lifting. Finally, Theorem 3.6 yields the following consequence. Corollary 4.2. Let M be a finitely generated quasi-projective left R-module with S = EndR(M ), N ∈ Gen(M ), and let α, β be the above functions. Then α and β restrict to order-preserving mutually inverse bijections between the sets of coclosed left S-submodules of HomR(M, N ) and coclosed left R-submodules of N . Remark 4.3. Our theory of cosmall Galois connections is clearly dualizable to one of some naturally defined essential Galois connections. As a consequence, one may obtain a lattice- theoretic version of Zelmanowitz's results from [12]. References [1] T. Albu and C. Nastasescu, Relative finiteness in module theory, Marcel Dekker, 1984. [2] J. Clark, C. Lomp, N. Vanaja and R. Wisbauer, Lifting modules, Frontiers in Mathematics, Birkhauser, 2006. [3] S. Crivei and G. Olteanu, GAP algorithms for finite abelian groups and applications, Carpathian J. Math. 24 (2008), 310 -- 316. [4] S. Crivei and S¸. S¸uteu Szollosi, Subgroup lattice algorithms related to extending and lifting abelian groups, Int. Electron. J. Algebra 2 (2007), 54 -- 70. [5] N.V. Dung, D.V. Huynh, P.F. Smith and R. Wisbauer, Extending modules, Pitman Research Notes, 313, Longman Scientific and Technical, 1994. [6] M. Ern´e, J. Koslowski, A. Melton and G.E. Strecker, A primer on Galois connections. In: Proceedings of the 1991 Summer Conference on General Topology and Applications in Honor of Mary Ellen Rudin and Her Work, Annals of the New York Academy of Sciences, 704, 1993, pp. 103 -- 125. [7] L. Ganesan and N. Vanaja, Modules for which every submodule has a unique coclosure, Comm. Algebra 30 (2002), 2355 -- 2377. [8] J.L. Garc´ıa Hern´andez and J.L. G´omez Pardo, On endomorphism rings of quasiprojective modules, Math. Z. 196 (1997), 87 -- 108. [9] P. Grzeszczuk and E.R. Puczy lowski, On Goldie and dual Goldie dimension, J. Pure Appl. Algebra 31 (1984), 47 -- 54. 10 S. CRIVEI, H. INANKIL, M.T. KOS¸AN, AND G. OLTEANU [10] D. Keskin, On lifting modules, Comm. Algebra 28 (2000), 3427 -- 3440. [11] R. Wisbauer, Foundations of ring and module theory, Gordon and Breach, 1991. [12] J.M. Zelmanowitz, Correspondences of closed submodules, Proc. Amer. Math. Soc. 124 (1996), 2955 -- 2960. Faculty of Mathematics and Computer Science,"Babes¸-Bolyai" University, Str. Mihail Kogalni- ceanu 1, 400084 Cluj-Napoca, Romania E-mail address: [email protected] Department of Mathematics, Gebze Institute of Technology, C¸ ayirova Campus, 41400 Gebze- Kocaeli, Turkey E-mail address: [email protected] Department of Mathematics, Gebze Institute of Technology, C¸ ayirova Campus, 41400 Gebze- Kocaeli, Turkey E-mail address: [email protected] Department of Statistics-Forecasts-Mathematics, "Babes¸-Bolyai" University, Str. T. Mihali 58-60, 400591 Cluj-Napoca, Romania E-mail address: [email protected]
1003.2192	3	1003	2011-05-17T18:27:33	The arity gap of order-preserving functions and extensions of pseudo-Boolean functions	[ "math.RA", "math.CO" ]	The aim of this paper is to classify order-preserving functions according to their arity gap. Noteworthy examples of order-preserving functions are so-called aggregation functions. We first explicitly classify the Lov\'asz extensions of pseudo-Boolean functions according to their arity gap. Then we consider the class of order-preserving functions between partially ordered sets, and establish a similar explicit classification for this function class.	math.RA	math	THE ARITY GAP OF ORDER-PRESERVING FUNCTIONS AND EXTENSIONS OF PSEUDO-BOOLEAN FUNCTIONS MIGUEL COUCEIRO, ERKKO LEHTONEN, AND TAM ´AS WALDHAUSER Abstract. The aim of this paper is to classify order-preserving functions ac- cording to their arity gap. Noteworthy examples of order-preserving functions are so-called aggregation functions. We first explicitly classify the Lov´asz ex- tensions of pseudo-Boolean functions according to their arity gap. Then we consider the class of order-preserving functions between partially ordered sets, and establish a similar explicit classification for this function class. ] . A R h t a m [ 3 v 2 9 1 2 . 3 0 0 1 : v i X r a 1. Introduction In this paper, we study the arity gap of functions of several variables. Essentially, the arity gap of a function f : An → B (n ≥ 2) that depends on all of its variables can be defined as the minimum decrease in the number of essential variables when variables of f are identified. Salomaa [18] showed that the arity gap of any Boolean function is at most 2. This result was extended to functions defined on arbitrary finite domains by Willard [21], who showed that the same upper bound holds for the arity gap of any function f : An → B, provided that n > max(A, 3). In fact, he showed that if the arity gap of such a function f equals 2, then f is totally symmetric. This line of research culminated into a complete classification of functions f : An → B according to their arity gap (see Theorem 2.5), originally presented in [4] in the setting of functions with finite domains; in [6] it was observed that this result holds for functions with arbitrary, possibly infinite domains. Salomaa's [18] result on the upper bound for the arity gap of Boolean func- tions mentioned above was strengthened in [3], where Boolean functions were com- pletely classified according to their arity gap. Using tools provided by Berman and Kisielewicz [1] and Willard [21], in [4] a similar explicit classification was estab- lished for all pseudo-Boolean functions, i.e., functions f : {0, 1}n → R. As it turns out, this leads to analogous classifications of wider classes of functions. In [5], this result on pseudo-Boolean functions was the key step in showing that among lat- tice polynomial functions only truncated ternary medians have arity gap 2; all the others have arity gap 1. Similar techniques are used in Section 3 to derive explicit descriptions of the arity gap of well-known extensions of pseudo-Boolean functions to the whole real line, namely, Owen and Lov´asz extensions. In Section 4 we consider the arity gap of order-preserving functions. To this extent, we present a complete classification of functions over arbitrary domains according to their arity gap (originally established in [4] for functions over finite domains), which is then used to derive a dichotomy theorem based on the arity gap (and the so-called quasi-arity), and to explicitly determine those order-preserving functions that have arity gap 1 and those that have arity gap 2. 1 2 MIGUEL COUCEIRO, ERKKO LEHTONEN, AND TAM ´AS WALDHAUSER Aggregation functions became a widely studied class of order-preserving func- tions. Thus, as a by-product of our general results, we obtain an explicit classifica- tion of these functions according to their arity gap, which we present in the end of Section 4. 2. Preliminaries: arity gap and the simple minor relation Throughout this paper, let A and B be arbitrary sets with at least two elements. A B-valued function (of several variables) on A is a mapping f : An → B for some positive integer n, called the arity of f . The A-valued functions on A are called operations on A. Operations on {0, 1} are called Boolean functions. We denote the set of real numbers by R. Functions f : {0, 1}n → R are referred to as pseudo- Boolean functions. For a natural number n ≥ 1, we denote [n] = {1, . . . , n}. The i-th variable is said to be essential in f : An → B, or f is said to depend on xi, if there is a pair ((a1, . . . , ai−1, ai, ai+1, . . . , an), (a1, . . . , ai−1, bi, ai+1, . . . , an)) ∈ An × An, called a witness of essentiality of xi in f , such that f (a1, . . . , ai−1, ai, ai+1, . . . , an) 6= f (a1, . . . , ai−1, bi, ai+1, . . . , an). The number of essential variables in f is called the essential arity of f , and it is denoted by ess f . If ess f = m, we say that f is essentially m-ary. For n ≥ 2, define An We also define A1 Consider f : An → B. Any function g : An → B satisfying f An = := {(a1, . . . , an) ∈ An : ai = aj for some i 6= j}. = := A. Note that if A has less than n elements, then An = = gAn = = An. = is called a support of f . The quasi-arity of f , denoted qa f , is defined as the minimum of the essential arities of the supports of f , i.e., qa f = ming ess g, where g ranges over the set of all supports of f . If qa f = m, we say that f is quasi-m-ary. A function f : An → B is said to be obtained from g : Am → B by simple variable substitution, or f is a simple minor of g, if there is a mapping σ : {1, . . . , m} → {1, . . . , n} such that f (x1, . . . , xn) = g(xσ(1), . . . , xσ(m)) for all (x1, . . . , xn) ∈ An. The simple minor relation constitutes a quasi-order ≤ on the set of all B-valued functions of several variables on A which is given by the following rule: f ≤ g if and only if f is obtained from g by simple variable substitution. If f ≤ g and g ≤ f , we say that f and g are equivalent, denoted f ≡ g. If f ≤ g but g 6≤ f , we denote f < g. It can be easily observed that if f ≤ g then ess f ≤ ess g, with equality if and only if f ≡ g. For background, extensions and variants of the simple minor relation, see, e.g., [2, 7, 8, 9, 12, 13, 17, 20, 22]. For f : An → B, i, j ∈ {1, . . . , n}, i 6= j, we define fi←j : An → B to be the simple minor of f given by the substitution of xj for xi, that is, fi←j (x1, . . . , xn) = f (x1, . . . , xi−1, xj, xi+1, . . . , xn). Note that on the right-hand side of the above equality, xj occurs twice, namely both at the i-th and the j-th positions. We denote ess< f = max g<f ess g, THE ARITY GAP OF ORDER-PRESERVING FUNCTIONS 3 and we define the arity gap of f by gap f = ess f − ess< f . It is easily observed that gap f = min i6=j (ess f − ess fi←j ), where i and j range over the set of indices of essential variables of f . In the sequel, whenever we consider the arity gap of some function f , we will assume that all variables of f are essential. This is not a significant restriction, because every nonconstant function is equivalent to a function with no inessential variables and equivalent functions have the same arity gap. Salomaa [18] proved that the arity gap of every Boolean function with at least two essential variables is at most 2. This result was generalized by Willard [21, Lemma 1.2] in the following theorem. Theorem 2.1. Let A be a finite set. Suppose f : An → B depends on all of its variables. If n > max(A, 3), then gap f ≤ 2. In [3], Salomaa's result was strengthened into an explicit classification of Boolean functions in terms of arity gap: Theorem 2.2. Assume that f : {0, 1}n → {0, 1} depends on all of its variables. We have gap f = 2 if and only if f is equivalent to one of the following Boolean functions: • x1 ⊕ x2 ⊕ · · · ⊕ xn ⊕ c, • x1x2 ⊕ x1 ⊕ c, • x1x2 ⊕ x1x3 ⊕ x2x3 ⊕ c, • x1x2 ⊕ x1x3 ⊕ x2x3 ⊕ x1 ⊕ x2 ⊕ c, where ⊕ denotes addition modulo 2 and c ∈ {0, 1}. Otherwise gap f = 1. Based on this, a complete classification of pseudo-Boolean functions according to their arity gap was presented in [4]: Theorem 2.3. For a pseudo-Boolean function f : {0, 1}n → R which depends on all of its variables, gap f = 2 if and only if f satisfies one of the following conditions: • n = 2 and f is a nonconstant function satisfying f (0, 0) = f (1, 1), • f = g ◦ h, where g : {0, 1} → R is injective and h : {0, 1}n → {0, 1} is a Boolean function with gap h = 2, as listed in Theorem 2.2. Otherwise gap f = 1. Remark 2.4. It is noteworthy that there is a complete one-to-one correspondence between pseudo-Boolean functions and set functions, i.e., functions v : 2[n] → R for some n ≥ 1. This correspondence is based on the natural order-isomorphism between {0, 1}n and the power set 2[n] of [n]. For a pseudo-Boolean function f : {0, 1}n → R we can associate a set function vf : 2[n] → R given by vf (T ) = f (eT ), where eT denotes the characteristic vector of T ⊆ [n]. Conversely, for a set function v : 2[n] → R, let fv : {0, 1}n → R be the pseudo-Boolean function defined by fv(eT ) = v(T ). Clearly, fvf = f and vfv = v for every pseudo-Boolean function f : {0, 1}n → R and every set function v : 2[n] → R. The study of the arity gap of functions An → B culminated into the character- ization presented in Theorem 2.5, originally proved in [4]. We need to introduce some terminology to state the result. Let 2A be the power set of A, and define oddsupp : Sn≥1 An → 2A by oddsupp(a1, . . . , an) = (cid:8)ai : {j ∈ [n] : aj = ai} is odd(cid:9). 4 MIGUEL COUCEIRO, ERKKO LEHTONEN, AND TAM ´AS WALDHAUSER A partial function f : S → B, S ⊆ An, is said to be determined by oddsupp if f = f ∗ ◦ oddsuppS for some function f ∗ : 2A → B. Theorem 2.5. Suppose that f : An → B, n ≥ 2, depends on all of its variables. (i) For 3 ≤ p ≤ n, gap f = p if and only if qa f = n − p. (ii) For n 6= 3, gap f = 2 if and only if qa f = n − 2 or qa f = n and f An = is determined by oddsupp. (iii) For n = 3, gap f = 2 if and only if there is a nonconstant unary function h : A → B and i1, i2, i3 ∈ {0, 1} such that f (x1, x0, x0) = h(xi1 ), f (x0, x1, x0) = h(xi2 ), f (x0, x0, x1) = h(xi3 ). (iv) Otherwise gap f = 1. Remark 2.6. The notion of a function's being determined by oddsupp is due to Berman and Kisielewicz [1]. Willard [21] showed that if f : An → B where A is finite, n > max(A, 3) and gap f = 2, then f is determined by oddsupp. Remark 2.7. While Theorem 2.5 was originally stated and proved in the setting of functions with finite domains, its proof presented in [4] does not make use of any assumption on the cardinalities of the domain and codomain -- as long as they contain at least two elements. Hence the theorem immediately generalizes for functions with arbitrary domains. 3. The arity gap of Lov´asz and Owen extensions In this section, we consider well-known extensions of pseudo-Boolean functions and generalize Theorem 2.3 accordingly. For further background on pseudo-Boolean functions, we refer the reader to Hammer and Rudeanu [11]. As it is well-known, every pseudo-Boolean function can be uniquely represented by a multilinear polynomial expression. A common way to construct such repre- sentations makes use of the notion of "Mobius transform". Let v : 2[n] → R be a set function. The Mobius transform (or Mobius inverse) of v is the map mv : 2[n] → R given by mv(S) = X (−1)S−T v(T ), for all S ⊆ [n]. T ⊆S In view of Remark 2.4, we say that m : 2[n] → R is the Mobius transform of f : {0, 1}n → R if m = mvf . Theorem 3.1 ([11]). Let f : {0, 1}n → R be a pseudo-Boolean function. Then (1) f (x) = X mvf (S) Y xi, for all x ∈ {0, 1}n. S⊆[n] i∈S Remark 3.2. Theorem 3.1 motivates the terminology "Mobius inverse of v" since it implies in particular that for every S ⊆ [n], v(S) = P mv(T ). T ⊆S The following result is well known and easy to verify (see, e.g., [15] for the case of order-preserving pseudo-Boolean functions). THE ARITY GAP OF ORDER-PRESERVING FUNCTIONS 5 Lemma 3.3. Let f : {0, 1}n → R be a pseudo-Boolean function and consider its corresponding set function vf . If xi is inessential in f , then mvf (S) = 0 whenever i ∈ S. In particular, f depends on xi if and only if xi appears in the multilinear polynomial representation (1) of f . There are several ways of extending a pseudo-Boolean function f : {0, 1}n → R to a function on R. Perhaps the most natural is the multilinear polynomial extension. The Owen extension [16] (or multilinear extension) of a pseudo-Boolean function f : {0, 1}n → R is the mapping Pf : Rn → R defined by Pf (x) = X mvf (S) Y xi, for all x ∈ Rn. S⊆[n] i∈S Clearly, f coincides with the restriction of Pf to {0, 1}n. Another extension of pseudo-Boolean functions to functions on R is the so-called "Lov´asz extension". This terminology is due to Singer [19] who refined a result by Lov´asz [14] concerning convex functions. The Lov´asz extension of a pseudo-Boolean function f : {0, 1}n → R is the mapping Ff : Rn → R defined by Ff (x) = X mvf (S) ^ xi, for all x ∈ Rn. S⊆[n] i∈S Observe that the Lov´asz extension of a pseudo-Boolean function f is the unique extension of f which is linear on the "standard simplices" Rn σ = {x ∈ Rn : xσ(1) ≤ xσ(2) ≤ · · · ≤ xσ(n)}, for any permutation σ on [n] (see [10]). Remark 3.4. The defining expressions of Owen and Lov´asz extensions differ only in the fact that the connecting operations between variables are the product and the minimum, respectively. In the sequel, this observation can be used to translate the results concerning Lov´asz extensions into analogous results about Owen extensions. Remark 3.5. Every function F : Rn → R of the form (2) F (x) = X m(S) ^ xi, S⊆[n] i∈S where m : 2[n] → R is the Lov´asz extension of a unique pseudo-Boolean function, namely, f = F {0,1}n . Therefore, we shall refer to any map of the form (2) as a Lov´asz extension. Theorem 3.6. Let f : {0, 1}n → R be a pseudo-Boolean function. Then the i-th variable is essential in f if and only if the i-th variable is essential in Ff . Proof. As observed, f coincides with Ff on {0, 1}n, and thus if the i-th variable is inessential in Ff , then the i-th variable is inessential in f . Conversely, if the i-th variable is inessential in f , then by Lemma 3.3 it follows that xi does not appear in the defining expression of Ff . Hence, the i-th variable is inessential in Ff . (cid:3) Corollary 3.7. Let f : {0, 1}n → R be a pseudo-Boolean function. Then gap f = gap Ff . In particular, gap Ff ≤ 2. Using Theorems 2.2 and 2.3, we obtain the following explicit descriptions of those Lov´asz extensions that have arity gap 2. 6 MIGUEL COUCEIRO, ERKKO LEHTONEN, AND TAM ´AS WALDHAUSER Theorem 3.8. Assume that F : Rn → R is a Lov´asz extension that depends on all of its variables. Then gap F = 2 if and only if F is of one of the following forms: (i) F ≡ a − b 2 X S⊆[n] (cid:0)(−2)S · ^ xi(cid:1), i∈S (ii) F ≡ a + (b − a)x1 + (a − b)(x1 ∧ x2), (iii) F ≡ a + (b − a)(cid:0)(x1 ∧ x2) + (x1 ∧ x3) + (x2 ∧ x3)(cid:1) + 2(a − b)(x1 ∧ x2 ∧ x3), (iv) F ≡ a + (b − a)(x1 + x2) + (a − b)(cid:0)(x1 ∧ x2) + (x1 ∧ x3) + (x2 ∧ x3)(cid:1) + 2(b − a)(x1 ∧ x2 ∧ x3), (v) F ≡ a + (b − a)x1 + (c − a)x2 + (2a − b − c)(x1 ∧ x2), for some a, b, c ∈ R. Otherwise gap F = 1. Note that since F is assumed to depend on all of its variables, for functions of the form (i) -- (iv) it holds that a 6= b, and for functions of the form (v) it holds that {a, b, c} 6= {a}. Proof. Let f : {0, 1}n → R be the pseudo-Boolean function determined by F . By Theorems 2.2 and 2.3, gap f = 2 if and only if (i) f ≡ (b − a)(x1 ⊕ · · · ⊕ xn) + a, (ii) f ≡ (b − a)(x1x2 ⊕ x1) + a, (iii) f ≡ (b − a)(x1x2 ⊕ x1x3 ⊕ x2x3) + a, (iv) f ≡ (b − a)(x1x2 ⊕ x1x3 ⊕ x2x3 ⊕ x1 ⊕ x2) + a, or (v) f : {0, 1}2 → R is nonconstant such that f (0, 0) = f (1, 1), say, f (0, 0) = f (1, 1) = a, f (1, 0) = b and f (0, 1) = c, where ⊕ denotes addition modulo 2, and a, b, c ∈ R. The theorem now follows by computing the Mobius transform of vf in each possible case. (cid:3) Corollary 3.9. A nondecreasing Lov´asz extension F : Rn → R has arity gap 2 if and only if (3) F ≡ a + (b − a)(cid:0)(x1 ∧ x2) + (x1 ∧ x3) + (x2 ∧ x3)(cid:1) + 2(a − b)(x1 ∧ x2 ∧ x3). Otherwise gap F = 1. Techniques similar to those developed in this section were successfully used in [5] to classify the class of lattice polynomial functions, i.e., functions which can be obtained as compositions of the lattice operations and variables (projections) and constants. A well-known example of a lattice polynomial function on a distributive lattice A is the median function med : A3 → A given by med(x1, x2, x3) = (x1 ∧ x2) ∨ (x1 ∧ x3) ∨ (x2 ∧ x3) = (x1 ∨ x2) ∧ (x1 ∨ x3) ∧ (x2 ∨ x3). As shown in [5], lattice polynomial functions with arity gap 2 are exactly the trun- cated median functions. Theorem 3.10 ([5]). Let f : An → A be a lattice polynomial function on a bounded distributive lattice A. Then gap f = 2 if and only if for some a, b ∈ A, a < b. Otherwise gap f = 1. f ≡ (a ∨ med(x1, x2, x3)) ∧ b, In the next section, we extend these results to the more general class of order- preserving maps between possibly different ordered sets A and B. THE ARITY GAP OF ORDER-PRESERVING FUNCTIONS 7 4. The arity gap of order-preserving functions Let (A; ≤) be a partially ordered set. We say that (A; ≤) is • upwards directed if every pair of elements of A has an upper bound, • downwards directed if every pair of elements of A has a lower bound, • bidirected if (A; ≤) is both upwards directed and downwards directed, • pseudo-directed if every pair of elements of A has an upper bound or a lower bound. Remark 4.1. In the above definitions, existence of a least upper bound or a greatest lower bound is not stipulated. Therefore, an upwards (or downwards) directed poset is not the same thing as a semilattice, nor is a bidirected poset the same thing as a lattice. However, every semilattice is either upwards or downwards directed, and every lattice and every bounded poset is bidirected. Moreover, every upwards directed or downwards directed poset is pseudo-directed. Let (A; ≤A) and (B; ≤B) be partially ordered sets. A function f : An → B is said to be order-preserving (with respect to the partial orders ≤A and ≤B) if for all a, b ∈ An, f (a) ≤B f (b) whenever a ≤A b, where a ≤A b denotes the componentwise ordering of tuples, i.e., a ≤A b if and only if ai ≤A bi for all i ∈ {1, . . . , n}. Lemma 4.2. Let (A; ≤A) be a pseudo-directed poset, and let f : An → B a function. If xi is essential in f then there are elements a1, . . . , an, bi ∈ A such that ai <A bi and f (a1, . . . , ai−1, ai, ai+1, . . . , an) 6= f (a1, . . . , ai−1, bi, ai+1, . . . , an). Moreover, if B is partially ordered by ≤B and f is order-preserving with respect to ≤A and ≤B, then f (a1, . . . , ai−1, ai, ai+1, . . . , an) <B f (a1, . . . , ai−1, bi, ai+1, . . . , an). Proof. Since xi is essential in f , there exist elements a1, . . . , ai−1, a′, b′, ai+1, . . . , an in A such that f (a1, . . . , ai−1, a′, ai+1, . . . , an) 6= f (a1, . . . , ai−1, b′, ai+1, . . . , an). By the assumption that (A; ≤) is pseudo-directed, a′ and b′ have an upper bound or a lower bound. Assume first that a′ and b′ have an upper bound c. We clearly have that (4) (5) f (a1, . . . , ai−1, a′, ai+1, . . . , an) 6= f (a1, . . . , ai−1, c, ai+1, . . . , an) f (a1, . . . , ai−1, b′, ai+1, . . . , an) 6= f (a1, . . . , ai−1, c, ai+1, . . . , an). or The claim thus follows by choosing bi := c and ai := a′ if (4) holds or ai := b′ if (5) holds. Otherwise a′ and b′ have a lower bound, and a similar argument shows that the claim holds also in this case. If f is order-preserving with respect to ≤A and ≤B, then we have in fact that f (a1, . . . , ai−1, ai, ai+1, . . . , an) <B f (a1, . . . , ai−1, bi, ai+1, . . . , an). (cid:3) Lemma 4.3. Let (A; ≤A) be a bidirected poset, let (B; ≤B) be any poset, and let f : An → B (n ≥ 2) be an order-preserving function that depends on all of its vari- ables. Then, for all i, j ∈ {1, . . . , n} (i 6= j), xj is essential in fi←j. Furthermore, 8 MIGUEL COUCEIRO, ERKKO LEHTONEN, AND TAM ´AS WALDHAUSER if i < j, then there exist elements c, d, a1, . . . , an ∈ A such that c <A d and (6) f (a1, . . . , ai−1, c, ai+1, . . . , aj−1, c, aj+1, . . . , an) <B f (a1, . . . , ai−1, d, ai+1, . . . , aj−1, d, aj+1, . . . , an). Proof. Assume, without loss of generality, that i = 1, j = 2. Since x1 is essential in f , by Lemma 4.2 there exist elements a1, . . . , an, b1 ∈ A such that a1 <A b1 and f (a1, a2, . . . , an) <B f (b1, a2, . . . , an). By the assumption that (A; ≤) is bidirected, there exist a lower bound c of a1 and a2 and an upper bound d of b1 and a2. Again, by the monotonicity of f , f1←2(a1, c, a3, . . . , an) = f (c, c, a3, . . . , an) ≤B f (a1, a2, a3, . . . , an) <B f (b1, a2, a3, . . . , an) ≤B f (d, d, a3, . . . , an) = f1←2(a1, d, a3, . . . , an), which shows that x2 is essential in f1←2 and inequality (6) holds. (cid:3) Proposition 4.4. Let (A; ≤A) be a bidirected poset, let (B; ≤B) be any poset, and let f : An → B (n ≥ 2) be an order-preserving function that depends on all of its variables. Then qa f ≥ n − 1 and f An = is not determined by oddsupp. Proof. Suppose first, on the contrary, that qa f = n − p for some p ≥ 2. Let g be a support of f with essential arity n− p. Then g has at least two inessential variables, say xi and xj, and these variables are clearly inessential in gi←j as well. But, since fi←j = gi←j, this constitutes a contradiction to Lemma 4.3 which asserts that xj is essential in fi←j. Suppose then, on the contrary, that f An = is determined by oddsupp. Then = = f ∗ ◦ oddsupp for some f ∗ : 2A → B. We clearly have that for all c, f An d, a3, . . . , an ∈ A, oddsupp(c, c, a3, . . . , an) = oddsupp(d, d, a3, . . . , an) (note that (c, c, a3, . . . , an), (d, d, a3, . . . , an) ∈ An =); hence f (c, c, a3, . . . , an) = f (d, d, a3, . . . , an). This contradicts Lemma 4.3. (cid:3) Proposition 4.5. Let (A; ≤A) be a bidirected poset, let (B; ≤B) be any poset, and let f : A3 → B be an order-preserving function that depends on all of its variables. Then gap f = 2 if and only if there is a nonconstant order-preserving unary function h : A → B such that f (x1, x0, x0) = f (x0, x1, x0) = f (x0, x0, x1) = h(x0). Proof. By Theorem 2.5, the condition is sufficient. For necessity, assume that gap f = 2. Then, by Theorem 2.5, there is a nonconstant unary function h : A → B and i1, i2, i3 ∈ {0, 1} such that f (x1, x0, x0) = h(xi1 ), f (x0, x1, x0) = h(xi2 ), f (x0, x0, x1) = h(xi3 ). We claim that i1 = i2 = i3 = 0. Suppose, on the contrary, that i1 = 1. By Lemma 4.3, there exist elements a, b, c ∈ A such that b <A c and f (a, b, b) <B f (a, c, c), but this is a contradiction to f (a, b, b) = h(a) = f (a, c, c). Similarly, we can derive a contradiction from the assumption that i2 = 1 or i3 = 1. The monotonicity of h follows from the monotonicity of f . For, if a ≤A b, then h(a) = f (a, a, a) ≤B f (b, b, b) = h(b). (cid:3) Theorem 4.6. Let (A; ≤A) be a bidirected poset, let (B; ≤B) be any poset, and let f : An → B (n ≥ 2) be an order-preserving function that depends on all of THE ARITY GAP OF ORDER-PRESERVING FUNCTIONS 9 its variables. Then gap f = 2 if and only if n = 3 and there is a nonconstant order-preserving unary function h : A → B such that f (x1, x0, x0) = f (x0, x1, x0) = f (x0, x0, x1) = h(x0). Otherwise gap f = 1. Proof. Immediate consequence of Theorem 2.5 and Propositions 4.4 and 4.5. (cid:3) By imposing stronger assumptions on the underlying posets, we obtain more stringent descriptions of order-preserving functions with arity gap 2. Lemma 4.7. Let (A; ≤A) and (B; ≤B) be lattices, and let h : A → B be a lattice homomorphism. Let f : A3 → B be an order-preserving function such that f (x1, x0, x0) = f (x0, x1, x0) = f (x0, x0, x1) = h(x0). If the homomorphic image of (A; ≤A) by h is a distributive sublattice of (B; ≤B), then f = med(cid:0)h(x1), h(x2), h(x3)(cid:1), where med denotes the ternary median function on Im h. Proof. By the monotonicity of f and the assumption that A is a lattice, we have that for all a1, a2, a3 ∈ A, h(a1 ∧ a2) = f (a1 ∧ a2, a1 ∧ a2, a3) ≤ f (a1, a2, a3) ≤ f (a1 ∨ a2, a1 ∨ a2, a3) = h(a1 ∨ a2). A similar argument shows that for all i, j ∈ {1, 2, 3}, we have h(ai ∧ aj) ≤ f (a1, a2, a3) ≤ h(ai ∨ aj). By the assumption that B is a lattice, it follows from the above inequalities that h(a1 ∧ a2) ∨ h(a2 ∧ a3) ∨ h(a1 ∧ a3) ≤ f (a1, a2, a3) ≤ h(a1 ∨ a2) ∧ h(a2 ∨ a3) ∧ h(a1 ∨ a3). Since h is a lattice homomorphism, we have that (7) (8) h(a1 ∧ a2) ∨ h(a2 ∧ a3) ∨ h(a1 ∧ a3) = (cid:0)h(a1) ∧ h(a2)(cid:1) ∨ (cid:0)h(a2) ∧ h(a3)(cid:1) ∨ (cid:0)h(a1) ∧ h(a3)(cid:1), h(a1 ∨ a2) ∧ h(a2 ∨ a3) ∧ h(a1 ∨ a3) = (cid:0)h(a1) ∨ h(a2)(cid:1) ∧ (cid:0)h(a2) ∨ h(a3)(cid:1) ∧ (cid:0)h(a1) ∨ h(a3)(cid:1). By the assumption that Im h is a distributive sublattice of B, the right-hand sides of (7) and (8) are equal, and they are actually equal to med(cid:0)h(a1), h(a2), h(a3)(cid:1). We conclude that f (a1, a2, a3) = med(cid:0)h(a1), h(a2), h(a3)(cid:1). (cid:3) Corollary 4.8. Let (A; ≤A) be a chain and let (B; ≤B) be any lattice. Let f : An → B be an order-preserving function. Then gap f = 2 if and only if n = 3 and f = med(cid:0)h(x1), h(x2), h(x3)(cid:1) for some nonconstant order-preserving unary func- tion h : A → B (here med denotes the median function on Im h). Otherwise gap f = 1. 10 MIGUEL COUCEIRO, ERKKO LEHTONEN, AND TAM ´AS WALDHAUSER Proof. If f = med(cid:0)h(x1), h(x2), h(x3)(cid:1), where h is as described in the statement, then clearly gap f = 2. For the converse implication, assume that gap f = 2. By Theorem 4.6, n = 3 and there is a nonconstant order-preserving unary function h : A → B such that f (x1, x0, x0) = f (x0, x1, x0) = f (x0, x0, x1) = h(x0). Since every order-preserving function h is a lattice homomorphism from a chain A to any lattice B and the homomorphic image of A by h is a chain and hence a distribu- tive sublattice of B, it follows from Lemma 4.7 that f = med(h(x1), h(x2), h(x3)). (cid:3) The last claim follows from Theorem 4.6, which asserts that gap f ≤ 2. To illustrate the use of the results obtained in this section, we present an alter- native proof of Theorem 3.10. Proof of Theorem 3.10. It is well-known that lattice polynomial functions are order- preserving. Therefore Theorem 4.6 applies, and gap f ≤ 2. Assume, without loss of generality, that ess f = n. Suppose that gap f = 2. Then, by Theorem 4.6, n = 3 and there is a nonconstant order-preserving unary function h : A → A such that f (x1, x0, x0) = f (x0, x1, x0) = f (x0, x0, x1) = h(x0). Since f is a polynomial function, h is a polynomial function as well, and hence h(x) = (a ∨ x) ∧ b for some a, b ∈ A, a < b. In particular, h is a lattice homomor- phism. Since A is a distributive lattice, Im h is a distributive sublattice of A, and Lemma 4.7 then implies that f = med(cid:0)h(x1), h(x2), h(x3)(cid:1) = h(cid:0)med(x1, x2, x3)(cid:1). Clearly, if f has the above form, then gap f = 2. Since gap f ≤ 2, the last claim of the theorem follows. (cid:3) As mentioned, the class of order-preserving functions includes the noteworthy class of aggregation functions. Traditionally, an aggregation function on a closed real interval [a, b] ⊆ R is defined as a mapping M : [a, b]n → [a, b] which is nonde- creasing and fulfills the boundary conditions M (a, . . . , a) = a and M (b, . . . , b) = b. From Corollary 4.8, we obtain the following. Corollary 4.9. Let M : [a, b]n → [a, b] be an aggregation function on a real interval [a, b]. Then gap M = 2 if and only if n = 3 and M = med(cid:0)h(x1), h(x2), h(x3)(cid:1) for some nonconstant order-preserving unary function h : [a, b] → [a, b] satisfying h(a) = a, h(b) = b. Otherwise gap f = 1. Acknowledgements The third author acknowledges that the present project is supported by the National Research Fund, Luxembourg, and cofunded under the Marie Curie Actions of the European Commission (FP7-COFUND), and supported by the Hungarian National Foundation for Scientific Research under grant no. K77409. THE ARITY GAP OF ORDER-PRESERVING FUNCTIONS 11 References [1] J. Berman, A. Kisielewicz, On the number of operations in a clone, Proc. Amer. Math. Soc. 122 (1994) 359 -- 369. [2] M. Couceiro, On the lattice of equational classes of Boolean functions and its closed intervals, J. Mult.-Valued Logic Soft Comput. 18 (2008) 81 -- 104. [3] M. Couceiro, E. Lehtonen, On the effect of variable identification on the essential arity of functions on finite sets, Int. J. Found. Comput. Sci. 18 (2007) 975 -- 986. [4] M. Couceiro, E. Lehtonen, Generalizations of ´Swierczkowski's lemma and the arity gap of finite functions, Discrete Math. 309 (2009) 5905 -- 5912. [5] M. Couceiro, E. Lehtonen, The arity gap of polynomial functions over bounded distributive lattices, 40th IEEE International Symposium on Multiple-Valued Logic (ISMVL 2010), IEEE Computer Society, Los Alamitos, 2010, pp. 113 -- 116. [6] M. Couceiro, E. Lehtonen, T. Waldhauser, Decompositions of functions based on arity gap, arXiv:1003.1294. [7] M. Couceiro, M. Pouzet, On a quasi-ordering on Boolean functions, Theoret. Comput. Sci. 396 (2008) 71 -- 87. [8] O. Ekin, S. Foldes, P. L. Hammer, L. Hellerstein, Equational characterizations of Bool- ean function classes, Discrete Math. 211 (2000) 27 -- 51. [9] A. Feigelson, L. Hellerstein, The forbidden projections of unate functions, Discrete Appl. Math. 77 (1997) 221 -- 236. [10] M. Grabisch, J.-L. Marichal, R. Mesiar, E. Pap, Aggregation Functions, Encyclopedia of Mathematics and Its Applications, vol. 127, Cambridge University Press, Cambridge, 2009. [11] P. L. Hammer, S. Rudeanu, Boolean Methods in Operations Research and Related Areas, Springer-Verlag, Berlin, 1968. [12] E. Lehtonen, Descending chains and antichains of the unary, linear, and monotone subfunc- tion relations, Order 23 (2006) 129 -- 142. [13] E. Lehtonen, ´A. Szendrei, Equivalence of operations with respect to discriminator clones, Discrete Math. 309 (2009) 673 -- 685. [14] L. Lov´asz, Submodular function and convexity. In: A. Bachem, M. Grotschel, B. Korte (eds.), Mathematical programming. The state of the art. Bonn 1982, Springer-Verlag, Berlin -- Heidelberg -- New York -- Tokyo, 1983, pp. 235 -- 257. [15] T. Murofushi, M. Sugeno, A theory of fuzzy measures: representations, the Choquet inte- gral, and null sets, J. Math. Anal. Appl. 159 (1991) 532 -- 549. [16] G. Owen, Multilinear extensions of games, Management Science 18 (1972) 64 -- 79. [17] N. Pippenger, Galois theory for minors of finite functions, Discrete Math. 254 (2002) 405 -- 419. [18] A. Salomaa, On essential variables of functions, especially in the algebra of logic, Ann. Acad. Sci. Fenn. Ser. A I. Math. 339 (1963) 3 -- 11. [19] I. Singer, Extensions of functions of 0 -- 1 variables and applications to combinatorial opti- mization, Numer. Funct. Anal. Optim. 7 (1985) 23 -- 62. [20] C. Wang, Boolean minors, Discrete Math. 141 (1991) 237 -- 258. [21] R. Willard, Essential arities of term operations in finite algebras, Discrete Math. 149 (1996) 239 -- 259. [22] I. E. Zverovich, Characterizations of closed classes of Boolean functions in terms of forbidden subfunctions and Post classes, Discrete Appl. Math. 149 (2005) 200 -- 218. (M. Couceiro) Mathematics Research Unit, University of Luxembourg, 6, rue Richard Coudenhove-Kalergi, L -- 1359 Luxembourg, Luxembourg E-mail address: [email protected] (E. Lehtonen) Computer Science and Communications Research Unit, University of Luxembourg, 6, rue Richard Coudenhove-Kalergi, L -- 1359 Luxembourg, Luxembourg E-mail address: [email protected] (T. Waldhauser) Mathematics Research Unit, University of Luxembourg, 6, rue Rich- ard Coudenhove-Kalergi, L -- 1359 Luxembourg, Luxembourg and Bolyai Institute, Uni- versity of Szeged, Aradi v´ertan´uk tere 1, H -- 6720 Szeged, Hungary E-mail address: [email protected]
1711.06670	3	1711	2019-07-30T03:56:45	Frobenius-Perron Theory of Endofunctors	[ "math.RA", "math.QA" ]	We introduce the Frobenius-Perron dimension of an endofunctor of a k-linear category and provide some applications.	math.RA	math	FROBENIUS-PERRON THEORY OF ENDOFUNCTORS J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Abstract. We introduce the Frobenius-Perron dimension of an endofunctor of a k-linear category and provide some applications. 0. Introduction The spectral radius (also called the Frobenius-Perron dimension) of a matrix is an elementary and extremely useful invariant in linear algebra, combinatorics, topology, probability and statistics. The Frobenius-Perron dimension has become a crucial concept in the study of fusion categories and representations of semismi- ple weak Hopf algebras since it was introduced by Etingof-Nikshych-Ostrik [ENO] in early 2000 (also see [EG, EGO, Ni]). In this paper several Frobenius-Perron type invariants are proposed to study derived categories, representations of finite dimensional algebras, and complexity of algebras and categories. Throughout let k be an algebraically closed field, and let everything be over k. 0.1. Definitions. The first goal is to introduce the Frobenius-Perron dimension of an endofunctor of a category. If an object V in a fusion category C is considered as the associated tensor endofunctor V ⊗C −, then our definition of the Frobenius- Perron dimension agrees with the definition given in [ENO], see [Example 2.11] for details. Our definition applies to the derived category of projective schemes and finite dimensional algebras, as well as other abelian and additive categories [Definitions 2.3 and 2.4]. We refer the reader to Section 2 for the following invariants of an endofunctor: Frobenius-Perron dimension (denoted by fpd, and fpdn for n ≥ 1), Frobenius-Perron growth (denoted by fpg), Frobenius-Perron curvature (denoted by fpv), and Frobenius-Perron series (denoted by FP). One can further define the above invariants for an abelian or a triangulated category. Note that the Frobenius-Perron dimension/growth/curvature of a category can be a non-integer, see Proposition 5.12(1), Example 8.7, and Remark 5.13(5) for non- integral values of fpd, fpg, and fpv respectively. If A is an abelian category, let Db(A) denote the bounded derived category of A. On the one hand it is reasonable to call fpd a dimension function since fpd(Db(Mod −k[x1,··· , xn])) = n 2000 Mathematics Subject Classification. Primary 18E30, 16G60, 16E10, Secondary 16E35. Key words and phrases. Frobenius-Perron dimension, derived categories, embedding of cate- gories, tame and wild dichotomy, complexity. 1 2 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU [Proposition 4.3(1)], but on the other hand, one might argue that fpd should not be called a dimension function since fpd(Db(coh(Pn))) =(1, ∞, n = 1 n ≥ 2 [Propositions 6.5 and 6.7]. In the latter case, fpd is an indicator of representation type of the category of coh(Pn), namely, coh(Pn) is tame if n = 1, and is of wild representation type for all n ≥ 2. A similar statement holds for projective curves in terms of genus [Proposition 6.5]. We can define the Frobenius-Perron ("fp") version of several other classical in- variants fp global dimension (denoted by fpgldim) [Definition 2.7(1)], fp Kodaira dimension (denoted by fpκ) [CG]. The first one is defined for all triangulated categories and the second one is defined for triangulated categories with Serre functor. In general, the fpgldim A does not agree with the classical global dimension of A [Theorem 7.8]. The fp version of the Kodaira dimension agrees with the classical definition for smooth projective schemes [CG]. Our second goal is to provide several applications. 0.2. Embeddings. In addition to the fact that the Frobenius-Perron dimension is an effective and sensible invariant of many categories, this invariant increases when the "size" of the endofunctors and categories increase. Theorem 0.1. Suppose C and D are k-linear categories. Let F : C → D be a fully faithful functor. Let σC and σD be endofunctors of C and D respectively. Suppose that F ◦ σC is naturally isomorphic to σD ◦ F . Then FP(u, t, σC) ≤ FP(u, t, σD). By taking σ to be the suspension functor of a pre-triangulated category, we have the following immediate consequence. (Note that the fp-dimension of a triangulated category T is defined to be fpd(Σ), where Σ is the suspension of T .) Corollary 0.2. Let T2 be a pre-triangulated category and T1 a full pre-triangulated subcategory of T2. Then the following hold. (1) fpd T1 ≤ fpd T2. (2) fpg T1 ≤ fpg T2. (3) fpv T1 ≤ fpv T2. (4) If T2 has fp-subexponential growth, so does T1. Fully faithful embeddings of derived categories of projective schemes have been investigated in the study of Fourier-Mukai transforms, birational geometry, and noncommutative crepant resolutions (NCCRs) by Bondal-Orlov [BO1, BO2], Van den Bergh [VdB], Bridgeland [Bri], Bridgeland-King-Reid [BKR] and more. Note that if fpgldim(T ) < ∞, then fpg(T ) = 0. If fpg(T ) < ∞, then fpv(T ) ≤ 1. Hence, fpd, fpgldim, fpg and fpv measure the "size", "representation type", or "complexity" of a triangulated category T at different levels. Corollary 0.2 has many consequences concerning non-existence of fully faithful embeddings provided that we compute the invariants fpd, fpg and fpv of various categories efficiently. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 3 0.3. Tame vs wild. Here we mention a couple of more applications. First we extend the classical trichotomy on the representation types of quivers to the fpd. Theorem 0.3. Let Q be a finite quiver and let Q be the bounded derived category of finite dimensional left kQ-modules. (1) kQ is of finite representation type if and only if fpd Q = 0. (2) kQ is of tame representation type if and only if fpd Q = 1. (3) kQ is of wild representation type if and only if fpd Q = ∞. By the classical theorems of Gabriel [Ga1] and Nazarova [Na], the quivers of respectively. finite and tame representation types correspond to the ADE and eAeDeE diagrams The above theorem fails for quiver algebras with relations [Proposition 5.12]. As we have already seen, fpd is related to the "size" of a triangulated category, as well as, the representation types. We will see soon that fpg is also closely connected with the complexity of representations. When we focus on the representation types, we make some tentative definitions. Let T be a triangulated category (such as Db(Modf.d. −A)). (i) We call T fp-trivial, if fpd T = 0. (ii) We call T fp-tame, if fpd T = 1. (iii) We call T fp-potentially-wild, if fpdT > 1. Further, (iiia) T is fp-finitely-wild, if 1 < fpd T < ∞. (iiib) T is fp-locally-finitely-wild, if fpd T = ∞ and fpdn(T ) < ∞ for all n. (iiic) T is fp-wild, if fpd1 T = ∞. There are other notions of tame/wildness in representation theory, see for example, [GKr, Dr2]. Following the above definition, fpd provides a quantification of the tame-wild dichotomy. By Theorem 0.3, finite/tame/wild representation types of the path algebra kQ are equivalent to the fp-version of these properties of Q. Let A be a quiver algebra with relations and let A be the derived category Db(Modf.d. −A). Then, in general, finite/tame/wild representation types of A are NOT equivalent to the fp-version of these properties of A [Example 5.5]. It is natural to ask Question 0.4. For which classes of algebras A, is the fp-wildness of A equivalent to the classical and other wildness of A in representation theory literature? 0.4. Complexity. The complexity of a module or of an algebra is an important invariant in studying representations of finite dimensional algebras [AE, Ca, CDW, GLW]. Let A be the quiver algebra kQ/(R) with relations R. The complexity of A is defined to be the complexity of the A-module T := A/Jac(A), namely, cx(A) = cx(T ) := lim sup n→∞ logn(dim Extn A(T, T )) + 1. Let GKdim denote the Gelfand-Kirillov dimension of an algebra (see [KL] and [MR, Ch. 13]). Under some reasonable hypotheses, one can show cx(A) = GKdim ∞Mn=0 Extn A(T, T )! . It is easy to see that cx(A) is an derived invariant. We extend the definition of the complexity to any triangulated category [Definition 8.2(4)]. 4 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Theorem 0.5. Let A be the algebra kQ/(R) with relations R and let A be the bounded derived category of finite dimensional left A-modules. Then fpg(A) ≤ cx(A) − 1. The equality fpg(A) = cx(A)− 1 holds under some hypotheses [Theorem 8.4(2)]. 0.5. Frobenius-Perron function. If T is a triangulated category with Serre func- tor S, we have a fp-function which is defined by f p : Z2 −→ R≥0 f p(a, b) := fpd(Σa ◦ Sb) ∈ R≥0. Then fpd(T ) is the value of the fp-function at (1, 0). Examples 5.1 and 5.4 respectively. The fp-function for the projective line P1 and the quiver A2 are given in the The statements in Theorem 0.3, Questions 0.4 and 7.11 indicate that f p(1, 0) predicts the representation type of T for certain triangulated categories. It is expected that values of the fp-function at other points in Z2 are sensitive to other properties of T . 0.6. Other properties. The paper contains some basic properties of fpd. Let us mention one of them. Proposition 0.6 (Serre duality). Let C be a Hom-finite category with Serre functor S. Let σ be an endofunctor of C. (1) If σ has a right adjoint σ!, then (2) If σ is an equivalence with quasi-inverse σ−1, then fpd(σ) = fpd(σ! ◦ S). fpd(σ) = fpd(σ−1 ◦ S). (3) If C is n-Calabi-Yau, then we have a duality, for all i, fpd(Σi) = fpd(Σn−i). 0.7. Computations. Our third goal is to develop methods for computation. To use fp-invariants, we need to compute as many examples as possible. In general it is extremely difficult to calculate useful invariants for derived categories, as the definitions of these invariants are quite sophisticated. We develop some techniques for computing fp-invariants. In Sections 4 to 8, we compute the fp-dimension for some non-trivial examples. 0.8. Conventions. (1) Usually Q means a quiver. (2) T is a (pre)-triangulated category with suspension functor Σ = [1]. (3) If A is an algebra over the base field k, then Modf.d. −A denote the category (4) If A is an algebra, then we use A for the abelian category Modf.d. −A. (5) When A is an abelian category, we use A for the bounded derived category of finite dimensional left A-modules. Db(A). FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 5 This paper is organized as follows. We provide background material in Section 1. The basic definitions are introduced in Section 2. Some basic properties are given in Section 3. Section 4 deals with some derived categories of modules over commutative rings. In Section 5, we work out the fp-theories of the projective line and quiver A2, as well as an example of non-integral fpd. In Section 6, we develop some techniques to handle the fpd of projective curves and prove the tame-wild dichotomy of projective curves in terms of fpd. Theorem 0.3 is proved in Section 7 where representation types are discussed. Section 8 focuses on the complexity of algebras and categories. We continue to develop the fp-theory in our companion paper [CG]. 1. Preliminaries 1.1. Classical Definitions. Let A be an n × n-matrix over complex numbers C. The spectral radius of A is defined to be ρ(A) := max{r1,r2,··· ,rn} ∈ R where {r1, r2,··· , rn} is the complete set of eigenvalues of A. When each entry of A is a positive real number, ρ(A) is also called the Perron root or the Perron- Frobenius eigenvalue of A. When applying ρ to the adjacency matrix of a graph (or a quiver), the spectral radius of the adjacency matrix is sometimes called the Frobenius-Perron dimension of the graph (or the quiver). Let us mention a classical result concerning the spectral radius of simple graphs. A finite graph G is called simple if it has no loops and no multiple edges. Smith [Sm] formulated the following result: Theorem 1.1. [DoG, Theorem 1.3] Let G be a finite, simple, and connected graph with adjacency matrix A. (1) ρ(A) = 2 if and only if G is one of the extended Dynkin diagrams of type eAeDeE. (2) ρ(A) < 2 if and only if G is one of the Dynkin diagrams of type ADE. To save space we refer to [DoG] and [HPR] for the diagrams of the ADE and eAeDeE quivers/graphs. the spectral radius in the following way. In order to include some infinite-dimensional cases, we extend the definition of Let A := (aij )n×n be an n× n-matrix with entries aij in R := R∪{±∞}. Define A′ = (a′ ij )n×n where a′ ij = aij xij −xij aij 6= ±∞, aij = ∞, aij = −∞. In other words, we are replacing ∞ in the (i, j)-entry by a finite real number, called xij , in the (i, j)-entry. And every xij is considered as a variable or a function mapping R → R. Definition 1.2. Let A be an n × n-matrix with entries in R. The spectral radius of A is defined to be (E1.2.1) ρ(A) := lim inf all xij→∞ ρ(A′) ∈ R. 6 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Remark 1.3. It also makes sense to use lim sup instead of lim inf in (E1.2.1). We choose to take lim inf in this paper. Here is an easy example. Example 1.4. Let A =(cid:18)1 −∞0 2 (cid:19). Then A′ =(cid:18)1 −x12 2 (cid:19). It is obvious that ρ(A) = lim x12→∞ 2 = 2. 0 ρ(A′) = lim x12→∞ 1.2. k-linear categories. If C is a k-linear category, then HomC(M, N ) is a k- module for all objects M, N in C. C(M, N ) are k- modules for all i ≥ 0. Let dim be the k-vector space dimension. Remark 1.5. One can use a dimension function other than dim. Even when a category C is not k-linear, it might still make sense to define some dimension function dim on the Hom-sets of the category C. The definition of Frobenius-Perron dimension given in the next section can be modified to fit this kind of dim. If C is also abelian, then Exti 1.3. Frobenius-Perron dimension of a quiver. In this subsection we recall some known elementary definitions and facts. Definition 1.6. Let Q be a quiver. (1) If Q has finitely many vertices, then the Frobenius-Perron dimension of Q is defined to be fpd Q := ρ(A(Q)) where A(Q) is the adjacency matrix of Q. (2) Let Q be any quiver. The Frobenius-Perron dimension of Q is defined to be fpd Q := sup{fpd Q′} where Q′ runs over all finite subquivers of Q. See [ES, Propositions 2.1 and 3.2] for connections between fpd of a quiver and its representation types, as well as its complexity. We need the following well-known facts in linear algebra. Lemma 1.7. (1) Let B be a square matrix with nonnegative entries and let A (2) Let A := (aij )n×n and B := (bij )n×n be two square matrices such that be a principal minor of B. Then ρ(A) ≤ ρ(B). 0 ≤ aij ≤ bij for all i, j. Then ρ(A) ≤ ρ(B). Let Q be a quiver with vertices {v1,··· , vn}. An oriented cycle based at a vertex vi is called indecomposable if it is not a product of two oriented cycles based at vi. For each vertex vi let θi be the number of indecomposable oriented cycles based at vi. Define the cycle number of a quiver Q to be The following result should be well-known. Θ(Q) := max{θi ∀ i}. Theorem 1.8. Let Q be a quiver and let Θ(Q) be the cycle number of Q. (1) fpd(Q) = 0 if and only if Θ(Q) = 0, namely, Q is acyclic. (2) fpd(Q) = 1 if and only if Θ(Q) = 1. (3) fpd(Q) > 1 if and only if Θ(Q) ≥ 2. The proof is not hard, and to save space, it is omitted. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 7 2. Definitions Throughout the rest of the paper, let C denote a k-linear category. A functor between two k-linear categories is assumed to preserve the k-linear structure. For simplicity, dim(A, B) stands for dim HomC(A, B) for any objects A and B in C. The set of finite subsets of nonzero objects in C is denoted by Φ and the set of subsets of n nonzero objects in C is denoted by Φn for each n ≥ 1. It is clear that Φ =Sn≥1 Φn. We do not consider the empty set as an element of Φ. Definition 2.1. Let φ := {X1, X2,··· , Xn} be a finite subset of nonzero objects in C, namely, φ ∈ Φn. Let σ be an endofunctor of C. (1) The adjacency matrix of (φ, σ) is defined to be A(φ, σ) := (aij)n×n, where aij := dim(Xi, σ(Xj )) ∀i, j. (2) An object M in C is called a brick [AS, Definition 2.4, Ch. VII] if HomC(M, M ) = k. If C is a pre-triangulated category, an object M in C is called an atomic object if it is a brick and satisfies (E2.1.1) HomC(M, Σ−i(M )) = 0, ∀ i > 0. (3) φ ∈ Φ is called a brick set (respectively, an atomic set) if each Xi is a brick (respectively, atomic) and dim(Xi, Xj) = δij for all 1 ≤ i, j ≤ n. The set of brick (respectively, atomic) n-object subsets is denoted by Φn,b (respectively, Φn,a). We write Φb =Sn≥1 Φn,b (respec- tively, Φa =Sn≥1 Φn,a). Define the b-height of C to be hb(C) = sup{n Φn,b is nonempty} and the a-height of C (when C is pre-triangulated) to be ha(C) = sup{n Φn,a is nonempty}. Remark 2.2. (1) A brick may not be atomic. Let A be the algebra khx, yi/(x2, y2 − 1, xy + yx). This is a 4-dimensional Frobenius algebra (of injective dimension zero). There are two simple left A-modules S0 := A/(x, y − 1), and S1 := A/(x, y + 1). Let Mi be the injective hull of Si for i = 0, 1. (Since A is Frobenius, Mi is projective.) There are two short exact sequences 0 −→ S0 −→ M0 f −−→ S1 −→ 0 and g 0 −→ S1 −−→ M1 −→ S0 −→ 0. It is easy to check that HomA(Mi, Mj) = k for all 0 ≤ i, j ≤ 1. Let A be the derived category Db(Modf.d. −A) and let X be the complex ··· −→ 0 −→ M0 g◦f −−→ M1 −→ 0 −→ ··· 8 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU An easy computation shows that HomA(X, X) = k = HomA(X, X[−1]). So X is a brick, but not atomic. (2) A brick object is called a schur object by several authors, see [CC] and [CKW]. It is also called endo-simple by others, see [vR1, vR2]. (3) An atomic object in a triangulated category is close to being a point-object defined by Bondal-Orlov [BO1, Definition 2.1]. A point-object was defined on a triangulated category with Serre functor. In this paper we do not automatically assume the existence of a Serre functor in general. When C is not a pre-triangulated category, we can not even ask for (E2.1.1). In that case we can only talk about bricks. Definition 2.3. Retain the notation as in Definition 2.1, and we use Φb as the testing objects. When C is a pre-triangulated category, Φb is automatically replaced by Φa unless otherwise stated. (1) The nth Frobenius-Perron dimension of σ is defined to be fpdn(σ) := sup φ∈Φn,b{ρ(A(φ, σ))}. If Φn,b is empty, then by convention, fpdn(σ) = 0. (2) The Frobenius-Perron dimension of σ is defined to be φ∈Φb{ρ(A(φ, σ))}. n {fpdn(σ)} = sup fpd(σ) := sup (3) The Frobenius-Perron growth of σ is defined to be fpg(σ) := sup φ∈Φb{lim sup n→∞ logn(ρ(A(φ, σn)))}. By convention, logn 0 = −∞. (4) The Frobenius-Perron curvature of σ is defined to be (ρ(A(φ, σn)))1/n}. φ∈Φb{lim sup fpv(σ) := sup n→∞ This is motivated by the concept of the curvature of a module over an algebra due to Avramov [Av]. (5) We say σ has fp-exponential growth (respectively, fp-subexponential growth) if fpv(σ) > 1 (respectively, fpv(σ) ≤ 1). Sometimes we prefer to have all information from the Frobenius-Perron dimen- sion. We make the following definition. Definition 2.4. Let C be a category and σ be an endofunctor of C. (1) The Frobenius-Perron theory (or fp-theory) of σ is defined to be the set (2) The Frobenius-Perron series (or fp-series) of σ is defined to be {fpdn(σm)}n≥1,m≥0. FP(u, t, σ) := fpdn(σm)tmun. ∞Xm=0 ∞Xn=1 Remark 2.5. To define Frobenius-Perron dimension, one only needs have an as- signment τ : Φn → Mn×n(Mod−k), for every n ≥ 1, satisfying the property that if φ1 is a subset of φ2, then τ (φ1) is a principal submatrix of τ (φ2). FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 9 Then we define the adjacency matrix of φ ∈ Φn to be A(φ, τ ) = (aij )n×n where aij = dim (τ (φ))ij ∀i, j. Then the Frobenius-Perron dimension of τ is defined in the same way as in Definition 2.3. If there is a sequence of τm, the Frobenius-Perron series of {τm} is defined in the same way as in Definition 2.4 by replacing σm by τm. See Example 2.6 next. (1) Let A be a k-linear abelian category. For each m ≥ 1 and Example 2.6. φ = {X1,··· , Xn}, define Em : φ −→ (Extm A (Xi, Xj))n×n . A(Xi, Xj) denote Hom0 By convention, let Ext0 A(Xi, Xj). Then, for each m ≥ 0, one can define the Frobenius-Perron dimension of Em as mentioned in Remark 2.5. (2) Let A be the k-linear abelian category Modf.d. −A where A is a finite di- mensional commutative algebra over a base field k. For each m ≥ 1 and φ = {X1,··· , Xn}, define Tm : φ −→(cid:0)TorA m(Xi, Xj)(cid:1)n×n . By convention, let TorA 0 (Xi, Xj) denote Xi ⊗A Xj. Then, for each m ≥ 0, one can define the Frobenius-Perron dimension of Tm as mentioned in Remark 2.5. Definition 2.7. (1) Let A be an abelian category. The Frobenius-Perron di- mension of A is defined to be fpd A := fpd(E1) where E1 := Ext1 Perron theory of A is the collection A(−,−) is defined as in Example 2.6(1). The Frobenius- {fpdm(En)}m≥1,n≥0 where En := Extn A(−,−) is defined as in Example 2.6(1). (2) Let T be a pre-triangulated category with suspension Σ. The Frobenius- Perron dimension of T is defined to be fpd T := fpd(Σ). The Frobenius-Perron theory of T is the collection {fpdm(Σn)}m≥1,n∈Z. The fp-global dimension of T is defined to be fpgldimT := sup{n fpd(Σn) 6= 0}. If T possesses a Serre functor S, the Frobenius-Perron S-theory of T is the collection {fpdm(Σn ◦ Sw)}m≥1,n,w∈Z. Remark 2.8. (1) The Frobenius-Perron dimension (respectively, Frobenius- Perron theory, fp-global dimension) can be defined for suspended categories [KV] and pre-n-angulated categories [GKO] in the same way as Definition 2.7(2) since there is a suspension functor Σ. 10 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU (2) When A is an abelian category, another way of defining the Frobenius- Perron dimension fpd A is as follows. We first embed A into the derived category Db(A). The suspension functor Σ of Db(A) maps A to A[1] (so it is not a functor of A). The adjacency matrix A(φ, Σ) is still defined as in Definition 2.1(1) for brick sets φ in A. Then we can define fpd(Σ A) := sup φ∈Φb,φ⊂A{ρ(A(φ, Σ))} as in Definition 2.3(2) by considering only the brick sets in A. Now fpd(A) agrees with fpd(Σ A). The following lemma is clear. Lemma 2.9. Let A be an abelian category and n ≥ 1. Then fpdn(Db(A)) ≥ fpdn(A). A similar statement holds for fpd, fpg and fpv. Proof. This follows from the fact that there is a fully faithful embedding A → Db(A) and that E1 on A agrees with Σ on Db(A). (cid:3) by Qσ C, as follows: For any category C with an endofunctor σ, we define the σ-quiver of C, denoted (1) the vertex set of Qσ objects in Φ1,a when C is pre-triangulated), and (respectively, in Φ1,a), where nX,Y = dim(X, σ(Y )). C consists of bricks in Φ1,b in C (respectively, atomic C consists of nX,Y -arrows from X to Y , for all X, Y ∈ Φ1,b (2) the arrow set of Qσ If σ is E1, this quiver is denoted by QE1 The following lemma follows from the definition. C , which will be used in later sections. Lemma 2.10. Retain the above notation. Then fpd σ ≤ fpd Qσ C . The fp-theory was motivated by the Frobenius-Perron dimension of objects in tensor or fusion categories [EG], see the following example. Example 2.11. First we recall the definition of the Frobenius-Perron dimension given in [EG, Definitions 3.3.3 and 6.1.6]. Let C be a finite semisimple k-linear tensor category. Suppose that {X1,··· , Xn} is the complete list of non-isomorphic simple objects in C. Since C is semisimple, every object X in C is a direct sum X = X ⊕ai i nMi=1 V ⊗C Xj ∼= nMi=1 i for some integers ai ≥ 0. The tensor product on C makes its Grothendieck ring Gr(C) a Z+-ring [EG, Definition 3.1.1]. For every object V in C and every j, write (E2.11.1) X ⊕aij for some integers aij ≥ 0. In the Grothendieck ring Gr(C), the left multiplication i=1 aij Xi. Then, by [EG, Definition 3.3.3], the Frobenius- Perron dimension of V is defined to be by V sends Xj to Pn (E2.11.2) FPdim(V ) := ρ((aij )n×n). In fact the Frobenius-Perron dimension is defined for any object in a Z+-ring. Next we use Definition 2.3(2) to calculate the Frobenius-Perron dimension. Let σ be the tensor functor V ⊗C − that is a k-linear endofunctor of C. If φ is a brick FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 11 subset of C, then φ is a subset of φn := {X1,··· , Xn}. For simplicity, assume that φ is {X1,··· , Xs} for some s ≤ n. It follows from (E2.11.1) that HomC(Xi, σ(Xj)) = k⊕aij , ∀ i, j. Hence the adjacency matrix of (φn, σ) is A(φn, σ) = (aij )n×n and the adjacency matrix of (φ, σ) is a principal minor of A(φn, σ). By Lemma 1.7(1), ρ(A(φ, σ)) ≤ ρ(A(φn, σ)). By Definition 2.3(2), the Frobenius-Perron di- mension of the functor σ = V ⊗C − is fpd(V ⊗C −) = sup φ∈Φb{ρ(A(φ, σ))} = ρ(A(φn, σ)) = ρ((aij )n×n), which agrees with (E2.11.2). This justifies calling fpd(V ⊗C−) the Frobenius-Perron dimension of V . For simplicity, "Frobenius-Perron" is abbreviated to "fp". 3. Basic properties 3.1. Embeddings. It is clear that the fp-series and the fp-dimensions are invari- ant under equivalences of categories. We record this fact below. Recall that the Frobenius-Perron series FP(u, t, σ) of an endofunctor σ is defined in Definition 2.4(2). Lemma 3.1. Let F : C → D be an equivalence of categories. Let σC and σD be endofunctors of C and D respectively. Suppose that F ◦ σC is naturally isomorphic to σD ◦ F . Then FP(u, t, σC) = FP(u, t, σD). Let R+ denote the set of non-negative real numbers union with {∞}. Let f (u, t) := fm,ntmun and g(u, t) := gm,ntmun ∞Xm,n=0 ∞Xm,n=0 be two elements in R+[[u, t]]. We write f ≤ g if fm,n ≤ gm,n for all m, n. Theorem 3.2. Let F : C → D be a faithful functor that preserves brick subsets. (1) Let σC and σD be endofunctors of C and D respectively. Suppose that F ◦ σC is naturally isomorphic to σD ◦ F . Then FP(u, t, σC) ≤ FP(u, t, σD). (2) Let τC and τD be assignments of C and D respectively satisfying the property in Remark 2.5. Suppose that ρ(A(φ, τC)) ≤ ρ(A(F (φ), τD )) for all φ ∈ Φn,b(C) and all n. Then FP(u, t, τC) ≤ FP(u, t, τD). Proof. (1) For every φ = {X1,··· , Xn} ∈ Φn(C), let F (φ) be {F (X1),··· , F (Xn)} in Φn(D). By hypothesis, if φ ∈ Φn,b(C), then F (φ) is in Φn,b(D). Let A = (aij) (respectively, B = (bij)) be the adjacency matrix of (φ, σC) (respectively, of (F (φ), σD)). Then, by the faithfulness of F , aij = dim(Xi, σC(Xj)) ≤ dim(F (Xi), F (σC(Xj))) = dim(F (Xi), σD(F (Xj))) = bij. By Lemma 1.7(2), (E3.2.1) ρ(A(φ, σC )) =: ρ(A) ≤ ρ(B) := ρ(A(F (φ), σD)). 12 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU By definition, (E3.2.2) Similarly, for all n, m, fpdn(σm fpdn(σC) ≤ fpdn(σD). C ) ≤ fpdn(σm (2) The proof of part (2) is similar. D ). The assertion follows. (cid:3) Theorem 0.1 follows from Theorem 3.2. 3.2. (a-)Hereditary algebras and categories. Recall that the global dimension of an abelian category A is defined to be gldim A := sup{n Extn A(X, Y ) 6= 0, for some X, Y ∈ A}. The global dimension of an algebra A is defined to be the global dimension of the category of left A-modules. An algebra (or an abelian category) is called hereditary if it has global dimension at most one. There is a nice property concerning the indecomposable objects in the derived category of a hereditary abelian category (see [Ke1, Section 2.5]). Lemma 3.3. Let A be a hereditary abelian category. Then every indecomposable object in the derived category D(A) is isomorphic to a shift of an object in A. Note that every brick (or atomic) object in an additive category is indecompos- able. Based on the property in the above lemma, we make a definition. Definition 3.4. An abelian category A is called a-hereditary (respectively, b- hereditary) if every atomic (respectively, brick) object X in the bounded derived category Db(A) is of the form M [i] for some object M in A and i ∈ Z. The object M is automatically a brick object in A. If α is an auto-equivalence of an abelian category A, then it extends naturally to an auto-equivalence, denoted by α, of the derived category A := Db(A). The main result in this subsection is the following. Recall that the b-height of A, de- noted by hb(A), is defined in Definition 2.1(3) and that the Frobenius-Perron global dimension of A, denoted by fpgldimA, is defined in Definition 2.7(2). Theorem 3.5. Let A be an a-hereditary abelian category with an auto-equivalence α. For each n, define n′ = min{n, hb(A)}. Let A be Db(A). (1) If m < 0 or m > gldim A, then fpd(Σm ◦ α) = 0. As a consequence, fpgldimA ≤ gldim A. (2) For each n, (E3.5.1) fpdn(α) ≤ fpdn(α) ≤ max 1≤i≤n′{fpdi(α)}. If gldim A < ∞, then (E3.5.2) fpdn(α) = max 1≤i≤n′{fpdi(α)}. (3) Let g := gldim A < ∞. Let β be the assignment (X, Y ) → (Extg A(X, α(Y ))). Then (E3.5.3) fpdn(Σg ◦ α) = max 1≤i≤n′{fpdi(β)}. (4) For every hereditary abelian category A, we have fpd(A) = fpd(A). (E3.5.4) A(φ, α) = A11 ∗ ∗ . . . ∗ 0 A22 ∗ . . . ∗ 0 0 A33 . . . ∗ ··· 0 ··· 0 ··· 0 . . . 0 ··· Ass  FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 13 Proof. (1) Since A is a-hereditary, every atomic object in A is of the form M [i]. is in A. Then, for i ≤ j, Case 1: m < 0. Write φ as {M1[d1],··· , Mn[dn]} where di is decreasing and Mi aij = HomA(Mi[di], (Σm ◦ α)Mj[dj ]) = HomA(Mi, α(Mj)[dj − di + m]) = 0 since dj − di + m < 0. Thus the adjacency matrix A := (aij )n×n is strictly lower triangular. As a consequence, ρ(A) = 0. By definition, fpd(Σm ◦ α) = 0. Case 2: m > gldim A. Write φ as {M1[d1],··· Mn[dn]} where di is increasing and Mi is in A. Then, for i ≥ j, aij = HomA(Mi[di], (Σm ◦ α)Mj[dj ]) = HomA(Mi, α(Mj)[dj − di + m]) = 0 since dj − di + m > gldim A. Thus the adjacency matrix A := (aij)n×n is strictly upper triangular. As a consequence, ρ(A) = 0. By definition, fpd(Σm ◦ α) = 0. (2) Let F be the canonical fully faithful embedding A → A. By Theorem 3.2 and (E3.2.2), fpdn(α) ≤ fpdn(α). For the other assertion, write φ as a disjoint union φd1 ∪···∪ φds where di is strictly decreasing and the subset φdi consists of objects of the form M [di] for M ∈ A. For any objects X ∈ φdi and Y ∈ φdj for i < j, HomA(X, Y ) = 0. Thus the adjacency matrix of (φ, α) is of the form where each Aii is the adjacency matrix A(φdi , α). For each φdi , we have which implies that A(φdi , α) = A(φdi [−di], α) = A(φdi [−di], α) ρ(Aii) ≤ fpdsi(α) ≤ max 1≤j≤n′ fpdj(α) where si is the size of Aii and n′ = min{n, hb(A)}. By using the matrix (E3.5.4), ρ(A(φ, α)) = max i {ρ(Aii)} ≤ max 1≤j≤n′ fpdj(α). Then (E3.5.1) follows. Suppose now that g := gldim A < ∞. Let φ ∈ Φn,a(A). Pick any M ∈ Φ1,b(A). Then, for g′ ≫ g, φ′ := φ ∪ {M [g′]} ∈ Φn+1,a(A). By Lemma 1.7(1), ρ(A(φ′, α)) ≥ ρ(A(φ, α)). Hence fpdn(α) is increasing as n increases. Therefore (E3.5.2) follows from (E3.5.1). (3) The proof is similar to the proof of part (2). Let F be the canonical fully faithful embedding A → A. By Theorem 3.2(2) and (E3.2.2), fpdn(β) ≤ fpdn(Σg ◦ α). By the argument at the end of proof of part (2), fpdn(Σg ◦ α) increases when n increases. Then max 1≤j≤n′ fpdj(β) ≤ fpdn(Σg ◦ α). For the other direction, write φ as a disjoint union φd1 ∪ ··· ∪ φds where di is strictly increasing and φdi consists of objects of the form M [di] for M ∈ A. For 14 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU objects X ∈ φdi and Y ∈ φdj for i < j, HomA(X, Σg(α(Y ))) = 0. Let γ = Σg ◦ α. Then the adjacency matrix of (φ, γ) is of the form (E3.5.4), namely, A(φ, γ) = A11 ∗ ∗ . . . ∗ 0 A22 ∗ . . . ∗ 0 0 A33 . . . ∗ ··· 0 ··· 0 ··· 0 . . . 0 ··· Ass  where each Aii is the adjacency matrix A(φdi , γ). For each φdi , we have which implies that A(φdi , γ) = A(φdi [−di], γ) = A(φdi [−di], β) ρ(Aii) ≤ fpdsi(β) ≤ max 1≤j≤n′ fpdj(β) where si is the size of Aii. By using matrix (E3.5.4), ρ(A(φ, γ)) = max i {ρ(Aii)} ≤ max 1≤j≤n′ fpdj(β). The assertion follows. (4) Take α to be the identity functor of A and g = 1 (since A is hereditary). By (E3.5.3), we have fpdn(Σ) = max 1≤i≤n′{fpdi(E1)}. By taking supn, we obtain that fpd(E1) = fpd(Σ). The assertion follows. (cid:3) 3.3. Categories with Serre functor. Recall from [Ke2, Section 2.6] that if a Hom-finite category C has a Serre functor S, then there is a natural isomorphism HomC(X, Y )∗ ∼= HomC(Y, S(X)) for all X, Y ∈ C. A (pre-)triangulated Hom-finite category C with Serre functor S is called n-Calabi-Yau if there is a natural isomorphism S ∼= Σn. (In [Ke2, Section 2.6] it is called weakly n-Calabi-Yau.) We now prove Proposition 0.6. Proposition 3.6 (Serre duality). Let C be a Hom-finite category with Serre functor S. Let σ be an endofunctor of C. (1) If σ has a right adjoint σ!, then (2) If σ is an equivalence with quasi-inverse σ−1, then fpd(σ) = fpd(σ! ◦ S). fpd(σ) = fpd(σ−1 ◦ S). (3) If C is (pre-)triangulated and n-Calabi-Yau, then we have a duality fpd(Σi) = fpd(Σn−i) for all i. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 15 Proof. (1) Let φ = {X1,··· , Xn} ∈ Φn,b and let A(φ, σ) be the adjacency matrix with (i, j)-entry aij = dim(Xi, σ(Xj)). By Serre duality, aij = dim(Xi, σ(Xj)) = dim(σ(Xj ), S(Xi)) = dim(Xj, (σ! ◦ S)(Xi)), which is the (j, i)-entry of the adjacency matrix A(φ, σ! ◦ S). Then ρ(A(φ, σ)) = ρ(A(φ, σ! ◦ S)). It follows from the definition that fpdn(σ) = fpdn(σ! ◦ S) for all n ≥ 1. The assertion follows from the definition. (2,3) These are consequences of part (1). (cid:3) 3.4. Opposite categories. Lemma 3.7. Let σ be an endofunctor of C and suppose that σ has a left adjoint σ∗. Consider σ∗ as an endofunctor of the opposite category Cop of C. Then fpdn(σ C) = fpdn(σ∗ Cop ). for all n. Proof. Let φ := {X1,··· , Xn} be a brick subset of C (which is also a brick subset of Cop). Then dimC(Xi, σ(Xj)) = dimC(σ∗(Xi), Xj) = dimCop (Xj, σ∗(Xi)), which implies that the adjacency matrix of σ∗ as an endofunctor of Cop is the transpose of the adjacency matrix of σ. The assertion follows. (cid:3) Definition 3.8. (1) Two pre-triangulated categories (Ti, Σi), for i = 1, 2, are called fp-equivalent if fpdn(Σm 1 ) = fpdn(Σm 2 ) (2) Two algebras are fp-equivalent if their bounded derived categories of finitely for all n ≥ 1, m ∈ Z. generated modules are fp-equivalent. (3) Two pre-triangulated categories with Serre functors (Ti, Σi, Si), for i = 1, 2, are called fp-S-equivalent if fpdn(Σm for all n ≥ 1, m, k ∈ Z. 1 ◦ Sk 1 ) = fpdn(Σm 2 ◦ Sk 2 ) Proposition 3.9. Let T be a pre-triangulated category. (1) T and T op are fp-equivalent. (2) Suppose S is a Serre functor of T . Then (T , S) and (T op, Sop) are fp-S- equivalent. Proof. (1) Let Σ be the suspension of T . Then T op is also pre-triangulated with suspension functor being Σ−1 = Σ∗ (restricted to T op). The assertion follows from Lemma 3.7. (2) Note that the Serre functor of T op is equal to S−1 = S∗ (restricted to T op). The assertion follows by Lemma 3.7. (cid:3) Corollary 3.10. Let A be a finite dimensional algebra. (1) A and Aop are fp-equivalent. (2) Suppose A has finite global dimension. In this case, the bounded derived category of finite dimensional A-modules has a Serre functor. Then A and Aop are fp-S-equivalent. 16 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Proof. (1) Since A is finite dimensional, the k-linear dual induces an equivalence of triangulated categories between Db(Modf.d. −A)op and Db(Modf.d. −Aop). The assertion follows from Proposition 3.9(1). (cid:3) The proof of (2) is similar by using Proposition 3.9(2) instead. There are examples where T and T op are not triangulated equivalent, see Ex- ample 3.12. In this paper, a k-algebra A is called local if A has a unique maximal ideal m and A/m ∼= k. The following lemma is easy and well-known. Lemma 3.11. Let A be a finite dimensional local algebra over k. Let A be the category Modf.d. −A and A be Db(A). (1) Let X be an object in A such that HomA(X, X[−i]) = 0 for all i > 0. Then (2) Every atomic object in A is of the form M [n] where M is a brick object in X is of the form M [n] where M is an object in A and n ∈ Z. A and n ∈ Z. Namely, A is a-hereditary. Proof. (2) is an immediate consequence of part (1). We only prove part (1). On the contrary we suppose that H m(X) 6= 0 and H n(X) 6= 0 for some m < n. Since X is a bounded complex, we can take m to be minimum of such integers and n to be the maximum of such integers. Since A is local, there is a nonzero map from H n(X) → H m(X), which induces a nonzero morphism in HomA(X, X[m − n]). This contradicts the hypothesis. Example 3.12. Let m, n be integers ≥ 2. Define Am,n to be the algebra (cid:3) khx1, x2i/(xm 1 , xn 2 , x1x2). It is easy to see that Am,n is a finite dimensional local connected graded algebra generated in degree 1 (with deg x1 = deg x2 = 1). If Am,n is isomorphic to Am′,n′ as algebras, by [BZ, Theorem 1], these are isomorphic as graded algebras. Suppose f : Am,n → Am′,n′ is an isomorphism of graded algebras and write f (x1) = ax1 + bx2, f (x2) = cx1 + dx2. Then the relation f (x1)f (x2) = 0 forces b = c = 0. As a consequence, m = m′ and n = n′. So we have proven that (1) Am,n is isomorphic to Am′,n′ if and only if m = m′ and n = n′. Next we claim that (2) if m 6= n, then the derived category Db(Modf.d. −Am,n) is not triangulated equivalent to Db(Modf.d. −Aop m,n). Let m, n, m′, n′ be integers ≥ 2. Suppose that Db(Modf.d. −Am,n) is triangulated equivalent to Db(Modf.d. −Am′,n′). Since Am,n is local, by [Ye, Theorem 2.3], every tilting complex over Am,n is of the form P [n] where P is a progenerator over Am,n. As a consequence, Am,n is Morita equivalent to Am′,n′. Since both Am,n and Am′,n′ are local, Morita equivalence implies that Am,n is isomorphic to Am′,n′ . By part (1), m = m′ and n = n′. In other words, if (m, n) 6= (m′, n′), then Db(Modf.d. −Am,n) is not triangulated equivalent to Db(Modf.d. −Am′,n′). As a consequence, if m 6= n, then Db(Modf.d. −Am,n) is not triangulated equivalent to Db(Modf.d. −An,m). By definition, Aop We can show that Db(Modf.d. −A) is dual to Db(Modf.d. −Aop) by using the In other words, Db(Modf.d. −A)op is triangulated equivalent to k-linear dual. Db(Modf.d. −Aop). Therefore the following is a consequence of part (2). m,n ∼= An,m. Therefore the claim (2) follows. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 17 (3) Suppose m 6= n and let A be Db(Modf.d. −Am,n). Then A is not tri- angulated equivalent to Aop. But by Proposition 3.9(1), A and Aop are fp-equivalent. 4. Derived category over a commutative ring Throughout this section A is a commutative algebra and A = Db(Mod−A). (In other sections A usually denotes Db(Modf.d. −A).) Lemma 4.1. Let A be a commutative algebra. Let X be an atomic object in A. Then X is of the form M [i] for some simple A-module M and some i ∈ Z. As a consequence, Mod−A is a-hereditary. Proof. Consider X as a bounded above complex of projective A-modules. Since A is commutative, every f ∈ A induces naturally a morphism of X by multiplication. For each i, H i(X) is an A-module. We have natural morphisms of A-algebras A → HomA(X, X) → EndA(H i(X)). By definition, HomA(X, X) = k. Thus HomA(X, X) = A/m for some ideal m of A that has codimension 1. Hence the A-action on H i(X) factors through the map A → A/m. This means that H i(X) is a direct sum of A/m. for some s, t > 0. If m < n, then Let n = sup X and m = inf X. Then H m(X) = (A/m)⊕s and H n(X) = (A/m)⊕t HomA(X, X[m − n]) ∼= HomA(X[n], X[m]) ∼= HomA(H n(X), H m(X)) 6= 0 which contradicts (E2.1.1). Therefore m = n and X = M [n] for M := H n(X). Since X is atomic, M has only one copy of A/m. (cid:3) Lemma 4.2. Let A be a commutative algebra. Let X and Y be two atomic objects in A. Then HomA(X, Y ) 6= 0 if and only if there is an ideal m of A of codimension 1 such that X ∼= A/m[m] and Y ∼= A/m[n] for some 0 ≤ n − m ≤ projdim A/m. Proof. By Lemma 4.1, X ∼= A/m[m] for some ideal m of codimension 1 and some integer m. Similarly, Y ∼= A/n[n] for ideal n of codimension 1 and integer n. m = n. Further, Extn−m projdim A/m. The converse can be proved in a similar way. If m 6= n, then clearly HomA(X, Y ) = 0. Hence (A/m, A/m) ∼= HomA(X, Y ) 6= 0 implies that 0 ≤ n− m ≤ Suppose HomA(X, Y ) 6= 0. (cid:3) A If A is an affine commutative ring over k, then every simple A-module is 1- dimensional. Hence (A/m)[i] is a brick (and atomic) object in A for every i ∈ Z and every maximal ideal m of A. The fp-global dimension fpgldim(A) is defined in Definition 2.7(2). Proposition 4.3. Let A be an affine commutative domain of global dimension g < ∞. (1) fpd(A) = g. (3) fpgldim(A) = g. (2) fpd(Σi) =(cid:0)g i(cid:1) for all i. Proof. (1) By Lemma 4.1, every atomic object is of the form M [i] for some M ∼= A/m where m is an ideal of codimension 1, and i ∈ Z. It is well-known that (E4.3.1) dim Exti A(A/m, A/m) =(cid:18)g i(cid:19), ∀ i. 18 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU If m1 and m2 are two different maximal ideals, then (E4.3.2) Exti A(A/m1, A/m2) = 0 for all i. Let φ be an atomic n-object subset. We can decompose φ into a disjoint union φA/m1 ∪ ··· ∪ φA/ms where φA/m consists of objects of the form A/m[i] for i ∈ Z. It follows from (E4.3.2) that the adjacency matrix is a block-diagonal matrix. Hence, we only need to consider the case when φ = φA/m after we use the reduction similar to the one used in the proof of Theorem 3.5. Let φ = φA/m = {A/m[d1],··· , A/m[dm]} where di is increasing. By Lemma 4.2, we have di+1−di > g, or di + g < di+1, for all i = 1,··· , m − 1. Under these conditions, the adjacency matrix is lower triangular with each diagonal being g. Thus fpd(Σ) = g. The proof of (2) is similar. (3) is a consequence of (2). (cid:3) Suggested by Theorem 3.5, we could introduce some secondary invariants as follows. The stabilization index of a triangulated category T is defined to be SI(T ) = min{n fpdn′ T = fpd T , ∀ n′ ≥ n}. The global stabilization index of T is defined to be GSI(T ) = max{SI(T ′) for all thick triangulated full subcategories T ′ ⊆ T }. It is clear that both stabilization index and global stabilization index can be defined for an abelian category. Similar to Proposition 4.3, one can show the following. Suppose that A is affine. For every i, let di := sup{dim Exti A(A/m, A/m) maximal ideals m ⊆ A}. Proposition 4.4. Let A be an affine commutative algebra. Then, for each i, fpd(Σi) = di < ∞ and ρ(A(φ, Σi)) ≤ di for all φ ∈ Φn,a. As a consequence, for each integer i, the following hold. (1) fpd(Σi) = fpd1(Σi). Hence the stabilization index of A is 1. (2) fpd(Σi) is a finite integer. Theorem 4.5. Let A be an affine commutative algebra and A be Db(Mod A). Let T be a triangulated full subcategory of A with suspension ΣT . Let i be an integer. T ). As a consequence, the global stabilization index of A T ) = fpd1(Σi (1) fpd(Σi is 1. T ) is a finite integer. (2) fpd(Σi (3) If T is isomorphic to Db(Modf.d. −B) for some finite dimensional algebra B, then B is Morita equivalent to a commutative algebra. Proof. (1,2) These are similar to Proposition 4.4. (3) Since B is finite dimensional, it is Morita equivalent to a basic algebra. So we can assume B is basic and show that B is commutative. Write B as a kQ/(R) where Q is a finite quiver with admissible ideal R ⊆ (kQ)≥2. We will show that B is commutative. First we claim that each connected component of Q consists of only one vertex. Suppose not. Then Q contains distinct vertices v1 and v2 with an arrow α : v1 → v2. Let S1 and S2 be the simple modules corresponding to v1 and v2 respectively. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 19 Then {S1, S2} is an atomic set in T . The arrow represents a nonzero element in Ext1 B(S1, S2). Hence HomT (S1, S2[1]) ∼= Ext1 B(S1, S2) 6= 0. By Lemma 4.2, S1 is isomorphic to a complex shift of S2. But this is impossible. Therefore, the claim holds. It follows from the claim in the last paragraph that B = B1 ⊕ ··· ⊕ Bn where each Bi is a finite dimensional local ring corresponding to a vertex, say vi. Next we claim that each Bi is commutative. Without loss of generality, we can assume Bi = B. Now let ι be the fully faithful embedding from ι : T := Db(Modf.d. −B) −→ A := Db(Mod−A). Let S be the unique simple left B-module. Then, by Lemma 4.1, there is a maximal ideal m of A such that ι(S) = A/m[w] for some w ∈ Z. After a shift, we might assume that ι(S) = A/m. The left B-module B has a composition series such that each simple subquotient is isomorphic to S, which implies that, as a left A- module, ι(B) is generated by A/m in A. By induction on the length of B, one sees that, for every n ∈ Z, H n(ι(B)) is a left A/md-module for some d ≫ 0 (we can take d = length(BB)). Since HomA(ι(B), ι(B)[−i]) = HomT (B, B[−i]) = 0 for all i > 0, the proof of Lemma 3.11(2) shows that ι(B) ∼= M [m] for some left A/md-module M and m ∈ Z. Since there are nonzero maps from S to B and from B to S, we have nonzero maps from A/m to ι(B) and from ι(B) to A/m. This implies that m = 0. Since B is local (and then B/mB is 1-dimensional for the maximal ideal mB), this forces that M = A/I where I is an ideal of A containing md. Finally, Bop = EndB(B) ∼= EndA(A/I, A/I) = EndA(A/I, A/I) ∼= A/I which is commutative. Hence B is commutative. (cid:3) In this section we give three examples. 5. Examples 5.1. Frobenius-Perron theory of projective line P1 := Proj k[t0, t1]. Example 5.1. Let coh(P1) =: A denote the category of coherent sheaves on P1. It is well-known (and follows from [BB, Example 3.18]) that Claim 5.1.1: Every brick object X in A (namely, satisfying HomP1(X, X) = k) is either O(m) for some m ∈ Z or Op for some p ∈ P1. P1(Op,Op) = 1. either φ = {O(m)} or φ = {Op}. Let E1 be the functor Ext1 case, ρ(A(φ, E1)) = 0 because Ext1 ρ(A(φ, E1)) = 1 because Ext1 Let φ be in Φn,b(coh(P1)). If n = 1 or φ is a singleton, then there are two cases: P1(−,−). In the first P1(O(m),O(m)) = 0, and in the second case, If φ > 1, then O(m) can not appear in φ as HomP1 (O(m),O(m′)) 6= 0 and HomP1(O(m),Op) 6= 0 for all m ≤ m′ and p ∈ P1. Hence, φ is a collection of Op for finitely many distinct points p's. In this case, the adjacency matrix is the identity n × n-matrix and ρ(A(φ, E1)) = 1. Therefore (E5.1.1) fpdn(coh(P1)) = fpd(coh(P1)) = 1 20 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU for all n ≥ 1. Since coh(P1) is hereditary, by Theorem 3.5(3,4), we obtain that (E5.1.2) for all n ≥ 1. fpdn(Db(coh(P1))) = fpd(Db(coh(P1))) = 1 Let K2 be the Kronecker quiver (E5.1.3) • 9 • By a result of Beilinson [Bei], the derived category Db(Modf.d. −kK2) is trian- gulated equivalent to Db(coh(P1)). As a consequence, (E5.1.4) fpd(Db(Modf.d. −kK2)) = fpd(Db(coh(P1))) = 1. It is easy to see, or by Theorem 1.8(1), where fpd of a quiver is defined in Definition 1.6. fpd K2 = 0 This implies that (E5.1.5) fpd(Db(Modf.d. −kK2)) > fpd K2. Next we consider some general auto-equivalences of Db(coh(P1)). Let (m) : coh(P1) → coh(P1) be the auto-equivalence induced by the shift of degree m of the graded modules over k[t0, t1] and let Σ be the suspension functor of Db(coh(P1)). Then the Serre functor S of Db(coh(P1)) is Σ◦ (−2). Let σ be the functor Σa◦ (b) for some a, b ∈ Z. By Theorem 3.5(1), fpdn(Σa ◦ (b)) = 0, ∀ a 6= 0, 1. For the rest we consider a = 0 or 1. By Theorem 3.5(2,3), we only need to consider fpd on coh(P1). If φ is a singleton {O(n)}, then the adjacency matrix is A(φ, σ) = dim(Op, Σa(Op)) = 1, It is easy to see from the above computation that for a = 0, 1. (E5.1.6) fpd1(Σa ◦ (b)) = a = 0, b < 0, 1 a = 0, b ≥ 0, b + 1 a = 1, b ≥ −1, 1 −b − 1 a = 1, b < −1. A(φ, σ) = dim(O, ΣaO(b)) = 0 a = 0, b < 0, b + 1 a = 0, b ≥ 0, 0 a = 1, b ≥ −1, −b − 1 a = 1, b < −1. This follows from the well-known computation of H i P1(O(m)) for i = 0, 1 and m ∈ Z. (It also follows from a more general computation [AZ, Theorem 8.1].) If φ = {Op} for some p ∈ P1, then the adjacency matrix is % % 9 FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 21 Now we consider the case when n > 1. If φ ∈ Φn,b(coh(P1)), φ is a collection of Op for finitely many distinct p's. In this case, the adjacency matrix A(φ, Σa ◦ (b)) is the identity n × n-matrix for a = 0, 1, and ρ(A(φ, σ)) = 1. Therefore (E5.1.7) fpdn(Σa ◦ (b)) = 1 for all n > 1, when restricted to the category coh(P1). It follows from Theorem 3.5(2,3) that Claim 5.1.2: Consider Σa ◦ (b) as an endofunctor of Db(coh(P1)). For a, b ∈ Z and n ≥ 1, we have (E5.1.8) fpdn(Σa ◦ (b)) = a 6= 0, 1, 0 a = 0, b < 0, 1 a = 0, b ≥ 0, b + 1 a = 1, b ≥ −1, 1 −b − 1 a = 1, b < −1.   Since S = Σ ◦ (−2), we have the following (also see Figure 1, next page) 0 a + b 6= 0, 1, 1 a + b = 0, b > 0, −(2b − 1) a + b = 0, b ≤ 0, 1 a + b = 1, b ≤ 0, a + b = 1, b > 0. 2b − 1 fpdn(Σa ◦ Sb) = fpdn(Σa+b ◦ (−2b)) = (E5.1.9) Claim 5.1.3: Since Db(coh(P1)) and Db(Modf.d. −kK2) are equivalent, the fp- theory of Db(Modf.d. −kK2) agrees with (E5.1.9) and Figure 1 (next page). 5.2. Frobenius-Perron theory of the quiver A2. We start with the following example. Example 5.2. Let A be the Z-graded algebra k[x]/(x2) with deg x = 1. Let C := gr−A be the category of finitely generated graded left A-modules. Let σ := (−) be the degree shift functor of C. It is clear that σ is an autoequivalence of C. Let A be the additive subcategory of C generated by σn(A) = A(n) for all n ∈ Z. Note that A is not abelian and that every object in A is of the formLn∈Z A(n)⊕pn for some integers pn ≥ 0. Since the Hom-set in the graded module category consists of homomorphisms of degree zero, we have HomA(A, A(n)) =(k n = 0, 1 otherwise 0 . In the following diagram each arrow represents 1-dimensional Hom for all possible Hom-set for different objects A(n) (E5.2.1) ··· −→ A(−2) −→ A(−1) −→ A(0) −→ A(1) −→ A(2) −→ ··· (where the number of arrows from A(m) to A(n) agrees with dim Hom(A(m), A(n))). It is easy to see that the set of indecomposable objects is {A(n)}n∈Z, which is also the set of bricks in A. 22 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Figure 1. fp-S-theory for P1, where f (a, b) = fpd(Σa ◦ Sb). Lemma 5.3. Retain the notation as in Example 5.2. When restricting σ onto the category A, we have, for every m ≥ 1, fpdm(σn) =(1 n = 0, 1, otherwise 0 (E5.3.1) . Proof. When n = 0, (E5.3.1) is trivial. Let n = 1. For each set φ ∈ Φm,b, we can assume that φ = {A(d1), A(d2, ),··· , A(dm)} for a strictly increasing sequence {di i = 1, 2,··· , m}. For any i < j, the (i, j)-entry of the adjacency matrix is aij = dim(A(di), A(dj + 1)) = 0. Thus A(φ, σ) is a lower triangular matrix with aii = dim(A(di), A(di + 1)) = 1. Hence ρ(A(φ, σ)) = 1. So fpdm(σ) = 1. Similarly, fpdm(σn) = 0 when n > 1 as dim(A(di), A(di + 2)) = 0 for all i. Let n < 0. Let φ = {A(d1), A(d2, ),··· , A(dm)} ∈ Φm,b where di are strictly decreasing. Then aij = dim(A(di), A(dj + n)) = 0 for all i ≤ j. Thus ρ(A(φ, σn)) = 0 and (E5.3.1) follows in this case. (cid:3) FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 23 Example 5.4. Consider the quiver A2 (E5.4.1) •2 ←− •1. Let Pi (respectively, Ii) be the projective (respectively, injective) left kA2-modules corresponding to vertices i, for i = 1, 2, It is well-known that there are only three indecomposable left modules over A2, with Auslander-Reiten quiver (or AR-quiver, for short) (E5.4.2) P2 −→ P1(= I2) −→ I1 where each arrow represents a nonzero homomorphism (up to a scalar) [Sc1, Ex.1.13, pp.24-25]. The AR-translation (or translation, for short) τ is determined by τ (I1) = P2. Let T be Db(Modf.d. −kA2). The Auslander-Reiten theory can be extended from the module category to the derived category. It is direct that, in T , we have the AR-quiver of all indecomposable objects (E5.4.3) P2[−1] ❅ . . . ❅❘ P1[−1] I1[−1] ✒ ❅ P1 = I2 ✒ ❅ P2[1] ✒ ❅ ❅❘ ❅❘ P2 I1 ❅❘ P1[1] I1[1] ✒ . . . The above represents all possible nonzero morphisms (up to a scalar) between non-isomorphic indecomposable objects in T . Note that T has a Serre functor S and that the AR-translation τ can be extended to a functor of T such that S = Σ◦τ [RVdB, Proposition I.2.3] or [Cr, Remarks (2), p. 23]. After we identifying P2[i] ↔ A(3i), P1[i] ↔ A(3i + 1), I1[i] ↔ A(3i + 2), (E5.4.3) agrees with (E5.2.1). Using the above identification, at least when re- stricted to objects, we have (E5.4.4) (E5.4.5) (E5.4.6) Σ(A(i)) ∼= A(i + 3), τ (A(i)) ∼= A(i − 2), S(A(i)) ∼= A(i + 1). It follows from the definition of the AR-quiver [ARS, VII] that the degree of τ is −2, see also [AS, Picture on p. 131]. Equation (E5.4.5) just means that the degree of τ is −2. By equation (E5.4.6), the Serre functor S satisfies the property of σ defined in Example 5.2. By Lemma 5.3 or (E5.3.1), we have fpdn(Σa ◦ Sb) = fpdn(σ3a+b) =(1 0 3a + b = 0, 1, otherwise. Therefore the fp-S-theory of T is given as above, and given as in Figure 2 (next page). In particular, we have proven fpgldim(Db(Modf.d. −kA2)) = fpd(Db(Modf.d. −kA2)) = fpd(Σ) = 0, which is less than gldim kA2 = 1. 24 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Figure 2. fp-S-theory for A2 5.3. An example of non-integral Frobenius-Perron dimension. In the next example, we "glue" K2 in (E5.1.3) and A2 in (E5.4.1) together. Example 5.5. Let G2 be the quiver (E5.5.1) • 1 β γ α • 2 consisting of two vertices 1 and 2, with arrow α : 2 → 1 and β, γ : 1 → 2 satisfying relations (E5.5.2) R : βα = γα = 0, αβ = αγ = 0. Note that (G2, R) is a quiver with relations. The corresponding quiver algebra with relations is a 5-dimensional algebra A = ke1 + ke2 + kα + kβ + kγ satisfying, in addition to (E5.5.2), (E5.5.3) e2 1 = e1, e2 2 = e2, 1 = e1 + e2, % % b b (E5.5.4) and (E5.5.5) FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 25 e2α = 0, αe1 = 0, e1β = e1γ = 0, βe2 = γe2 = 0, αe2 = α = e1α, βe1 = β = e2β, γe1 = γ = e2γ. We can use the following matrix form to represent the algebra A A =(cid:18) ke1 kβ + kγ ke2(cid:19) . kα For each i = 1, 2, let Si be the left simple A-module corresponding to the vertex i and Pi be the projective cover of Si. Then P1 ∼= Ae1 is isomorphic to the first kβ + kγ(cid:19); and P2 ∼= Ae2 is isomorphic to the second column column of A, namely,(cid:18) ke1 ke2(cid:19). of A, namely,(cid:18) kα We will show that the Frobenius-Perron dimension of the category of finite di- mensional representations of (G2, R) is √2, by using several lemmas below that contain some detailed computations. Lemma 5.6. Let V = (V1, V2) be a representation of (G2, R). Let W = im α and K = ker α. Take a k-space decomposition V2 = W⊕K where W ∼= W . Then there is a decomposition of (G2, R)-representations V ∼= (W ⊕T, W ⊕K) ∼= (W , W )⊕(T, K) where α is the identity when restricted to W (and identifying W with W ) and is zero when restricted to K, where β and γ are zero when restricted to W . Proof. Since W = im α, V2 ∼= W ⊕ K where K = ker α and W ∼= W . Write V1 = W ⊕ T for some k-subspace T ⊆ V1. The assertion follows by using the relations in (E5.5.2). (cid:3) Recall that A2 is the quiver given in (E5.4.1) and K2 is the Kronecker quiver given in (E5.1.3). By the above lemma, the subrepresentation (W, W ) (where we identify W with W ) is in fact a representation of (cid:18)ke1 subrepresentation (T, K) is a representation of(cid:18) ke1 0 kβ + kγ ke2(cid:19) (∼= kK2). 0 kα ke2(cid:19) (∼= kA2) and the Let In be the n × n-identity matrix. Let Bl(λ) denote the block matrix  λ 0 . . . 0 0 1 λ . . . 0 0 0 1 . . . 0 0 ··· ··· . . . ··· ··· 0 0 . . . λ 0 0 0 . . . 1 λ .  Lemma 5.7. Suppose k is of characteristic zero. The following is a complete list of indecomposable representations of (G2, R). (1) P2 ∼= (k, k), where α = I1 and β = γ = 0. (2) Xn(λ) = (K, K) with dim K = n, where α = 0, β = In and γ = Bl(λ) for some λ ∈ k. (3) Yn = (K, K) with dim K = n, where α = 0, β = Bl(0) and γ = In. (4) S2,n = (T, K) with dim T = n and dim K = n + 1, where α = 0, β = (In, 0) and γ = (0, In). 26 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU (5) S1,n = (T, K) with dim T = n+1 and dim K = n, where α = 0, β = (In, 0)τ and γ = (0, In)τ . As a consequence, kG2/(R) is of tame representation type [Definition 7.1]. Proof. (1) By Lemma 5.6, this is the only case that could happen when α 6= 0. Now we assume α = 0. (2,3,4,5) If α = 0, then we are working with representations of Kronecker quiver K2 (E5.1.3). The classification follows from a classical result of Kronecker [Ben, Theorem 4.3.2]. By (1-5), for each integer n, there are only finitely many 1-parameter families of indecomposable representations of dimension n. Therefore A is of tame represen- tation type. (cid:3) The following is a consequence of Lemma 5.7 and a direct computation. Lemma 5.8. Retain the hypotheses of Lemma 5.7. The following is a complete list of brick representations of (G2, R). (1) P2 ∼= (k, k), where α = I1 and β = γ = 0. (2) X1(λ) = (k, k), where α = 0, β = I1 and γ = λI1 for some λ ∈ k. (3) Y1 = (k, k), where α = 0, β = 0 and γ = I1. (4) S2,n for n ≥ 0. (5) S1,n for n ≥ 0. The set Φ1,b consists of the above objects. Let X1(∞) denote Y1. We have the following short exact sequences of (G2, R)- representations 0 → S1 → P2 → S2 → 0, 0 → S2 → X1(λ) → S1 → 0, 0 → Sn+1 2 → S2,n → Sn 1 → 0, 2 → S1,n → Sn+1 0 → Sn 1 → 0, 0 → S2 2 → S2,n → S1,n−1 → 0, where n ≥ 1 for the last exact sequence, and have the following nonzero homs, where A = kG2/(R), HomA(X1(λ), S1,n) 6= 0, ∀ n ≥ 1 HomA(S2,n, X1(λ)) 6= 0, ∀ n ≥ 1 HomA(S2,m, S2,n) 6= 0, ∀ m ≤ n HomA(S1,n, S1,m) 6= 0, ∀ m ≤ n HomA(S2,n, S1,m) 6= 0, ∀ m + n ≥ 1 Lemma 5.9. Retain the hypotheses of Lemma 5.7. The following is the complete list of zero hom-sets between brick representations of G2 in both directions. (1) HomA(X1(λ), X1(λ′)) = HomA(X1(λ′), X1(λ)) = 0 if λ 6= λ′ in k ∪ {∞}. (2) HomA(S1, S2) = HomA(S2, S1) = 0. As a consequence, if φ ∈ Φn,b for some n ≥ 2, then φ = {S1, S2} or φ = {X1(λi)}n i=1 for different parameters {λ1,··· , λn}. We also need to compute the Ext1 A-groups. Lemma 5.10. Retain the hypotheses of Lemma 5.7. Let λ 6= λ′ be in k ∪ {∞}. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 27 A(X1(λ), X1(λ)) = HomA(X1(λ), X1(λ)) = k. A(X1(λ), X1(λ′)) = HomA(X1(λ), X1(λ′)) = 0. (1) Ext1 (2) Ext1 (3) (cid:18)Ext1 Ext1 (4) Ext1 (5) dim Ext1 (6) dim Ext1 A(S1, S1) Ext1 A(S2, S1) Ext1 A(P2, P2) = 0. A(S1, S2) A(S2, S2)(cid:19) =(cid:18)0 k⊕2 0 (cid:19) . k A(S2,n, S2,n) ≤ 1 for all n. A(S1,n, S1,n) ≤ 1 for all n. Remark 5.11. In fact, one can show the following stronger version of Lemma 5.10(5,6). (5') Ext1 (6') Ext1 A(S2,n, S2,n) = 0 for all n. A(S1,n, S1,n) = 0 for all n. Proof of Lemma 5.10. (1,2) Consider a minimal projective resolution of X1(λ) P1 → P2 fλ−→ P1 → X1(λ) → 0 where fλ sends e2 ∈ P2 to γ − λβ ∈ P1. More precisely, we have (cid:18) ke1 kβ + kγ(cid:19) e1→α −−−−→(cid:18) kα ke2(cid:19) e2→γ−λβ −−−−−−→(cid:18) ke1 kβ + kγ(cid:19) −→ P1/(k(γ − λβ)) −→ 0. Applying HomA(−, X1(λ′)) to the truncated projective resolution of the above, we obtain the following complex k 0 ←−−−− k g ←−−−− k ←− 0. If g is zero, this is case (1). If g 6= 0, this is case (2). tions of S1 and S2. (3) The proof is similar to the above by considering minimal projective resolu- (4) This is clear since P2 is a projective module. (5,6) Let S be either S2,n or S1,n. By Example 5.1, fpd(Modf.d −kK2) = 1. This implies that dim Ext1 kK2(S, S) ≤ 1 where S is considered as an indecomposable K2-module. Let us make a comment before we continue the proof. Following a more careful analysis, one can actually show that Ext1 kK2 (S, S) = 0. Using this fact, the rest of the proof would show the assertions (5',6') in Remark 5.11. Now we continue the proof. There is a projective cover P b 1 −→ S so that ker f is a direct sum of finitely many copies of S2. Since P2 is the projective cover of S2, we have a minimal projective resolution f −→ P a 2 −→ P b 1 −→ S −→ 0 for some a, b. In the category Modf.d −kK2, we have a minimal projective resolution of S 0 −→ Sa 2 −→ P b 1 −→ S −→ 0 28 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU where S2 is a projective kK2-module. Hence we have a morphism of complexes −−−−→ P a 2 −−−−→ P b 1 −−−−→ S −−−−→ 0 y= y= y h f 0 −−−−→ Sa 2 −−−−→ P b 1 −−−−→ S −−−−→ 0 Applying HomA(−, S) to above, we obtain that · · · ←−−−−− HomA(P a 2 , S) ←−−−−− HomA(P b 1 , S) ←−−−−− HomA(S, S) ←−−−−− 0 g x   x   = x   = · · · ←−−−−− HomA(Sa 2 , S) ←−−−−− HomA(P b 1 , S) ←−−−−− HomA(S, S) ←−−−−− 0 Note that g is an isomorphism. Since dim Ext1 kK2 (S, S) ≤ 1, the cokernel of f has dimension at most 1. Since g is an isomorphism, the cokernel of h has dimension at most 1. This implies that Ext1 (cid:3) A(S, S) has dimension at most 1. Proposition 5.12. Let A be the category Modf.d. −A where A is in Example 5.5. As a consequence, fpd A = √2. (1) fpdn A =(√2 n = 2, (3) fpd A ≥ √2. (2) SI(A) = 2. n 6= 2. 1 Proof. (1) This is a consequence of Lemmas 5.9 and 5.10. Parts (2,3) follow from part (1). (cid:3) Remark 5.13. Let A be the algebra given in Example 5.5. We list some facts, comments and questions. (1) The algebra A is non-connected N-graded Koszul. (2) The minimal projective resolutions of S1 and S2 are ··· −→ P ⊕4 1 −→ P ⊕4 2 −→ P ⊕2 1 −→ P ⊕2 2 −→ P1 −→ S1 −→ 0, and ··· −→ P ⊕4 2 −→ P ⊕2 2 −→ P1 −→ P2 −→ S2 −→ 0. (3) For i ≥ 0, we have Exti 0 1 −→ P ⊕2 A(S1, S1) =(k⊕2i/2 A(S1, S2) =(0 A(S2, S2) =(k⊕2i/2 A(S2, S1) =(0 0 Exti Exti Exti k⊕2(i+1)/2 i is even is odd i is even is odd i is even is odd k⊕2(i−1)/2 i is even is odd. (4) One can check that every algebra of dimension 4 or less has either infinite or integral fpd. Hence, A is an algebra of smallest k-dimension that has finite FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 29 non-integral (or irrational) fpd. It is unknown if there is a finite dimensional algebra A such that fpd(Modf.d. −A) is transcendental. (5) Several authors have studied the connection between tame-wildness and complexity [BS, ES, Fa, FW, Kul, Ri]. The algebra A is probably the first explicit example of a tame algebra with infinite complexity. (6) It follows from part (3) that the fp-curvature of A := Db(Modf.d. −A) is √2 (some details are omitted). As a consequence, fpg(A) = ∞. By Theorem 8.3, the complexity of A is ∞. We don't know what fpd A is. 6. σ-decompositions We fix a category C and an endofunctor σ. For a set of bricks B in C (or a set of atomic objects when C is triangulated), we define fpdn B (σ) = sup{ρ(A(φ, σ)) φ := {X1,··· , Xn} ∈ Φn,b, and Xi ∈ B ∀i}. Let Λ := {λ} be a totally ordered set. We say a set of bricks B in C has a σ-decomposition {Bλ}λ∈Λ (based on Λ) if the following holds. (1) B is a disjoint unionSλ∈Λ Bλ. (2) If X ∈ Bλ and Y ∈ Bδ with λ < δ, HomC(X, σ(Y )) = 0. The following lemma is easy. Lemma 6.1. Let n be a positive integer. Suppose that B has a σ-decomposition {Bλ}λ∈Λ. Then fpdn B(σ) ≤ sup λ∈Λ,m≤n{fpdm Bλ(σ)}. Proof. Let φ be a brick set that is used in the computation of fpdn B(σ). Write (E6.1.1) where λi is strictly increasing and φλi = φ ∩ Bλi . For any objects X ∈ φλi and Y ∈ φλj , where λi < λj, by definition, HomC(X, σ(Y )) = 0. Listing the objects in φ in the order that suggested by (E6.1.1), then the adjacency matrix of (φ, σ) is of the form φ = φλ1 ∪ ··· ∪ φλs A(φ, σ) = A11 ∗ ∗ . . . ∗ 0 A22 ∗ . . . ∗ 0 0 A33 . . . ∗ ··· 0 ··· 0 ··· 0 . . . 0 ··· Ass  where each Aii is the adjacency matrix A(φλi , σ). By definition, where si is the size of Aii, which is no more than n. Therefore ρ(Aii) ≤ fpdsi Bλi (σ) ρ(A(φ, σ)) = max i {ρ(Aii)} ≤ sup λ∈Λ,m≤n{fpdm Bλ(σ)}. The assertion follows. (cid:3) We give some examples of σ-decompositions. Example 6.2. Let A be an abelian category and A be the derived category Db(A). Let [n] be the n-fold suspension Σn. 30 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU (1) Suppose that α is an endofunctor of A and α is the induced endofunctor of A. For each n ∈ Z, let Bn := {M [−n] M is a brick in A} and B :=Sn∈Z Bn. If Mi[−ni] ∈ Bni , for i = 1, 2, such that n1 < n2, then HomA(M1[−n1], α(M2[−n2])) = Extn1−n2 Then B has a α-decomposition {Bn}n∈Z based on Z. (2) Suppose g := gldim A < ∞. Let σ be the functor Σg ◦ α. For each n ∈ Z, let Bn := {M [n] M is a brick in A} and B :=Sn∈Z Bn. If Mi[ni] ∈ Bni, Then B has a σ-decomposition {Bn}n∈Z based on Z. HomA(M1[n1], σ(M2[n2])) = Extn2−n1+g for i = 1, 2, such that n1 < n2, then (M1, α(M2)) = 0. (M1, α(M2)) = 0. A A Example 6.3. Let C be a smooth projective curve and let A be the the category of coherent sheaves over C. Every coherent sheaf over C is a direct sum of a torsion subsheaf and a locally free subsheaf. Define B0 = {T is a torsion brick object in A}, B−1 = {F is a locally free brick object in A} and Let σ be the functor E1 := Ext1 B = B−1 ∪ B0. A(−,−). If F ∈ B−1 and T ∈ B0, then Ext1 A(F, T ) = 0. Hence, B has an E1-decomposition based on the totally ordered set Λ := {−1, 0}. The next example is given in [BB]. Example 6.4. Let C be an elliptic curve. Let A be the the category of coherent sheaves over C and A be the derived category Db(A). The slope of a coherent sheaf X 6= 0 [BB, Definition 4.6] is defined to be First we consider coherent sheaves. Let Λ be the totally ordered set Q ∪ {+∞}. µ(X) := χ(X) rk(X) ∈ Λ where χ(X) is the Euler characteristic of X and rk(X) is the rank of X. If X and Y are bricks such that µ(X) < µ(Y ), by [BB, Corollary 4.11], X and Y are semistable, and thus by [BB, Proposition 4.9(1)], HomA(Y, X) = 0. By Serre duality (namely, Calabi-Yau property), HomA(X, Y [1]) = Ext1 (E6.4.1) Write B = Φ1,b(A) and Bλ be the set of (semistable) bricks with slope λ. Then A(X, Y ) = 0 when X ∈ Bλ and Y ∈ Bν with B = Sλ∈Λ Bλ. By (E6.4.1), Ext1 λ < ν. Hence B has an E1-decomposition. By Lemma 6.1, for every n ≥ 1, A(X, Y ) = HomA(Y, X)∗ = 0. fpdn(E1) = fpdn B(E1) ≤ sup λ∈Λ,m≤n{fpdm Bλ(E1)}. Next we compute fpdn Bλ(E1). Let SSλ be the full subcategory of A con- sisting of semistable coherent sheaves of slope λ. By [BB, Summary], SSλ is an abelian category that is equivalent to SS∞. Therefore one only needs to compute fpdn B∞(E1) in the category SS∞. Note that SS∞ is the abelian category of tor- sion sheaves and every brick object in SS∞ is of the form Op for some p ∈ C. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 31 In this case, A(φ, E1) is the identity matrix. Consequently, ρ(A(φ, E1)) = 1. This shows that fpdn Bλ(E1) = fpdn B∞ (E1) = 1 for all n ≥ 1. It is clear that fpdn(E1) ≥ fpdn B∞(E1) = 1. Combining with Lemma 6.1, we obtain that fpdn(E1) = 1 for all n. (The above approach works for functors other than E1.) Finally we consider the fp-dimension for the derived category A. It follows from Theorem 3.5(3) that for all n ≥ 1. By definition, fpdn(Σ) = fpdn(E1) = 1 fpd(A) = fpd(A) = 1. As we explained before fpd is an indicator of the representation types of cate- gories. Drozd-Greuel studied a tame-wild dichotomy for vector bundles on projective curves [DrG] and introduced the notion of VB-finite, VB-tame and VB-wild sim- ilar to the corresponding notion in the representation theory of finite dimensional algebras. In [DrG] Drozd-Greuel showed the following: Let C be a connected smooth projective curve. Then (a) C is VB-finite if and only if C is P1. (b) C is VB-tame if and only if C is elliptic (that is, of genus 1). (c) C is VB-wild if and only if C has genus g ≥ 2. We now prove a fp-version of Drozd-Greuel's result [DrG, Theorem 1.6]. We thank Max Lieblich for providing ideas in the proof of Proposition 6.5(3) next. Proposition 6.5. Suppose k = C. Let X be a connected smooth projective curve and let g be the genus of X. (1) If g = 0 or X = P1, then fpd Db(coh(X)) = 1. (2) If g = 1 or X is an elliptic curve, then fpd Db(coh(X)) = 1. (3) If g ≥ 2, then fpd Db(coh(X)) = ∞. Proof. (1) The assertion follows from (E5.1.4). (2) The assertion follows from Example 6.4. (3) By Theorem 3.5(4), fpd(Db(coh(X))) = fpd(coh(X)). Hence it suffices to For each n, let {xi}n show that fpd(coh(X)) = ∞. i=1 be a set of n distinct points on X. By [DrG, Lemma 1.7], we might further assume that 2xi 6∼ xj + xk for all i 6= j, as divisors on X. Write Ei := O(xi) for all i. By [DrG, p.11], HomOX(Ei,Ej) = 0 for all i 6= j, which is also a consequence of a more general result [HL, Proposition 1.2.7]. It is clear that HomOX(Ei,Ei) = k for all i. Let φn be the set {E1, . . . ,En}. Then it is a brick set of non-isomorphic vector bundles on X (which are stable with rank(Ei) = deg(Ei) = 1 for all i). Define the sheaf Hij = Hom(Ei,Ej) for all i, j. Then deg(Hij ) = 0. By the Riemann-Roch Theorem, we have 0 = deg(Hij) = χ(Hij ) − rank(Hij )χ(OX) = dim HomOX(Ei,Ej) − dim Ext1 = δij − dim Ext1 OX(Ei,Ej) + (g − 1), OX(Ei,Ej) − (1 − g) 32 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Define the matrix An with entries aij := dim Ext1 which implies that dim Ext1 OX(Ei,Ej) = g − 1 + δij . This formula was also given in [DrG, p.11 before Lemma 1.7] when i 6= j. OX(Ei,Ej) = g − 1 + δij, which is the adjacency matrix of (φn, E1). This matrix has entries g along the diagonal and entries g − 1 everywhere else. Therefore the vector (1, . . . , 1) is an eigenvector for this matrix with eigenvalue n(g− 1) + 1. So ρ(An) ≥ n(g − 1) + 1 ≥ n + 1. Since we can define φn for arbitrarily large n, we must have fpd(coh(X)) = ∞. Question 6.6. Let X be a smooth irreducible projective curve of genus g ≥ 2. Is fpdn(X) finite for each n? If yes, do these invariants recover g? (cid:3) Proposition 6.7. Suppose k = C. Let Y be a smooth projective scheme of dimen- sion at least 2. Then fpd1(coh(Y)) = fpd(coh(Y)) = fpd1(Db(coh(Y))) = fpd(Db(coh(Y))) = ∞. Proof. It is clear that fpd1(coh(Y)) is smallest among these four invariants. suffices to show that fpd1(coh(Y)) = ∞. It is well-known that Y contains an irreducible projective curve X of arbitrarily large (either geometric or arithmetic) genus, see, for example, [CFZ, Theorem 0.1] or [Ch, Theorems 1 and 2]. Let OX be the coherent sheaf corresponding to the curve X and let g be the arithmetic genus of X. In the abelian category coh(X), we have It dim Ext1 OX(OX,OX) = dim H 1(X,OX) = g. Since coh(X) is a full subcategory of coh(Y), we have dim Ext1 OY(OX,OX) ≥ dim Ext1 OX(OX,OX) = g. By taking φ = {OX}, one sees that fpd1(coh(Y)) ≥ fpd1(coh(X)) ≥ g for all such X. Since g can be arbitrarily large, the assertion follows. (cid:3) 7. Representation types 7.1. Representation types. We first recall some known definitions and results. Definition 7.1. Let A be a finite dimensional algebra. (1) We say A is of finite representation type if there are only finitely many isomorphism classes of finite dimensional indecomposable left A-modules. (2) We say A is tame or of tame representation type if it is not of finite represen- tation type, and for every n ∈ N, all but finitely many isomorphism classes of n-dimensional indecomposables occur in a finite number of one-parameter families. (3) We say A is wild or of wild representation type if, for every finite dimensional k-algebra B, the representation theory of B can be embedded into that of A. The following is the famous trichotomy result due to Drozd [Dr1]. Theorem 7.2 (Drozd's trichotomy theorem). Every finite dimensional algebra is either of finite, tame, or wild representation type. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 33 Remark 7.3. (1) An equivalent and more precise definition of a wild algebra is the following. An algebra A is called wild if there is a faithful exact embedding of abelian categories (E7.3.1) Emb : Modf.d. −khx, yi → Modf.d. −A that respects indecomposability and isomorphism classes. (2) A stronger notion of wildness is the following. An algebra A is called strictly wild, also called fully wild, if Emb in (E7.3.1) is a fully faithful embedding. (3) It is clear that strictly wild is wild. The converse is not true. We collect some celebrated results in terms of representation types of path alge- bras [Ga1] and [Na, DoF]. Theorem 7.4. Let Q be a finite connected quiver. (1) [Ga1] The path algebra kQ is of finite representation type if and only if the underlying graph of Q is a Dynkin diagram of type ADE. (2) [Na, DoF] The path algebra kQ is of tame representation type if and only if the underlying graph of Q is an extended Dynkin diagram of type eAeDeE. Our main goal in this section is to prove Theorem 0.3. We thank Klaus Bongartz for suggesting the following lemma. Lemma 7.5. [B1] Let A be a finite dimensional algebra that is strictly wild. Then, for each integer a > 0, there is a finite dimensional brick left A-module N such that dim Ext1 Proof. Let V be the vector spaceLa i=1 kxi and let B be the finite dimensional al- gebra khV i/(V ⊗2). By [B2, Theorem 2(i)], there is a fully faithful exact embedding A(N, N ) ≥ a. Since A is strictly wild, there is a fully faithful exact embedding Modf.d. −B −→ Modf.d. −khx, yi. Hence we have a fully faithful exact embedding Modf.d. −khx, yi −→ Modf.d. −A. (E7.5.1) F : Modf.d. −B −→ Modf.d. −A. Let S be the trivial B-module B/B≥1. dim Ext1 injection It follows from an easy calculation that B(S, S) = dim(V )∗ = a. Since F is fully faithful exact, F induces an F : Ext1 B(S, S) −→ Ext1 A(F (S), F (S)). Thus dim Ext1 brick. The assertion follows by taking N = F (S). A(F (S), F (S)) ≥ a. Since S is simple, it is a brick. Hence, F (S) is a (cid:3) Proposition 7.6. (1) Let A be a finite dimensional algebra that is strictly wild, then (2) If A := kQ is wild, then fpd1(E1) = fpd(A) = fpd(A) = ∞. fpd1(E1) = fpd(A) = fpd(A) = ∞. 34 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Proof. (1) For each integer a, by Lemma 7.5, there is a brick N in A such that A(N, N ) ≥ a. Hence fpd1(E1) ≥ a. Since a is arbitrary, fpd1(E1) = ∞. Ext1 Consequently, fpd(A) = ∞. By Lemma 2.9, fpd(A) = ∞. (2) It is well-known that a wild path algebra is strictly wild, see a comment of Gabriel [Ga2, p.149] or [Ar, Proposition 7]. The assertion follows from part (1). (cid:3) The following lemma is based on a well-understood AR-quiver theory for acyclic quivers of finite representation type and the hammock theory introduced by Brenner [Bre]. We refer to [RV] if the reader is interested in a more abstract version of the hammock theory. For a class of quivers including all ADE quivers, there is a convenient (though not essential) way of positioning the vertices as in [AS, Example IV.2.6]. A quiver Q is called well-positioned if the vertices of Q are located so that all arrows are strictly from the right to the left of the same horizontal distance. For example, the following quiver Dn is well-positioned: 1 2 ··· n − 2 n − 1 n Lemma 7.7. Let Q be a quiver such that (i) the underlying graph of Q is a Dynkin diagram of type A, or D, or E, and that (ii) Q is well-positioned. Let A = kQ and let M, N be two indecomposable left A-modules in the AR-quiver of A. Then the following hold. (1) There is a standard way of defining the order or degree for indecomposable left A-modules M , denoted by deg M , such that all arrows in the the AR- quiver have degree 1, or equivalently, all arrows are from the left to the right of the same horizontal distance. As in (E5.4.2), when Q = A2, deg P2 = 0, deg P1 = 1 and deg I1 = 2. (2) If HomA(M, N ) 6= 0, then deg M ≤ deg N . (3) The degree of the AR-translation τ is −2. (4) If Ext1 (5) There is no oriented cycle in the E1-quiver of A := Modf.d. −kQ, denoted (6) fpd(A) = 0. A(M, N ) 6= 0, then deg M ≥ deg N + 2. by QE1 A , defined before Lemma 2.10. Proof. (1) This is a well-known fact in AR-quiver theory. For each given quiver Q as described in (i,ii), one can build the AR-quiver by using the Auslander-Reiten translation τ and the Knitting Algorithm, see [Sc1, Ch. 3]. Some explicit examples are given in [Ga3, Ch. 6] and [Sc1, Ch. 3]. (2) This follows from (1). Note that the precise dimension of HomA(M, N ) can be computed by using hammock theory [Bre, RV]. Some examples are given in [Sc1, Ch. 3]. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 35 (3) This follows from the definition of the translation τ in the AR-quiver theory [ARS, VII]. See also, [Cr, Remarks (2), p. 23]. (4) By Serre duality, Ext1 R(M, N ) = HomA(N, τ M )∗ [RVdB, Proposition I.2.3] or [Cr, Lemma 1, p. 22]. If Ext1 R(M, N ) 6= 0, then, by Serre duality and part (2), deg N ≤ deg τ M = deg M − 2. Hence deg M ≥ deg N + 2. (5) In this case, every indecomposable module is a brick. Hence the E1-quiver QE1 A has the same vertices as the AR-quiver. By part (4), if there is an arrow from M to N in the quiver QE1 A , then deg M ≥ deg N + 2. This means that all arrows A are from the right to the left. Therefore there is no oriented cycle in QE1 in QE1 A . (cid:3) (6) This follows from part (5), Theorem 1.8(1) and Lemma 2.10. Theorem 7.8. Let Q be a finite quiver whose underlying graph is a Dynkin diagram of type ADE and let A = kQ. Then fpd(A) = fpd(A) = fpgldim(A) = 0. Proof. Since the path algebra A is hereditary, A is a-hereditary of global dimension 1. By Theorem 3.5(3), fpd(A) = fpd(A). If Q1 and Q2 are two quivers whose underlying graphs are the same, then, by Bernstein-Gelfand-Ponomarev (BGP) re- flection functors [BGP], Db(Modf.d. −kQ1) and Db(Modf.d. −kQ2) are triangulated equivalent. Hence we only need prove the statement for one representative. Now we can assume that Q satisfies the hypotheses (i,ii) of Lemma 7.7. By Lemma 7.7(6), fpd(A) = 0. Therefore fpd(A) = 0, or equivalently, fpd(Σ) = 0. By Theorem 3.5(1), fpd(Σi) = 0 for all i 6= 0, 1. Therefore fpgldim(A) = 0. 7.2. Weighted projective lines. To prove Theorem 0.3, it remains to show part (2) of the theorem. Our proof uses a result of [CG] about weighted projective lines, which we now review. Details can be found in [GL, Section 1]. (cid:3) For t ≥ 1, let p := (p0, p1,··· , pt) be a (t + 1)-tuple of positive integers, called the weight sequence. Let D := (λ0, λ1,··· , λt) be a sequence of distinct points of the projective line P1 over k. We normalize D so that λ0 = ∞, λ1 = 0 and λ2 = 1 (if t ≥ 2). Let S := k[X0, X1,··· , Xt]/(X pi i − X p1 1 + λiX p0 0 , i = 2,··· , t). The image of Xi in S is denoted by xi for all i. Let L be the abelian group of rank 1 generated by −→xi for i = 0, 1,··· , t and subject to the relations p0−→x0 = ··· = pi−→xi = ··· = pt−→xt =: −→c . The algebra S is L-graded by setting deg xi = −→xi . The corresponding weighted projective line, denoted by X(p, D) or simply X, is a noncommutative space whose category of coherent sheaves is given by the quotient category coh(X) := grL −S grL f.d. −S where grL −S is the category of noetherian L-graded left S-modules and grL is the full subcategory of grL −S consisting of finite dimensional modules. The weighted projective lines are classified into the following three classes: f.d. −S domestic tubular wild if p is (p, q), (2, 2, n), (2, 3, 3), (2, 3, 4), (2, 3, 5); if p is (2, 3, 6), (3, 3, 3), (2, 4, 4), (2, 2, 2, 2); otherwise. X is  Db(coh(X)) ∼=  Db(Modf.d. −keAp,q) Db(Modf.d. −keDn) Db(Modf.d. −keE6) Db(Modf.d. −keE7) Db(Modf.d. −keE8) if p = (p, q), if p = (2, 2, n), if p = (2, 3, 3), if p = (2, 3, 4), if p = (2, 3, 5). X(M, M ) = 1. 36 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU Let X be a weighted projective curve. Let V ect(X) be the full subcategory of coh(X) consisting of all vector bundles. Similar to the elliptic curve case [Example 6.4], one can define the concepts of degree, rank and slope of a vector bundle on a weighted projective curve X, see [LM, Section 2] for details. For each µ ∈ Q∪{∞}, let V ectµ(X) be the full subcategory of V ect(X) consisting of all vector bundles of slope µ. Lemma 7.9. Let X = X(p, D) be a weighted projective line. (1) coh(X) is noetherian and hereditary. (2) (3) Let M be a generic simple object in coh(X). Then Ext1 (4) fpd1(coh(X)) ≥ 1. (5) If X is tubular or domestic, then Ext1 (6) If X is domestic, then Ext1 X(X, Y ) = 0 for all X ∈ V ectµ′ (X) and Y ∈ V ectµ(X) with µ′ < µ. X(X, Y ) = 0 for all X ∈ V ectµ′(X) and Y ∈ V ectµ(X) with µ′ ≤ µ. As a consequence, fpd(Σ V ectµ′ (X)) = 0 for all µ < ∞. (7) Suppose X is tubular. Then every indecomposable vector bundle X is semi- stable. (8) Suppose X is tubular and let µ ∈ Q. Then each V ectµ(X) is a uniserial cate- gory. Accordingly indecomposables in V ectµ(X) decomposes into Auslander- Reiten components, which all are tubes of finite rank. Proof. (1) This is well-known. (2) [GL, 5.4.1]. (3) Let M be a generic simple object. Then M is a brick and Ext1(M, M ) = 1. (4) Follows from (3) by taking φ := {M}. (5) This is [Sc2, Corollary 4.34(i)] since tubular is also called elliptic in [Sc2]. (6) This is [Sc2, Comments after Corollary 4.34] since domestic is also called parabolic in [Sc2]. The consequence is clear. (7) [GL, Theorem 5.6(i)]. (8) [GL, Theorem 5.6(iii)]. (cid:3) We will use the following result which is proved in [CG]. Theorem 7.10. [CG] Let X be a weighted projective line. (1) If X is domestic, then fpd Db(coh(X)) = 1. (2) If X is tubular, then fpd Db(coh(X)) = 1. (3) If X is wild, then fpd Db(coh(X)) ≥ dim HomX(OX,OX(−→ω )) where −→ω is the dualizing element [GL, Sec. 1.2]. There is a similar statement for smooth complex projective curves [Proposition 6.5]. The authors are interested in answering the following question. Question 7.11. Let X be a wild weighted projective line. What is the exact value of fpdn Db(coh(X))? FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 37 7.3. Tubes. The following example is studied in [CG], which is dependent on direct linear algebra calculations. Example 7.12. [CG] Let ξ be a primitive nth root of unity. Let Tn be the algebra khg, xi . Tn := (gn − 1, xg − ξgx) This algebra can be expressed by using a group action. Let G be the group {g gn = 1} ∼= Z/(n) acting on the polynomial ring k[x] by g · x = ξx. Then Tn is naturally isomor- phic to the skew group ring k[x] ∗ G. Let −−−→An−1 denote the cycle quiver with n vertices, namely, the quiver with one oriented cycle connecting n vertices. It is also known that Tn is isomorphic to the path algebra of the quiver −−−→An−1. Then fpd(Modf.d. −Tn) = 1 by [CG]. 7.4. Proof of Theorem 0.3. Proof of Theorem 0.3. Part (1) follows from Theorems 7.4(1) and 7.8. Part (3) follows from Proposition 7.6(2). It remains to deal with part (2). (2) By Theorem 7.4(2), Q must be of type either −−−→An−1, or eAp,q, or eDn, or eE6,7,8. If Q is of type −−−→An−1, the assertion follows from Example 7.12. If Q is of type eAp,q, eDn, or eE6,7,8, the assertion follows from Lemma 7.9(2) and Theorem 7.10(1). (cid:3) 8. Complexity The concept of complexity was first introduced by Alperin-Evens in the study of group cohomology in 1981 [AE]. Since then the study of complexity has been extended to finite dimensional algebras, Frobenius algebras, Hopf algebras and commutative algebras. First we recall the classical definition of the complexity for finite dimensional algebras and then give a definition of the complexity for triangulated categories. We give the following modified (but equivalent) version, which can be generalized. Definition 8.1. Let A be a finite dimensional algebra and T = A/J(A) where J(A) is the Jacobson radical of A. Let M be a finite dimensional left A-module. (1) The complexity of M is defined to be cx(M ) := lim sup n→∞ logn(dim Extn A(M, T )) + 1. (2) The complexity of the algebra A is defined to be cx(A) := cx(T ). In the original definition of complexity by Alperin-Evens [AE] and in most of other papers, the dimension of n-syzygies is used instead of the dimension of the Extn-groups, but it is easy to see that the asymptotic behavior of these two series are the same, therefore these give rise to the same complexity. It is well-known that cx(M ) ≤ cx(A) for all finite dimensional left A-modules M . Next we introduce the notion of a complexity for a triangulated category which is partially motivated by the work in [BHZ, Section 4]. Definition 8.2. Let T be a pre-triangulated category. Let d be a real number. 38 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU (1) The left subcategory of complexity less than d is defined to be (2) The right subcategory of complexity less than d is defined to be dT :=(cid:26)X ∈ T Td :=(cid:26)X ∈ T lim n→∞ dim HomT (X, Σn(Y )) nd−1 lim n→∞ dim HomT (Y, Σn(X)) nd−1 = 0,∀ Y ∈ T(cid:27) . = 0,∀ Y ∈ T(cid:27) . (3) The complexity of T is defined to be (4) The Frobenius-Perron complexity of T is defined to be cx(T ) := inf {d dT = T } fpcx(T ) := fpg(Σ) + 1. Note that it is not hard to show that cx(T ) = inf {d Td = T } . Theorem 8.3. Let T be a pre-triangulated category. Then fpcx(T ) ≤ cx(T ). Proof. Let d be any number strictly larger than cx(T ). We need to show that fpcx(T ) ≤ d. Let φ ∈ Φm,a be an atomic set and let X :=LXi∈φ Xi. Then, by definition, dim HomT (X, Σn(X)) lim n→∞ nd−1 = 0. Then there is a constant C such that dim HomT (X, Σn(X)) < Cnd−1 for all n > 0. Since each Xi is a direct summand of X, we have aij (n) := dim HomT (Xi, Σn(Xj)) < Cnd−1 for all i, j. This means that each entry aij (n) in the adjacency matrix of A(φ, Σn) is less than Cnd−1. Therefore ρ(A(φ, Σn)) < mCnd−1. By Definition 2.3(3), fpg(Σ) ≤ d − 1. Thus fpcx(T ) ≤ d as desired. (cid:3) We will prove that the equality fpcx(T ) = cx(T ) holds under some extra hy- potheses. Let A be a finite dimensional algebra with a complete list of simple left A-modules {S1,··· , Sw}. We use n for any integer and i, j for integers between 1 and w. Define, for i ≤ j, pij(n) := min{dim Extn A(Si, Sj), dim Extn A(Sj, Si)} and Pn := max{pij(n) i ≤ j}. We say A satisfies averaging growth condition (or AGC for short) if there are positive integers C and d, independent of the choices of n and (i, j), such that A(Si, Sj) ≤ C max{Pn−d, Pn−d+1,··· , Pn+d} dim Extn (E8.3.1) for all n and all 1 ≤ i, j ≤ w. Theorem 8.4. Let A be a finite dimensional algebra and A = Db(Modf.d. −A). (1) cx(A) = cx(A). As a consequence, cx(A) is a derived invariant. (2) If A satisfies AGC, then fpcx(A) = cx(A) = cx(A). As a consequence, if A is local or commutative, then fpcx(A) = cx(A) = cx(A). FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 39 We will prove Theorem 8.4 after the next lemma. Let T be a pre-triangulated category with suspension Σ. We use X, Y, Z for objects in T . Fix a family φ of objects in T and a positive number d. Define (E8.4.1) (E8.4.2) (E8.4.3) (E8.4.4) d(φ) =(cid:26)X ∈ T (φ)d =(cid:26)X ∈ T d(φ) =(X ∈ T (φ)d =(X ∈ T lim n→∞ lim n→∞ dim HomT (X, Σn(Y )) nd−1 dim HomT (Y, Σn(X)) nd−1 = 0,∀ Y ∈ φ(cid:27) , = 0,∀ Y ∈ φ(cid:27) , lim n→∞Pi≤n dim HomT (X, Σi(Y )) n→∞Pi≤n dim HomT (Y, Σi(X)) lim nd nd = 0,∀ Y ∈ φ) , = 0,∀ Y ∈ φ) . Lemma 8.5. The following are full thick pre-triangulated subcategories of T closed under direct summands. (1) d(φ). (2) (φ)d. (3) d(φ). (4) (φ)d. Proof. We only prove (1). The proofs of other parts are similar. Suppose X ∈ d(φ). Using the fact limn→∞ (n+1)d−1 = 1, we see that X[1] = Σ(X) is in d(φ). Similarly, X[−1] is in d(φ). If f : X1 → X2 be a morphism of objects in d(φ), and let X3 be the mapping cone of f , then, for each Y ∈ φ, we have an exact sequence nd−1 → HomT (X1, Σn−1(Y )) → HomT (X3, Σn(Y )) → HomT (X2, Σn(Y )) → which implies that X3 ∈ d(φ). Therefore d(φ) is a thick pre-triangulated subcate- gory of T . If X ∈ d(φ) and X = Y ⊕ Z, it is clear that Y, Z ∈ d(φ). Therefore d(φ) is closed under taking direct summands. (cid:3) Proof of Theorem 8.4. (1) Let c = cx(A). For every d < c, we have that lim sup n→∞ dim Extn nd−1 A(T, T ) = ∞ which implies that T 6∈ dA. Therefore d ≤ cx(A). Conversely, let d > c. It follows from the definition that lim sup n→∞ dim Extn nd−1 A(T, T ) = 0. This means that T ∈ ({T})d. Since T generates A, we have A = ({T})d. Again, since T generates A, we have A = Ad = dA. By definition, d ≥ cx(A) as desired. (2) Assume that A satisfies AGC. Let c1 = fpcx(A), c2 = lim sup n→∞ c3 = lim sup n→∞ logn(C max{Pn−d, Pn−d+1,··· , Pn+d}) + 1, logn(Pn) + 1, c4 = cx(A) = cx(A). 40 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU By calculus, we have c2 = c3. Let φ be the atomic set of simple objects {Si}w i=1. Then ρ(φ, Σn) ≥ pij(n), for all i, j, by Lemma 1.7(2). So ρ(φ, Σn) ≥ Pn. As a consequence, c1 ≥ c3. Let T = A/J =Lw i=1 Sdi for some finite numbers {di}w i=1. Let D be maxi{di}. By AGC, namely, (E8.3.1), didj dim Extn dim Extn A(Si, Sj) i A(T, T ) =Xi,j ≤ w2DC max{Pn−d, Pn−d+1,··· , Pn+d} which implies that c4 = cx(A) = cx(T ) ≤ c2. Combining with Theorem 8.3, we have c1 = c2 = c3 = c4 as desired. If A is local, then there is only one simple module S1. Then (E8.3.1) is automatic. A(Si, Sj) = 0 for all n and all i 6= j. Again, in this If A is commutative, then Exti case, (E8.3.1) is obvious. The consequence follows from the main assertion. (cid:3) For all well-studied finite dimensional algebras A, (E8.3.1) holds. For example, the algebra A in Example 5.5 satisfies AGC. This can be shown by using the computation given in Remark 5.13(3). It is natural to ask if every finite dimensional algebra satisfies AGC. Lemma 8.6. (1) Let A be an abelian category and A = Db(A). If gldim A < (2) Let T be a pre-triangulated category. If fpgldimT < ∞, then fpcx(T ) = 0. ∞, then fpcx(A) = 0. Proof. Both are easy and proofs are omitted. (cid:3) We conclude with examples of non-integral fpg of a triangulated category. Example 8.7. (1) Let α be any real number in {0}∪{1}∪[2,∞). By [KL, Theorem 1.8, or p. 14], there is a finitely generated algebra R with GKdim R = α. More precisely, [KL, Theorem 1.8] implies that there is a 2-dimensional vector space V ⊂ R that generates R such that, there are positive integers a < b, for every n > 0, anα < dim(k1 + V )n < bnα. Define a filtration F on R by FiR = (k1 + V )i, ∀ i. Let A be the associated graded algebra gr R with respect to this grading. Then A is connected graded and generated by two elements in degree 1 and satisfying, for every n > 0, (E8.7.1) anα < dim Ai < bnα. nXi=0 To match up with the definition of complexity, we further assume that there are c < d such that, for every n > 0, (E8.7.2) cnα−1 < dim An < dnα−1. This can be achieved, for example, by replacing A by its polynomial extension A[t] (with deg t = 1) and replacing α by α + 1. Next we make A a differential graded (dg) algebra by setting elements in Ai to have cohomological degree i and dA = 0. For this dg algebra, we denote the FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 41 derived category of left dg A-modules by A. Let O be the object AA in A. By the definition of the cohomological degree of A, we have HomA(O, ΣiO) = Ai, ∀ i. Let T be the full triangulated subcategory of A generated by O. (E8.7.3) implies that O is an atomic object. Now using (E8.7.3) together with (E8.7.2), we obtain that (E8.7.3) (E8.7.4) fpcx(T ) ≥ α. By (E8.7.2)-(E8.7.3), we have that, for every d > α, O ∈ d({O}). Since O generates T , we have d({O}) = T . The last equation means that O ∈ (T )d. Since O generates T , we have (T )d = T . By definition, d > cx(T ). Combining these with Theorem 8.3 and (E8.7.4), we have, for every d > α, α ≤ fpcx(T ) ≤ cx(T ) < d which implies that fpcx(T ) = cx(T ) = α. This construction implies that (E8.7.5) HomT (O, Σi(O))! = GKdim A = α. GKdim ∞Mi=0 logical degree i. (2) We now consider an extreme case. Let a := {ai}∞ non-negative integers with a0 = 1. Define B to be the dg algebraL Bi such that (i) dim Bi = ai for all i. In particular, B0 = k. Elements in Bi have cohomo- (ii) (Li>0 Bi)2 = 0. i=0 be any sequence of In this case, GKdim B = 0. Similar to part (1), the derived category of left dg B-modules is denoted by B. Let O be the object BB in B. Then (iii) Differential dB = 0. HomB(O, ΣiO) = Bi, ∀ i, and O is an atomic object. Let T be the full triangulated subcategory of B generated by O. The argument in part (1) shows that fpcx(T ) = lim sup Now let r be any real number ≥ 1 and let ai = (1 fpcx(T ) = r. Let r be any real number ≥ 1 and ai = ⌊ri⌋ for all i ≥ 0. Then fpcx(T ) =(1 . Using a similar method (with details omitted), fpv(T ) = r. . Then we have i = 0 i ≥ 1 logn(an) + 1. ⌊ir−1⌋ n→∞ r = 1 ∞ r > 1 Acknowledgments. The authors would like to thank Klaus Bongartz, Christof Geiss, Ken Goodearl, Claus Michael Ringel, and Birge Huisgen-Zimmermann for many useful conversations on the subject, and thank Max Lieblich for the proof of Proposition 6.5(3). J. Chen was partially supported by the National Natu- ral Science Foundation of China (Grant No. 11571286) and the Natural Science Foundation of Fujian Province of China (Grant No. 2016J01031). Z. Gao was partially supported by the National Natural Science Foundation of China (Grant No. 61401381). E. Wicks and J.J. Zhang were partially supported by the US 42 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU National Science Foundation (Grant Nos. DMS-1402863 and DMS-1700825). X.- H. Zhang was partially supported by the National Natural Science Foundation of China (Grant No. 11401328). H. Zhu was partially supported by a grant from Jiangsu overseas Research and Training Program for university prominent young and middle-aged Teachers and Presidents, China. References [AE] J. Alperin and L. Evens, Representations, resolutions, and Quillen's dimension theorem, J. Pure Appl. Algebra 22 (1981), 1-9. [Ar] S. Ariki, Hecke algebras of classical type and their representation type, Proc. London Math. Soc. (3) 91 (2005), no. 2, 355 -- 413. [AZ] M. Artin, J.J. Zhang, Noncommutative projective schemes, Adv. Math. 109 (1994), 228 -- 287. [AS] I. Assem, D. Simson and A. Skowroski, Elements of the representation theory of associative algebras, Vol. 1. Techniques of representation theory. London Mathematical Society Student Texts, 65. Cambridge University Press, Cambridge, 2006. [ARS] M. Auslander, I. Reiten and S.O. Smalo, Representation theory of Artin algebras, Corrected reprint of the 1995 original. Cambridge Studies in Advanced Mathematics, 36 Cambridge University Press, Cambridge, 1997. [Av] L.L. Avramov, Infinite free resolutions, in: Six Lectures on Commutative Algebra, in: Prog. Math., vol. 166, Birkhauser Verlag, Basel, Boston, Berlin, 1998, 1-118. [BHZ] Y.-H. Bao, J.-W. He and J.J. Zhang, Pertinency of Hopf actions and quotient categories of Cohen-Macaulay algebras, J. Noncommut. Geom. (accepted for publication). [BZ] J. Bell and J.J. Zhang, An isomorphism lemma for graded rings, Proc. Amer. Math. Soc., 145 (2017), no. 3, 989 -- 994 [Bei] A. A. Beilinson, Coherent sheaves on Pn and problems in linear algebra, Funktsional. Anal. i Prilozhen. 12(3) (1978), 68 -- 69. [Ben] D.J. Benson, Representations and cohomology I. Basic representation theory of finite groups and associative algebras. Cambridge Studies in Advanced Mathematics, 30. Cambridge Uni- versity Press, Cambridge, 1991. [BS] P.A. Bergh and Ø. Solberg, Relative support varieties, Q. J. Math. 61 (2010), no. 2, 171 -- 182. [BGP] I.N. Bernstein, I.M. Gel'fand and V.A. Ponomarev, Coxeter functors and Gabriel's theo- rem, Russ. Math. Surv. 28 (1973), no. 2, 17 -- 32 (English). [BO1] A. Bondal and D. Orlov, Reconstruction of a variety from the derived category and groups of autoequivalences, Compositio Math. 125 (2001), no. 3, 327 -- 344. [BO2] A. Bondal and D. Orlov, Derived categories of coherent sheaves, In Proceedings of the International Congress of Mathematicians, Vol. II (Beijing, 2002), pages 47 -- 56. Higher Ed. Press, Beijing, 2002. [B1] K. Bongartz, priviate communications. [B2] K. Bongartz, Representation embeddings and the second Brauer-Thrall conjecture, preprint, (2017), arXiv:1611.02017. [Bre] S. Brenner, A combinatorial characterization of finite Auslander-Reiten quivers, Proceed- ings ICRA 4, Ottawa 1984, Lecture Notes in Mathematics 1177 (Springer, Berlin, 1986), pp. 13 -- 49. [Bri] T. Bridgeland, Flops and derived categories, Invent. Math., 147(3), (2002), 613 -- 632. [BKR] T. Bridgeland, A. King, and M. Reid, The MacKay correspondence as an equivalence of derived categories, J. Amer. Math. Soc., 14(3) (2001), 535 -- 554. [BB] K. Bruning and I. Burban, Coherent sheaves on an elliptic curve, (English summary) Inter- actions between homotopy theory and algebra, 297 -- 315, Contemp. Math., 436, Amer. Math. Soc., Providence, RI, 2007. [Ca] J.F. Carlson, The decomposition of the trivial module in the complexity quotient category, J. Pure Appl. Algebra 106 (1996), 23 -- 44. [CDW] J.F. Carlson, P. W. Donovan and W. W. Wheeler, Complexity and quotient categories for group algebras, J. Pure Appl. Algebra 93 (1994), 147 -- 167. [CC] A.T. Carroll and C. Chindris, Moduli spaces of modules of Schur-tame algebras, (English summary) Algebr. Represent. Theory 18 (2015), no. 4, 961 -- 976. FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 43 [Ch] J.A. Chen, On genera of smooth curves in higher-dimensional varieties, Proc. Amer. Math. Soc. 125 (1997), no. 8, 2221 -- 2225. [CG] J.M. Chen, Z.B. Gao, E. Wicks, J. J. Zhang, X-.H. Zhang and H. Zhu, Frobenius-Perron theory for projective schemes, ArXiv e-prints, (2019). arXiv:1907.02221. [CKW] C. Chindris, R. Kinser and J. Weyman, Module varieties and representation type of finite- dimensional algebras, Int. Math. Res. Not. IMRN (2015), no. 3, 631 -- 650. [CFZ] C. Ciliberto, F. Flamini and M. Zaidenberg, Gaps for geometric genera, Arch. Math. (Basel) 106 (2016), no. 6, 531 -- 541. [Cr] W. Crawley-Boevey, Lectures on representations of quivers, Lectures in Oxford in 1992, available at www.amsta.leeds.ac.uk/ pmtwc/quivlecs.pdf. [DoG] M.A. Dokuchaev, N.M. Gubareni, V.M. Futorny, M.A. Khibina and V.V. Kirichenko, Dynkin diagrams and spectra of graphs, Sao Paulo J. Math. Sci., 7 (2013), no. 1, 83 -- 104. [DoF] P. Donovan and M.R. Freislich, The representation theory of finite graphs and associated algebras, Number 5 in Carleton Mathematical Lecture Notes. Carleton University, Ottawa, Ont., 1973. [Dr1] J.A. Drozd, Tame and wild matrix problems, Representation theory, II (Proc. Second Inter- nat. Conf., Carleton Univ., Ottawa, Ont., 1979), Lecture Notes in Math., vol. 832, Springer, Berlin, 1980, pp. 242 -- 258. [Dr2] Y.A. Drozd, Derived tame and derived wild algebras, Algebra Discrete Math. (2004), no. 1, 57 -- 74. [DrG] Y.A. Drozd and G.-M. Greuel, Tame and wild projective curves and classification of vector bundles, J. Algebra 246 (2001), no. 1, 1 -- 54. [ES] K. Erdmann and Ø. Solberg, Radical cube zero selfinjective algebras of finite complexity, J. Pure Appl. Algebra 215 (2011), no. 7, 1747 -- 1768. [EG] P. Etingof, S. Gelaki, D. Nikshych and V. Ostrik, Tensor categories, Mathematical Surveys and Monographs, 205. American Mathematical Society, Providence, RI, 2015. [EGO] P. Etingof, S. Gelaki and V. Ostrik, Classification of fusion categories of dimension pq, Int. Math. Res. Not., 2004, no. 57, 3041 -- 3056. [ENO] P. Etingof, D. Nikshych and V. Ostrik, On fusion categories, Ann. of Math., (2) 162 (2005), no. 2, 581 -- 642. [Fa] R. Farnsteiner, Tameness and complexity of finite group schemes, Bulletin of the London Mathematical Society, 39 (2007) no. 1, 63 -- 70. [FW] J. Feldvoss and S. Witherspoon, Support varieties and representation type of self-injective algebras, Homology Homotopy Appl. 13 (2011), no. 2, 197 -- 215. [Ga1] P. Gabriel, Unzerlegbare Darstellungen. I, (German. English summary) Manuscripta Math. 6 (1972), 71 -- 103; correction, ibid. 6 (1972), 309. [Ga2] P. Gabriel, Repr´esentations ind´ecomposables, (French) S´eminaire Bourbaki, 26e ann´ee (1973/1974), Exp. No. 444, pp. 143 -- 169. Lecture Notes in Math., Vol. 431, Springer, Berlin, 1975. [Ga3] P. Gabriel, Auslander-Reiten sequences and representation-finite algebras, Representation theory, I (Proc. Workshop, Carleton Univ., Ottawa, Ont., 1979), pp. 1 -- 71, Lecture Notes in Math., 831, Springer, Berlin, 1980. [GL] W. Geigle and H. Lenzing, A class of weighted projective curves arising in representation theory of finite-dimensional algebras, Singularities, representation of algebras, and vector bundles (Lambrecht, 1985), 265 -- 297, Lecture Notes in Math., 1273, Springer, Berlin, 1987. [GKO] C. Geiss, B. Keller and S. Oppermann, nangulated categories, J. Reine Angew. Math. 675 (2013) 101 -- 120. [GKr] C. Geiss and H. Krause, On the notion of derived tameness, (English summary) J. Algebra Appl. 1 (2002), no. 2, 13 -- 157. [GLW] J.-Y. Guo, A. Li, Q. Wu, Selfinjective Koszul algebras of finite complexity, Acta Math. Sinica, English Series 25 (2009), 2179 -- 2198. [HPR] D. Happel, U. Preiser, and C.M. Ringel, Binary polyhedral groups and Euclidean diagrams. Manuscripta Math., 31 (1-3), (1980), 317 -- 329. [HL] D. Huybrechts and M. Lehn, The geometry of moduli spaces of sheaves, Second edition. Cambridge Mathematical Library. Cambridge University Press, Cambridge, 2010. [Ke1] B. Keller, Derived categories and tilting, Handbook of tilting theory, 49 -- 104, London Math. Soc. Lecture Note Ser., 332, Cambridge Univ. Press, Cambridge, 2007. 44 J.M.CHEN, Z.B. GAO, E. WICKS, J. J. ZHANG, X-.H. ZHANG AND H. ZHU [Ke2] B. Keller, Calabi-Yau triangulated categories. Trends in representation theory of algebras and related topics, 467 -- 489, EMS Ser. Congr. Rep., Eur. Math. Soc., Zrich, 2008. [KV] B. Keller and D. Vossieck Sous les cat´egories driv´ees (French) [Beneath the derived cate- gories] C. R. Acad. Sci. Paris S´er. I Math. 305 (1987), no. 6, 225 -- 228. [KL] G.R. Krause and T.H. Lenagan, Growth of algebras and Gelfand-Kirillov dimension, Re- search Notes in Mathematics, Pitman Adv. Publ. Program, 116 (1985). [Kul] J. Kulshammer, Representation type of Frobenius-Lusztig kernels, Q. J. Math. 64 (2013), no. 2, 471 -- 488. Corrigendum "Representation type of Frobenius-Lusztig kernels" Q. J. Math. 66 (2015), no. 4, 1139. [LM] H. Lenzing, H. Meltzer, Sheaves on a weighted projective line of genus one, and representa- tions of a tubular algebra, Representations of algebras (Ottawa, Canada, 1992), CMS Conf. Proc. 14, 313 -- 337, (1994). [MR] J.C. McConnell and J.C . Robson, "Noncommutative Noetherian Rings," Wiley, Chichester, 1987. [Ni] D. Nikshych, Semisimple weak Hopf algebras, J. Algebra, 275 (2004), no. 2, 639 -- 667. [Na] L.A. Nazarova, Representations of quivers of infinite type, Izvestiya Akademii Nauk SSSR. Seriya Matematicheskaya, 37 (1973), 752 -- 791. [RVdB] I. Reiten and M. Van den Bergh, Noetherian hereditary abelian categories satisfying Serre duality, J. Amer. Math. Soc. 15 (2002), no. 2, 295 -- 366. [Ri] J. Rickard, The representation type of self-injective algebras, Bull. London Math. Soc. 22 (1990), 540 -- 546. [RV] C.M. Ringel and D. Vossieck, Hammocks, Proc. London Math. Soc. (3) 54 (1987), no. 2, 216 -- 246. [Sc1] R. Schiffler, Quiver representations, CMS Books in Mathematics/Ouvrages de Math´ematiques de la SMC. Springer, Cham, 2014. [Sc2] O. Schiffmann, Lectures on Hall algebras, Geometric methods in representation theory. II, 1 -- 141, S´emin. Congr., 24-II, Soc. Math. France, Paris, 2012. [Sm] J.H. Smith, Some properties of the spectrum of a graph, in: Combinatorial Structures and their Applications (Eds. R. Guy, H. Hanani, N. Sauer, J. Schonheim), Gordon and Breach (New York), 1970, 403 -- 406. [VdB] M. Van den Bergh, Three-dimensional flops and noncommutative rings, Duke Math. J. 122(3), (2004), 423 -- 455. [vR1] A.-C. van Roosmalen, Abelian 1-Calabi-Yau categories, Int. Math. Res. Not. IMRN 2008, no. 6, Art. ID rnn003, 20pp. [vR2] A.-C. van Roosmalen, Numerically finite hereditary categories with Serre duality, Trans. Amer. Math. Soc. 368 (2016), no. 10, 7189 -- 7238. [Ye] A. Yekutieli, Dualizing complexes, Morita equivalence and the derived Picard group of a ring, J. London Math. Soc. (2) 60 (1999), no. 3, 723 -- 746. Chen: School of Mathematical Sciences, Xiamen University, Xiamen 361005, Fujian, China E-mail address: [email protected] Gao: Department of Communication Engineering, Xiamen University, Xiamen 361005, Fujian, China E-mail address: [email protected] Wicks: Department of Mathematics, Box 354350, University of Washington, Seattle, Washington 98195, USA E-mail address: [email protected] J.J. Zhang: Department of Mathematics, Box 354350, University of Washington, Seattle, Washington 98195, USA E-mail address: [email protected] X.-H. Zhang: College of Sciences, Ningbo University of Technology, Ningbo, 315211, Zhejiang, China E-mail address: [email protected] FROBENIUS-PERRON THEORY OF ENDOFUNCTORS 45 Zhu: Department of Information Sciences, the School of Mathematics and Physics, Changzhou University, Changzhou 213164, China E-mail address: [email protected]
1003.0956	8	1003	2018-03-13T11:12:18	Signatures of hermitian forms and the Knebusch Trace Formula	[ "math.RA" ]	Signatures of quadratic forms have been generalized to hermitian forms over algebras with involution. In the literature this is done via Morita theory, which causes sign ambiguities in certain cases. In this paper, a hermitian version of the Knebusch Trace Formula is established and used as a main tool to resolve these ambiguities. The last page is an erratum for the published version. We inadvertently (I) gave an incorrect definition of adjoint involutions; (II) omitted dealing with the case $(H\times H, \widehat{\phantom{m}}\,)$. As $W(H\times H, \widehat{\phantom{m}}\,)= W(R\times R, \widehat{\phantom{m}}\,)=0$, the omission does not affect our reasoning or our results. For the sake of completeness we point out where some small changes should be made in the published version.	math.RA	math	SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA VINCENT ASTIER AND THOMAS UNGER Abstract. Signatures of quadratic forms have been generalized to hermitian forms over algebras with involution. In the literature this is done via Morita theory, which causes sign ambiguities in certain cases. In this paper, a hermitian version of the Knebusch Trace Formula is established and used as a main tool to resolve these am- biguities. 1. Introduction In this paper we study signatures of hermitian forms over central simple algebras with involution of any kind, defined over formally real fields. These generalize the classical signatures of quadratic forms. Following [3] we do this via extension to real closures and Morita equivalence. This leads to the notion of M-signature of hermitian forms in Section 3.2. We study its properties, make a detailed analysis of the impact of choosing different real closures and different Morita equivalences and show in particular that sign changes can occur. This motivates the search for a more intrinsic notion of signature, where such sign changes do not occur. In Section 3.3 we define such a signature, the H-signature, which only depends on the choice of a tuple of hermitian forms, mimicking the fact that in quadratic form theory the form h1i always has positive signature. The H-signature generalizes the definition of signature in [3] and is in particular well-defined when the involution be- comes hyperbolic after scalar extension to a real closure of the base field, addressing an issue with the definition proposed in [3]. Our main tool is a generalization of the Knebusch trace formula to M-signatures of hermitian forms, which we establish in Section 5. In Section 7 we show that the total H-signature of a hermitian form is a continuous map and in Section 8 we prove the Knebusch trace formula for H-signatures. 2. Preliminaries 2.1. Algebras with Involution. The general reference for this section is [11, Chap- ter I]. 2000 Mathematics Subject Classification. 16K20, 11E39, 13J30. Key words and phrases. Central simple algebras, involutions, hermitian forms, signatures, Knebusch trace formula. 1 2 VINCENT ASTIER AND THOMAS UNGER Let F be a field of characteristic different from 2. An F-algebra with involution is a pair (A, σ) consisting of a finite-dimensional F-algebra A with centre Z(A) and an F-linear map σ : A → A such that σ(xy) = σ(y)σ(x) for all x, y ∈ A and σ2 = idA. The involution σ is either of the first kind or of the second kind. In the first case A is simple, Z(A) = F and σF = idF. In the second case there are two possibilities: either A is simple, Z(A) = K = F( √d) for some d ∈ F× and σK is the nontrivial F-automorphism of K, or (A, σ) ≃ (B × Bop,b) with B a simple F-algebra, Z(A) ≃ F × F a double-field andb the exchange involution, defined by [(x, yop) = (y, xop) for all x, y ∈ B. We call (A, σ) degenerate if it is isomorphic to (B × Bop,b). Consider the F-subspaces Sym(A, σ) = {a ∈ A σ(a) = a} and Skew(A, σ) = {a ∈ A σ(a) = −a} of A. Then A = Sym(A, σ) ⊕ Skew(A, σ). Assume that σ is of the first kind. Then dimF(A) = m2 for some positive integer m. Furthermore, σ is either orthogonal (or, of type +1) if dimF Sym(A, σ) = m(m + 1)/2, or symplectic (or, of type −1) if dimF Sym(A, σ) = m(m − 1)/2. If σ is of the second kind, then dimF(A) = 2m2 for some positive integer m and dimF Sym(A, σ) = dimF Skew(A, σ) = m2. Involutions of the second kind are also called unitary. Let σ and τ be two involutions on A that have the same restriction to Z(A). By the Skolem-Noether theorem they differ by an inner automorphism: τ = Int(u) ◦ σ for some u ∈ A×, uniquely determined up to a factor in F×, such that σ(u) = u if σ and τ are both orthogonal, both symplectic or both unitary and σ(u) = −u if one of σ, τ is orthogonal and the other symplectic. Here Int(u)(x) := uxu−1 for x ∈ A. 2.2. ε-Hermitian Spaces and Forms. The general references for this section are [10, Chapter I] and [19, Chapter 7], both for rings with involution. Treatments of the central simple and division cases can also be found in [6] and [14], respectively. Let (A, σ) be an F-algebra with involution. Let ε ∈ {−1, 1}. An ε-hermitian space over (A, σ) is a pair (M, h), where M is a finitely generated right A-module (which is automatically projective since A is semisimple) and h : M × M −→ A is a sesquilinear form such that h(y, x) = εσ(h(x, y)) for all x, y ∈ M. We call (M, h) a hermitian space when ε = 1 and a skew-hermitian space when ε = −1. If (A, σ) is a field equipped with the identity map, we say (skew-) symmetric bilinear space instead of (skew-) hermitian space. Consider the left A-module M∗ = HomA(M, A) as a right A-module via the involu- tion σ. The form h induces an A-linear map h∗ : M → M∗, x 7→ h(x, ·). We call (M, h) nonsingular if h∗ is an isomorphism. All spaces occurring in this paper are assumed to be nonsingular. We often simply write h instead of (M, h) and speak of a form instead of a space. By Wedderburn theory, M decomposes into a direct sum of k simple right A-modules, for some k ∈ N which we call the rank of h. If A = D is a division algebra (so that M ≃ Dn for some integer n) such that (D, σ, ε) , (F, idF, −1), then h can be diagonalized: there exist invertible elements a1, . . . , an ∈ Sym(D, σ) such that, after a change of basis, h(x, y) = Pn i=1 σ(xi)aiyi, for SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 3 all x = (x1, . . . , xn), y = (y1, . . . , yn) ∈ Dn. In this case we use the shorthand notation h = ha1, . . . , aniσ, which resembles the usual notation for diagonal quadratic forms. If A is not a division algebra we can certainly consider diagonal hermitian forms h = ha1, . . . , aniσ defined on the free A-module An, where n ∈ N and a1, . . . , an are invertible elements in Sym(A, σ). We call n the dimension of h. Note that not all hermitian forms over (A, σ) are diagonalizable. Witt cancellation and Witt decomposition hold for ε-hermitian spaces over (A, σ). Let Wε(A, σ) denote the Witt group of Witt classes of ε-hermitian spaces over (A, σ). When ε = 1 we drop the subscript and simply write W(A, σ). We denote the usual Witt ring of F by W(F). We find it convenient to identify forms over (A, σ) with their classes in Wε(A, σ). 0. Lemma 2.1. Let (A, σ) be an F-algebra with involution. (i) If σ is of the first kind and dimF(A) is odd, then W(A, σ) ≃ W(F) and W−1(A, σ) = (ii) If σ is of the first kind and dimF(A) is even, then W−1(A, σ) ≃ W(A, τ) for some (iii) If σ is of the second kind, then W−1(A, σ) ≃ W(A, σ). (iv) If (A, σ) is degenerate, then W(A, σ) = 0. involution τ of opposite type to σ. Proof. (i) It follows from the assumptions that A ≃ Mn(F) and that σ is necessarily or- thogonal. By hermitian Morita theory (see Section 2.4) we have that W(A, σ) ≃ W(F) and W−1(A, σ) ≃ W−1(F, idF). Since skew-symmetric forms over F are hyperbolic we have W−1(F, idF) = 0. (ii) By [11, 3.A] there exists an involution τ of opposite type to σ since A has involutions of both types under our assumptions. As observed earlier we then have τ ◦ σ = Int(u) for some u ∈ A× with σ(u) = −u. Let h be a skew-hermitian form over (A, σ). Then uh is a hermitian form over (A, τ). The one-to-one correspondence h 7→ uh respects isometries, orthogonal sums and hyperbolicity and so induces the indicated isomorphism. (iii) Let u ∈ Z(A), u , 0 be such that σ(u) = −u. For example, let u = √d if Z(A) = F( √d) and let u = (−1, 1) if Z(A) ≃ F × F. The one-to-one correspondence h 7→ uh induces the indicated isomorphism. (iv) Assume that (A, σ) ≃ (B × Bop,b) and let h : M × M → B × Bop be hermitian with respect tob. Let e1 = (1, 0) and e2 = (0, 1). Then M = M1 ⊕ M2 with Mi := Mei (i = 1, 2) and h is hyperbolic since hM1×M1 Corollary 3.7.3]), which can be verified by direct computation. = M1 (cf. [10, Chapter I, (cid:3) = 0 and M⊥1 (cid:3) In view of Lemma 2.1(iv) degenerate F-algebras with involution are not interest- ing in the context of this paper. Therefore we redefine F-algebra with involution to mean non-degenerate F-algebra with involution. Observe that such an algebra with involution may become degenerate over a field extension of F. 2.3. Adjoint Involutions. The general reference for this section is [11, 4.A]. 4 VINCENT ASTIER AND THOMAS UNGER Let (A, σ) be an F-algebra with involution. Let (M, h) be an ε-hermitian space over (A, σ). The algebra EndA(M) is again simple with centre Z(A) since M is finitely gener- ated [11, 1.10]. The involution adh on EndA(M), defined by h(x, f (y)) = h(adh( f )(x), y), for all x, y ∈ M, and all f ∈ EndA(M) is called the adjoint involution of h. The invo- lutions σ and adh are of the same kind and σ(α) = adh(α) for all α ∈ Z(A). In case adh and σ are of the first kind we also have type(adh) = ε type(σ). Furthermore, every involution on EndA(M) is of the form adh for some ε-hermitian form h over (A, σ) and the correspondence between adh and h is unique up to a multiplicative factor in F× in the sense that adh = adλh for every λ ∈ F×. Let (A, σ) be an F-algebra with involution. By a theorem of Wedderburn there exists a division algebra D (unique up to isomorphism) with centre Z(A) and a finite- dimensional right D-vector space V such that A ≃ EndD(V). Thus A ≃ Mm(D) for some positive integer m. Furthermore there exists an involution ϑ on D of the same kind as σ and an ε0-hermitian form ϕ0 over (D, ϑ) with ε0 ∈ {−1, 1} such that (A, σ) and (EndD(V), adϕ0) are isomorphic as algebras with involution. In matrix form adϕ0 is described as follows: adϕ0(X) = Φ0ϑt(X)Φ−1 0 , for all X ∈ Mm(D), where ϑt(X) := (ϑ(xi j))t for X = (xi j) and Φ0 ∈ GLm(D) is the Gram matrix of ϕ0. Thus ϑt(Φ0) = ε0Φ0. 2.4. Hermitian Morita Theory. We refer to [2, §1], [5], [6, Chapters 2 -- 3], [10, Chapter I, §9], or [13] for more details. Let (M, h) be an ε-hermitian space over (A, σ). One can show that the algebras with involution (EndA(M), adh) and (A, σ) are Morita equivalent: for every µ ∈ {−1, 1} there is an equivalence between the categories H µ(EndA(M), adh) and H εµ(A, σ) of non-singular µ-hermitian forms over (EndA(M), adh) and non-singular εµ-hermitian forms over (A, σ), respectively (where the morphisms are given by isometry), cf. [10, Chapter I, Theorem 9.3.5]. This equivalence respects isometries, orthogonal sums and hyperbolic forms. It induces a group isomorphism Wµ(EndA(M), adh) ≃ Wεµ(A, σ). The Morita equivalence and the isomorphism are not canonical. The algebras with involution (A, σ) and (D, ϑ) are also Morita equivalent. An ex- ample of such a Morita equivalence is obtained by composing the following three non-canonical equivalences of categories, the last two of which we will call scaling and collapsing. For computational purposes we describe them in matrix form. We follow the approach of [17]: H ε(A, σ) −−−→ H ε(Mm(D), adϕ0) scaling −−−−−−→ H ε0ε(Mm(D), ϑt) collapsing −−−−−−−−→ H ε0ε(D, ϑ). (1) Scaling: Let (M, h) be an ε-hermitian space over (Mm(D), adϕ0). Scaling is given is only determined up to a scalar factor in F× = adλϕ0 for any λ ∈ F× and that replacing Φ0 by λΦ0 results in a different by (M, h) 7−→ (M, Φ−1 since adϕ0 Morita equivalence. 0 h). Note that Φ−1 0 Collapsing: Recall that Mm(D) ≃ EndD(Dm) and that we always have M ≃ (Dm)k ≃ Mk, m(D) for some integer k. Let h : M × M −→ Mm(D) be an ε0ε-hermitian form with respect to ϑt. Then h(x, y) = ϑt(x)By, for all x, y ∈ Mk, m(D), where B ∈ Mk(D) satisfies SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 5 ϑt(B) = ε0εB, so that B determines an ε0ε-hermitian form b over (D, ϑ). Collapsing is then given by (M, h) 7−→ (Dk, b). 3. Signatures of Hermitian Forms 3.1. Signatures of Forms: the Real Closed Case. Let R be a real closed field, C = R( √ −1) (which is algebraically closed) and H = (−1, −1)R Hamilton's quaternion division algebra over R. We recall the definitions of signature for the various types of forms (all assumed to be nonsingular) over R, C, R× R and H. We will use them in the definition of the M-signature of a hermitian form over (A, σ) in Section 3.2. (a) Let b be a symmetric bilinear (or quadratic) form over R. Then b ≃ hα1, . . . , αni for some n ∈ N and αi ∈ {−1, 1}. We let sign b := Pn i=1 αi. By Sylvester's Law of Inertia, sign b is well-defined. (b) Let b be a skew-symmetric form over R. Then b is hyperbolic and we let sign b := 0. (c) Let h be a hermitian form over (C, −), where √ √ −1. Then h ≃ hα1, . . . , αni− for some n ∈ N and αi ∈ {−1, 1}. By a theorem of Jacobson [7], h is up to isometry uniquely determined by the symmetric bilinear form bh := 2×hα1, . . . , αni defined over R. We let sign h := 1 −1 = − 2 sign bh = signhα1, . . . , αni. involution. Then h is hyperbolic by Lemma 2.1(iv) and we let sign h := 0. (d) Let h be a hermitian form over (R × R,b), where [(x, y) = (y, x) is the exchange (e) Let h be a hermitian form over (H, −), where − denotes quaternion conjugation. Then h ≃ hα1, . . . , αni− for some n ∈ N and αi ∈ {−1, 1}. By a theorem of Jacobson [7], h is up to isometry uniquely determined by the symmetric bilinear form bh := 4 × hα1, . . . , αni defined over R. We let sign h := 1 (f) Let h be a skew-hermitian form over (H, −), where − denotes quaternion conjuga- tion. Then h is a torsion form [19, Chapter 10, Theorem 3.7] and we let sign h := 0. 4 sign bh = signhα1, . . . , αni. to (c) and (d), respectively, by Lemma 2.1. Note that the cases of skew-hermitian forms over (C, −) and (R × R,b) can be reduced Remark 3.1. The signature maps defined in (a), (c) and (e) above give rise to unique group isomorphisms W(R) ≃ Z, W(C, −) ≃ Z, and W(H, −) ≃ Z such that signh1i = 1, = 1, respectively. In addition, we have the group isomor- signh1i− 10.5]. phisms W−1(R, idR) = W(R × R,b) = 0, W−1(H, −) ≃ Z/2Z. See also [10, Chapter I, = 1 and signh1i− 3.2. The M-Signature of a Hermitian Form. Our approach in this section is inspired by [3, §3.3,§3.4]. We only consider hermitian forms over (A, σ), cf. Lemma 2.1. Let F be a formally real field and let (A, σ) be an F-algebra with involution. Con- sider an ordering P ∈ XF, the space of orderings of F. By a real closure of F at P we mean a field embedding ι : F → K, where K is real closed, ι(P) ⊆ K2 and K is algebraic over ι(F). 6 VINCENT ASTIER AND THOMAS UNGER Let h be a hermitian form over (A, σ). Choose a real closure ι : F → FP of F at P, and use it to extend scalars from F to FP: W(A, σ) −→ W(A ⊗F FP, σ ⊗ id), h 7−→ h ⊗ FP := (idA ⊗ ι)∗(h), where the tensor product is along ι. The extended algebra with involution (A⊗F FP, σ⊗ id) is Morita equivalent to an algebra with involution (DP, ϑP), chosen as follows: (i) If σ is of the first kind, DP is equal to one of FP or HP := (−1, −1)FP . Furthermore, we may choose (DP, ϑP) = (FP, idFP ) in the first case and (DP, ϑP) = (HP, −) in the second case by Morita theory (scaling). (ii) If σ is of the second kind, recall that Z(A) = F( √d). Now if d <P 0, then DP is equal to FP( √ −1), whereas if d >P 0, then DP is equal to FP × FP and A ⊗F FP is a direct product of two simple algebras. Furthermore, we may choose (DP, ϑP) = (FP( √ second case, again by Morita theory (scaling). −1), −) in the first case and (DP, ϑP) = (FP × FP,b) in the Note that ϑP is of the same kind as σ in each case. The extended involution σ⊗ idFP is adjoint to an εP-hermitian form ϕP over (DP, ϑP) where εP = −1 if one of σ and ϑP is orthogonal and the other is symplectic, whereas εP = 1 if σ and ϑP are of the same type, i.e. both orthogonal, symplectic or unitary. Now choose any Morita equivalence M : H (A ⊗F FP, σ ⊗ id) −→ H εP(DP, ϑP) with (DP, ϑP) ∈ {(FP, idFP ), (HP, −), (FP( √ analysis above. This Morita equivalence induces an isomorphism, which we again denote by M , −1), −), (FP × FP,b)}, which exists by the (2) (3) M : W(A ⊗F FP, σ ⊗ id) ∼−→ WεP(DP, ϑP). Definition 3.2. Let P ∈ XF. Fix a real closure ι : F → FP of F at P and a Morita equivalence M as above. Define the M-signature of h at (ι, M ), denoted signM ι h, as follows: where sign M (h ⊗ FP) can be computed as shown in Section 3.1. signM ι h := sign M (h ⊗ FP), This definition relies on two choices: firstly the choice of the real closure ι : F → FP of F at P and secondly the choice of the Morita equivalence M . Note that there is no canonical choice for M . We now study the dependence of the M-signature on the choice of ι and M . Let ι1 : F → L1 and ι2 : F → L2 be two real closures of F at P, and let (D1, ϑ1) and ε1 play the role of (DP, ϑP) and εP, respectively, obtained above when ι is replaced by ι1. Let M1 : H (A ⊗F L1, σ ⊗ id) −→ H ε1(D1, ϑ1) be a fixed Morita equivalence. By the Artin-Schreier theorem [19, Chapter 3,Theorem 2.1] there is a unique isomorphism ρ : L1 → L2 such that ρ ◦ ι1 = ι2. It extends to an isomorphism id ⊗ ρ : (A ⊗F L1, σ ⊗ id) → (A ⊗F L2, σ ⊗ id). The isomorphism ρ also extends canonically to D1 ∈ {L1, (−1, −1)L1, L1( √ −1), L1 × L1}. Consider the L2-algebra with involution (D2, ϑ2) := SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 7 (ρ(D1), ρ◦ϑ1◦ρ−1). We define ρ(M1) to be the Morita equivalence from (A⊗F L2, σ⊗id) to (D2, ϑ2), described by the following diagram: H (A ⊗F L1, σ ⊗ id) (id⊗ρ)∗ H (A ⊗F L2, σ ⊗ id) M1 ρ(M1) H ε1(D1, ϑ1) ρ∗ / H ε1(D2, ϑ2) Proposition 3.3 (Change of Real Closure). With notation as above we have for every h ∈ W(A, σ), sign M1(h ⊗ L1) = sign ρ(M1)(h ⊗ L2), in other words signM1 ι1 h = signρ(M1) ι2 h. Proof. The statement is trivially true when ε1 = −1, by cases (b), (d) and (f) in Sec- tion 3.1, so we may assume that ε1 = 1. Consider the diagram W(A ⊗F L1, σ ⊗ id) W(D1, ϑ1) W(A, σ) (id⊗ρ)∗ (id⊗ι1)∗ 6♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗ (id⊗ι2)∗ M1 ∼ ρ(M1) ∼ ρ∗ Z sign $■■■■■■■■■■ :✉✉✉✉✉✉✉✉✉✉ sign W(A ⊗F L2, σ ⊗ id) / W(D2, ϑ2) The left triangle commutes by the definition of ρ. The square commutes by the defini- tion of ρ(M1). The right triangle commutes since ρ∗(h1iϑ1 ) = h1iϑ2 and by Remark 3.1. The statement follows. (cid:3) (cid:3) Proposition 3.4 (Change of Morita Equivalence). Let M1 and M2 be two different Morita equivalences as in (2). Then there exists δ ∈ {−1, 1} such that for every h ∈ W(A, σ), signM1 ι h = δ signM2 ι h. Proof. Note that ε1 = ε2. The statement is trivially true when ε1 = −1, by cases (b), (d) and (f) in Section 3.1, so we may assume that ε1 = 1. The two different Morita equivalences give rise to two different group isomorphisms mi : W(A ⊗F FP, σ ⊗ id) ∼−→ Z (i = 1, 2), by (3) and Remark 3.1. The map m1 ◦ m−1 equal to idZ or −idZ. 2 is an automorphism of Z and is therefore (cid:3) (cid:3) Propositions 3.3 and 3.4 immediately imply / / / / / $ 6 ( / : 8 VINCENT ASTIER AND THOMAS UNGER Corollary 3.5. Let ι1 : F → L1 and ι2 : F → L2 be two real closures of F at P and let M1 and M2 be two different Morita equivalences as in (2). Then there exists δ ∈ {−1, 1} such that for every h ∈ W(A, σ), signM1 ι1 h = δ signM2 ι2 h. The following result easily follows from the properties of Morita equivalence: Proposition 3.6. ι (h1 ⊥ h2) = signM ι h1 + (i) Let h be a hyperbolic form over (A, σ), then signM ι h = 0. (ii) Let h1 and h2 be hermitian forms over (A, σ), then signM signM ι h2. (iii) The M-signature at (ι, M ), signM ι , induces a homomorphism of additive groups (iv) Let h be a hermitian form over (A, σ) and q a quadratic form over F, then ι h, where signP q denotes the usual signature of the W(A, σ) −→ Z. signM ι quadratic form q at P. (q⊗ h) = signP q· signM Definition 3.7. Let h be a hermitian form over (A, σ). From Definition 3.2 and Sec- tion 3.1 it follows that signM ι h is automatically zero whenever P belongs to the fol- lowing subset of XF, which we call set of nil-orderings: Nil[A, σ] := {P ∈ XF DP = HP} {P ∈ XF DP = FP} {P ∈ XF DP = FP × FP} if σ is orthogonal if σ is symplectic , if σ is unitary where the square brackets indicate that Nil[A, σ] depends only on the Brauer class of A and the type of σ. 3.3. The H-Signature of a Hermitian Form. It follows from Corollary 3.5 that signM is uniquely defined up to a choice of sign. We can arbitrarily choose the sign of ι the signature of a form at each ordering P. See for instance Remark 3.13 for a way to change sign using Morita equivalence (scaling). A more intrinsic definition is therefore desirable, in particular when considering the total signature map of a hermitian form XF → Z since such arbitrary changes of sign would prevent it from being continuous. We are thus led to define a signature that is independent of the choice of ι and M . Lemma 3.8. Let P ∈ XF \ Nil[A, σ]. Let ι1 : F → L1 and ι2 : F → L2 be two real closures of F at P and let M1 and M2 be two different Morita equivalences as in (2). Let h0 ∈ W(A, σ) be such that signM1 h0 , 0 and let δk ∈ {−1, 1} be the sign of signMk ιk h0 for k = 1, 2. Then ι1 δ1 signM1 ι1 h = δ2 signM2 ι2 h, for all h ∈ W(A, σ). SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 9 Proof. Let δ ∈ {−1, 1} be as in Corollary 3.5. Then, we have for all h ∈ W(A, σ) that signM2 h0. It follows that ι2 δ1 = δδ2. Thus δ1 signM1 ι1 h and in particular that signM2 ι2 h = δδ2 signM1 ι1 h0 = δ signM1 ι1 h = δ2 signM2 ι2 h = δ signM1 ι1 h. (cid:3) (cid:3) We will show in Theorem 6.4 that there exists a finite tuple H = (h1, . . . , hs) of diagonal hermitian forms of dimension one over (A, σ) such that for every P ∈ XF \ Nil[A, σ] there exists h0 ∈ H such that signM Definition 3.9. Let h ∈ W(A, σ) and let P ∈ XF. We define the H-signature of h at P as follows: If P ∈ Nil[A, σ], define signH P h := 0. If P < Nil[A, σ], let i ∈ {1, . . . , s} be the least integer such that signM ι hi , 0 (for any ι and M , cf. Corollary 3.5), let δι,M ∈ {−1, 1} be the sign of signM signH ι hi and define ι h0 , 0. P h := δι,M signM ι h. Lemma 3.8 shows that this definition is independent of the choice of ι and M (but it does depend on the choice of H). A choice of Morita equivalence which is convenient for computations of signatures is given by (1) with (A, σ) replaced by (A ⊗F FP, σ ⊗ id). We denote this Morita equivalence by N and now describe the induced isomorphisms of Witt groups: W(A ⊗F FP, σ ⊗ id) ξ∗P / W(Mm(DP), adϕP ) scaling / W(Mm(DP), ϑP t) collapsing / W(DP, ϑP) h ⊗ FP ✤ / ξ∗P(h ⊗ FP) ✤ / Φ−1 P ξ∗P(h ⊗ FP) ✤ / N (h ⊗ FP), (4) where h is a hermitian form over (A, σ), P ∈ XF \ Nil[A, σ] (so that εP = 1), ξ∗P is the group isomorphism induced by some fixed isomorphism ξP : (A ⊗F FP, σ ⊗ id) ∼−→ (Mm(DP), adϕP), ι h1iσ = sign ϕP. and ΦP is the Gram matrix of the form ϕP. Observe that sign ϕP can be computed as in Section 3.1. Lemma 3.10. Let P ∈ XF \ Nil[A, σ], let ι : F → FP be a real closure of F and let N and ϕP be as above. Then signN Proof. We extend scalars from F to FP via ι, h1iσ 7−→ h1iσ ⊗ FP = h1⊗ 1iσ⊗id and push h1 ⊗ 1iσ⊗id through the sequence (4), h1 ⊗ 1iσ⊗id 7−→ ξ∗P(h1 ⊗ 1iσ⊗id) = hξP(1 ⊗ 1)iadϕP P iϑP t . (Note that ξP(1 ⊗ 1) = Im, the m × m-identity matrix in Mm(DP) since ξP is an algebra homomorphism.) By collapsing, the matrix Φ−1 P now corresponds to a quadratic form over FP, a hermitian form over (FP( √ −1), −) or a hermitian form over (HP, −). In either case Φ−1 Remark 3.11. It follows from Lemma 3.10 that the signature defined in [3, §3.3, §3.4] is actually signH P with H = (h1iσ). It is now clear that this signature cannot be computed when signN ι h1iσ = 0, i.e. when σ ⊗ idFP ≃ adϕP is hyperbolic. In contrast, P is congruent to ΦP. Thus signN = hImiadϕP 7−→ Φ−1 ι h1iσ = sign ϕP. P hImiadϕP = hΦ−1 (cid:3) (cid:3) / / / / / / 10 VINCENT ASTIER AND THOMAS UNGER if we take H = (h1, . . . , hs), as described before Definition 3.9 we are able to compute the signature in all cases. Note that we may choose h1 = h1iσ, so that Definition 3.9 generalizes the definition of signature in [3, §3.3, §3.4]. Example 3.12. Let (A, σ) = (M4(R), adϕ), where ϕ = h1, −1, 1, −1i. Then sign ϕ = 0. Consider the dimension one hermitian forms h =* 1 1 −1 1!+σ , h1 =* 1 −1!+σ 1!+σ , and h2 =* 1 1 1 −1 1 over (A, σ). Then signN h = −2, signN h1 = 0 and signN h2 = 4, where we suppressed the index ι since R is real closed. Let H1 = (h1) and H = (h1, h2), then signH1 h is not defined, whereas signH h = −2. Observe that taking H = (h1, −h2) instead would result in signH h = 2. Remark 3.13. Let a ∈ A× be such that σ(a) = εa with ε ∈ {−1, 1}. The Morita equiva- lence scaling by a, H (A, σ) −→ H ε(A, Int(a) ◦ σ), h 7−→ ah induces an isomorphism ζa : W(A, σ) −→ Wε(A, Int(a) ◦ σ), h 7−→ ah. It is clear that signM ι h = sign ι M ◦(ζ−1 a ⊗id) ah. Consider the special case where a ∈ F×. Thus ε = 1 and Int(a) ◦ σ = σ. Assume that a <P 0. Then signM ζa(h) = sign M (ah ⊗ FP) = sign M (−h ⊗ FP) = − signM ι h, where the last equality follows from Proposition 3.6(iii). The same computation shows that signH P h for any choice of H. Thus, scaling by a changes the sign of the signature, which is contrary to what is claimed in [3, p. 662]. P ζa(h) = − signH ι Remark 3.14. For any choice of H, P and ι as in Definition 3.9, there exists a Morita equivalence M ′ such that signH h for any h ∈ W(A, σ) (i.e. such that signM ′ hi > 0 with hi as in Definition 3.9). Indeed, for M as in Definition 3.9, it suffices to take M ′ = δι,M M . P h = signM ′ ι ι It remains to be shown that a tuple H as described just before Definition 3.9 does ex- ist. In order to reach this conclusion we first need to develop more theory in Sections 4 and 5. 4. Signatures of Involutions Let (A, σ) be an F-algebra with involution of any kind with centre a field K. Con- sider the involution trace form Tσ : A × A −→ K, (x, y) 7−→ TrdA(σ(x)y), where TrdA denotes the reduced trace of A. If σ is of the first kind, Tσ is a symmetric bilinear form over F. If σ is of the second kind, Tσ is a hermitian form over (K, σK). Let P ∈ XF. The signature of the involution σ at P is defined by signP σ := psignP Tσ and is a nonnegative integer, since signP Tσ is always a square; cf. Lewis and Tignol [15] for involutions of the first kind and Qu´eguiner [18] for involutions of the second kind. We call the involution σ positive at P if signP σ , 0. Example 4.1. SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 11 (i) Let (A, σ) = (Mn(F), t), where t denotes transposition. Then Tσ ≃ n2×h1i. Hence signP σ = n for all P ∈ XF. (ii) Let (A, σ) = ((a, b)F, −), where − denotes quaternion conjugation. Then Tσ ≃ h2i ⊗ h1, −a, −b, abi. Hence signP σ = 2 for all P ∈ XF such that a <P 0, b <P 0 and signP (iii) Let (A, σ) = (F( √d), −), where − denotes conjugation. Then Tσ ≃ h1iσ. We have signPh1iσ = 1 2 signPh1, −di, cf. [19, Chapter 10, Examples 1.6(iii)]. Hence signP σ = 0 for all P ∈ XF such that d >P 0 and signP σ = 1 for all P ∈ XF such that d <P 0. = 0 for all other P ∈ XF. τ) = (signP σ)(signP τ) for all P ∈ XF. Remark 4.2. Let (A, σ) and (B, τ) be two F-algebras with involution. (i) Consider the tensor product (A⊗F B, σ⊗τ). Then Tσ⊗τ ≃ Tσ⊗Tτ and so signP(σ⊗ (ii) If (A, σ) ≃ (B, τ), then Tσ ≃ Tτ so that signP σ = signP τ for all P ∈ XF. Remark 4.3. Pfister's local-global principle holds for algebras with involution (A, σ) and also for hermitian forms h over such algebras, [16]. Remark 4.4. The map sign σ is continuous from XF (equipped with the Harrison topology, see [12, Chapter VIII 6] for a definition) to Z (equipped with the discrete topology). Indeed: define the map √ on Z by setting √k = −1 if k is not a square in Z. Since Z is equipped with the discrete topology, this map is continuous. Since Tσ is a symmetric bilinear form or a hermitian form over (K, σK), the map sign Tσ is continuous from XF to Z (by [12, VIII, Proposition 6.6] and [19, Chapter 10, Ex- ample 1.6(iii)]). Thus, by composition, sign σ = psign Tσ is continuous from XF to Z. Lemma 4.5. Let P ∈ XF. If P ∈ Nil[A, σ], then signP σ = sign ϕP = 0. Other- wise, signP σ = λP sign ϕP, where λP = 1 if (DP, ϑP) = (FP, idFP ) or (DP, ϑP) = (FP( √ −1), −) and λP = 2 if (DP, ϑP) = (HP, −). Proof. This is a reformulation of [15, Theorem 1] and part of its proof for involutions of the first kind and [18, Proposition 3] for involutions of the second kind. (cid:3) Lemma 4.6. Let (M, h) be a hermitian space over (A, σ), let P ∈ XF, let ι : F → FP be a real closure of F at P and let M be a Morita equivalence as in (2). If P ∈ Nil[A, σ], then signP adh = signM ι h, with λP as defined in Lemma 4.5. In particular, signM Proof. Assume first that P ∈ Nil[A, σ]. Then signM ι h = 0. Consider the adjoint involution adh on EndA(M). Since h is hermitian, σ and adh are of the same type. Furthermore, A and EndA(M) are Brauer equivalent by [11, 1.10]. Thus Nil[A, σ] = Nil[EndA(M), adh]. By Lemma 4.5 we conclude that signP adh = 0. ι h = 0. Otherwise, signP adh = λP signM ι h = 0 if and only if signP adh = 0. (cid:3) Next, assume that P ∈ XF \ Nil[A, σ]. Without loss of generality we may re- place F by FP. Consider a Morita equivalence M ′ : H (A, σ) −→ H (D, ϑ) with (D, ϑ) = (F, id), (D, ϑ) = (H, −) or (D, ϑ) = (F( √ −1), −). Let (N, b) be the hermitian 12 VINCENT ASTIER AND THOMAS UNGER space over (D, ϑ) corresponding to (M, h) under M ′. Then signM ′ h = sign b. By [2, Remark 1.4.2] we have (EndA(M), adh) ≃ (EndD(N), adb) so that sign adh = sign adb. By [15, Theorem 1] and [18, Proposition 3] we have sign adb = λ sign b with λ = 1 if (D, ϑ) = (F, id) or (D, ϑ) = (F( √ −1), −) and λ = 2 if (D, ϑ) = (H, −). We con- clude that sign adh = λ signM ′ h = λ signM h, where the last equality follows from Corollary 3.5. (cid:3) (cid:3) 5. The Knebusch Trace Formula for M-Signatures We start with two preliminary sections in order not to overload the proof of Theo- rem 5.1 below. 5.1. Hermitian Forms over a Product of Rings with Involution. Let (A, σ) = (A1, σ1)× · · · × (At, σt), where A, A1, . . . , At are rings and σ, σ1, . . . , σt are involutions. We write an element a ∈ A indiscriminately as (a1, . . . , at) or a1 + · · · + at with ai ∈ Ai for i = 1, . . . , t. Writing 1A = (e1, . . . , et), the elements e1, . . . , et are central idempotents, and the coordinates of a ∈ A are given by A −→ A1 × · · · × At, a 7−→ (ae1, . . . , aet). Note that eie j = 0 whenever i , j. We assume that σ(1) = 1 and thus σ(ei) = ei for i = 1, . . . , t. Let M be an A-module and let h : M × M → A be a hermitian form over (A, σ). Following [9, Proof of Lemma 1.9] we can write M = Mei, m = (me1, . . . , met), tai=1 i=1 Mei is the A-module with set of elements Qt where `t i=1 Mei, whose sum is de- fined coordinate by coordinate and whose product is defined by (m1e1, . . . , mtet)a = (m1e1a1, . . . , mtetat) for m1, . . . , mt ∈ M and a ∈ A. Define hi = hMei. Then hi(xei, yei) = σ(ei)h(x, y)ei = h(x, y)e2 hi : Mei × Mei → Ai is a hermitian form over (Ai, σi). We also have h(xei, ye j)eie j h(xe1 + · · · + xet, ye1 + · · · + yet) = = h(x, y)ei and i tXi, j=1 tXi=1 = h(xei, yei)ei which proves that h = h1 ⊥ . . . ⊥ ht. 5.2. Algebraic Extensions and Real Closures. We essentially follow [19, Chapter 3, Lemma 2.6, Lemma 2.7, Theorem 4.4]. Let P ∈ XF and let FP denote a real closure of F at P. Let L be a finite extension of F. Writing L = F[X]/(R) for some R ∈ F[X] and R = R1 · · · Rt as a product of pairwise distinct irreducibles in FP[X] with deg R1 = · · · = deg Rr = 1 and deg Rr+1 = SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 13 · · · = deg Rt = 2, we obtain canonical FP-isomorphisms L ⊗F FP ≃ FP[X]/(R1 · · · Rt) and ω L ⊗F FP −→ F1 × · · · × Ft, (5) where Fi = FP[X]/(Ri) is a real closed field for 1 ≤ i ≤ r and is algebraically closed for r + 1 ≤ i ≤ t. We write 1 = (e1, . . . , et) in F1 × · · · × Ft and define ωi(x) = ω(x)ei for x ∈ L ⊗F FP, the projection of ω(x) on its i-th coordinate. Let ι0 : FP → L ⊗F FP be the canonical inclusion. Then ωi ◦ ι0 : FP → Fi is an isomorphism of fields and of FP-modules for 1 ≤ i ≤ r. In particular, for 1 ≤ i ≤ r, Fi is naturally an FP-module of dimension one, and it is easily seen that TrFi/FP (ωi ◦ ι0)−1 and so TrFi/FP is an isomorphism of fields. Let ι1 : L → L⊗F FP be the canonical inclusion. Then ωi ◦ ι1 : L → Fi (i = 1, . . . , r) denote the r different embeddings of ordered fields corresponding to the orderings Qi on L that extend P. In other words, if {Q1, . . . , Qr} are the different extensions of P to L, then for every 1 ≤ i ≤ r, the map = ι1 L / L ⊗F FP ωi / Fi is a real closure of L at Qi. Since TrFi/FP is an isomorphism of fields, it follows that the map ι1 L is also a real closure of L at Qi. / L ⊗F FP ωi TrFi/FP/ / FP / Fi 5.3. The Knebusch Trace Formula. Let (A, σ) be an F-algebra with involution. Let L/F be a finite extension. The trace TrL/F : L → F induces an A-linear homomorphism which induces a group homomorphism (transfer map) TrA⊗F L = idA ⊗ TrL/F : A ⊗F L −→ A Tr∗A⊗F L : W(A ⊗F L, σ ⊗ id) −→ W(A, σ), (M, h) 7−→ (M, TrA⊗F L ◦ h), cf. [1, p. 362]. The following theorem is an extension of a result due to Knebusch [8, Proposi- tion 5.2], [19, Chapter 3, Theorem 4.5] to F-algebras with involution. The proof fol- lows the general lines of Knebusch's original proof. Theorem 5.1. Let P ∈ XF. Let L/F be a finite extension of ordered fields and let h be a hermitian form over (A ⊗F L, σ ⊗ id). Fix a real closure ι : F → FP and a Morita equivalence M as in (2). Then, with notation as in Section 5.2, signM ι (Tr∗A⊗F Lh) = rXi=1 sign(ωi◦ι0)(M ) ωi◦ι1 h. Proof. By definition of signature we have signM ι (Tr∗A⊗F Lh) = sign M [(Tr∗A⊗F Lh) ⊗F FP]. (6) / / / / 14 VINCENT ASTIER AND THOMAS UNGER Consider the commutative diagram TrL/F L ⊗F FP L ⊗F FP TrL⊗FP /FP F ⊗F FP / FP It induces a commutative diagram TrA⊗L A A ⊗F L ⊗F FP ⊗F FP / A ⊗F FP which in turn induces a commutative diagram of Witt groups A ⊗F L ⊗F FP idA⊗TrL⊗FP /FP / W(A ⊗F L, σ ⊗ id) Tr∗A⊗L W(A, σ) W(A ⊗F L ⊗F FP, σ ⊗ id ⊗ id) (idA⊗TrL⊗FP/FP )∗ / W(A ⊗F FP, σ ⊗ id) where the vertical arrows are the canonical restriction maps. Thus sign M [(Tr∗A⊗F Lh) ⊗F FP] = sign M [(idA ⊗ TrL⊗FP/FP )∗(h ⊗F FP)]. (7) With reference to Section 5.2 consider the diagram A ⊗F L ⊗F FP idA⊗ω A ⊗F (F1 × · · · × Ft) ∼ / idA⊗F TrL⊗FP /FP id A ⊗F FP idA⊗TrF1×···×Ft /FP / A ⊗F FP id (A ⊗F F1) × · · · × (A ⊗F Ft) Pt i=1 idA⊗TrFi/FP / A ⊗F FP where commutativity of the first square follows from the isomorphism (5) of FP- algebras, whereas commutativity of the second square follows from [4, p. 137]. We push h ⊗ FP through the induced commutative diagram of Witt groups: / h′1 ⊥ . . . ⊥ h′t h ⊗ FP (idA⊗ω)∗ h′ ✤ ❴ ❴ (idA ⊗ TrL⊗FP/FP )∗(h ⊗ FP) ✤ id /Pt i=1(idA ⊗ TrFi/FP )∗(h′i) where the image of h′ equals the orthogonal sum h′1 ⊥ . . . ⊥ h′t by Section 5.1. Thus sign M [(idA ⊗ TrL⊗FP/FP )∗(h ⊗F FP)] = We have to consider the following two cases: tXi=1 sign M [(idA ⊗ TrFi/FP )∗(h′i)]. (8) / / / / / / / / / / / / / ✤ / / / / SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 15 Case 1: Assume that 1 ≤ i ≤ r. Observe that h′i sign M [(idA ⊗ TrFi/FP )∗(h′i)] = sign M [(idA ⊗ (TrFi/FP ))∗ ◦ (idA ⊗ (ωi ◦ ι1))∗(h)]. The form (idA ⊗ (ωi ◦ ι1))∗(h) is defined over A ⊗F Fi, and the commutative diagram = (idA ⊗ (ωi ◦ ι1))∗(h). Then Fi together with Proposition 3.3 gives TrFi/FP 7♦♦♦♦♦♦ '❖❖❖❖❖❖ id FP ωi◦ι0 Fi sign M [(idA ⊗ (TrFi/FP ))∗ ◦ (idA ⊗ (ωi ◦ ι1))∗(h)] (9) Case 2: Assume that r + 1 ≤ i ≤ t. Since Fi is algebraically closed, it follows from Morita theory that the Witt group W(A ⊗F Fi, σ ⊗ id) is torsion and so h′i is a torsion form. Therefore (idA ⊗ TrFi/FP )∗(h′i) is also a torsion form and thus has signature zero. = sign(ωi ◦ ι0)(M )[(idA ⊗ (ωi ◦ ι1))∗(h)] We conclude that equations (6)-(9) yield the Knebusch trace formula. (cid:3) (cid:3) 6. Existence of Forms with Nonzero Signature Theorem 6.1. Let (A, σ) be an F-algebra with involution and let P ∈ XF \ Nil[A, σ]. Fix a real closure ι : F → FP and a Morita equivalence M as in (2). There exists a hermitian form h over (A, σ) such that signM ι h , 0. Proof. Let P ∈ XF \ Nil[A, σ]. Then A ⊗F FP ≃ Mm(DP) for some m ∈ N, where DP = FP, HP or FP( √ −1) if σ is orthogonal, symplectic or unitary, respectively. In each case there exists a positive involution τ on Mm(DP) (namely, transposition, conjugate transposition and quaternion conjugate transposition, respectively, cf. Example 4.1). By Lemma 4.6 the hermitian form h1iτ over (Mm(DP), τ) has nonzero signature since τ is the adjoint involution of the form h1iτ. After scaling we obtain a dimension one hermitian form h0 over (A ⊗F FP, σ ⊗ id) such that sign M (h0) , 0 by Proposition 3.4. The form h0 is already defined over a finite extension L of ι(F), contained in FP. Thus we can consider h0 as a form over (A⊗F L, σ⊗id) and we have sign M (h0⊗ FP) , 0. In other words, if Q1 is the ordering L∩ F×2 P on L, then for any real closure κ1 : L → L1 of L at Q1 and any Morita equivalence M1 as in (2), but starting from H (A⊗F L1, σ⊗ id), we have signM1 h0 , 0 by Corollary 3.5. κ1 Let X = {Q ∈ XL P ⊂ Q}. By [19, Chapter 3, Lemma 2.7], X is finite, say X = {Q1, Q2, . . . , Qr}. Thus there exist a2, . . . , ar ∈ L× such that {Q1} = H(a2, . . . , ar) ∩ X. q = 0 for ℓ , 1. It follows that signM1 κ1 q = 2r−1 h0 , 0 and (q ⊗ h0) = 0 for ℓ , 1, where κℓ : L → Lℓ is any real closure of L at Qℓ and Mℓ Consider the Pfister form q := hha2, . . . , arii = h1, a2i⊗· · ·⊗h1, ari. Then signQ1 q· signM1 and signQℓ signMℓ κℓ is any Morita equivalence as in (2), but starting from H (A ⊗F Lℓ, σ ⊗ id). (q⊗ h0) = signQ1 κ1 7 ' 16 VINCENT ASTIER AND THOMAS UNGER Now Tr∗A⊗F L(q ⊗ h0) is a hermitian form over (A, σ). By the trace formula, Theo- rem 5.1, we have signM ι (cid:0)Tr∗A⊗F L(q ⊗ h0)(cid:1) = rXi=1 sign(ωi◦ι0)(M ) ωi◦ι1 (q ⊗ h0) = sign(ω1◦ι0)(M ) ω1◦ι1 (q ⊗ h0) , 0. ι (cid:3) (cid:3) h = 0 for all hermitian forms h over (A1, σ1); h = 0 for all hermitian forms h over (A2, σ2); Taking h := Tr∗A⊗F L(q ⊗ h0) proves the theorem. Corollary 6.2. Let (A1, σ1) and (A2, σ2) be F-algebras with involution of the same type. Assume that A1 and A2 are Brauer equivalent. Let P ∈ XF, let ι : F → FP be a real closure of F at P and let Mℓ be any Morita equivalence as in (2), but starting from H (Aℓ ⊗F FP, σℓ ⊗ id) for ℓ = 1, 2. Then the following statements are equivalent: (i) signM1 (ii) signM2 (iii) signP ϑ = 0 for all involutions ϑ on A1 of the same type as σ1; (iv) signP ϑ = 0 for all involutions ϑ on A2 of the same type as σ2. Proof. By Theorem 6.1, the first two statements are equivalent to P ∈ Nil[A1, σ1] = Nil[A2, σ2]. Thus (i) ⇔ (ii). (i) ⇒ (iii) Let ϑ be as in (iii). Then ϑ = adh1iϑ and ϑ = Int(a)◦ σ1 for some invertible a ∈ Sym(A1, σ1). Thus, with notation as in Remark 3.13 and using Lemma 4.6 and Proposition 3.4 we have for any Morita equivalence M , ι signP ϑ = λP signM ι ι h1iϑ = λP signM ◦(ζa⊗id) ζ−1 a (h1iϑ) = λP signM1 ζ−1 a (h1iϑ) = λP signM1 ha−1iσ1 ι ι which is zero by the assumption. (ii) ⇒ (iv): This is the same proof as (i) ⇒ (iii) after replacing (A1, σ1) by (A2, σ2). For the remainder of the proof we may assume without loss of generality that A2 is a division algebra and that A1 ≃ Mm(A2) for some m ∈ N. (iii) ⇒ (iv): Let ϑ be any involution on A2. By the assumption, signP(t ⊗ ϑ) = 0, where t denotes the transpose involution. Since t is a positive involution, it follows that signP ϑ = 0, cf. Remark 4.2. (iv) ⇒ (ii): The assumption implies that signM2 h = 0 for every hermitian form h of dimension 1 over (A2, σ2), which implies (ii) since all hermitian forms over (A2, σ2) can be diagonalized. (cid:3) Remark 6.3. Note that statement (iii) in Corollary 6.2 is equivalent to: signM1 for all diagonal hermitian forms h of dimension one over (A1, σ1). h = 0 (cid:3) ι ι Theorem 6.4. Let (A, σ) be an F-algebra with involution. There exists a finite set H = {h1, . . . , hs} of diagonal hermitian forms of dimension one over (A, σ) such that for every P ∈ XF \ Nil[A, σ], real closure ι : F → FP and Morita equivalence M as in (2) there exists h ∈ H such that signM ι h , 0. SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 17 Proof. For every P ∈ XF, choose a real closure ιP : F → FP and a Morita equivalence MP as in (2). By Corollary 3.5 we may assume without loss of generality and for the sake of simplicity that the map ιP is an inclusion. The algebra A ⊗F FP is isomorphic to a matrix algebra over DP, where DP ∈ {FP, HP, FP( √ −1), FP × FP}. There is a finite extension LP of F, LP ⊂ FP such that A ⊗F LP is isomorphic to a matrix algebra over EP, where EP ∈ {LP, (−1, −1)LP, LP( √ Since P ∈ UP we can write XF =SP∈XF UP. We know from [19, Chapter 3, Lemma 2.7, Theorem 2.8] that signQ(Tr∗LP/Fh1i) is the number of extensions of Q to LP. Thus, −1), LP × LP}, and P extends to LP. Let UP := {Q ∈ XF Q extends to LP}. UP =(cid:16)sign(Tr∗LP/Fh1i)(cid:17)−1 ({1, . . . , k}), where k is the dimension of the quadratic form Tr∗LP/Fh1i, and so UP is clopen in XF (and in particular compact). Therefore, and since XF is compact, there exists a finite number of orderings P1, . . . , Pℓ in XF such that XF = Sℓ i=1 UPi. Now let P ∈ {P1, . . . , Pℓ} and let LP be as before. By Theorem 6.1 we have that for every MQ Q ∈ UP\Nil[A, σ] there exists a hermitian form hQ over (A, σ) such that sign ιQ hQ , 0. By Corollary 6.2 and Remark 6.3 we may assume that hQ is diagonal of dimension one. Consider the total signature map µQ : XF −→ Z, P 7−→ signMP µ−1 Q (Z \ {0}). UP \ Nil[A, σ] = [Q∈UP\Nil[A,σ] hQ. Then (10) ιP Consider the continuous map λP : XLP −→ XF, R 7−→ R ∩ F. We have Q ∈ UP \ Nil[A, σ] if and only if some extension Q′ of Q to LP is in XLP \ Nil[A ⊗F LP, σ ⊗ id] (this follows from the fact that the ordered fields (F, Q) and (LP, Q′) have a common real closure) if and only if Q ∈ λ(cid:0)XLP \ Nil[A ⊗F LP, σ ⊗ id](cid:1). We observe that XLP \ Nil[A ⊗F LP, σ ⊗ id] is clopen and compact since Nil[A ⊗F LP, σ ⊗ id] is either ∅ or the whole of XLP, which follows from the fact that A ⊗F LP is a matrix algebra over one of LP, (−1, −1)LP, LP( √ −1), LP × LP. UP \ Nil[A, σ] = λ(cid:0)XLP \ Nil[A ⊗F LP, σ ⊗ id](cid:1) UP =(cid:0)UP ∩ Nil[A, σ](cid:1) ∪ [Q∈UP\Nil[A,σ] is compact and thus closed. Thus UP ∩ Nil[A, σ] is open in UP. Using (10) we can write µ−1 Q (Z \ {0}). Hence, Now µ−1 Q (Z \ {0}) = (sign adhQ)−1(Z \ {0}) by Lemma 4.6, which is open since sign adhQ is continuous by Remark 4.4. Thus, since UP is compact, there exist Q1, . . . , Qt ∈ UP \ Nil[A, σ] such that UP =(cid:0)UP ∩ Nil[A, σ](cid:1) ∪ t[i=1 µ−1 Qi (Z \ {0}). 18 VINCENT ASTIER AND THOMAS UNGER Corollary 6.5. Let (A, σ) be an F-algebra with involution. The set Nil[A, σ] is clopen in XF. In other words, for every Q ∈ UP\ Nil[A, σ] one of sign Now let HP = {hQ1, . . . , hQt}. Letting H =Sℓ Proof. By Theorem 6.4 we have Nil[A, σ] =Ts follows from Lemma 4.6 and Remark 4.4. MQ ιQ hQi (i = 1, . . . , t) is nonzero. i=1 HPi finishes the proof. (cid:3) (cid:3) i=1{P ∈ XF signMP ιP (cid:3) hi = 0}. The result (cid:3) At this stage we have established all results that are needed for the definition of the H-signature in Definion 3.9. In the final two sections we show that the total H- signature of a hermitian form is continuous and we reformulate the Knebusch trace formula in terms of the H-signature. 7. Continuity of the Total H-Signature Map of a Hermitian Form Let h be a hermitian form over (A, σ). With reference to Definition 3.9 we denote by signH h the total H-signature map of h: XF −→ Z, P 7−→ signH P h. Lemma 7.1. Let H = (h1, . . . , hs) be as in Definition 3.9. There is a finite partition of XF into clopens XF = Nil[A, σ] ∪ ℓ [i=1 Zi, such that for every i ∈ {1, . . . , ℓ} one of the total H-signature maps signH h1, . . . , signH hs is constant non-zero on Zi. Proof. For r = 1, . . . , s, let By Lemma 4.6 we have Yr := {P ∈ XF signH P hi = 0, i = 1, . . . , r}. Yr = r\i=1 {P ∈ XF signP adhi = 0}. Thus each Yr is clopen. We have Y0 := XF ⊇ Y1 ⊇ · · · ⊇ Ys−1 ⊇ Ys = Nil[A, σ] and therefore, XF = (Y0 \ Y1) ∪(Y1 \ Y2) ∪ · · · ∪(Ys−1 \ Ys) ∪ Nil[A, σ]. Let r ∈ {0, . . . , s − 1} and consider Yr \ Yr+1. By the definition of Y1, . . . , Ys the map signH hr+1 is never 0 on Yr \ Yr+1. Furthermore, signH hr+1 only takes a finite number of values k1, . . . , km. Now observe that there exists a λ ∈ {1, 2} such that signH hr+1 = sign adhr+1 1 λ on Yr \ Yr+1. This follows from Lemma 4.6 and Definition 3.9 for P ∈ Yr \ Yr+1. SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 19 Therefore, (cid:0)signH hr+1(cid:1)−1(ki) ∩ (Yr \ Yr+1) =(cid:0)sign adhr+1(cid:1)−1(λki) ∩ (Yr \ Yr+1), which is clopen by Remark 4.4. It follows that Yr \ Yr+1 is covered by finitely many disjoint clopen sets on which the map signH hr+1 has constant non-zero value. The result follows since the sets Yr \ Yr+1 for r = 0, . . . , s − 1 form a partition of XF \ Nil[A, σ]. (cid:3) (cid:3) Theorem 7.2. Let h be a hermitian form over (A, σ). The total H-signature of h, is continuous. signH h : XF −→ Z, P 7−→ signH P h Proof. We use the notation and the conclusion of Lemma 7.1. Since Nil[A, σ] and 1, . . . , ℓ. the sets Zi are clopen, it suffices to show that (signH h)Zi is continuous for every i = Let i ∈ {1, . . . , ℓ}, ki ∈ Z \ {0} and j ∈ {1, . . . , s} be such that signH h j = ki on Zi. Let k ∈ Z. Then (cid:0)(signH h)Zi(cid:1)−1(k) = {P ∈ Zi signH = {P ∈ Zi ki signH = {P ∈ Zi ki signH = {P ∈ Zi signH P h = k} P h = kik} P h = k signH P h j} P (ki × h ⊥ −(k × h j)) = 0}. (cid:0)(signH h)Zi(cid:1)−1(k) = {P ∈ Zi signP adki×h⊥−(k×h j ) = 0}, (cid:3) It follows from Lemma 4.6 that which is clopen by Remark 4.4. (cid:3) 8. The Knebusch Trace Formula for H-Signatures Theorem 8.1. Let H = (h1, . . . , hs) be as in Definition 3.9. Let P ∈ XF. Let L/F be a finite extension of ordered fields and let h be a hermitian form over (A ⊗F L, σ ⊗ id). Then, with H ⊗ L := (h1 ⊗ L, . . . , hs ⊗ L), we have signH signH⊗L Q h. Proof. By Theorem 5.1 (and using its notation), we know that P (Tr∗A⊗F Lh) = XP⊆Q∈XL rXi=1 (Tr∗A⊗F Lh) = signM ι sign(ωi◦ι0)(M ) h, ωi◦ι1 for any ι and M . Fix a real closure ι : F → FP and choose a Morita equivalence M such that signM P (cf. Remark 3.14). We only have to check that sign(ωi◦ι0)(M ) ι signH⊗L = signH for i = 1, . . . , r. ωi◦ι1 = Qi 20 VINCENT ASTIER AND THOMAS UNGER By definition of M , there is a k ∈ {1, . . . , s} such that signM ι hk > 0. To check that sign(ωi◦ι0)(M ) = signH⊗L Qi ωi◦ι1 and signM check that sign(ωi◦ι0)(M ) This follows from the fact that ωi◦ι1 (h j ⊗ L) = 0 for j = 1, . . . , k − 1 and sign(ωi◦ι0)(M ) ι h j = 0 for 1 ≤ j ≤ k− 1 for i = 1, . . . , r, it suffices to (hk ⊗ L) > 0. ωi◦ι1 sign(ωi◦ι0)(M ) ωi◦ι1 (hℓ ⊗ L) = signM ι hℓ for every 1 ≤ ℓ ≤ s, which we verify in the remainder of the proof. By definition, sign(ωi◦ι0)(M ) ωi◦ι1 Consider the commutative diagram (hℓ ⊗ L) = sign(ωi ◦ ι0)(M )[(idA ⊗ (ωi ◦ ι1))∗(hℓ ⊗ L)]. F ι 7♦♦♦♦♦♦ '❖❖❖❖❖❖❖ FP ι0 / L ⊗F FP ωi / Fi id L ι1 / L ⊗F FP / Fi ωi Thus, by Proposition 3.3, sign(ωi ◦ ι0)(M )[(idA ⊗ (ωi ◦ ι1))∗(hℓ ⊗ L)] = sign(ωi ◦ ι0)(M )[(idA ⊗ (ωi ◦ ι0 ◦ ι))∗(hℓ)]. Finally the commutative diagram 7♦♦♦♦♦♦ ωi◦ι0◦ι F '❖❖❖❖❖❖ ι Fi (ωi◦ι0)−1 FP together with Proposition 3.3 yields sign(ωi ◦ ι0)(M )[(idA ⊗ (ωi ◦ ι0 ◦ ι))∗(hℓ)] = sign M [(idA ⊗ ι)∗(hℓ)] = signM ι hℓ, which concludes the proof. (cid:3) (cid:3) References [1] E. Bayer-Fluckiger and H. W. Lenstra, Jr. Forms in odd degree extensions and self-dual nor- mal bases. Amer. J. Math., 112 (1990), no. 3, 359 -- 373. [2] E. Bayer-Fluckiger, R. Parimala, Galois cohomology of the classical groups over fields of [3] E. Bayer-Fluckiger, R. Parimala, Classical groups and the Hasse principle, Ann. of Math. (2) cohomological dimension ≤ 2, Invent. Math. 122 (1995), no. 2, 195 -- 229. 147 (1998), no. 3, 651 -- 693. [4] N. Bourbaki, ´El´ements de math´ematique. 23. Premi`ere partie: Les structures fondamentales de l'analyse. Livre II: Alg`ebre. Chapitre 8: Modules et anneaux semi-simples. Actualit´es Sci. Ind. no. 1261. Hermann, Paris (1958). [5] A. Frohlich and A.M. McEvett, Forms over rings with involution. J. Algebra 12 (1969), 79 -- 104. [6] N. Grenier-Boley, Groupe de Witt d'une alg`ebre simple centrale `a involution, PhD thesis, Universit´e de Franche-Comt´e (2004). [7] N. Jacobson, A note on hermitian forms, Bull. Amer. Math. Soc. 46 (1940), 264 -- 268. / / 7 ' / / 7 ' SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA 21 [8] M. Knebusch, On the uniqueness of real closures and the existence of real places. Comment. Math. Helv. 47 (1972), 260 -- 269. [9] M. Knebusch, A. Rosenberg, and R. Ware, Structure of Witt rings and quotients of Abelian group rings. Amer. J. Math. 94 (1972), 119 -- 155. [10] M.-A. Knus, Quadratic and Hermitian forms over rings. Grundlehren der Mathematischen Wissenschaften 294, Springer-Verlag, Berlin (1991). [11] M.-A. Knus, A.S. Merkurjev, M. Rost, J.-P. Tignol, The Book of Involutions, Coll. Pub. 44, Amer. Math. Soc., Providence, RI (1998). [12] T.Y. Lam, Introduction to quadratic forms over fields, Graduate Studies in Mathematics 67, American Mathematical Society, Providence, RI (2005). [13] D.W. Lewis, Forms over real algebras and the multisignature of a manifold, Advances in Math. 23 (1977), no. 3, 272 -- 284. [14] D.W. Lewis, The isometry classification of Hermitian forms over division algebras, Linear Algebra Appl. 43 (1982), 245 -- 272. [15] D.W. Lewis, J.-P. Tignol, On the signature of an involution, Arch. Math. 60 (1993), 128 -- 135. [16] D.W. Lewis, T. Unger, A local-global principle for algebras with involution and hermitian forms, Math. Z. 244 (2003), 469 -- 477. [17] D.W. Lewis, T. Unger, Hermitian Morita theory: a matrix approach, Irish Math. Soc. Bull. 62 (2008), 37 -- 41. [18] A. Qu´eguiner. Signature des involutions de deuxi`eme esp`ece. Arch. Math. 65 (1995), no. 5, 408 -- 412. [19] W. Scharlau, Quadratic and Hermitian Forms, Grundlehren der Mathematischen Wis- senschaften 270, Springer-Verlag, Berlin (1985). School of Mathematical Sciences, University College Dublin, Belfield, Dublin 4, Ireland E-mail address: [email protected], [email protected] 8 1 0 2 r a M 3 1 ] . A R h t a m [ 8 v 6 5 9 0 . 3 0 0 1 : v i X r a SIGNATURES OF HERMITIAN FORMS AND THE KNEBUSCH TRACE FORMULA: ERRATUM VINCENT ASTIER AND THOMAS UNGER (I) Our description of the adjoint involution adϕ0 at the end of §2.3 should be corrected as follows: adϕ0 (X) = Φ−1 0 ϑt(X)Φ0. Consequently, the definition of scaling in §2.4 should be corrected as follows: (M, h) 7→ (M, Φ0h). The second line of equation (3.3) should be corrected as follows: h ⊗ FP 7−→ ξ∗P(h ⊗ FP) 7−→ ΦPξ∗P(h ⊗ FP) 7−→ N (h ⊗ FP). In the proof of Lemma 3.10, the following corrections should be made: Firstly, the end of the displayed sequence of maps should be corrected as follows: hImiadϕP 7−→ ΦPhImiadϕP = hΦPiϑt P . Secondly, the last two sentences of the proof should be: "By collapsing, the matrix −1), −) ΦP now corresponds to a quadratic form over FP, a hermitian form over (FP( √ or a hermitian form over (HP, −). Thus signN h1iσ = sign ϕP." ι (b) Lines 4, 5 of p. 930 should read: exchange involution. Then h is hyperbolic by Lemma 2.1(iv) and we let sign h := 0. For the sake of completeness we point out where some small changes should be made. (a) In §3.1, the following case should be added: (II) In [1, §3.1] we inadvertently omitted dealing with the case (H × H,b). As W(H × H,b) = W(R × R,b) = 0, the omission does not affect our reasoning or our results. (g) Let h be a hermitian form over (H × H,b), where [(x, y) = (y, x) is the Note that the cases of skew-hermitian forms over (C, −), (R×R,b) and (H×H,b) In addition, we have the group isomorphisms W−1(R, idR) = W(R × R,b) = W(H × H,b) = 0, W−1(H, −) ≃ Z/2Z. If σ is of the second kind, recall that Z(A) = F( √d). Now if d <P 0, then DP is equal to FP( √ −1), whereas if d >P 0, then DP is equal to FP × FP or HP × HP and A ⊗F FP is a direct product of two simple algebras. Furthermore, we may choose (DP, ϑP) = (FP( √ can be reduced to (c), (d) and (g), respectively, by Lemma 2.1. (c) The penultimate sentence of Remark 3.1 should read: (d) On p. 930, (ii) should read: −1), −) in the first case and (DP, ϑP) = (FP × FP,b), (DP, ϑP) = (HP × HP,b) in the second case, again by Morita theory (e) On p. 931, line 1 should read resp. (scaling). (f) On p. 931, line 20 should read (DP, ϑP) ∈ {(FP, idFP ), (HP, −), (FP( √ D1 ∈ {L1, (−1, −1)L1, L1( √ {P ∈ XF DP = FP × FP or DP = HP × HP} if σ is unitary −1), L1 × L1, (−1, −1)L1 × (−1, −1)L1} −1), −), (FP × FP,b), (HP × HP,b)} (g) On p. 933, line 6 should read 2 VINCENT ASTIER AND THOMAS UNGER (i) On p. 942, line 22 should read −1), FP × FP, HP × HP} (h) On p. 942, line 20 should read DP ∈ {FP, HP, FP( √ EP ∈ {LP, (−1, −1)LP, LP( √ LP, (−1, −1)LP, LP( √ (j) On p. 943, line 7 should read −1), LP × LP, (−1, −1)LP × (−1, −1)LP} −1), LP × LP, (−1, −1)LP × (−1, −1)LP References [1] Astier, V., Unger, T.: Signatures of hermitian forms and the Knebusch trace formula, Math. Ann. 358, 925 -- 947 (2014) School of Mathematics and Statistics, University College Dublin, Belfield, Dublin 4, Ireland E-mail address: [email protected], [email protected]
1010.3408	1	1010	2010-10-17T13:09:12	Non-commutative Hom-Poisson algebras	[ "math.RA", "math-ph", "math-ph" ]	A Hom-type generalization of non-commutative Poisson algebras, called non-commutative Hom-Poisson algebras, are studied. They are closed under twisting by suitable self-maps. Hom-Poisson algebras, in which the Hom-associative product is commutative, are closed under tensor products. Through (de)polarization Hom-Poisson algebras are equivalent to admissible Hom-Poisson algebras, each of which has only one binary operation. Multiplicative admissible Hom-Poisson algebras are Hom-power associative.	math.RA	math	NON-COMMUTATIVE HOM-POISSON ALGEBRAS DONALD YAU Abstract. A Hom-type generalization of non-commutative Poisson algebras, called non- commutative Hom-Poisson algebras, are studied. They are closed under twisting by suitable self- maps. Hom-Poisson algebras, in which the Hom-associative product is commutative, are closed under tensor products. Through (de)polarization Hom-Poisson algebras are equivalent to admissi- ble Hom-Poisson algebras, each of which has only one binary operation. Multiplicative admissible Hom-Poisson algebras are Hom-power associative. 1. Introduction A Poisson algebra (A, {, }, µ) consists of a commutative associative algebra (A, µ) together with a Lie algebra structure {, }, satisfying the Leibniz identity: {µ(x, y), z} = µ({x, z}, y) + µ(x, {y, z}). Poisson algebras are used in many fields in mathematics and physics. In mathematics, Poisson al- gebras play a fundamental role in Poisson geometry [18], quantum groups [4, 5], and deformation of commutative associative algebras [8]. In physics, Poisson algebras are a major part of deforma- tion quantization [11], Hamiltonian mechanics [3], and topological field theories [17]. Poisson-like structures are also used in the study of vertex operator algebras [7]. One way to generalize Poisson algebras is to omit the commutativity requirement. Such a structure is called a non-commutative Poisson algebra. When the associative product happens to be commu- tative, one has a Poisson algebra. Some classification results of finite dimensional non-commutative Poisson algebras can be found in [12]. One can also think of a non-commutative Poisson algebra as a special case of a Leibniz pair [6]. The purpose of this paper is to study a twisted generalization of non-commutative Poisson alge- bras, called non-commutative Hom-Poisson algebras. In a non-commutative Hom-Poisson algebra A, there is a linear self-map α (the twisting map) and two binary operations {, } (the Hom-Lie bracket) and µ (the Hom-associative product). The associativity, the Jacobi identity, and the Leib- niz identity in a non-commutative Poisson algebra are replaced by their Hom-type (i.e., α-twisted) analogues in a non-commutative Hom-Poisson algebra. In particular, (A, µ, α) is a Hom-associative algebra [13], and (A, {, }, α) is a Hom-Lie algebra [10]. If the twisting map is the identity map, then a non-commutative Hom-Poisson algebra reduces to a non-commutative Poisson algebra. Most of our results are about Hom-Poisson algebras, which are non-commutative Hom-Poisson algebras in which the Hom-associative products are commutative. Hom-Poisson algebras were de- fined in [15] by Makhlouf and Silvestrov. It is shown in [15] that Hom-Poisson algebras play the Date: April 24, 2018. 2000 Mathematics Subject Classification. 17A15,17B63. Key words and phrases. Non-commutative Hom-Poisson algebras, admissible Hom-Poisson algebras, Hom-power associative algebra. 1 2 DONALD YAU same role in the deformation of commutative Hom-associative algebras as Poisson algebras do in the deformation of commutative associative algebras. Other Hom-type algebras are studied in [13, 14] and [19] - [26] and the references therein. The rest of this paper is organized as follows. In section 2 non-commutative Hom-Poisson algebras are defined. It is shown that, just like Poisson algebras, Hom-Poisson algebras are closed under tensor products in a non-trivial way (Theorem 2.9). It should be noted that Hom-Lie algebras are not closed under tensor products in any non-trivial way. In section 3 it is shown that non-commutative Hom- Poisson algebras are closed under suitable twistings by weak morphisms (Theorem 3.2). This is a unique feature for Hom-type algebras, as non-commutative Poisson algebras are not closed under such twistings. Using a special case of this result, several classes of (non-commutative) Hom-Poisson algebras are constructed. In section 4 it is shown that Hom-Poisson algebras are equivalent to admissible Hom-Poisson algebras, each of which has one twisting map and only one binary operation. The correspondence between Hom-Poisson algebras and admissible Hom-Poisson algebras is the Hom-version of the correspondence between Poisson algebras and admissible Poisson algebras [9, 16]. In section 5 it is shown that multiplicative admissible Hom-Poisson algebras are Hom-power associative. 2. Basic properties of non-commutative Hom-Poisson algebras In this section, we introduce (non-commutative) Hom-Poisson algebras and study tensor products of Hom-Poisson algebras. 2.1. Notations. We work over a fixed field k of characteristic 0. If V is a k-module and µ : V ⊗2 → V is a bilinear map, then µop : V ⊗2 → V denotes the opposite map, i.e., µop = µτ , where τ : V ⊗2 → V ⊗2 interchanges the two variables. For a linear self-map α : V → V , denote by αn the n-fold composition of n copies of α, with α0 ≡ Id. Let us begin with the basic definitions regarding Hom-algebras. Definition 2.2. (1) By a Hom-module we mean a pair (A, α) in which A is a k-module and α : A → A is a linear map, called the twisting map. (2) By a Hom-algebra we mean a triple (A, µ, α) in which (A, α) is a Hom-module and µ : A⊗2 → A is a bilinear map, called the multiplication. A Hom-algebra (A, µ, α) and the corresponding Hom-module (A, α) are often abbreviated to A. (3) A Hom-algebra (A, µ, α) is said to be multiplicative if αµ = µα⊗2. It is called commu- tative if µ = µop. Unless otherwise specified, an algebra (A, µ) with µ : A⊗2 → A is also regarded as a Hom-algebra (A, µ, Id) with identity twisting map. Given a Hom-algebra (A, µ, α), we often use the abbreviation for x, y ∈ A. µ(x, y) = xy Let us now recall the Hom-type generalizations of associative and Lie algebras from [13]. Definition 2.3. Let (A, µ, α) be a Hom-algebra. NON-COMMUTATIVE HOM-POISSON ALGEBRAS 3 (1) The Hom-associator asA : A⊗3 → A is defined as (2) The Hom-algebra A is called a Hom-associative algebra if it satisfies the Hom- asA = µ(µ ⊗ α − α ⊗ µ). (2.3.1) associative identity (3) The Hom-Jacobian JA : A⊗3 → A is defined as asA = 0. JA = µ(µ ⊗ α)(Id + σ + σ2), (2.3.2) (2.3.3) where σ(x ⊗ y ⊗ z) = z ⊗ x ⊗ y. The sum (Id + σ + σ2) is called a cyclic sum. (4) The Hom-algebra A is called a Hom-Lie algebra if it satisfies the Hom-Jacobi identity JA = 0. (2.3.4) When the twisting map α is the identity map, the above definitions reduce to the usual definitions of the associator, an associative algebra, the Jacobian, and a Lie algebra. Examples of Hom- associative and Hom-Lie algebras can be found in [13, 19, 20]. In terms of elements x, y, z ∈ A, the Hom-associator and the Hom-Jacobian are asA(x, y, z) = (xy)α(z) − α(x)(yz), JA(x, y, z) = (xy)α(z) + (zx)α(y) + (yz)α(x). Let us recall the definition of a non-commutative Poisson algebra [12]. Definition 2.4. A non-commutative Poisson algebra (A, {, }, µ) consists of (1) a Lie algebra (A, {, }) and (2) an associative algebra (A, µ) such that the Leibniz identity {x, yz} = {x, y}z + y{x, z} is satisfied for all x, y, z ∈ A. A Poisson algebra is a non-commutative Poisson algebra (A, {, }, µ) in which µ is commutative. A morphism of non-commutative Poisson algebras is a linear map that is a morphism of the underlying Lie algebras and associative algebras. In a non-commutative Poisson algebra (A, {, }, µ), the Lie bracket {, } is called the Poisson bracket, and µ is called the associative product. The Leibniz identity says that {x, −} is a derivation with respect to the associative product. It can be rewritten in element-free form as where (1 2)(x ⊗ y ⊗ z) = y ⊗ x ⊗ z. {, }(Id ⊗ µ) = µ ({, } ⊗ Id + (Id ⊗ {, })(1 2)) , Hom-Poisson algebras are first introduced in [14] by Makhlouf and Silvestrov. We now define the Hom-type generalization of a non-commutative Poisson algebra. Definition 2.5. A non-commutative Hom-Poisson algebra (A, {, }, µ, α) consists of (1) a Hom-Lie algebra (A, {, }, α) and (2) a Hom-associative algebra (A, µ, α) 4 DONALD YAU such that the Hom-Leibniz identity {, }(α ⊗ µ) = µ ({, } ⊗ α + (α ⊗ {, })(1 2)) (2.5.1) is satisfied. A Hom-Poisson algebra is a non-commutative Hom-Poisson algebra (A, {, }, µ, α) in which µ is commutative [14]. In a non-commutative Hom-Poisson algebra (A, {, }, µ, α), the operations {, } and µ are called the Hom-Poisson bracket and the Hom-associative product, respectively. In terms of elements x, y, z ∈ A, the Hom-Leibniz identity says {α(x), yz} = {x, y}α(z) + α(y){x, z}, (2.5.2) where as usual µ(x, y) is abbreviated to xy. By the anti-symmetry of the Hom-Poisson bracket {, }, the Hom-Leibniz identity is equivalent to {xy, α(z)} = {x, z}α(y) + α(x){y, z}. A (non-commutative) Poisson algebra is exactly a multiplicative (non-commutative) Hom-Poisson algebra with identity twisting map. Let us now provide some basic properties of non-commutative Hom-Poisson algebras. Every associative algebra (A, µ) has a non-commutative Poisson algebra structure in which the Poisson bracket is the commutator bracket. The following result is the Hom-type analogue of this observation. Proposition 2.6. Let (A, µ, α) be a Hom-associative algebra. Then A− = (A, [, ] = µ − µop, µ, α) is a non-commutative Hom-Poisson algebra. Proof. It is proved in [13] (Proposition 1.6) that (A, [, ], α) is a Hom-Lie algebra. Indeed, the com- mutator bracket [, ] is obviously anti-symmetric. One can write out all twelve terms (in terms of µ) in the Hom-Jacobian of A− and observe that their sum is zero. To check the Hom-Leibniz identity (2.5.2) for A−, we compute as follows: [x, y]α(z) + α(y)[x, z] − [α(x), yz] = (xy)α(z) − (yx)α(z) + α(y)(xz) − α(y)(zx) − α(x)(yz) + (yz)α(x) = asA(x, y, z) − asA(y, x, z) + asA(y, z, x). Since asA = 0, we conclude that A− satisfies the Hom-Leibniz identity. (cid:3) Next we study tensor products of Hom-Poisson algebras. For motivation, recall that Lie algebras are not closed under tensor products. The same is true for Hom-Lie algebras. On the other hand, Poisson algebras are closed under tensor products. In the rest of this section, we show that Hom- Poisson algebras are also closed under tensor products. We need the following preliminary result. Lemma 2.7. Let (A, µ, α) be a commutative Hom-associative algebra. Then the expression (xy)α(z) is invariant under permutations of x, y, z ∈ A. In other words, we have for all the permutations π on three letters. µ(µ ⊗ α) = µ(µ ⊗ α)π NON-COMMUTATIVE HOM-POISSON ALGEBRAS 5 Proof. Pick x, y, z ∈ A. Then (xy)α(z) = (yx)α(z) = α(y)(xz) = (xz)α(y). So the expression (xy)α(z) is invariant under the transpositions (1 2) and (2 3). The proof is complete because these two transpositions generate the symmetric group on three letters. (cid:3) Definition 2.8. Let (A, {, }, µ, α) be a quadruple in which (A, α) is a Hom-module and {, }, µ : A⊗2 → A are bilinear operations. Define its Hom-associator asA and Hom-Jacobian JA using µ and {, }, respectively, i.e., asA = µ(µ ⊗ α − α ⊗ µ), JA = {, }({, } ⊗ α)(Id + σ + σ2), where σ(x ⊗ y ⊗ z) = z ⊗ x ⊗ y. Now we are ready to prove that Hom-Poisson algebras are closed under tensor products. Theorem 2.9. Let (Ai, {, }i, µi, αi) be Hom-Poisson algebras for i = 1, 2, and let A = A1 ⊗ A2. Define the operations α : A → A and µ, {, } : A⊗2 → A by: α = α1 ⊗ α2, µ(x1 ⊗ x2, y1 ⊗ y2) = µ1(x1, y1) ⊗ µ2(x2, y2), {x1 ⊗ x2, y1 ⊗ y2} = {x1, y1}1 ⊗ µ2(x2, y2) + µ1(x1, y1) ⊗ {x2, y2}2 for xi, yi ∈ Ai. Then (A, {, }, µ, α) is a Hom-Poisson algebra. Proof. That (A, µ, α) is a commutative Hom-associative algebra follows from the commutativity and Hom-associativity of both µi. Also, the commutativity of the µi and the anti-symmetry of the {, }i imply the anti-symmetry of {, }. It remains to prove the Hom-Jacobi identity and the Hom-Leibniz identity in A. To simplify the typography, we abbreviate µ1, µ2, and µ using juxtaposition and drop the sub- scripts in {, }i and αi. Pick x = x1 ⊗ x2, y = y1 ⊗ y2, and z = z1 ⊗ z2 ∈ A. Then {{x, y}, α(z)} = {{x1, y1} ⊗ x2y2, α(z1) ⊗ α(z2)} + {x1y1 ⊗ {x2, y2}, α(z1) ⊗ α(z2)} = {{x1, y1}, α(z1)} ⊗ (x2y2)α(z2) + {x1, y1}α(z1) ⊗ {x2y2, α(z2)} + {x1y1, α(z1)} ⊗ {x2, y2}α(z2) + (x1y1)α(z1) ⊗ {{x2, y2}, α(z2)} The cyclic sum over x, y, z of the term a is 0 by Lemma 2.7 and the Hom-Jacobi identity in A1. Likewise, the cyclic sum over x, y, z of the term d is 0 by Lemma 2.7 and the Hom-Jacobi identity in A2. It follows that the Hom-Jacobian JA(x, y, z) consists of only the cyclic sum over x, y, z of the terms b and c. Using the Hom-Leibniz identity in the Ai, the terms b and c become a {z {z c } } b {z {z d } } b = {x1, y1}α(z1) ⊗ {x2y2, α(z2)} = {x1, y1}α(z1) ⊗ α(x2){y2, z2} + {x1, y1}α(z1) ⊗ {x2, z2}α(y2), c = {x1y1, α(z1)} ⊗ {x2, y2}α(z2) = α(x1){y1, z1} ⊗ {x2, y2}α(z2) + {x1, z1}α(y1) ⊗ {x2, y2}α(z2). 6 DONALD YAU Therefore, the Hom-Jacobian of A applied to x ⊗ y ⊗ z is: JA(x, y, z) = {x1, y1}α(z1) ⊗ α(x2){y2, z2} + {x1, y1}α(z1) ⊗ {x2, z2}α(y2) + α(x1){y1, z1} ⊗ {x2, y2}α(z2) + {x1, z1}α(y1) ⊗ {x2, y2}α(z2) + {z1, x1}α(y1) ⊗ α(z2){x2, y2} + {z1, x1}α(y1) ⊗ {z2, y2}α(x2) + α(z1){x1, y1} ⊗ {z2, x2}α(y2) + {z1, y1}α(x1) ⊗ {z2, x2}α(y2) + {y1, z1}α(x1) ⊗ α(y2){z2, x2} + {y1, z1}α(x1) ⊗ {y2, x2}α(z2) + α(y1){z1, x1} ⊗ {y2, z2}α(x2) + {y1, x1}α(z1) ⊗ {y2, z2}α(x2) . b1 {z b3 c1 c3 {z {z {z {z {z c5 b5 } } } } } } b2 {z b4 c2 c4 {z {z {z {z {z c6 b6 } } } } } } Using the anti-symmetry of {, }i and the commutativity of µi, observe that the following sums are 0: b1 + c6, b2 + c3, b3 + c2, b4 + c5, b5 + c4, and b6 + c1. This shows that (A, {, }, α) satisfies the Hom-Jacobi identity JA = 0. Finally, we check the Hom-Leibniz identity in A. Using the Hom-associativity and the Hom- Leibniz identity in the Ai and Lemma 2.7, we compute as follows: {xy, α(z)} = {x1y1 ⊗ x2y2, α(z1) ⊗ α(z2)} = {x1y1, α(z1)} ⊗ (x2y2)α(z2) + (x1y1)α(z1) ⊗ {x2y2, α(z2)} = α(x1){y1, z1} ⊗ α(x2)(y2z2) + {x1, z1}α(y1) ⊗ (x2z2)α(y2) + α(x1)(y1z1) ⊗ α(x2){y2, z2} + (x1z1)α(y1) ⊗ {x2, z2}α(y2) = (α(x1) ⊗ α(x2)) ({y1, z1} ⊗ y2z2 + y1z1 ⊗ {y2, z2}) + ({x1, z1} ⊗ x2z2 + x1z1 ⊗ {x2, z2}) (α(y1) ⊗ α(y2)) = α(x){y, z} + {x, z}α(y). This shows that A satisfies the Hom-Leibniz identity. (cid:3) Setting αi = IdAi in Theorem 2.9, we recover the following well-known result about Poisson algebras. Corollary 2.10. Let (Ai, {, }i, µi) be Poisson algebras for i = 1, 2, and let A = A1 ⊗ A2. Define the operations µ, {, } : A⊗2 → A by: µ(x1 ⊗ x2, y1 ⊗ y2) = µ1(x1, y1) ⊗ µ2(x2, y2), {x1 ⊗ x2, y1 ⊗ y2} = {x1, y1}1 ⊗ µ2(x2, y2) + µ1(x1, y1) ⊗ {x2, y2}2 for xi, yi ∈ Ai. Then (A, {, }, µ) is a Poisson algebra. 3. Twistings of non-commutative Hom-Poisson algebras In this section, we first observe that non-commutative Hom-Poisson algebras are closed under twisting by suitable self-maps. A special case of this observation is that non-commutative Poisson algebras give rise to multiplicative non-commutative Hom-Poisson algebras via twisting by self- morphisms (Corollary 3.4). To study whether these twistings of non-commutative Poisson algebras NON-COMMUTATIVE HOM-POISSON ALGEBRAS 7 actually give rise to non-commutative Poisson algebras, we introduce the concept of rigidity in Definition 3.5. Twistings and rigidity of some classes of Poisson algebras are then studied. Definition 3.1. Let (A, {, }, µ, α) be a quadruple in which (A, α) is a Hom-module and {, }, µ : A⊗2 → A are bilinear operations. (1) A is multiplicative if α{, } = {, }α⊗2 and αµ = µα⊗2. (2) Let (B, {, }B, µB, αB) be another such quadruple. A weak morphism f : A → B is a linear map such that f {, } = {, }Bf ⊗2 and f µ = µBf ⊗2. A morphism f : A → B is a weak morphism such that f α = αBf . Note that a quadruple (A, {, }, µ, α) is multiplicative if and only if the twisting map α : A → A is a morphism. The following result says that (non-commutative) Hom-Poisson algebras are closed under twisting by self-weak morphisms. Theorem 3.2. Let (A, {, }, µ, α) be a (non-commutative) Hom-Poisson algebra and β : A → A be a weak morphism. Then Aβ = (A, {, }β = β{, }, µβ = βµ, βα) is also a (non-commutative) Hom-Poisson algebra. Moreover, if A is multiplicative and β is a morphism, then Aβ is a multiplicative (non-commutative) Hom-Poisson algebra. Proof. If µ is commutative, then clearly so is µβ. The rest of the proof applies whether µ is commutative or not. Since β is compatible with µ and {, }, the Hom-associators and the Hom-Jacobians of A and Aβ are related as: and β2asA = (βµ) (βµ ⊗ βα − βα ⊗ βµ) = asAβ β2JA = (β{, })(β{, } ⊗ βα)(Id + σ + σ2) = JAβ . This implies that (A, µβ, βα) is a Hom-associative algebra and that (A, {, }β, βα) is a Hom-Lie algebra. Likewise, applying β2 to the Hom-Leibniz identity (2.5.1) in A, we obtain (β{, })(βα ⊗ βµ) = (βµ) (β{, } ⊗ βα + (βα ⊗ β{, })(1 2)) , which is the Hom-Leibniz identity in Aβ. For the multiplicativity assertion, suppose A is multiplicative and β is a morphism. Let ν denote either µ or {, }. Then (βα)νβ = (βα)(βν) = βνα⊗2β⊗2 = νβ(αβ)⊗2 = νβ(βα)⊗2. 8 DONALD YAU This shows that Aβ is multiplicative. (cid:3) Two special cases of Theorem 3.2 follow. The following result says that each multiplicative (non-commutative) Hom-Poisson algebra gives rise to a derived sequence of multiplicative (non- commutative) Hom-Poisson algebras. Corollary 3.3. Let (A, {, }, µ, α) be a multiplicative (non-commutative) Hom-Poisson algebra. Then is a multiplicative (non-commutative) Hom-Poisson algebra for each integer n ≥ 0. An =(cid:16)A, {, }(n) = αn{, }, µ(n) = αnµ, αn+1(cid:17) Proof. The multiplicativity of A implies that αn : A → A is a morphism. By Theorem 3.2 Aαn = An is a multiplicative (non-commutative) Hom-Poisson algebra. (cid:3) The following observation is the α = Id special case of Theorem 3.2. It gives a twisting construc- tion of multiplicative (non-commutative) Hom-Poisson algebras from (non-commutative) Poisson algebras. A result of this form was first given by the author in [20] for G-Hom-associative algebras. This twisting construction highlights the fact that the usual category of (non-commutative) Poisson algebras is not closed under twisting by self-morphisms, which, in view of Theorem 3.2, is in strong contrast with the category of (non-commutative) Hom-Poisson algebras. This is a major concep- tual difference between (non-commutative) Hom-Poisson algebras and (non-commutative) Poisson algebras. Corollary 3.4. Let (A, {, }, µ) be a (non-commutative) Poisson algebra and β : A → A be a mor- phism. Then is a multiplicative (non-commutative) Hom-Poisson algebra. Aβ = (A, {, }β = β{, }, µβ = βµ, β) In the setting of Corollary 3.4, two natural questions arise: Is the triple A′ β = (A, {, }β = β{, }, µβ = βµ) (3.4.1) a (non-commutative) Poisson algebra? If so, is it isomorphic to A itself? To study these questions, we introduce the following concepts. Definition 3.5. Let (A, {, }, µ) be a non-commutative Poisson algebra. (1) Given a morphism β : A → A, the triple A′ β in (3.4.1) is called the β-twisting of A. A twisting of A is a β-twisting of A for some morphism β : A → A. (2) The β-twisting A′ β of A is called trivial if {, }β = 0 = µβ. A′ β is called non-trivial if either {, }β 6= 0 or µβ 6= 0. (3) A is called rigid if every twisting of A is either trivial or isomorphic to A. The reader is cautioned not to confuse the above notion of rigidity with Gerstenhaber's [8]. Every non-commutative Poisson algebra A has at least one trivial twisting A′ β, which is obtained by taking β = 0. As we will see below, it is possible for a β-twisting to be trivial even if β is not the zero map. On the other hand, if a non-commutative Poisson algebra A has either a non-zero NON-COMMUTATIVE HOM-POISSON ALGEBRAS 9 Lie bracket or a non-zero associative product, then the IdA-twisting is non-trivial. As we will see below, it is possible for a β-twisting to be isomorphic to A even if β is not the identity map. The following result gives some basic criteria for non-rigidity. Proposition 3.6. Let (A, {, }, µ) be a non-commutative Poisson algebra. Suppose there exists a morphism β : A → A such that either: (1) µβ = βµ is not associative or (2) {, }β = β{, } does not satisfy the Jacobi identity. Then A is not rigid. Proof. The β-twisting A′ satisfy the Jacobi identity. For the same reason, the β-twisting A′ β is non-trivial, since otherwise µβ would be associative and {, }β would β cannot be isomorphic to A. (cid:3) Using Proposition 3.6 we now consider some classes of non-commutative Poisson algebras and study their (non-)rigidity. Example 3.7 (Free associative algebras are not rigid). Let A = (k(S), µ) be the free unital associative algebra on a non-empty set S. Equip A with the non-commutative Poisson algebra structure in which the Poisson bracket is the commutator bracket (Proposition 2.6). We claim that A is not rigid. By Proposition 3.6 it suffices to show that µα = αµ is not associative for some morphism α on A. Pick an element X ∈ S, and let α : A → A be the morphism determined for Y ∈ S by α(Y ) =(1 + X if Y = X, if Y 6= X. Y For n ≥ 1, we have The associator of µα applied to (X, X, α(X)) is: µα(µα(X, X), α(X)) − µα(X, µα(X, α(X))) αn(X) = n + X. =(cid:0)α2(X)(cid:1)3 = (2 + X)3 − (1 + X)(2 + X)(3 + X) − α(X)α2(X)α3(X) 6= 0. So µα is not associative, and A is not rigid. (cid:3) Example 3.8 (Matrix algebras are not rigid). Let A = (Mn(k), µ) be the associative algebra of n × n matrices over k for some n ≥ 2. Equip A with the non-commutative Poisson algebra structure in which the Poisson bracket is the commutator bracket (Proposition 2.6). We claim that A is not rigid. By Proposition 3.6 it suffices to show that µα = αµ is not associative for some morphism α on A. To construct such a morphism, consider the diagonal matrix D = 1/2 0 1 0 ... ... 0 0   · · · · · · . . . · · · . 0 0 ... 1   From here on, a matrix (aij ) ∈ A with aij = 0 whenever i ≥ 3 or j ≥ 3 is abbreviated to its upper left 2 × 2 submatrix α((aij )) = D(aij )D−1 = a12/2 · · · · · · a22 ... . . . · · · an2 a1n/2 a2n ... ann .   a11 2a21 ... 2an1   a21 a22!. a11 a12 1 0! ∈ A. X = 1 1 10 DONALD YAU The inverse of D is the same as D except that its (1, 1)-entry is 2. There is a morphism α : A → A given by Consider such a matrix Then a quick computation shows that whereas µα(µα(X, X), α(X)) = α2(X 3) = 3 1/2 1 !, µα(X, µα(X, α(X))) = α(cid:0)Xα(X)α2(X)(cid:1) = 5 8 6 3/8 1/4!. So µα is not associative, and A is not rigid. (cid:3) Example 3.9 (Hom-Poisson algebras from linear Poisson structures). Let g be a finite- dimensional Lie algebra, and let (S(g), µ) be its symmetric algebra. If {ei}n i=1 is a basis of g, then S(g) is the polynomial algebra k[e1, . . . , en]. Suppose the structure constants for g are given by Then the symmetric algebra S(g) becomes a Poisson algebra with the Poisson bracket [ei, ej] = ck ijek. n Xk=1 ijek(cid:18) ∂F ∂ei ck {F, G} = 1 2 n Xi,j,k=1 ∂G ∂ej − ∂F ∂ej ∂G ∂ei(cid:19) (3.9.1) for F, G ∈ S(g). This Poisson algebra structure on S(g) is called the linear Poisson structure. Note that {ei, ej} = [ei, ej]. Let α : g → g be a Lie algebra morphism. It extends to an associative algebra morphism α : S(g) → S(g) determined by α(F (e1, . . . , en)) = F (α(e1), . . . , α(en)). Moreover, we claim that α respects the Poisson bracket (3.9.1), i.e., α is a morphism of Poisson algebras on S(g). Indeed, suppose that α{F, G} = {α(F ), α(G)} and α{H, G} = {α(H), α(G)} for some F, G, H ∈ S(g). Then the Leibniz identity {F H, G} = {F, G}H + F {H, G} NON-COMMUTATIVE HOM-POISSON ALGEBRAS 11 implies Likewise, if α{F H, G} = α({F, G})α(H) + α(F )α({H, G}) = {α(F ), α(G)}α(H) + α(F ){α(H), α(G)} = {α(F )α(H), α(G)} = {α(F H), α(G)}. α{F, G} = {α(F ), α(G)} and α{F, H} = {α(F ), α(H)}, then the Leibniz identity implies {F, GH} = {F, G}H + G{F, H} α{F, GH} = {α(F ), α(GH)}. Therefore, since S(g) is the polynomial algebra k[e1, . . . , en], to check that α{F, G} = {α(F ), α(G)} for all F, G ∈ S(g), it suffices to show α{ei, ej} = {α(ei), α(ej)} (3.9.2) for all i, j ∈ {1, . . . , n}. The desired identity (3.9.2) is true because {ei, ej} = [ei, ej] and α is a Lie algebra morphism on g. Therefore, given a Lie algebra morphism α : g → g, the extended map α : S(g) → S(g) is a morphism of Poisson algebras. By Corollary 3.4 there is a multiplicative Hom-Poisson algebra S(g)α = (S(g), {, }α = α{, }, µα = αµ, α), which reduces to the linear Poisson structure S(g) if α = Id. (cid:3) Example 3.10 (S(sl2) is not rigid). In this example, we show that the symmetric algebra (S(sl2), µ) on the Lie algebra sl2, equipped with the linear Poisson structure (3.9.1), is not rigid in the sense of Definition 3.5. The Lie algebra sl2 has a basis {e, f, h}, with respect to which the Lie bracket is given by [h, e] = 2e, [h, f ] = −2f, [e, f ] = h. To show that S(sl2) = k[e, f, h] is not rigid, consider the Lie algebra morphism α : sl2 → sl2 given by α(e) = λe, α(f ) = λ−1f, α(h) = h, where λ ∈ k is a fixed scalar with λ 6= 0, 1. As in Example 3.9, denote by α : S(sl2) → S(sl2) the extended map, which is a Poisson algebra morphism. By Proposition 3.6, the Poisson algebra S(sl2) is not rigid if µα = αµ is not associative. We have µα(µα(e, h), h) − µα(e, µα(h, h)) = α2(e)α2(h)α(h) − α(e)α2(h)α2(h) = (λ2 − λ)eh2, which is not 0 in the symmetric algebra S(sl2) because λ 6= 0, 1. Therefore, µα is not associative, and the linear Poisson structure on S(sl2) is not rigid. (cid:3) 12 DONALD YAU Example 3.11 (Hom-Poisson algebras from Poisson manifolds). In this example, we discuss how multiplicative Hom-Poisson algebras arise from Poisson manifolds, which include symplectic manifolds and Poisson-Lie groups. The reader is referred to, e.g., [4, 18] for discussion of Poisson manifolds. The ground field in this example is the field of real numbers. Let M be a Poisson manifold. This means that M is a smooth manifold and that the commutative associative algebra (C∞(M ), µ) (under point-wise multiplication) of smooth R-valued functions on M is equipped with a Poisson algebra structure {, } : C∞(M ) × C∞(M ) → C∞(M ). Let N be another Poisson manifold. A Poisson map ϕ : M → N between two Poisson manifolds is a smooth map such that for all f, g ∈ C∞(N ). Given such a Poisson map, the induced map {f, g}ϕ = {f ϕ, gϕ} ϕ∗ : C∞(N ) → C∞(M ), ϕ∗(f ) = f ϕ is a morphism of Poisson algebras. Let ϕ : M → M be a Poisson map. By Corollary 3.4 the morphism ϕ∗ : C∞(M ) → C∞(M ) yields a multiplicative Hom-Poisson algebra C∞(M )ϕ∗ = (C∞(M ), {, }ϕ∗ = ϕ∗{, }, µϕ∗ = ϕ∗µ, ϕ∗) . If ϕ = IdM , then C∞(M )Id∗ is the original Poisson algebra C∞(M ). (cid:3) We now give sufficient conditions that guarantee that the Poisson algebra C∞(M ) of smooth functions on M is not rigid in the sense of Definition 3.5. Corollary 3.12. Let M be a Poisson manifold. Suppose there exist a Poisson map ϕ : M → M , x ∈ M , and f ∈ C∞(M ) such that the matrix Φ = f (ϕ2(x)) f (ϕ3(x)) f (ϕ(x)) f (ϕ2(x))! has non-zero trace and non-zero determinant. Then the Poisson algebra C∞(M ) is not rigid. Proof. Using Proposition 3.6 and the notations in Example 3.11, it suffices to show that µϕ∗ is not associative. That the matrix Φ has non-vanishing trace and determinant means f (ϕ2(x)) 6= 0 and Using the conditions in (3.12.1), we have: (cid:0)f (ϕ2(x))(cid:1)2 6= f (ϕ(x))f (ϕ3(x)). (3.12.1) µϕ∗ (µϕ∗ (f, f ), f ϕ) (x) =(cid:0)f (ϕ2(x))(cid:1)3 6= f (ϕ(x))f (ϕ2(x))f (ϕ3(x)) = µϕ∗ (f, µϕ∗ (f, f ϕ)) (x). This shows that µϕ∗ is not associative, so C∞(M ) is not rigid. (cid:3) The following example illustrates how the non-rigidity criterion in Corollary 3.12 can be applied. NON-COMMUTATIVE HOM-POISSON ALGEBRAS 13 Example 3.13 (C∞(R2n) is not rigid). The Euclidean space R2n is a Poisson manifold with Poisson structure n {f, g} = Xi,j=1(cid:18) ∂f ∂xi ∂g ∂xi+n − ∂f ∂xi+n ∂g ∂xi(cid:19) for f, g ∈ C∞(R2n). Using Corollary 3.12 we show that the Poisson algebra C∞(R2n) is not rigid in the sense of Definition 3.5. Consider the map ϕ : R2n → R2n defined as ϕ(a1, . . . , a2n) = (a1 + c1, . . . , a2n + c2n), where the cj ∈ R are fixed scalars, not all of which are zero, say, ci 6= 0. The map ϕ is a Poisson map by Chain Rule. Let f ∈ C∞(R2n) be the function and let 0 be the origin in R2n. For each k ≥ 0, we have f (a1, . . . , a2n) = ai, So and ϕk(0) = (kc1, . . . , kc2n). f (ϕ2(0)) = 2ci 6= 0 f (ϕ2(0))2 − f (ϕ(0))f (ϕ3(0)) = (2ci)2 − ci(3ci) = c2 i 6= 0. Therefore, by Corollary 3.12 the Poisson algebra C∞(R2n) is not rigid. (cid:3) Example 3.14 (Every Poisson structure on the Heisenberg algebra is rigid). Over the ground field of complex numbers, the Poisson algebra structures on the Heisenberg algebra are classified by Goze and Remm in [9] (section 2). In this example, we show that these Poisson algebras are all rigid in the sense of Definition 3.5. Let us first recall the Goze-Remm classification of Poisson algebra structures on the Heisenberg algebra [9]. Let H be the complex Heisenberg algebra, which is the three-dimensional complex Lie algebra with a basis {X, Y, Z} such that [X, Y ] = Z, [X, Z] = 0 = [Y, Z]. (3.14.1) The isomorphism classes of Poisson algebra structures on H are divided into two families. First there is a one-parameter family of Poisson algebra structures on H, P3,1(ζ) = {XY = ζZ}, where ζ is any complex number. The notation above means that in the Poisson algebra P3,1(ζ), the commutative associative product satisfies and the unspecified binary products of basis elements are all zero. The only other isomorphism class of Poisson algebra structure on H is the Poisson algebra XY = ζZ = Y X, (3.14.2) P3,2 = {X 2 = Z}. We aim to show that the Poisson algebras P3,1(ζ) and P3,2 are all rigid in the sense of Definition 3.5. 14 DONALD YAU Consider the Lie algebra morphisms on the Heisenberg algebra. It follows from the Heisenberg algebra relations (3.14.1) that a linear map α : H → H is a Lie algebra morphism if and only if its matrix with respect to the basis {X, Y, Z} is α =  a11 a21 a31 a12 a22 a32 0 0 b   with b = a11a22 − a21a12, (3.14.3) where the aij are arbitrary complex numbers. If α is a morphism on either P3,1(ζ) or P3,2, then we obtain further relations among the aij. (i) We first show that P3,1(ζ) is rigid. If α : P3,1(ζ) → P3,1(ζ) is a Poisson algebra morphism, then applying α to the relations (3.14.2) yields so Likewise, applying α to the relations in P3,1(ζ) yields ζb = ζ(a11a22 + a21a12), ζa21a12 = 0. X 2 = 0 = Y 2 (3.14.4) Conversely, if a Lie algebra morphism α : H → H satisfies (3.14.4) and (3.14.5), then it is a Poisson algebra morphism on P3,1(ζ). We consider the two cases: ζ = 0 and ζ 6= 0. ζa11a21 = 0 = ζa12a22. (3.14.5) • If ζ = 0, then (3.14.4) and (3.14.5) do not impose further relations on α in (3.14.3). Since P3,1(0) has a trivial associative product µ, every α-twisting of P3,1(0) also has a trivial Hom- associative product µα. With α as in (3.14.3), the induced Hom-Lie bracket as in Corollary 3.4 is determined by If b = 0, then [, ]α = 0 and the α-twisting [X, Y ]α = bZ. of P3,1(0) is trivial. If b 6= 0, then the map P3,1(0)′ α = (P3,1(0), [, ]α, µα) P3,1(0) → P3,1(0)′ α with   X 7→ X, Y 7→ Y, Z 7→ bZ is an isomorphism of Poisson algebras. In other words, the Poisson algebra P3,1(0) is rigid. • If ζ 6= 0, then the relations (3.14.4) and (3.14.5) become a21a12 = a11a21 = a12a22 = 0. NON-COMMUTATIVE HOM-POISSON ALGEBRAS 15 This means that α takes one of the following three forms: α1 =  α2 =  α3 =  0 a11 a22 0 a31 a32 a11 a12 0 0 a31 a32 0 0 a21 a22 a31 a32 0 0 a11a22 ,   0 0 0 0 0 0     with a12 6= 0, with a21 6= 0. For α = α2 or α3, we have α(Z) = 0, so [, ]α = 0 = µα. This implies that the α2-twisting and the α3-twisting of P3,1(ζ) (with ζ 6= 0) are both trivial. For α = α1, if a11 = 0 or a22 = 0, then α(Z) = 0. This implies that [, ]α = 0 = µα, so the α1-twisting of P3,1(ζ) (with ζ 6= 0) is trivial when a11a22 = 0. On the other hand, suppose a11a22 6= 0. Since µα(X, Y ) = ζa11a22Z = µα(Y, X), [X, Y ]α = a11a22Z, the map P3,1(ζ) → P3,1(ζ)′ α with   X 7→ X, Y 7→ Y, Z 7→ a11a22Z is an isomorphism of Poisson algebras. We have shown that when ζ 6= 0, every twisting of P3,1(ζ) is either trivial or isomorphic to P3,1(ζ) itself. Therefore, P3,1(ζ) is rigid. (ii) Next we show that P3,2 is rigid. Suppose α : P3,2 → P3,2 is a morphism of Poisson algebras. Applying α to the relations we obtain Y 2 = 0 and X 2 = Z, Conversely, if a Lie algebra morphism α : H → H satisfies (3.14.6), then it is a Poisson algebra morphism on P3,2. Therefore, the Poisson algebra morphisms α on P3,2 are a12 = 0 and a11a22 = a2 11. (3.14.6) and α4 =  0 0 a21 a22 a31 a32 0 0 0   with a11 6= 0. α5 =  0 a11 0 0 a21 a11 a31 a32 a2 11   Since α4(Z) = 0, we have [, ]α4 = 0 = µα4, so the α4-twisting of P3,2 is trivial. For α = α5, we have α(Z) = a2 11Z and µα(X, X) = a2 11Z = [X, Y ]α. P3,2 → (P3,2)′ X 7→ X, Y 7→ Y, Z 7→ a2 11Z DONALD YAU α with   16 So the map is an isomorphism of Poisson algebras. We have shown that every twisting of P3,2 is either trivial or isomorphic to P3,2 itself. Therefore, P3,2 is rigid. (cid:3) 4. Admissible Hom-Poisson algebras A Poisson algebra has two binary operations, the Lie bracket and the commutative associative product. It is shown in [9, 16] that Poisson algebras can be described using only one binary operation via the polarization-depolarization process. The purpose of this section is to extend this alternative description of Poisson algebras to Hom-Poisson algebras. In other words, we will show that a Hom- Poisson algebra can be described using only the twisting map and one binary operation. We first define the Hom-algebras that correspond to Hom-Poisson algebras. Definition 4.1. Let (A, µ, α) be a Hom-algebra. Then A is called an admissible Hom-Poisson algebra if it satisfies asA(x, y, z) = 1 3 {(xz)α(y) − (zx)α(y) + (yz)α(x) − (yx)α(z)} (4.1.1) for all x, y, z ∈ A, where asA is the Hom-associator (2.3.1) of A. As usual in (4.1.1) the product µ is denoted by juxtapositions of elements in A. An admissible Hom-Poisson algebra with α = Id is exactly an admissible Poisson algebra as defined in [9]. To compare Hom-Poisson algebras and admissible Hom-Poisson algebras, we need the following function, which generalizes a similar function in [16]. Definition 4.2. Let (A, µ, α) be a Hom-algebra. Define the quadruple P (A) =(cid:18)A, {, } = 1 2 (µ − µop), • = 1 2 (µ + µop), α(cid:19) , called the polarization of A. We call P the polarization function. (4.2.1) The following result says that admissible Hom-Poisson algebras, and only these Hom-algebras, give rise to Hom-Poisson algebras via polarization. It is the Hom-version of [16] (Example 2). Theorem 4.3. Let (A, µ, α) be a Hom-algebra. Then the polarization P (A) is a Hom-Poisson algebra if and only if A is an admissible Hom-Poisson algebra. The proof will be given below. Assuming Theorem 4.3 for the moment, first we observe that the polarization function is actually a bijection from admissible Hom-Poisson algebras to Hom-Poisson algebras. To prove this statement, we introduce the following function. Definition 4.4. Let (A, {, }, •, α) be a quadruple in which (A, α) is a Hom-module and {, }, • : A⊗2 → A are binary operations. Define the Hom-algebra P −(A) = (A, µ = {, } + •, α) , (4.4.1) called the depolarization of A. We call P − the depolarization function. NON-COMMUTATIVE HOM-POISSON ALGEBRAS 17 The following result says that there is a bijective correspondence between admissible Hom-Poisson algebras and Hom-Poisson algebras via polarization and depolarization. It is the Hom-version of [9] (Proposition 3). Corollary 4.5. The polarization and the depolarization functions P : {admissible Hom-Poisson algebras} ⇄ {Hom-Poisson algebras} : P − are the inverses of each other. Proof. If (A, µ, α) is an admissible Hom-Poisson algebra, then P (A) is a Hom-Poisson algebra by Theorem 4.3. We have P −(P (A)) = A because µ = 1 2 (µ − µop) + 1 2 (µ + µop). Conversely, suppose (A, {, }, •, α) is a Hom-Poisson algebra. To see that P −(A) is an admissible Hom-Poisson algebra, note that the anti-symmetry of {, } and the commutativity of • imply that {, } = • = 1 2 1 2 (({, } + •) − ({, } + •)op) , (({, } + •) + ({, } + •)op) . So the Hom-algebra P −(A) has the property that P (P −(A)) = A, which is a Hom-Poisson algebra. It follows from Theorem 4.3 that P −(A) is an admissible Hom-Poisson algebra. Since P −P and P P − are both identity functions, P and P − are the inverses of each other. (cid:3) It should be noted that both the polarization and the depolarization functions preserve multi- plicativity. So the polarization of a multiplicative admissible Hom-Poisson algebra is a multiplicative Hom-Poisson algebra. Conversely, the depolarization of a multiplicative Hom-Poisson algebra is a multiplicative admissible Hom-Poisson algebra. We now prove Theorem 4.3 with a series of Lemmas. To prove the "if" part of Theorem 4.3, we need some preliminary results. The following observation says that admissible Hom-Poisson algebras are Hom-flexible. It is the Hom-version of [9] (Proposition 4). The notion of Hom-flexibility is a Hom-type generalization of the usual definition of flexibility and was first introduced in [13]. Lemma 4.6. Every admissible Hom-Poisson algebra (A, µ, α) is Hom-flexible, i.e., asA(x, y, z) + asA(z, y, x) = 0 (4.6.1) for all x, y, z ∈ A. Proof. The required identity (4.6.1) follows immediately from the defining identity (4.1.1), in which the right-hand side is anti-symmetric in x and z. (cid:3) Next we observe that in an admissible Hom-Poisson algebra the cyclic sum of the Hom-associator is trivial. Lemma 4.7. Let (A, µ, α) be an admissible Hom-Poisson algebra. Then SA(x, y, z) def = asA(x, y, z) + asA(z, x, y) + asA(y, z, x) = 0 (4.7.1) for all x, y, z ∈ A. 18 DONALD YAU Proof. Using the defining identity (4.1.1), we have: asA(x, y, z) = 1 3 ((yz)α(x) + (xz)α(y) − (yx)α(z) − (zx)α(y)) = − + 1 3 1 3 ((zy)α(x) − (yz)α(x) + (xy)α(z) − (xz)α(y)) ((xy)α(z) − (yx)α(z) + (zy)α(x) − (zx)α(y)) = −asA(z, x, y) + asA(x, z, y) = −asA(z, x, y) − asA(y, z, x). The last equality above follows from Hom-flexibility (Lemma 4.6). Therefore, we conclude that SA = 0. (cid:3) Next we show that the polarization of an admissible Hom-Poisson algebra is commutative Hom- associative. Lemma 4.8. Let (A, µ, α) be an admissible Hom-Poisson algebra. Then is a commutative Hom-associative algebra. (cid:18)A, • = 1 2 (µ + µop), α(cid:19) Proof. It is obvious that • = (µ + µop)/2 is commutative. To show that the Hom-associator asP (A) = •(• ⊗ α − α ⊗ •) is trivial, pick x, y, z ∈ A. We write µ using juxtaposition of elements in A. Expanding asP (A) in terms of µ, we have: 4asP (A)(x, y, z) = (xy)α(z) + (yx)α(z) + α(z)(xy) + α(z)(yx) − α(x)(yz) − α(x)(zy) − (yz)α(x) − (zy)α(x) = asA(x, y, z) − asA(z, y, x) + (yx)α(z) − (yz)α(x) − asA(z, x, y) + (zx)α(y) + asA(x, z, y) − (xz)α(y) (4.8.1) Using (4.1.1) and Hom-flexibility (Lemma 4.6), we can combine six of the eight terms above as follows: asA(x, y, z) − asA(z, y, x) + (yx)α(z) − (yz)α(x) + (zx)α(y) − (xz)α(y) = asA(x, y, z) + asA(x, y, z) − 3asA(x, y, z) (4.8.2) = −asA(x, y, z) By Lemmas 4.6 and 4.7, the other two terms in the last line in (4.8.1) become: −asA(z, x, y) + asA(x, z, y) = −asA(z, x, y) − asA(y, z, x) Using (4.8.2) and (4.8.3) in (4.8.1), we conclude that asP (A) = 0. = asA(x, y, z). (4.8.3) (cid:3) Now we observe that the polarization of an admissible Hom-Poisson algebra is a Hom-Lie algebra. NON-COMMUTATIVE HOM-POISSON ALGEBRAS 19 Lemma 4.9. Let (A, µ, α) be a Hom-algebra. Then 4JP (A)(x, y, z) = asA(x, y, z) + asA(z, x, y) + asA(y, z, x) − asA(y, x, z) − asA(z, y, x) − asA(x, z, y) (4.9.1) for all x, y, z ∈ A, where JP (A) is the Hom-Jacobian (2.3.3) of the polarization P (A) (4.2.1). More- over, if A is an admissible Hom-Poisson algebra, then is a Hom-Lie algebra. (cid:18)A, {, } = 1 2 (µ − µop), α(cid:19) Proof. Since {, } = (µ − µop)/2, one can expand 4JP (A) = 4{, }({, } ⊗ α)(Id + σ + σ2) in terms of µ. The resulting twelve terms are then written in terms of the Hom-associator asA. The result is the right-hand side of (4.9.1). For the second assertion, first note that {, } is clearly anti-symmetric. Next observe that the identity (4.9.1) can be rewritten as 4JP (A)(x, y, z) = SA(x, y, z) − SA(y, x, z), where SA is the cyclic sum of the Hom-associator defined in (4.7.1). Since SA = 0 in an admissible Hom-Poisson algebra (Lemma 4.7), we conclude that JP (A) = 0. In other words, (A, {, }, α) satisfies the Hom-Jacobi identity. (cid:3) The following result says that the polarization of an admissible Hom-Poisson algebra satisfies the Hom-Leibniz identity (2.5.2). Lemma 4.10. Let (A, µ, α) be a Hom-algebra. Then the polarization P (A) satisfies 4 ({α(x), y • z} − {x, y} • α(z) − α(y) • {x, z}) = asA(x, y, z) + asA(z, y, x) + asA(x, z, y) + asA(y, z, x) (4.10.1) − asA(y, x, z) − asA(z, x, y) for all x, y, z ∈ A. Moreover, if A is an admissible Hom-Poisson algebra, then the polarization P (A) satisfies the Hom-Leibniz identity. Proof. Since {, } = (µ − µop)/2 and • = (µ + µop)/2, the left-hand side of (4.10.1) can be expanded in terms of µ into twelve terms. The result can be written in terms of the Hom-associator asA, which turns out to be the right-hand side of (4.10.1). For the second assertion, suppose that A is an admissible Hom-Poisson algebra. Then Hom- flexibility (Lemma 4.6) implies that the right-hand side of (4.10.1) is 0. We conclude that which is the Hom-Leibniz identity in the polarization P (A). (cid:3) {α(x), y • z} = {x, y} • α(z) + α(y) • {x, z}, Next we show that only admissible Hom-Poisson algebras can give rise to Hom-Poisson algebras via polarization. Lemma 4.11. Let (A, µ, α) be a Hom-algebra such that the polarization P (A) is a Hom-Poisson algebra. Then A is an admissible Hom-Poisson algebra. 20 DONALD YAU Proof. We need to prove the identity (4.1.1). Pick x, y, z ∈ A. We will express the Hom-associator asA in several different forms and compare them. On the one hand, the Hom-Jacobi identity JP (A) = 0 and (4.9.1) imply that asA(x, y, z) = asA(y, x, z) − asA(y, z, x) − asA(z, x, y) + asA(z, y, x) + asA(x, z, y). Moreover, the Hom-Leibniz identity in P (A) and (4.10.1) imply that asA(x, y, z) = asA(y, x, z) − asA(y, z, x) + asA(z, x, y) − asA(z, y, x) − asA(x, z, y). Adding (4.11.1) and (4.11.2) and dividing the result by 2, we obtain (4.11.1) (4.11.2) asA(x, y, z) = asA(y, x, z) − asA(y, z, x), (4.11.3) which we will use in a moment. On the other hand, since µ = {, } + •, we can expand the Hom-associator asA in terms of {, } and • as follows: asA(x, y, z) = (xy)α(z) − α(x)(yz) = {{x, y}, α(z)} + {x • y, α(z)} + {x, y} • α(z) + (x • y) • α(z) (4.11.4) − {α(x), {y, z}} − {α(x), y • z} − α(x) • {y, z} − α(x) • (y • z) Since the polarization P (A) is assumed to be a Hom-Poisson algebra, we have: 0 = asP (A)(x, y, z) = (x • y) • α(z) − α(x) • (y • z), 0 = {x, z} • α(y) − α(y) • {x, z} = {x • y, α(z)} − α(x) • {y, z} − {α(x), y • z} + {x, y} • α(z), (4.11.5) {{x, z}, α(y)} = {{x, y}, α(z)} − {α(x), {y, z}}. Using the identities (4.11.5) in (4.11.4), we obtain: 4asA(x, y, z) = 4{{x, z}, α(y)} = (xz)α(y) − (zx)α(y) − α(y)(xz) + α(y)(zx) = (xz)α(y) − (zx)α(y) + asA(y, x, z) − (yx)α(z) − asA(y, z, x) + (yz)α(x) = (xz)α(y) − (zx)α(y) + (yz)α(x) − (yx)α(z) + asA(x, y, z), where the last equality follows from (4.11.3). Finally, subtracting asA(x, y, z) in the above calculation and dividing the result by 3, we obtain the desired identity (4.1.1). (cid:3) Proof of Theorem 4.3. If A is an admissible Hom-Poisson algebra, then Lemmas 4.8, 4.9, and 4.10 imply that the polarization P (A) is a Hom-Poisson algebra. The converse is Lemma 4.11. (cid:3) 5. Hom-power associativity The purpose of this section is to show that multiplicative admissible Hom-Poisson algebras are Hom-power associative. Power associativity of admissible Poisson algebras is proved in [9] (Propo- sition 6). Let us first recall the definition of a Hom-power associative algebra from [26]. Definition 5.1. Let (A, µ, α) be a Hom-algebra, x ∈ A, and n be a positive integer. NON-COMMUTATIVE HOM-POISSON ALGEBRAS 21 (1) The nth Hom-power xn ∈ A is defined inductively as x1 = x, xn = xn−1αn−2(x) for n ≥ 2. (2) For positive integers i and j, define xi,j = αj−1(xi)αi−1(xj). (3) A is called nth Hom-power associative if xn = xn−i,i for all x ∈ A and i ∈ {1, . . . , n − 1}. (5.1.1) (5.1.2) (5.1.3) (4) A is called Hom-power associative if A is nth Hom-power associative for all n ≥ 2. By definition for all n ≥ 2. x2 = xx, xn = xn−1,1 If the twisting map α is the identity map, then xn = xn−1x, xi,j = xixj , and nth Hom-power associativity reduces to xn = xn−ixi (5.1.4) for all x ∈ A and i ∈ {1, . . . , n − 1}. Therefore, Hom-powers and (nth) Hom-power associa- tivity become Albert's right powers and (nth) power associativity [1, 2] if α = Id. Examples of Hom-power associative algebras include multiplicative right Hom-alternative algebras and non- commutative Hom-Jordan algebras. Other results for Hom-power associative algebras can be found in [26]. By definition, power associativity involves infinitely many defining identities, namely, (5.1.4) for all n. A well-known result of Albert [1] says that an algebra (A, µ) is power associative if and only if it is third and fourth power associative, i.e., the condition (5.1.4) holds for n = 3 and 4. Moreover, for (5.1.4) to hold for n = 3 and 4, it is necessary and sufficient that (xx)x = x(xx) and ((xx)x)x = (xx)(xx) for all x ∈ A. The Hom-versions of these statements are also true. More precisely, the author proved in [26] that a multiplicative Hom-algebra (A, µ, α) is Hom-power associative if and only if it is third and fourth Hom-power associative, which in turn is equivalent to x2α(x) = α(x)x2 and x4 = α(x2)α(x2) (5.1.5) for all x ∈ A. The following result is the Hom-version of [9] (Proposition 6). Theorem 5.2. Every multiplicative admissible Hom-Poisson algebra is Hom-power associative. Proof. As discussed above, by a result in [26] it suffices to prove the two equalities in (5.1.5). Hom- flexibility (Lemma 4.6) implies that 0 = asA(x, x, x) = x2α(x) − α(x)x2, 22 DONALD YAU which proves the first identity in (5.1.5). To prove the other equality in (5.1.5), note that Hom- flexibility implies that: 0 = asA(α(x), x2, α(x)) = (α(x)x2)α2(x) − α2(x)(x2α(x)). Together with the first identity in (5.1.5), we have: x4 def= (x2α(x))α2(x) = (α(x)x2)α2(x) = α2(x)(x2α(x)) = α2(x)(α(x)x2). (5.2.1) Using multiplicativity and (5.2.1), the defining identity (4.1.1) applied to (α(x), α(x), x2) says that: 0 = 3asA(α(x), α(x), x2) − (α(x)x2)α2(x) + (x2α(x))α2(x) − (α(x)x2)α2(x) + (α(x)α(x))α(x2 ) = 3(cid:8)(α(x)α(x))α(x2 ) − α2(x)(α(x)x2)(cid:9) − x4 + (α(x)α(x))α(x2 ) = 4α(x2)α(x2) − 4x4. We have proved the second identity in (5.1.5). (cid:3) References [1] A.A. Albert, On the power-associativity of rings, Summa Brasil. Math. 2 (1948) 21-32. [2] A.A. Albert, Power-associative rings, Trans. Amer. Math. Soc. 64 (1948) 552-593. [3] V.I. Arnold, Mathematical methods of classical mechanics, Grad. Texts in Math. 60, Springer, Berlin, 1978. [4] V. Chari and A.N. Pressley, A guide to quantum groups, Cambridge Univ. Press, Cambridge, 1994. [5] V.G. Drinfel'd, Quantum groups, in: Proc. ICM (Berkeley, 1986), p.798-820, AMS, Providence, RI, 1987. [6] M. Flato, M. Gerstenhaber, A.A. Voronov, Cohomology and deformation of Leibniz pairs, Lett. Math. Phys. 34 (1995) 77-90. [7] E. Frenkel and D. Ben-Zvi, Vertex algebras and algebraic curves, Math Surveys and Monographs 88, 2nd ed., AMS, Providence, RI, 2004. [8] M. Gerstenhaber, On the deformation of rings and algebras, Ann. Math. 79 (1964) 59-103. [9] M. Goze and E. Remm, Poisson algebras in terms of non-associative algebras, J. Alg. 320 (2008) 294-317. [10] J.T. Hartwig, D. Larsson, and S.D. Silvestrov, Deformations of Lie algebras using σ-derivations, J. Alg. 295 (2006) 314-361. [11] M. Kontsevich, Deformation quantization of Poisson manifolds, Lett. Math. Phys. 66 (2003) 157-216. [12] F. Kubo, Finite-dimensional non-commutative Poisson algebras, J. Pure Appl. Alg. 113 (1996) 307-314. [13] A. Makhlouf and S. Silvestrov, Hom-algebra structures, J. Gen. Lie Theory Appl. 2 (2008) 51-64. [14] A. Makhlouf and S. Silvestrov, Hom-algebras and Hom-coalgebras, J. Alg. Appl. 9 (2010) 1-37. [15] A. Makhlouf and S. Silvestrov, Notes on formal deformations of Hom-associative and Hom-Lie algebras, to appear in Forum Math., arXiv:0712.3130v1. [16] M. Markl and E. Remm, Algebras with one operation including Poisson and other Lie-admissible algebras, J. Alg. 299 (2006) 171-189. [17] P. Schaller and T. Strobl, Poisson structure induced (topological) field theories, Mod. Phys. Lett. A 9 (1994) 3129-3136. [18] I. Vaisman, Lectures on the geometry of Poisson manifolds, Birkhauser, Basel, 1994. [19] D. Yau, Enveloping algebras of Hom-Lie algebras, J. Gen. Lie Theory Appl. 2 (2008) 95-108. [20] D. Yau, Hom-algebras and homology, J. Lie Theory 19 (2009) 409-421. [21] D. Yau, The Hom-Yang-Baxter equation, Hom-Lie algebras, and quasi-triangular bialgebras, J. Phys. A 42 (2009) 165202 (12pp). [22] D. Yau, Hom-bialgebras and comodule Hom-algebras, Int. Elect. J. Alg. 8 (2010) 45-64. [23] D. Yau, Hom-Maltsev, Hom-alternative, and Hom-Jordan algebras, arXiv:1002.3944. [24] D. Yau, Hom-Novikov algebras, arXiv:0909.0726. [25] D. Yau, The Hom-Yang-Baxter equation and Hom-Lie algebras, arXiv:0905.1887. [26] D. Yau, Hom-power associative algebras, arXiv:1007.4118. NON-COMMUTATIVE HOM-POISSON ALGEBRAS 23 Department of Mathematics, The Ohio State University at Newark, 1179 University Drive, Newark, OH 43055, USA E-mail address: [email protected]
1312.0413	2	1312	2013-12-21T20:43:34	G\"odel algebras: interactive dualities and their applications	[ "math.RA" ]	We present a technique for deriving certain new natural dualities for any variety of algebras generated by a finite Heyting chain. The dualities we construct are tailored to admit a transparent translation to the more pictorial Priestley/Esakia duality and back again. This enables us to combine the two approaches and so to capitalise on the virtues of both, in particular the categorical good behaviour of a natural duality: we thereby demonstrate the fullness, or not, of each of our dualities; we obtain new results on amalgamation; and we also provide a simple treatment of coproducts.	math.RA	math	G ODEL ALGEBRAS: INTERACTIVE DUALITIES AND THEIR APPLICATIONS L. M. CABRER AND H. A. PRIESTLEY Dedicated to Brian Davey in celebration of his 65th birthday Abstract. We present a technique for deriving certain new natural dualities for any variety of algebras generated by a finite Heyting chain. The dualities we construct are tailored to admit a transparent translation to the more pictorial Priestley/Esakia duality and back again. This enables us to combine the two approaches and so to capitalise on the virtues of both, in particular the categorical good behaviour of a natural duality: we thereby demonstrate the fullness, or not, of each of our dualities; we obtain new results on amalgamation; and we also provide a simple treatment of coproducts. 3 1 0 2 c e D 1 2 ] . A R h t a m [ 2 v 3 1 4 0 . 2 1 3 1 : v i X r a 1. Introduction This paper focuses on the classes of algebras Gn = ISP(Cn), where Cn is the n-element Heyting chain, and on natural dualities for them. As we recall below, the classes Gn are important both within and beyond duality theory. It is highly appropriate that this topic should feature in the Special Issue of Algebra Universalis in honour of Brian Davey: through his work, the varieties Gn have been influential as a example within natural duality theory for nearly 40 years, beginning with his trail-blazing 1976 paper [7]. The advances to which study of these varieties have made a contribution are well chronicled, and fully referenced, by Davey and Talukder [9, Section 1]. Here we add a further chapter to the saga. Although the emphasis will be squarely on the varieties Gn, which have very special features which work to our advantage, glimpses also open up of future avenues of investigation with wider scope. Many varieties which provide illuminating test case examples for duality theory are also of relevance to logic. In particular any variety A of Heyting algebras is the algebraic counterpart of an intermediate logic LA (an axiomatic extension of intuitionistic propositional logic, IPC). The varieties Gn are the proper subvarieties of the variety G of Godel algebras (also known as Godel/Dummett algebras, L-algebras, or pre-linear Heyting algebras), viz. the Heyting algebras that satisfy the pre-linearity equation (a → b) ∨ (b → a) ≈ (cid:62). They are the equivalent alge- braic semantics of the Godel/Dummett extension of IPC obtained by adding the linearity axiom (x → y) ∨ (y → x) (see [15, 13] and also [20] for further historical references). In general, if a quasivariety A is the equivalent algebraic semantics for a sentential logic LA, then there are connections between algebraic properties in A and logical properties of LA. For example, differ- ent amalgamation properties in A correspond to different interpolation properties in LA (see for example [5]). This leads us to study amalgamation in Godel algebras, as reported below. Duality theory for Heyting algebras is intimately connected with Kripke semantics for IPC. Paralleling these relational semantics there is the topological duality we shall refer to as Priest- ley/Esakia duality [14]. This specialises Priestley duality for the category D of bounded distribu- tive lattices to Heyting algebras and provides a primary tool for the study of such algebras. It is well known that a Heyting algebra belongs to Gn if and only if the order of its associated Esakia space is a forest of depth at most n − 2. Further details are given in Section 4. An overarching 2010 Mathematics Subject Classification. Primary: 06D50; Secondary: 08C20, 06D20 03G25. Key words and phrases. Godel algebra, Heyting algebra, natural duality, Esakia duality, amalgamation, coprod- uct . The first author was supported by a Marie Curie Intra European Fellowship within the 7th European Community Framework Program (ref. 299401-FP7-PEOPLE-2011-IEF). . 1 2 L. M. CABRER AND H. A. PRIESTLEY objective of this paper may be seen as the development of a closer tie-up than hitherto available between natural dualities for the varieties Gn and the Priestley/Esakia duality. We have aimed to make our results accessible to those interested in Godel algebras per se and in the applications to amalgamation and coproducts we give in Section 6. Nonetheless, we cannot make our account fully self-contained and shall refer to Clark and Davey's text [4] for background on the fundamentals of natural duality theory. The following summary of certain key events is directed at those already conversant with this theory. It will enable us to set in context for such readers what we achieve in this paper. It was established in [7] that each variety Gn is endodualisable (so that the alter ego (Cn; End Cn, T) yields a duality); in particular Gn is dually equivalent to a category of Boolean spaces acted on by a monoid of continuous maps. While endodualisability of Gn ensures that the alter ego is of an amenable type, the endomorphism monoid End Cn grows exponentially as n increases. This issue was addressed by Davey and Talukder [9, Section 2], with consideration of optimality within the realm of dualities for Gn based on endomorphisms. But there is another approach worthy of consideration. It was to obtain more tractable dualities than those supplied by the NU Du- ality Theorem, as it applies in particular to distributive lattice-based algebras, that Davey and Werner [10] devised their 'piggyback method'. This leads to a reasonably economical choice of alter ego, which in the case of Gn contains (the graphs of) the members of a family containing both endomorphisms and partial endomorphisms. While natural duality theory was set up from the outset to encompass alter egos containing partial operations, the perception has always been that total structures are to be preferred, whenever possible. However, with a recent spurt of progress in understanding alter egos, the role of partial operations is steadily becoming less mysterious. Moreover, adding partial operations to upgrade a duality to a strong, and hence full, duality is a standard technique for creating full dualities for varieties of lattice-based algebras. Hence includ- ing certain partial endomorphisms in an alter ego for a variety Gn may be desirable on grounds of economy, and also the best option for achieving fullness when a full duality is wanted. The piggyback strategy leans heavily on Priestley duality for D and it is natural to go one step further and to seek to relate the piggyback natural duality for a D-based quasivariety A to Priestley duality applied to U(A), where U : A → D is the obvious forgetful functor. Indeed, the germ of the idea for the piggyback method is already present in Davey's proof of his duality for Gn [7, Theorem 2.4]. Esakia, in his review MR0412063 (54 #192) of [7], observed that it would be interesting to compare Davey's duality for Gn with the duality in Esakia's own 1974 paper [14]. Davey had already taken the first step here: within the proof of [7, Theorem 2.4] he shows how to pass from the natural dual of an algebra A ∈ Gn to the Priestley/Esakia dual of its D-reduct. He does this by identifying the latter with a Priestley space obtained as the quotient of the natural dual space of A, where the equivalence relation and the ordering on the quotient are determined by the action of the endomorphisms of Cn. The present authors in [3] discussed an analogous process in the context of a piggyback duality for any finitely generated quasivariety of D-based algebras. This translation was developed to facilitate an analysis of coproducts in finitely generated D-based quasivarieties. Here we take these ideas further. For the dualities we present for each class Gn, we are able to set up a simple two-way translation, in a functorial way, between the dual categories involved. Simplicity stems from our choices of alter ego; these result a quotienting process which is particularly easy to visualise. But what is much more significant for applications is the bi-directional nature of our translation. We shall refer to the dualities for which such a translation is available as interactive. With an interactive duality to hand we have, in a strong sense, a 'best of both worlds' scenario: we can harness both the categorical virtues of a natural duality and the merits of Priestley duality, specifically its pictorial character and the fact that its is a strong natural duality. We now outline the structure of this paper and indicate, in somewhat more detail than above, what we achieve. In Section 2 we conduct a detailed analysis of the elements in the hom-sets Gn(A, Cn) for A ∈ Gn. This includes investigation of the images of these maps in D(U(A), 2) under Φω : x (cid:55)→ ω ◦ x, where ω : U(Cn) → {0, 1} is the D-homomorphism which sends the top element of Cn to 1 and all other elements to 0. Some of our results are well known (surjectivity of Φω, for example, is a key component in the validation of Davey and Werner's piggyback duality) 3 and certain ingredients are common to our treatment of endomorphisms and that of Davey [7], but other results are new. In particular Lemmas 2.1 -- 2.3 go beyond what appears in the existing literature, and suggest that particular partial endomorphisms and endomorphisms of Cn may be good candidates for inclusion in an alter ego for Cn tailored to a smooth translation between the associated natural duality for Gn and Priestley duality. In Section 3 we combine duality theory's Test Algebra Lemma with an adaptation of the proof of the traditional Piggyback Duality Theorem to obtain a family of new dualities for Gn (for n (cid:62) 4). Each includes in the alter ego n − 3 partial endomorphisms and one of n − 2 endomorphisms. In Section 4 we confirm that our new dualities are indeed tailor-made for two-way translation. We adapt the strategy used in [3, Theorem 2.4] to pass, with the aid of ω, from the natural dual of an algebra to the Priestley dual of its reduct. More significantly, the results in Section 2 allow us also to go back again (Theorem 4.6). We demonstrate the power of our interactive dualities first in Section 5. We are able to show that for each n only one of our 2n−3 × (n − 2) dualities is full. (We recall that Davey [7] showed that End Cn does not yield a full duality on Gn when n (cid:62) 4 and it is known that the entire monoid Endp Cn of partial and total endomorphisms of Cn yields a full duality.) Section 6 is devoted to two different applications: to amalgamation and to coproducts. It follows from results of Maksimova [18] that Gn fails to satisfy the amalgamation property if n (cid:62) 4, so that not every V -formation admits amalgamation in Gn. We use our two-way translation to determine which V -formations do admit amalgamation in Gn. This result is based on the categorical properties of natural dualities and the fact that Priestley duality maps injective homomorphisms to surjective continuous maps. Finally we extend our work on coproducts [3] by adding Godel algebras to the catalogue of examples provided there. We employ our two-way translation to give a procedure for describing coproducts in Gn and, in certain cases, in G too. Our method provides a simple alternative to the procedure presented by D'Antona and Marra [6] in the case of finite Godel algebras (they employ solely Priestley/Esakia duality) and by Davey [7, Section 5] for algebras in Gn (he uses only his natural duality for Gn). We elect to formulate our principal results about dualities for Gn under the assumption n (cid:62) 4 since to encompass n = 2, 3 would complicate the statements and contribute little that is new. However, mutatis mutandis, the special cases can be fitted into our general scheme, and we make brief comments as we proceed to confirm this. 2. Godel algebras An algebra A = (A;∧,∨,→,⊥,(cid:62)) is a Heyting algebra if (A;∧,∨,⊥,(cid:62)) is a bounded distribu- tive lattice and a ∧ b (cid:54) c if and only if a (cid:54) b → c. A basic reference for the algebraic properties of Heyting algebras is [1, Chapter IX]. We shall denote the variety of Heyting algebras by H. We make this, and likewise any other class of algebras with which we work, into a category by taking as morphisms all homomorphisms. It will be important that any Heyting algebra has a reduct in the variety D of bounded dis- tributive lattices. More precisely, we have a forgetful functor U : H → D that on objects sends A = (A;∧,∨,→,⊥,(cid:62)) to (A;∧,∨,⊥,(cid:62)) and sends any morphism, regarded as a map, to the same map. Heyting algebras are rather special amongst algebras with reducts in D in that the implication → is uniquely determined by the underlying order. The variety G of Godel algebras is the subvariety of H consisting of those algebras which satisfy the pre-linearity equation (a → b) ∨ (b → a) ≈ (cid:62). It is a consequence of pre-linearity that every subdirectly irreducible Godel algebra is a chain. Moreover G is generated as a variety by any infinite Heyting chain, and its proper subvarieties are precisely the varieties generated by finite chains [16]. Consider the n-element chain with elements 0, 1, . . . , n− 1 labelled so that 0 < 1 < ··· < n− 1. Then we define Cn = ({0, 1, . . . , n − 1};∧,∨,→,⊥,(cid:62)), where the constants ⊥ and (cid:62) are taken to be the bounds 0 and n − 1 and a → b =(cid:40)(cid:62) if a (cid:54) b, if b < a. b 4 L. M. CABRER AND H. A. PRIESTLEY Trivially, every homomorphic image of Cn is a chain and therefore isomorphic to a subalgebra of Cn, whence it follows that ISP(Cn) = HSP(Cn). Thus Gn, defined earlier to be the quasivariety generated by Cn, is also the variety generated by Cn. The lattice of subvarieties of the variety G is the chain 1 is generated by the trivial algebra and G 2 is term-equivalent to the variety of Boolean Here G algebras. G 1 ⊆ G 2 ⊆ ··· ⊆ G. In [17] it is proved that a Heyting algebra A is a Godel algebra if and only if the set of its prime lattice filters forms a forest, that is, the set of prime filters that contain a given prime filter forms a chain (see [16]). Henceforth, following the natural duality approach, we work with homomorphisms into 2 rather than with prime filters: A is a Godel algebra if and only if D(U(A), 2), ordered pointwise, is a forest. As we have noted already, A belongs to Gn if and only if the forest D(U(A), 2) has depth at most n − 2. This result is well known but hard to attribute; it can be seen as a consequence of Lemmas 2.1 and 2.2 below. For completeness we belatedly recall the definition of depth. Assume we have a poset P with the property that for every p ∈ P the up-set ↑p does not contain an infinite ascending chain. Then for p ∈ P we define d(p) = max{C − 1 C ⊆ ↑p and C is a chain}. If {d(p) p ∈ P} is bounded above, then the depth of P is defined to be sup{d(p) p ∈ P}. We note for future use the fact that in a poset of finite depth the order relation determines and is determined by the associated covering relation, which we denote by(cid:6). An algebra A in a quasivariety A = ISP(M) is determined by the homomorphisms from A into M. This fact underlies the centrality in natural duality theory of the hom-sets A(A, M) for A ∈ A. Accordingly we shall assemble a number of results about homomorphisms from an algebra A ∈ Gn into Cn. We indicated already in Section 1 the importance of the D-homomorphism ω : U(Cn) → 2 defined by ω((cid:62)) = 1 and ω(k) = 0 for k < (cid:62). This reflects the key role played by the pre-image of the constant (cid:62) in the study of homomorphisms between Heyting algebras. For any A ∈ Gn and any f ∈ Gn(A, Cn) the map ω ◦ f ∈ D(U(A), 2) and f−1((cid:62)) = (ω ◦ f )−1(1). Lemma 2.1. Let A ∈ Gn and x ∈ Gn(A, Cn). For each i ∈ ran x \ {0} let ui : A → 2 be defined by ui(a) = 1 if and only if x(a) (cid:62) i. Then the assignment i (cid:55)→ ui determines a bijection between ran x \ {0} and the subset ↑(ω ◦ x) of D(U(A), 2). In particular ↑(ω ◦ x) (cid:54) n − 1 for each x ∈ Gn(A, Cn). Moreover, if x, y ∈ Gn(A, Cn), then Proof. Certainly, for each i ∈ ran x \ {0}, the map ui is a D-homomorphism for which ui (cid:62) ω ◦ x. To see that the map i (cid:55)→ ui is injective, we argue as follows. Let i, j ∈ ran x\{0} and assume that i < j. Then there exists a ∈ A for which x(a) = j. This implies that uj(a) = 1 and ui(a) = 0, so ui (cid:54)= uj. Now let u ∈ ↑(ω ◦ x). Let k = min{x(a) a ∈ A and u(a) = 1}. Note that k (cid:54)= 0. We claim that u = uk. Certainly u (cid:54) uk. Now assume that b ∈ A is such that uk(b) = 1, that is, x(b) (cid:62) k. Choose a ∈ A such that u(a) = 1 and x(a) = k. Then u(a → b) (cid:62) (w ◦ x)(a → b) = ω(x(a) → x(b)) = ω((cid:62)) = 1. It follows that u(b) (cid:62) u(a ∧ b) = u(a ∧ (a → b)) = u(a) ∧ u(a → b) = 1. (cid:3) Therefore uk (cid:54) u. ω ◦ x(cid:6) ω ◦ y ⇐⇒ x −1((cid:62)) and ran x − 1 = ran y. −1((cid:62)) ⊆ y The final claim in the following lemma appears in [10, Section 3.5] (see also the proof of [7, Theorem 2.4]), but the lemma gives additional information. We shall denote the power set of {0, . . . , n − 1} by ℘n. Lemma 2.2. Let A ∈ Gn. Let T = { (u, V ) ∈ D(U(A), 2) × ℘n 0, n − 1 ∈ V and ↑u + 1 = V }. Then there exist well-defined and mutually inverse maps ιA : Gn(A, Cn) → T and γA : T → Gn(A, Cn). 5 The first of these is defined by ιA : x (cid:55)→ (ω ◦ x, ran x), for x ∈ Gn(A, Cn). The map γA is defined in the following way: let V ∈ ℘n be such that V = {i0, i1, . . . im}, where i0 = 0, im = n − 1 and ij < ij+1 in Cn for 0 (cid:54) j (cid:54) m. Then, for a ∈ A, γA(u, V )(a) = ik, where k = { v ∈ ↑u v(a) = 1}. In particular, the map x (cid:55)→ ω ◦ x is a surjection from Gn(A, Cn) to D(U(A), 2). Proof. Lemma 2.1 tells us that ιA is a map from Gn(A, Cn) into T . Since ↑u + 1 = V , the map γA(u, V ) is well defined for each (u, V ) ∈ T . It is straightforward to check that γA(u, V ) is a homomorphism from A to Cn for each (u, V ) ∈ T and that γA and ιA are mutually inverse. (cid:3) Combining the fact that the map x (cid:55)→ ω ◦ x is a surjection from Gn(A, Cn) to D(U(A), 2) with Lemma 2.1, we obtain an alternative proof of the well-known fact that ↑u (cid:54) n − 1 for each A ∈ Gn and u ∈ D(U(A), 2). As a consequence of Lemma 2.2, we can also describe the endomorphisms of Cn; cf. [9, Lemma 2.2]. In particular, e (cid:55)→ ran e sets up a bijection from End Cn to { V ∈ ℘n 0, n − 1 ∈ V }. h1 hn−2 g1 gn−3 Figure 1. Endomorphisms and partial endomorphisms of Cn We shall make use in the next sections of certain endomorphisms and partial endomorphisms of Cn. For 1 (cid:54) i (cid:54) n − 2 we let hi be the unique endomorphism of Cn with ran hi = Cn \ {i}. More precisely, hi(x) = x + 1 if i (cid:54) x < n − 1 and hi(x) = x otherwise. (These endomorphisms 1 (cid:54) i (cid:54) n − 3 we define the partial endomorphism gi with domain Cn \ {i} as follows: also appear in [4, Section 2], with hi denoted ei, but the use we make of them is different.) For gi(x) =(cid:40)x if x (cid:54)= i + 1, if x = i + 1. i These maps are indeed partial endomorphisms, none of which extends to an endomorphism. For 1 (cid:54) i (cid:54) n − 3, let fi : Cn \ {i + 1} → Cn \ {i} be the inverse of gi. For 1 (cid:54) i (cid:54) n − 4 the map fi is a non-extendable partial endomorphism; fn−3 extends to hn−3. Figure 1 depicts h1, hn−2, g1 and gn−3; corresponding diagrams of f1 and fn−3 are obtained from those of g1 and gn−3 by left-to-right reflection. We fix for future use the following notation: for n (cid:62) 4, Σn = {f1, g1} × ··· × {fn−3, gn−3} × {h1, . . . , hn−2}. In Lemma 2.1 we described, for any given A ∈ Gn, the covering relation on the distinct elements of the set { ω ◦ x x ∈ Gn(A, Cn)}. Below we complement this result by demonstrating when elements ω ◦ x and ω ◦ y coincide. Lemma 2.3. Fix n (cid:62) 4 and σ ∈ Σn. Let A ∈ Gn and let x, y ∈ Gn(A, Cn) be such that x (cid:54)= y. Then the following statements are equivalent: 6 L. M. CABRER AND H. A. PRIESTLEY (1) x−1((cid:62)) = y−1((cid:62)); (2) ω ◦ x = ω ◦ y; (3) there exists a finite sequence z0 = x, z1, . . . , zN = y of elements of Gn(A, Cn) with the property that, for each 0 (cid:54) j < N , there is some ij ∈ {1, . . . , n− 3} such that zj+1 = σij ◦ zj or zj = σij ◦ zj+1. Since fi and gi are inverses of each other, without loss of generality we may assume that σi = gi Proof. Conditions (1) and (2) are equivalent since ω(k) = (cid:62) if and only if k = (cid:62). for each i (cid:54) n − 3. Since gi(k) = (cid:62) if and only if k = (cid:62), (3) implies (1). It remains to show that (2) implies (3). By Lemma 2.1 and condition (2), ran x = ran y. Let m = ran x = ran y and V = {0, . . . , m − 2} ∪ {n − 1}. Now let u = γA(ω ◦ x, V ) as defined in Lemma 2.2. It is easy to see that either x = u or there exists a sequence of j1, . . . , j(cid:96) of elements of {1, . . . , n − 3} such that u = gj1 ◦ gj2 ◦ ··· ◦ gj(cid:96) ◦ x. Similarly, y = u or u = gk1 ◦ gk2 ◦ ··· ◦ gkm ◦ y for some k1, . . . , km ∈ {1, . . . , n − 3}. Since x (cid:54)= y we cannot have both x = u and y = u. Considering the (cid:3) three remaining possibilities in turn it is easy to see that (3) holds in each case. 3. Natural dualities for Gn In what follows we shall use Priestley duality as an ancillary tool. We shall assume familiarity with basic facts concerning this prototypical natural duality (to be found in [4] and [12, Chap- ter 11]), but we do need to establish notation. We denote the category of Priestley spaces by P. We can express D and P as, respectively, ISP(2) and IScP+( 2∼), where 2 = ({0, 1};∧,∨, 0, 1) and 2∼ = ({0, 1}; (cid:54), T); here T is the discrete topology and IScP+( 2∼) is the class of isomorphic copies of closed substructures of powers of 2∼. Here we shall use non-generic symbols H and K for the hom-functors H = D(−, 2) and K = P(−, 2∼) which set up a dual equivalence between D and P. Given L ∈ D, the evaluation map kL : L → KH(L) is defined by kL(a)(u) = u(a), for a ∈ L and u ∈ H(L); this map is an isomorphism. We refer to the Priestley space H(L) as the Priestley dual of L. We now turn to natural dualities more generally. We shall confine attention to the varieties Gn = ISP(Cn) that interest us, referring to [4] any reader who requires an account in a more general setting. We note at the outset that it will suffice for our purposes to consider a more restricted form of alter ego than is allowed for in [4]. (We also remark that we have no need in this paper to consider natural dualities which are multisorted.) We consider a topological = (Cn; G, H, T), where G ⊆ End Cn, H ⊆ Endp Cn \ End Cn (the (non-total) structure Cn∼ partial endomorphisms of Cn), and T is the discrete topology. We refer to Cn∼ as an alter ego for Cn. We define Xn to be the topological quasivariety generated by Cn∼ ): a belongs to Xn if and only if it is isomorphic to a topological structure of the same type as Cn∼ ; here operations and partial operations are lifted pointwise. closed substructure of a power of Cn∼ The superscript + serves to indicate that the empty structure is included in Xn. The morphisms of Xn are the continuous structure-preserving maps. , viz. Xn = IScP+(Cn∼ We define hom-functors D : Gn → Xn and E : Xn → Gn as follows: (cid:40)D(A) = Gn(A, Cn) (cid:40)E(X) = Xn(X, Cn∼ D : Gn → Xn, E : Xn → Gn, ) D(f ) = − ◦ f, E(ϕ) = − ◦ ϕ; here Gn(A, Cn) is considered as a substructure of (Cn∼ ) inherits its algebra struc- ture pointwise from Cn. A crucially important fact is that these functors are well defined. This is a consequence of our assumption that we include in Cn∼ only operations and partial operations which are algebraic. Moreover, for each A ∈ Gn, the evaluation map eA, given by eA(a)(x) = x(a) (for a ∈ A and x ∈ D(A)), is an embedding from A to ED(A). Likewise, for each X ∈ Xn, the map εX given by εX(ϕ)(α) = α(ϕ) (for ϕ ∈ X and α ∈ E(X)) is an embedding. In categorical terms, (D, E, e, ε) is a dual adjunction between Gn and Xn with the unit and counit of the ad- junction given by the evaluation maps. (See [4, Chapter 2] for a justification of these assertions )A and Xn(X, Cn∼ 7 in a general setting.) Let A ∈ Gn. We say that Cn∼ (or just G ∪ H) yields a duality on A if eA is an isomorphism from A to ED(A) and that G ∪ H yields a duality on Gn if it yields a duality on each A ∈ Gn. For later use, we say that a dualising alter ego Cn∼ yields a full duality on Gn if DE(X) ∼= X for all X ∈ Xn. We shall need the following result. It is obtained by specialising the Test Algebra Lemma to the very particular situation that concerns us. See [4, Section 8.1] for the general version of this result and contextual discussion. We reiterate that Gn is endodualisable so that the assumptions of Lemma 3.1 are met when G = End Cn. Lemma 3.1. (Test Algebra Lemma, special case) Let (Cn; G, H, T) be an alter ego of Cn which is such that G ⊆ End Cn, H ⊆ Endp Cn \ End Cn and G ∪ H yields a duality on Gn. Let e ∈ G. Then (G \ {e}) ∪ H yields a duality on Gn provided it yields a duality on the single algebra Cn. Proof. The Test Algebra Lemma in its general form tells us that we can discard e from G without destroying the duality so long as (G \ {e}) ∪ H yields a duality on graph e, regarded as an algebra (cid:3) in Gn. But the graph of any endomorphism is isomorphic to Cn. We contrast the use of Cn as a test algebra with that employed in [9, proof of Theorem 2.4]. There Davey and Talukder identify a particular generating set G for End Cn. They then show that the duality it yields is optimal by showing that G \ {e} does not yield a duality on the particular algebra Cn, for any e ∈ G. This means that they are using the Test Algebra Lemma to guide the choice of an algebra that witnesses indispensability of each member of their set G. We use the Test Algebra Lemma in the opposite direction: the lemma tells us that to prove that a given endomorphism e can be dropped from a dualising alter ego it suffices to test this on the single algebra graph e -- we do not have to verify that (G \ {e}) ∪ H yields a duality on every A ∈ Gn. Henceforth, unless indicated otherwise, we shall consider varieties Gn for which n (cid:62) 4. We include the endomorphism h1 (as defined in Section 2) in our alter ego for Cn, rather than any alternative endomorphism, because this makes it particularly easy to establish Claim 4 of the proof of Proposition 3.2. We adopt a more even-handed attitude to endomorphisms in Theorem 3.3. The proof of the proposition draws very heavily on the ideas used to prove the Piggyback Duality Theorem [4, Theorem 7.2.1], as this applies to a quasivariety ISP(M), where M is a finite algebra with a reduct in D. Proposition 3.2. Let the partial endomorphisms g1, . . . , gn−3 and endomorphism h1 be defined as in Section 2. Then {g1, . . . gn−3, h1} yields a duality on the algebra Cn. Proof. Observe that the evaluation map eCn : Cn → ED(Cn) is injective, and the evaluation map kU(Cn) : U(Cn) → KHU(Cn) is an isomorphism, and so surjective. We want to show that eCn is surjective. Now we bring in the critical, but entirely elementary, observation that it will suffice to construct an injective map ∆ : ED(Cn) → KHU(Cn) (see [4, proof of Piggyback Duality Theorem 7.2.1] or [10]). Recall that ω : U(Cn) → 2 denotes the D-morphism with ω−1(1) = {n − 1} and that for each u ∈ HU(Cn) we can find x ∈ D(Cn) such that u = ω ◦ x. We may now attempt to define ∆ as follows. Given ϕ ∈ ED(Cn) let ω ◦ x = ω ◦ y =⇒ ω(ϕ(x)) = ω(ϕ(y)). Suppose first that y = gi ◦ x for some gi. Then ϕ(x) = (cid:62) if and only if ϕ(y) = gi(ϕ(x)) = (cid:62). We argue likewise when x = gi ◦ y. Hence, by Lemma 2.3, ϕ(x) = (cid:62) if and only if ϕ(y) = (cid:62). Since ω(j) = 1 if and only if j = (cid:62), our claim is proved. (cid:0)∆(ϕ)(cid:1)(u) =(cid:0)∆(ϕ)(cid:1)(ω ◦ x) = ω(ϕ(x)). We now establish a series of claims. These combine with the observations above to prove the proposition. 1. ∆ is a well-defined map. We have already observed that every element of HU(Cn) is of the form ω ◦ x for some x ∈ D(Cn). We must now check that, for x and y in D(Cn) = End Cn and ϕ ∈ ED(Cn), 8 L. M. CABRER AND H. A. PRIESTLEY 2. ∆(ϕ) is order-preserving for each ϕ ∈ ED(Cn). −1 For 1 (cid:54) i (cid:54) n − 1, let ui : Cn → 2 be the map determined by u (1) = ↑i. It is trivial to i check that the set {u1, . . . , un−1} coincides with HU(Cn) and u1 > u2 > ··· > un−1. Let 1 (cid:54) i < j < n − 1, so uj < ui. Assume that (∆(ϕ))(uj) = 1. We wish to show that (∆(ϕ))(ui) = 1. For each k such that i (cid:54) k (cid:54) j, let xk = γA(uk, ({0} ∪ {n − k . . . , n − 1} )), where the map γA is as defined in Lemma 2.2. It follows that ω◦ xi = ui and ω◦ xj = uj. Then 1 = (∆(ϕ))(uj) = ω(ϕ(xj)), that is, ϕ(xj) = (cid:62). Clearly xk = h1 ◦ xx+1 whenever i (cid:54) k < j. Since ϕ(xj) = (cid:62), h1((cid:62)) = (cid:62), and ϕ preserves h1, it follows that ϕ(xi) = (cid:62). Therefore (∆(ϕ))(ui) = ω(ϕ(xi)) = 1. 3. For each ϕ ∈ ED(Cn) the map ∆(ϕ) : HU(Cn) → 2∼ is continuous. This is immediate because Cn is finite. 4. ∆ is injective. Suppose that ϕ, ψ ∈ ED(Cn) with ϕ (cid:54)= ψ. Pick x ∈ D(Cn) such that ϕ(x) (cid:54)= ψ(x) in Cn. 1(ψ(x)) = (cid:62) 1 denotes the j-fold composition of h1 if j > 0 and the identity map Without loss of generality, assume that ϕ(x) < ψ(x). Let j = (n−1)−ψ(x). Then hj and hj if j = 0). Since ϕ and ψ preserve h1, 1(ϕ(x)) (cid:54)= (cid:62) (where hj (∆(ψ))(ω ◦ hj 1 ◦ x) = ω(ψ(hj 1 ◦ x)) = ω(hj 1(ψ(x))) = ω(hj 1(ψ(x))) = 1 and 1(ϕ(x))) = ω(hj 1(ϕ(x))) = 0. (cid:3) 1 ◦ x)) = ω(hj 1 ◦ x) = ω(ϕ(hj (∆(ϕ))(ω ◦ hj Therefore ∆(ϕ) (cid:54)= ∆(ψ). The following theorem supplies a family of alter egos each of which dualises Gn. In Section 5, we shall see that, even if the natural dualities presented in Theorem 3.3 are closely connected, they have significantly different properties. We recall that the definition of Σn was given in Section 2. Theorem 3.3. Let σ ∈ Σn. Then Cσ n∼ Proof. We first note that Lemma 3.1 and Proposition 3.2 combine to tell us that the alter ego (Cn; g1, . . . , gn−3, h1, T) yields a duality on Gn. For any i, the maps gi and fi are interchangeable because their graphs are mutual converses. We note that h1 = f1 ◦ ··· ◦ fi−1 ◦ hi for 2 (cid:54) i (cid:54) n − 2. Hence (see [4, Section 2.4]) h1 is entailed (cid:3) by f1, . . . , fn−3 and hi. Therefore Cσ n∼ = (Cn; σ, T) yields a duality on Gn = ISP(Cn). yields a duality for any choice of σ from Σn. We remark that we could use the Test Algebra Lemma to prove that each of the dualities presented in Theorem 3.3 is optimal; cf. [9, Theorem 2.4]. The technique is standard and we do not include details here. We briefly consider G 3. Here End C3 = {idC3, h1}, and there are no non-extendable endo- morphisms to consider. We could define Σ3 = {h1} and so bring n = 3 within the scope of Theorem 3.3. But this adds nothing that is new: already in [7] the alter ego (C3; h1, T) was shown to yield a duality on G 3. For n = 2 there is even less that is worth saying, since End(C2) = {idC2} and there are no non-extendable partial endomorphisms. The duality for G 2 associated with Σ2, defined to be ∅, is just Stone duality for Boolean algebras. 4. From natural duality to Priestley/Esakia duality and back again The main objective in this section is to investigate how the dualities presented in Theorem 3.3 fa- cilitate translation from the categorically well-behaved natural duality set-up to the more pictorial representation afforded by Priestley/Esakia duality for Heyting algebras. Before demonstrating how the translation operates we briefly recall the Priestley/Esakia duality. This has a long his- tory, and has been rediscovered and reformulated many times. By way of reference we note here Esakia's paper [14] and also the recent paper [2]. The relative pseudocomplement in a Heyting algebra is uniquely determined by the underlying lattice order. More precisely, we may assert that the forgetful functor U : H → D is faithful and U : H → U(H) is part of a categorical equivalence (actually an isomorphism); the inverse 9 V : U(H) → H maps each bounded distributive lattice L that admits a relative pseudocomplement to the unique Heyting algebra A such that U(A) = L. An algebra L ∈ D can be identified with its second dual K(X), where X = H(L). There exists a Heyting algebra B with U(B) = K(X) if and only if the Priestley space X = (X; (cid:54), T) has the property that ↓O is T-open whenever O is T-open; if this condition is satisfied we call X an Esakia space. Given Esakia spaces X and Y, a continuous order-preserving map ϕ : Y → X is such that K(ϕ) : K(X) → K(Y) preserves the relative pseudocomplement if and only ϕ is an Esakia morphism, meaning that ϕ(↑y) = ↑ϕ(y) for all y ∈ Y . In summary, there is a dual equivalence between the category of Heyting algebras and the category of Esakia spaces (with Esakia morphisms), obtained by restricting the duality between D and P to the subcategory U(H) and a certain subcategory E of P. As observed earlier, a Heyting algebra is a Godel algebra if and only if the associated Esakia space is a forest. In our formulation of the duality trees grow downwards. Restricting the func- tors HU and VK(cid:22) E we obtain a dual equivalence between the category G of Godel algebras and the category F of Esakia spaces whose order structure is a forest and Esakia morphisms. The category Gn is dually equivalent to the full subcategory Fn of Esakia spaces whose objects are forests of depth at most n − 2. Figure 2 summarises the various dual equivalences relating to Priestley/Esakia duality and their restrictions to full subcategories, shown by unlabelled vertical arrows. Gn Fn G F H VK(cid:22) E HU E K(cid:22) E U V U(H) D H(cid:22) U(H) HK P Figure 2. Priestley/Esakia duality for Godel algebras For fixed n (cid:62) 4 and each choice of σ ∈ Σn, we shall use Dσ and Eσ to denote the functors of Cn. Our immediate aim is to relate the dual space Dσ(A) to determined by the alter ego Cσ n∼ the Priestley/Esakia dual HU(A). Some word of explanation is needed before we demonstrate how ) whose class of objects is I(Dσ(Gn)). to do this. Let Yσ denote the full subcategory of IScP+(Cσ n∼ From Theorem 3.3 and the Priestley/Esakia duality for Gn, it is straightforward to see that Fn and Yσ are equivalent categories. Therefore one may ask: why present a description of Dσ(A) from HU(A), and vice versa, if this can be obtained using A as a stepping stone? The answer is that to prove the trivial fact that Fn and Yσ are equivalent is not our final goal. We want to reveal the very special connection between these categories which will be our primary tool in the final sections of the paper. Assume we have any finitely generated (quasi)variety A of distributive lattice-based algebras with forgetful functor U : A → D. In [3, Section 2] we presented a procedure for passing from the natural dual of an algebra A ∈ A to the Priestley dual HU(A) when the natural duality under consideration was obtained by the piggyback method. Here we carry out an analogous process, but now based on any of the dualities we established in Theorem 3.3. We shall do this by proving a variant of [3, Theorem 2.4], as this theorem applies to the special case in which A = Gn. This result -- Theorem 4.3 -- achieves more than the direct specialisation of the general result. The reason for this lies in the way in which, for A ∈ Gn, the layers of the Priestley space HU(A) are derived from the action of the maps gi (or fi) on D(A), and how the lifting of the chosen endomorphism hj relates these layers. (It is convenient to visualise HU(A) as being comprised of layers, each layer consisting of the elements at a particular depth; see Fig. 3 relating to Example 4.4.) n = IScP+(Cσ ), n∼ n−2, TX), where the partial Before we begin we recap on the form taken by the objects of the dual category Xσ where σ ∈ Σn. These are topological structures X = (X; σX 1 , σX 2 , . . . , σX 10 L. M. CABRER AND H. A. PRIESTLEY x X on X/(cid:39)σ X as follows: [x](cid:39)σ X [y](cid:39)σ X ⇐⇒ x (cid:54)(cid:39)σ X z and σX and let (cid:54)σ X is a partial order. (for 1 (cid:54) i (cid:54) n − 3) and the operation σX n−2 are obtained by pointwise lifting operations σX i of the corresponding operations σi on Cn and the domain of each partial operation is a closed substructure of X (see [4, Chapter 2] for details). Let (cid:39)σ X be the binary relation defined on X by x (cid:39)σ X y if and only if either x = y or there exists a sequence z0 = x, . . . , zN = y ∈ X such that, for ij ◦ zj+1. each j ∈ {0, . . . , N − 1}, there exists ij ∈ {1, . . . , n− 3} such that zj+1 = σX Then (cid:39)σ X is motivated by Lemma 2.3, which can be recast as follows. Lemma 4.1. Let σ ∈ Σn. Let A ∈ Gn and X = Dσ(A) and x, y ∈ X. Then X y. X is an equivalence relation on X. The definition of (cid:39)σ ij ◦ zj or zj = σX X does not destroy antisymmetry, so (cid:54)σ −1((cid:62)) ⇐⇒ ω ◦ x = ω ◦ y ⇐⇒ x (cid:39)σ The cluttered notation adopted below is temporarily necessary because we shall work simulta- X by neously with more than one alter ego. We denote the equivalence class of x ∈ X under (cid:39)σ [x](cid:39)σ X be the reflexive, transitive closure of(cid:6)σ −1((cid:62)) = y X . We now define a relation(cid:6)σ X(cid:6)σ antisymmetric relation(cid:6)σ Lemma 4.2. Let σ, τ ∈ Σn. Let A be an algebra in Gn and let X = Dσ(A) and X(cid:48) = Dτ (A) be the associated dual spaces. Then (i) (cid:39)σ (ii) (cid:54)σ for which(cid:6)σ Moreover, for any σ ∈ Σn, the relation (cid:54)σ We now prove (ii). Let x, y ∈ X be such that [x](cid:39)(cid:6)σ n−2(z) ∈ [y](cid:39). So x(cid:6)τ Proof. (i) follows directly from Lemma 4.1. Since we now know that the equivalence relations on X = Gn(A, Cn) obtained from σ and τ are the same we shall write simply (cid:39) for the relation and [x](cid:39) for the equivalence class of an element x in X. X [y](cid:39). Then x (cid:54)(cid:39) y and there exists z ∈ [x](cid:39) n−2(z) ∈ [y](cid:39). Since hj(n − 2) = n − 1 for any j ∈ {1, . . . , n − 2}, it follows that, for n−2(z))(a) = 1 ⇐⇒ z(a) ∈ {n − 2, n − 1} ⇐⇒ (ω ◦ τ X (ω ◦ σX (cid:48) n−2(z) (cid:39) σX By Lemma 4.1, τ X X coincides with (cid:54)τ Therefore (cid:54)σ X(cid:48). X(cid:48) y. We deduce that(cid:6)σ X y and ∃z(cid:0)x (cid:39)σ X is a partial order on X/(cid:39)σ X and (cid:39)τ X and (cid:54)τ X(cid:48) are equal; X(cid:48) are equal. X. Taking the reflexive, transitive closure of the n−2(z))(a) = 1. X and(cid:6)τ X y(cid:1), n−2(z) (cid:39)σ X of depth at most n − 2, X is the associated covering relation. for which σX a ∈ A, X(cid:48) are equal. X The final assertions follow from Lemma 2.1 and the way in which the order on X/(cid:39) is defined.(cid:3) Theorem 4.3. Let σ ∈ Σn. Let A ∈ Gn and X = Dσ(A) be its dual space. Let Y = X/(cid:39)σ and let T be the quotient topology derived from the topology of Dσ(A). Then Y = (Y ; (cid:54)σ X, T) is a Priestley space isomorphic to HU(A). Proof. By Lemma 4.2 we may assume that σ = (g1, . . . , gn−3, h1). We shall write (cid:39) in place of (cid:39)σ and omit subscripts from equivalence classes and from the order and covering relations on X/(cid:39). We know that the map Φω : x (cid:55)→ ω◦x from D(A) to HU(A) = (Z; (cid:54)Z, TZ) is surjective. Arguing just as in the proof of [3, Theorem 2.3] we proved that (Z; TZ) is homeomorphic to the quotient space (X/ ker(Φω); TX/ ker(Φω)). From the definition of Φω, we have Φω(x) = Φω(x) if and only if ω ◦ x = ω ◦ y. By Lemma 4.1, ker(Φω) coincides with the relation (cid:39) described in terms of the liftings of g1, . . . , gn−3. So we have identified (D(A)/(cid:39) ; T) with (Z; TZ). It remains to reconcile the order of the quotient space with that of HU(A). Since we are working with posets of finite depth it suffices to consider the covering relations. First suppose that [x](cid:6) [y]. 1 (z). Since i (cid:54) h1(i) for each i ∈ Cn, we have, for a ∈ A, So there exists z such that x (cid:39) z, y (cid:39) hX and hence ω ◦ x (cid:54) ω ◦ y in HU(A). ω(x(a)) = ω(z(a)) (cid:54) ω(h1(z(a))) = (ω ◦ hX 1 (z))(a)) = ω(y(a)) X 11 1 (z))) = j = d(ω ◦ y), we have ω ◦ (hX 1 ◦z) = {0}∪↑(n−j) and so ω◦x = ω◦z < ω◦(hX Conversely, assume that ω◦y covers ω◦x in HU(A). Assume d(ω◦y) = j and d(ω◦x) = j +1. By Lemma 2.2, there exists z ∈ Gn(A, Cn) such that ω◦ z = ω◦ x and ran(z) = {0}∪↑(n− 1− j). By Lemma 4.1, z (cid:39) x. Now observe that ran(hX 1 (z)). 1 (z)) = ω ◦ x. Therefore Since ↑(ω ◦ x) is a chain and d(ω ◦ (hX 1 (z) (cid:39) y. Hence [x] (cid:54) [y]. (cid:3) hX the proof of [7, Theorem 2.4]. But there is an important point to note. The relations (cid:39) and(cid:6) A retrospective look at [7] is due here. There are clear similarities between our proof of The- orem 4.3 and Davey's original proof of endodualisability of Gn [7, Theorem 2.4]. Lemma 4.1 establishes that, for each A ∈ Gn, our relation (cid:39)σ D(A) coincides with the relation R defined in are defined using the lifting of the (partial and non-partial) operations σi. Therefore, they are available in every space in X = IScP+(Cσ ), and not only those of the form D(A) for some A ∈ Gn. n∼ This difference becomes crucial in the following sections when we determine which dualities are full and which V -formations admit amalgamation. We now take a break from theory to discuss how translation works in practice. Example 4.4. Fix n = 5 and σ = (g1, g2, h3). We illustrate the passage from the natural dual to the Priestley/Esakia dual for the algebra C5. Here the elements of X = Dσ(C5) are exactly the endomorphisms of C5, on which g1, g2 and h3 act by composition. We label each endomorphism e of C5 by writing ran e \ {0, 4} (which uniquely determines e) as a string, as indicated in Fig. 3. For endomorphisms e and f we have (e, f ) ∈ graph gX if and only if i ∀a ∈ C5 (cid:0)(e(a), f (a)) ∈ graph gi(cid:1) for i = 1, 2. In order for this to hold it is necessary that i /∈ dom e and i + 1 /∈ ran f . In the figure, the solid arrows arrows indicate the action of h3. Dashed and dotted arrows indicate, respectively, the action of g1 and of g2. ∅ 2 1 3 x (cid:55)→ [x](cid:39) 12 13 23 123 Dσ(C5) = G 5(C5, C5) ∅ 1 2 3 12 13 23 123 HU(C5) Figure 3. Translation applied to the algebra C5 in G 5 Of course we have a special situation here because C5 is a chain. In general each layer of the dual space will not be a single (cid:39)-equivalence class. We elected here to use the endomorphism h3 in the alter ego, rather than the alternatives h1 and h2 supplied by Theorem 3.3, because the action of h3 on End C5 is especially simple. However one feature of the translation is present whichever of h1, h2 and h3 we include the alter ego: (cid:39) is determined solely by g1 and g2, whereas the ordering amongst (cid:39)-equivalence classes is determined solely by hi, whichever value of i we choose. 12 L. M. CABRER AND H. A. PRIESTLEY We should draw attention, however, to Theorem 5.1 below in which we show that in any application in which we need a full (or equivalently a strong) duality for Gn, then we must use h1 rather than any other hi. We now seek to demonstrate that the process for passing from D(Cn) to HU(Cn) (for n (cid:62) 4) is much less transparent using a duality based solely on endomorphisms (as in [7, 9]) than when we use any of the variants supplied by Theorem 3.3. Example 4.5. We shall consider the alter ego (C5; h1, h2, h3, T) for G 5. As shown by Davey and Talukder [9, Theorem 2.4], this yields an optimal duality. In Fig. 4, the action of h1, h2 and h3 on End Cn is shown by dashed, dotted and solid arrows, respectively. It can be seen that these maps do encode (cid:39) and that, on the associated quotient, we can recover the ordering of HU(C5). What does emerge clearly from this example is that translation from an endomorphism-based duality to Priestley/Esakia duality can be quite complicated, even on very simple objects. Moreover fully reconciling our approach with that in [7] is not a trivial exercise in practice, though the theory ensures that it is, of course, possible. ∅ 2 3 1 12 13 23 123 Figure 4. The dual of C5 for the duality for G 5 with alter ego (C5; h1, h2, h3, T) Theorem 4.6 presents the other half of the two-way translation process between Fn and Yσ, for a given σ ∈ Σn, as it applies to an object HU(A), where A ∈ Gn. Lemma 2.2 sets up, for the given algebra A, mutually inverse bijections ιA and γA between Gn(A, Cn) and a specified subset of HU(A) × ℘n. Starting from the Esakia space HU(A), we form suitable pairs with elements of ℘n and HU(A). We then form a topological structure X by equipping our set of pairs with a i and establish that X is isomorphic to the natural dual space Dσ(A). topology and operations σX We carry out this construction using only the order and topological structure of HU(A) (see the remarks following the theorem for the significance of this). The Priestley/Esakia duality applied to A ∈ Gn tells us that the evaluation map kA : A → KHU(A) is an isomorphism and the sets of the form kA(a), as a ranges over A, are precisely the clopen up-sets in HU(A). Moreover, we recall that D(A) is topologised with the subspace topology induced by the product topology on (Cn)A, where the topology on Cn is discrete. These observations underlie the way topology is handled in the theorem. Theorem 4.6. Let A be an algebra in Gn and let HU(A) = (Y ; (cid:54), T). Let X = { (y, U ) ∈ Y × ℘n U = d(y) + 1 and 0, n − 1 ∈ U }. Define partial maps gX restrictions: i and f X i on X as follows, where the domains are given by the indicated 13 gX i (y, U ) = (y, gi(U )) f X i (y, U ) = (y, fi(U )) if i /∈ U, if i + 1 /∈ U, and total maps hX j given by j (y, U ) =(cid:40)(y, hj(U )) (s(y), hj(U )) hX if n − 2 /∈ U, otherwise; here s denotes the function which, on HU(A) -- a forest of finite depth -- maps each non-maximal point to its unique upper cover and fixes each maximal point. For each clopen up-set W of HU(A) and each i ∈ Cn, let AW,i = { (y, U ) ∈ X i ∈ U and ↓i ∩ U = ↑x ∩ W + 1}. 1 , σX Let σ ∈ Σn. Then Dσ(A) ∼= X = (X; σX the family of sets of the form AW,i. Proof. Lemma 2.2 sets up mutually inverse bijections ιA : Gn(A, Cn) → X and γA : X → Gn(A, Cn), where ιA : x (cid:55)→ (ω ◦ x, ran x). From the description of γA given there, for each a ∈ A and i ∈ Cn, we have (γA(y, U ))(a) = i if and only if i ∈ U and ↓i ∩ U − 1 = {u ∈ ↑y u ∈ kA(a)}. Thus n−2, TX), where TX is the topology generated by 1 , . . . , σX and hence ιA determines a homeomorphism between (X; TX) and Dσ(A). In what follows gD(A) , f D(A) and hD(A) j i i denote the lifting of the (partial) maps gi, fi and hj to γA(AkA(a),i) = { x ∈ D(A) x(a) = i} Let i ∈ {1, . . . , n − 3} and x ∈ Gn(A, Cn). Then x ∈ dom(gD(A) Gn(A, Cn) for each i ∈ {1, . . . , n − 3} and j ∈ {1, . . . , n − 2}. that is, if i /∈ ran x. In this case ιA(gD(A) i i (x)) = ιA(gi ◦ x) = (ω ◦ gi ◦ x, ran(gi ◦ x)) = (ω ◦ f, gi(ran x)) = gX i (ιA(x)). ) if and only if ran x ⊆ dom gi, The proof that ιA ◦ f D(A) Then i = f X i ◦ ιA is similar. Finally, let j ∈ {1, . . . , n − 2} and x ∈ Gn(A, Cn). ιA(hD(A) j Here we have two cases. (x)) = ιA(hj ◦ x) = (ω ◦ hj ◦ x, hj(ran x)). If n − 2 /∈ ran x then ω ◦ hj ◦ x = ω ◦ x. ω ◦ hj ◦ x = s(ω ◦ x). We deduce that ιA preserves σi for 1 (cid:54) i (cid:54) n − 2. A remark is in order here on what we have really achieved in Theorem 4.6. We have already observed that our 'going back' construction builds (up to isomorphism) Dσ(A) ∈ Yσ solely from the topology and order of the Esakia space HU(A) (as encoded by the map s). That is, the construction is performed without directly involving the algebra A. This means that we can carry out the procedure on any Esakia space in Fn, regardless of whether or not the space is explicitly represented in the form HU(A) (as it can be, certainly). If n − 2 ∈ ran x then (cid:3) We introduced the category Yσ earlier but did not then give an explicit description of the equivalence between this category and Fn indicated in Fig. 5. We can now remedy this omission. Define Fσ : Yσ → Fn on objects by letting X; (cid:54)σ Now define Fσ(η), for each η ∈ Yσ(X, Y), as follows: X) = [η(x)](cid:39)σ Fσ(X) = (X/(cid:39)σ Fσ(η)([x](cid:39)σ X, TX/(cid:39)σ X). (for x ∈ X). The fact that Fσ(η) is well defined follows from the definition of (cid:39)σ Y and the fact that η preserves σ1, . . . , σn−3. Since η preserves σn−2, it is straightforward to check that Fσ(η) is an Esakia morphism. Then Fσ is a functor naturally equivalent to HU(cid:22) Eσ(cid:22) Y Gn Yσ . 14 L. M. CABRER AND H. A. PRIESTLEY Yσ Dσ Eσ Gσ Fσ Gn HU(cid:22) Gn VK(cid:22) Fn Fn Figure 5. The equivalence set up by the functors Fσ and Gσ As with Fσ, the assignment Y (cid:55)→ Gσ(Y) = (X; σX n−1, TX) can be extended to a functor from Fn into Yσ. In this case we will not present explicitly the action of Gσ on maps. For our purposes, it is enough to observe that Theorems 4.3 and 4.6 imply that Gσ can be made into a functor naturally equivalent to DσV K(cid:22) Fn ; equivalently, Fσ together with Gσ determine a categorical equivalence between Yσ and Fn. 1 , . . . , σX In the same way as we did in Section 3 we end this section with a comment about G 3 and G = (C3; h1, T) for C3, Davey [7, pp. 126 -- 127] shows how to obtain 2. Using the alter ego C3∼ H(A) from D(A) he first observes that H(A) and D(A) are homeomorphic as topological spaces. In our terms this means that x (cid:39) y if and only if x = y. This is actually the natural way to then the order Davey defines on D(A) is exactly the reflexive (transitive) closure of(cid:6). This is define (cid:39) in the absence of partial endomorphisms. Moreover, if we identify D(A)/(cid:39) with D(A) only one side of the translation; the other direction can be obtained by the same construction and argument used in Theorem 4.6. For G 2, the translation between the duality yielded by (C2; T) and the Priestley/Esakia duality is essentially that whereby a Boolean space is regarded as a special case of a Priestley space. 5. The quest for full dualities Our first application of the translations developed in Section 4 is to pick out the full dualities from among the dualities we developed in Theorem 3.3. In Theorem 5.1 we show that, for n (cid:62) 4, all choices of σ ∈ Σn except σ = (g1, . . . , gn−3, h1) lead to dualities which are not full. Our strategy is similar to one used by Davey; see his proof that End Cn fails to dualise Gn fully when n (cid:62) 4 [7, p. 127]. 2 and for G The primary tool for establishing that a natural duality is full is to establish that it is strong (see [4, Chapter 3] for the definitions and discussion). The dualities for G 3 yielded by the alter egos (C2; T) and (C3; h1, T) contain no partial operations. They are known to be strong (see [4, Theorem 4.2.3(ii)] for the case n = 3) and hence full. When n = 4, the fact that Cσ , with n∼ σ = (g1, h1), determines a (strong and hence) full duality for G 4 was proved by Davey and Talukder in [9, Theorem 6.1]. We shall show that, for any n (cid:62) 4, the dualising set {g1, . . . , gn−3, h1} yields a full duality. Our proof uses Theorems 4.3 and 4.6. It is this technique, and the fact that fullness is obtained directly, and not via strongness, that we wish to accentuate here. We note also that Davey's proof of fullness of his endomorphism-based duality for G 3 [7, pp. 126 -- 127] may be seen as essentially a very special case of our method. Our full dualities are necessarily strong; see [8, pp. 13 -- 14], where Godel algebras are called relative Stone Heyting algebras. (The paper [8], a stepping stone along the way to the final resolution in the negative of the longstanding "ful equals strong?" question, identifies various well-known varieties for which non-strong full dualities cannot be found.) As an aside, we note that a small generating set for the monoid Endp Cn is needed if axiomatisation of the dual category is to feasible. We can claim to have set up as good a full duality as is possible for addressing the axiomatisation problem for general n. However in this paper we shall not seek an extension of [9, Theorem 6.1], which relies both on a suitable generating set for Endp C4 being found by hand and on standardness arguments (see [9, Section 3]). Theorem 5.1. Let n (cid:62) 4 and σ ∈ Σn. Then the alter ego Cσ n∼ σ = (g1, . . . , gn−3, h1). 15 fully dualises Gn if and only if n and Fσ : Yσ n (cid:54)= IScP+(Cσ n∼ whose universe X is {j + 1, . . . , n − 1}. Observe that dom σX Proof. Assume first that σ (cid:54)= (g1, . . . , gn−3, h1). We divide the problem into two cases. Both proofs employ the same tool. Since the functors Gσ : Fn → Yσ n → Fn determine a categorical equivalence, X is isomorphic to Gσ(Fσ(X)) for every X ∈ Yσ n. Therefore we present a space X in IScP+(Cσ ) and we observe that X is not isomorphic to Gσ(Fσ(X)), which proves n∼ that Yσ ), that is, Cσ does not yield a full duality. In the proof below we shall omit n∼ subscripts and superscripts, for example from (cid:39), where these are clear from the context. Case 1: Assume that σi = fi for some i ∈ {1, . . . , n − 3}. Let j be such that σj = fj and σi = gi for any j < i (cid:54) n − 3. Let X be the sub- space of Cσ i = X if i (cid:54)= j and n∼ j = {j + 2, . . . , n − 1}. Now assume X ∈ Yσ. By assumption, the quotient space X/(cid:39) dom σX has two elements {j + 1, . . . , n − 2} and {n − 1}. Then Fσ(X) = ({ [n − 2](cid:39), [n − 1](cid:39) }; (cid:54), T), where T is the discrete topology and [n − 2](cid:39) < [n − 1](cid:39). The universe of Gσ(Fσ(X)) is {([n − 1](cid:39),∅), ([n − 2](cid:39),{1}), . . . , ([n − 2](cid:39),{n − 2})}. It follows that Gσ(Fσ(X)) = n − 1 and X = n − 1 − j. Therefore Gσ(Fσ(X)) and X are not isomorphic. Case 2: Assume that σi = gi for 1 (cid:54) i (cid:54) n − 3 and σn−2 = hi, where i (cid:54)= 1. With this assumption hi(1) = 1, and so {1, n − 1} is a closed subuniverse of Cσ n∼ whose universe X is {1, n − 1}. Observe that dom σX . Let X be the subspace of Cσ i = X if i (cid:54)= 1 and n∼ 1 = {n − 1}. Now assume X ∈ Yσ. Since σi = gi for 1 (cid:54) i (cid:54) n − 3, the definition of (cid:39) dom σX implies that X/(cid:39) has two classes {1} and {n − 1}. Because hi(1) = 1 and hi(n − 1) = n − 1, the space Fσ(X) is ({ [1](cid:39), [n − 1](cid:39) }; =, T), where T is the discrete topology. The universe of Z = Gσ(Fσ(X)) is {([1](cid:39),∅), ([n − 1](cid:39),∅)} and dom σZ 1 = {n − 1}, the spaces Z and X are not isomorphic. determines a duality on Gn, for each non-empty set S, the space (Cσ )S is isomorphic to the dual space of the S-generated free algebra FGn (S) in n∼ Gn [4, Corollary 2.24]. Therefore, to prove that Cσ determines a full duality, it is enough to prove n∼ that each closed substructure X of Dσ(FGn (S)), for some set S, is isomorphic to the dual space of some algebra in Gn. Now assume σ = (g1, . . . , gn−3, h1). Since Cσ n∼ 1 = Z. Since dom σX Let us fix a non-empty set S and a closed substructure X of (Cσ n∼ )S be such that x (cid:39) y. We claim that y ∈ X. By the definition of (cid:39) there is no loss of generality in assuming that x = gX i (y) or y = gX i (x), since X is a closed substructure of (Cσ i (y) then, for each s ∈ S, we n∼ have xs = gi(ys) and hence xs ∈ dom fi and fi(xs) = ys. Now observe that for i ∈ {1, . . . , n − 1}, the partial endomorphism fi equals gi−1 ◦ gi−2 ◦ ··· ◦ g2 ◦ g1 ◦ h1 ◦ gn−3 ◦ gn−2 ◦ ··· ◦ gi+2 ◦ gi+1. Then )S, it follows directly that y ∈ X. If x = gX i (x) for some i ∈ {1, . . . , n − 3}. If y = gX )S. Let x ∈ X and y ∈ (Cσ n∼ y = gX i−1 ◦ gX i−2 ◦ ··· ◦ gX 2 ◦ gX 1 ◦ hX 1 ◦ gX n−3 ◦ gX n−2 ◦ ··· ◦ gX i+2 ◦ gX i+1(x), and we have established our claim. Let A = FGn (S) and let Y = (Dσ(A)/(cid:39); (cid:54), TY) be as defined in Theorem 4.3. The fact that, for each x ∈ X, the class [x](cid:39) is contained in X implies that X/(cid:39) ⊆ Dσ(A)/(cid:39). Let Z be the substructure of Y whose universe is X/(cid:39). Since X is a closed subset of Dσ(A) and TY is the quotient topology, X/(cid:39) is a closed subset of ( Dσ(A)/(cid:39); TY ). From the fact that X is closed under h1 and the definition of (cid:54) in Dσ(A)/(cid:39), it follows that X/(cid:39) is an up-set in (Dσ(A)/(cid:39); (cid:54)). We conclude that Z belongs to Fn. Therefore there exists B ∈ Gn such that HU(B) ∼= Z. By Theorem 4.6, Dσ(B) ∼= Gσ(Z). By Theorems 4.3 and 4.6, the map ιA : Dσ(A) → GσFσ(Dσ(A)), defined by ιA(x) = (ω ◦ x, ran x), sets up an isomorphism. Since for x ∈ X the class [x](cid:39) is contained in X, the restriction of ιA to X determines a bijection between X and Gσ(Z). Moreover, ιA(cid:22) X is an isomorphism between X and Gσ(Z) in the category IScP+(Cσ ). Putting the components of the proof together n∼ (cid:3) we deduce that X ∼= Gσ(Z) ∼= Dσ(B). 16 L. M. CABRER AND H. A. PRIESTLEY 6. Applications: amalgamation and coproducts Our second application of our interactive duality theory concerns amalgamation in varieties of Godel algebras. Our main result here is Theorem 6.4. Along the way we expose the interrelation between satisfaction of the amalgamation property in a finitely generated quasivariety and the existence of an alter ego which is a total structure yielding a strong duality: the result given in Lemma 6.1 is not new, but we do present a simpler and more self-contained proof of it than that of [4, Lemma 5.3]. Given a class of algebras A, a V-formation is a quintuple (A, B, C, fB, fC) where A, B, C ∈ A, and fB ∈ A(A, B) and fC ∈ A(A, C) are injective homomorphisms. A V -formation (A, B, C, fB, fC) admits amalgamation if there exist an algebra D ∈ A and embeddings hB : B → D and hC : C → D such that hB ◦ fB = hC ◦ fC. The class A has the amalgamation property if every V -formation admits amalgamation. Let A be a quasivariety. It is well known that A admits all colimits and in particular pushouts, also known as fibred coproducts (see [19, Chapter III] for definitions and properties of colimit, pushout, and coproduct). Let (A, B, C, fB, fC) be a V -formation in A. Then (A, B, C, fB, fC) admits amalgamation if and only if the pushout maps pB : B → B(cid:96)A C and pC : B → B(cid:96)A C are embeddings. With this in mind, we shall focus below on the properties of pushouts, and particularly on when pushout maps are embeddings. Lemma 6.1. Let A = ISP(M) be the quasivariety generated by a finite algebra M and assume that there is a total structure M∼ which yields a strong duality on A. Then A has the amalgamation Proof. Let D : A → IScP+(M∼ ) and E : IScP+(M∼ ) → A be the functors determined by M∼ . By [4, Theorem 6.1.2 and Exercise 6.1], under our assumptions a homomorphism h in A is an embedding if and only if D(h) is surjective. property. Let (A, B, C, fB, fC) be a V -formation in A. Then D(fB) and D(fC) are surjections. Let X ←− D(C). That is, X is the subspace of D(B) × D(C) −→ D(A) be the fibred product D(B) whose universe is D(fB ) D(fC ) X = { (x, y) ∈ D(B) × D(C) D(fB)(x) = D(fC)(y)}. As we mentioned in Section 1, in [18] Maksimova proved that Gn fails to satisfy the amalgama- Since D(fB) and D(fC) are surjective, the projection maps πB : X → D(B) and πC : X → D(C) are also surjective. Since the duality is full, it follows that E(X) is the pushout of (A, B, C, fB, fC) and the pushout maps are E(πB) ◦ eB : B → E(X) and E(πC) ◦ eC : C → E(X). Since E(πB) ◦ eB (cid:3) and E(πC) ◦ eC are embeddings, the V -formation (A, B, C, fB, fC) admits amalgamation. tion property if n (cid:62) 4. Combining this with Lemma 6.1, we obtain the following. Corollary 6.2. The variety Gn, with n (cid:62) 4, does not admit a total (single-sorted ) strong duality. Given a natural duality, the problem of describing which maps between dual structures cor- respond to embeddings between algebras is not as simple in general as it is when the duality is strong and based on a total structure. However for the Godel algebra varieties Gn the task of describing such maps is greatly facilitated by two features which work to our advantage. Firstly, we can call on the two-way translation between our natural dualities and the Priestley/Esakia duality. Secondly, the latter duality is the restriction to Heyting algebras of Priestley duality, which is strong and has a total structure as the alter ego. Lemma 6.3. Let A, B ∈ Gn with n (cid:62) 2 and f : A → B be a homomorphism. For each σ ∈ Σn, the following statements are equivalent: (1) f is an embedding; (2) for each x ∈ Dσ(A), there exists y ∈ Dσ(B) such that Dσ(f )(y) (cid:39)σ Proof. The map f is an embedding if and only if HU(f ) is surjective. Since the functor HU is naturally isomorphic to FσDσ, the map HU(f ) is surjective if and only if Fσ(Dσ(f )) is surjective. Dσ (A) x. From the observations we made about the functor Fσ in Section 4 we have Fσ(Dσ(f ))([z](cid:39)σ [Dσ(f )(z)](cid:39)σ . Therefore Fσ(Dσ(f )) is surjective if and only if (2) holds. Dσ (B) Dσ (A) 17 ) = (cid:3) We are now ready to prove the main result of this section. Theorem 6.4. Let (A, B, C, fB, fC) be a V -formation in Gn with n (cid:62) 4. Let σ = (g1, . . . , gn−3, h1). Then the following statements are equivalent: (1) (A, B, C, fB, fC) admits amalgamation; (2) the dual maps Dσ(fB) and Dσ(fC) satisfy the following conditions: (a) for each x ∈ Dσ(B), there exist y ∈ Dσ(B) and z ∈ Dσ(C) such that x (cid:39)σ (b) for each x(cid:48) ∈ Dσ(B) such that x(cid:48) Dσ(fB)(y) = Dσ(fB)(z); and and Dσ(fC)(y(cid:48)) = Dσ(fB)(z(cid:48)). ∈ Dσ(C), there exist y(cid:48) ∈ Dσ(C) and z(cid:48) Dσ (B) y and Dσ (C) y(cid:48) (cid:39)σ , Dσ and Eσ, respectively. , D and E instead of Cσ n∼ Proof. In this proof we only consider σ = (g1, . . . , gn−3, h1), therefore we shall omit the super- scripts and write Cn∼ As in Lemma 6.1, let X be the subspace of D(B) × D(C) whose universe is { (x, y) ∈ D(B) × ←− D(C). By Theo- D(C) D(fB)(x) = D(fC)(y)}, that is, the fibred product D(B) is full. Therefore E(X) is the pushout of (A, B, C, fB, fC) rem 5.1, the duality determined by Cn∼ with pushout maps E(πB)◦ eB : B → E(X) and E(πC)◦ eC : C → E(X). Now observe that πB sat- isfies condition (2) in Lemma 6.3 if and only if (a) holds. This proves that E(πB) is an embedding (cid:3) if and only if (a) holds. Similarly, E(πC) is an embedding if and only if (b) holds. −→ D(A) D(fB ) D(fC ) Gn. observations give us access to certain coproducts in G. In particular we can calculate coproducts Our last application concerns coproducts of Godel algebras. The coproduct of a family of algebras depends not only on the algebras involved but also on the class in which we form the coproduct. In particular, given a set K of algebras in Gn ⊆ G, their coproduct in G might be different from their coproduct in Gn. Nonetheless, if the set K is finite, the coproduct formed in G coincides with the coproduct formed in Gm, for a suitably large m; in particular, it is enough to consider m = 2+(cid:80)A∈K(kA−2), where kA is the minimal k such that A ∈ Gk for all A ∈ K. These in G of any finite set of algebras belonging to(cid:83)n Natural dualities, when available, are a good tool for the study of coproducts: under such a duality, the dual space of the coproduct of a family of algebras corresponds to the cartesian product of their dual spaces. This fact was already noted in [7, Section 5], where it was applied to determine finitely generated free algebras in Gn and in G, with the calculations performed wholly within the natural duality setting. The main difficulty one encounters in attempting to use a natural duality to calculate the coproduct in Gn or in G of a family of algebras K ⊆ Gn lies in finding effective descriptions of the dual spaces of the algebras A ∈ K and then of the algebra dual to the resulting cartesian product. On the other hand, using Priestley/Esakia duality instead, as D'Antona and Marra do in [6], brings different challenges. While dual spaces may be easy to visualise, the duality functor (viz. the restriction of HU to Gn; see Fig. 2) does not in general convert a coproduct in Gn to a cartesian product, so that the dual space of a coproduct is hard to describe. D'Antona and Marra proceed in the following way to describe the coproduct of any finite family K of finite (equivalently, finitely generated) Godel algebras. They first find the Priestley/Esakia dual of each algebra in K. Then, using the fact that product distributes over coproduct in the category of Esakia spaces (see [11, p. 391], where it is proved that, in every variety of Heyting algebras, coproduct distributes over product), the problem reduces to describing the product of two finite trees in a suitable category. In bare outline, the authors' general strategy to construct the product of two trees is: first to represent each tree by a family of ordered partitions; second, to construct another family of ordered partitions from the original ones by suitably shuffling and merging these. Finally they must reverse the process to obtain a tree from the resulting set of ordered partitions. Our two-way translation allows us to work with a natural duality and Priestley/Esakia in combination, and so gives a simpler approach than one based on either of these duality techniques alone. We present our method as a 5-step procedure. We shall fix σ ∈ Σn to be (g1, . . . , gn−3, hn−2) 18 L. M. CABRER AND H. A. PRIESTLEY Figure 6. Calculating C3(cid:96)G5 C4: Steps 1 and 2 K, where K ⊆ Gn: be legitimate. To calculate(cid:96)Gn 1. For each A ∈ K, determine HU(A). 2. Use Theorem 4.6 to construct GσHU(A) from HU(A), for each A ∈ K. 3. Form the cartesian product X =(cid:81){ GσHU(A) A ∈ K}. 4. Determine (cid:39)σ X, (cid:54)σ 5. VK(X/(cid:39)σ X,(cid:6)σ X) as in Theorem 4.3. X; (cid:54)σ X, TX/(cid:39)σ X and (X/(cid:39)σ X) gives(cid:96)Gn X, TX/(cid:39)σ K. because this gives particularly simple pictures in examples, but any other choice of σ would also Of course the procedure described above does not constitute an algorithm unless restricted to a finite family of finite algebras. Some comments on our strategy as compared with that in [6] should be made. Our procedure allows us on the one hand to be flexible and to calculate coproducts in any subvariety Gk that contains the algebras, and on the other hand, most significantly, we replace the involved procedure of calculating shuffles and merges simply by the computation of a cartesian product of structured spaces. It might be seen as a disadvantage that we need to know at the outset in which G´odel subvariety Gn we need to work in order that the coproduct calculated in Gn coincides with that calculated in G. But this is a minor issue: since, as we observed before, a suitable n can easily be found once the finite family of algebras is fixed. From the perspective of the procedure developed in [6], one can similarly cut down to a finitely generated subvariety Gk. However to do so one needs first to calculate the Priestley/Esakia dual of the coproduct in G and then to truncate the resulting forest to obtain a forest of depth k − 2. uvxyzv∅u1u2u3z∅y1y2y3x12x13x23D(C4)D(C3)HU(C3)HU(C4)119 We now illustrate how our method works in practice. The first example we choose is the one given in [6, Examples 1 and 2]. This is done to enable the reader to compare and contrast the two procedures. (Observe that in [6] the order in the Priestley/Esakia spaces is the dual of the one considered in this paper.) σ = (g1, g2, h3). Example 6.5. Let us determine C3(cid:96)G C4. First observe C3(cid:96)G C4 belongs to G Step 1: Let HU(C3) = ({u, v}; (cid:54)) where u < v and HU(C3) = ({x, y, z}; (cid:54)) with x < y < z (see Fig. 6). Step 2: Figure 6 also shows Gσ(HU(C3)) and Gσ(HU(C4)). 5. We take v∅z∅v∅y1v∅y2v∅y3v∅x12v∅x13v∅x23u2z∅u2y1u2y2u2y3u2x12u2x13u2x23u1z∅u1y1u1y2u1y3u1x12u1x13u1x23u3z∅u3y1u3y2u3y3u3x12u3x13u3x231Figure7.CalculatingC3'G5C4:Step4120 L. M. CABRER AND H. A. PRIESTLEY Figure 8. Calculating C3(cid:96)G5 C4 : Step 4 Figure 9. Calculating C3(cid:96)G4 C4: Step 2 Step 3: The cartesian product Gσ(HU(C3)) × Gσ(HU(C3)) is represented in Fig. 7. Step 4: Figure 8 depicts the calculation of Fσ(Gσ(HU(C3)×Gσ(HU(C4)) and therefore isomorphic to HU(C3(cid:96)G5 Step 5: The lattice whose Priestley dual is shown in Fig. 8 is (the reduct of the Godel algebra) isomorphic to ⊥ ⊕(cid:0)(⊥ ⊕ (C3 × C3 × C2)) × C4 × C3(cid:1), where ⊕ denotes linear sum. C4). Example 6.6. Since C3 and C4 lie in G C4. Here Step 1 is the same as in Example 6.5. Steps 2 -- 4 are shown in Figs. 9 -- 11. We see that the tree X = Gσ(FσHU(C3) × FσHU(C4)) is (isomorphic to) a truncation of the one obtained in Step 4 of Example 6.5. It is only for n (cid:62) 5 that the coproduct in Gn coincides with that in G. Finally, to 4, we can also calculate C3(cid:96)G4 v∅z∅v∅y1v∅y2v∅y3u1z∅u2z∅u3∅u1y1u2y2u3y3u2y1u3y1y3y2v∅x12v∅x13v∅x23u2x12u3x13u3x23u1y2u1y3u2y3u1x12u1x13u2x23u2x13u3x12u1x231v∅u1u2z∅y1y2x12D(C4)D(C3)121 Figure 10. Calculating C3(cid:96)G4 C4: Step 3 Figure 11. Calculating C3(cid:96)G4 C4: Step 4 complete Step 5 we observe that the dual lattice of X is which has 82 elements. ⊥ ⊕ ((⊥ ⊕ (C2 × C2 × C2)) × C3 × C3), References [1] R. Balbes and Ph. Dwinger, Distributive Lattices. Missouri University Press (1974) [2] G. Bezhanishvili, N. Bezhanishvili, D. Gabelaia and A. Kurz, Bitopological dualities for distributive lattices and Heyting algebras. Math. Struct. Comput. Sci. 20, 359 -- 393 (2010) u1x12u2x12u1z∅u1y1u1y2u2y1v∅y1v∅z∅u2z∅u2y2v∅y2v∅x121v∅z∅v∅y1v∅y2u1z∅u2z∅u1y1u2y2u2y1v∅x12u2x12u1y2u1x12122 L. M. CABRER AND H. A. PRIESTLEY [3] L.M. Cabrer and H.A. Priestley, Coproducts of distributive lattice-based algebras. Algebra Universalis (to appear), arxiv:1308.4650 [4] D.M. Clark and B.A. Davey, Dualities for the Working Algebraist. Cambridge University Press (1998) [5] J. Czelakowski and D. Pigozzi, Amalgamation and interpolation in abstract algebraic logic. In Models, Algebras, and Proofs (eds. X. Caicedo and C.H. Montenegro), Lecture Notes in Pure and Appl. Mathematics Vol. 203, 187 -- 265 (1999) [6] O.M. D'Antona and V. Marra, Computing coproducts of finitely presented Godel algebras. Ann. Pure Appl. Logic 142, 202 -- 211 (2006) [7] B.A. Davey, Dualities for equational classes of Brouwerian algebras and Heyting algebras. Trans. Amer. Math. Soc. 221, 119 -- 146 (1976) [8] B.A. Davey, M. Haviar and T. Niven, When is a full duality strong?, Houston J. Math. (electronic) 33, 1 -- 22 (2007) [9] B.A. Davey and M.R. Talukder, Dual categories for endodualisable Heyting algebras: optimization and ax- iomatization. Algebra Universalis 53, 331-355 (2005) [10] B.A. Davey and H. Werner, Piggyback-Dualitaten. Bull. Austral. Math. Soc. 32, 1 -- 32 (1985) [11] B.A. Davey and H. Werner, Distributivity of coproducts over products. Algebra Universalis 12, 387 -- 394 (1981) [12] B.A. Davey and H.A. Priestley, Introduction to Lattices and Order, 2nd edn. Cambridge University Press (2002) [13] M.A.E. Dummett, A propositional calculus with a denumerable matrix. J. Symb. Logic 24, 96 -- 107 (1959) [14] L.L. Esakia, Topological Kripke models. Soviet Math. Dokl. 15, 147-151 (1974) [15] K. Godel, Zum intuitionistischen Aussagenkalkul. Anz. Akad. Wiss. Wien 69, 65 -- 66 (1932). Reprinted in Godel's collected Works Vol.1, Oxford University Press (1986), pp. 222 -- 225 added [16] T. Hecht and T. Katrin´ak, Equational classes of relative Stone algebras. Notre Dame J. Formal Logic 13, 248-254 (1972) [17] A. Horn, Free L-algebras. J. Symb. Logic 34, 475-480 (1969) [18] L.L. Maksimova, Craig's theorem in superintuitionistic logics and amalgamable varieties of pseudoboolean algebras, Algebra i Logika 16, 643 -- 681 (1977) [19] S. Mac Lane, Categories for the Working Mathematician. Grad. Texts in Math. Vol. 5, Springer-Verlag (1969) [20] J. von Plato, Skolem's discovery of Godel -- Dummett logic. Studia Logica 73, 153 -- 157 (2003) E-mail address: [email protected] (lmc) Dipartimento di Statistica, Informatica, Applicazioni, Universit`a degli Studi di Firenze, 59 Viale Morgani, 50134, Italy E-mail address: [email protected] Mathematical Institute, University of Oxford, Radcliffe Observatory Quarter, Oxford OX2 6GG, United Kingdom
0803.2800	4	0803	2010-02-11T18:30:58	Lie algebras of smooth sections	[ "math.RA", "math.DG" ]	Lie algebras of smooth sections are Lie algebras obtained from bundles of Lie algebras, where the latter are vector bundles of which the fibers are Lie algebras. We also consider the $\operatorname{C}^k$-sections for $k \in \mathbb{N}$. This paper studies the derivations, the centroid and the isomorphisms of such Lie algebras and generalizes some facts from Pierre Lecomte's publications in 1979 and 1980 to the case where the fiber is perfect or centerfree and it gives some more explicit proofs.	math.RA	math	Lie algebras of smooth sections Hasan Gundogan July 2007 Abstract Lie algebras of smooth sections are Lie algebras obtained from bundles of Lie algebras, where the latter are vector bundles of which the fibers are Lie algebras. We also consider the Ck-sections for k ∈ N. This paper, which is essentially my diploma thesis from May 2007 at Technische Universitat Darmstadt, studies the derivations, the centroid and the isomorphisms of such Lie algebras and generalizes some facts from Lecomte's publications [Le79] and [Le80] to the case where the fiber is perfect or centerfree and it gives some more explicit proofs. Contents 1 Introduction 1.1 Motivation and requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Remarks concerning notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Definitions, notions, former results 2.1 Associative algebras and Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Fiber bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Bundles of Lie algebras and the Lie algebra of sections . . . . . . . . . . . . . . . . . . . . 2.4 Lie connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Local operators and the Peetre Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . Cech cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6 2.7 Structure of Lie algebras of Ck-sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7.1 Center and commutator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7.2 Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7.3 Centroid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7.4 Isomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 2 3 3 13 20 22 24 25 26 26 27 34 38 1 Introduction 1.1 Motivation and requirements There are two main goals in the analysis of Lie algebras of smooth sections: On the one hand, the infinite- dimensional Lie algebras are not yet as well studied as the finite-dimensional ones and the methods used in finite-dimensional Lie theory are difficult to adapt to the infinite-dimensional case. I will discuss some properties of the Lie algebras of smooth sections in Chapter 3, especially their derivations, centroids and isomorphisms. A solid knowledge of analysis, linear algebra, topology and some knowledge of the theory of manifolds, e.g. a priori to be acquired in [Ne05], is required. However, there is no knowledge of Lie algebra theory required, for the relevant parts will be explained. 1 1.2 Remarks concerning notation I will follow some notational conventions which ought to be clarified. • The formulae A ⊆ B and B ⊇ A will mean that each element of the set A is an element of the set B and equality is not excluded. In order to describe a subset relation excluding equality I will write A ( B or B ) A. • All vector spaces and (bi)linear maps are considered over a field K ∈ {R, C}. • The set {0, 1, 2, . . .} is denoted by N and the set {0, 1, 2, . . .} ∪ {∞} is denoted by N. • A Ck-map, k ∈ N, is a k-times continuously differentiable function. A function is also called "smooth" instead of C∞. • A manifold M will always be smooth, finite-dimensional over R, hausdorff, paracompact and con- nected. • Let (e1, . . . , em) denote the canonical basis of Rm. Then (U, ξ) being a chart of a manifold M with dim M = m, x ∈ U and f : U → E being a C1-map between U and a finite-dimensional vector space E, we define the notation (∂ξi )x f := ∂ξi f (x) := d dt(cid:12)(cid:12)(cid:12)(cid:12)t=0 f(cid:0)ξ−1(ξ(x) + tei)(cid:1) . i=1 X i∂ξi for certain functions X 1, . . . X m ∈ C∞(U, R). For f ∈ C∞(M, R) we define X.f ∈ C∞(M, R) by (X.f )(x) := (Txf )(Xx) and we see that the two different definitions of (∂ξi )x are linked by the formula (∂ξi )x (f ) = ((∂ξi ) .f ) (x). As a basis of TxM we can take(cid:0)(∂ξ1 )x , . . . , (∂ξm )x(cid:1), where (∂ξi )x := Tξ(x)(cid:0)ξ−1(cid:1) (ei). We write ∂ξi for the map U → TxM , x 7→ (∂ξi)x. Any vector field X ∈ V(M ) takes the local formPm If f is a Ck-function and α = (α1, . . . , αm) ∈ Nm a multi-index with α :=Pm i=1 αi ≤ k, then we write: ξ f := ∂α1 ∂α ξ1 · · · ∂αm ξm f. For a multi-index α ∈ Nm we also define α! :=Qm can be stated as follows for x, y ∈ U and some z ∈ [0, 1] · (x − y) + y ⊆ U : i=1 αi. Then, if ξ(U ) is convex, the Taylor Formula f (x + y) = Xα<kQm i=1 ξi(y)αi α! · ∂α ξ f (x) + Xα=kQm i=1 ξi(y)αi α! · ∂α ξ f (z). Furthermore, we define the notation γ ≤ α for multi-indices α, γ ∈ Nm, which means γi ≤ αi for all i ∈ {1, . . . , m}, and we define the multinomial coefficients for multi-indices γ ≤ α ∈ Nm: (cid:18)α γ(cid:19) := α! γ!(α − γ)! . For multi-indices γ ≤ α ∈ Nm and any canonical basis vector ei of Rm, the multinomial coefficients satisfy the formula γ + ei(cid:19) =(cid:18)α (cid:18)α + ei γ(cid:19) +(cid:18) α γ + ei(cid:19) . • A Lie group is a manifold equipped with a smooth group multiplication whose inversion is smooth. • The neutral element of a group is denoted by 1 and if g is an element of the group, then λg is the multiplication by g from the left and ρg is the multiplication by g from the right. 2 • Depending on the context, the symbol 1 denotes the identity map of a vector space, the endomor- phism of a vector bundle where each 1x is the identity of the x-fiber, or a constant map with value 1 on a manifold. • For numbers m, n ∈ N\ {0}, the space of all m×n-matrices with entries in K is denoted by Mm,n(K) and is canonically identified with the space of linear functions Kn → Km. Note that, if we have m = 0 or n = 0, then Mm,n(K) is the vector space containing the unique linear map Kn → 0 or 0 → Kn, respectively. • If f : M → E is a Ck-map between a manifold and a finite-dimensional vector space, n ∈ N with n ≤ k and x ∈ M , then we define the notation jn x (f ) = 0 :⇐⇒ (f (x) = 0 jn−1 x (f ) = 0 and T n x f = 0 if n = 0 if n > 0. • If, for each point x ∈ M of a set, Tx = T (x) : N → P is a map and Ax = A(x) ∈ N is a point, then we denote the map M → P , x 7→ Tx(Ax) by T · A. If, for each point x ∈ M of a set, Tx = T (x) : N → P is a map and Sx = S(x) : L → N is a map, then we denote the map M → P L, x 7→ Tx ◦ Sx by T ◦ S. 2 Definitions, notions, former results 2.1 Associative algebras and Lie algebras Definition 2.1. Let A be a vector space. 1. A is called an algebra if equipped with a bilinear map A × A → A, (x, y) 7→ x · y, which we call the multiplication of the algebra A. 2. An algebra A is called associative, if x · (y · z) = (x · y) · z is satisfied for all x, y, z ∈ A. In this case the expression x · y · z is clear without ambiguity. An associative algebra A may possess an element 1 ∈ A with 1 · x = x · 1 = x for all x ∈ A, called unity element. 3. An algebra g is called a Lie algebra and its multiplication map is also written [·, ·] : g × g → g, (x, y) 7→ [x, y], if the following two conditions are satisfied: (a) [·, ·] is alternating, i.e. [x, x] = 0 for all x ∈ g. (b) the Jacobi identity, i.e. [[x, y] , z] + [[z, x] , y] + [[y, z] , x] = 0 for all x, y, z ∈ g. 4. If S, T are subsets of an algebra A, we define S · T to be the vector space generated by the set {s · t : s ∈ S, t ∈ T }. In the Lie algebra case we also write [S, T ] instead of S · T . 5. Let B be a vector subspace of an algebra A. It is called left ideal of A, if A · B ⊆ B, i.e. x · y ∈ B for all x ∈ A, y ∈ B. It is called right ideal of A, if B · A ⊆ B, i.e. y · x ∈ B for all x ∈ A, y ∈ B. It is called (two-sided) ideal of A, symbolically B E A, if it is a left ideal and a right ideal of A. In the Lie algebra case, the conditions A · B ⊆ B and B · A ⊆ B are equivalent. 6. A vector subspace B of an algebra A is called a subalgebra of A, symbolically B ≤ A, if x · y ∈ B for all x, y ∈ B, i.e. the multiplication of A induces a multiplication of B. Of course a subalgebra of A is an algebra, too. 7. Let g be a Lie algebra. The vector subspace [g, g] is called the commutator of g and g is perfect if g = [g, g]. The vector subspace { x ∈ g [x, y] = 0 for all y ∈ g} is called the center of g, denoted by z(g), and g is abelian if z(g) = g or, equivalently, [g, g] = 0. The Lie algebra g is called centerfree, if z(g) = 0. 3 8. A Lie algebra g is nilpotent, if its lower central series g0, g1, g2, . . ., which is defined by g0 := g and gn+1 := [g, gn] for n ∈ N, becomes zero eventually. 9. A Lie algebra g is solvable, if its derived series g(0), g(1), g(2), . . ., which is defined by g(0) := g and g(n+1) :=(cid:2)g(n), g(n)(cid:3) for n ∈ N, becomes zero eventually. Remark 2.2. In our case of K being a field of characteristic 0, an algebra multiplication is alternating if and only if it is skew-symmetric: On the one hand we have, for all x, y ∈ g : [x, y] = − [y, x] =⇒ [x, x] = − [x, x] =⇒ 2 [x, x] = 0 =⇒ [x, x] = 0. On the other hand, for all x, y, z ∈ g: [z, z] = 0 =⇒ 0 = [x + y, x + y] = [x, x] + [x, y] + [y, x] + [y, y] = [x, y] + [y, x] =⇒ [x, y] = − [y, x] . The next lemma (of which I leave out the elementary proof) shows four "canonical" ways to construct algebras. Lemma 2.3. 1. Let V be an arbitrary vector space. Then its vector space endomorphisms form an associative algebra where the multiplication of the algebra is ◦, the composition of functions. This associative algebra is called the endomorphism algebra of V , denoted by End(V )1. 2. Let A be an associative algebra. A equipped with the map [·, ·] : A × A → A, (a, b) 7→ a · b − b · a, called the commutator bracket of A, is a Lie algebra, denoted by AL. 3. A being an algebra, the vector space { f ∈ End(A) f (a · b) = f (a) · b + a · f (b) for all a, b ∈ A} is a Lie subalgebra of End(A)L, called the Lie algebra of derivations of A, denoted by Der(A). 4. A being an algebra and I E A an ideal, the quotient vector space A/I equipped with the well-defined induced multiplication A/I × A/I −→ A/I (a + I, b + I) 7−→ (a · b) + I is an algebra, called quotient algebra of A modulo I, denoted by A/I. Definition 2.4. The set of Lie algebra morphisms of g and h is Hom(g, h) := { f : g → h linear map f ([x, y]) = [f (x), f (y)] for all x, y ∈ g} . The set of Lie algebra isomorphimsm of g and h is Iso(g, h) := { f : g → h linear isomorphism f ([x, y]) = [f (x), f (y)] for all x, y ∈ g} . In the special case of g = h we even have a group Aut(g) := Iso(g, g) with neutral element 1 = idg, which is called group of Lie algebra automorphisms of g. Now it is time to give some important examples of algebras. Example 2.5. 1. Every vector space V can be equipped with the trivial Lie bracket : [x, y] := 0 for all x, y ∈ V . 1The symbol End(V ) will also be used to describe only the set of vector space endomorphisms of V . 4 2. S being a set and A being an algebra, the set AS of all mappings S → A can be pointwisely equipped with an algebra structure. An important subalgebra of such an algebra is Ck(M, K) ≤ KM , where M is a smooth manifold, K is equipped with the field multiplication as algebra multiplication and k ∈ N. 3. If (g, [·, ·]) is a Lie algebra and ϕ : V → g is an isomorphism of vector spaces, then we can define a Lie bracket on V as follows: [·, ·]ϕ : V × V −→ V (v, w) 7−→ ϕ−1 [ϕ(v), ϕ(w)] . Then ϕ becomes an isomorphism of the Lie algebras(cid:16)V, [·, ·]ϕ(cid:17) and (g, [·, ·]). 4. If V is a vector space, we call gl(V ) := End(V )L the general Lie algebra of V . 5. If V is a vector space and dim(V ) < ∞, we call sl(V ) := { f ∈ End(V ) tr(f ) = 0} the special Lie algebra of V . In fact, sl(V ) is a Lie subalgebra of gl(V ): tr ([f, g]) = tr(f ◦ g − g ◦ f ) = tr(f ◦ g) − tr(g ◦ f ) = tr(f ◦ g) − tr(f ◦ g) = 0 This calculation even shows that [gl(V ), gl(V )] ⊆ sl(V ). 6. The squared matrices over K with n rows form the associative algebra Mn,n(K) with respect to the multiplication of matrices. It is isomorphic to End (Kn). The corresponding Lie algebra (Mn,n(K))L isomorphic to gl (Kn) is denoted by gln(K) and called general linear Lie algebra of order n. Analogously, sln(K) := { A ∈ Mn(K) tr(A) = 0} is called special linear Lie algebra of order n and is isomorphic to sl(Kn). We identify End (Kn) , gl (Kn) , sl (Kn) with Mn(K), gln(K), sln(K), re- spectively. 7. Let β : V × V → K be a bilinear map. The set o(V, β) := { f ∈ End(V ) β(f (x), y) + β(x, f (y)) = 0} is a K-Lie subalgebra of gl(V ), called the Lie algebra of β-skew-symmetric endomorphisms of V . There are some important examples of this type of Lie algebra: (a) For V = Kn and β(x, y) =Pn algebra of order n, which can be identified with i=1 xiyi we write on(K) := o(V, β). This is the orthogonal Lie the set of skew-symmetric n × n-matrices. The space son(K) := on(K) ∩ sln(K) is a Lie subalgebra of on(K), called special orthogonal Lie algebra of order n.2 i=1 xiyi we write un(C) := o(V, β). This is the unitary Lie algebra (b) For V = Cn and β(x, y) =Pn of order n, which can be identified with (cid:8) A ∈ Mn(K) A + AT = 0(cid:9) , n A ∈ Mn(C) A + AT = 0o , the set of complex skew-hermitian n × n-matrices. Note that this is not a complex, but a real Lie algebra, since it is not a C-vector space. The space sun(C) := un(C) ∩ sln(C) is a Lie subalgebra of un(C), called special unitary Lie algebra of order n. 2In our considered case of K ∈ {R, C}, there is no difference between son(K) and on(K). The case son(K) ( on(K) is only possible if char(K) = 2. 5 (c) For V = K2n and β(x, y) = Pn symplectic Lie algebra of order n, which can be identified with i=1 xiyn+i − xn+iyi we write sp2n(K) := o(V, β). This is the Remark 2.6. B C −AT(cid:19) ∈ M2 (Mn(K)) ∼= M2n(K)(cid:12)(cid:12)(cid:12)(cid:12) B = BT , C = CT(cid:27) . (cid:26)(cid:18)A 1. One can show that for any Lie algebra g there is an associative algebra with unity U(g), such that g can be embedded into (U(g))L via an injective morphism of Lie algebras ηg : g → (U(g))L (cf. Part III of [Ne94]). If one demands a universal property3 of U(g) and ηg, then (U(g), ηg) is unique up to algebra isomorphism, i.e. an isomorphism of vector spaces preserving the multiplications, and (U(g), ηg) is called universal enveloping algebra of g. 2. The Ado Theorem states that each finite-dimensional Lie algebra is isomorphic to a Lie subalgebra of gln(K) for certain n ∈ N, i.e. each finite-dimensional Lie algebra is, up to isomorphism, a Lie algebra of squared matrices. Definition 2.7. Let g be a Lie algebra. 1. If V is a vector space, then a representation of g on V is a morphism of Lie algebras 2. A g-module is a vector space V with a K-bilinear map ρ : g −→ gl(V ). µ : g × V −→ V (x, v) 7−→ µ(x, v) =: x.v [x, y] .v = x.(y.(v)) − y.(x.(v)) satisfying the equation for all x, y ∈ g and v ∈ V . It is not difficult to obtain a bijective correspondence between the concepts of g-module and repre- sentation of g: Given a module with multiplication µ : g × V → V one can define a representation by r(µ) : g −→ gl(V ) x 7−→ (v 7→ µ(x, v)). Given a representation ρ : g −→ gl(V ) one can define a module structure on V by m(ρ) : g × V −→ V (x, v) 7−→ ρ(x)(v). It is easy to verify that r(m(ρ)) = ρ and m(r(µ)) = µ for every representation ρ and every module multiplication µ. A linear map f : V → W between two modules is called module morphism, if f (x.v) = x.f (v) for all x ∈ g, v ∈ V . The vector space of module morphisms V → W is denoted by Homg(V, W ) and we write Endg(V ) := Homg(V, V ). A vector subspace W ⊆ V of a module which is invariant under g is itself a module, called submodule. Images and kernels of module morphisms, quotients of modules modulo submodules and intersections of submodules are again modules. Note that the direct sum (in the sense of vector subspaces) of two submodules is also a submodule. 3 If A is an associative algebra with unity and α : g → AL is a morphism of Lie algebras, then there is a unique morphism of associative algebras α′ : U (g) → A such that α′(1) = 1 and α′ ◦ ηg = α. 6 3. A module V 6= 0 is called (a) simple, if 0 is its only proper submodule. (b) semisimple, if for each of its submodules W there is a complementary submodule W ′ ⊆ V such that W ⊕ W ′ = V . (c) trivial, if g.V = 0. Some useful facts about simple and semisimple modules are presented in the following two lemmas, shown e.g. in [Ne02]: Lemma 2.8. 1. Submodules and quotient modules of semisimple modules are semisimple, too. 2. The following statements are equivalent for a Lie algebra g and a g-module V : (a) V is semisimple. (b) V is a sum of simple modules. (c) V is a direct sum of simple modules. Lemma 2.9. (Schur Lemma) Let V, W be simple g-modules of a Lie algebra g. 1. Homg(V, W ) = 0, if V is not isomorphic to W . 2. Each non-zero element of Endg(V ) is invertible. 3. If dim(V ) < ∞ and K = C, then Endg(V ) = C · 1. Definition 2.10. Let g be a Lie algebra. We define the adjoint representation 4 of g to be the map ad : g −→ Der(g) x 7−→ adx := (y 7→ [x, y]) . The Jacobi identity for g ensures that the range is properly chosen. Note that z(g) = ker(ad). The adjoint representation of g corresponds to the g-module structure on g, where the multiplication is the Lie bracket of g. In this sense the ideals of g are the submodules of g. We use this change of perspective in order to define the notions "simple" and "semisimple" for Lie algebras. Definition 2.11. Let g be a Lie algebra and consider it as a g-module with respect to the adjoint representation. 1. g is a simple Lie algebra, if it is a non-trivial simple g-module. 2. g is a semisimple Lie algebra, if it is the direct sum of non-trivial simple g-modules, i.e. of simple Lie algebras. 3. g is a reductive Lie algebra, if it is the direct sum of simple g-modules., i.e. ideals of g. Example 2.12. The Lie algebras sln(K) for n ≥ 2 and son(K) = on(K) for n ≥ 5 and sp2n(K) for n ≥ 1 are simple. The Lie algebras gln(K) for n ≥ 1 are reductive and dim z (gln(K)) = 1. For proofs, cf. e.g. [Ne02]. Remark 2.13. If g = a ⊕ b is a decomposition of a Lie algebra into a direct sum of ideals, then [a, b] ⊆ a ∩ b = 0. 4ad represents g on g: ad[x,y](z) = [[x, y] , z] = − [y, [x, z]] + [x, [y, z]] = [adx, ady] (z) =⇒ ad[x,y] = [adx, ady]. 7 The relation between reductive and semisimple Lie algebras is described in the following lemma. For a proof, cf. e.g. [Ne02]. Lemma 2.14. Let g be a Lie algebra. 1. If g is semisimple, then it is reductive, perfect, centerfree and Der(g) = ad(g). So ad : g → Der(g) is an isomorphism of Lie algebras. 2. If g is reductive, then g = [g, g] ⊕ z(g) and [g, g] is semisimple. In particular, a reductive Lie algebra is semisimple if and only if it is centerfree if and only if it is perfect. Example 2.15. Let g be a two-dimensional K-vector space with basis (x1, x2). The unique skew- symmetric bilinear map [·, ·] : g × g → g determined by the relation [x1, x2] := x1 is a Lie bracket. The Lie algebra (g, [·, ·]) is centerfree, but not perfect, thus not reductive. Let h be a five-dimensional K-vector space with basis (y1, . . . , y5). The unique skew-symmetric bilinear map {·, ·} : h × h → h determined by the relations {y1, y2} := y1, {y1, y3} := y2, {y1, y4} := y3, {y2, y3} := y4, {y2, y4} := y5, {y1, y5} := {y2, y5} := {y3, y4} := {y3, y5} := {y4, y5} := 0. is a Lie bracket. The Lie algebra (h, {·, ·}) possesses the non-zero center K · y5, but it is perfect, thus not reductive. An interesting vector subspace of End(g) is the centroid of g. Definition 2.16. The commutant of the adjoint representation or centroid of a Lie algebra g is defined as follows: Cent(g) : = { f ∈ End(g) [f, adx] = 0 for all x ∈ g} = { f ∈ End(g) f ◦ adx = adx ◦f for all x ∈ g} = { f ∈ End(g) f [x, y] = [x, f (y)] for all x, y ∈ g} . Note that for f ∈ Cent(g) and x, y ∈ g we have: [f (x), y] = − [y, f (x)] = −f [y, x] = f [x, y] = [x, f (y)] . For a finer analysis of the structure of Cent(g), we introduce special elements in an endomorphism algebra. Definition 2.17. Let End(V ) be the endomorphism algebra of a vector space V and g a Lie algebra. 1. A map f ∈ End(V ) is nilpotent, if there exists n ∈ N such that f n = 0.5 A map f ∈ End(V ) is semisimple, if for each f -invariant subspace W ⊆ V there exists a complementary f -invariant subspace W ′ ⊆ V such that W ⊕ W ′ = V . 2. We define the subsets N(g) := { f ∈ Cent(g) f nilpotent}, S(g) := { f ∈ Cent(g) f semisimple} and the vector subspace J(g) := { ϕ ∈ Cent(g) adx ◦ϕ = 0 = ϕ ◦ adx for all x ∈ g}. Lemma 2.18. Let g be a Lie algebra. 1. Cent(g) is an associative subalgebra of End(g). 2. We have the inclusions Cent(g) ◦ Der(g) ⊆ Der(g) and [Cent(g), Der(g)] ⊆ Cent(g) or, to say it in other words, Der(g) is a Cent(g)-module and Der(g) acts by module derivations on Cent(g). 3. We have the inclusion [Cent(g), Cent(g)] ⊆ Hom(g/ [g, g] , z(g)). In particular, if g is perfect or centerfree, then Cent(g) is abelian. In the latter case, if g is of finite dimension, then N(g), S(g) are associative subalgebras of Cent(g) with Cent(g) = N(g) ⊕ S(g). 5Note that in the case of dim V = d < ∞ this is equivalent to f d = 0. 8 4. If g is of dimension d ∈ N and semisimple, then N(g) = 0, thus Cent(g) = S(g). Proof. 1. Clearly, Cent(g) is a vector subspace of End(g). If f, g ∈ Cent(g), then for all x, y ∈ g we have: (f g) [x, y] = f (g [x, y]) = f ([x, g(y)]) = [x, f (g(y))] = [x, (f g)(y)] . So, f g is also in Cent(g). Therefore Cent(g) is an associative subalgebra of End(g). 2. Let f ∈ Cent(g), g ∈ Der(g) and x, y ∈ g. Then: [f, g] [x, y] = f (g [x, y]) − g(f [x, y]) = f ([gx, y] + [x, gy]) − g([x, f y]) = f ([gx, y]) + f ([x, gy]) − [gx, f y] − [x, gf y] = [gx, f y] + [x, f gy] − [gx, f y] − [x, gf y] = [x, f gy] − [x, gf y] = [x, f gy − gf y] = [x, [f, g] (y)] . This implies [Cent(g), Der(g)] ⊆ Cent(g). We may also calculate: (f g) [x, y] = f ([gx, y] + [x, gy]) = [(f g)x, y] + [x, (f g)y] . We thus get Cent(g) ◦ Der(g) ⊆ Der(g). 3. Let f, g ∈ Cent(g) and x, y ∈ g. We calculate: [[f, g] (x), y] = [f (gx) − g(f x), y] = f [gx, y] − g [f x, y] = f g [x, y] − gf [x, y] = [f x, gy] − [f x, gy] = 0. Therefore [f, g] (g) ⊆ z(g). If z ∈ [g, g] is arbitrarily chosen, then there exist finitely many elements x1, y1, x2, y2, . . . , xn, yn ∈ g such that z =Pn nXi=1 [f, g] (z) = i=1 [xi, yi] and thus we obtain f g [xi, yi] − gf [xi, yi] = 0 implying [f, g] ([g, g]) = 0 and so [f, g] identifies with a vector space morphism g/ [g, g] → z(g). Now let Cent(g) be abelian and g of finite dimension. Obviously, the subsets N(g), S(g) are closed under the multiplication by scalars in K. Let f, g ∈ N(g) and n, m ∈ N such that f n = gm = 0. Obviously, f ◦ g is also nilpotent. Since Cent(g) is abelian, we have [f, g] = 0 and so we can apply the Binomial Theorem, yielding (f + g)n+m+1 = = k n+m−1Xk=0 (cid:18)n + m − 1 n−1Xk=0(cid:18)n + m − 1 (cid:19) f kgn+m−1−k n − 1 − k(cid:19) f n−1−kgm+k + m−1Xk=0(cid:18)n + m − 1 n + k (cid:19) f n+kgm−k−1 = 0 + 0 = 0. Thus f + g is also nilpotent. So N(g) is an associative subalgebra of Cent(g). Let f, g ∈ S(g). We want to show that f + g and f ◦ g are also semisimple. If K = R, then we use Corollary V.2.8 of [Ne94] to say that h ∈ End(g) would be semisimple if and only if its complexification hC := idC⊗g ⊗h was semisimple. So we only consider the case K = C. By Lemma V.3.3 of [Ne94], "semisimple" means the same as "diagonalizable" for K = C. Since Cent(g) is abelian, we have [f, g] = 0 and so we can apply a theorem about simultaneous diagonalzation (cf. e.g. Theorem 11.B.15 of [StWi90]): Diagonalizable endomorphisms of a vector space of finite dimension are simultaneously diagonalizable if and only if they commute. So f and g are diagonal 9 with respect to a fixed basis, thus also f + g and f ◦ g, so they are semisimple, too. This shows that S(g) is an associative subspace of Cent(g). The Jordan Theorem V.3.5 of [Ne94] yields that each element f ∈ Cent(g) decomposes into the sum of a nilpotent vector space endomorphism fn and a semisimple one fs and because of the fact that f commutes with any adx for x ∈ g this is also satisfied for fn and fs. Since Cent(g) is abelian, any two endomorphisms n, s with f = n + s, where n ∈ N(g) and s ∈ S(g), commute with f and so the theorem also implies that n = fn and s = fs. This shows Cent(g) = N(g) ⊕ S(g). 4. If f ∈ N(g), then f d(g) = 0. On the other hand, f (g) is an ideal of g because for x, y ∈ g we have [y, f (x)] = f [x, y] ∈ f (g). Since f (g)n ⊆ f n(g) for all n ∈ N, this implies f (g)d = 0, so f (g) is a nilpotent ideal of g. But g is semisimple, so this ideal is trivial and f = 0. Remark 2.19. 1. Note that any finite-dimensional associative commutative C-algebra A with unity element 1 can be realized as the centroid of a perfect and centerfree C-Lie algebra g. We show this in several steps. (a) Let k be an finite-dimensional simple C-Lie algebra, e.g. k = sl2(C). Then it is central, i.e. Cent(k) = Endk(k) = K · 1, by the Schur Lemma. We define g := k ⊗ A, understood as the tensor product of K-vector spaces. By setting [(x ⊗ a), (y ⊗ b)] := [x, y] ⊗ ab for all generating elements x, y ∈ k and a, b ∈ A and extending [·, ·] to a C-bilinear map g × g → g, we obtain a Lie algebra (g, [·, ·]). (b) Since k is simple, it is perfect. This implies that g is also perfect: An element v ∈ g takes the i=1 xi ⊗ ai for certain x1, . . . , xn ∈ k and a1, . . . , an ∈ A and each xi can be written j=1 [yij, zij] for some yi1, zi1, . . . , yiri , ziri ∈ k. So we have form v =Pn as xi =Pri v = [yij, zij] ⊗ ai = riXj=1 nXi=1 (c) Furthermore, g is centerfree because k is so: If v =Pm and a1, . . . , am ∈ A, then, for all x ∈ k, we have: riXj=1 nXi=1  0 = [v, x ⊗ 1] = [xj ⊗ aj, x ⊗ 1] = [yij ⊗ ai, zij ⊗ 1] . [xj , x] ⊗ aj. mXj=1 j=1 xj ⊗ aj ∈ z(g) for some x1, . . . , xm ∈ k mXj=1 nXi=1 But this is only possible, if, for all j ∈ {1, . . . , m}, we have aj = 0 or xj ∈ z(k) = 0, thus v = 0. (d) We may consider k ⊆ g by the embedding k ֒→ g, x 7→ x ⊗ 1 and g becomes a k-module by restriction of the adjoint representation. By [Ne02], there are x1, . . . , xn, x1, . . . , xn ∈ k such i=1 ad(xi) ◦ ad(xi) is an element in Cent(k) and, by [Hu72] on page 122, it is equal to λ · 1 for some λ ∈ C\ {0} and so we may assume, after normalizing: thatPn ad(xi) ◦ ad(xi) = 1. If a ∈ A is an arbitrary element, then fa :=Pn i=1 ad(xi ⊗ a) ◦ ad(xi ⊗ a) is in Cent(g) and fa(x⊗b) = x⊗ab for all x ∈ k and b ∈ A. So Cent(g) = Endg(g) is generated by endomorphisms of the type ad(x) ⊗ 1, where x ∈ k, and those of the type fa, where a ∈ A. We obtain: Cent(g) = Endg(g) ∼= Endk(k) ⊗ A ∼= K ⊗ A ∼= A. 10 2. If g is an arbitary Lie algebra, then the subsets N(g), S(g) ⊆ Cent(g) are not necessarily closed under the addition in End(g): Let g be the abelian complex Lie algebra C2. Then Cent(g) = End(g) is, as associative algebra, isomorphic to M2,2(C). We can write and the two matrices on the right hand site are nilpotent, but the matrix on the left hand site is not. And we can write 0 1 (cid:18)0 1 1 0(cid:19) =(cid:18)0 1 (cid:18)0 0(cid:19) =(cid:18)0 0 0(cid:19) +(cid:18)0 0(cid:19) +(cid:18) 0 0(cid:19) −1 0(cid:19) 0 1 2 1 1 and the two matrices on the right hand site are semisimple, but the matrix on the left hand site is not. 3. Let g and h be Lie algebras. The quotient Lie algebra g/ [g, g] and the Lie subalgebra z(h) ⊆ h are abelian and so the Lie algebra morphisms g/ [g, g] → z(h) are just the linear maps g/ [g, g] → z(h). In particular, we have Hom(g/ [g, g] , z(h)) = 0 if and only if g is perfect or h is centerfree. There is a natural isomorphism between linear maps ϕ : g/ [g, g] → z(g) and linear maps ϕ : g → z(g) with adx ◦ϕ = 0 = ϕ ◦ adx for all x ∈ g: Given ϕ, we set ϕ(x) := ϕ(x + [g, g]). Given ϕ, the map ϕ is well-defined by ϕ(x + [g, g]) := ϕ(x) because x + [g, g] = y + [g, g] implies the existence of a1, b1, . . . , an, bn ∈ g such that x − y =Pn nXi=1 ϕ(x) − ϕ(y) = i=1 [ai, bi] and thus ϕ[ai, bi] = nXi=1 ϕ ◦ adai (bi) = 0. We obtain: Hom(g/ [g, g] , z(g)) ∼= { ϕ ∈ Hom(g, z(g)) adx ◦ϕ = 0 = ϕ ◦ adx for all x ∈ g} = J(g). 4. Let g be a Lie algebra such that z(g) ⊆ [g, g]. For all f ∈ Cent(g), g ∈ J(g), x, y ∈ g we have: f g([x, y]) = f (0) = 0 [f g(x), y] = f [g(x), y] = f (0) = 0. gf [x, y] = g [f (x), y] = 0 [gf (x), y] = [g(f (x)), y] = 0 g2(x) = g(g(x)) ∈ g(z(g)) ⊆ g([g, g]) = 0. Thus, in this case, J(g) is a two-sided ideal of Cent(g) with J(g)2 = 0. Definition 2.20. A Lie algebra g is called decomposable if it is the direct sum of two proper ideals. If there is no such decomposition, g is called indecomposable. Remark 2.21. A simple Lie algebra is automatically indecomposable, since 0 is its only proper ideal. If an indecomposable Lie algebra g is reductive, then it is one-dimensional and trivial or it is simple because g = z(g) ⊕ [g, g] implies g = z(g) or g = [g, g] and in the latter case a decomposition of the semisimple Lie algebra [g, g] into simple Lie algebras may only have one summand. In general, indecomposable Lie algebras are not reductive, e.g. the Lie algebra g in Example 2.15. Lemma 2.22. Let g be a Lie algebra of finite dimension with abelian centroid, e.g. g is perfect or i=1 gi into a direct sum of non-zero ideals gi, where centerfree. Then there exists a decomposition g =Ln each gi is an indecomposable Lie algebra and this decomposition is unique except for the order. 11 Proof. The existence of a decomposition into indecomposable non-zero ideals is proven by mathematical induction over the finite dimension of g. The only difficult part is the one of the uniqueness: Let j=1 hj be two such decompositions with corresponding systems of orthogonal projections p1, . . . , pn and q1, . . . , qm. The projections are in Cent(g), as we can show with the following easy calculation: i=1 gi =Lm g =Ln pi([x, y]) = pi([xi, yi] + [x′, y′]) = [xi, yi] = [xi + x′, yi] = [x, pi(y)] , to the decompositionLn where z = zi + z′ is the decompositions of an element z ∈ g into gi and its complement with respect i=1 gi. Since Cent(g) is abelian, we have pi ◦ qj = qj ◦ pi for i ∈ {1, . . . , n} and j ∈ {1, . . . , m}. Now fix an i. Then pi induces a projection hj → hj for each j ∈ {1, . . . , m} and so, by the indecomposability of the hj , we obtain pi = 0 on hj or pi = 1 on hj . But from this we deduce that gi = im(pi) is the direct sum of some hj's. We conclude, by the indecomposability of gi, that there is exactly one j ∈ {1, . . . , m} such that gi = hj. The following theorem is due to M´edina and Revoy, cf. [MeRe93], and gives us information about the centroids of Lie algebras which are not necessarily perfect or centerfree. Theorem 2.23. Let g be a Lie algebra of finite dimension such that z(g) ⊆ [g, g]. 1. There are indecomposable idempotents p1, . . . , pn ∈ Cent(g) which are pairwise orthogonal, i.e. pi ◦ pj = δij pi for i, j ∈ {1, . . . , n}, satisfying Pn 2. Setting gi := pi(g) for i ∈ {1, . . . , n} we have a decomposition g = Ln non-zero ideals. i=1 pi = 1. i=1 gi into indecomposable 3. Each Cent(gi) is isomorphic to pi ◦ Cent(g) ◦ pi, its quotient Cent(gi)/ J(gi) modulo the maximal nilpotent ideal J(gi) is a field. i,j=1 Cij into ideals, each Cij is a Cent(gi)-Cent(gj)-bimodule and, if i 6= j ∈ {1, . . . , n}, then the vector i6=j=1 Cij 4. Setting Cij := pi ◦ Cent(g) ◦ pj for i, j ∈ {1, . . . , n} we have the decomposition Cent(g) =Ln space Hom(gj / [gj, gj] , z(gi)) is isomorphic to Cij . Furthermore, J(g) =Ln i=1 J(gi) ⊕Ln is also a decomposition into ideals. We also need the following result, Propsition 22.1 of [La01], to show a more general proposition about the uniqueness of the decomposition of a finite-dimensional Lie algebra into indecomposable non-zero ideals. Theorem 2.24. Let R be a ring with unity element 1. Suppose there exists a decomposition of 1 ∈ R into a sum of indecomposable idempotents, say 1 = c1 + . . . + cr, such that cicj = δij ci and ci ∈ Z(R) for all i, j ∈ {1, . . . , n}. Then the decomposition is unique except for the order. e.g. g is reductive and dim z(g) ≤ 1. Then this decomposition is unique except for the order. Proposition 2.25. Let g be a finite-dimensional Lie algebra such that z(g) ⊆ [g, g] with a decomposition i=1 gi such that Hom(gi/ [gi, gi] , z(gj)) = 0 if i 6= j ∈ {1, . . . , r}, into indecomposable non-zero ideals Ln Proof. By Theorem 2.23, we have R := Cent(g) ∼=Ln Cent(g)/ J(g) ∼= Ln g into indecomposable non-zero ideals, say g = Ln i=1 J(gi) and the quotient i=1 Cent(gi)/ J(gi) is commutative. Now suppose there are two decompositions of j=1 hj with corresponding systems of orthogonal projections p1, . . . pn and q1, . . . , qm. Let f ∈ Cent(g) and f = f1 + . . . + fn its decomposition into elements of the Cent(gi) ⊆ Cent(g). Then, for fixed i ∈ {1, . . . , n}, we have i=1 Cent(gi), J(g) =Ln i=1 gi = Lm nXk=1 pi ◦ f = nXk=1 pi ◦ fk = pi ◦ fi = fi = fi ◦ pi = fk ◦ pi = f ◦ pi. So the pi are in Z(R) and, by a dual argument, all the qj are so. By Theorem 2.24, this implies {p1, . . . , pn} = {q1, . . . , qm}, thus {g1, . . . , gn} = {h1, . . . , hm}. 12 Remark 2.26. In general, the decomposition of a finite-dimensional Lie algebra g into indecomposable non-zero ideals is not unique and the requirements of the preceding proposition are sharp: Let g = g1 ⊕ g2 be a decomposition into indecomposable non-zero ideals with corresponding indecomposable projections p1, p2 and Hom(g1/ [g1, g1] , z(g2)) 6= 0. If 0 6= ϕ : g → g is a morphism mapping g1 to z(g2) and the 1 := (1 + ϕ) ◦ p1 ◦ (1 + ϕ)−1 = p1 + ϕ is also an indecomposable spaces [g1, g1] and g2 to 0, then p′ idempotent of Cent(g). Thus g = im p′ 1) is another decomposition of g into indecomposable non-zero ideals. 1 + im(1 − p′ Remark 2.27. If g1, . . . , gn are the unique indecomposable non-zero ideals of a Lie algebra g and f ∈ Aut(g) is a Lie algebra automorpism, then of course g1, . . . , gn are the unique indecomposable non- zero ideals of f (g) and thus there is a permutation σ ∈ Sn such that f (gi) = gσ(i) for all i ∈ {1, . . . , n}. Lemma 2.28. Let g be a Lie algebra of finite dimension. 1. If K = C, then g is indecomposable if and only if S(g) = C · 1. 2. If K = R, then g is indecomposable if and only if either S(g) = R · 1 or S(g) = R · 1 + R · J for a complex structure J on g, i.e J 2 = −1. In the second case, J and −J are the only complex structures on g compatible with the Lie bracket. Proof. 1. For any f ∈ Cent(g) the eigenspaces of f are ideals of g: If y 6= 0 is an eigenvector of f for the eigenvalue λ and x ∈ g then f [x, y] = [x, f y] = [x, λy] = λ [x, y], i.e. [x, y] is also an eigenvector of f for the eigenvalue λ. If K = C, then the semisimple endomorphisms in Cent(g) are exactly the diag- onalizable endomorphisms in Cent(g). Thus g is indecomposable if and only if each endomorphism in S(g) has exactly one eigenvalue if and only if S(g) = C · 1. 2. If K = R and g is decomposable into the direct sum of non-trivial ideals g1 and g2 with corresponding orthogonal projections p1 and p2, then p1p2 = 0 although p1 6= 0 6= p2, thus Cent(g) has zero-divisors and is neither isomorphic as associative R-algebras to R nor to C. If K = R and g is indecomposable, then any endomorphism in S(g) admits exactly one real eigenvalue or exactly two non-real, complex conjugate eigenvalues. If every endomorphism in S(g) admits exactly one real eigenvalue, then S(g) = R · 1. In the other case, if any f ∈ S(g) admits the non-real complex eigenvalues a + ib and a − ib, then J := 1 b (f − a1) ∈ S(g) is a complex structure of g because of the following calculation for the eigenvector x ∈ g for the eigenvalue a ± ib: J 2(x) = ( = 1 b2 (f − a1)2)(x) = 1 b2 ((a ± ib)2x − 2a(a ± ib)x + a2x) = 1 b2 (f 2(x) − 2af (x) + a2x) 1 b2 (a2 ± 2iab − b2 − 2a2 ∓ 2iab + a2)(x) = −x. The space of semisimple endomorphims in Cent(g) in this case is the space of real linear combination of 1 and J, thus isomorphic to C as associative algebras over R. Since the complex conjugation is the only R-linear algebra automorphism on C different from the identity, the only complex structures on S(g) are J and −J. 2.2 Fiber bundles The definitions of various bundles in this subsection are based on [tD91], [Hu75] and [BiCr01]. Our notion "bundle" is what sometimes is called "fiber bundle", i.e. the fibers of a bundle are all "of the same type". 13 Definition 2.29. 1. Let E, M, F be manifolds such that there is surjective submersion π : E → M and each fiber Ex := π−1(x) for x ∈ M is diffeomorphic to F . If there is an open cover U = (Ui)i∈I of M and a family of smooth maps Φ = (ϕi)i∈I , each ϕi : π−1(Ui) → F inducing a diffeomorphism (π, ϕi) : π−1(Ui) → Ui × F (local triviality), then the sextuple (E, M, π, F, U, Φ) is called a bundle with bundle space E over the base space M with projection π, fiber F and bundle atlas (U, Φ). A pair (Ui, ϕi) is called bundle chart. As a shorter formulation we will say "π : E → M is a bundle with fiber F " without explicit mention of the bundle atlas. 2. A bundle π′ : E′ → M ′ is a subbundle of a bundle π : E → M provided E′ ⊆ E and M ′ ⊆ M and π′(x) = π(x) for all x ∈ E′. Remark 2.30. Since we only consider paracompact manifolds as base spaces, we may always assume that the bundle atlas of a bundle and a corresponding6 atlas of the base spaces are locally finite. For each bundle there is a class of interesting functions, the so-called sections. Definition 2.31. Let π : E → M be a bundle and k ∈ N. A Ck-map X : M −→ E x 7−→ X(x) = Xx is called Ck-section of the bundle, if π ◦ X = idM . The set of all Ck-sections of the bundle π : E → M is denoted by Γk(E) and we also write Γ(E) instead of Γ∞(E). We now give a first definition of a vector bundle. Definition 2.32. Let V be a finite-dimensional vector space and π : E → M a bundle with fiber V with a vector space structure on each fiber Ex for x ∈ M , isomorphic to V by (ϕi)Ex : Ex → V for each i ∈ I. Then the sextuple (E, M, π, V, U, Φ) is called a vector bundle. As a shorter formulation we will say "π : E → M is a vector bundle with fiber V ". Definition 2.33. Let πi : Ei → N be a vector bundle with fiber Vi for i = 1, 2. 1. A vector bundle morphism from π1 : E1 → N to π2 : E2 → N is a map κ : E1 → E2 such that the following diagram E1 κ E2 π2 } π1 N commutes and κ : (E1)x → (E2)x is a vector space morphism for all x ∈ N . The space of vector bundle morphisms from π1 : E1 → N to π2 : E2 → N is denoted by Hom(E1, E2). A vector bundle morphism from π1 : E1 → N to π1 : E1 → N is also called vector bundle endomorphism of π1 : E1 → N . 2. A vector bundle morphism is called vector bundle map, if fibers are bijectively mapped on each other and, thus, the fibers are isomorphic vector spaces. 3. A vector bundle morphism κ is a vector bundle isomorphism, if there exists a vector bundle mor- phism ι ∈ Hom(E2, E1), such that κ is inverse to ι. In this case the vector bundles π1 : E1 → N and π2 : E2 → N are called isomorphic. Note that vector bundle isomorphisms are vector bundle maps. 6The cover of the base space is the same. 14 / / } A second definition of a vector bundle needs the definition of a principal bundle and a bundle associated to a principal bundle. Definition 2.34. Let G be a Lie group and π : P → M a bundle with fiber G such that π is identic to the canonical projection of a smooth right action R : P × G −→ P (p, g) 7−→ p · g = Rg(p) = Rp(g). If the following diagram commutes for all i ∈ I, g ∈ G π−1(Ui) Rg π−1(Ui) ϕi G ϕi ρg / G, then the septuple (P, M, π, G, R, U, Φ) is called a principal bundle with structure group G and bundle action R. As a shorter formulation we will say "π : P → M is a G-principal bundle" without explicit mention of the bundle action. Remark 2.35. Let π : P → M be a G-principal bundle. 1. G acts freely on P by R, i.e. p · g = p for some p ∈ P , g ∈ G implies g = 1. To see this, let p · g = p for some p ∈ P , g ∈ G and let (U, ϕ) be a bundle chart in x := π(p). We calculate: ϕ(p) = ϕ(p · g) = ϕ(p)g =⇒ 1 = g. It is easy to show now that R acts simply transitively on its orbits, i.e. Rp : G → P is injective. for all p ∈ P the map 2. The condition that the surjective submersion π : P → M is identic to the canonical projection πR : P → P/G of the action R can be weakened in the following sense: If they both are surjective submersions, then M and P/G are already diffeomorphic. To show this, let h be a map P/G → M such that the following diagram commutes: P πR π =z z z z z z z h z P/G M . The map h is smooth because πR is a submersion and h is surjective because π is. It is even bijective because h(cid:0)πR(p)(cid:1) = h(cid:0)πR(q)(cid:1) implies π(p) = π(q) and there exists a bundle chart (U, ϕ) and an element g ∈ G such that (π(p), ϕ(p)) = (π(q), ϕ(q) · g) = (π(q), ϕ(q · g)) yielding p = q · g, thus πR(p) = πR(q). Furthermore, for any x ∈ M there is an open neighbourhood U ⊆ M such that there exists a section X : U → π−1(U ). Then h−1 maps y ∈ U on πR(X(y)) ∈ P/G and is smooth. So h is a diffeomorphism P/G → M . If π : P → M is a G-principal bundle and F is a manifold on which G acts from the left, then we can construct a new bundle with fiber F by replacing the structure group G by F . Definition 2.36. Let π : P → M be a G-principal bundle and F a manifold with a left action L : G × F −→ F (g, f ) 7−→ g · f = Lg(f ) = Lf (g). 15 / / / / / = By defining a left action as follows L′ : G × (P × F ) −→ P × F (g, (p, f )) 7−→ g · (p, f ) :=(cid:0)p · g−1, g · f(cid:1) . and considering the orbit space B := (P × F )/G, we obtain a bundle π′ : B → M with fiber F , called the bundle associated to the G-principal bundle π : P → M with fiber F .7 Let us examine its structure: The bundle projection is π′ : B −→ M [(p, f )] := G · (p, f ) 7−→ π(p) and this is well-defined because π(p) = π(p · g−1) for all p ∈ P, g ∈ G. The bundle atlas (Ui, ϕi)i∈I of the G-principal bundle π : P → M induces a bundle atlas (Ui, ϕ′ i)i∈I of π′ : B → M by ϕ′ i : π′−1(Ui) −→ F [(p, f )] 7−→ ϕi(p) · f and this is well-defined because [(p, f )] = [(q, e)] yields the existence of g ∈ G such that p · g−1 = q and g · f = e and then ϕi(q) · e = ϕi(p · g−1) · g · f = ϕi(p) · g−1 · g · f = ϕi(p) · f. Finally, each fiber Bx = π′−1(x) for x ∈ M is diffeomorphic to F : π′−1(x) = { [(p, f )] ∈ (P × F )/G π(p) = x} = { [(p, f )] ∈ (P × F )/G p ∈ Px} = (Px × F )/G ∼= F. Now we can define a vector bundle as being a bundle associated to a principal bundle. Definition 2.37. Given a finite-dimensional vector space V and a Lie subgroup G of GL(V ) acting on V 8 and a G-principal bundle π : P → M , the bundle π′ : B → M associated to the G-principal bundle π : P → M with fiber V is called a vector bundle. Remark 2.38. 1. A vector bundle π′ : B → M in the sense of Definition 2.37 allows a vector space : Bx → V structure on each fiber Bx for x ∈ M , defined by the condition that all restrictions (ϕ′ are vector space isomorphisms and this is well-defined because i)Bx (ϕ′ i)−1 Bx (λϕ′ i(G · (p, v)) + ϕ′ i(G · (q, w))) = (ϕ′ = (ϕ′ = (ϕ′ i)−1 Bx i)−1 Bx i)−1 Bx (λϕi(p)(v) + ϕi(q)(w)) (ϕi(p)(λv + w)) (ϕ′ i(G · (p, λv + w))) = G · (p, λv + w) does not depend on the choice of the bundle chart (Ui, ϕ′ i), where x ∈ Ui, λ ∈ K, G · (p, v), G · (q, w) ∈ Bx. Therefore every vector bundle in the sense of Definition 2.37 is a vector bundle in the sense of Definition 2.32. 2. On the other hand, let π : E → M be a vector bundle with fiber V in the sense of Definition 2.32. By setting P := [x∈M Iso (V, Ex) , 7In this formulation we neglect the dependence of the bundle structure on the actions R and L. 8This condition may be replaced by the existence of an smooth group morphism Ξ : G → GL(V ) (a so-called Lie group representation) and G acting on V as follows: g.v := Ξ(g)(v). 16 we obtain a surjective submersion ρ : P → M by setting ρ(f ) := x if f ∈ Iso (V, Ex), where the smooth structure on P is induced by the bundle atlas (Ui, ϕi)i∈I of the bundle π : E → M by declaring each map ρ−1(Ui) −→ Ui × GL(V ) f 7−→(cid:16)x, (ϕi)Eρ(f ) ◦ f(cid:17) to be a diffeomorphism. We obtain a GL(V )-principal bundle ρ : P → M with bundle action P × GL(V ) −→ P (f, T ) 7−→ f ◦ T and we define ρ′ : B = (P × V )/ GL(V ) → M to be the bundle associated to this principal bundle with fiber V . It is isomorphic to the vector bundle π : E → M via κ : B −→ E GL(V ) · (f, v) 7−→ f (v). We will see that, under certain conditions, the structure group may be reduced to a closed subgroup G of GL(V ). Definition 2.39. Let π′ : B → M be a vector bundle with fiber V associated to the G-principal bundle π : P → M with bundle atlas (Ui, ϕ′ i)i∈I and (Ui, ϕi)i∈I , respectively. We also define an index set as follows: I := { (i, j) ∈ I × I Ui ∩ Uj 6= ∅}. For (i, j) ∈ I and x ∈ Ui ∩ Uj we define an element gji(x) ∈ G by (π, ϕj ) ◦ (π, ϕi)−1 : (Ui ∩ Uj) × G −→ (Ui ∩ Uj) × G (x, g) 7−→ (x, gji(x)g) . Note that, if G ≤ GL(V ) or G is injectively representated in GL(V ), this element can also be defined by (cid:0)π′, ϕ′ j(cid:1) ◦ (π′, ϕ′ i)−1 : (Ui ∩ Uj) × V −→ (Ui ∩ Uj) × V (x, v) 7−→ (x, gji(x)(v)) . We obtain smooth functions gji : Ui ∩Uj → G for (i, j) ∈ I, called transition maps. The family (gji)(i,j)∈I is called cocyle relative to the bundle atlas (Ui, ϕi)i∈I . As a shorter formulation we will say "π′ : B → M is a vector bundle with fiber V and cocycle (gji)(i,j)∈I ." We have: gji(x) · gik(x) · gkj(x) = 1 for all x ∈ Ui ∩ Uj ∩ Uk. (1) Given another bundle atlas (Ui, ψi)i∈I (the cover U is the same!) and the cocyle (kji)(i,j)∈I relative to this bundle atlas, we define smooth mappings hi : Ui → G for i ∈ I by (π, ψi) ◦ (π, ϕi)−1 : Ui × G −→ Ui × G (x, g) 7−→ (x, hi(x)g) . The two cocyles are then linked by the formula kji(x) = hj(x) · gji(x) · hi(x)−1 for all x ∈ Ui ∩ Uj. (2) Cech cohomology deals with abstract cocyles fulfilling the conditions (1) and (2) and the equivalence classes of cohomological cocyles. We will see details in Subsection 2.6. 17 Remark 2.40. If G ≤ GL(V ) is a closed subgroup (and hence a Lie subgroup) containing the images of the transition maps of a vector bundle π : E → M , then one can show (cf. Theorem I.6.4.1 of [Hu75]) that π : E → M is isomorphic to the bundle associated to a G-principal bundle with base space M and fiber V . Another equivalent way to define vector bundles is given by Proposition I.5.5.1 of [Hu75]. Proposition 2.41. Let V be a finite-dimensional vector space, G a closed subgroup of GL(V ), M a manifold, (Ui)i∈I an open cover of M and gji : Ui ∩ Uj → G for (i, j) ∈ I a familiy of maps meeting the condition (1), where I := { (i, j) ∈ I × I Ui ∩ Uj 6= ∅}. Then there is a vector bundle π : E → M with fiber V , bundle atlas (Ui, ϕi)i∈I and cocyle (gji)(i,j)∈I and it is unique up to isomorphism. Knowing different ways to define and construct vector bundles, it is now time to give some important examples of vector bundles. Example 2.42. 1. Let V be a finite-dimensional vector space and M a manifold. Then, by setting E := M × V , we obtain a vector bundle π : E → M, (m, v) 7→ m with fiber V , called the trivial vector bundle with base space M and fiber V . 2. Let M be a m-dimensional smooth manifold. Then the tangent bundle T M is the disjoint union of tangent spaces Sx∈M TxM provided with a topology and a smooth structure like in [Ne05], Definition II.1.8. If π : T M → M , π(TxM ) = {x} is the natural projection, then π : T M → M is a vector bundle with fiber Rm. The set of smooth sections of this bundle is denoted by V(M ) or Γ(T M ) and these sections are called vector fields. 3. Let πℓ : Eℓ → M be a vector bundle with fiber Vℓ, bundle atlas (Ui, ϕℓ ji)(i,j)∈I for ℓ = 1, 2. Theorem 2.41 yields the existence of the vector bundle Π : Hom(E1, E2) → M with fiber Hom(V1, V2), bundle atlas (Ui, Φi)i∈I and cocyle (Gji)(i,j)∈I , where we define the mappings Gji : Ui ∩ Uj → GL(Hom(V1, V2)) by i )i∈I and cocyle (gℓ Gji(x)(f ) := g2 ji(x) ◦ f ◦ g1 ji(x)−1 for f ∈ Hom(V1, V2), x ∈ Ui ∩ Uj and (i, j) ∈ I. In order to discuss sections of bundles locally, we introduce the following notation. Definition 2.43. Let (U, ϕ) be a bundle chart of a vector bundle π : E → M with fiber V . The local form of a section X ∈ Γk(π−1(U )), where k ∈ N, is X ϕ ∈ Ck(U, V ), defined by claiming the following diagram to be commutative: π−1(U ) (π,ϕ) / U × V 9t t t t t t t t t (idU ,X ϕ) X U. t t If N is a set and η : Γk(π−1(U )) → N is a map, then the corresponding local form of η is denoted by ηϕ : Ck(U, V ) → N . A very useful type of cover of the base space of a bundle is the so-called Palais cover. Definition 2.44. If π : E → M is a bundle, then(cid:0)(Vi, ψi, ξi, ρi)i∈J , (Jt)r (a) (Vi)i∈J is a locally finite cover of M , t=1(cid:1) is called Palais cover, if (b) (Vi, ψi)i∈J is a bundle atlas of E, 18 / O O 9 (c) (Vi, ξi)i∈J is an atlas of M , (d) (ρi : M → [0, 1])i∈J is a partition of unity such that supp(ρi) is a compact subset of Vi for all i ∈ J, (e) {J1, . . . , Jr} is a partition of J such that i, j ∈ Jt and Vi ∩ Vj 6= ∅ already implies i = j for all t ∈ {1, . . . , r}. If the base space of a bundle is paracompact, then there is always a Palais cover. Theorem 2.45. A bundle with paracompact base space possesses a Palais cover. Proof. Since M is paracompact, there is surely (Vj, ψj, ξj, ρj)j∈J fulfilling the conditions (a)-(d). By Theorem I in Chapter 1.2 of [GrHaVa72], there is a finite set S and a refinement (Vsk)s∈S,k∈N of (Vj )j∈J , such that for each s ∈ S we have Vsk ∩ Vsℓ = ∅ if k 6= ℓ ∈ N. By restriction of the ψj, ξj and ρj we obtain induced mappings ψsk, ξsk, ρsk and a Palais cover(cid:16)(Vsk, ψsk, ξsk, ρsk)s∈S,k∈N , ({s0, s1, s2, . . .})s∈S(cid:17). The following lemma will be very useful in some future proofs. Lemma 2.46. If we have a vector bundle π : E → M with fiber V , a section A ∈ Γ(E), a point x ∈ M and an integer s ∈ N such that js x(A) = 0, then there is an integer r ∈ N and there are sections A1, . . . , Ar ∈ Γ(E) and functions a1, . . . , ar ∈ C∞(M, R) with a1(x) = . . . = ar(x) = 0 such that A = rXi=1 as+1 i Ai. (3) Proof. Let n be the vector space dimension of V and m the dimension of M . The lemma will be proven in two steps. 1. We consider the case of M being a convex open neigbourhood U ⊆ Rm of x = 0 and E = U × Kn and A ∈ C∞(U, Kn) with compact support F ⊆ U . Then, by the Taylor Formula, there is a family of functions Aα ∈ C∞ (U, Kn), α ∈ Nm, α = s + 1, such that A(y) = Xα=s+1 1 α! · yα1 1 · · · yαm m · Aα(y) for all y ∈ U . Let W ⊆ U be an open set with F ⊆ W and ρ : U → [0, 1] be a smooth function with compact support such that ρW ≡ 1. Then we have A(y) = ρ(y)s+2 · A(y) = Xα=s+1 (ρ(y)y1)α1 · · · (ρ(y)ym)αm · 1 α! · ρ(y) · Aα(y) (4) for all y ∈ U . By the Multinomial Theorem, we have (t1 + . . . + tm)k = Xα=k(cid:18) k α!(cid:19) tα1 1 · · · tαm m for t1, . . . , tm ∈ R and, by the inversion formula presented in [MoHeRaCo94], there are, for α ∈ Nm, scalars λij = λij (α) ∈ R and µj = µj(α) ∈ R, where i ∈ {1, . . . , m} and j ∈ {1, . . . , p} for an integer p = p(α) ∈ N, such that tα1 1 · · · tαm m = pXj=1 µj (λ1jt1 + . . . + λmj tm)α . 19 Equation (4) then turns into: A(y) = Xα=s+1 = Xα=s+1 λ1jρ(y)y1 + . . . + λmjρ(y)ym  p(α)Xj=1 } aα(y)s+1 · eAα(y). {z =:aα(y) s+1 · µj α! · ρ(y) · Aα(y) =: eAα(y) {z } for this local case. Since eAα is of compact support and aα(0) = 0 for all α ∈ Nm with α = s + 1, we have shown (3) 2. For the general case, we take an x-neigbourhood U ⊆ M , a bundle chart (U, ϕ), a chart (U, ξ) of M such that ξ(x) = 0 and ξ(U ) is convex and a smooth function ρ : M → [0, 1] such that supp(ρ) is a compact subset of U and ρW ≡ 1 for an x-neighbourhood W ⊆ U . By the argument in the first step, we my write (ρ · A)ϕ ◦ ξ−1 = eAi for smooth functions eA1, . . . eAr ∈ C∞(ξ(U ), V ) and ea1, . . . ,ear ∈ C∞(ξ(U ), R) with ea1(x) = . . . = i)ϕ(y) := eAi(ξ(y)) for ear(x) = 0. For i ∈ {1, . . . r}, we define Ai ∈ Γ(E) by Ai := ρ · A′ all y ∈ U and we define ai ∈ C∞(M, R) with a1(x) = . . . = ar(x) = 0 by ai := ρ · (eai ◦ ξ). This yields A =Pr i, where (A′ i Ai on M . i=1 as+1 i rXi=1eas+1 2.3 Bundles of Lie algebras and the Lie algebra of sections Now we are ready to define a central objects of this diploma thesis: Bundles of Lie algebras and the corresponding Lie algebra of Ck-sections. The definitions of this subsection are taken from [Le80]. Definition 2.47. Let k be a Lie algebra and M a manifold. A bundle of Lie algebras π : L → M with fiber k is a vector bundle π : L → M with fiber k which has an Aut(k)-valued cocyle (gji)(i,j)∈I . We now define additional structure on a bundle of Lie algebras and on the corresponding space of Ck-sections. Definition 2.48. Let π : L → M be a bundle of Lie algebras with fiber k and k ∈ N. 1. We justify the name "bundle of Lie algebras" by defining a Lie bracket on each fiber Lx for x ∈ M by a bundle chart (Ui, ϕi) in x and the Lie bracket [·, ·] on k (cf. Example 2.5.2): [·, ·] : Lx × Lx −→ Lx (v, w) 7−→ [v, w]ϕi = (ϕi)−1 Lx [ϕi(v), ϕi(w)] . In order to see that this is well-defined, we make use of the fact that all maps gij(x) for (i, j) ∈ I are isomorphisms of Lie algebras: [v, w]ϕj = (ϕj)−1 Lx = (ϕi)−1 Lx = [v, w]ϕi [ϕj(v), ϕj (w)] = (ϕi)−1 Lx [gij(x)(ϕj (v)), gij (x)(ϕj (w))] = (ϕi)−1 Lx (gij(x) [ϕj(v), ϕj (w)]) [ϕi(v), ϕi(w)] . 20 2. With the help of the Lie algebra structure on each fiber of L, we pointwisely define a Lie bracket on the vector space Γk(L) by [X, Y ] (x) := [X(x), Y (x)] for X, Y ∈ Γk(L), x ∈ M. 3. Γk(L) turns out into a topological Lie algebra by embedding it into the space Ck(M, L) equipped with the Ck-compact-open topology, i.e. embed it into the topological productQk by X 7→(cid:0)T iX(cid:1)k i=0 , where each C(cid:0)T iM, T iL(cid:1) is equipped with the compact-open topology. Definition 2.49. The Lie algebra Γk(L) is called Lie algebra of Ck-sections. In the case of k = ∞ we call Γ∞(L) = Γ(L) also Lie algebra of smooth sections. i=0 C(cid:0)T iM, T iL(cid:1) Remark 2.50. 1. In the case of M := {0} we have L = {0} × k ∼= k and Γk(L) = Ck ({0} , {0} × k) ∼= k. So the finite-dimensional Lie algebras are special Lie algebras of Ck-sections. 2. The Lie algebra Γk(L) is a Lie algebra of order zero in the following sense: For sections X, Y ∈ Γk(L) and x ∈ M , the expression [X, Y ] (x) depends on X, Y only via their "zeroth order parts", namely X(x), Y (x). The Lie algebra of smooth vector fields V(M ) = Γ(T M ) is a Lie algebra of order 1. There is a useful lemma about Lie algebras of Ck-sections, which we want to prove right now. Lemma 2.51. Let π : L → M be a bundle of Lie algebras with fiber k and k ∈ N. Then for all x ∈ M the evaluation map evx : Γk(L) −→ Lx X 7−→ Xx is a surjective morphism of Lie algebras. Proof. The function evx is clearly linear. It is also compatible with the Lie bracket: evx [X, Y ] = [X, Y ]x = [Xx, Yx] = [evx(X), evx(Y )] . For the proof of the surjectivity, choose an arbitrary u ∈ Lx, a bundle chart (U, ϕ) such that x ∈ U and let ρ : U → R be a smooth map with compact support contained in U and ρ(x) = 1. We define X ∈ Γ(L) ⊆ Γk(L) via Xy :=(ϕ−1 Lx 0 (ρ(y) · ϕ(u)) for y ∈ U. for y ∈ M \U. Then evx(X) = Xx = u. Remark 2.52. Let π : L → M be a bundle of Lie algebras with fiber k and (U, ϕ) a bundle chart of L and (U, ξ) a corresponding chart of M . Any directional derivative ∂ξi for i ∈ {1, . . . , dim(M )} is a derivation of the poinwise Lie algebra structure on C∞(U, k), the local forms of the sections in Γ(cid:0)π−1(U )(cid:1). This follows from the Chain rule, the fact that any bilinear map β : V × V → V for dim(V ) < ∞ is smooth and the rule T(a,b)β(v, w) = β(v, b) + β(a, w). Thus we have the equation: ∂ξi [Aϕ, Bϕ] = [∂ξiAϕ, Bϕ] + [Aϕ, ∂ξi Bϕ] for all A, B ∈ Γ(cid:0)π−1(U )(cid:1). There are certain bundles associated to a bundle of Lie algebras which will help us to analyze the structure of a Lie algebra of Ck-sections. 21 Definition 2.53. Let π : L → M be a bundle of Lie algebras with fiber k associated to an Aut(k)-principal bundle ρ : P → M . 1. Let I E k be a characteristic ideal of k, i.e. an ideal which is Aut(k)-stable. We define : L[I] → M to be the bundle associated to ρ : P → M with fiber I. It can be identified with a subbundle of π : L → M as follows: A typical element of L[I]x ⊆ L[I] takes the form [(ψ, i)] :=(cid:8)(ψf −1, f (i)) : f ∈ Aut(k)(cid:9) for a Lie algebra isomorphism ψ : k → Lx and an element i ∈ I. The element ψ(i) ∈ Lx is clearly well-defined and so we may embed L[I] into L. In the cases of I being [k, k] and z(k) 9 we also write [L, L] instead of L[[k, k]], called the commutator of L and z(L) instead of L[z(k)], called the center ∼= [Lx, Lx] and z(L)x ∼= z(Lx) via [(ψ, i)] 7→ ψ(i) for i ∈ [k, k] and i ∈ z(k), of L. Note that [L, L]x respectively. 2. Let J ⊆ End(k) be a subalgebra which is invariant under the left action Aut(k) × End(k) −→ End(k) (f, g) 7−→ f ◦ g ◦ f −1. We define : L(J) → M to be the bundle associated to ρ : P → M with fiber J. It can be identified with a subbundle of Π : Hom(L, L) → M as follows: A typical element of L(J)x ⊆ L(J) takes the form [(ψ, j)] := Aut(k) · (ψ, j) =(cid:8)(ψf −1, f jf −1) : f ∈ Aut(k)(cid:9) for a Lie algebra isomorphism ψ : k → Lx and an endomorphism j : k → k, j ∈ J. The map ψ ◦ j ◦ ψ−1 : Lx −→ Lx is clearly well-defined and, by applying this construction to all classes [(ψ, j)] for ψ ∈ Iso(k, Lx), where x runs through M and j ∈ J is fixed, we obatin an element of Hom(L, L). In the cases of J being Der(k) and Cent(k) 10 we also write Der (L) instead of L(Der(k)) and Cent(L) instead of L(Cent(k)). Note that Der(L)x ∼= Der(Lx) and Cent(L)x ∼= Cent(Lx) via [(ψ, j)] 7→ ψ ◦ j ◦ ψ−1 for j ∈ Der(k) and j ∈ Cent(k), respectively. 2.4 Lie connections We briefly repeat the definition of a covariant derivative and define the Lie connection corresponding to a fixed bundle of Lie algebras. Definition 2.54. Let π : E → M be a vector bundle with fiber V . A covariant derivative or connection of E is a linear map ∇ : Γ(T M ) −→ End(Γ(E)) X 7−→ ∇X satisfying the following conditions: 9 These ideals are characteristic: (a) i = Pj [xj, yj ] =⇒ f (i) = Pj [f xj, f yj] for all f ∈ Aut(k), i ∈ [k, k], xj, yj ∈ k. (b) (cid:2)f −1x, i(cid:3) = 0 =⇒ f (cid:2)f −1x, i(cid:3) = 0 =⇒ [x, f (i)] = 0 for all f ∈ Aut(k), i ∈ z(k), x ∈ k. 10 These algebras are Aut(k)-invariant: (a) f gf −1 [x, y] = f g((cid:2)f −1x, f −1y(cid:3)) = f ((cid:2)gf −1x, f −1y(cid:3) + (cid:2)f −1x, gf −1y(cid:3)) = (cid:2)f gf −1x, y(cid:3) + (cid:2)x, f gf −1y(cid:3) for all f ∈ Aut(k), g ∈ Der(k), x, y ∈ k. (b) (f gf −1 ◦ adx)(y) = f gf −1 [x, y] = f g((cid:2)f −1x, f −1y(cid:3)) = (f g ◦ adf −1x)(f −1y) = (f ◦ adf −1x)(gf −1y) = f (cid:2)f −1x, gf −1y(cid:3) = (cid:2)x, f gf −1y(cid:3) = (adx ◦f gf −1)(y) for all f ∈ Aut(k), g ∈ Cent(k), x, y ∈ k. 22 (a) ∇X (aA) = (X.a)A + a∇X A for all X ∈ Γ(T M ), a ∈ C∞(M, K), A ∈ Γ(E). (b) ∇(aX+bY )A = a∇X A + b∇Y A for all X, Y ∈ Γ(T M ), a, b ∈ C∞(M, K), A ∈ Γ(E). Remark 2.55. Given a point x ∈ M , a vector field X ∈ Γ(T M ) and sections A, B ∈ Γ(E) it is, by using property (b) of Definition 2.54, easy to see that ((∇X A)B)(x) depends on X only via Xx: Firstly we show that ((∇X A)B)(x) depends on X at most via XU , where U ⊆ M is any open neighbourhood of x. In fact, if ρ : M → [0, 1] is a smooth map with ρM\U ≡ 0 and ρW ≡ 1 for a smaller x-neighbourhood W ⊆ U , then (X − X ′)U ≡ 0 implies (∇X Y )(x) − (∇X ′ Y )(x) = (∇X−X ′ Y )(x) = ρ(x) · (∇X−X ′ Y )(x) = (∇ρ·(X−X ′)Y )(x) = (∇0Y )(x) = 0. Thus we can now reduce the problem to a problem in a chart (U, ξ): Let X, X ′ be two vector fields i=1 Z i∂ξi for certain functions identical in x and ρ : M → [0, 1] as above. We can write X − X ′ = Pm Z 1, . . . Z m ∈ C∞(U, R). We note that Z i(x) = 0 for all i and conclude by calculating: (∇X Y )(x) − (∇X ′ Y )(x) = (∇ρ·(X−X ′)Y )(x) =(cid:16)∇ρ·(Pm Y! (x) = = mXi=1(cid:0)ρ · Z i(cid:1) · ∇∂ξi i=1 Zi∂ξi)Y(cid:17) (x) Z i(x) ·(cid:16)∇∂ξi mXi=1 Y(cid:17) (x) = 0. Locally, a covariant derivative ist given by the so-called Christoffel symbols. Definition 2.56. Let (U, ϕ) be a bundle chart of the vector bundle π : E → M with fiber V and (U, ξ) a corresponding chart of M and let ∇ be a covariant derivative of E. Then there are unique mappings Γϕ i ∈ C∞(U, End(V )) for i ∈ {1, . . . , dim(M )} such that (∇X A)ϕ = X i (∂ξi Aϕ + Γϕ i · Aϕ) dim(M)Xi=1 for all sections A and all vector fields X with local formPm symbols of ∇ with respect to (U, ϕ). i=1 X i∂ξi. The maps Γϕ i are called Christoffel For a given bundle of Lie algebras we now define a Lie connection. Definition 2.57. Let π : L → M be a bundle of Lie algebras with fiber k and ∇ a covariant derivative of L. We call ∇ a Lie connection if for all vector fields X ∈ Γ(T M ) and for all sections A, B ∈ Γ(L) we have i.e. ∇ is a map Γ(T M ) → Der(Γ(L)). ∇X [A, B] = [∇X A, B] + [A, ∇X B] , Whether a covariant derivative of a given bundle of Lie algebras is a Lie connection, can be decided by examining its Christoffel symbols. Lemma 2.58. Let π : L → M be a bundle of Lie algebras with fiber k and ∇ a covariant derivative of L. Then ∇ is a Lie connection if and only if for all bundle charts (U, ϕ) of L and all corresponding charts (U, ξ) of M the Christoffel symbols are in in C∞(U, Der(k)). Proof. The covariant derivative ∇ is a Lie connection if and only if for all bundle charts (U, ϕ) of L, all corresponding charts (U, ξ) of M , all sections A, B ∈ Γ(cid:0)π−1(U )(cid:1) and any integer i ∈ {1, . . . , dim(M )} we have i [Aϕ, Bϕ] = [∂ξiAϕ + Γϕ i (Aϕ) , Bϕ] + [Aϕ, ∂ξi Bϕ + Γϕ ∂ξi [Aϕ, Bϕ] + Γϕ i (Bϕ)] . (5) 23 Due to Remark 2.52 we have So the equation (5) turns into ∂ξi [Aϕ, Bϕ] = [∂ξi Aϕ, Bϕ] + [Aϕ, ∂ξi Bϕ] . Γϕ i (x) [Aϕ(x), Bϕ(x)] = [Γϕ i (x) (Aϕ(x)) , Bϕ(x)] + [Aϕ(x), Γϕ i (x) (Bϕ(x))] for all x ∈ U . But the last equation is, due to Lemma 2.51, equivalent to Γϕ i ∈ C∞(U, Der(k)). Finally we want to show the existence of Lie connections. Proposition 2.59. Let π : L → M be a bundle of Lie algebras with fiber k. Then there exists a covariant derivative ∇ of L. Proof. Let π : L → M be a bundle of Lie algebras with fiber k and (Uj, ϕj)j∈J a locally finite bundle atlas of L with corresponding atlas (Uj, ξj)j∈J of M . For each j ∈ J we define a covariant derivative ∇Uj on π−1(Uj) by: X A(cid:17)ϕj (cid:16)∇Uj := dim(M)Xi=1 X i∂(ξj )i Aϕ for X ∈ Γ (T U ) and A ∈ Γ(cid:0)π−1(Uj)(cid:1). This covariant derivative is a Lie connection (cf. Lemma 2.58). After choosing a smooth partition (ρj : M → [0, 1])j∈J of 1 subordinated to the cover (Uj)j∈J , i.e. supp(ρj) is a compact subset of Uj for all j ∈ J, we define a covariant derivative ∇ of L by ∇ :=Xj∈J ρj · ∇Uj . Since ∇ is locally the finite sum of Lie connections, it is also a Lie connection. 2.5 Local operators and the Peetre Theorems Definition 2.60. Let π1 : E1 → M , π2 : E2 → M be two vector bundles and k1, k2 ∈ N. A linear operator T : Γk1 (E1) → Γk2 (E2) is called local, if for any X ∈ Γk1 (E1) and any open set U ⊆ M the condition Xx = 0 ∈ (E1)x for all x ∈ U implies (T X)x = 0 ∈ (E2)x for all x ∈ U . The meaning of an operator to be local is shown in the following lemma. Lemma 2.61. If Y, Z ∈ Γk1 (E1) are identical sections on an open set U ⊆ M and we have a local operator T : Γk1(E1) → Γk2(E2), then T Y, T Z ∈ Γk1 (E2) are identical on U . Proof. The section X := Y − Z ∈ Γk1 (E1) vanishs on U . The locality of T yields that T X = T Y − T Z also vanishs on U . So T YU = T ZU . Example 2.62. Let ∇ : Γ(T M ) −→ End(Γ(E)) be a covariant derivative on a vector bundle π : E → M with fiber V and X ∈ Γ(T M ) a vector field. Then ∇X is a local operator: Let Y ∈ Γ(E) be a section zero on an open set U ⊆ M , let x ∈ U be a point and let ρ : M → [0, 1] be a smooth map such that ρM\U ≡ 0 and ρW ≡ 1 for a smaller x-neighbourhood W ⊆ U . We easily calculate: (∇X Y )(x) = ρ(x)(∇X Y )(x) = (∇X (ρ · Y ))(x) − ((X.ρ)Y ) (x) = 0 − 0 = 0. The Peetre Theorem, proven in [Na68], p. 175-176, can be stated as follows: 24 Theorem 2.63. (Peetre Theorem, Version 1) Let M be a smooth m-dimensional manifold and πi : Ei → M a vector bundle with fiber Vi for i = 1, 2. If T : Γ (E1) → Γ (E2) is a local operator, then for any point x ∈ M there exists an open neighbourhood U ⊆ M , bundle charts (U, ϕ1), (U, ϕ2) of E1, E2, respectively, a chart (U, ξ) of M , a number n ∈ N and a familiy of functions fα ∈ C∞ (U, Hom(V1, V2)), α ∈ Nm, α ≤ n, such that for all X ∈ Γ(cid:0)π−1 1 (U )(cid:1) we have fα ·(cid:0)∂α (T X)ϕ2 = Xα≤n ξ X ϕ1(cid:1) . (6) Definition 2.64. The formula (6) says that T is a differential operator of order at most n on U . By a modification of the proof of Theorem 2.63 one gets the following result (cf. [Le79], p.52): Theorem 2.65. (Peetre Theorem, Version 2) Let M be a smooth m-dimensional manifold and πi : Ei → M a vector bundle with fiber Vi and ki ∈ N for i = 1, 2. If T : Γk1 (E1) → Γk2 (E2) is a local operator, then T is a differential operator of order at most k1 − k2. In particular, if k1 = k2 then T is a differential operator of order 0 and if k1 < k2 then T = 0. Furthermore, we obtain T = 0 if T : Γk1 (E1) → Γ(E2) ⊆ Γk1+1(E2) is local for k1 ∈ N. Remark 2.66. In the situations of the above theorems, if T : Γk(E1) → Γk(E2) is a differential operator of order 0 for k ∈ N, then it can be identified with a Ck-section of the bundle Hom(E1, E2) as follows: Fix an x ∈ M and an open neighbourhood U ⊆ M of x such that π−1 1 (U ) is trivial. For vectors v ∈ (E1)x we choose sections X : U → π−1 1 (U ) such that Xx = v and define the linear map τx : (E1)x −→ (E2)x v 7−→ (T X)x . depends on X only via Xx because T is an operator of order 0. The required section is the map M → Hom(E1, E2), x 7→ τx. Cech cohomology 2.6 Cech cohomology is one of the important cohomology theories in algebraic topology. It can be defined for any topological space and a presheaf of groups on this space. However, we will only consider it for a manifold M and a discrete group G, which can be understood as a Lie group. The following definitions are due to [tD91]. Definition 2.67. Let Cov(M ) := { V ∈ P(O(M ))S {U ∈ V} = M} denote the set of collections of open subsets of M who cover M and for V ∈ Cov(M ) we define V ∗ V := { (U, V ) ∈ V × V U ∩ V 6= ∅}. 1. A familiy of smooth functions (gV U : U ∩V → G)(U,V )∈V∗V is called a cocyle related to V ∈ Cov(M ), if it satisfies the equation gV U (x) · gUW (x) · gW V (x) = 1 1 for all x ∈ U ∩ V ∩ W for all U, V, W ∈ V. The set of these families is denoted by Z (M, V, G). 2. Two cocycles (gV U )(U,V )∈V∗V, (kV U )(U,V )∈V∗V ∈ Z 1 (M, V, G) are called cohomologous, if there exists a familiy of smooth mappings (hU : U → G)U∈V such that the relation kV U (x) = hV (x) · gV U (x) · (hU (x))−1 is satisfied for all x ∈ U ∩V for all U, V ∈ V. The relation "cohomologous" is an equivalence relation on Z 1 (M, V, G) and the set of the corresponding equivalence classes is denoted by H (M, V, G). 1 25 1 3. If G is abelian, two cocycles (gV U )(U,V )∈V∗V, (kV U )(U,V )∈V∗V ∈ Z 1 wisely and we get a group structure on Z normal subgroup of Z on the quotient Z (M, V, G), denoted by B (M, V, G) = H (M, V, G)/ B 1 (V, G) can be multiplied point- (M, V, G). The set of cocyles x 7→ hV (x) · (hU (x))−1 is a (M, V, G) and this gives rise to the group structure 1 (M, V, G). 1 1 1 4. Cov(M ) is a directed set by the refinement relation V ≺ W :⇐⇒ ∀W ∈ W ∃V ∈ V : W ⊆ V and so we can define the first Cech cohomology group of M with values in G to be11 1 H (M, G) := 1 H (M, V, G). lim−→ V∈Cov(M) In general, if G is not necessarily abelian, we obtain the first Cech cohomology set H group structure. 1 (M, G) without 1 If G is abelian and discrete, then the Cech cohomology groups H (M, G) are isomorphic to the singular cohomology groups H1(M, G) (cf. e.g [EiSt57], Chapter IX), which can be easily "calculated" in some 1 (M, G) ∼= Hom(H1(M ), G) ∼= Hom(π1(M ), G), where H1(M ) cases. There are also group isomorphies H is the first singular homology group of M with values in Z and π1(M ) is the fundamental group of M . Remark 2.68. We give some first cohomology groups which are calculated in any elementary algebraic topology lecture: 1. If G is an abelian group and M is a simply connected manifold, then H1(M, G) = 0. 2. If G is an abelian group, then H1(Sn, G) =(G if n = 1 if n > 1. 0 3. H1(CP n, Z/2Z) = H1(HP n, Z/2Z) = 0. 2.7 Structure of Lie algebras of Ck-sections From now on, facts from [Le80] are explained in detail and generalized. We have k ∈ N and a bundle of Lie algebras π : L → M with fiber k. The dimensions of M and k are denoted by m and d, respectively. Lecomte only considered the case of z(k) = 0 and we will replace this condition with weaker ones. 2.7.1 Center and commutator The center of Γk(L) is easy to calculate. Proof. If X ∈ Γk (z(L)) then, for all x ∈ M , we have Xx ∈ z(Lx) and thus [X, Y ]x = [Xx, Yx] = 0 for Proposition 2.69. For k ∈ N we have z(cid:0)Γk(L)(cid:1) = Γk (z(L)). all Y ∈ Γk(L) and x ∈ M . This proves z(cid:0)Γk(L)(cid:1) ⊇ Γk (z(L)). If X ∈ z(cid:0)Γk(L)(cid:1), x ∈ M and u ∈ Lx, Xx ∈ z(Lx). This proves z(cid:0)Γk(L)(cid:1) ⊆ Γk (z(L)). then, by Lemma 2.51, there is a section Y ∈ Γk(L) such that Yx = u, thus [Xx, u] = [X, Y ]x = 0 proving Corollary 2.70. Γk(L) is centerfree if and only if k is centerfree. 11This definition is not rigorous. If (Wj )j∈J is a refinement of (Vi)i∈I , then, for the definition of the direct limit, it is (M, G) does not necessary to have a function f : J → I such that f (j) = i implies Wj ⊆ Vi. But one can show that H depend on the choice of such functions. 1 26 For the calculation of the commutator of Γk(L) we first have to prove a technical lemma. Lemma 2.71. There exists an r ∈ N such that, for each bundle chart (U, ϕ) and each X ∈ Γk([L, L]) with compact support contained in U , there are sections Y1, Z1, . . . , Yr, Zr ∈ Γk(L) with compact supports contained in U with X = rXt=1 [Yt, Zt] . Proof. Let ([u1, w1] , . . . [ur, wr]) be a basis of [k, k]. Let F ⊆ M be compact such that supp X ⊆ F ◦ ⊆ F ⊆ U and ρ : M → [0, 1] a smooth function with compact support contained in U such that ρF ≡ 1. Then there are functions f1, . . . , fr ∈ Ck(M, K) with supports contained in F such that we have, for each x ∈ U : For t ∈ {1, . . . , r} we define Yt, Zt ∈ Γk(L) by X ϕ(x) = rXt=1 ft(x) [ut, wt] . Yt(x) = Zt(x) = 0 for x ∈ M \U Y ϕ t (x) = ft(x)ut and Z ϕ t (x) = ρ(x)wt for x ∈ U. t=1 0 =Pr For x ∈ M \F we have X(x) = 0 =Pr rXt=1 X ϕ(x) = rXt=1 [ft(x)ut, 1 · wt] = [Y ϕ t (x), Z ϕ t (x)] . t=1 [Yt(x), Zt(x)]. For x ∈ F we have t=1 [Yt, Zt]. Thus X =Pr Proposition 2.72. For k ∈ N we have(cid:2)Γk(L), Γk(L)(cid:3) = Γk ([L, L]). Proof. Let (cid:0)(Vi, ψi, ξi, ρi)i∈J , (Jt)r section X ∈ Γk(L) takes the form X = Pr t=1 Xt for Xt = Pi∈Jt setting Yp :=Pi∈Jt p ∈ Γk(L) and Zp :=Pi∈Jt p for i 6= j are disjoint, we obtain Xt =Ps (cid:2)Γk(L), Γk(L)(cid:3) ⊇ Γk([L, L]). In order to show(cid:2)Γk(L), Γk(L)(cid:3) ⊆ Γk([L, L]) we consider X =Pp Yp, Zp ∈ Γk(L). For arbitary x ∈ M we have X(x) =Pp Corollary 2.73. Γk(L) is perfect if and only if k is perfect. the supports of Y i Y i p , Z j ρiX, where t ∈ {1, . . . , r}. We fix X ∈ Γk([L, L]) and t ∈ {1, . . . , r}. Then, by Lemma 2.71, there are, for each i ∈ Jt, sections Y i 1 , Z i t=1(cid:1) be a Palais cover (cf. Definition 2.44 and Theorem 2.45). Any p(cid:3). By s ∈ Γk(L) with compact supports contained in Vi, such that ρiX =Ps p ∈ Γk(L) for each p ∈ {1, . . . , s} and recalling that t=1 Xt and p=1(cid:2)Y i p=1 [Yp, Zp], yielding X =Pr 1, . . . , Y i s=1 [Yp(x), Zp(x)] ∈ [Lx, Lx] = [L, L]x. s=1 [Yp, Zp] for appropriate sections p , Z i s , Z i Z i 2.7.2 Derivations Now we want to calculate the derivations of the Lie algebras of Ck-sections by applying Version 2 of the Peetre Theorem. Theorem 2.74. Let k ∈ N. If k is perfect or centerfree, then Der(cid:0)Γk(L)(cid:1) ∼= Γk (Der(L)) as Lie algebras. Proof. Let D be a derivation of Γk(L). We want to show that D is automatically a local operator. Let X ∈ Γk(L) be zero on an open set U ⊆ M . 27 • Suppose z(k) = 0. Then for all x ∈ U and Y ∈ Γk(L) with supp(Y ) ⊆ U we have [X, Y ] = 0 and therefore: [(DX)x, Yx] = (D [X, Y ])x − [Xx, (DY )x] = 0 − 0 = 0. By Lemma 2.51, this implies [(DX)x, v] = 0 for all v ∈ Lx, yielding (DX)x ∈ z(Lx) = 0 for all x ∈ U . • Suppose [k, k] = k. Since Γk(L) is also perfect, there exist sections Y1, Z1, . . . , Yr, Zr ∈ Γk(L) such t=1 [Yt, Zt]. Now, if x ∈ U then there is an x-neigbourhood W ⊆ U and a smooth map ρ : M → [0, 1] such that ρW ≡ 1 and ρM\U ≡ 0. We set X ′ := (1 − ρ)2 · X. Thus X = X ′ on M t=1 [(1 − ρ) · Yt, (1 − ρ) · Zt] yielding that X =Pr and we obtain X =Pr rXt=1 (DX)x = [D((1 − ρ) · Yt)(x), (1 − ρ)(x) · Zt(x)] + [(1 − ρ)(x) · Yt(x), D((1 − ρ) · Zt)(x)] = 0. In both cases we have (DX)U ≡ 0, thus D is local. By applying Theorem 2.65 and Remark 2.66 we have a linear map Ψ : Der(cid:0)Γk(L)(cid:1) → Γk (Hom(L, L)) well-defined by Ψ(D)(Xx) := (DX)x for X ∈ Γk(L), x ∈ M . Since we have Ψ(D) [Xx, Yx] = (D [X, Y ])x = ([DX, Y ])x + ([X, DY ])x = [Ψ(D)(Xx), Yx] + [Xx, Ψ(D)(Yx)] for X, Y ∈ Γk(L) and x ∈ M , the image of Ψ is contained in Γk(Der(L)). Furthermore, the map Ψ is compatible with the Lie brackets on Der(cid:0)Γk(L)(cid:1) and Γk(Der(L)): Finally, the function Φ : Γk(Der(L)) → Der(cid:0)Γk(L)(cid:1) defined by (Φ (D) (X))x := Dx(Xx) for a section D ∈ Γk(Der(L)), X ∈ Γk(L), x ∈ M , is the inverse of Ψ. We conclude that Ψ is an isomorphism of Lie algebras. Ψ [D, D′] (Xx) = ([D, D′] X)x = (D(D′X))x − (D′(DX))x = [Ψ(D), Ψ(D′)] (Xx). In order to analyze Γ(L) = Γ∞(L) we define the notion of an x-derivation of L. Definition 2.75. Let x ∈ M . 1. An x-derivation of L is a linear map δ : Γ(L) → Lx such that the following condition is satisfied for all X, Y ∈ Γ(L): δ [X, Y ] = [δ(X), Yx] + [Xx, δ(Y )] . The vector space of all x-derivations of L is denoted by Dx(L), the unionSx∈M Dx(L) is denoted by D(L) and we have a projection p : D(L) −→ M defined by p(Dx(L)) = {x} for x ∈ M . 2. Note that Der(L) can be embedded into D(L) via the natural injection i : Der(L) −→ D(L), where D ∈ Der(Lx) is mapped to the x-derivation i(D) : Γ(L) → Lx, X 7→ D(Xx). Theorem 2.76. Let k be perfect or centerfree. Then every x-derivation of L is a differential operator of order at most 1 and the triple (D(L), p, M ) admits a vector bundle structure such that Der (Γ(L)) can be naturally identified with Γ(D(L)). In addition, there exists a short exact sequence of vector bundles as follows: 0 −→ Der (L) i−→ D(L) σ−→ T M ⊗ Cent(L) −→ 0. 28 Proof. Let δ ∈ Dx(L) be an x-derivation. By Lemma 2.46, every section X ∈ Γ(L) with j1 the form x(X) = 0 takes X = f 2 i Xi (7) rXi=1 for functions fi ∈ C∞(M, K), fi(x) = 0 and Xi ∈ Γ(L). • Suppose z(k) = 0. If Y ∈ Γ(L) is an arbitrary smooth section, then: rXi=1 δ(cid:2)f 2 i Xi, Y(cid:3) −(cid:2)fi(x)2Xi(x), δ(Y )(cid:3) } {z =0  δ [fiXi, fiY ] = [δ(fiXi), fi(x)Y (x)] + [fi(x)Xi(x), δ(fiY )] [δ(X), Y (x)] = = = thus δ(X) ∈ z(Lx) = 0. i Xi(cid:1) , Y (x)(cid:3) = rXi=1 rXi=1(cid:2)δ(cid:0)f 2 rXi=1 rXi=1 0 + 0 = 0, • Suppose [k, k] = k. Since Γ(L) is also perfect, there exist, for each i ∈ {1, . . . , r}, sections Yi1, Zi1, . . . , Yis, Zis ∈ Γ(L) such that Xi =Ps δ(X) = δ [fiYit, fiZit] t=1 [Yit, Zit]. Then: rXi=1 sXt=1 rXi=1 rXi=1 i Xi(cid:1) = δ(cid:0)f 2 sXt=1 = [δ(fiYit), fi(x)Zit(x)] + [fi(x)Yit(x), δ(fiZit)] = rXi=1 sXt=1 0 = 0. For both cases, we conclude: δ(X) = 0 if j1 x(X) = 0. (8) Let (U, ϕ) be a bundle chart in x and (U, ξ) a corresponding chart of M . Since π : L → M is defined by an Aut(k)-valued cocyle, ϕ restricted to any fiber is an isomorphism of Lie algebras, therefore the Lie bracket on Γ(cid:0)π−1(U )(cid:1) corresponds to the natural pointwise Lie bracket on C∞(U, k) meaning [X, Y ]ϕ = [X ϕ, Y ϕ] for X, Y ∈ Γ(cid:0)π−1(U )(cid:1). Note, by relation (8), that δ has a local form without terms of order greater than 1: X ϕ δϕ 7−→ D (X ϕ(x)) + Si (∂ξi X ϕ(x)) (9) for certain D, S1, . . . , Sm ∈ End(k). Let us show now that a local form of the type (9) is satisfied for an x-derivation δϕ if and only if D ∈ Der(k) and S1, . . . , Sm ∈ Cent(k). Equation (9) implies the following two equations: mXi=1 δϕ [X ϕ, Y ϕ] = D ([X ϕ(x), Y ϕ(x)]) + = D ([X ϕ(x), Y ϕ(x)]) + Si (∂ξi [X ϕ, Y ϕ] (x)) Si ([∂ξiX ϕ(x), Y ϕ(x)] + [X ϕ(x), ∂ξi Y ϕ(x)]) . mXi=1 mXi=1 mXi=1(cid:2)Si (∂ξi X ϕ(x)) , Y ϕ(x)(cid:3) +(cid:2)X ϕ(x), Si (∂ξi Y ϕ(x))(cid:3) . + (10) (11) 29 [δϕ (X ϕ) , Y ϕ(x)] + [X ϕ(x), δϕ (Y ϕ)] = [D (X ϕ(x)) , Y ϕ(x)] + [X ϕ(x), D (Y ϕ(x))] By comparing the equations (10) and (11) for constant sections X, Y we obtain D ([X ϕ(x), Y ϕ(x)]) = [D (X ϕ(x)) , Y ϕ(x)] + [X ϕ(x), D (Y ϕ(x))] , thus D is a derivation of k. Knowing this we compare the equations (10) and (11) for constant X and arbitrary Y and obtain, after the cancelation of the derivation part: Si [X ϕ(x), ∂ξi Y ϕ(x)] = mXi=1 Since for any i ∈ {1, . . . , m} we have mXi=1(cid:2)X ϕ(x), Si (∂ξi Y ϕ(x))(cid:3) . ∂ξiY ϕ(x) = d dt(cid:12)(cid:12)(cid:12)(cid:12)t=0 Y ϕ(cid:0)ξ−1(ξ(x) + tei)(cid:1) = (TxY ϕ)(cid:2)(cid:0)Tξ(x)ξ−1(cid:1) (ei)(cid:3) and Tξ(x)ξ−1 is a linear bijection, there is, for any x ∈ U and v ∈ k, a smooth map Y ϕ : U → k with ∂ξiY ϕ(x) = δij v for j ∈ {1, . . . , m}. Therefore we may conclude that each Si is contained in the centroid of k. We obtain a bijective correspondence: [x∈U Dx (U × k) −→ U × (Der(k) × Cent(k)m) δϕ 7−→(cid:0)p (δϕ) ,(cid:0)D,(cid:0)S1, . . . , Sm(cid:1)(cid:1)(cid:1) . (12) By extending this construction to all charts of a maximal bundle atlas of M , we construct trivializations of D(L). In order to show that we obtain a bundle structure on D(L) we will show that the transitions are linear: Let (U, ϕ) and (V, ψ) be bundle charts of π : L → M in x with corresponding charts (U, ξ) endomorphisms of k such that for all X ϕ, X ψ ∈ C∞(U ∩ V, k) we have and (V, η) of M with ξ(x) = η(x) = 0 and let D, eD ∈ Der(k), S1, . . . , Sm,eS1, . . . ,eSm ∈ Cent(k) be the δϕ (X ϕ) = D (X ϕ(x)) + and By the definition of the local forms, we have δψ(cid:0)X ψ(cid:1) = eD(cid:0)X ψ(x)(cid:1) + Si (∂ξi X ϕ(x)) mXi=1 mXi=1 eSi(cid:0)∂ηi X ψ(x)(cid:1) . for a transition map g : U ∩ V → Aut(k) of the principal bundle to which π : L → M is associated. By considering the commutative diagram and thus ◦ δψ(cid:17)(cid:0)X ψ(cid:1) ◦ δϕ(cid:17) (X ϕ) = δ(X) =(cid:16)(cid:0)ψLx(cid:1)−1 (cid:16)(cid:0)ϕLx(cid:1)−1 (cid:0)ϕLx(cid:1)−1 (X ϕ(x)) = X(x) =(cid:0)ψLx(cid:1)−1(cid:0)X ψ(x)(cid:1) , δψ(cid:0)X ψ(cid:1) = g(x) (δϕ (X ϕ)) and X ϕ(x) = g(x)−1(cid:0)X ψ(cid:1) k = Tϕ(x)k TxX ϕ Txξ TxM / Rm <z z z z g(x) k = Tψ(x)k z z Txη z z z , TxX ψ TxM 30 o o / < o o we may calcuate ∂ξi X ϕ(x) = TxX ϕ(cid:0)T0ξ−1(ei)(cid:1) =(cid:0)g(x)−1 ◦ TxX ψ ◦ T0η−1 ◦ Txξ(cid:1)(cid:0)T0ξ−1(ei)(cid:1) =(cid:0)g(x)−1 ◦ TxX ψ(cid:1)(cid:0)T0η−1(ei)(cid:1) = g(x)−1(cid:0)∂ηiX ψ(x)(cid:1) mXi=1 eSi(cid:0)∂ηi X ψ(x)(cid:1) = δϕ (X ϕ) = g(x) (δϕ (X ϕ)) = g(x) D (X ϕ(x)) + and obtain, by the definition of D, eD, S1, . . . , Sm,eS1, . . . ,eSm: eD(cid:0)X ψ(x)(cid:1) + mXi=1 Si (∂ξi X ϕ(x))! =(cid:8)g(x) ◦ D ◦ g(x)−1(cid:9)(cid:0)X ψ(x)(cid:1) + =(cid:8)g(x) ◦ D ◦ g(x)−1(cid:9)(cid:0)X ψ(x)(cid:1) + mXi=1(cid:8)g(x) ◦ Si(cid:9) (∂ξi X ϕ(x)) mXi=1(cid:8)g(x) ◦ Si ◦ g(x)−1(cid:9)(cid:0)∂ηi X ψ(x)(cid:1) . (13) For contstant sections X this equation implies eD(cid:0)X ψ(x)(cid:1) =(cid:8)g(x) ◦ D ◦ g(x)−1(cid:9)(cid:0)X ψ(x)(cid:1) , thus eD = g(x) ◦ D ◦ g(x)−1. Knowing this, equation (13) also implies mXi=1 eSi(cid:0)∂ηiX ψ(x)(cid:1) = mXi=1(cid:8)g(x) ◦ Si ◦ g(x)−1(cid:9)(cid:0)∂ηiX ψ(x)(cid:1) Theorem 2.41, a vector bundle p : D(L) → M with fiber Der(k) × Cent(k)m. There is an isomorphism of Lie algebras Φ : Der(Γ(L)) → Γ(D(L)), defined as follows: If D ∈ Der(Γ(L)), then Φ(D) is the map and, by inserting appropriate sections for X, we may conclude that eSi = g(x) ◦ Si ◦ g(x)−1 for all i ∈ {1, . . . , m}. So the transitions D 7→ eD and Si → eSi are linear isomorphisms and we have, by M → D(L), x 7→ evx ◦D. If X ∈ Γ(D(L)), X ∈ Γ(L) and x ∈ M , then(cid:0)Φ−1(X)X(cid:1)x = (Xx) X. Pi vi ⊗ Ai corresponds to the endomorphism α 7→ Pi α(vi) · Ai. For any α ∈ TxM ∗ and w ∈ Lx let Let us define the map σ : D(L) → T M ⊗Cent(L). If δ ∈ Dx(L), then we define σ(δ) ∈ TxM ⊗End(Lx) as follows: We identify TxM ⊗ End(Lx) = Hom(TxM ∗, End(Lx)) in the natural way, i.e. the element a ∈ C∞(M, R) and A ∈ Γ(L) such that Txa = α and A(x) = w. Then σ(δ) is pointwisely defined by Now let us check that σ is well-defined and valued in T M ⊗ Cent(L) by using (8) and (12): {σ(δ)(α)} (w) := δ(aA) − a(x)δ(A). ({σ(δ)(α)} (w))ϕ = (δ(aA) − a(x)δ(A))ϕ = D(a(x)Aϕ(x)) + Si(∂ξi (aAϕ)(x)) − a(x) D(Aϕ(x)) + Si(∂ξi Aϕ(x))! mXi=1 mXi=1 = = mXi=1 mXi=1 Si (∂ξi a(x) · Aϕ(x) + a(x)∂ξi Aϕ(x) − a(x)∂ξi Aϕ(x)) Si (∂ξi a(x) · ϕ(w)) = ∂ξi a(x) · Si (ϕ(w)) . mXi=1 By leaving the local coordinates we see that {σ(δ)(α)} only depends on α and w and we also obtain σ(δ) ∈ TxM ⊗ Cent(Lx). The above calculation also shows that σ maps exactly the elements without local part in Cent(k)m to zero, i.e. im i = ker σ. 31 We finally show the surjectivity of σ : D(L) → T M ⊗ Cent(L): Let x ∈ M and (U, ϕ) a bundle chart in x with corresponding chart (U, ξ) of M and ρ : M → [0, 1] a smooth map with supp ρ ⊆ U and ρW ≡ 1 for ◦ Si ◦ ϕ ∈ Cent(Lx) an x-neighbourhood W ⊆ U and (S1, . . . , Sm) ∈ Cent(k)m. We define fi :=(cid:0)ϕLx(cid:1)−1 and δ ∈ Dx(L) is defined by δ(X) := ρ ·(cid:0)ϕLx(cid:1)−1(cid:0)Pm i=1 Si (∂ξi X ϕ(x))(cid:1) for X ∈ Γ(L). Then the local part of σ(δ) is (S1, . . . , Sm). The short exact sequence in Theorem 2.76 naturally induces another short exact sequence. Corollary 2.77. The following induced sequence is exact: 0 −→ Γ(Der (L)) i−→ Γ(D(L)) σ−→ Γ(T M ⊗ Cent(L)) −→ 0, where (i(X))(x) := i(X(x)) and (σ(X))(x) := σ(X(x)). Definition 2.78. The map σ : D(L) → T M ⊗ Cent(L) from Theorem 2.76 is called symbol map. For a global description of Der(Γ(L)) we firstly extend the notion of Lie connections. Fix a Lie connection ∇ : Γ(T M ) → Der(Γ(L)). Remark 2.79. The symbol of a derivation ∇X of Γ(L) for a vector field X ∈ Γ(T M ) is X ⊗1: Let x ∈ M be a point, a ∈ C∞(M, R) a smooth map and A ∈ Γ(L) a section of L. Then we identify ∇X ∈ Der(Γ(L)) with (y 7→ evy ◦∇X ) ∈ Γ(D(L)) and calculate: [σ (∇X (x)) (Txa)] (Ax) = (∇X (x)) (aA) − a(x) (∇X (x)) (A) = (∇X (aA)) (x) − a(x) (∇X A) (x) = ((X.a)A + a∇X (A)) (x) − (a∇X A) (x) = ((X.a)A) (x) = (Txa)(Xx) · Ax and therefore σ (∇X (x)) = Xx ⊗ 1x. Thus, by the short exact sequence of Corollary 2.77, we see that not all derivations of Γ(L) can be described like that. It is necessary to extend ∇ in an appropriate way. Definition 2.80. The extension of ∇ to Γ(T M ⊗ Cent(L)) is the linear map ∇ : Γ(T M ⊗ Cent(L)) −→ Der(Γ(L)) = Γ(D(L)) Y 7−→ ∇Y defined as follows: Let Y ∈ Γ(T M ⊗ Cent(L)) be a section and (U, ξ) a chart of M . Then there are unique elements S1, . . . , Sm ∈ Γ(Cent(π−1(U ))) such that We define ∇Y locally by YU = ∂ξi ⊗ Si. (∇Y )U := Si ◦ ∇∂ξi . (14) We show that this is well-defined: Let (U, ξ) and (V, η) be two charts of M and x ∈ U ∩ V . Then Yx takes the formPm i=1 (∂ξi )x ⊗ si =Pm j=1(cid:0)∂ηj(cid:1)x ⊗ rj for endomorphisms s1, . . . , sm, r1, . . . , rm ∈ Cent(Lx). Note that, by the uniqueness of these forms, we can write mXi=1 mXi=1 rj = Yx(Txηj). 32 By writing the matrix Tη(x)(cid:0)ξ ◦ η−1(cid:1) as A = (aij) we find the following relation between the basis vectors (∂ξi )x and(cid:0)∂ηj(cid:1)x: (cid:0)∂ηj(cid:1)x = Tx(cid:0)ξ−1 ◦ ξ(cid:1) Tη(x)η−1(ej) = Tξ(x)ξ−1(Aej) = Tξ(x)ξ−1 mXk=1 akj ek! = akj Tξ(x)ξ−1(ek) mXk=1 = akj (∂ξk )x . mXk=1 We write Tξ(x)(cid:0)η ◦ ξ−1(cid:1) = A−1 = (aij ) and calculate: Yx(Txηj ) ◦ akj∇(∂ξk )x = mXk=1 (Txηj)(cid:0)(∂ξi )x(cid:1) · si ◦ akj∇(∂ξk )x (∂ξi )x ⊗ si! (Txηj) ◦ akj∇(∂ξk )x mXj,k=1 mXi=1 mXi,j,k=1 aij · akj · si ◦ ∇(∂ξk )x = mXj=1 rj ◦ ∇(∂ηj )x = = = mXj=1 mXi,j,k=1 mXi=1 si ◦ ∇(∂ξi)x . So we have a well-defined extension. Lemma 2.81. The extension of ∇ satisfies the following properties for all Y ∈ Γ(T M ⊗ Cent(L)), a ∈ C∞(M, R), A, B ∈ Γ(L), X ∈ Γ(T M ), S ∈ Γ(Cent(L)) and x ∈ M : 1. ∇Y [A, B] = [∇Y A, B] + [A, ∇Y B], i.e. ∇Y ∈ Der(Γ(L)), 2. (∇Y (aA))x = (Yx(Txa)) {A} + (a∇Y A)x, i.e. ∇Y (aA) = (Y.a){A} + a∇Y A, i.e. σ (∇Y ) = Y, 3. ∇aY A = a∇Y A, 4. ∇X⊗SA = S ◦ ∇X A. In particular, we have ∇X⊗1 = ∇X . Proof. Let (U, ξ) be a chart of M . We show the properties by concluding locally: 1. 2. ∇Y [A, B] = = mXi=1 mXi=1 Si ◦ ∇∂ξi [A, B] = Si ◦h∇∂ξi A, Bi + mXi=1 mXi=1 Si ◦(cid:16)h∇∂ξi Si ◦hA, ∇∂ξi Bi(cid:17) A, Bi +hA, ∇∂ξi Bi = [∇Y A, B] + [A, ∇Y B] . aA(cid:17)x = Si(x) ◦ ((∂ξi a) A)x + mXi=1 mXi=1 Si(x) ◦(cid:16)a∇∂ξi A(cid:17)x = Yx(Txa)[A] + (a∇Y A)x . (∇Y (aA))x = mXi=1 Si(x) ◦(cid:16)∇∂ξi i=1 ∂ξi ⊗ Si then aY =Pm mXi=1 ∇aY = 3. If Y =Pm i=1 ∂ξi ⊗ aSi, thus: aSi ◦ ∇∂ξi = a mXi=1 Si ◦ ∇∂ξi = a∇Y . 33 4. If X =Pm i=1 X i∂ξi then implying X ⊗ S = mXi=1 X i∂ξi! ⊗ S = mXi=1 ∂ξi ⊗ X iS ∇X⊗SA = mXi=1 X iS ◦ ∂ξi A = S ◦ ∇X A. Remark 2.82. 1. It is possible to define the extension of a Lie connection ∇ : Γ(T M ) → Der(Γ(L)) = Γ(D(L)) without the help of charts of M : For x ∈ M and X ∈ Γ(T M ) the x-derivation (∇X )x depends on X only via Xx and so we can understand ∇ as a map T M → D(L). We now extend ∇ in a natural way to a map ∇ : T M ⊗ Cent(L) → D(L) by mapping Xx ⊗ Sx to Sx ◦ (∇X )x (cf. Equation (14)). 2. It is not difficult to show that any linear map L : Γ(T M ⊗ Cent(L)) → Der(Γ(L)) satisfying the four conditions of the preceding lemma is the extension of exactly one Lie connection ∇. We can now formulate and prove a theorem about the global structure of Der(Γ(L)). Theorem 2.83. Fix a Lie connection ∇ on L. If k is perfect or centerfree, then a linear function D : Γ(L) → Γ(L) is a derivation of Γ(L) if and only if there are sections Y ∈ Γ(T M ⊗ Cent(L)) and D0 ∈ Γ(Der(L)) such that D = ∇Y + i(D0) and this decomposition is unique. Proof. Obviously, for any Y ∈ Γ(T M ⊗ Cent(L)) and D0 ∈ Γ(Der(L)) the linear map ∇Y + i(D0) is a derivation of Γ(L). On the other hand, fix D ∈ Der(Γ(L)) = Γ(D(L)) and set Y := σ(D) ∈ Γ(T M ⊗ Cent(L)). Since σ (∇Y ) = Y, the linear map D − ∇Y : Γ(L) → Γ(L) has symbol 0 and thus identifies with a differential operator of order zero which is a section of Hom(L, L). Since D and ∇Y are derivations of Γ(L), this section takes its values in Der(Γ(L)). Therefore D − ∇Y can be written as i(D0) for a D0 ∈ Γ(Der(L)). The uniqueness of the decomposition follows from the fact that Y = σ(D) is the only element in Γ(T M ⊗ Cent(L)) such that σ(∇Y ) = Y. 2.7.3 Centroid In this subsection we will calculate the centroid of Γk(L). We begin with a technical lemma. Lemma 2.84. Let X ∈ Γ(L) be a section which is zero on an open set U ⊆ M and x ∈ U . Then there exists a compact x-neigbourhood F and there are functions Dt ∈ Der(Γ(L)), Xt ∈ Γ(L) with XtF ≡ 0 for t ∈ {1, . . . , r} such that X = rXt=1 Dt(Xt). Proof. Let F be a compact x-neigbourhood contained in U . Since M is paracompact, there exists a Palais cover(cid:0)(Vi, ψi, ξi, ρi)i∈J , (Jt)r t=1(cid:1) refining the open cover {U, M \F } of M (cf. Definition 2.44 and 34 Theorem 2.45). For any i ∈ J and Y ∈ Γ(cid:0)π−1(Vi)(cid:1) recall that Y ψi : Rm ⊇ ξi(Vi) → k denotes the mapping such that the following diagram commutes: L ⊇ π−1(Vi) (ξi◦π,ψi) Y M ⊇ Vi ξi / ξi(Vi) × k (id,Y ψi) ξi(Vi). The Lie bracket on Γ(cid:0)π−1(Vi)(cid:1) corresponds to the natural pointwise Lie bracket on C∞(ξi(Vi), k), i.e. [Y, Z]ψi = [Y ψi , Z ψi] for Y, Z ∈ Γ(cid:0)π−1(Vi)(cid:1), for all i ∈ J. We set Y := ρiX and choose a section X i ∈ Γ(cid:0)π−1(Vi)(cid:1) by claiming ∂(ξi)1 (X i)ψi = Y ψi. With Di : Γ(cid:0)π−1(Vi)(cid:1) → Γ(cid:0)π−1(Vi)(cid:1) defined by (Di)ψi := ∂(ξi)1 one obtains ρiX = Y = Di(X i). Note that in the case of supp(ρi) ⊆ U we can choose X i ≡ 0 because XU ≡ 0 and each Vi is a subset X i we obtain a well-defined derivation Dt and a well-defined section Xt for any t ∈ {1, . . . , r} and of exactly one, U or M \F . By setting Dt := Pi∈Jt Di Xi∈Jt rXt=1 Xi∈Jt rXt=1 Dt(Xt) = Di and Xt := Pi∈Jt rXt=1 Xi∈Jt X i!! = ρiX! = X. Theorem 2.85. Any endomorphism T : Γ(L) → Γ(L) with T ◦ Der(Γ(L)) ⊆ Der(Γ(L)) is a differential operator of order zero and can be identified with a smooth section of Hom(L, L). Proof. We want to show that T is local. Let X ∈ Γ(L) be zero on an open set U ⊆ M . Lemma 2.84 yields that there exist D1, . . . , Dr ∈ Der(Γ(L)) and X1, . . . , Xr ∈ Γ(L), each Xt zero on an x-neighbouhood F , such that X =Pr t=1 Dt(Xt). We obtain T XF = ((T ◦ Dt)(Xt))F = 0 = 0 (15) rXt=1 rXt=1 because any D ∈ Der(Γ(L)) can be identified with a section of D(L) by D 7→ XD, where XD(x) := evx ◦D, and hence is local. By (15) we conclude that T is local. Theorem 2.63 yields that T is a differential operator. By Theorem 2.76, any derivation D ∈ Der(Γ(L)) is a differential operator of order at most one and so is any T ◦ D. If T were of order n > 0, then there would be a bundle chart (U, ϕ) with corresponding chart (U, ξ) of M and functions fα ∈ C∞ (U, End(k)), α ∈ Nm, α ≤ n, such that for all X ∈ Γ(cid:0)π−1(U )(cid:1) we would have ξ X ϕ(cid:1) (T X)ϕ = Xα≤n fα ·(cid:0)∂α and there would be x′ ∈ U , u ∈ k and α′ = (α′ may assume, without loss of generality, that ξ(x′) = 0. By defining D ∈ Der(cid:0)Γ(cid:0)π−1(U )(cid:1)(cid:1) by Dϕ := ∂ξ1 and X ∈ Γ(cid:0)π−1(U )(cid:1) by X ϕ(x) := ξ1(x) ·Qm fα ·(cid:0)∂α (T (DX))ϕ = Xα≤n ξ (DX)ϕ(cid:1) = Xα≤n i · u we would obtain i=1 ξi(x)α′ m) ∈ Nm with α′ = n such that fα′ (x′)(u) 6= 0. We 1, . . . , α′ =0 in x′, if α6=α′ =⇒ (T (DX))ϕ(x′) = fα′ (x′) (α′! · u) 6= 0, fα · (cid:0)∂α ξ ∂ξ1 X ϕ(cid:1) } {z contradicting to the facts that T ◦ D is a differential operator of order at most one and j1 T is of order zero. x′(X) = 0. So 35 / / / O O O O Theorem 2.86. If k is perfect or centerfree, then Cent(cid:0)Γk(L)(cid:1) ∼= Γk (Cent(L)) as associative algebras. Proof. In the smooth case we use Lemma 2.18.2 and Theorem 2.85 to see that all T ∈ Cent(Γ(L)) identify with smooth sections of Hom(L, L). For k ∈ N we will firstly show that any A ∈ Cent(Γk(L)) is local: Let X ∈ Γk(L) be zero on an open set U ⊆ M . • Suppose z(k) = 0. Then for all x ∈ U and Y ∈ Γk(L) we have: [(AX)x, Yx] = [Xx, (AY )x] = 0. This implies (AX)x ∈ z(Lx) = 0 for all x ∈ U . • Suppose [k, k] = k. Then, by Corollary 2.73, the Lie algebra Γk(L) is also perfect. So there exist t=1 [Yt, Zt]. Now, if x ∈ U then there is an x-neigbourhood W ⊆ U and a smooth map ρ : M → [0, 1] such that ρW ≡ 1 and ρM\U ≡ 0. We sections Y1, Z1, . . . , Yr, Zr ∈ Γk(L) such that X = Pr set X ′ := (1 − ρ) · X, thus X = X ′ on M and we obtain X =Pr t=1 [Yt, (1 − ρ) · Zt], yielding (AX)x = [(AYt)(x), (1 − ρ)(x) · Zt(x)] = 0. rXt=1 In both cases we have (AX)U ≡ 0, thus A is local. So we can use Theorem 2.65 to see that all T ∈ Cent(Γk(L)) identify with Ck-sections of Hom(L, L). We know now that, for k ∈ N, there is a linear map Ψ : Cent(cid:0)Γk(L)(cid:1) → Γk (Hom(L, L)) well-defined by Ψ(A)(Xx) := (AX)x for X ∈ Γk(L), x ∈ M . Since we have Ψ(A) [Xx, Yx] = (A [X, Y ])x = ([AX, Y ])x = [Ψ(A)(Xx), Yx] for X, Y ∈ Γk(L) and x ∈ M , the image of Ψ is contained in Γk(Cent(L)). Furthermore, the map Ψ is compatible with the composition of functions on Cent(cid:0)Γk(L)(cid:1) and Γk(Cent(L)): Finally, the function Φ : Γk(Cent(L)) → Cent(cid:0)Γk(L)(cid:1) defined by (Φ (A) (X))x := Ax(Xx) for a section A ∈ Γk(Cent(L)), X ∈ Γk(L), x ∈ M , is the inverse of Ψ. We conclude that Ψ is an isomorphism of associative algebras. Ψ(A ◦ A′)(Xx) = ((A ◦ A′)X)x = (A(A′X))x = (Ψ(A) ◦ Ψ(A′))(Xx). We can now deduce an important structure theorem. Theorem 2.87. Let k be indecomposable. Then Γk(L) is so, if Hom(k/ [k, k] , z(k)) = 0. Proof. Let Γk(L) = I1 ⊕ I2 be a decomposition into a direct sum of ideals and P 1 be the projection onto I1 parallel to I2. Note that P 1 ∈ Cent(Γk(L)) = Γk(Cent(L)) because for any X, Y ∈ Γk(L) decomposing into X1, Y1 ∈ I1 and X2, Y2 ∈ I2 we have: P 1 [X, Y ] = P 1([X1, Y1] + [X2, Y2]) = [X1, Y1] = [X1 + X2, Y1] =(cid:2)X, P 1Y(cid:3) . Since P 1 ◦ P 1 = P 1 on Γk(L), each P 1 indecomposability of Lx ∼= k, this implies P 1 x 7→ P 1 x 7−→ ϕi ◦ P 1 So Γk(L) is indecomposable. x ∈ Cent(Lx), where x ∈ M , is a projection in Cent(Lx) and, by x = 1x. The map M → Cent(Lx) ⊆ Cent(L), x is continuous and for each bundle atlas (Ui, ϕi)i∈I , where each Ui is connected, the mapping x has discrete image. But all this is only possible if P 1 = 0 or P 1 = 1, i.e. I1 = 0 or I2 = 0. x = 0 or P 1 (16) 36 Remark 2.88. If Hom(k/ [k, k] , z(k)) = 0, then, by Lemma 2.18.3, we have Cent(k) = N(k) ⊕ S(k) and, since Aut(k) acts on N(k) and S(k) by conjugation, we also have the following decomposition: Γk(Cent(L)) = Γk(N(L)) ⊕ Γk(S(L)), where N(L), S(L) denote the corresponding subbundles L(N(k)), L(S(k)) of Cent(L) = L(Cent(k)) (cf. Definition 2.53.2), respectively. Consider the case S(k) = K · 1. Then S(L) is a bundle over M with fiber K · 1 and, for any x ∈ M , the fiber S(L)x is naturally isomorphic to S(Lx). Thus any section X ∈ Γk (S(L)) takes the form M → S(L), x 7→ Xx = f (x)1x for some f ∈ Ck(M, K), hence: Γk(S(L)) = Ck(M, K) · 1. (17) k = Ln H(k) := Tn ing to the decomposition k =Ln Remark 2.89. Let k be reductive and dim z(g) ≤ 1, e.g., k = gln(K) or Hom(k/ [k, k] , z(k)) = 0. If i=1 ki is a decomposition into a direct sum of indecomposable non-zero ideals, then we define i=1 { f ∈ Aut(k) f (ki) = ki}, the normal subgroup of Aut(k) stabilizing this decomposition. This is well-defined because the decomposition into a direct sum of indecomposable non-zero ideals is, by Proposition 2.25, unique except for the order. We denote the quotient group Aut(k)/ H(k) by G(k). Let k·k : End(k) → R+ be the Frobenius norm on End(k) with respect to a basis (v1, . . . , vd) correspond- ij, where f (vj) = aij vi for i, j ∈ {1, . . . , d}. This norm induces, by dim k = d < ∞, the compact-open topology. So the subspace Aut(k) ⊆ End(k) becomes a topological group. i=1 ki, i.e. kf k :=qPd i,j=1 a2 By Remark 2.27, the matrices corresponding to the automorphisms in Aut(k) are "permutated block diagonal matrices". Therefore, if f ∈ H(k) with f (vj) = aij vi for i, j ∈ {1, . . . , d} and we set ε := 0.5 · mini,j∈{1,...,d} aij , then the ε-ball around f in Aut(k) with respect to k·k is entirely contained in H(k). So H(k) ⊆ Aut(k) is open and G(k) equipped with the corresponding quotient topology is discrete. The following general theorem about first Cech cohomology sets is Theorem V.5 of [Ne04]. We will use it to show Lemma 2.91 which will be useful for two structure theorems. Theorem 2.90. If M is a smooth manifold and q : A → G is a smooth and surjective morphism of Lie groups with kernel H, then there exists an exact sequence of morphisms of pointed sets as follows: 1 1 −→ C∞(M, H) −→ C∞(M, A) −→ C∞(M, G) −→ H 1 (M, H) −→ H (M, A) −→ H 1 (M, G). 1 Lemma 2.91. Let A be a Lie group, H E A open and G := A/H such that H 1 H (M, A) ∼= H (M, H). 1 (M, G) = 1. Then Proof. Since H 1 1 (M, G) = 1, the map H 1 (M, H) → H (M, A) is, by Theorem 2.90, surjective. of a perfect or centerfree Lie algebra k and let H Theorem 2.92. Let k =Ln there is a decomposition into a direct sum of indecomposable non-zero ideals Γk(L) =Ln i=1 ki be a decomposition into a direct sum of indecomposable non-zero ideals (M, G(k)) be trivial (e.g. M simply connected). Then i=1 Γk(Li), where each πLi : Li → M is a subbundle of π : L → M with fiber ki and this decomposition is unique except for the order. 1 1 Proof. Lemma 2.91 yields H H(k) and we can construct the subbundles πLi : Li → M with fiber ki. We also have 1 (M, Aut(k)) ∼= H (M, H(k)), so the bundle π : L → M admits a cocycle in Γk(L) = Γk(Li), nMi=1 which is a decomposition into a direct sum of indecomposable non-zero ideals because each ki is inde- composable (cf. Theorem 2.87). We conclude by showing the uniqueness of this decomposition (except for the order). 37 Let Γk(L) = I1 ⊕ I2 be a decomposition into a direct sum of ideals and P 1 be the projection onto I1 parallel to I2 and for i ∈ {1, . . . , n} let Pi be the projection onto Γk(Li) associated to the decom- i=1 Γk(Li). Calculations totally analogous to (16) show that P 1 and each Pi are position Γk(L) = Ln in Cent(Γk(L)) = Γk(Cent(L)). For all x ∈ M we have (cid:2)Pi, P 1(cid:3) (x) = (cid:2)Pi(x), P 1(x)(cid:3) = 0 because any i ∈ {1, . . . , n}. This implies P 1(cid:0)Γk(Li)(cid:1) = Pi(I1) and, due to the indecomposability of the ideals, Cent(Lx) is abelian by Hom(k/ [k, k] , z(k)) = 0 and Lemma 2.18.3. Therefore P 1 commutes with Pi for P 1 Γk(Li) = 0 or P 1 Γk(Li) = 1 for any i ∈ {1, . . . , n}, resulting in I1 = tMℓ=1 Γk(cid:0)Ljℓ(cid:1) for some 1 ≤ j1 < . . . < jt ≤ n. Thus the decomposition Γk(L) =Ln A statement concerning complex structures of real Lie algebras of Ck-sections is given by the following i=1 Γk(Li) is unique. proposition. Proposition 2.93. Let K = R and k be indecomposable with Hom(k/ [k, k] , z(k)) = 0. Then Γk(L) admits 1 (M, Z/2Z) ∼= Hom(π1(M ), Z/2Z) is trivial (e.g. M simply at most two complex structures and, if H connected), then Γk(L) admits a complex structure if and only if k does. Proof. A complex structure J ∈ Cent(cid:0)Γk(L)(cid:1) = Γk(Cent(L)) induces a complex structure Jx on each fiber Cent(Lx), thus on k. Lemma 2.28.2 implies that k possesses at most two complex structures, thus, by the connectedness of M , there can be at most two different Ck-complex structures, namely J and −J, on Γk(L). Let J0 be a complex structure on k. We turn k into a complex Lie algebra by defining the multiplication by a complex scalar via (a + bi) · x := ax + bJ0(x) for a, b ∈ R and x ∈ k. If σ ∈ AutR (k), a, b ∈ R and x ∈ k, then σ((a + bi) · x) = σ(ax + bJ0(x)) = aσ(x) + bσ(J0(x)). (cid:0)σ ◦ J0 ◦ σ−1(cid:1)2 Thus σ ∈ AutR (k) is in AutC (k) if and only if σ◦J0 = J0◦σ. Otherwise we have σ◦J0 ◦σ−1 = −J0 because = −1 and J0, −J0 are the only complex structures on k by Lemma 2.28.2. So AutC (k) is 1 a subgroup of index 2 of AutR (k), thus normal and AutR (k) / AutC (k) ∼= Z/2Z. Since H (M, Z/2Z) = 1, 1 we have H (M, AutC (k)) by Lemma 2.91, yielding the existence of an AutC (k)-valued cocyle of L and this turns π : L → M into a bundle of Lie algebras with complex fiber k and Γk(L) into a complex Lie algebra. 1 (M, AutR(k)) ∼= H Conversely, if Γk(L) admits a complex structure, then, by Theorem 2.86, this induces a complex structure on each fiber k, thus on k. 2.7.4 Isomorphisms Now we will discuss the isomorphisms of Lie algebras of Ck-sections. In the light of Theorem 2.92 it suffices to do this for indecomposable Lie algebras only. The main key to the analysis of the isomorphisms is Lemma 2.96. We want to show some lemmas before. In this subsection M, N are smooth manifolds with positive dimensions m, n, respectively Lemma 2.94. For k ∈ N the topology on M is the same as the initial topology of the maps in Ck(M, K). Proof. Let OM denote the topology on M . The initial topology of the maps in Ck(M, K), denoted by Oi, is the coarsest topology on M such that each f ∈ Ck(M, K) is continuous, so that Oi ⊆ OM . Let V be a neighbourhood of x ∈ M with respect to OM . Then, by paracompactness of M , there is a smooth function ρ : M → [0, 1] ⊆ K such that ρM\V ≡ 0 and ρW ≡ 1 for a smaller x-neighbourhood W ⊆ V . 38 But this yields W ⊆ ρ−1(B(0.5; 1)) ⊆ V , where B(r; z) denotes the open ball with radius r around z in K. Thus V is also a neighbourhood of x ∈ M with respect to Oi and we conclude OM ⊆ Oi, so Oi = OM . Lemma 2.95. There exists a non-constant proper map ϕ ∈ C∞(M, K), i.e. ϕ-preimages of compact subspaces are compact. Proof. Let U ( M be a non-empty open subset with compact closure, K ⊆ U a non-empty compact subset and ϕ ∈ C∞(M, [0, 1]) ⊆ C∞(M, K) a map such that supp ρ ⊆ U and ρK ≡ 1. For any compact C ⊆ K, the set ϕ−1(C) is a closed subset of U , hence compact. Lemma 2.96. Let v : Ck(M, K) → Cℓ(N, K) be an isomorphism of associative algebras for some 0 6= k, ℓ ∈ N. Then k = ℓ and there is a Ck-diffeomorphism λ : M → N such that v(f ) = f ◦ λ−1 for all f ∈ Ck(M, K). Proof. We set A := AM := Ck(M, K) and AU := AM Furthermore, for x0 ∈ M we write Nx0 := N M V ⊆ N and N N y0 for y0 ∈ N are understood in the obvious analogous manner. x0 := { f ∈ A f (x0) = 0}. The symbols AN , AN U := { f ∈ A supp f ⊆ U } for open U ⊆ M . V for open Since any Nx0 obviously is a vector subspace of A and A · Nx0 ⊆ Nx0, it is even an ideal of A. Its codimension12 is 1 because, evidently, we have Nx0 ⊕K·1 = A. We will show that in fact every ideal I E A of codimension 1 takes the form Nx for a some x ∈ M .13 For this, it suffices to show I ⊆ Nx for some x ∈ M because all Nx are ideals of codimension 1. Since, for x0 6= x1, we have (Nx0 ∩ Nx1 ) ⊕ K · 1 6= A, the point x ∈ M corresponding to the ideal I is uniquely determined. and f = f0 · f f0 ∈ I because f0 ∈ I and I is an ideal of A. Thus AU ⊆ I. Assume there exists an ideal I E A of codimension 1 such that for all x ∈ M we have I 6⊆ Nx, i.e., there exists f0 ∈ I with f0(x) 6= 0. By continuity, there is even an admissible14 open x-neighbourhood U ⊆ M with f0(x′) 6= 0 for all x′ ∈ U . Note that M can, in our case, be covered by admissible open sets, is a well-defined function in A = Ck(M, K) say M =Sγ∈Γ Uγ. If U ⊆ M is open and f ∈ AU then f and dim(cid:0)K · ϕ ⊕ K · ϕ2(cid:1) = 2, there are a, b ∈ K such that aϕ + bϕ2 =: f1 ∈ I is not zero and f −1 ({0}) = (bϕ)−1 ({0}) ∪ ϕ−1(cid:0)(cid:8)− a Now let ϕ ∈ A be a non-locally constant proper map, which exists by Lemma 2.95. Since codimAI = 1 ({0}) ({0}) = (aϕ)−1 ({0}) and aϕ is proper if b = 0. We define an open set U1 := f −1 is compact because f −1 f −1 1 (K\ {0}). The 1 compact subset f −1 ({0}) ⊆ M is covered by finitely many admissible open sets U2, U3, . . . , Ur ∈ (Uγ)γ∈Γ. Let (Vβ)β∈B be a locally finite refinement of (U1, . . . , Ur) such that there is a smooth partition of unity i=1 Bi and let β ∈ Bi for some i ∈ {1, . . . , r} imply Vβ ⊆ Ui. Then, for any i ∈ {1, . . . r} and β ∈ Bi, we have ρ · A ⊆ AUi . We obtain (ρβ : M → [0, 1])β∈B subordinate to (Vβ )β∈B. Let B1, . . . , Br ⊆ B be subsets such that B = Sr b(cid:9)(cid:1) and bϕ is proper in the case of b 6= 0 and f0 1 1 1 Xβ∈Bi ρβ · A ⊆ AUi ⊆ I, rXi=1 A = 1 · A = Xβ∈B (ρβ · A) ⊆ rXi=1 but this contradicts the fact that I has codimension 1. This shows that for any ideal I E A with codimension 1 there exists x ∈ M such that I = Nx. Since v is an isomorphism of associative algebras, v(Nx) is, for any x ∈ M , an ideal of AN with for some unique y ∈ N . We write y =: λ(x) and obtain a mapping codimension 1, thus equal to N N y λ : M → N . A dual argumentation with v−1 instead of v leads to a mapping eλ : N → M such that 12The "codimension" of an ideal I E A = AM or J E AN is always meant to be the vector space dimension of the quotient algebra A/I or AN /J, respectively. 13The proof will also show the analogous statemanet for the ideals J E AN with codimension 1. 14We call an open set U admissible, if there is a function f ∈ I such that f (y) 6= 0 for all y ∈ U . 39 f(cid:0)λ−1(y)(cid:1) = 0 λ ◦eλ = idN and eλ ◦ λ = idM , so λ is bijective. Note that we can perform the following calculation for f ∈ A and y ∈ N : =⇒ f ∈ Nλ−1(y) =⇒ v(f ) ∈ N N y =⇒ v(f )(y) = 0. But this already implies that f ◦ λ−1 = v(f ) for any f ∈ A because we may replace the mapping f by f − r · 1 for arbitrary r ∈ K in the above calculation. We will now show that λ is a Ck-diffeomorphism. By Lemma 2.94, the topologies on M and N can be described as the initial topologies of the maps in Ck(M, K) and Cℓ(N, K), respectively. Since v−1(g) = g ◦ λ ∈ Ck(M, K) for all g ∈ Cℓ(N, K), the map λ : M → N is continuous. Let (U ′ j)j∈J a locally finite atlas of N . We modify the chart maps by multiplying them with the maps ρi : M → [0, 1] and j : N → [0, 1], respectively, of smooth partitions of unity subordinate to the atlases. Then ρiUi ≡ 1 and j still covering the whole manifold. We obtain new atlases, j Vj denoted by (Ui, ϕi)i∈I and (Vj , ψj)j∈J , where each chart map is defined on the whole manifold. Now let t (ψj(Vj ∩ λ(Ui))) ⊆ R. t ∈ {1, . . . , n}, i ∈ I, j ∈ J and define A := ϕi(cid:0)Ui ∩ λ−1(Vi)(cid:1) ⊆ Rm and D := e∗ i)i∈I be a locally finite atlas of M and (V ′ t ◦ ψj ◦ λ ◦ (ϕi)−1 : A → D is equal to ≡ 1 for open sets Ui ⊆ U ′ i and Vj ⊆ V ′ The map e∗ j , ψ′ i , ϕ′ thus, since t,i and j were arbirarily chosen, λ : M → N is Ck. v−1 (e∗ t ◦ ψj) ◦ (ϕi)−1 ∈ Ck (A, D) , ∈Ck(M,R) {z } By the symmetry of the arguments, λ−1 is a Cℓ-map and, by the Inverse Mapping Theorem, even a max(k, ℓ)-times continuously differentiable diffeomorphism. Since the composition with λ turns Cℓ(N, K)- maps into Ck(M, K)-maps and λ−1 turns Ck(M, K)-maps into Cℓ(N, K)-maps, so Ck(M, K) = Cℓ(M, K) and Ck(N, K) = Cℓ(N, K), thus k = ℓ, completing the proof. The following Lemma will be used in Theorem 2.98. Lemma 2.97. 1. Let V be a vector space of finite dimension, (U, ξ) a chart of M and T : U → End(V ), f : U → V smooth functions. Then, for any multi-index α ∈ Nm, we have: ξ T · ∂α−γ ξ f(cid:17) . ∂α ξ (T · f ) =Xγ≤α(cid:18)α Xγ≤α γ(cid:19)(cid:16)∂γ =(1, if α = 0 (−1)α−γ γ!(α − γ)! 0 otherwise = δα,0 α! . (18) (19) 2. We have, for α ∈ Nm: Proof. 1. The proof is done by mathematical induction over α. The claim is trivially true for α = 0. For α = 1 there exists a canonical basis vector ei of Rm such that α = ei and ∂α ξ = ∂ξi . Then ∂ξi (T · f ) = ∂ξi T · f + T · ∂ξif and the claim is also true. Now consider the claim shown for all multi-indices β ∈ Nm with β = n > 0 and fix α ∈ Nm with α = n + 1. There exists a canonical basis vector ei of Rm such that β := α − ei ∈ Nm and β = n. By induction hypothesis, we have ∂β ξ (T · f ) =Xγ≤β(cid:18)β γ(cid:19) ∂γ ξ T · ∂β−γ ξ f, 40 thus ∂α T · ∂β−γ ξ ξ ξ ξ ξ T · ∂β−γ ξ T · ∂β−γ ξ (T · f ) =Xγ≤β(cid:18)β γ(cid:19) ∂ξi(cid:16)∂γ =Xγ≤β(cid:18)β γ(cid:19) ∂γ+ei = Xei≤δ≤α(cid:18)α − ei δ − ei(cid:19) ∂δ ξ f + Xei≤δ≤α−ei(cid:18)(cid:18)α − ei δ(cid:19)(cid:16)∂δ f(cid:17) =Xγ≤β(cid:18)β γ(cid:19)(cid:16)∂γ+ei f +Xγ≤β(cid:18)β γ(cid:19) ∂γ f + Xγ≤α−ei(cid:18)α − ei γ (cid:19) ∂γ δ (cid:19)(cid:19) ∂δ δ − ei(cid:19) +(cid:18)α − ei =Xδ≤α(cid:18)α ξ T · ∂β−γ+ei f(cid:17) . ξ T · ∂α−δ ξ T · ∂α−δ ξ T · ∂α = 1 · ∂0 f ξ ξ ξ f + ∂γ ξ T · ∂β−γ+ei ξ f(cid:17) ξ T · ∂α−γ ξ f ξ T · ∂α−δ ξ f + 1 · ∂α ξ T · ∂0 ξ f So the claim is true for all multi-indices α ∈ Nm. 2. For a given multi-index α ∈ Nm we define a mapping Fα : Rm → R, x 7→Qm . By the Taylor Formula and the fact that ∂γF ≡ 0 for each γ ∈ Nm with γi > αi for an index i ∈ {1, . . . , m}, we obtain: i=1 xαi i γ! 1 γ! i=1(−yi)αi · ∂γFα(x) + 0 Fα(x − y) =Xγ≤αQm (−yi)αi! · α! · mYi=1 =Xγ≤α (−1)αi! · 1 · mYi=1 mYj=1 =Xγ≤α (α − γ)! 1 γ! (α − γ)! j xαj −γj  . mYj=1 1αj−γj =Xγ≤α δα,0 α! = Fα(0) α! This yields, by setting x = y = (1, . . . , 1) ∈ Rm: (−1)α−γ γ!(α − γ)! . Theorem 2.98. Let π : L → M and : E → N be two bundles of Lie algebras with fiber k and g, respectively, dim k = d, dim g = e, Hom(k/ [k, k] , z(k)) = 0, Hom(g/ [g, g] , z(g)) = 0 and S(k) = K · 1, S(g) = K · 1, so that both Lie algebras are indecomposable by Lemma 2.28. Suppose there exists an isomorphism of Lie algebras µ : Γk(L) → Γℓ(E) for some 0 6= k, ℓ ∈ N. Then k = ℓ and (a) if k, ℓ ∈ N, then µ is induced in a natural way by a Ck-isomorphism of vector bundles κ : L → E, i.e., if κ′ : M → N is the map underlying to κ (that means κ′ ◦ π = ◦ κ), then for all X ∈ Γk(L) we have µ(X) = κ ◦ X ◦ (κ′)−1. In particular, the manifolds M, N are Ck-diffeomorphic and the Lie algebras k, g are isomorphic. (b) if k, ℓ = ∞, then the manifolds M, N are diffeomorphic and the Lie algebras k, g are isomorphic. After identifying the manifolds and the Lie algebras, µ turns into a linear differential operator of order at most e − 1 taking the following local form on a bundle chart (U, ϕ) of π : L → M with corresponding chart (U, ξ) of M : Aϕ µϕ 7−→ Xα<d 1 α! N α ·(cid:0)µ0 · ∂α ξ Aϕ(cid:1) , (20) where µ0 is a smooth function U → Aut(g) and we use the notation N α = N α1 multi-indices α = (α1, . . . , αm) ∈ Nm and smooth functions N1, . . . , Nm ∈ C∞(U, N(g)). 1 ◦ . . . ◦ N αm m for 41 Proof. The isomorphism of Lie algebras µ : Γk(L) → Γℓ(E) induces an isomorphism of associative algebras application of (17), we see that there is a function v(f ) ∈ Cℓ(N, K) such that we have 2.86, with an isomorphism Γk(Cent(L)) → Γℓ(Cent(E)) and, by applying Lemma 2.18.3 to the fibers of eµ : Cent(cid:0)Γk(L)(cid:1) → Cent(cid:0)Γℓ(E)(cid:1) byeµ(T ) := µ◦T ◦µ−1 for T ∈ Cent(cid:0)Γk(L)(cid:1). This identifies, by Theorem Cent(E), we deduce that eµ induces an isomorphism of associative algebras v : Ck(M, K) → Cℓ(N, K) as follows: By (17), an arbitary element of S(cid:0)Γk(L)(cid:1) takes the form f · 1, where f ∈ Ck(M, K). By another where Nf ∈ N(cid:0)Γℓ(E)(cid:1) = Γℓ(N(E)) is nilpotent and this decomposition is unique. Note that for constant functions f ≡ c, the map eµ(f · 1) = c ·eµ(1) = c · 1 is semisimple and thus, by uniqueness of the decomposition (21), we have Nf = 0 for constant functions f . The morphism property of v is shown by the following calculations: eµ(f · 1) = v(f ) · 1 + Nf , (21) and eµ(f · 1) ·eµ(g · 1) =eµ((f · 1) · (g · 1)) =eµ(f g · 1) = v(f g) · 1 + Nf g {z eµ(f · 1) ·eµ(g · 1) = (v(f ) · 1 + Nf ) · (v(g) · 1 + Ng) = v(f )v(g) · 1 + v(f ) · Ng + v(g) · Nf + Nf Ng } Furthermore, the Binomial Theorem yields, for f ∈ Ck(M, K) and r ∈ N: nilpotent . Nf r = t(cid:19) v(f )r−t (Nf )t . S(Γk(E)) are injective. By applying Lemma 2.96 to v : Ck(M, K) → Cℓ(N, K), we obtain k = ℓ and the existence of a Ck-diffeomorphism λ : M → N such that v(f ) = f ◦ λ−1 for all f ∈ Ck(M, K). We now identifiy the manifolds M, N via λ and the associative algebras Ck(M, K), Ck(N, K) via v, so that we have the isomorphisms µ : Γk(L) → Γk(E) and rXt=1(cid:18)r Note that v is bijective because eµ S(Γk(L)) and eµ−1 eµ : Γk(Cent(L)) = Cent(cid:0)Γk(L)(cid:1) −→ Γk(Cent(E)) = Cent(cid:0)Γk(E)(cid:1). For all sections A ∈ Γk(L), all points (µ (f eA))x = (µ ((f e · 1) (A)))x = ((eµ (f e · 1)) (µ(A)))x = ((f e · 1 + Nf e) (µ(A)))x t(cid:19) f (x)e−t (Nf )t x ∈ M and all mappings f ∈ Ck(M, K) with f (x) = 0, we calculate: = (f eµ(A))x + (Nf e(µ(A)))x = f (x)e (µ(A))x + x (µ(A))x = 0 + 0 = 0. eXt=1(cid:18)e (22) Suppose a section A ∈ Γk(L) be zero on an open set U ⊆ M , x ∈ U and let ρ : M → [0, 1] be a smooth function and W ⊆ U a smaller x-neighbourhood such that ρM\U ≡ 0 and ρW ≡ 1. Then A = (1 − ρ)eA and (22) shows (µ(A))x = 0. Since x was arbitrarily chosen, µ(A) is also zero on U . So µ : Γk(L) → Γk(E) is local and, by the Peetre Theorems 2.63 and 2.65, a differential operator. If k ∈ N, then Theorem 2.65 even implies that µ is a differential operator of order 0 inducing a bundle isomorphism κ : L → E (cf. Definition 2.33 and Remark 2.66). This proves (a). Now assume k = ∞. Let A ∈ Γ(L) be a section with je−1 x (A) = 0. By Lemma 2.46, the section A locally takes the form for functions fi ∈ C∞(M, K), fi(x) = 0 and Ai ∈ Γ(L). By using relation (22), we see that A = f e i Ai rXi=1 (µ(A))x = rXi=1 [µ (f e i Ai)]x = 0 = 0. rXi=1 42 This proves that the order of the differential operator µ is at most e−1 because je−1 (A) = 0 already implies (µ(A))x = 0. Let (U, ϕ) be a bundle chart of π : L → M with corresponding chart (U, ξ) of M such that and N ϕ into: ξ(U ) is convex. We have the local forms µϕ : Ck(U, k) → Ck(U, g),eµϕ : Ck(U, Cent(k)) → Ck(U, Cent(g)) f ∈ Ck(U, N(g)) of µ, eµ and Nf ∈ Γk(N(E)), respectively. Locally, the decomposition (21), turns x eµϕ(f · 1) = f · 1 + N ϕ f . The Taylor Formula yields, for each x ∈ U and A ∈ Γ(L): je−1 x y 7−→ Aϕ(y) − Xα<eQm i=1(ξi(y) − ξi(x))αi α! For x, y ∈ U , i ∈ {1, . . . , m} and α ∈ Nm with α < e we write: · ∂α ξ Aϕ(x) = 0. (23) ξi,x(y) := ξi(y) − ξi(x), Ξα,x(y) := φα,x(y) := mYi=1 mYi=1 (ξi(y) − ξi(x))αi , (ξi(y) − ξi(x))αi · ∂α ξ Aϕ(x). Then we define smooth mappings N1, . . . , Nm : U → N(g) and µ0 : U → Hom(k, g) (in the sense of Lie algebra morphisms), by setting for x ∈ U , u ∈ k and the constant map cu : U → g, y 7→ u: µ0(x)(u) := µϕ(cu)(x) and We calculate: ξ Aϕ(x)(cid:1)! Ni(x) := N ϕ ξi,x (x) =eµϕ(ξi,x · 1)(x) − (ξi,x · 1)(x) =eµϕ(ξi,x · 1)(x) (ξi(y) − ξi(x))αi · 1!(cid:0)∂α µϕ (φα,x) = µϕ y 7→ mYi=1 =eµϕ (Ξα,x · 1) · µϕ(cid:0)y 7→ ∂α ξ Aϕ(x)(cid:1) =  ·(cid:0)µ0 · ∂α mYi=1eµϕ(cid:0)ξαi i,x · 1(cid:1) ξ Aϕ(cid:1) {z } i ! ·(cid:0)µ0 · ∂α = mYi=1 ξ Aϕ(cid:1) = N α ·(cid:0)µ0 · ∂α 7−→ Xα<e N α ·(cid:0)µ0 · ∂α ξ Aϕ(cid:1) . Aϕ µϕ 1 α! N αi =N αi i ξ Aϕ(cid:1) . 43 Thus, by (23) and the fact that µ is of order at most e − 1, we have the local form (24) It remains to show that each µ0(x), where x ∈ U , is bijective. This will be done in two steps. First, we verify the following identitiy for A ∈ Γ(L) and B ∈ Γ(E) with µ(A) = B (implying µϕ (Aϕ) = Bϕ): P (A) := Xα<e (−1)α α! ∂α ξ (N α · (µ0 · Aϕ)) = Xα<e (−1)α α! ∂α ξ (N α · Bϕ) =: Q (B) . (25) Note: If S1, . . . , Sm ∈ N(g) and α ≥ e, then Sα = Sα1 1 ◦ . . . ◦ Sαm m = 0 (26) because Sα can be written as a sum of α-th powers of linear combinations of the Si and the e-th power of a nilpotent morphism cotained in N(g) is zero. Therefore we can perform the following manipulations: (−1)α α! Xα<e ∂α ξ (N α · Bϕ) (24) β<e = Xα<e = Xα<e = Xα+β<e = Xα+β<e (26) (18) γ≤α (−1)α α! (−1)α α!β! ∂α ∂α 1 β! ξ N α · Xβ<e N β ·(cid:16)µ0 · ∂β ξ Aϕ(cid:17)(cid:17) ξ (cid:16)N α+β ·(cid:16)µ0 · ∂β ξ Aϕ(cid:17) ξ (cid:16)(cid:0)N α+β ◦ µ0(cid:1) · ∂β ξ (cid:0)N α+β ◦ µ0(cid:1) · ∂α+β−γ ∂α ∂γ ξ (−1)α α!β! (−1)α β!γ!(α − γ)! ξ Aϕ(cid:17) Aϕ. Now we perform the following substitutions: α′ := α + β − γ, β′ := γ and γ′ := β, thus α + β = α′ + β′, α − γ = α′ − γ′ and γ ≤ α ⇐⇒ γ′ ≤ α′. And so we can finally show (25): (−1)α α! Xα<e ∂α γ≤α ξ (N α · Bϕ) = Xα+β<e = Xα′+β′<e = Xα′+β′<e = Xβ′<e γ′≤α′ (19) (−1)α β!γ!(α − γ)! Aϕ ∂γ ξ ξ (cid:0)N α+β ◦ µ0(cid:1) · ∂α+β−γ ξ (cid:16)N α′+β′ ∂β′ (−1)α′−γ′ · (−1)β′ γ′!β′!(α′ − γ′)! Xγ′≤α′ (−1)β′ β′! (−1)β′ (−1)α′−γ′ γ′!(α′ − γ′)! · · (µ0 · Aϕ)(cid:17) . ξ (cid:16)N β′ β′! ∂β′ ξ Aϕ ◦ µ0(cid:17) · ∂α′ ξ (cid:16)N α′+β′ ∂β′ ◦ µ0(cid:17) · ∂α′ ξ Aϕ By the definition of P and µ0, if Aϕ(x) = u = A′ϕ(x) for A, A′ ∈ Γ(L) and x ∈ U , then: P (A)(x) = Xα<e (−1)α α! ∂α ξ (N α(x) (µ0(x) (Aϕ(x)))) = Xα<e (−1)α α! ξ (N α(x) (µϕ(cu)(x))) = P (A′)(x). ∂α So P (A)(x) ∈ g depends on Aϕ only via Aϕ(x) ∈ k and we can define a linear map Px : k → g by setting Px(Aϕ(x)) := P (A)(x) for x ∈ U . We may also define linear maps Qx : g → g for x ∈ U by Qx(Bϕ(x)) := Q(B)(x) for x ∈ U due to an analogous element as above with the N α instead of µ0. Since any Qx is a sum of 1 and a nilpotent linear map (see the sum in the third term of (25) evaluated in x), 44 it is bijective. We define a smooth mapping η0 : U → Hom(g, k) (in the sense of Lie algebra morphisms) by setting for x ∈ U , v ∈ g and the constant map cv : U → g, x′ → v: η0(x)(v) :=(cid:0)µ−1(cid:1)ϕ (cv)(x). We now fix x ∈ U and v ∈ g. Since µ is surjective, there exists A ∈ Γ(L) such that µϕ (Aϕ) = cv, thus Aϕ(x) = η0(x)(v). By (25), we have Px (η0(x)(v)) = Qx(v). The injectivity of Qx implies the injectivity of η0(x) : g → k, yielding e ≤ d. By the symmetry of the arguments, µ0(x) : k → g is also injective, yielding d ≤ e. So µ0(x) is an isomorphism of Lie algebras. Remark 2.99. If the Lie algebra k is complex simple, then it is central, i.e. Cent(k) = C · 1, by the Schur Lemma. If the Lie algebra k is real simple, then Proposition X.1.5 of [He78] says that k satisfies exactly one of the following two conditions: A. k admits a complex structure and the complexification kC = k ⊗R C is the direct sum of two simple isomorphic ideals, hence kC is not a simple C-Lie algebra. B. kC is a simple C-Lie algebra. If the real simple Lie algebra k is the Lie algebra associated to a compact Lie group, then its complexifi- caton kC is complex simple by Lemma X.1.3 of [He78], so we are in case B and have Cent(k) = R · 1. Corollary 2.100. If we have one of the following two cases: 1. the Lie algebra k is complex simple, 2. the Lie algebra k is real simple and associated to a compact Lie group, be identified with some µ0 ∈ Γk (Autλ(L)), i.e. a Ck-section in Γk (Hom(L, L)) where for all x ∈ M the map µ0(x) : Lx → Lλ(x) is an isomorphism of Lie algebras. The bundle Autλ(L) is isomorphic to then for any µ ∈ Aut(cid:0)Γk(L)(cid:1), where k ∈ N, there is a Ck-diffeomorphism λ : M → M such that µ can Aut(L) := AutidM (L) by(cid:0)f : Lx → Lλ(x)(cid:1) 7−→(cid:0)µ−1 ◦ f : Lx → Lx(cid:1). Corollary 2.101. In both of the cases of Corollary 2.100, the Lie algebra of Ck-sections, where k ∈ N, of the trivial bundle L = M × k is naturally isomorphic to Ck(M, k) and, since Diff k(M ) and Γk (Aut(L)) ∼= Ck(M, Aut(k)) can be naturally embedded into Aut(Ck(M, k)) as a subgroup and a normal subgroup, respectively, we obtain the isomorphism Aut(Ck(M, k)) ∼= Ck(M, Aut(k)) ⋊ Diff k(M ). References [BiCr01] [EiSt57] Richard L. Bishop, Richard J. Crittenden; Geometry of Manifolds; American Math- ematical Society (2001). Samuel Eilenberg, Norman Steedrod; Foundations of Algebraic Topology; Prince- ton University Press (1957). [GrHaVa72] Werner Greub, Stephen Halperin, Ray Vanstone; Connections, Curvature, and Co- homology; Volume I; Academic Press (1972). [He78] [Hu72] Sigurd Helgason; Differential Geometry, Lie Groups, and Symmetric Spaces; Academic Press (1978). J. E. Humphreys; Introduction to Lie Algebras and Representation Theory; Springer-Verlag (1972). 45 [Hu75] [La01] [Le79] [Le80] Dale Husemoller; Fibre Bundles; Springer Verlag (1975). Tsit-Yuen Lam; A First Course in Noncommutative Rings; Springer-Verlag (2001). Pierre Lecomte; Alg`ebres de Lie d'ordre z´ero sur une vari´et´e; Th`ese de doctorat, Liege (1979). Pierre Lecomte; Sur l'alg`ebre de Lie des sections d'un fibr´e en alg`ebres de Lie; Annales de l'institut Fourier, 30 no. 4 (1980), pp. 35-50. [MeRe93] Alberto M´edina, Philippe Revoy; Alg`ebres de Lie orthogonales; Modules orthog- onaux; Communications in Algebra, 21(7) (1993), p. 2301. [MoHeRaCo94] Daniel Moak, Konrad Heuvers, K. P. S. Bhaskara Rao, Karen Collins; An inversion relation of multinomial type; Discrete Mathematics, 131(1-3) (1994), pp. 195-204. [Na68] [Ne02] [Ne04] [Ne05] [Ne94] [StWi90] [tD91] [Wo06] Raghavan Narasimhan; Analysis on real and complex manifolds; North Holland Publishing Company (1968). Karl Hermann Neeb; Skript zur Vorlesung Lie-Algebren im SS 02 an der TUD. http://www.mathematik.tu-darmstadt.de:8080/ags/ag5/mitglieder/ ...professoren/neeb/skripten/Lie-Algebren-SS02.pdf Karl Hermann Neeb; Notes on non-abelian cohomology; Manuscript (2004). Karl Hermann Neeb; Lecture Notes on Manifolds and Transformation Groups; WS 05 - Technische Universitat Darmstadt. http://www.mathematik.tu-darmstadt.de:8080/ags/ag5/mitglieder/ ...professoren/neeb/skripten/MFK-WS2005.pdf Karl Hermann Neeb; Skript zur Vorlesung Lie-Algebren. http://www.mathematik.tu-darmstadt.de:8080/ags/ag5/mitglieder/ ...professoren/neeb/skripten/Lie-Algebren-WS94.pdf Uwe Storch, Hartmut Wiebe; Lehrbuch der Mathematik, Band II: Lineare Al- gebra; BI Wissenschaftsverlag (1990). Tammo tom Dieck; Topologie; Gruyter (1991). Christoph Wockel; Infinite Dimensional Lie Theory for Gauge Groups; Disser- tation, Darmstadt (2006). 46
1604.06066	1	1604	2016-04-20T19:14:37	On the Galois correspondence for Hopf Galois structures	[ "math.RA" ]	We study the question of the surjectivity of the Galois correspondence from subHopf algebras to subfields given by the Fundamental Theorem of Galois Theory for abelian Hopf Galois structures on a Galois extension of fields with Galois group G', a finite abelian p-group. Applying the connection between regular subgroups of the holomorph of a finite abelian p-group (G, +) and associative, commutative nilpotent algebra structures A on (G, +) of Caranti, et. al., we show that if A gives rise to a H-Hopf Galois structure on L/K, then the K-subHopf algebras of H correspond to the ideals of A. Among the applications, we show that if G and G' are both elementary abelian p-groups, then the only Hopf Galois structure on L/K of type G for which the Galois correspondence is surjective is the classical Galois structure on L/K.	math.RA	math	ON THE GALOIS CORRESPONDENCE FOR HOPF GALOIS STRUCTURES LINDSAY N. CHILDS Abstract. We study the question of the surjectivity of the Ga- lois correspondence from subHopf algebras to subfields given by the Fundamental Theorem of Galois Theory for abelian Hopf Ga- lois structures on a Galois extension of fields with Galois group Γ, a finite abelian p-group. Applying the connection between regular subgroups of the holomorph of a finite abelian p-group (G, +) and associative, commutative nilpotent algebra structures A on (G, +), we show that if A gives rise to a H-Hopf Galois structure on L/K, then the K-subHopf algebras of H correspond to the ideals of A. As applications, we show that if G and Γ are both elementary abelian p-groups, then the only Hopf Galois structure on L/K of type G for which the Galois correspondence is surjective is the classical Galois structure. Also, if Γ is elementary abelian of order pn and p > n there exists an H-Hopf Galois structure on L/K for which there are exactly n + 1 K-sub-Hopf algebras (but approxi- mately npn intermediate subfields for n large). By contrast, if Γ is cyclic of order pn, p odd, then for every Hopf Galois structure on L/K, the Galois correspondence is surjective. 1. Introduction The Fundamental Theorem of Galois Theory (FTGT) of Chase and Sweedler [CS69] states that if L/K is a H-Hopf Galois extension of fields for H a K-Hopf algebra H, then there is an injection F from the set of K-sub-Hopf algebras of H to the set of intermediate fields K ⊆ E ⊆ L given by sending a K-subHopf algebra J to F (J) = LJ . The strong form of the FTGT holds if the injection is also a surjection. For a classical Galois extension of fields with Galois group Γ, the FTGT holds in its strong form. It is known from [GP87] that if L/K is a (classical) Galois extension with non-abelian Galois group Γ, then there is a Hopf Galois structure on L/K so that F maps onto the subfields E of L that are normal over K. So if Γ is not a Hamiltonian group [Ha59, 12.5], then L/K has a Hopf Galois structure for which the strong form of the FTGT does not hold. In particular, the strong Date: May 21, 2019. 1 2 LINDSAY N. CHILDS form fails extremely for the unique [By04] non-classical Hopf Galois structure on L/K when Γ is a non-abelian simple group. Suppose L/K is a classical Galois extension with Galois group Γ. If L/K is also H-Hopf Galois for some K-Hopf algebra H, then L ⊗K K is L ⊗K H-Hopf Galois over L, and by [GP87], L ⊗K H ∼= LN for some regular subgroup N of Perm(G) normalized by λ(Γ), the image in Perm(Γ) of the left regular representation of Γ on Γ. If G is an abstract group and α : G → Perm(Γ) is an injective homomorphism from G to Perm(Γ) whose image α(G) = N, then we say that H has type G. Implicit in [GP87], and made explicit by Crespo, Rio and Vela ([CRV16], Proposition 2.2), is that the K-subHopf algebras of H correspond to the subgroups of α(G) that are normalized by λ(Γ) in Perm(Γ). Nearly all of the examples examining the success or failure of the strong form of the FTGT for a non-classical Hopf Galois structure on a classical Galois extension L/K with Galois group Γ involve non- abelian groups. Perhaps the only wholly abelian example of failure in the literature is in [CRV15], 3.1, where Γ ∼= C2 × C2 and L/K has a Hopf Galois structure by H, a K-Hopf algebra which is a K-form of LC4. Then by classical Galois theory, there are three intermediate subfields between K and L, but LC4 has only one intermediate L-Hopf algebra, so H can have at most one intermediate K-subHopf algebra. Hence the strong form of the FTGT cannot hold for that Hopf Galois structure. Here we assume that L/K is a Galois extension with Galois group an abelian p-group Γ of order pn. We look at Hopf Galois structures on L/K by K-Hopf algebras H of type G where G is abelian, also of order pn. There is a sequence of correspondences associated to H: • the abelian K-Hopf algebra H corresponds by base change and Galois descent (as in [GP87]) to the L-Hopf algebra L ⊗K H, which equals LN for some regular abelian subgroup N of Perm(Γ) normalized by λ(Γ); • if α : G → Perm(Γ) is a regular embedding with image N, then α corresponds to a regular embedding β : Γ → Hol(G), as shown in [By96]; • writing G = (G, +), the regular subgroup β(Γ) of Hol(G) corre- sponds to an associative, commutative nilpotent ring structure A on (G, +), as shown in [CDVS06] and [FCC12]. Our main result implies that under these correspondences, the K- sub-Hopf algebras of H correspond to ideals of A. As a consequence, we show that for Γ ∼= G an elementary abelian p-group, the only case ON THE GALOIS CORRESPONDENCE 3 where the FTGT holds in its strong form is if H is the classical Galois structure on L/K. This paper and [FCC12], [Ch15] and [Ch16] demonstrate in different ways the usefulness of the correspondence of [CDVS06] in the Hopf Galois theory of Galois extensions of fields whose Galois group is a finite abelian p-group. 2. Translating to and from the holomorph Let L/K be a Galois extension with Galois group Γ and let G be a group of the same cardinality as Γ. The H-Hopf Galois structures on L/K of type G correspond to images of regular embeddings α : G → Perm(Γ) where α(G) is normalized by λ(G). As shown in [By96], a regular embedding α : G → Γ whose image α(G) is normalized by λ(Γ) corresponds to a regular embedding β : Γ → Hol(G), where Hol(G) = ρ(G) · Aut(G) ⊂ Perm(G) is the normalizer of λ(G) in Perm(G). Here ρ : G → Perm(G)is the right regular representations of G in Perm(G). The relationship be- tween α and β is as follows: Let β : Γ → Hol(G) be a regular embedding. Define b : Γ → G by b(γ) = β(γ)(eG) for γ in Γ, g in G, where eG is the identity element of G. Then β(γ)(g) = (b(λ(γ))b−1)(g) = (C(b)λ(γ))(g) Define α : G → Perm(Γ) by α(g)(γ) = (b−1(λ(g))b)(γ) = (C(b−1)λ(g))(γ). The following formulas will be useful below. Proposition 2.1. Suppose β : Γ → Hol(λ(G)) is an regular embed- ding, and let α = C(b−1)λG : G → Perm(Γ) be the regular embedding corresponding to β. Then for all γ in Γ and g in G, there is some h in G so that and β(γ)λ(g)β(γ)−1 = λ(h) λ(γ)α(g)λ(γ)−1 = α(h). Proof. The first formula follows because β maps Γ into Hol(G), the nor- malizer of λ(G) in Perm(G). Since C(b−1)(β)(γ) = λ(γ) and C(b−1)λ(g) = α(g), the second formula follows from the first by applying C(b−1) to the first formula. (cid:3) 4 LINDSAY N. CHILDS 3. On the Galois correspondence for Hopf Galois structures Let L/K be a Galois extension of fields with Galois group Γ, and suppose α : G → P erm(Γ) is a regular embedding such that λ(Γ) normalizes α(G). Then by descent, H = L[α(G)]G is a K-Hopf algebra and there is an action of H on L making L/K into an H-Hopf Galois extension. (See [GP87].) The Fundamental Theorem of Galois Theory of Chase and Sweedler [CS69] gives an injection F from the set of K- subHopf algebras of H to the set of intermediate fields F with K ⊆ F ⊆ L by H ′ 7→ LH ′ = {x ∈ Lhx = ǫ(h)x for all h in H ′}. From Proposition 2.2 of [CRV16], the K-subHopf algebras of H corre- spond to λ(Γ)-invariant subgroups of α(G). Thus to study the image of F , we look at the λ(Γ) invariant subgroups of α(G). We do this for G a finite abelian p-group by utilizing a characteri- zation of regular subgroups of Hol(G) due to Caranti, Della Volta and Sala [CDVS06] as extended in Proposition 2 of [FCC12]: Proposition 3.1. Let (G, +) be a finite abelian p-group. Then each regular subgroup of Hol(G) is isomorphic to the group (G, ◦) induced from a structure (G, +, ·) of a commutative, associative nilpotent ring (hereafter, "nilpotent") on (G, +), where the operation ◦ is defined by g ◦ h = g + h + g · h. The idea is the following: Let (G, +) be an abelian group of order pn , and suppose that A = (G, +, ·) is a nilpotent ring structure on (G, +) yielding the operation ◦. Define τ : (G, ◦) → Hol(G, +) by τ (g)(x) = g ◦ x. Then τ (g)(0) = g, and τ (g)τ (g ′)(x) = τ (g)(g ′ ◦ x) = g ◦ (g ′ ◦ x) = (g ◦ g ′) ◦ x = τ (g ◦ g ′)(x), so τ is an isomorphism from (G, ◦) into Hol(G, +) whose image τ (G, ◦) = T is a regular subgroup of Hol(G). This process is reversible: given a regular subgroup T of Hol(G, +), there is a nilpotent ring structure (G, +, ·) on G, which defines the ◦ operation as above and yields a unique isomomorphism τ : (G, ◦) → T so that τ (g)(x) = g ◦ x. Now suppose L/K be a Galois extension with Galois group Γ, a finite abelian p-group of order pn, and ξ : Γ → (G, ◦) is some isomorphism. Given A and τ , let β = τ ξ : Γ → T , an embedding of Γ onto T in Hol(G). Then β yields the map b : Γ → G ON THE GALOIS CORRESPONDENCE 5 by b(γ) = β(γ)(0) = τ (ξ(γ))(0) = ξ(γ) ◦ 0 = ξ(γ). Thus b is an isomorphism from Γ to (G, ◦). Then b defines an embed- ding α : (G, +) → Perm(Γ), and we have the relations of Proposition 2.1: for all γ in Γ, g in G, there is some h in G so that and λ(γ)α(g) = α(h)λ(γ) β(γ)λ(g) = λ(h)β(γ). Theorem 3.2. Suppose the nilpotent algebra A = (G, +, ·) yields the regular embedding α : (G, +) → Perm(Γ) whose image is normalized by λ(Γ). Then the lattice (under inclusion) of λ(Γ)-invariant subgroups of α(G) is isomorphic to the lattice of ideals of A. Proof. First, α : G → P erm(Γ) is an injective homomorphism from (G, +) to Perm(Γ). So if J is an additive subgroup of G, then α(J) is a subgroup of α(G) ⊂ Perm(Γ). Conversely, if J is a subset of G and α(J) is a subgroup of α(G), then for s, t in J, α(s + t) = α(s)α(t) is in α(J), so J is an additive subgroup of G. So there is a bijection between subgroups of (G, +) and subgroups of α(G). Clearly J1 ⊆ J2 iff α(J1) ⊆ α(J2), so the bijection is lattice-preserving. Suppose the image α(G) of α is normalized by λ(Γ), so for all γ in Γ, g in G, there is some h in G so that This equation holds iff λ(γ)α(g)λ(γ)−1 = α(h). β(γ)λG(g) = λG(h)β(γ). Recalling that A = (G, +, ·) = (G, ◦), factor β = τ ξ where ξ : Γ → A = (G, ◦) is an isomorphism and τ : A = (G, ◦) → Hol(G) an embedding. Let ξ(γ) = k in A. Then the last equation is and applying this to x in G gives τ (k)λG(g) = λG(h)τ (k), τ (k)(g + x) = h + τ (k)(x), k ◦ (g + x) = h + k ◦ x. Viewing this equation in A where a ◦ b = a + b + a · b, we have k + (g + x) + k · g + k · x = h + k + x + k · x. This last equation reduces to h = g + k · g. 6 LINDSAY N. CHILDS Now suppose J is an ideal of A and g is in J. Then k · g is in J, so h is in J, and so λ(γ) conjugates α(g) in α(J) to an element of α(J). So α(J) is normalized by λ(Γ) in Perm(Γ). Conversely, suppose J is an additive subgroup of (G, +, ·) = A and α(J) is normalized by λ(Γ). Then for all γ in G, g in J, λ(γ)α(g)λ(γ)−1 = α(h) and α(h) is in α(J). So h is in J. Hence for all k = ξ(γ) in G, and g in J, h = g + k · g is in J. Now J is an additive subgroup of A, so k · g is in J for all k in G, g in J. Thus J is an ideal of A. (cid:3) 4. Examples Theorem 3.2 transforms the problem of describing the image of the Galois correspondence map F on a H-Hopf Galois structure on L/K to the study of the ideals of the nilpotent algebra associated to H. In this section we look at some examples. Theorem 4.1. Let L/K be a Galois extension of fields with Galois group Γ an elementary abelian p-group of order pn. Suppose L/K has a Hopf Galois structure by an abelian Hopf algebra H of type G where G is an elementary abelian p-group. Let A be a nilpotent ring structure yielding the regular subgroup T ∼= (G, ◦) ⊂ Hol(G) corresponding to H, where (G, ◦) ∼= Γ. Then the H-Hopf Galois structure on L/K satisfies the strong form of the FTGT if and only if A is the trivial nilpotent algebra satisfying A2 = 0, if and only if T = ρ(G), iff H is the classical Galois structure by KΓ) on L/K. Proof. If A2 = 0, then (G, ◦) = (G, +), so the regular subgroup T acts on G by τ (g)(h) = g◦h = g+h, hence T = λ(G). Since G is abelian, the corresponding Hopf Galois structure on L/K is the classical structure by the K-Hopf algebra K[Γ]. So the Galois correspondence holds in its strong form. For the converse, view (G, +) as an n-dimensional Fp-vector space. Suppose A2 6= 0. Then for some a, b in A, ab 6= 0. Then the subspace Fpa does not contain ab. For if ab = ra for r 6= 0 in Fp, then a = sba for s 6= 0 in Fp. Then a = (sb)a = (sb)2a = . . . = (sb)n+1a = 0 since A is nilpotent of dimension n, hence (sb)n+1 = 0. Thus the subspace Fpa is not an ideal of A. The subgroup α(Fpa) of α(G) is then not normalized by λ(Γ). But Γ ∼= G, so there are bijections between subgroups of α(G), subgroups of G, subgroups of Γ and (by classical Galois theory) subfields of L ON THE GALOIS CORRESPONDENCE 7 containing K. If some subgroup of α(G) is not normalized by λ(Γ), then the number of K- subHopf algebras of H = L[α(G)]G is strictly smaller than the number of subfields between K and L. So the Galois correspondence for the H-Hopf Galois structure on L/K does not hold in its strong form. (cid:3) By choosing a particular nilpotent algebra structure on (Fn p , +) we can see how badly the Galois correspondence can fail to be surjective. Let A be the primitive n-dimensional nilpotent Fp-algebra generated p , +) and so the multiplication p , +). Let p , ◦) where the operation ◦ is defined using the multiplication by z with zn+1 = 0. Then (A, +) ∼= (Fn on A yields a nilpotent Fp-algebra structure on (G, +) = (Fn G = (Fn on A by a ◦ b = a + b + a · b. Theorem 4.2. Let G be an elementary abelian p-group of order pn. Let A be a primitive Fp-algebra structure A on G, and let (G, ◦) be the corresponding group structure on Fn p . Suppose L/K is a Galois extension of fields with Galois group Γ ∼= (G, ◦). Then the primitive nilpotent Fp-algebra A corresponds to an H-Hopf Galois structure on L/K for some K-Hopf algebra H, and the K-subHopf algebras of H form a descending chain H = H1 ⊃ H2 ⊃ . . . ⊃ Hn ⊃ K. Hence the Galois correspondence F for H maps onto exactly n+1 fields F with K ⊆ F ⊆ L. Proof. Given Theorem 3.2, we just need to show that ideals of A are Ji = hzii for i = 1, . . . , n. Suppose J is an ideal of A and has an element r(zk + zk+rb) of minimal degree k, where r 6= 0 in Fp, b in A. Then J also contains zk + zk+rb and (zk + zk+rb)(−zrb) = −zk+rb − zk+2rb2, hence their sum, zk − zk+2rb2 = zk + zk+r′ b′ where r′ > r. Repeating this argument until r′ > n shows that J contains zk, hence J ⊇ Jk = hzki. Since Jk = hzki contains every element of degree ≥ k, J = Jk. Thus A has exactly n + 1 ideals. Since the correspondence between ideals of A and λ(Γ) invariant subgroups of α(G) is lattice-preserving, we have a single filtration α(G) = α(J1) ⊃ α(J2) ⊃ . . . ⊃ α(Jn) ⊃ 0. 8 LINDSAY N. CHILDS If H is the corresponding K- of λ(G)-invariant subgroups of α(G). Hopf algebra making L/K into a Hopf Galois extension, then H has a unique filtration of K-sub-Hopf algebras, H = H1 ⊃ H2 ⊃ . . . ⊃ Hn ⊃ K. (cid:3) For A a primitive nilpotent Fp-algebra with An+1 = 0, the corre- sponding group (G, ◦) is isomorphic (by a 7→ 1 + a) to the group of principal units of the truncated polynomial ring Fp[x]/(xn+1Fp[x]. The structure of that group is described in Corollary 3 of [Ch07]. In par- ticular (G, ◦), hence Γ, is an elementary abelian p-group if and only if p > n. In Theorem 4.2, when p > n, then L/K is classically Galois with Galois group Γ ∼= (Fn p , +). So the number of subgroups of Γ, and hence the the number of subfields E with K ⊆ E ⊆ L, is equal to the number of subspaces of Fn p , namely n Xr=1 (pn − 1)(pn − p) · · · (pn − pr−1) (pr − 1)(pr − p) · · · (pr − pr−1) ∼ npn for n large. So the Galois correspondence map F is extremely far from being surjective for a Hopf Galois structure corresponding to a nilpotent algebra structure A with dim(A/A2) = 1. By contrast, Proposition 4.3. Let L/K be a Galois extension of fields with Galois group Γ cyclic of order pn, p odd. Let the K-Hopf algebra H give a Hopf Galois structure on L/K. Then H has type G where G ∼= Γ, and the Galois correspondence for that Hopf Galois structure holds in its strong form. Proof. From [Ko98] it is known that if Γ is cyclic of order pn then every Hopf Galois structure must have type G ∼= Γ. So let G be cyclic of order pn, which we identify with (Z/pnZ, +). Then we view Hol(G) = G ⋊ Aut(G) as the set of pairs (a, g) where a and g are modulo pn and (g, p) = 1, or equivalently as the set of matrices (cid:18)g a 0 1(cid:19) in Aff1(Z/pnZ) ⊂ GL2(Z/pnZ), acting on s in G by (cid:18)g a 0 1(cid:19)(cid:18)s 1(cid:19) = (cid:18)gs + a 1 (cid:19) . ON THE GALOIS CORRESPONDENCE 9 View Γ as the free Z/pnZ-module with basis y. From Proposition 2 of [Ch11], the pn−1 regular embeddings β : Γ = (Z/pnZ)y → Hol(G) are determined by β(y) where β(y) = (cid:18)1 + pd −1 1 (cid:19) 0 for some d modulo pn−1. So in the notation of the last section, τ (−1) = (cid:18)(1 + pd −1 1 (cid:19) 0 and acts on s in G as above. That action defines the operation ◦ on G by (−1) ◦ s = (1 + pd)s − 1 = −1 + s + pds. The multiplication on (G, +) to make (G, +, ·) = A a nilpotent algebra is then defined by (−1) · s = (−1) ◦ s − ((−1) + s) = (−1 + s + pds) + 1 − s = pds. By distributivity, for every r, s in Z/pnZ, −r · s = rspd. Replacing d by −d, let Ad be the commutative nilpotent algebra struc- ture on (Z/pnZ, +) with multiplication r · s = rspd for all r, s in Z/pnZ. It is then easy to check that the ideals of Ad are the principal ideals generated by pr, for r = 0, . . . , n. Since those are also the additive subgroups of (Ad, +) = (Z/pnZ, +), it follows by Theorem 3.2 that for every Hopf Galois structure on L/K, the Galois correspondence holds in its strong form. (cid:3) References [By96] [By04] [CDVS06] [CS69] [Ch89] [Ch07] N. P. Byott, Uniqueness of Hopf Galois structure of separable field extensions, Comm. Algebra 24 (1996), 3217 -- 3228. N. P. Byott, Hopf-Galois structures on field extensions with simple Galois groups, Bull. London Math. Soc. 36 (2004), 23 -- 29. A. Caranti, F. Dalla Volta, M. Sala, Abelian regular subgroups of the affine group and radical rings. Publ. Math. Debrecen 69 (2006), 297 -- 308. S. U. Chase, M. E. Sweedler, Hopf Algebras and Galois Theory, Springer LNM 97 (1969). L. N. Childs, On the Hopf Galois theory for separable field extensions, Comm. Algebra 17 (1989), 809 -- 825. L. N. Childs, Some Hopf Galois structures arising from elementary abelian p-groups, Proc. Amer. Math. Soc. 135 (2007), 3453 -- 3460. 10 [Ch11] [Ch15] [Ch16] [CRV15] [CRV16] [FCC12] [GP87] [Ha59] [Ko98] LINDSAY N. CHILDS L. N. Childs, Hopf Galois structures on Kummer extensions of prime power degree, New York J. Math. 17 (2011). 5174. L. N. Childs, On Abelian Hopf Galois structures and finite commu- tative nilpotent rings, New York J. Math., 21 (2015), 205 -- 229. L. N. Childs, Obtaining abelian Hopf Galois structures from finite commutative nilpotent rings, arxiv: 1604.05269 T. Crespo, A. Rio, M. Vela, From Galois to Hopf Galois: theory and practice, Contemp. Math. 649 (2015), 29 -- 46. T. Crespo, A. Rio, M. Vela, On the Galois correspondence theorem in separable Hopf Galois theory, Pub. Mat. Debrecen 60 (2016), 221 -- 234. S. C. Featherstonhaugh, A. Caranti, L. N. Childs, Abelian Hopf Ga- lois structures on prime-power Galois field extensions, Trans. Amer. Math. Soc. 364 (2012), 3675 -- 3684. C. Greither, B. Pareigis, Hopf Galois theory for separable field exten- sions, J. Algebra 106 (1987), 239 -- 258. M. Hall, The Theory of Groups, Macmillan, New York, 1959. T.Kohl, Classification of the Hopf Galois structures on prime power radical extensions, J. Algebra 207 (1998), 525 -- 546. Department of Mathematics and Statistics, University at Albany, Albany, NY 12222 E-mail address: [email protected]
1611.01488	1	1611	2016-11-04T18:48:57	On bases that are closed under multiplication	[ "math.RA" ]	It is well known that there is no basis of the field for real numbers regarded as a vector space over any proper subfield that is closed under multiplication. Mabry has extended this result to bases of arbitrary proper field extensions. The aim of this short communication is to notice that the proof of the result concerning the reals may be adjusted to a larger class of algebras (including full matrix algebras); thereby we subsume Mabry's result.	math.RA	math	ON BASES THAT ARE CLOSED UNDER MULTIPLICATION TOMASZ KANIA To appear in the American Mathematical Monthly Abstract. It is well known that there is no basis of the field for real numbers regarded as a vector space over any proper subfield that is closed under multiplication. Mabry has extended this result to bases of arbitrary proper field extensions. The aim of this short communication is to notice that the proof of the result concerning the reals may be adjusted to a larger class of algebras (including full matrix algebras); thereby we subsume Mabry's result. Let k be a field and let A be a k-algebra, i.e., A is a vector space over the field k which is at the same time a (possibly non-unital or non-commutative) ring whose operations are compatible with the vector-space operations. (A ring is unital when it is has an identity element with respect to multiplication.) For example, for every fixed natural number n, the Cartesian product kn of n copies of k is a paradigm example of a unital k-algebra when furnished with the operations of addition, multiplication and scalar multiplication defined component-wise. Since A is a vector space over k on its own, we may talk about bases in A (that is, maximal linearly independent subsets of A). By linear algebra, a subset H ⊂ A is a basis if and only if every vector x ∈ A may be written uniquely as a linear combination of vectors in H. (By the Kuratowski -- Zorn lemma, every vector space has a basis.) Let us say that a basis H of A is closed under multiplication if x · y ∈ H whenever x and y are in H. Such bases have been studied by Mabry ([3]) in the context of field extensions. More specifically, Mabry proved that if k1 is a proper subfield of another field k2, then no basis of the latter field, regarded as a k1-algebra, is closed under multiplication. L´opez-Permouth, Moore and Szabo ([2]) studied bases in unital algebras that consist of invertible elements and they have obtained an alternative proof of Mabry's result on field extensions. The aim of this note is to extend Mabry's result from a different angle -- we shall give a quick proof of a more general yet still very simple result that will encompass not only field extensions but many other algebras too. Before we do so, we introduce a piece of terminology. An ideal of a k-algebra A is a linear subspace of A that is closed under multiplying its elements by arbitrary elements of A from left and right. The k-codimension of a subspace 2010 Mathematics Subject Classification. 13P10, 15A03 (primary), and 20M25 (secondary). Key words and phrases. Hamel basis, basis closed under multiplication, ideal of codimension one, simple algebra. The author acknowledges with thanks funding received from the European Research Council / ERC Grant Agreement No. 291497. 1 2 TOMASZ KANIA V of a vector space W over k is the k-dimension of the quotient space W/V . Armed with this terminology, we are ready to present the main result of this note. Main Theorem. Let k be a field and suppose that A is a k-algebra. Suppose further that A does not have ideals of k-codimension one. Then no basis of A is closed under multiplication. In the case where A is a proper field extension of k (which implies dimk A > 2), A does not have ideals other than {0} and A (algebras with this property are called simple), so in particular it does not have ideals of k-codimension 1. Indeed, if a ∈ A is a non-zero element in A, then for any b ∈ A one has b = ba−1a, so A itself is the smallest ideal containing a. Consequently, we derive Mabry's result as a corollary and the conclusion of our main result extends to the class of all simple k-algebras. It is well-known that the k-algebras Mn(k) consisting of all n × n-matrices over k (n ∈ N) are prototypical examples of simple k-algebras. Another familiar example is the R-algebra of quaternions. On the other hand, the Cartesian products kn (n > 1) do have bases closed under multiplication, namely (1, 0, 0, . . .), (1, 1, 0, . . .), . . . , (1, 1, 1, . . .). Semigroup algebras constitute another class of k-algebras having bases closed under multi- plication. Indeed, fix a field k and a semigroup (S, ·) (that is, a set with an associative bi- nary operation). Denote by k[S] the vector space over k comprising all functions f : S → k that assume at most finitely many non-zero values. For each s ∈ S define δs to be the function that takes value 1 on s and 0 otherwise. Then the operation δs · δt = δst (s, t ∈ S) extends uniquely to an associative operation on k[S] that makes it a k-algebra. Clearly {δs : s ∈ S} is a basis for k[S] that is closed under multiplication. Proof of the Main Theorem. Let A be a k-algebra with a basis H closed under multiplica- tion. Consider the map f : A → k given by f (x) = X ch, h∈H where x = Ph∈H chh (ch ∈ k, h ∈ H) is the unique basis expansion for x ∈ A. (This map is well-defined as there are at most finitely many non-zero scalars ch in the basis expansion of x.) Then f is a linear functional on A. As H is closed under multiplication, f is multiplicative too, as already observed [1, Problem 7 in Section 14.6] in the case where A is the field of reals regarded as an algebra over the field of rational numbers. Indeed, given two elements x = Ph∈H chh and y ∈ Pg∈H dgg in A, where ch, dg ∈ k, h, g ∈ H (recall that there are at most finitely many non-zero scalars ch and dh, where h ∈ H), we have chdg = f (x) · f (y), f (x · y) = f (X X as the basis expansion is unique. h∈H g∈H chdg h · g {z}∈H ) = X X h∈H g∈H In the case where A has k-dimension at least two, there exist two distinct elements h, h′ ∈ H, so one has f (h − h′) = 0, hence f has non-trivial kernel. As f is non-zero, linear ON BASES THAT ARE CLOSED UNDER MULTIPLICATION 3 and multiplicative, it is a homomorphism of k-algebras. Consequently, the kernel of f is a proper two-sided ideal of A. Moreover, as f is a surjection onto k, the k-codimension of ker f is equal to 1. (cid:3) Acknowledgements. We wish to thank the anonymous referee for valuable comments that improved the presentation of this note significantly. References [1] M. Kuczma, An introduction to the theory of functional equations and inequalities. Cauchy's equation and Jensen's inequality. Second edition. Edited and with a preface by Attila Gil´anyi. Birkhuser Verlag, Basel, 2009. [2] S. L´opez-Permouth, J. Moore and S. Szabo, Algebras having bases consisting entirely of units, in: Groups, Rings and Group Rings, Contempoaray Mathematics, Vol. 499, American Mathematical Society, Providence, RI, 2009, pp. 219 -- 228. [3] R. D. Mabry, No nontrivial Hamel basis is closed under multiplication, Aequat. Math. 71 (2006), 294 -- 299. Mathematics Institute, University of Warwick, Gibbet Hill Rd, Coventry, CV4 7AL, England E-mail address: [email protected]
1805.00766	1	1805	2018-05-02T12:34:00	The Baker-Campbell-Hausdorff formula via mould calculus	[ "math.RA" ]	The well-known Baker-Campbell-Hausdorff theorem in Lie theory says that the logarithm of a noncommutative product e X e Y can be expressed in terms of iterated commutators of X and Y. This paper provides a gentle introduction t{\'o} Ecalle's mould calculus and shows how it allows for a short proof of the above result, together with the classical Dynkin explicit formula [Dy47] for the logarithm, as well as another formula recently obtained by T. Kimura [Ki17] for the product of exponentials itself. We also analyse the relation between the two formulas and indicate their mould calculus generalization to a product of more exponentials.	math.RA	math	The Baker-Campbell-Hausdorff formula via mould calculus Yong Li∗1, David Sauzin†2, and Shanzhong Sun‡1 1Department of Mathematics, Capital Normal University, Beijing 100048 P. R. China 2CNRS UMR 8028 IMCCE, 77 av. Denfert-Rochereau, 75014 Paris, France Abstract The well-known Baker-Campbell-Hausdorff theorem in Lie theory says that the logarithm of a noncommutative product eX eY can be expressed in terms of iterated commutators of X and Y . This paper provides a gentle introduction to ´Ecalle's mould calculus and shows how it allows for a short proof of the above result, together with the classical Dynkin explicit formula [Dy47] for the logarithm, as well as another for- mula recently obtained by T. Kimura [Ki17] for the product of exponentials itself. We also analyse the relation between the two formulas and indicate their mould calculus generalization to a product of more exponentials. Contents 1 Introduction 2 Mould calculus for pedestrians 3 The BCH Theorem and Dynkin's formula 4 Alternative formulas for etX etY and its logarithm 5 Generalization to an arbitrary number of factors 6 Relation between the two kinds of mould expansion 2 4 10 11 14 16 ∗Partially supported by NSFC (No.11131004, 11271269, 11771303), Email: [email protected] †Email: [email protected] ‡Partially supported by NSFC (No.11131004, 11271269, 11771303), Email: [email protected] 1 1 Introduction Let A be a noncommutative associative algebra with unit. In the associative algebra A [[t]] of all power series in an indeterminate t with coefficients in A , one can take the exponential of any series without constant term in t and the logarithm of any series with constant term 1A . In this context, the famous Baker-Campbell-Hausdorff theorem (BCH theorem, for short) can be phrased as log(etX etY ) ∈ Lie(X, Y )[[t]] for any X, Y ∈ A , (1) where Lie(X, Y ) is the Lie subalgebra of A generated by X and Y , i.e. the smallest subspace which contains X and Y and is stable under commutator (see e.g. [BF12] and references therein). In fact, using the notation [A, B] or adA B for a commutator AB − BA, one has log(etX etY ) = t(X + Y ) + t2 2 [X, Y ] + t3 12 ([X, [X, Y ]] + [Y, [Y, X]]) − t4 24 [Y, [X, [X, Y ]]] + · · · , where the coefficient of each power of t can be written in terms of nested commutators involving X and Y only, and there is a remarkable explicit formula due to Dynkin [Dy47]: log(eX eY ) =X (−1)k−1 k tσ σ [X p1 Y q1 · · · X pk Y qk ] p1!q1! · · · pk!qk! (2) with summation over all k ∈ N∗ and (p1, q1), · · · , (pk, qk) ∈ N × N \ {(0, 0)}, where σ := p1 + q1 + · · · + pk + qk and [X p1 Y q1 · · · X pk Y qk ] := adp1 Y if qk ≥ 1 and adp1 X X if qk = 0 (in which case pk ≥ 1). Of course, the contribution of the terms with qk ≥ 2, or with pk ≥ 2 and qk = 0, is zero. Y · · · adpk−1 X adqk−1 X adq1 Y · · · adpk X adq1 Y Our aim is to revisit the BCH theorem and the Dynkin formula in the light of ´Ecalle's so-called "mould calculus". We will show how mould calculus allows one to prove these results with little effort, as well as an interesting formula which was recently obtained by T. Kimura [Ki17] in relation to the BCH theorem and the Zassenhaus formula and reads etX etY = 1A + ∞Xr=1 ∞Xn1,...,nr=1 1 nr(nr + nr−1) · · · (nr + · · · + n1) Dn1 · · · Dnr with Dn := tn (n − 1)! adn−1 X (X + Y ) for each n ≥ 1. (3) We will also show how formula (3) and a little knowledge of mould calculus immediately imply the BCH theorem, and how the results can be generalized to a product of more than two exponentials. It seems hard to prove all these facts using the methods of [Ki17], which rely on a lot of explicit combinatorial computations, whereas almost no computation is needed when using a tiny part of mould machinery. In a nutshell, the point is that the rational coefficients in (3) make up a "symmetral mould"-in fact, a very classical one in 2 mould calculus-and that Dynkin's formula (2) is in essence a typical "Lie mould expan- sion" involving an "alternal" mould; we will explain in due time what "mould expansions", "symmetrality" and "alternality" are and how they relate to the Lie theory. We will also define a new operation in mould calculus, which gives the relation between the rational coefficients appearing in formulas (2) and (3). Mould calculus was set up by J. ´Ecalle in the 1980s as part of his resurgence theory ([Ec81], [Ec92]). Originally, ´Ecalle developed resurgence theory as a tool to study analytic classification problems within dynamical system theory, first for one-dimensional holomor- phic germs, and then for much larger classes of discrete dynamical systems or vector fields, allowing him to tackle the Dulac conjecture about the finiteness of limit cycles of planar analytic vector fields. It soon turned out that resurgence theory has its own merits not only in mathematics but also in physics. For example, quantum resurgence was developed by ´Ecalle himself ([Ec84]) and Voros ([Vo83]) to study the spectrum of Schrodinger operators, and it was continued by Pham and his collaborators (e.g. [DDP93]) as an essential aspect of exact WKB analysis. The mathematical side of resurgence theory has evolved steadily ([Sa16]). Recently, resurgence theory has been at the forefront in such diverse topics in mathematical physics as BPS spectrum ([GMN13]), supersymmetric field theories ([BD16] and references therein), resurgence and quantization as Riemann-Hilbert correspondence ([Ko17]), topological strings and Gromov-Witten theory ([CMS17], [CSV17]), to name a few. Resurgence theory deals with analytic functions which enjoy a certain property of an- alytic continuation ("endlessly continuable functions"), which form an algebra, and which typically appear as Borel transforms of certain divergent series. In his systematic study of the singularities of these functions, their monodromies and Stokes data, ´Ecalle discovered an infinite family of derivations acting on them, which generate a free Lie algebra. Mould calculus first appeared as a convenient combinatorial tool to manipulate these derivations. Later on, ´Ecalle also used mould calculus to study formal classification problems in dy- namical system theory, without any relation to resurgence theory. Mould calculus has since been used in various branches of mathematics, for example in the theory of multiple zeta values ([Ec03], [Sc12], [BE17], [BS17]), in conjugacy problems for formal or analytic differential equations [Me09], [Sa09], in combinatorial Hopf algebras related to symmetric functions [Th11], in conjugacy problems in Lie algebras motivated by classical and quantum dynamics [PS17], in the study of Rayleigh-Schrodinger series [NP18]. In the present paper, we do not assume any familiarity with mould calculus on the part of the reader, and we introduce the most basic ideas about moulds. The BCH formula can be seen as an application, and we hope that readers can find other interesting applications in mathematics or physics. 3 The paper is organized as follows. – Section 2 is a gentle introduction to mould calculus, containing the basic definitions and properties that we will require in our applications. – Section 3 gives short proofs of the BCH theorem (Theorem A) and Dynkin's formula (Theorem B) based on mould calculus. – Section 4 gives a short proof of Kimura's formula (Theorem C) via mould calculus, as well as another derivation of the BCH theorem (Corollary 4.5). – Section 5 indicates how to generalize the previous results to the case of a product of more factors etX1 · · · etXN , with arbitrary N ≥ 2 (Theorems B' and C'). – Section 6 defines a new operation in mould calculus, that we call σ-composition, which allows us to relate the mould used for Dynkin's formula and the one used for Kimura's formula. 2 Mould calculus for pedestrians Throughout the article we use the notation N = {0, 1, 2, . . .}, N∗ = {1, 2, 3, . . .}. In this section, we denote by k a field of characteristic zero (it will be Q in our later applications) and by N a nonempty set (in our applications, it will be either a finite set or N∗). 2.1 The mould algebra Viewing N as an alphabet (the elements of which we call "letters"), we denote by N the corresponding set of "words" (or "strings"): N := {n = n1 · · · nr r ∈ N, n1, . . . , nr ∈ N }. The concatenation law (a1 · · · ar, b1 · · · bs) ∈ N ×N 7→ a1 · · · ar b1 · · · bs ∈ N yields a monoid structure, with the empty word ∅ as unit. Definition 2.1. A k-valued mould on N is a function N → k. The set of all moulds is denoted by kN . Given a mould M , it is customary to denote by M n the value it takes on a word n. Mould multiplication is defined by the formula (M × N )n := X(a,b) such that n=a b 4 M aN b for n ∈ N , (4) for any two moulds M, N ∈ kN . For instance, (M × N )n1n2 = M ∅N n1n2 + M n1 N n2 + M n1n2N ∅. It is immediate to check that kN is an associative k-algebra, noncommutative if N has more than one element, whose unit is the mould 1 defined by 1∅ = 1 and 1n = 0 for n 6= ∅. We say that a mould M has order ≥ p if M n = 0 for each word n of length < p. Clearly, if ord M ≥ p and ord N ≥ q, then ord(M × N ) ≥ p + q. In particular, if M ∅ = 0, then ord M ×k ≥ k for each k ∈ N∗, hence the moulds eM :=Xk∈N 1 k! M ×k and log(1 + M ) := Xk∈N∗ (−1)k−1 k M ×k (5) are well-defined (because, for each n ∈ N , only finitely many terms contribute to (eM )n or (log(1 + M ))n). We thus get mutually inverse bijections { M ∈ kN M ∅ = 0 } exp ⇄ log { M ∈ kN M ∅ = 1 }. 2.2 Comoulds and mould expansions Moulds are meant to provide the coefficients of certain multi-indexed expansions in an associative algebra A. To deal with infinite expansions, we require this A to be a complete filtered associative algebra, i.e. there is an order function ord : A → N ∪ {∞} compatible with sum and product,1 such that every family (Ai)i∈I of A is formally summable provided, for each p ∈ N, all the Ai's have order ≥ p except finitely many of them. See [Sa09] or [PS17] for the details. For the present paper, the reader may think of A = A [[t]] with the order function relative to powers of t, where A is an associative algebra as in the introduction. Assumption 2.2. We suppose that we are given a family (Bn)n∈N in A such that all the Bn's have order ≥ 1 and, for each p ∈ N, only finitely many of them are not of order ≥ p. Definition 2.3. We call associative comould generated by (Bn)n∈N the family (Bn)n∈N defined by B∅ := 1A and Bn1···nr := Bn1 · · · Bnr for all r ≥ 1 and n1, . . . , nr ∈ N . 1We assume ord(A + B) ≥ min{ord A, ord B} and ord(AB) ≥ ord A + ord B for any A, B ∈ A, and ord A = ∞ iff A = 0. 5 Lemma 2.4. The formula M ∈ kN 7→ M B := Xn∈N M nBn ∈ A (6) defines a morphism of associative algebras. Moreover, M ∅ = 0 ⇒ (eM )B = eM B, M ∅ = 1 ⇒ (log M )B = log(M B). (7) Proof. Observe that the family (M nBn)n∈N is formally summable in A thanks to our assumption on the Bn's. The property Ba b = BaBb for all a, b ∈ N entails (M × N )B = (M B)(N B), (8) whence M ×kB = (M B)k for all k ∈ N, and (7) follows. It is the right-hand side in (6) that is called a mould expansion. Example 2.5. Suppose we are given X, Y ∈ A , an associative algebra. Take k = Q, N = Ω := {x, y}, a two-letter alphabet, and A = A [[t]]. We then consider the associative comould generated by Bx := tX, By := tY. (9) Trivially, tX = IxB and tY = IyB, where Ix, Iy ∈ QΩ are defined by x :=( 1 if ω is the one-letter word x 0 else, I ω y :=( 1 if ω is the one-letter word y 0 else. I ω We thus get etX = eIxB, etY = eIy B, and etX etY = SΩB with SΩ := eIx × eIy , log(etX etY ) = TΩB with TΩ := log SΩ. (10) By (4) and (5), we get 1 p!q! if ω is of the form xpyq with p, q ∈ N (11) 0 else, Sω Ω = thus the first part of (10) is just another way of writing etX etY =P tp+q In the general case, retaining from the associative algebra structure of A only the underlying Lie algebra structure, i.e. using only commutators (with the notation adA B = [A, B]), one can define another kind of mould expansion: p!q! X pY q. Definition 2.6. We call Lie comould generated by (Bn)n∈N the family (B[ n ])n∈N of A defined by B[∅] := 0 and B[n1···nr] := adBn1 · · · adBnr−1 Bnr = [Bn1 , [· · · [Bnr−1, Bnr ] · · · ]]. 6 We define the Lie mould expansion associated with a mould M ∈ kN by the formula M [B] := Xn∈N \{∅} 1 r(n) M nB[ n ] ∈ A, (12) where r(n) denotes the length of a word n. Division by r(n) is just a normalization choice whose convenience will appear in Sec- tion 2.3. In Section 3, we will prove the BCH theorem by showing how to pass from the second part of (10) to a Lie mould expansion. 2.3 Symmetrality and alternality One can get a morphism property for Lie mould expansions analogous to (8) by imposing restrictions to the moulds that we use: they must be "alternal". A tightly related notion is that of "symmetral" mould. The definition of both notions relies on word shuffling. Recall that the shuffling of two words a = ω1 · · · ωℓ and b = ωℓ+1 · · · ωr is the set of all the words n which can be obtained by interdigitating the letters of a and those of b while preserving their internal order in a or b, i.e. the words which can be written n = ωτ (1) · · · ωτ (r) with a permutation τ such that2 τ −1(1) < · · · < τ −1(ℓ) and τ −1(ℓ + Definition 2.7. A mould M ∈ kN is said to be alternal if M ∅ = 0 and Xn∈N sh(cid:16)a, b n (cid:17)M n = 0 for any two nonempty words a, b. (13) A mould M ∈ kN is said to be symmetral if M ∅ = 1 and Xn∈N sh(cid:16)a, b n (cid:17)M n = M aM b for any two words a, b. (14) Example 2.8. It is obvious that any mould M whose support is contained in the set of one-letter words (i.e. r(n) 6= 1 ⇒ M n = 0) is alternal. For instance, the moulds Ix and Iy of Example 2.5 are alternal. An elementary example of symmetral mould is E defined by 2Indeed, τ −1(i) is the position in n of ωi, the i-th letter of a b. 7 and b. For instance, if n, m, p, q are four distinct elements of N , 1) < · · · < τ −1(r). We define the shuffling coefficient sh(cid:0) a, b permutations τ , and we set sh(cid:0) a, b nmqpm(cid:17) = 0, mnqmp(cid:17) = 1, We also define, for arbitrary words n and a, sh(cid:0) a, ∅ n (cid:1) to be the number of such n (cid:1) := 0 whenever n does not belong to the shuffling of a nmmqp(cid:17) = 2. sh(cid:16) nmp, mq sh(cid:16) nmp, mq n (cid:1) = 1 if a = n, 0 else. n (cid:1) = sh(cid:0) ∅, a sh(cid:16) nmp, mq En := 1 r(n)! . Indeed, since the total number of words obtained by shuffling of any a, b ∈ N r(a)(cid:1), (counted with multiplicity) is(cid:0)r(a b) n (cid:17)En = sh(cid:16)a, b Xn∈N r(a b)! r(a)!r(b)! · 1 r(a b)! = EaEb. We shall see later that the moulds eIx, eIy and SΩ involved in (10) are symmetral, and that TΩ is alternal. In this paper,3 we are interested in the shuffling coefficients because of the following classical relation between the Lie comould and the associative comould: B[ n ] = X(a,b)∈N ×N (−1)r(b)r(a) sh(cid:16)a, b n (cid:17) B eb a for all n ∈ N , (15) where, for an arbitrary word b = b1 · · · bs, we denote byeb the reversed word: eb = bs · · · b1 (we omit the proof-see [vW66], [Re93], [PS17]). An immediate and useful consequence is Lemma 2.9. If M is an alternal mould, then M [B] = M B, i.e. Xn∈N \{∅} 1 r(n) M nB[ n ] = Xn∈N M nBn. Proof. Putting together (12) and (15), we get M [B] = Pn6=∅Pa,b Now, sh(cid:0) a, b n (cid:1) 6= 0 ⇒ r(n) = r(a) + r(b), hence r(a)+r(b)(cid:18)Xn∈N M [B] = Xr(a)+r(b)≥1 n (cid:17)M n(cid:19)B eb a =Xa6=∅ sh(cid:16)a, b (−1)r(b) r(a) (−1)r(b) r(a) r(n) sh(cid:0) a, b n (cid:1)M n B eb a. M aBa = M B (the internal sum is M a when b = ∅ and it does not contribute when a or b 6= ∅ because of (13), nor when a = ∅ because of the factor r(a)). Any mould expansion associated with an alternal mould thus belongs to the (closure of the) Lie subalgebra of A generated by the Bn's, since it can be rewritten as a Lie mould expansion, involving only commutators of the Bn's. Lemma 2.9 is related to the classical Dynkin-Specht-Wever projection lemma in the context of free Lie algebras (see e.g. [Re93]). One should also mention that the concepts 3In ´Ecalle's work, the initial motivation for the definition of alternality and symmetrality is the situation when A is an algebra of operators (acting on an auxiliary algebra) and each Bn acts as a derivation: in that case, the B[ n ]'s satisfy a modified Leibniz rule which involves the shuffling coefficients, whence it follows that M B is itself a derivation if M is an alternal mould, and an algebra automorphism if M is symmetral. Here we do not assume anything of that kind on A and the Bn's but rather follow the spirit of "Lie mould calculus" as advocated in [PS17]. 8 of symmetrality and alternality are related to certain combinatorial Hopf algebras, as em- phasized by F. Menous in his work on the renormalization theory in perturbative quantum field theory-see e.g. [Me09] and footnote 4. Hopf-algebraic aspects of mould calculus are also touched upon in [Sa09], [PS17] and [NP18]. For our applications, we require a last general result from mould calculus (see e.g. [Sa09] for a proof): Lemma 2.10. • The product of two symmetral moulds is symmetral. • The logarithm of a symmetral mould is alternal. • The exponential of an alternal mould is symmetral. Example 2.11. The mould I defined by I n =( 1 if r(n) = 1 0 else, (16) is alternal (being supported in one-letter words). The symmetral mould E of Example 2.8 is eI . In fact, the set of all symmetral moulds is a group for mould multiplication, the set of all alternal moulds is a Lie algebra for mould commutator, and we get the analogue of (8) for Lie mould expansions: M , N alternal ⇒ [M, N ][B] =(cid:2)M [B], N [B](cid:3). Let us also mention a manifestation of the antipode of the Hopf algebra related to moulds:4 M alternal ⇒ S(M ) = −M, M symmetral ⇒ S(M ) = multiplicative inverse of M , where S(M )n1···nr := (−1)rM nr···n1. All these facts are mentioned in ´Ecalle's works and can be proved by Hopf-algebraic techniques or by direct computation. 4 Denote by k N the linear span of the set of words, i.e. the k-vector space consisting of all formal sums c = P cn n with finitely many nonzero coefficients cn ∈ k. Now, k N is a Hopf algebra if we define multiplication by extending (a, b) 7→ a b := P sh(cid:0) a, b n (cid:1)n by bilinearity, comultiplication by extending a ⊗ b by linearity, and antipode by extending n1 · · · nr 7→ (−1)rnr · · · n1 by linearity (the unit n 7→ P n=a b is ∅ and the counit is c 7→ c∅). The set of moulds can be identified with the set of linear forms on k N , if we identify M ∈ kN with c 7→ P M ncn. The associative algebra structure (4) of kN is then dual to the coalgebra structure of k N , and alternal moulds appear as infinitesimal characters of k N (linear forms M such that M (c c′) = M (c)c′ ∅ + c∅M (c′)) and symmetral moulds as characters (linear forms M such that M (∅) = 1 and M (c c′) = M (c)M (c′)). 9 3 The BCH Theorem and Dynkin's formula Let A be an associative algebra. We now use mould calculus to prove Theorem A. Suppose X, Y ∈ A . Let Ψ = etX etY ∈ A [[t]]. Then log Ψ ∈ Lie(X, Y )[[t]], where Lie(X, Y ) is the Lie subalgebra of A generated by X and Y . Theorem B (Dynkin, [Dy47]). In the above situation, log Ψ =X (−1)k−1 k tσ σ [X p1 Y q1 · · · X pk Y qk ] p1!q1! · · · pk!qk! (17) with summation over all k ∈ N∗ and (p1, q1), · · · , (pk, qk) ∈ N × N \ {(0, 0)}, where σ := p1 + q1 + · · · + pk + qk and [X p1 Y q1 · · · X pk Y qk] := adp1 Y if qk ≥ 1 and adp1 Y · · · adpk X adqk−1 X adq1 X adq1 Y · · · adpk−1 X X if qk = 0. Y Proof of Theorem A. Half of the work has already been done in Example 2.5! With the two- letter alphabet Ω = {x, y}, Bx = tX and By = tY , we have log Ψ = TΩB with TΩ = log SΩ, SΩ = eIx × eIy . The mould SΩ is symmetral, because Ix and Iy are alternal (they are supported in the set of one-letter words) hence eIx and eIy are symmetral by Lemma 2.10 and so is their product. It follows, still by Lemma 2.10, that TΩ is alternal. Lemma 2.9 then shows that log Ψ = TΩ[B]. (18) In particular, being expressed as a Lie mould expansion, log Ψ lies in Lie(X, Y )[[t]]. Proof of theorem B. With the same notation as previously, by definition, hence (18) yields T ω Ω =Xk≥1 log Ψ =Xk≥1 (−1)k−1 k (−1)k−1 k ω=ω1···ωk Xω1,..., ωk∈Ω\{∅} Xω1,..., ωk∈Ω\{∅} Sω1 Ω · · · Sωk Ω for each word ω, 1 r(ω1)+···+r(ωk) Sω1 Ω · · · Sωk Ω B[ω1···ωk]. Inserting (11), we exactly get (17). Mould calculus also allows us to express the inner derivation associated with log Ψ: Corollary 3.1. The inner derivation of A [[t]] associated with Z := log(etX etY ) is X adq1 Y · · · adpk p1!q1! · · · pk!qk! X adq1 Y · · · adpk p1!q1! · · · pk!qk! X adqk [adp1 adp1 tσ σ Y adZ =X (−1)k−1tσ k (19) with summation over all k ∈ N∗ and (p1, q1), · · · , (pk, qk) ∈ N × N \ {(0, 0)}, where σ := p1 + q1 + · · · + pk + qk and with the same bracket notation as in Theorem A. =X (−1)k−1 k X adqk Y ] 10 Proof. Working in the associative algebra End A [[t]] with the comould and the Lie comould associated with Ax := adtX and Ay := adtY , we get adZ = TΩ[A] (i.e. the second part of (19)) from (17) because A[ω] = adB[ω]. Lemma 2.9 then entails adZ = TΩA, i.e. the first part of (19) (which could have been obtained directly from adZ = log(eadtX eadtY ). 4 Alternative formulas for etXetY and its logarithm In this section, we take N := N∗ = {1, 2, 3, . . .} as our alphabet, and k := Q as base field. We now show how to find Kimura's formula (3) from mould calculus. 4.1 An alternative mould expansion for etX etY Theorem C ([Ki17]). Let X, Y ∈ A as in Theorem A. Then Ψ = etX etY can be written Ψ = 1A + ∞Xr=1 ∞Xn1,...,nr=1 tn (n − 1)! 1 nr(nr + nr−1) · · · (nr + · · · + n1) Dn1 · · · Dnr with Dn := adn−1 X (X + Y ) for each n ≥ 1. (20) (21) The rest of section 4.1 is devoted to a new proof of this formula. Lemma 4.1. Ψ = etX etY is the unique element of A [[t]] such that Ψt=0 = 1A , t∂tΨ = DΨ, where D := t etX (X + Y ) e−tX . (22) Proof. The fact that Ψ satisfies (22) is straightforward. On the other hand, if Ψ ∈ A [[t]] is also solution to (22), then ord( Ψ − Ψ) ≥ 1 and it is easy to see that in fact ord( Ψ − Ψ) = ∞ because t∂t( Ψ − Ψ) = D( Ψ − Ψ) and ord D ≥ 1; hence Ψ − Ψ = 0. Let N := N∗ and consider the associative comould associated with the family (Dn)n∈N defined by (21). We have D = Xn∈N Dn = ID, (23) where D in the left-hand side is the element of A [[t]] defined in (22), while the right-hand side is the mould expansion associated with the mould I defined by (16). The proof of (23) is essentially the Hadamard lemma: adX can be written LX − RX, where LX and RX are the operators of left-multiplication and right-multiplication by X and they commute, hence et adX = et(LX −RX ) = etLX e−tRX , and etLX and e−tRX are the operators of left- multiplication and right-multiplication by etX and e−tX , whence et adX A = etX A e−tX for any A ∈ A [[t]]. (24) In particular, etX (X + Y )e−tX =Pn∈N tn−1 (n−1)! adn−1 X (X + Y ). 11 Lemma 4.2. For any mould S ∈ QN , where ∇S is the mould defined by t∂t(SD) = (∇S)D, (∇S)n1···nr := (n1 + · · · + nr)Sn1···nr for each word n1 · · · nr ∈ N . Proof. Obvious, since Dn ∈ tnA for each n ∈ N . Lemma 2.4, formula (23) and Lemma 4.2 inspire us to look for a solution to (22) in the form of a mould expansion: Ψ = SD will be solution to (22) if S ∈ QN is solution to the mould equation S∅ = 1, ∇S = I × S (25) (indeed: we have (∇S)D = t∂tΨ on the one hand, and (I × S)D = (ID)(SD) = DΨ on the other hand, and S∅ = 1 ensures ord(Ψ − 1A ) ≥ 1 because ord Dn ≥ 1 for all nonempty word n). Now the second part of (25) is equivalent to (n1 + · · · + nr)Sn1···nr = Sn2···nr for each nonempty word n1 · · · nr ∈ N , (26) thus the mould equation (25) has a unique solution: the mould SN ∈ QN defined by Sn1···nr N := 1 nr(nr + nr−1) · · · (nr + · · · + n1) for each n1 · · · nr ∈ N . (27) In conclusion, SN is a solution to (25), thus SN D is a solution to (22), thus SN D = Ψ = etX etY (28) and formula (20) is proved. Remark 4.3. For any alphabet N and base field k, an arbitrary function φ : N → k gives rise to a linear operator ∇φ : kN → kN defined by the formula (∇φ M )n1···nr =(cid:0)φ(n1) + · · · + φ(nr)(cid:1)M n1···nr (29) (with the convention that an empty sum is 0). The reader can check that ∇φ is a mould derivation, i.e. it satifies the Leibniz rule ∇φ(M × N ) = (∇φ M ) × N + M × ∇φ N . Here, we have used the mould derivation associated with the inclusion map N∗ ֒→ Q. 4.2 An alternative Lie mould expansion for log(etX etY ) The mould SN that we have just constructed happens to be a very common and useful object of mould calculus (see e.g. [Ec81] or [Sa09, §13]). It is well-known that it is symmetral; we give the proof for the sake of completeness. 12 Lemma 4.4. The mould SN defined by the formula (27) is symmetral. Proof. We prove the property (14) for M = SN by induction on r(a) + r(b). The property holds when a = ∅ or b = ∅ because S∅ N = 1. In particular it holds when r(a) + r(b) = 0. Suppose now that a and b are arbitrary nonempty words. Using the notation n := n1 + · · · + nr, 'n := n2 · · · nr for any nonempty word n1 · · · nr, we multiply the right-hand side of (14) by a + b: we get (a+b)Sa N Sb N = aSa N Sb N +bSa N Sb N = S'a N Sb N +Sa N S'b N =Xc c (cid:1)Sc sh(cid:0) 'a, b N +Xc c (cid:1)Sc sh(cid:0) a, 'b N , (30) where we have used (26) and the induction hypothesis. On the other hand, multiplying the left-hand side of (14) by a + b, we get (a + b)Xn sh(cid:0) a, b n (cid:1)Sn N =Xn n sh(cid:0) a, b n (cid:1)Sn N =Xn sh(cid:0) a, b n (cid:1)S'n N (31) (using (26) again). The last sum can be split into two according to the first letter of n, n (cid:1) which must come either from the first letter of a or from the first letter of b for sh(cid:0) a, b n (cid:1) = sh(cid:0) a, 'b c (cid:1), to be nonzero: either n = a1c and sh(cid:0) a, b c (cid:1), or n = b1c and sh(cid:0) a, b therefore (30) and (31) coincide, which proves (14) with M = SN . n (cid:1) = sh(cid:0) 'a, b We are now in a position to obtain a new formula for log Ψ, on which its Lie character is manifest-the new formula thus contains the BCH theorem: Corollary 4.5. Let TN := log SN ∈ QN . Then, with the notation of Theorem C, we have log Ψ = TN [D], i.e. log(etX etY ) =Xr≥1 ∞Xn1,...,nr=1 1 r T n1···nr N [Dn1 , [· · · [Dnr−1 , Dnr ] · · · ]] ∈ Lie(X, Y )[[t]]. Proof. From Theorem C and Lemma 2.4 we deduce log Ψ = log(SN D) = TN D. (32) By Lemmas 2.10 and 4.4, TN is alternal. We conclude by Lemma 2.9. From the definition TN = words of small length: (−1)k−1 k ∞Pk=1 (SN − 1)×k, we can write down the coefficients for T n1 = Sn1 = 1 n1 1 T n1n2 = Sn1n2 − 2 Sn1 Sn2 = n1 − n2 2n1n2(n1 + n2) T n1n2n3 = Sn1n2n3 − 1 2 T n1n2n3n4 = Sn1n2n3n4 − Sn1n2Sn3 − 1 2 Sn1Sn2n3 + Sn1 Sn2 Sn3 1 3 1 2 Sn1Sn2n3n4 − 1 2 Sn1n2Sn3n4 − Sn1n2n3Sn4 1 2 + 1 3 Sn1Sn2 Sn3n4 + 1 3 Sn1 Sn2n3 Sn4 + 1 3 Sn1n2 Sn3 Sn4 − Sn1Sn2 Sn3Sn4 1 4 13 (omitting the subscript N to lighten notation). The low powers of t in log Ψ = TN [D] can then be extracted from the Lie mould expansion and we recover the classical BCH series: · · · · · · log Ψ = + T n1 Dn1 + ∞Xn1=1 ∞Xn1,n2,n3,n4=1 ∞Xn1,n2=1 1 2 T n1n2[Dn1 , Dn2 ] + ∞Xn1,n2,n3=1 1 3 T n1n2n3[Dn1 , [Dn2 , Dn3]] 1 4 T n1n2n3n4[Dn1 , [Dn2 , [Dn3 , Dn4]]] + · · · = t(X + Y ) + t2 2 [X, Y ] + − t3 12 ([(X + Y ), [X, Y ]]) − t3 3! t4 24 [X, [X, Y ]] + t4 4! [X, [X, [X, Y ]]] + t5 5! [X, [X, [X, [X, Y ]]]] + · · · ([(X + Y ), [X, [X, Y ]]]) − t5 120 [[X, Y ], [X, [X, Y ]]] [(X + Y ), [(X + Y ), [X, [X, Y ]]]] − − t5 80 [(X + Y ), [X, [X, [X, Y ]]]] + · · · t5 240 [[X, Y ], [(X + Y ), [X, Y ]]] + · · · [(X + Y ), [(X + Y ), [(X + Y ), [X, Y ]]]] + · · · + + t5 720 t5 720 = t(X + Y ) + t2 2 [X, Y ] + t3 12 ([X, [X, Y ]] + [Y, [Y, X]]) − − + t5 720 t5 360 [X, [X, [X, [X, Y ]]]] − [Y, [X, [X, [X, Y ]]]] + t5 720 t5 120 [Y, [Y, [Y, [Y, X]]]] + [Y, [X, [Y, [X, Y ]]]] + t4 24 [Y, [X, [X, Y ]]] [X, [Y, [Y, [Y, X]]]] [X, [Y, [X, [Y, X]]]] + · · · . t5 360 t5 120 5 Generalization to an arbitrary number of factors One of the merits of the mould calculus approach is that the formulas are easily generalized to the case of Ψ = etX1 · · · etXN ∈ A [[t]], where A us an associative algebra and X1, . . . , XN ∈ A for some N ≥ 2. 5.1 Mould expansion of the first kind Theorem B'. Let NN ∗ := { p ∈ NN p1 + · · · + pN ≥ 1 }. We have log Ψ =X (−1)k−1 k tσ σ (cid:2)X p1 1 · · · X 1 1! · · · p1 p1 p1 N · · · X N N ! · · · pk pk 1 · · · X 1 1! · · · pk N ! with summation over all k ∈ N∗ and p1, · · · , pk ∈ NN ∗ , where σ := denote nested commutators as before. 14 pk N N (cid:3) NPj=1 kPi=1 pi j and the bracket Proof. Let Ω := {x1, . . . , xN } be an N -element set. We consider the associative comould generated by the family Bx1 := tX1, . . . , BxN := tXN ∈ A [[t]]. (33) We can write tX1 = I1B, . . . , tXN = IN B, with moulds I1, . . . , IN ∈ QΩ defined by j :=(1 if ω is the one-letter word xj 0 else I ω for j = 1, . . . , N . Hence Ψ = SΩB with SΩ := eI1 × · · · × eIN , log Ψ = TΩB with SΩ := log SΩ. (34) The moulds I1, . . . , IN are alternal (being supported in one-letter words), hence Lemma 2.10 entails that their exponentials are symmetral, and also SΩ, while TΩ is alternal. We deduce that (−1)k−1 k Xω1,..., ωk∈Ω\{∅} 1 r(ω1)+···+r(ωk) Sω1 Ω · · · Sωk Ω B[ω1···ωk]. The conclusion stems from the fact that log Ψ = TΩ[B] =Xk≥1 Ω = p1! · · · pN ! Sω 1 if ω is of the form xp1 1 · · · xpN N with (p1, . . . , pN ) ∈ NN else. 0 5.2 Mould expansion of the second kind Theorem C'. In the above situation, Ψ = etX1 · · · etXN can also be written Ψ = 1A + ∞Xr=1 with Dn := tn ∞Xn1,...,nr=1 NXj=1 1 nr(nr + nr−1) · · · (nr + · · · + n1) Xm1,...,mj−1∈N m1+···+mj−1=n−1 · · · admj−1 adm1 X1 Xj−1 m1! · · · mj−1! Dn1 · · · Dnr (35) Xj for each n ≥ 1. (36) Note that formula (35) involves exactly the same rational coefficients as in the case N = 2. The only difference in the formula is that the Dn's of (21) have been generalized to the Dn's which are defined in (36) and read Dn := t(X1 + · · · + XN ) tn adn−1 (n − 1)! X1 X2 + · · · + tn for n = 1, XN for n > 1. (37) Xm1+···+mN−1=n−1 · · · admN−1 adm1 XN−1 X1 m1! · · · mN −1! 15 Proof. We have Ψt=0 = 1A and t∂tΨ = tX1 etX1 · · · etXN + t etX1 X2 etX2 · · · etXN + · · · + t etX1 · · · etXN−1 XN etXN = DΨ, where D := t NXj=1 AdetX1 · · · AdetXj−1 Xj with the notation AdE A = EAE−1 for any A ∈ A [[t]] whenever E is an invertible element of A [[t]]. Moreover, we observe that there is no other solution in A [[t]] to the system Ψt=0 = 1A , t∂tΨ = DΨ, (38) because ord D ≥ 1. j=1 eadtX1 · · · eadtXj−1 Xj =Pn≥1 Thanks to (24), we compute D = tPN Dn. Let us thus take N = N∗ as alphabet and consider the associative comould generated by (Dn)n∈N , so that D can be rewritten as the mould expansion ID, with the same mould as in (16). Lemmas 2.4 and 4.2 show that a mould expansion Ψ = SD is solution to (38) if S ∈ QN is solution to the mould equation (25) (indeed: (∇S)D = t∂tΨ on the one hand, and (I × S)D = (ID)(SD) = DΨ on the other hand, and S∅ = 1 ensures ord(Ψ − 1A ) ≥ 1 because ord Dn ≥ 1 for all nonempty word n). But we already know that S = SN defined by (27) is the unique solution to (25), hence Ψ = SN D, (39) which is equivalent to (35). Notice that, in view of Section 4.2, the mould SN is symmetral, the mould TN = log SN is alternal, whence i.e. log Ψ = TN D = TN [D], (40) log(etX1 · · · etXN ) =Xr≥1 ∞Xn1,...,nr=1 1 r T n1···nr N [Dn1, [· · · [Dnr−1, Dnr ] · · · ]] which thus belongs to Lie(X1, . . . , XN )[[t]], in accordance with the BCH theorem. 6 Relation between the two kinds of mould expansion In our application to products of two or more exponentials, we have seen two different kinds of mould expansion. The first kind involves an N -element alphabet Ω := {x1, . . . , xN } and the comould generated by the family (Bω)ω∈Ω defined by (33). For the second one, the alphabet is N := N∗ and the comould is generated by the family (Dn)n∈N which is defined 16 by (37) and boils down to the Dn's of (21) when N = 2. A natural question is: What is the relation between both kinds of mould expansion? i.e. can one pass from the representation of the product Ψ as SΩB in (34) to its representation as SN D in (39), or from log Ψ = TΩB in (34) to log Ψ = TN D in (40)? In this section, we will answer this question by defining a new operation on moulds, which allows one to pass directly from SN to SΩ, or from TN to TΩ. We take N = 2 for simplicity but the generalization to arbitrary N is easy. We start by giving a mould expansion of the first kind for the Dn's themselves. Lemma 6.1. Let Ω := {x, y}. The formula ω ∈ Ω 7→ U ω := 1 if ω = x (−1)q p!q! if ω is of the form xpyxq for some p, q ∈ N (41) 0 else  defines an alternal mould U ∈ QΩ such that Dn = UnB for each n ∈ N∗, (42) where the left-hand side is defined by (21) and the right-hand side is the mould expansion (for the comould generated by (9)) associated with Un := restriction of U to the words of length n. (n−1)! adn−1 Ix Proof. In view of (8), we have adM B(N B) = [M B, N B] = [M, N ]B = (adM N )B for any M, N ∈ QΩ, hence (21) can be rewritten as Dn = 1 Un := 1 (Ix + Iy). Since Ix and Iy are alternal and the set of all alternal moulds is stable under mould commutator (as mentioned at the end of Section 2.3), we see that this mould Un is alternal. Since the support of Un is contained in the set of words of IxB(cid:0)(Ix + Iy)B(cid:1) = UnB with length n, the formula U :=Pn≥1 Un makes sense and defines an alternal mould (and Un now appears as the restriction of this U to the set of words of length n). There only remains to check (41). (n−1)! adn−1 Now, adIx = L − R, where L and R are the operators of left-multiplication and right- multiplication by Ix, which commute, hence the binomial theorem yields Un = Xp+q=n−1 (−1)q p!q! LpRq(Ix + Iy) = Xp+q=n−1 (−1)q p!q! I ×p x × (Ix + Iy) × I ×q x , n = 1 if ω = x and n = 1, (−1)q p!q! i.e. U ω if ω is of the form xpyxq for some p, q ∈ N such that p + q = n − 1 (in which case p and q are uniquely determined), and 0 else. Our U thus coincides with the mould defined by (41). 17 In fact the proof just given shows that U = eadIx (Ix + Iy) = eIx × (Ix + Iy) × e−Ix. (43) This mould will allow us to relate D-mould expansions and B-mould expansions: Theorem D. Let N := N∗. Define a linear map M ∈ QN 7→ M ⊙ U ∈ QΩ by the formulas (M ⊙ U )∅ := M ∅, (M ⊙ U )ω :=Xs≥1 Xω=ω1··· ωs ω1,..., ωs∈Ω\{∅} Then M r(ω1)···r(ωs)U ω1 · · · U ωs for ω ∈ Ω \ {∅}. (45) (44) M D = (M ⊙ U )B for any M ∈ QN . Recall that r : Ω → N∗ = N is our notation for the length function. In (45), r(ω1) · · · r(ωs) is to be understood as a word of length s of N (and the sum is finite because the words ωj are nonempty, hence s ≤ r(ω)). Proof. By direct computation, using (42) to express Dn1···ns = Dn1 · · · Dns, M D = Xn∈N M nDn = M ∅ 1A +Xs≥1 Xn1,...,ns∈N M n1···ns = M ∅ 1A +Xs≥1 Xn1,...,ns∈N = M ∅ 1A +Xs≥1 Xω1,..., ωs∈Ω\{∅} = M ∅ 1A + Xω∈Ω\{∅} Xs≥1, ω1,... ωs∈Ω\{∅} ω=ω1··· ωs M n1···ns Xω1∈Ω Xω1,..., ωs∈Ω r(ω1)=n1 r(ω1)=n1,...,r(ωs)=ns U ω1 Bω1 · · · Xωs∈Ω r(ωs)=n1 U ωs Bωs · · · U ωs Bω1··· ωs U ω1 M r(ω1)···r(ωs)U ω1 · · · U ωs Bω1··· ωs M r(ω1)···r(ωs)U ω1 · · · U ωs!Bω = (M ⊙ U )B. The relations SN D = SΩB (which coincides with Ψ according to (10) and (28)) and TN D = TΩB (which coincides with log Ψ according to (28) and (32)) now appear as a manifestation of Theorem D and the following Theorem E. SN ⊙ U = SΩ, TN ⊙ U = TΩ. The proof of Theorem E is given at the end of this section. Our definition (44)–(45) of the mould operation '⊙' is a variant of ´Ecalle's mould com- position '◦' which is defined for any alphabet that is a commutative semigroup ([Ec84], [Sa09], [FFM17]). Here is a definition which encompasses both operations: 18 Definition 6.2. Given two alphabets Ω and N , and a map σ : Ω \ {∅} → N , we define the σ-composition (M, U ) ∈ kN × kΩ 7→ M ◦σ U ∈ kΩ by the formulas (M ◦σ U )∅ := M ∅, (M ◦σ U )ω :=Xs≥1 Xω=ω1··· ωs ω1,..., ωs∈Ω\{∅} M σ(ω1)···σ(ωs)U ω1 · · · U ωs for ω ∈ Ω \ {∅}. (46) (47) Thus, we recover the '⊙' composition in the special case when N = N∗ and σ(ω) = r(ω) (with arbitrary Ω), and ´Ecalle's composition '◦' when N = Ω is a commutative semigroup and σ(n1 · · · nr) = n1 + · · · + nr for any nonempty word of N . Some classical properties of the latter operation can be generalized as follows: (i) (M ◦σ U ) × (N ◦σ U ) = (M × N ) ◦σ U . (ii) eM ◦σ U = (eM ) ◦σ U if M ∅ = 0, log(M ◦σ U ) = (log M ) ◦σ U if M ∅ = 1. (iii) I ◦σ U = U − U ∅1Ω, where I is defined by (16) and 1Ω is the unit of kΩ. (iv) Denote by ιΩ : Ω ֒−→ Ω \ {∅} the inclusion map. If φ : N → k is a function such that φ ◦ σ maps the concatenation in Ω to the addition in k, then (∇φM ) ◦σ U = ∇ψ(M ◦σ U ) for all M ∈ kN , with ψ := φ ◦ σ ◦ ιΩ, (48) where ∇φ and ∇ψ are the mould derivations defined by (29). (v) If U is alternal and σ(ω1 · · · ωr) = σ(ωτ (1) · · · ωτ (r)) for every permutation τ and for any ω1, . . . , ωr ∈ Ω, then M alternal ⇒ M ◦σ U alternal, M symmetral ⇒ M ◦σ U symmetral. (vi) Suppose (Bω)ω∈Ω satisfies Assumption 2.2. Then the formula Dn := Pω∈σ−1(n) defines a family (Dn)n∈N which also satisfies Assumption 2.2, and U ωBω M D = (M ◦σ U )B for any M ∈ kN . (vii) Suppose that τ : N \ {∅} → M is a map such that ψ := τ ◦ ιN ◦ σ satisfies ψ(ω1 · · · ωs) = τ (σ(ω1) · · · σ(ωs)) for any s ≥ 1 and ω1, . . . ωs ∈ Ω \ {∅}, then M ◦ψ (N ◦σ U ) = (M ◦τ N ) ◦σ U for any M ∈ kM, N ∈ kN , U ∈ kΩ. 19 (The proof of these properties is left to the reader.) Proof of Theorem E. Here Ω = {x, y}, N = N∗ and σ = r : Ω → N is word length. Since TN = log SN and TΩ = log SΩ, in view of (ii) it is sufficient to prove SN ◦σ U = SΩ. As noticed in Section 4.1, SN is a solution in QN to equation (25), which involves ∇ = ∇φ, with the notation φ : N ֒−→ Q for the inclusion map. Taking '⊙U ' of both sides of (25), we get (∇φSN ) ◦σ U = (I × SN ) ◦σ U. (49) We compute the left-hand side by means of (iv): φ◦σ(ω) = r(ω) is word length, in particular it maps concatenation in Ω to addition in Q, and φ ◦ σ ◦ ιΩ ≡ 1, hence the left-hand side is ∇1(SN ◦σ U ). Note that the mould derivation ∇1 is given by (∇1M )ω = r(ω)M ω. By (i) and (iii), the right-hand side of (49) is (I ◦σ U ) × (SN ◦σ U ) = U × (SN ◦σ U ). Therefore, SN ◦σ U is a solution to M ∅ = 1, ∇1M = U × M. (50) It is easy to see that (50) has no other solution in QΩ. On the other hand, by (10), SΩ = eIx × eIy , and ∇1 is a mould derivation which satisfies ∇1Ix = Ix and ∇1Iy = Iy, thus ∇1SΩ = ∇1(eIx) × eIy + eIx × ∇1(eIy ) = Ix × eIx × eIy + eIx × Iy × eIy = (Ix + eIx × Iy × e−Ix) × eIx × eIy = U × SΩ by (43). Therefore SΩ is a solution to (50), hence it must coincide with SN ◦σ U . Acknowledgments D.S. and Y.L. thank the Centro Di Ricerca Matematica Ennio De Giorgi and the Scuola Normale Superiore di Pisa for their kind hospitality, during which this work was completed. 20 References [BS17] [BD16] [BF12] [BE17] S. Baumard and L. Schneps, On the derivation representation of the fundamental Lie algebra of mixed elliptic motives, Annales Math´ematiques du Qu´ebec 41, 1 (2017), 43–62. A. Behtash, G. V. Dunne, T. Schafer, T. Sulejmanpasic and M. Unsal, Com- plexified path integrals, exact saddles, and supersymmetry, Phys. Rev. Lett. 116 (2016),011601. A. Bonfiglioli and R. Fulci, Topics in Noncommutative Algebra-The Theorem of Campbell, Baker, Hausdorff and Dynkin, Lecture Notes in Mathematics, 2034. Springer, Heidelberg, (2012). xxii+539 pp. O. Bouillot and J. ´Ecalle, Invariants of identity-tangent diffeomorphisms ex- panded as series of multitangents and multizetas, in Resurgence, physics and numbers, 109–232 (2017), CRM Series, 20, Ed. Norm., Pisa. [CSV17] R. Couso-Santamar´ıa, R. Schiappa and R.Vaz, On asymptotices and resurgent structures of enumerative Gromov-Witten invariants, Commun. Num. Theor. Phys. 11 (2017) 707–790, [1605.07473]. [CMS17] R. Couso-Santamar´ıa1, M. Marino and R. Schiappa, Resurgence matches quan- tization, J. Phys. A: Math. Theor. 50 (2017) 145402. [Dy47] E. B. Dynkin, Calculation of the coefficients in the Campbell-Hausdorff formula (Russian), Dokl. Akad. Nauk SSSR (N.S.), 57, (1947), 323–326. [DDP93] H. Dillinger, E. Delabaere and F. Pham, R´esurgence de Voros et p´eriodes des [Ec81] [Ec84] [Ec92] [Ec03] courbes hyperelliptiques, Annales de l'institut Fourier 43, 1 (1993), 163–199. J. ´Ecalle, Les fonctions r´esurgentes, Publ. Math. d'Orsay [Vol.1: 81-05, Vol.2: 81-06,Vol.3: 85-05], (1981,1985). J. ´Ecalle, Cinq applications des fonctions r´esurgentes, Publ. Math. d'Orsay, 84– 62, (1984). J. ´Ecalle, Introduction aux fonctions analysables et preuve constructive de la conjecture de Dulac, Actualit´es Math., Hermann, Paris (1992). J. ´Ecalle, ARI/GARI, la dimorphie et l'arithm´etique des multizetas: un premier bilan, Journal de Th´eorie des Nombres de Bordeaux 15, 2 (2003), 411–478. [FFM17] F. Fauvet, L. Foissy, and D. Manchon, The Hopf algebra of finite topologies and mould composition, Annales de l'Institut Fourier 67, 3 (2017), 911–945. [GMN13] D. Gaiotto, G. W. Moore and A. Neitzke, Wall-crossing, Hitchin Systems, and the WKB Approximation, Adv. in Math. 234, (2013), 239–403. [Ki17] T. Kimura, Explicit description of the Zassenhaus formula, Theor.Exp.Phys., (2017), 041A03. 21 [Ko17] M. Kontsevich, Resurgence and Quantization, Course given at IHES, Paris in April, 2017. [Ma15] M. Matone, An algorithm for the Baker-Campbell-Hausdorff formula, J. High Energy Phys. 05 (2015) 113, [arXiv:1502.06589]. [Me09] F. Menous, Formal differential equations and renormalization, in Renormal- ization and Galois theories, A. Connes, F. Fauvet, J.-P. Ramis (eds.), IRMA Lect.Math.Theor.Phys., 15 (2009), 229–246. [NP18] J.-C. Novelli, T. Paul, D. Sauzin, J.-Y. Thibon, "Rayleigh-Schrodinger series and Birkhoff decomposition", Letters in Mathematical Physics 108 (2018), 18 p. https://doi.org/10.1007/s11005-017-1040-1 [PS17] [Re93] [Sa08] [Sa09] [Sa16] [Sc12] [Th11] T. Paul and D. Sauzin, Normalization in Lie algebras via mould calculus and applications, Regular and Chaotic Dynamics 22, 6 (2017), 616–649. C. Reutenauer, Free Lie algebras, London Mathematical Society Monographs 7, Clarendon Press, Oxford University Press, New York 1993, xviii+269 pp. D. Sauzin, Initiation to mould calculus through the example of saddle-node sin- gularities, Rev. Semin. Iberoam. Mat. 3 (2008), no. 5-6, 147–160. D. Sauzin, Mould expansions for the saddle-node and resurgence monomials, in Renormalization and Galois theories, A. Connes, F. Fauvet, J.-P. Ramis (eds.), IRMA Lectures in Mathematics and Theoretical Physics 15, Zurich: European Mathematical Society, 83–163 (2009). D. Sauzin, Introduction to 1-summability and resurgence, in Divergent se- ries, summability and resurgence. I. Monodromy and resurgence, C. Mitschi, D. Sauzin. Lecture Notes in Mathematics, 2153. Springer, 2016. xxi+298 pp. L. Schneps, Double shuffle and Kashiwara-Vergne Lie algebras, Journal of Alge- bra 367 (2012), 54–74. J.-Y. Thibon, Noncommutative symmetric functions and combinatorial Hopf al- gebras, in: Asymptotics in dynamics, geometry and PDEs; generalized Borel summation. Vol. I, (2011) 219–258, CRM Series, 12, Ed. Norm., Pisa. [Vo83] A. Voros, The return of the quartic oscillator. The complex WKB method, An- nales de l'I. H. P., section A, tome 39, n◦ 3 (1983) 211–338. [vW66] W. von Waldenfels, "Zur Charakterisierung Liescher Elemente in freien Alge- bren," Arch. Math. (Basel) 17, 44–48 (1966). 22
1805.07812	3	1805	2018-12-12T20:01:53	Epsilon-strongly groupoid graded rings, the Picard inverse category and cohomology	[ "math.RA" ]	We introduce the class of partially invertible modules and show that it is an inverse category which we call the Picard inverse category. We use this category to generalize the classical construction of crossed products to, what we call, generalized epsilon-crossed products and show that these coincide with the class of epsilon-strongly groupoid graded rings. We then use generalized epsilon-crossed groupoid products to obtain a generalization, from the group graded situation to the groupoid graded case, of the bijection from a certain second cohomology group, defined by the grading and the functor from the groupoid in question to the Picard inverse category, to the collection of equivalence classes of rings epsilon-strongly graded by the groupoid.	math.RA	math	EPSILON-STRONGLY GROUPOID GRADED RINGS, THE PICARD INVERSE CATEGORY AND COHOMOLOGY PATRIK NYSTEDT Department of Engineering Science, University West, SE-46186 Trollhattan, Sweden JOHAN OINERT Department of Mathematics and Natural Sciences, Blekinge Institute of Technology, SE-37179 Karlskrona, Sweden H´ECTOR PINEDO Escuela de Matem´aticas, Universidad Industrial de Santander, Carrera 27 Calle 9, Edificio Camilo Torres Apartado de correos 678, Bucaramanga, Colombia Abstract. We introduce the class of partially invertible modules and show that it is an inverse category which we call the Picard inverse cat- egory. We use this category to generalize the classical construction of crossed products to, what we call, generalized epsilon-crossed products and show that these coincide with the class of epsilon-strongly groupoid graded rings. We then use generalized epsilon-crossed groupoid prod- ucts to obtain a generalization, from the group graded situation to the groupoid graded case, of the bijection from a certain second cohomol- ogy group, defined by the grading and the functor from the groupoid in question to the Picard inverse category, to the collection of equivalence classes of rings epsilon-strongly graded by the groupoid. E-mail addresses: [email protected]; [email protected]; [email protected]. 2010 Mathematics Subject Classification. 16W50, 16E99, 16D99, 14C22. Key words and phrases. groupoid graded ring, epsilon-strongly graded ring, partially invertible module, Picard groupoid, Picard inverse category, cohomology, Leavitt path algebra, partial skew groupoid ring. 1 2 EPSILON-STRONGLY GROUPOID GRADED RINGS 1. Introduction Almost 40 years ago Nastasescu and Van Oystaeyen [15] proved an ele- gant result that relates the collection of group graded equivalence classes of strongly graded rings to certain second cohomology groups. Namely, let S be a ring. We always assume that S is associative and equipped with a multi- plicative identity 1S. The ring S is called graded by a group G (or G-graded) if there is a set {Sg}g∈G of additive subgroups of S such that S = ⊕g∈GSg, and for all g, h ∈ G the inclusion SgSh ⊆ Sgh holds. The ring S is called strongly graded if for all g, h ∈ G the equality SgSh = Sgh holds. Given two G-graded rings S and T , a ring homomorphism f : S → T is called graded if for all g, h ∈ G the inclusion f (Sg) ⊆ Tg holds. Now, by [15, Proposition I.3.13] the collection of strongly graded rings can be parametrized by the so called generalized crossed products (A, F, f ), for rings A and group ho- momorphisms F from G to the Picard group Pic(A) of A. Indeed, for each g ∈ G we put F (g) = [Pg] (the A-bimodule isomorphism class of Pg) and we assume that F (e) = [A], where e denotes the identity element of G. The map f is a factor set associated with F . By this we mean a collection of A- bimodule isomorphisms fg,h : Pg ⊗A Ph → Pgh chosen so that the following diagram commutes Pg ⊗A Ph ⊗A Pp idPg ⊗fh,p −−−−−−→ Pg ⊗A Php fg,h⊗idPpy Pgh ⊗A Pp fgh,p−−−−→ fg,hp y Pghp for all g, h, p ∈ G. The multiplication in (A, F, f ) = ⊕g∈GPg is defined by the biadditive extension of the relation x·y = fg,h(x⊗y) for all x ∈ Pg, all y ∈ Ph and all g, h ∈ G. Let (A, F ) denote the collection of all generalized crossed products of the form (A, F, f ), where f is a factor set associated with F . Given D and D′ in (A, F ) put D ≈ D′ if there is an isomorphism of graded rings D → D′ which is simultaneously an A-bimodule isomorphism. Then ≈ is an equivalence relation on (A, F ) and we can define C(A, F ) = (A, F )/ ≈. Let U (Z(A)) denote the unit group of the center Z(A) = {a ∈ A ∀b ∈ A, ab = ba} of A. The classical cohomology groups H n(G, U (Z(A))), for n ≥ 0, can then be defined, and in particular the corresponding second cohomology group. Theorem 1 (Nastasescu and Van Oystaeyen [15]). If A is a ring, F : G → Pic(A) is a group homomorphism and f is a factor set associated with F , then the map H 2(G, U (Z(A))) → C(A, F ), defined by [q] 7→ qf , is bijective. Many natural examples of rings, such as rings of matrices, crossed product algebras defined by separable extensions and groupoid rings are not, in a natural way, graded by groups, but instead by groupoids (see e.g. [13, 14] or Section 6 of the present article). Let G be a groupoid, that is, a category where each morphism is an isomorphism. The classes of objects EPSILON-STRONGLY GROUPOID GRADED RINGS 3 and morphisms of G are denoted by G0 and G1, respectively. If g ∈ G1, then the domain and codomain of g is denoted by d(g) and c(g), respectively. We let G2 denote the set of all pairs (g, h) ∈ G1 × G1 that are composable, that is, such that d(g) = c(h). From now on, assume that G is small, that is, such that G1 is a set, and let S be a ring which is graded by G. Recall from [13, 14] that this means that there is a set {Sg}g∈G1 of additive subgroups of S such that S = ⊕g∈G1Sg and, for all g, h ∈ G1, SgSh ⊆ Sgh, if (g, h) ∈ G2, and SgSh = {0}, otherwise. In that case, S is called strongly graded if for all (g, h) ∈ G2 the equality SgSh = Sgh holds. By [14, Proposition 4.1] the collection of strongly groupoid graded rings can be parametrized by generalized groupoid crossed products (A, F, f ), for rings A and groupoid homomorphisms F from G to the Picard groupoid PIC. Here A denotes the direct product of the rings {F (e)}e∈G0 . Westman [20] (see also [19]) has developed a cohomology theory for groupoids which extends the classical group cohomology theory. In particular, the corresponding second cohomology group H 2(G, Z(A)) can be defined. The set C(A, F ) is defined analogously to the group graded case. Theorem 2 (Lundstrom [14]). If G is a groupoid, F : G → PIC is a functor and f is a factor set associated with F , then the map H 2(G, U (Z(A))) → C(A, F ), defined by [q] 7→ qf , is bijective. In [17] the authors of the present article introduced the class of epsilon- strongly group-graded rings. This class properly contains both the class of strongly graded rings and the class of unital partial crossed products. The main goal of the present article is to show a simultaneous generalization (see Theorem 3) of Theorem 1 and Theorem 2 that holds for an even wider family of rings, namely the class of epsilon-strongly groupoid graded rings. Indeed, let S be a ring which is graded by a small groupoid G. We say that S is epsilon-strongly graded by G if for each g ∈ G1, SgSg−1 is a unital ideal of Sc(g) such that for all (g, h) ∈ G2 the equalities SgSh = SgSg−1Sgh = SghSh−1Sh hold. Let ǫg denote the multiplicative identity element of SgSg−1 and put R = ⊕e∈G0Se. In this context it turns out that the multiplication map Sg ⊗R Sh → Sgh, for (g, h) ∈ G2, is injective with image equal to ǫgSgh. In particular this implies that the R-isomorphism classes of the modules [Sg] do not form a groupoid. Instead they form an inverse category PICcat which we call the Picard inverse category. The collection of epsilon-strongly groupoid graded rings can then be parametrized by generalized groupoid epsilon-crossed products (A, F, f ), for rings A and partial functors of inverse categories F : G → PICcat. Here A denotes the direct product of the rings {F (e)}e∈G0 and f is a partial factor set. On the other hand, in [9] the concept of a partial action was introduced as an efficient tool to study C ∗-algebras, permitting to characterize vari- ous important classes of them as crossed products by partial actions. The study of partial actions and partial representations of groups on algebras was initiated in [4], and extended to the groupoid situation in [2]. Recently, 4 EPSILON-STRONGLY GROUPOID GRADED RINGS Dokuchaev and Khrypchenko [5] have developed a cohomology theory for partial actions of groups on monoids which we extend to the groupoid set- ting. In particular, we define the corresponding second cohomology group H 2(G, Z(A)). The set C(A, F ) is defined analogously to the group graded case, and we obtain the following theorem, which is the main result of this article. Theorem 3. If G is a groupoid, F : G → PICcat is a partial functor of inverse categories and f is a partial factor set associated with F , then the map H 2(G, U (Z(A))) → C(A, F ), defined by [q] 7→ qf , is bijective. Here is a detailed outline of this article. In Section 2, we state our con- ventions on categories. In particular, we recall the definition of inverse categories. In that section, we also show that the collection of partial bi- jections between sets form an inverse category which we denote by BIJcat (see Proposition 13). Inside this category sits the well-known groupoid BIJ of bijections between sets. In Section 3, we show (see Proposition 17) that the collection of partial (commutative) ring isomorphisms ISOcat (ISOCcat) is an inverse subcategory of BIJcat. This category contains, in turn, the well-known groupoid ISO (ISOC) having all (commutative) rings as objects and ring isomorphisms as morphisms. We also show a result concerning central idempotents in rings that we need in subsequent sections. In Sec- tion 4, we recall the definition of (pre-)equivalence data and some proper- ties of such systems, we also introduce the Picard inverse category PICcat (see Definition 27 and Theorem 28). In Section 5, we define epsilon-strongly groupoid graded rings (see Definition 34). In Section 6, we provide examples of epsilon-strong groupoid gradings on partial skew groupoid rings, Leavitt path algebras and Morita rings. In Section 7, we define epsilon-strongly groupoid graded modules (see Definition 46) and we provide a characteri- zation of them (see Proposition 47) that generalizes a result previously ob- tained for strongly group graded modules. At the end of the section, after putting R = ⊕e∈G0Se, we show (see Proposition 48) that the multiplication maps mg,h : Sg ⊗R Sh → ǫgSgh = Sghǫh−1, for (g, h) ∈ G2, are R-bimodule isomorphisms. In particular this implies that for every g ∈ G1, the sextuple (ǫgR, ǫg−1R, Sg, Sg−1, mg,g−1, mg−1,g) is a set of equivalence data. In Section 8, we introduce generalized epsilon- crossed groupoid products (see Definition 53) and we show that they parame- trize the family of epsilon-strongly groupoid graded rings (see Proposition 55 and Proposition 56). In Section 9, we extend the construction of a partial cohomology theory for partial actions of groups on commutative monoids, from [5], to partial actions of groupoids. In Section 10, we use the results in the previous sections to prove Theorem 3. EPSILON-STRONGLY GROUPOID GRADED RINGS 5 2. Preliminaries on categories In this section, we state our conventions on categories. In particular, we recall the definition of inverse categories. We introduce the new notion of a partial functor of inverse categories (see Definition 4) and we show that the composition of two partial functors of inverse categories is again a partial functor of inverse categories (see Proposition 25). In this section, we also show that the collection of partial bijections between sets form an inverse category which we denote by BIJcat (see Proposition 13). Suppose that G is a category. The classes of objects and morphisms of G are denoted by G0 and G1, respectively. Recall that G is called small if G1 is a set. If g ∈ G1, then the domain and codomain of g are denoted by d(g) and c(g), respectively. If n ≥ 2, then we let Gn denote the set of all (g1, . . . , gn) ∈ ×n i=1G1 that are composable, that is, such that for every i ∈ {1, . . . , n − 1} the relation d(gi) = c(gi+1) holds. The category G is called a groupoid if to each g ∈ G1 there is a unique h ∈ G1 such that (g, h) ∈ G2, (h, g) ∈ G2, gh = c(g) and hg = d(g). In that case, we write h = g−1. A subcategory H of a groupoid G is called a subgroupoid if the restriction of the map (·)−1 on G1 to H1 coincides with the map (·)−1 on H1. Let G be an inverse category. Recall that this means that there to each g ∈ G1 is a unique h ∈ G1 such that (g, h) ∈ G2, (h, g) ∈ G2, ghg = g and hgh = h. In that case we write h = g∗. A subcategory H of G is called an inverse subcategory if the restriction of the map ∗ on G1 to H1 coincides with the map ∗ on H1. Note that if G is a groupoid, then G is an inverse category if we for each g ∈ G1 put g∗ = g−1. It turns out that, for our purposes, the usual notion of functor is too restrictive when considered for inverse categories. Therefore we make the following weakening. Definition 4. Suppose that G and H are inverse categories. A partial functor of inverse categories F : G → H is a pair of maps (F0, F1), where F0 : G0 → H0 and F1 : G1 → H1, that satisfy the following axioms. (I1) If g : a → b in G1, then F1(g) : F0(a) → F0(b). (I2) If a ∈ G0, then F1(ida) = idF0(a). (I3) If (g, h) ∈ G2, then F1(g)F1(h) = F1(g)F1(g∗)F1(gh) = F1(gh)F1(h∗)F1(h). By abuse of notation, we will write F for both F0 and F1 in the sequel. Remark 5. Note that if we replace axiom (I3) in Definition 4 by • If (g, h) ∈ G2, then F1(g)F1(h) = F1(gh). then F is an ordinary functor. Proposition 6. If G and H are inverse categories and F : G → H is an ordinary functor, then F is a partial functor of inverse categories. Proof. Take (g, h) ∈ G2. Then, since F is an ordinary functor, we get that F (g)F (h) = F (gh) = F (gg∗gh) = F (g)F (g∗)F (gh) and F (g)F (h) = F (gh) = F (ghh∗h) = F (gh)F (h∗)F (h). (cid:3) 6 EPSILON-STRONGLY GROUPOID GRADED RINGS Proposition 7. If F : G → H is a partial functor of inverse categories, then for every g ∈ G1 the relation F (g∗) = F (g)∗ holds. Proof. Take a morphism g : a → b in G1. If we put h = ida in (I3), then we get that F (g)F (ida) = F (g)F (g∗)F (g). From (I2) it follows that this relation can be written as (1) F (g) = F (g)F (g∗)F (g). By replacing g by g∗ in (1), we get that (2) F (g∗) = F (g∗)F (g)F (g∗). Equations (1) and (2) show that F (g∗) satisfy the axioms for F (g)∗. Since F (g)∗ is uniquely defined we thus get that F (g∗) = F (g)∗. (cid:3) Lemma 8. Suppose that F : G → H is a partial functor of inverse cate- gories. If (g, h) ∈ G2 are chosen so that gh = h (gh = g) , then F (g)F (h) = F (h) (F (g)F (h) = F (g)) . Proof. Suppose that (g, h) ∈ G2 satisfy gh = h. From (I3) and Proposi- tion 7, we get that F (g)F (h) = F (gh)F (h∗)F (h) = F (h)F (h)∗F (h) = F (h). Now suppose that (g, h) ∈ G2 satisfy gh = g. From (I3) and Proposition 7, we get that F (g)F (h) = F (g)F (g∗)F (gh) = F (g)F (g)∗F (g) = F (g). (cid:3) Proposition 9. If F : G → G′ and F ′ : G′ → G′′ are partial functors of inverse categories, then F ′ ◦ F : G → G′′ is a partial functor of inverse categories. Proof. Put F ′′ = F ′ ◦ F . It is easy to see that (I1) and (I2) hold for F ′′. Now we show (I3). To this end, take (g, h) ∈ G2. We wish to show that (3) and (4) F ′′(g)F ′′(h) = F ′′(g)F ′′(g∗)F ′′(gh) F ′′(g)F ′′(h) = F ′′(gh)F (h∗)F (h). First we show (3). Using (I3) for F ′ and F , we get that the left side of (3) equals F ′(F (g))F ′(F (h)) = F ′(F (g))F ′(F (g)∗)F ′(F (g)F (h)) = F ′(F (g))F ′(F (g)∗)F ′(F (g)F (g)∗F (gh)) Using Lemma 8 for the right side of (3) we get F ′(F (g))F ′(F (g)∗)F ′(F (gh)) = F ′(F (g))[F ′(F (g)∗)F ′(F (g)F (g)∗)]F ′(F (gh)). Using (I3) and Lemma 8, the last expression equals F ′(F (g))F ′(F (g)∗)[F ′(F (g)F (g)∗)F ′(F (g)F (g)∗)F ′(F (gh))] = F ′(F (g))F ′(F (g)∗)F ′(F (g)F (g)∗F (gh)) EPSILON-STRONGLY GROUPOID GRADED RINGS 7 which shows (3). Now we show (4). Using (I3) for F ′ and F , we get that the left side of (4) equals F ′(F (g))F ′(F (h)) = F ′(F (g)F (h))F ′(F (h)∗)F ′(F (h)) = F ′(F (gh)F (h∗)F (h))F ′(F (h)∗)F ′(F (h)). The right side of (4) equals (5) F ′(F (gh))F ′(F (h∗))F ′(F (h)). Using Lemma 8, we get that (5) equals F ′(F (gh))F ′(F (h∗)F (h))F ′(F (h∗))F ′(F (h)). Using (I3) and Lemma 8, the last expression equals F ′(F (gh)F (h∗)F (h))F ′(F (h∗)F (h))F ′(F (h∗)F (h))F ′(F (h∗))F ′(F (h)) which equals F ′(F (gh)F (h∗)F (h))F ′(F (h∗))F ′(F (h)) showing (4). (cid:3) Proposition 10. If F : G → H is a partial functor of inverse categories, where H is a groupoid, then F is an ordinary functor. Proof. Take (g, h) ∈ G2. From Proposition 7, it follows that F (g)F (h) = F (g)F (g∗)F (gh) = F (g)F (g)∗F (gh) = F (g)F (g)−1F (gh) = F (gh). (cid:3) Definition 11. Let BIJ denote the groupoid having the collection of all sets as objects and bijections between sets as morphisms. A f Y Definition 12. Let A and B be sets. By a partial bijection from B to A we mean a choice of subsets Y ⊆ B and X ⊆ A and a bijection f : Y → X. We will indicate this by writing X B . We will now define, what we call, the category of partial bijections BIJcat. The class of objects in BIJcat consists of all sets. The class of morphisms in BIJcat consists of all partial bijections X B . The domain and codomain of such a morphism is B and A, respectively. The identity morphism at A is defined to be A A. The empty partial bijection from B to A is denoted by A0B. Suppose that X A f Y B and X ′ D are morphisms in BIJcat. If B = C, then the composition of these morphisms is defined to be f (Y ∩X ′) A f Y ∩X ′ ◦ gg−1(Y ∩X ′) A(idA)A C gY ′ g−1(Y ∩X ′) D A f Y . Otherwise, the composition is defined to be A0D. We also define a map ∗ : (BIJcat)1 → (BIJcat)1 by (X Bf −1X A . A f Y B )∗ = Y Proposition 13. BIJcat is an inverse category. C hY ′′ Proof. Suppose that α = X BIJcat. First we check the axioms for identity elements: (idAA∩X ◦ f f −1(A∩X))f −1(A∩X) Aα = idA(A∩X) C and γ = X ′′ B , β = X ′ A(idA)A A B gY ′ A f Y B A D are morphisms in = X A (idX ◦ f )Y B = α and αB B(idB)B B = f (Y ∩B) A (f Y ∩B ◦ idBid−1 B (Y ∩B)) B (Y ∩B) id−1 B = Y B(f ◦ idY )X A = α. 8 EPSILON-STRONGLY GROUPOID GRADED RINGS Now we prove associativity. We get that (αβ)γ = X1 A (f Y ∩X ′ ◦ gg−1(Y ∩X ′))g−1(Y ∩X ′)∩X ′′ ◦ h , Y1 C Y1 where and X1 = (f Y ∩X ′ ◦ gg−1(Y ∩X ′))(g−1(Y ∩ X ′) ∩ X ′′), We also get that Y1 = h−1(g−1(Y ∩ X ′) ∩ X ′′). α(βγ) = X2 A f Y ∩g(Y ′∩X ′′) ◦ (gY ′∩X ′′ ◦ hh−1(Y ′∩X ′′))Y2 Y2 D where and X2 = f (Y ∩ g(Y ′ ∩ X ′′)) Y2 = (gY ′∩X ′′ ◦ hh−1(Y ′∩X ′′))−1(Y ∩ g(Y ′ ∩ X ′′)). Since composition of functions is associative, we only need to show that X1 = X2 and Y1 = Y2. First we show that X1 ⊆ X2. Take a ∈ X1. Then a = f (g(b)) for some b ∈ X ′′ such that g(b) ∈ Y ∩ X ′. Since g : Y ′ → X ′ we also get that b ∈ Y ′. Thus g(b) ∈ Y ∩ g(Y ′ ∩ X ′′) and hence a = f (g(b)) ∈ X2. Next we show that X2 ⊆ X1. Take c ∈ X2. Then c = f (g(d)), for some d ∈ Y ′ ∩ X ′′ such that g(d) ∈ Y . Since g : Y ′ → X ′ we get that g(d) ∈ Y ∩ X ′. Thus d ∈ g−1(Y ∩ X ′) ∩ X ′′ and hence c ∈ X1. Now we show that Y1 ⊆ Y2. Take a ∈ Y1. Then h(a) ∈ g−1(Y ∩X ′)∩X ′′ ⊆ Y ′ ∩ X ′′. Thus g(h(a)) ∈ g(Y ′ ∩ X ′′). Also g(h(a)) ∈ g(g−1(Y ∩ X ′)) ⊆ Y . Hence g(h(a)) ∈ Y ∩ g(Y ′ ∩ X ′′). So we get that a ∈ Y2. Next we show that Y2 ⊆ Y1: Y2 = (h−1h−1(Y ′∩X ′′) ◦ g−1 Y ′∩X ′′)(Y ∩ g(Y ′ ∩ X ′′)) = h−1h−1(Y ′∩X ′′)(g−1(Y ∩ g(Y ′ ∩ X ′′)) ∩ Y ′ ∩ X ′′) = h−1(g−1(Y ∩ g(Y ′ ∩ X ′′)) ∩ Y ′ ∩ X ′′) ∩ h−1(Y ′ ∩ X ′′). Since g(Y ′ ∩ X ′′) ⊆ X ′ and Y ′ ∩ X ′′ ⊆ X ′′ we thus get that Y2 ⊆ Y1. Finally, we show that BIJcat is an inverse category. To this end, first note that (6) and (7) αα∗ = X A f Y B (X A f Y B )∗ = X A f Y B Y Bf −1X A = X A idX X A α∗α = (X A f Y B )∗X A f Y B = Y Bf −1X A X A f Y B = Y BidY Y B. Thus, it follows that αα∗α = X A idX X A X A f Y B = X A f Y B = α and α∗αα∗ = Y BidY Y B Y Bf −1X A = Y Bf −1X A = α∗. EPSILON-STRONGLY GROUPOID GRADED RINGS 9 Next suppose that (8) and (9) where β = X ′ B gY ′ Bf −1X Y αβα = α βαβ = β A . From (6), (7) and (8) it follows that X ′ B gY ′ A = α∗ = α∗αα∗ = α∗αβαα∗ = Y BidY g−1(Y ∩X ′)∩X A = g(g−1(Y ∩X ′)∩X) gg−1(Y ∩X ′)∩X Y B B A . X A idX X A Thus we get that Y = g(g−1(Y ∩ X ′)∩ X) ⊆ X ′ and X = g−1(Y ∩ X ′)∩ X ⊆ Y ′. Analogously, from (9), it follows that X ′ ⊆ Y and Y ′ ⊆ X. Thus X ′ = Y and Y ′ = X and hence it follows that β = α∗. (cid:3) 3. Preliminaries on rings In this section, we introduce the category of partial (commutative) ring isomorphisms ISOcat (ISOCcat). This category contains the well-known groupoid ISO (ISOC) having all (commutative) rings as objects and ring isomorphisms as morphisms. We also show a result (see Proposition 18), concerning central idempotents in rings, that we need in subsequent sections. Let A be a ring. We always assume that A is associative and equipped with a multiplicative identity element 1A. Definition 14. Let ISO (ISOC) denote the category having all (commuta- tive) rings as objects and ring isomorphisms as morphisms. Remark 15. Recall that an ideal I of a ring A is said to be a unital ideal if I, viewed as a ring in itself, is unital. In this case, the multiplicative identity element of I is denoted by 1I and lies in the center of A. Indeed, let a ∈ A be arbitrary. Since I is an ideal of A it follows that 1I a, a1I ∈ I. Thus, 1I a = 1I a1I = a1I . Furthermore, if I and J are unital ideals of a ring A, then I ∩ J = IJ. Now we will define two ring versions of the inverse category BIJcat that we defined in the previous section. Definition 16. Let ISOcat (ISOCcat) denote the subcategory of BIJcat hav- ing (commutative) rings as objects and, as morphisms, all I B in BIJcat such that I and J are unital ideals of A and B respectively and f : J → I is a ring isomorphism. Note that the composition of two morphisms, I B and I ′ BgJ ′ . Define a map ∗ : (ISOcat)1 → (ISOcat)1 by restriction of the map ∗ defined on (BIJcat)1. This restricts, in turn, to a map ∗ : (ISOCcat)1 → (ISOCcat)1. C , in these categories equals f (J I ′) f J I ′ ◦ gg−1(J I ′) g−1(J I ′) C Af J Af J A The following is clear. Proposition 17. ISOcat and ISOCcat are inverse subcategories of BIJcat. 10 EPSILON-STRONGLY GROUPOID GRADED RINGS The center of A, denoted by Z(A), is the subring of A consisting of all elements a ∈ A with the property that for all b ∈ A the equality ab = ba holds. We let idem(A) denote the set of all central idempotents of A and we let ideal(A) denote the set of all unital ideals of A. Proposition 18. Let A be a ring. The map θ : idem(A) → ideal(A), defined by θ(x) = Ax, for x ∈ idem(A), is an isomorphism of multiplicative monoids. For all x ∈ idem(A) the equality Z(Ax) = Z(A)x holds. Proof. It is clear that θ is a homomorphism of multiplicative monoids. Take x, y ∈ idem(A) such that θ(x) = θ(y). Then Ax = Ay. Since both x and y are multiplicative identity elements for the same monoid, it follows that x = y. Thus θ is injective. Now we show that θ is surjective. Take I ∈ ideal(A). Recall that 1I ∈ Z(A), by Remark 15. By the idempotency of 1I we get that θ(1I ) = A1I = I. Thus the surjectivity of θ follows. For the last part, take x ∈ idem(A). The inclusion Z(A)x ⊆ Z(Ax) clearly holds. Take y ∈ Z(Ax). Then y = ax for some a ∈ A. Clearly, yx = ax2 = ax = y, since x is idempotent. Thus, it suffices to show that y ∈ Z(A). Take b ∈ A. Then, since x ∈ Z(A) and y ∈ Z(Ax), we get that yb = yxb = ybx = ybx = bxy = byx = by. (cid:3) 4. The Picard Inverse Category In this section, we recall the definition of (pre-)equivalence data and some properties of such systems. Then we introduce the Picard inverse category PICcat (see Definition 27 and Theorem 28). From [3, Definition (3.2)] we recall the following. Definition 19. A set of pre-equivalence data (I, J, P, Q, α, β) consists of rings I and J, an I-J-bimodule P , a J-I-bimodule Q an I-bimodule homo- morphism (10) α : P ⊗J Q → I and an I-bimodule homomorphism (11) β : Q ⊗I P → J such that the following two diagrams commute P ⊗J Q ⊗I P α⊗idP−−−−→ I ⊗I P Q ⊗I P ⊗J Q β⊗idQ−−−−→ J ⊗J Q (12) idP ⊗βy P ⊗J J y idQ⊗αy Q ⊗I I y −−−−→ P −−−−→ Q where the unlabelled arrows are the multiplication maps. We shall call it a set of equivalence data if α and β are isomorphisms. Now we gather some well-known properties concerning pre-equivalence data that we need in the sequel. EPSILON-STRONGLY GROUPOID GRADED RINGS 11 Proposition 20. If (I, J, P, Q, α, β) is a set of pre-equivalence data such that α (or β) is surjective, then the following assertions hold: (a) α (or β) is an isomorphism; (b) P and Q are generators as I-modules (or J-modules); (c) P and Q are finitely generated and projective J-modules (I-modules). Proof. See [3, Theorem (3.4)]. (cid:3) Proposition 21. If (I, J, P, Q, α, β) is a set of equivalence data, then the ring homomorphisms and EndJ (P ) ← I → EndJ (Q)op EndI (P )op ← J → EndI (Q), induced by the bimodule structure on P and Q, are isomorphisms. These isomorphisms restrict to ring isomorphisms EndI−J (P ) ← Z(I) → EndJ−I (Q) and EndI−J (P ) ← Z(J) → EndJ−I (Q), which in turn restrict to group isomorphisms AutI−J (P ) ← U (Z(I)) → AutJ−I(Q) and AutI−J (P ) ← U (Z(J)) → AutJ−I (Q). Proof. See [3, Theorem (3.5)]. (cid:3) Remark 22. Let (I, J, P, Q, α, β) be a set of equivalence data. It follows from Proposition 21 that there is a unique ring isomorphism γP : Z(J) → Z(I) with the property that for all p ∈ P and all b ∈ Z(J) the equality γP (b)p = pb holds. Definition 23. A set of partial equivalence data (A, B, I, J, P, Q, α, β) consists of rings A and B, unital ideals I and J of, respectively, A and B such that (I, J, P, Q, α, β) is a set of equivalence data, P is an I-J-bimodule and Q is a J-I-bimodule. Remark 24. Note that, with the notation and assumptions of Definition 23, P (resp. Q) extends uniquely to an A-B-bimodule (resp. B-A-bimodule). Thus, we may interchangeably think of P (resp. Q) as an A-B-bimodule or an I-J-bimodule (resp. B-A-bimodule or J-I-bimodule). Proposition 25. Suppose that (13) (A, B, I, J, P, Q, α, β) 12 and (14) EPSILON-STRONGLY GROUPOID GRADED RINGS (B, C, I ′, J ′, P ′, Q′, α′, β′) are sets of partial equivalence data. Then (A, C, I ′′, J ′′, P ′′, Q′′, α′′, β′′) is a set of partial equivalence data, where (15) I ′′ = γP (1J 1I ′)A, J ′′ = γ−1 P ′ (1J 1I ′)C, P ′′ = P ⊗B P ′, Q′′ = Q′ ⊗B Q, and for p ∈ P , p′ ∈ P ′, q ∈ Q, q′ ∈ Q′, we put α′′(p ⊗ p′ ⊗ q′ ⊗ q) = α(pα′(p′ ⊗ q′) ⊗ q) and β′′(q′ ⊗ q ⊗ p ⊗ p′) = β′(q′ ⊗ β(q ⊗ p)p′). Proof. We begin by noticing that P ′ ⊗C Q′ ∼= P ′ ⊗J ′ Q′. Indeed, J ′ is a unital ideal of C and hence J ′ = 1J ′C. Moreover, P ′ is a right J ′-module and Q′ is a left J ′-module. Thus, P ′ ⊗C Q′ ∋ p′ ⊗C q′ 7→ p′ ⊗J ′ q′ ∈ P ′ ⊗J ′ Q′ is a well-defined isomorphism of I ′-bimodules (and B-bimodules). The map α′′ : P ′′ ⊗C Q′′ → I ′′ is an isomorphism of A-bimodules since it is the composition of the following chain of A-bimodule isomorphisms: P ⊗B P ′ ⊗C Q′ ⊗B Q ∼= P ⊗B P ′ ⊗J ′ Q′ ⊗B Q ∼= P ⊗B I ′ ⊗B Q = P ⊗B 1I ′B ⊗B Q = P 1I ′ ⊗B B ⊗B Q ∼= P 1J 1I ′ ⊗B Q ∼= γP (1J 1I ′)P ⊗B Q ∼= γP (1J 1I ′)I = γP (1J 1I ′)A. Analogously, the map β′′ : Q′′⊗AP ′′ → J ′′ is an isomorphism of C-bimodules since it is the composition of the following chain of C-bimodule isomor- phisms: Q′ ⊗B Q ⊗A P ⊗B P ′ ∼= Q′ ⊗B J ⊗B P = Q′ ⊗B 1J B ⊗B P ′ = Q′ ⊗B B ⊗B 1J P ′ ∼= Q′ ⊗B 1J 1I ′P ′ ∼= Q′ ⊗B P ′γ−1 P ′ (1J 1I ′) ∼= J ′γ−1 P ′ (1J 1I ′) = γ−1 P ′ (1J 1I ′)C. From Proposition 18 it follows that I ′′ = γP (1J 1I ′)A and J ′′ = γ−1 P ′ (1J 1I ′)C are well-defined. Now we verify the diagrams in (12). By abuse of notation, we let m denote all of the various multiplication maps. Take p1, p2 ∈ P , 2 = p2 ⊗ p′′ 1, p′ p′ 2, 1 = q′ q′′ 2 ⊗ q2. Then, by making use of (12) for (13) and 2 ∈ P ′, q1, q2 ∈ Q and q′ 2 ∈ Q′. Put p′′ 1 = p1 ⊗ p′ 1 ⊗ q1 and q′′ 2 = q′ 1, q′ 1, p′′ EPSILON-STRONGLY GROUPOID GRADED RINGS 13 (14), we get that (m ◦ (α′′ ⊗ idP ′′ ))(p′′ 1 ⊗ q′′ 1 ⊗ p′′ 2 ) = (m ◦ (α′′ ⊗ idP ′′ ))(p1 ⊗ p′ 1 ⊗ q′ 1 ⊗ q1 ⊗ p2 ⊗ p′ 2) 1 ⊗ q′ 1 ⊗ q′ = α(p1α′(p′ = (m ◦ (α ⊗ idP ))(p1α′(p′ = (m ◦ (idP ⊗ β))(p1α′(p′ = p1α′(p′ = p1 ⊗ α′(p′ = p1 ⊗ (m ◦ (α′ ⊗ idP ′ ))(p′ = p1 ⊗ (m ◦ (idP ′ ⊗ β ′))(p′ = p1 ⊗ p′ = (m ◦ (idP ′′ ⊗ β ′′))(p′′ 1) ⊗ q1)p2 ⊗ p′ 2 1 ⊗ q′ 1) ⊗ q1 ⊗ p2) ⊗ p′ 2 1 ⊗ q′ 1) ⊗ q1 ⊗ p2) ⊗ p′ 2 1)β(q1 ⊗ p2) ⊗ p′ 2 1)β(q1 ⊗ p2)p′ 2 1 ⊗ q′ 1 ⊗ q′ 1 ⊗ β(q1 ⊗ p2)p′ 2) 1 ⊗ q′′ 1 ⊗ β(q1 ⊗ p2)p′ 2) 1 ⊗ β(q1 ⊗ p2)p′ 2) 1 ⊗ p′′ 2 ) 1β ′(q′ 1 ⊗ q′ and (m ◦ (idQ′′ ⊗ α′′))(q′′ 1 ⊗ p′′ 1 ⊗ q′′ 1 ⊗ q′ 2) ⊗ q2) 2 ) = q′ 1 ⊗ q1α(p1α′(p′ = q′ 1 ⊗ (m ◦ (idQ ⊗ α))(q1 ⊗ p1α′(p′ = q1 ⊗ (m ◦ (β ⊗ idQ))(q1 ⊗ p1α′(p′ = q′ 1 ⊗ q′ 1 ⊗ q′ 2) ⊗ q2) 2) ⊗ q2) 1 ⊗ q′ 1 ⊗ q′ 1 ⊗ β(q1 ⊗ p1α′(p′ 1 ⊗ β(q1 ⊗ p1)α′(p′ 1β(q1 ⊗ p1)α′(p′ 1α′(β(q1 ⊗ p1)p′ 1 ⊗ q′ 1 ⊗ q′ 2))q2 2)q2 2) ⊗ q2 2) ⊗ q2 = q′ = q′ = q′ = (m ◦ (idQ′ ⊗ α′))(q′ = (m ◦ (β ′ ⊗ idQ′ ))(q′ = β ′(q′ 1 ⊗ β(q1 ⊗ p1)p′ = (m ◦ (β ′′ ⊗ idQ′′ ))(q′′ 1 ⊗ β(q1 ⊗ p1)p′ 1 ⊗ β(q1 ⊗ p1)p′ 1)q′ 1 ⊗ p′′ 1 ⊗ q′′ 2 ⊗ q2 2 ), 1 ⊗ q′ 1 ⊗ q′ 2) ⊗ q2 2) ⊗ q2 which finishes the proof. (cid:3) To motivate the approach taken later, we now recall the definition of the Picard groupoid PIC. Definition 26. Let PIC denote the category having as objects all unital rings. A morphism in PIC from B to A is the collection of all A-B-bimodule isomorphism classes [P ], for invertible A-B-bimodules P . Given two such classes [P ] and [Q], where d([P ]) = B = c([Q]), we put [P ][Q] = [P ⊗B Q]. Then PIC is a groupoid. Indeed, if P is an invertible A-B-bimodule, then there is an invertible B-A-bimodule Q such that P ⊗B Q ∼= A (as A- bimodules) and Q ⊗A P ∼= B (as B-bimodules). Thus, if we put [P ]−1 = [Q], then, clearly [P ][P ]−1 = [A] and [P ]−1[P ] = [B]. Definition 27. Let P be an A-B-bimodule and suppose that I and J are unital ideals of, respectively, A and B, making P an I-J-bimodule. We will indicate this by writing I B is partially invertible if there is J A and maps α and β such that (A, B, I, J, P, Q, α, β) is a set of partial B. We say that I AP J AP J BQI 14 EPSILON-STRONGLY GROUPOID GRADED RINGS equivalence data. Let PART denote the collection of all partially invertible bimodules I B. Define an equivalence relation ∼ on PART by saying that I AP J AP J A′P ′J ′ B′ if (A, B, I, J) = (A′, B′, I ′J ′) and P ∼= P ′ as I -J -bimodules. B ∼ I ′ AP J AP J AP J AP J B]) = B and c([I B in PART will be denoted by [I B] of partially invertible modules I The equivalence class of I B]. The class of objects in PICcat consists of all rings. The class of morphisms in PICcat consists of all equivalence classes [I AP J B. Define the domain and codomain of a morphism [I B] in PICcat by the relations d([I B]) = A, respectively. Given a ring A, the identity morphism at A is defined to be the morphism [A A]. Given B P ′J ′ A P ′′J ′′ two morphisms [I AP J C ], where I ′′, P ′′ and J ′′ are defined in Proposition 25. It is clear that the morphisms of the form [A A], for rings A, satisfy the axioms for identity morphisms of PICcat. If [I B] ∈ (PICcat)1, then there is J A and maps α and β such that (A, B, I, J, P, Q, α, β) is a set of partial equivalence data. Put [I C ] in PICcat put [I B] and [I ′ AAA AP J C ] = [I ′′ BP ′J ′ AAA B][I ′ AP J AP J B]∗ = [J BQI A]. AP J AP J BQI Theorem 28. PICcat is an inverse category. Proof. First we show that the partial composition in (PICcat)1 is associative. Suppose that [I1 J3 A P1 D ] are morphisms in PICcat. We need to show that J2 C ] and [I3 J1 B ], [I2 B P2 C P3 (16) A P1 J1 B ][I2 B P2 C P3 J3 D ] = [I1 A P1 B P2 J2 C ][I3 C P3 (cid:16)[I1 J2 C ](cid:17) [I3 J1 B ](cid:16)[I2 J3 D ](cid:17) . By repeated application of the composition in PICcat it follows that (16) is equivalent to showing the equalities (17) and (18) γP1⊗BP2(γ−1 P2 (1J1 1I2)1I3) = γP1(1J1γP2(1J21I3)) γ−1 P3 (γ−1 P2 (1J11I2)1I3) = γ−1 P2⊗C P3 (1J1γP2(1J21I3)). First we show (17). Take p1 ∈ P1 and p2 ∈ P2. Then p1 ⊗ p2γ−1 P2 (1J11I2)1I3 = p1 ⊗ p2γ−1 P2 (1J11I2)1J21I3 = p1 ⊗ 1J11I2p21J21I3 = p1 ⊗ 1J1p21J21I3 = p11J1 ⊗ p21J21I3 = p11J1 ⊗ γP2(1J21I3)p2 = p11J1γP2(1J21I3) ⊗ p2 = γP1(1J1γP2(1J21I3))p1 ⊗ p2. Now we show (18). Take p2 ∈ P2 and p3 ∈ P3. Then 1J1γP2(1J21I3)p2 ⊗ p3 = 1J11I2γP2(1J21I3)p2 ⊗ p3 = 1J11I2p21J21I3 ⊗ p3 = 1J11I2p21I3 ⊗ p3 = p2γ−1 P2 = p2 ⊗ γ−1 P2 (1J11I2)1I3 p3 = p2 ⊗ p3γ−1 P3 (1J1 1I2)1I3 ⊗ p3 (γ−1 P2 (1J11I2)1I3), EPSILON-STRONGLY GROUPOID GRADED RINGS 15 as desired. Next we show the axioms for ∗. Take g = [I Then there is J set of partial equivalence data. Put g∗ = [J B] ∈ (PICcat)1. A and maps α and β such that (A, B, I, J, P, Q, α, β) is a AP J BQI gg∗ = [I AP J B][J BQI A] =(cid:20)γP (1J 1J )A A A (P ⊗B Q)1I A A ] = [I AI I A]. = [1I A Using this we get that BQI A]. Then γ−1 Q (1J 1J )A A (P ⊗B Q) (cid:21) gg∗g = [I AI I A][I AP J B] = [γI (1I 1I )A A (I ⊗A P ) γ−1 P (1I 1I )B B ] = [1I A A P 1J B B ] = [I AP B B ] = g and g∗gg∗ = [J BQI A][I AI I (Q ⊗A I) γ−1 I A (1I 1I )A B Q1I A A ] = [J BQI A] = g∗. Now we show uniqueness of g∗. To this end, first note that B A] =(cid:20)γQ(1I 1I )B B] =(cid:20)γQ(1I 1I )B B g∗g = [J BQI A][I AP J Next, suppose that (Q ⊗A P ) γ−1 P (1I 1I )B B B J 1J B B ] = [J BJ J B]. (cid:21) = [1J B (cid:21) = [1J B ghg = g and hgh = h (19) for some h = [K it follows that g∗ghgg∗ = g∗gg∗ and thus that [J Rewriting the last equality we get that A ] with h∗ = [L B M L AN K B ]. From the first equality in (19) A]. A] = [J B M L B][K BQI BJ J A ][I AI I (cid:20)γJ (1J γM (1L1I ))B B (J ⊗B M ⊗A I) γ−1 M ⊗A I (1J γM (1L1I ))A A (cid:21) = [J BQI A] (20) and thus that (21) and (22) γJ (1J γM (1L1I ))B = J γ−1 M ⊗AI (1J γM (1L1I ))A = I. Since γJ is the identity map Z(J) → Z(J), (21) implies that 1J γM (1L1I )B = J and hence, in particular, that J ⊆ K. Using that γI equals the identity map on Z(I), it follows that γ−1 M ⊗AI : Z(γM (1L1I )) → Z(1L1I ). From (22) it therefore, in particular, follows that I ⊆ L. From the second equality in (19) it follows, by symmetry, that J ⊆ K and L ⊆ I. Thus J = K and L = I and hence from (20) it follows that h = g∗. (cid:3) Example 29 (The Picard semigroup of a commutative ring). Let R be a unital commutative ring and let M be a finitely generated (central) R-bimodule of rank less than or equal to one, that is, rk(Mp) ≤ 1, for all p ∈ Spec(R). Let M ∗ = HomR(M, R) be the dual of M. Then by [8, Proposition 3.8, Lemma 3.9] there exists e ∈ idem(R) and an R-bimodule isomorphism α : M ⊗R M ∗ → Re, given by α(m ⊗ f ) := f (m), for all f ∈ M ∗ and m ∈ M. Moreover, by [8, Lemma 3.10] the isomorphisms classes of M and 16 EPSILON-STRONGLY GROUPOID GRADED RINGS M ∗ are elements of the Picard group Pic(Re). In particular, both M and M ∗ are unital Re-bimodules. From this we get that (R, R, Re, Re, M, M ∗, α, α) is a set of partial equivalence data. By [8, Proposition 3.8] the inverse subcategory of PICcat, whose only object is R and whose morphisms are of the form [Re R ], is a commutative inverse semigroup, denoted by PicS(R). It was defined in [8] and is called the Picard semigroup of R. R M Re Definition 30. Now we will define a partial functor of inverse categories L : PICcat → ISOCcat. If A is a ring, then put L(A) = Z(A). If [I B] is a Z(J) morphism in PICcat, then put L([I Z(B) where γP : Z(J) → Z(I) is the ring isomorphism defined in Remark 22. B]) = Z(I) Z(A)γP AP J AP J Proposition 31. The map L : PICcat → ISOCcat is a functor and hence, by Proposition 6, a partial functor of inverse categories. Proof. Take morphisms g = [I AP J C ] in PICcat. Then B] and h = [I ′ B P J ′ γ−1 P ′ (1J 1I′ )Z(C) P ⊗BP ′ γ Z(C) L(gh) = γP (1J 1I′ )Z(A) Z(A) and L(g)L(h) = γP (Z(J)Z(I ′)) Z(A) γP Z(J)Z(I ′) ◦ γP ′ −1 P ′ (Z(J )Z(I′)) γ γ−1 P ′ (Z(J)Z(I ′)) Z(C) . Note that and γP (Z(J)Z(I ′)) = γP (1J 1I ′)Z(A) P ′ (Z(J)Z(I ′)) = γ−1 γ−1 P ′ (1J 1I ′)Z(C). P ′ (Z(J)Z(I ′)). If a ∈ γ−1 Put γ1 = γP Z(J)Z(I ′) and γ2 = γP ′γ−1 p ∈ P and p′ ∈ P ′, then γP ⊗BP ′(a)p ⊗ p′ = p ⊗ p′a = p ⊗ γ2(a)p′ = pγ2(a) ⊗ p′ = (γ1 ◦ γ2)(a)p ⊗ p′. From Remark 22 it therefore follows that γP ⊗BP ′ = γ1 ◦ γ2. (cid:3) P ′ (1J 1I ′)Z(C), 5. Epsilon-strongly groupoid graded rings In this section, we recall the definition of groupoid graded rings and some of their properties. Then we define epsilon-strongly groupoid graded rings (see Definition 34) and provide a characterization of them which generalizes an analogous result for group graded rings (see Proposition 37). Through- out this section, S denotes a ring which is graded by a small groupoid G. Recall from [13, 14] that this means that there is a set of additive subgroups {Sg}g∈G of S such that S = ⊕g∈GSg and, for all g, h ∈ G1, SgSh ⊆ Sgh, if (g, h) ∈ G2, and SgSh = {0}, if (g, h) /∈ G2. In that case, S is called strongly graded if for all (g, h) ∈ G2 the equality SgSh = Sgh holds. Given two G- graded rings S and T , a ring homomorphism f : S → T is called graded if for all g ∈ G1 the inclusion f (Sg) ⊆ Tg holds. We put R = ⊕e∈G0Se. From the next result it follows that we may always assume that G0 is finite. EPSILON-STRONGLY GROUPOID GRADED RINGS 17 Proposition 32. With the above notation, we get that 1S ∈ R. If we put G′ 1 = {g ∈ G1 (1S )d(g), (1S )c(g) 6= 0}, then G′ is a subgroupoid of G such that G′ 0 = {e ∈ G0 (1S )e 6= 0} and G′ 0 is finite and S = ⊕g∈G′ Sg. 1 Proof. This follows from [13, Proposition 2.1.1]. (cid:3) Proposition 33. The ring S is strongly graded if and only if for all g ∈ G1 the inclusion 1Sc(g) ∈ SgSg−1 holds. Proof. This follows from [14, Lemma 3.2]. (cid:3) Definition 34. The ring S is said to be epsilon-strongly graded by G if, for each g ∈ G1, SgSg−1 is a unital ideal of Sc(g) such that for all (g, h) ∈ G2 the equalities SgSh = SgSg−1Sgh = SghSh−1Sh hold. Remark 35. It follows from Proposition 32 and Proposition 33 that if S is strongly graded, then S is epsilon-strongly graded. Remark 36. Suppose that S is epsilon-strongly graded by G and g ∈ G1. Then by the definition of R, the Sc(g)-ideal SgSg−1 is a unital ideal of R. Moreover if ǫg is its multiplicative identity element, then for r ∈ R we get that ǫgr, rǫg ∈ SgSg−1. Therefore ǫgr = (ǫgr)ǫg = ǫg(rǫg) = rǫg, which shows that ǫg ∈ Z(R), and SgSg−1 = ǫgSc(g) = ǫgR. We now wish to show an epsilon-analogue of Proposition 33. Proposition 37. The ring S is epsilon-strongly graded by G if and only if for each g ∈ G1 there is ǫg ∈ SgSg−1 such that for each s ∈ Sg the equalities ǫgs = s = sǫg−1 hold. Proof. First we show the "only if" statement. Suppose that S is epsilon- strongly graded. Take g ∈ G1. Let ǫg denote 1Sg Sg−1 . Take sg ∈ Sg. From Proposition 32 it follows that SgSg−1Sg = Sg. Therefore there is a positive integer n and a(i) g−1 ∈ Sg−1, for i ∈ {1, . . . , n}, such that g b(i) g ∈ Sg and b(i) g . Since ǫg = 1Sg Sg−1 and ǫg−1 = 1Sg−1 Sg , we get that g , c(i) g−1c(i) i=1 a(i) n n (ǫga(i) g b(i) g−1)c(i) g = g b(i) a(i) g−1c(i) g = sg g (b(i) a(i) g−1c(i) g ǫg−1) = g b(i) a(i) g−1c(i) g = sg. Now we show the "if" statement. Suppose that to each g ∈ G1 there is ǫg ∈ SgSg−1 such that for each s ∈ Sg the equalities ǫgs = s = sǫg−1 hold. Take (g, h) ∈ G2. Then, from Proposition 32, it follows that SgSh = ǫgSgSh ⊆ SgSg−1SgSh ⊆ SgSg−1Sgh ⊆ SgSd(g)h = SgSc(h)h = SgSh and SgSh = SgShǫh−1 ⊆ SgShSh−1Sh ⊆ SghSh−1Sh = Sgc(h)Sh = Sgd(g)Sh = SgSh. sg =Pn and ǫgsg = sgǫg−1 = Xi=1 Xi=1 n Xi=1 Xi=1 n 18 EPSILON-STRONGLY GROUPOID GRADED RINGS Moreover, it is clear that ǫg is the multiplicative identity element of SgSg−1. (cid:3) 6. Examples of epsilon-strongly groupoid graded rings In this section we present some examples of epsilon-strongly groupoid graded rings. 6.1. Partial skew groupoid rings. The notion of a partial action of a groupoid on a ring, as well as the construction of the corresponding partial skew groupoid ring, is due to Bagio and Paques [2]. Let G be a groupoid and suppose that B is a ring which is the product of a collection of rings {Be}e∈G0. Definition 38. A partial groupoid action of G on B is a collection of maps {θg}g∈G1, where, for each g ∈ G1, Bg is an ideal of Bc(g), Bc(g) is an ideal of B and θg : Bg−1 → Bg is a ring isomorphism satisfying the following three axioms: (G1) if e ∈ G0, then θe = idBe; (G2) if (g, h) ∈ G2, then θ−1 (G3) if (g, h) ∈ G2 and x ∈ θ−1 h (Bg−1 ∩ Bh) = B(gh)−1; h (Bg−1 ∩ Bh), then θg(θh(x)) = θgh(x). Note that conditions (G2) and (G3) are equivalent to the fact that θgh is an extension of θg ◦ θh. We say that θ is global if θgh = θg ◦ θh, for each (g, h) ∈ G2. Definition 39. Let {θg}g∈G1 be a partial groupoid action of G on B. Sup- pose that for each each g ∈ G1, Bg is unital, i.e. Bg is generated by an idempotent 1g which is central in Bc(g), and θg is a monoid isomorphism. In that case, we say that {θg}g∈G1 is a unital partial groupoid action of G on B. The partial skew groupoid ring B ⋆θ G, associated with a unital partial groupoid action {θg}g∈G1 of G on B, is the set of all finite formal sums bgδg, where bg ∈ Bg, with addition defined componentwise and mul- Pg∈G1 tiplication determined by the rule (23) (bgδg)(b′ hδh) = bgαg(b′ h1g−1)δgh, if (g, h) ∈ G2, and (rgδg)(r′ hδh) = 0, otherwise. There is a natural G-grading on B ⋆θ G. Indeed, if we put Sg = Bgδg for each g ∈ G1, then B ⋆θ G = ⊕g∈G1Sg is G-graded. For each g ∈ G the idempotent 1gδc(g) satisfies the conditions of Proposition 37. Thus, B ⋆θ G is an epsilon-strongly G-graded ring. Moreover, by [1, Proposition 2.5] one has that B ⋆θ G is strongly G-graded, if and only if, θ is global. EPSILON-STRONGLY GROUPOID GRADED RINGS 19 6.2. Leavitt path algebras. Let E = (E0, E1, r, s) be a directed graph, consisting of two countable sets E0, E1 and maps r, s : E1 → E0. The elements of E0 are called vertices and the elements of E1 are called edges. If both E0 and E1 are finite sets, then we say that E is finite. A path µ in E is a sequence of edges µ = µ1 . . . µn such that r(µi) = s(µi+1) for i ∈ {1, . . . , n − 1}. In such a case, s(µ) := s(µ1) is the source of µ, r(µ) := r(µn) is the range of µ and n is the length of µ. Definition 40 ([11]). Let E be any directed graph and let K be a uni- tal ring. The Leavitt path algebra of E with coefficients in K, denoted by LK(E), is the algebra generated by a set {v v ∈ E0} of pairwise orthogonal idempotents, together with a set of elements {f f ∈ E1} ∪ {f ∗ f ∈ E1}, which satisfy the following relations: (1) s(f )f = f r(f ) = f , for all f ∈ E1; (2) r(f )f ∗ = f ∗s(f ) = f ∗, for all f ∈ E1; (3) f ∗f ′ = δf,f ′r(f ), for all f, f ′ ∈ E1; (4) v = P{f ∈E1s(f )=v} f f ∗, for every v ∈ E0 for which s−1(v) is non- empty and finite. Here the ring K commutes with the generators. Remark 41. (a) Every path µ = µ1 . . . µn may be viewed as an element of LK(E). Given such an element µ, we put µ∗ := µ∗ 1 ∈ LK (E). The element µ∗ may also be thought of as a ghost path in E, as opposed to µ which is a real path. n . . . µ∗ (b) Note that every element x ∈ LK (E) may be written on the form i , for suitable ki ∈ K and (real) paths αi and βi satisfying i=1 kiαiβ∗ x = Pn r(αi) = r(βi), for i ∈ {1, . . . , n}. 6.2.1. The groupoid. Based on E, we define a groupoid G in the following way. The objects of G are the vertices of E, i.e. G0 = E0. An ordered pair of vertices, (u, v), is an arrow in G with d(u, v) = v and c(u, v) = u, if there is a path µ = µ1µ2 . . . µn such that u = s(µ) = s(µ1) and v = r(µ) = r(µn), where µi ∈ E1∪(E1)∗∪E0 for each i ∈ {1, 2, . . . , n}, and r(µi) = s(µi+1) for each i ∈ {1, 2, . . . , n − 1}. Two arrows (u, v) and (v′, w) and composable if only if v′ = v. In that case, their composition is defined to be equal to (u, v)(v, w) = (u, w). 6.2.2. The grading. Let W denote the set of finite real paths in E, and consider W as a subset of the ring LK(E). Lemma 42. If we, for each (u, v) ∈ G1, put S(u,v) = spanK{αβ∗ α, β ∈ W such that s(α) = u, r(α) = r(β), s(β) = v} then this turns S = LK (E) = ⊕(u,v)∈G1S(u,v) into a G-graded ring. 20 EPSILON-STRONGLY GROUPOID GRADED RINGS Proof. Clearly, LK(E) = P(u,v)∈G1 S(u,v) and this sum is in fact direct. Indeed, take a non-zero x ∈ S(u,v) ∩ S(a,b). Then x = ux and x = ax, and hence 0 6= x = ux = u(ax) = (ua)x. In particular, ua 6= 0 which means that u = a. Similarly, we may conclude that v = b. That is, (u, v) = (a, b). Let (u, v), (v′, w) ∈ G1 be arbitrary. If v′ = v then we get that S(u,v)S(v,w) ⊆ S(u,w). On the other hand, if v′ 6= v, then S(u,v)S(v′,w) = {0}. This shows that LK(E) is G-graded. (cid:3) Theorem 43. If E is a finite graph, then the Leavitt path algebra S = LK(E) is epsilon-strongly G-graded. Proof. Let (u, v) ∈ G1 be arbitrary. If u /∈ S(u,v)S(v,u), then we shall be interested in the following set: P(u,v) = {α αβ∗ ∈ S(u,v)}. For αi, αj ∈ P(u,v), we write αi ≤ αj if αi is an initial subpath of αj. Clearly, ≤ is a partial order on P(u,v). Moreover, using that E is finite, it is not difficult to see that there can only be a finite number of minimal elements of P(u,v) with respect to ≤. We collect all such minimal elements in the set M(u,v) = {α1, . . . , αk}. We are now ready to define ǫ(u,v) in the following way: ǫ(u,v) =(cid:26) u Pαj ∈M(u,v) if u ∈ S(u,v)S(v,u) otherwise αjα∗ j Note that, whenever αβ∗ ∈ S(u,v) is a non-zero monomial, i.e. r(α) = r(β), we get that αα∗ = αr(β)α∗ = αβ∗βα∗ ∈ S(u,v)S(v,u). In particular, αjα∗ j ∈ S(u,v)S(v,u) for each αj ∈ M(u,v). Hence, by construction, ǫ(u,v) ∈ S(u,v)S(v,u). Moreover, (ǫ(u,v))∗ = ǫ(u,v). Take any monomial γδ∗ ∈ S(u,v). First we show that ǫ(u,v)γδ∗ = γδ∗. Case 1: (u ∈ S(u,v)S(v,u)) Clearly, ǫ(u,v)γδ∗ = uγδ∗ = γδ∗. Case 2: (u /∈ S(u,v)S(v,u)) Note that there is some α′ ∈ M(u,v) such that γ = α′γ′. Thus, ǫ(u,v)γδ∗ = α′α′∗ + Xαj ∈M(u,v)\{α′} = α′α′∗α′γ′δ∗ + 0 = α′γ′δ∗ = γδ∗. αjα∗ j  γδ∗ It remains to show that γδ∗ǫ(v,u) = γδ∗. Note that γδ∗ ∈ S(u,v) ⇐⇒ δγ∗ ∈ S(v,u). It follows, from Case 1 and Case 2, that ǫ(v,u)δγ∗ = δγ∗. Using this we get that γδ∗ǫ(v,u) = γδ∗(ǫ(v,u))∗ = (ǫ(v,u)δγ∗)∗ = (δγ∗)∗ = γδ∗. This concludes the proof. (cid:3) EPSILON-STRONGLY GROUPOID GRADED RINGS 21 In general LK(E) need not be strongly G-graded, as the following example shows. Example 44. Let K be a unital ring and consider the Leavitt path algebra LK(E) associated with the following graph E: •v1 f1 •v2 f2 / •v3 A few short calculations reveal that • S(v2,v1)S(v1,v2) = Kf1f ∗ 1 • S(v1,v2)S(v2,v1) = Kv1 • S(v3,v2)S(v2,v3) = Kv3 • S(v2,v3)S(v3,v2) = Kf2f ∗ 2 • S(v3,v1)S(v1,v3) = {0} • S(v1,v3)S(v3,v1) = {0} and we may choose • ǫ(v2,v1) = f1f ∗ 1 • ǫ(v1,v2) = v1 • ǫ(v3,v2) = v3 • ǫ(v2,v3) = f2f ∗ 2 • ǫ(v1,v3) = ǫ(v3,v1) = 0. Clearly, (v1, v3) and (v3, v1) are composable, but {0} = S(v1,v3)S(v3,v1) 6= S(v1,v1). Thus, LK(E) is not strongly G-graded. Remark 45. Gon¸calves and Yoneda [10] have shown that each Leavitt path algebra may be viewed as a partial skew groupoid ring. Their observation gives rise to another example of an epsilon-strong groupoid grading on a Leavitt path algebra. 6.3. Morita rings. Let (A, B,A MB,B NA, ϕ, φ) be a strict Morita context. It consists of unital rings A and B, an A-B-bimodule M , a B-A-bimodule N , an A-A-bimodule epimorphism ϕ : M ⊗B N → A and a B-B-bimodule epimorphism φ : N ⊗A M → B. The associated Morita ring is the set with the natural addition and with a multiplication defined by S =(cid:18) A M N B (cid:19) (cid:18) a1 m1 b1 (cid:19) ∗(cid:18) a2 m2 n2 n1 b2 (cid:19) =(cid:18) a1a2 + ϕ(m1 ⊗ n2) n1a2 + b1n2 a1m2 + m1b2 φ(n1 ⊗ m2) + b1b2 (cid:19) for a1, a2 ∈ A, b1, b2 ∈ B, m1, m2 ∈ M and n1, n2 ∈ N . Let G be a group and I a non-empty set. Then the set I × G × I considered as morphisms, where the composition is given by the rule (i, g, j)(j, h, k) = (i, gh, k), o o / 22 EPSILON-STRONGLY GROUPOID GRADED RINGS for all i, j, k ∈ I and g, h ∈ G, is a groupoid. Using this groupoid and taking I = {1, 2} and G the infinite cyclic group generated by g we can define a grading on S by putting 0 S(1,e,1) =(cid:18) A 0 S(1,g,2) =(cid:18) 0 M 0 (cid:19) , S(2,e,2) =(cid:18) 0 0 (cid:19) , S(2,g−1,1) =(cid:18) 0 0 B (cid:19) , N 0 (cid:19) 0 0 0 and S(i,h,j) = {( 0 0 have that 0 0 )} , in any other case. Then for h ∈ G \ {e, g, g−1} we S(1,h−1,1)S(1,h,1) 6= S(1,e,1) and thus S is not strongly graded. However, S is epsilon-strongly graded. Indeed, it is easy to see that S(1,g,2)S(2,g−1,1) =(cid:18) im (ϕ) 0 S(2,g−1,1)S(1,g,2) =(cid:18) 0 0 (cid:19) =(cid:18) A 0 0 (cid:19) 0 B (cid:19) . 0 im (φ) (cid:19) =(cid:18) 0 0 0 0 0 and If we put ǫ(1,e,1) = ǫ(1,g,2) =(cid:18) 1A 0 0 (cid:19) 0 and ǫ(2,e,2) = ǫ(2,g−1,1) =(cid:18) 0 0 1B (cid:19) , 0 then, by Proposition 37 this yields an epsilon-strong (I × G × I)-grading on S. 7. Epsilon-strongly groupoid graded modules In this section, we define epsilon-strongly groupoid graded modules (see Definition 46) and we provide a characterization of them (see Proposition 47) that generalizes a result [15, Theorem I.3.4] previously obtained for strongly group graded modules. At the end of this section, we show (see Proposi- tion 48) that the multiplication maps mg,h : Sg ⊗R Sh → ǫgSgh = Sghǫh−1, for (g, h) ∈ G2, are R-bimodule isomorphisms. In particular this implies that for every g ∈ G1, the sextuple (ǫgR, ǫg−1R, Sg, Sg−1, mg,g−1, mg−1,g) is a set of equivalence data. Throughout this section, S denotes a ring which is graded by a small groupoid G, and we put R = ⊕e∈G0Se. For the entirety of this section also let M be a graded left (right) S-module. Recall that this means that there to each g ∈ G1 is an additive subgroup Mg of M such that M = ⊕g∈G1Mg, as additive groups, and for all g, h ∈ G1, the inclusion SgMh ⊆ Mgh (or MgSh ⊆ Mgh) holds, if (g, h) ∈ G2, and SgMh = {0} (or MgSh = {0}), otherwise. Recall that M is called strongly graded if for all (g, h) ∈ G2 the equality SgMh = Mgh (or MgSh = Mgh) holds. EPSILON-STRONGLY GROUPOID GRADED RINGS 23 Definition 46. We say that M is epsilon-strongly graded if, for each g ∈ G1, SgSg−1 is a unital ideal of Sc(g) such that for all (g, h) ∈ G2 the equality SgMh = SgSg−1Mgh (MgSh = MghSh−1Sh) holds. Proposition 47. The following assertions are equivalent: (a) The ring S is epsilon-strongly graded; (b) Every graded left S-module is epsilon-strongly graded; (c) Every graded right S-module is epsilon-strongly graded. Proof. Suppose that (a) holds. First we show that (b) holds. Let M be a G-graded left S-module and take (g, h) ∈ G2. Then SgSg−1Mgh ⊆ SgMg−1gh = SgMh = SgSg−1SgMh ⊆ SgSg−1Mgh. Next we show that (c) holds. Let M be a G-graded right S-module and take (g, h) ∈ G2. Then MghSh−1Sh ⊆ MgSh = MgShSh−1Sh ⊆ MghSh−1Sh. It is clear that (b) (or (c)) implies (a). (cid:3) Proposition 48. Suppose that S is epsilon-strongly graded, and let {ǫg}g∈G1 be the family of central idempotents of R provided by Proposition 37. Then for all (g, h) ∈ G2 the following assertions hold: (a) For every graded left S-module M the multiplication map mg,h : Sg ⊗Sd(g) Mh → ǫgMgh is an isomorphism of R-bimodules; (b) For every graded right S-module M the multiplication map m′ g,h : Mg ⊗Sd(g) Sh → Mghǫh−1 is an isomorphism of R-bimodules; (c) The multiplication map mg,h : Sg ⊗R Sh → ǫgSgh = Sghǫh−1 is an isomorphism of R-bimodules; (d) For every g ∈ G1, the sextuple (ǫgR, ǫg−1R, Sg, Sg−1, mg,g−1, mg−1,g) is a set of equivalence data. Proof. Take (g, h) ∈ G2. (a) Let M be a G-graded left S-module. From Proposition 47(ii) it follows that mg,h is surjective. Now we show that mg,h is injective. To this end, take a positive integer n and s(i) h ∈ Mh, for i ∈ {1, . . . , n}, i=1 s(i) h ∈ Sg ⊗Sd(g) Mh. Take a g−1 ∈ Sg−1, for j ∈ {1, . . . , m}, such g ∈ Sg and v(j) g ∈ Sg and l(i) g ⊗ l(i) ǫgs(i) g ⊗ l(i) h = n m Xi=1 Xj=1 g v(j) u(j) g−1s(i) g ⊗ l(i) h g ⊗ v(j) u(j) g−1mg,h(x) = 0. g v(j) g−1. Then positive integer m and u(j) such that mg,h(x) = 0, where x = Pn j=1 u(j) that ǫg =Pm Xi=1 Xi=1 g ⊗ v(j) u(j) g ⊗ l(i) s(i) Xj=1 Xi=1 g−1s(i) g l(i) h = h = x = n m = n n m Xj=1 24 EPSILON-STRONGLY GROUPOID GRADED RINGS g,h is surjective. Now we show that m′ (b) Let M be a G-graded right S-module. From Proposition 47(iii) it g,h is injective. To this g ∈ Mg and s(i) h ∈ Sh, for i ∈ {1, . . . , n}, g ⊗ s(i) h ∈ Mg ⊗Sd(g) Sh. Take a h ∈ Sh, for j ∈ {1, . . . , m}, such follows that m′ end, take a positive integer n and m(i) such that m′ positive integer m, and u(j) h−1v(j) g,h(x) = 0, where x = Pn i=1 l(i) h−1 ∈ Sh−1 and v(j) h . Then j=1 u(j) n n m g ⊗ s(i) l(i) h ǫh−1 = g ⊗ s(i) l(i) h u(j) h−1v(j) h g ⊗ s(i) l(i) h = Xi=1 h−1 ⊗ v(j) h u(j) g s(i) l(i) h = Xi=1 Xj=1 h−1 ⊗ v(j) m′ g,h(x)u(j) h = 0. n x = that ǫh−1 =Pm Xi=1 Xi=1 = n Xj=1 m m Xj=1 (c) and (d) follow immediately from (a) or (b). (cid:3) Remark 49. Take g ∈ G1. It is clear from the definition of epsilon-strongly groupoid graded rings that the sextuplet (ǫgR, ǫg−1R, Sg, Sg−1, mg,g−1, mg−1,g) is a set of pre-equivalence data with mg,g−1 and mg−1,g surjective. Thus, injectivity of the maps mg,g−1 and mg−1,g also follow from Proposition 20(a). 8. Generalized Epsilon-crossed products In this section, we introduce generalized epsilon-crossed groupoid prod- ucts (see Definition 53) and we show that they parametrize the family of epsilon-strongly groupoid graded rings (see Proposition 55 and Proposi- tion 56). Throughout this section G denotes a small groupoid with G0 finite. Definition 50. Suppose that F : G → PICcat is a partial functor of inverse categories. For each e ∈ G0 define the ring Ae by F (e) = Ae. For each g ∈ G1 put F (g) = hIg Ac(g) (Pg)Jg unital ideal Ig of Ac(g) and some unital ideal Jg of Ad(g), making Pg an Ig- Jg-bimodule. For the time being, assume that the bimodules Pg, for g ∈ G1, are fixed. From the equality F (g−1) = F (g)∗ it follows by the proof of Ad(g)i for some Ac(g)-Ad(g)-bimodule Pg, some Theorem 28 that Jg = Ig−1 so we may write F (g) = hIg each g ∈ G1, put ǫg = 1Ig . Ac(g) (Pg) Ig−1 Ad(g)i. For Proposition 51. Suppose that (g, h) ∈ G2. Then γPgh(ǫ(gh)−1 ǫh−1) = ǫgǫgh. In particular, ǫgPgh = Pghǫh−1. EPSILON-STRONGLY GROUPOID GRADED RINGS 25 Proof. From (15) it follows that Ac(g) Ac(g) (Pg) (Ig)Ig F (g)F (g−1)F (gh) =hIg =hIg =(cid:20)γIg (ǫgǫgh)Ac(g) =(cid:20)ǫgǫghAc(g) Ac(g) Ac(g) Ad(g) Ig−1 Ad(g)ihIg−1 Ac(g)ihIgh Ac(g) (Pg−1 )Ig (Pgh) (Ig ⊗Ac(g) Pgh) Ac(g) I(gh)−1 Ac(g)ihIgh Ad(h) i γ−1 Pgh Ad(h) (ǫgǫgh)Ad(h) (cid:21) (ǫgǫgh)Ad(h) γ−1 Pgh Ad(h) (cid:21) (Ig ⊗Ac(g) Pgh) (Pgh) I(gh)−1 Ad(h) i and I(gh)−1 Ac(g) (Pgh) Ad(h) ihIh−1 F (gh)F (h−1)F (h) =hIgh Ad(h) ihIh−1 =hIgh =(cid:20)γPgh (ǫ(gh)−1 ǫh−1 )Ac(g) (Pgh) I(gh)−1 Ac(g) Ac(g) Ad(h) Ad(h) (Ph) Ac(h) (Ph−1)Ih (Ih−1) Ac(h)ihIh Ad(h)i Ih−1 (Pgh ⊗Ad(h) Ih−1) ǫ(gh)−1 ǫh−1 Ad(h) Ad(h) Ih−1 Ad(h)i (cid:21) . Thus, from the equality F (g)F (g−1)F (gh) = F (gh)F (h−1)F (h) and Propo- sition 18, we get that γPgh(ǫ(gh)−1ǫh−1) = ǫgǫgh. Finally, ǫgPgh = ǫgǫghPgh = γPgh(ǫ(gh)−1ǫh−1)Pgh = Pghǫ(gh)−1ǫh−1 = Pghǫh−1. (cid:3) Remark 52. Let F : G → PICcat be a partial functor of inverse categories. Then for every (g, h) ∈ G2 there are Ac(g)-Ad(h)-bimodule isomorphisms Pg ⊗Ad(g) Ph ∼= Pg ⊗Ad(g) Pg−1 ⊗Ac(g) Pgh ∼= Ig ⊗Ac(g) Pgh ∼= ǫgPgh. Definition 53. Let F : G → PICcat be a partial functor of inverse cate- gories. A partial factor set associated with F is a family f = {fg,h (g, h) ∈ G2}, where each fg,h : Pg ⊗Ad(g) Ph → ǫgPgh = Pghǫh−1 is an isomorphism of Ac(g)-Ad(h)-bimodules, making the following diagram commutative Pg ⊗Ad(g) Ph ⊗Ad(h) Pr idPg ⊗fh,r −−−−−−→ Pg ⊗Ad(g) Phrǫr−1 (24) fg,h⊗idPry ǫgPgh ⊗Ad(h) Pr fgh,r−−−−→ fg,hr y ǫgPghrǫr−1 for all (g, h, r) ∈ G3. If f is a partial factor set associated with F , then we define the partial generalized epsilon-crossed product (F, f ) as the additive group ⊕g∈G1Pg with multiplication defined by the biadditive extension of the relations x · y = fg,h(x ⊗ y), if (g, h) ∈ G2, and x · y = 0, otherwise, for all x ∈ Pg and y ∈ Ph and all g, h ∈ G1. It is clear that if for each g ∈ G1 we put (F, f )g = Pg, then (F, f ) is a groupoid graded ring. Remark 54. We notice that the notion of partial factor sets already exists in the literature, in a close but different sense. Indeed, partial projective representations and their corresponding factor sets where introduced in [6]. 26 EPSILON-STRONGLY GROUPOID GRADED RINGS Later, in [7], these factor sets were called partial factor sets. For a detailed account of these notions, we refer the reader to the survey [18]. Proposition 55. If F : G → PICcat is a partial functor of inverse cate- gories, then the ring (F, f ) is epsilon-strongly G-graded. Proof. Put S = (F, f ). By (24) the multiplication is associative. By the definition of the multiplication, for all (g, h) ∈ G2, the equality SgSh = ǫgSgh = Sghǫh−1 holds. All that is left to show is that S has a multiplicative identity element. Take e ∈ G0 and put ce = fe,e(1Ae ⊗ 1Ae). Take ae ∈ Ae. Then, since fe,e is an Ae-bimodule homomorphism, it follows that aece = aefe,e(1Ae ⊗ 1Ae) = fe,e(ae ⊗ 1Ae) = fe,e(1Ae ⊗ ae) = fe,e(1Ae ⊗ 1Ae)ae = ceae. Thus, ce ∈ Z(Ae). Since fe,e is surjective there are a, a′ ∈ Ae such that fe,e(a ⊗ a′) = 1Ae. By Ae-bilinearity it follows that aa′ce = ceaa′ = 1Ae. Hence ce ∈ U (Z(Ae)). Now set ne = c−1 e . Then fe,e(ne ⊗ ne) = ne. Hence ne is a multiplicative identity element of S. In fact, take g ∈ G1 and x ∈ Pg. Then there is y ∈ Pg such that x = fc(g),g(nc(g) ⊗ y). Thus, by (24), we get that n :=Pe∈G0 n · x = nc(g) · x = fc(g),g(nc(g) ⊗ x) = fc(g),g(nc(g) ⊗ fc(g),g(nc(g) ⊗ y)) = fc(g),g(fc(g),c(g)(nc(g) ⊗ nc(g)) ⊗ y) = fc(g),g(nc(g) ⊗ y) = x. Analogously, x · n = x. (cid:3) Proposition 56. If S is an epsilon-strongly graded ring, then there is a partial functor of inverse categories F : G → PICcat and a partial factor set f associated with F such that S = (F, f ). ǫgSc(g) Sc(g) ǫg−1 Sd(g) Sd(g) ], g ∈ G1, and Proof. Define F : G → PICcat by F (g) = [ a partial factor set f associated with F by the multiplication maps fg,h : Sg ⊗Sd(g) Sh → ǫgSgh = Sghǫh−1 for (g, h) ∈ G2. The claim now follows immediately from Proposition 48(c). (cid:3) Definition 57. Let F and F ′ be partial functors of inverse categories from G to PICcat that coincide on G0. Take partial factor sets f and f ′ Sg associated with F and F ′ respectively and put F (g) = hIg F ′(g) = hI ′ Ad(g)i, Ad(g)i, g ∈ G1. A morphism from f to f ′ is a family g is an Ac(g)-Ad(g)-bimodule homo- α = (αg)g∈G1, where each αg : Pg → P ′ morphism, such that the diagram (Pg)Jg J ′ g g Ac(g) (P ′ g) Ac(g) (25) Pg ⊗Ad(g) Ph αg⊗αhy g ⊗Ad(g) P ′ P ′ h is commutative for all (g, h) ∈ G2. fg,h−−−−→ ǫgPgh αgh y f ′ gh−−−−→ ǫgP ′ gh EPSILON-STRONGLY GROUPOID GRADED RINGS 27 Lemma 58. With the above notation, a morphism α from F to F ′ induces a homomorphism of graded rings α from (F, f ) to (F ′, f ′). Moreover, if each αe, e ∈ G0, is surjective, then α(1) = 1. The map α is an isomorphism if and only if each αe, e ∈ G0, is bijective. Proof. Similar to the proof of [13, Lemma 4.2]. (cid:3) Proposition 59. The isomorphism class of (F, f ) does not depend on the choice of the bimodules Pg. Proof. Put F (g) = hIg Ac(g) (Pg)Jg there exists an Ac(g)-Ad(g)-bimodule homomorphism αg : Pg → P ′ now put f ′ associated with F and (25) commutes. g. If we gh ◦ fgh ◦ (αg ⊗ αh), for (g, h) ∈ G2, then f ′ is a factor set g,h = α−1 (cid:3) Ad(g)i = hIg Ac(g) (P ′ g)Jg Ad(g)i, for g ∈ G1. Then 9. Partial Cohomology of groupoids In this section, we extend the construction of a partial cohomology theory for partial actions of groups on commutative monoids, from [5], to partial actions of groupoids. We follow closely the presentation and the proofs in [5]. Partial actions of groupoids on rings were first studied in [2]. Partial actions of categories on sets and topological spaces have been introduced in [16]. For the rest of this section, let G be a groupoid and suppose that B is the product of a collection of commutative semigroups {Be}e∈G0. Remark 60. Let {θg}g∈G1 be a unital partial groupoid action of G on B1 Note that if e ∈ G0 and C and D are unital ideals of Be, then C ∩ D = CD so it follows from [2, Lemma 1.1] that the properties (G2) and (G3) can be replaced by (G2′) if (g, h) ∈ G2, then θg(Bg−1Bh) = BgBgh, and (G3′) if (g, h) ∈ G2 and x ∈ Bh−1Bh−1g−1, then θg(θh(x)) = θgh(x), respectively. A unital partial G-module is a pair (B, θ), where B is a commutative monoid and θ is a unital partial action of G on B. Definition 61. A morphism ψ : (B, θ) → (B′, θ′) of partial G-modules is a set of monoid homomorphisms ψ = {ψc(g) : Bc(g) → B′ c(g)}g∈G such that • ψc(g)(Bg) ⊆ B′ g, • θ′ g ◦ ψc(g) = ψc(g) ◦ θg on Bg−1, for all g ∈ G1. Recall that, if n ≥ 2, then we let Gn denote the set of all (g1, . . . , gn) ∈ i=1G1 that are composable, that is, such that for every i ∈ {1, . . . , n − 1} ×n the relation d(gi) = c(gi+1) holds. 1In the sense of Definition 39, but in this case {θg}g∈G1 is a family of semigroup isomorphisms. 28 EPSILON-STRONGLY GROUPOID GRADED RINGS We denote by pM od(G) the category of (unital) partial G-modules. Some- times, for convenience (B, θ) will be replaced by B. Suppose that B ∈ pM od(G). An n-cochain of G with values in B is a function f : Gn → B such that for every (g1, . . . , gn) ∈ Gn, f (g1, . . . , gn) is an invertible element of B(g1,...,gn) = Bg1Bg1g2 · · · Bg1···gn. Denote the set of n-cochains by C n(G, B). We let C 0(G, B) denote U (B), the group of units in B. Proposition 62. Let B ∈ pM od(G). Then C n(G, B) is an abelian group under pointwise multiplication. Proof. This is clear if n = 0. Now suppose that n ≥ 1. Define en ∈ C n(G, B) in the following way. Given (g1, . . . , gn) ∈ Gn, put en(g1, . . . , gn) = It is clear that en is an identity element of C n(G, B). 1g11g1g2 · · · 1g1···gn. Given f ∈ C n(G, B) and (g1, . . . , gn) ∈ Gn, put f −1(g1, . . . , gn) = f (g1, . . . , gn)−1, where the inverse is taken in B(g1,...,gn). en. It is clear that f f −1 = f −1f = (cid:3) Definition 63. Let B ∈ pM od(G) and let n be a non-negative integer. Define a map δn : C n(G, B) → C n+1(G, B) in the following way. Take b = (be)e∈ob(G) ∈ U (B) and g ∈ G1. Put δ0(b)(g) = θg(1g−1bd(g))b−1 c(g). Now suppose that n is a positive integer. Take f ∈ C n(G, B) and (g1, . . . , gn+1) in Gn+1. Put δn(f )(g1, . . . , gn+1) = θg1 (1g−1 1 f (g2, . . . , gn+1)) n Yi=1 f (g1, . . . , gigi+1, . . . gn+1)(−1)i! f (g1, . . . , gn)(−1)n+1 . Adapting the proofs of [6, Proposition 1.5] and [6, Proposition 1.7] to our situation, we get the following. Proposition 64. Suppose that B ∈ pM od(G), and that n is a non-negative integer. Then the following assertions hold: (a) The map δn is a well-defined homomorphism of groups C n(G, B) → C n+1(G, B) satisfying δn+1δn = en+2; (b) The map sending B to the sequence {δn : C n(G, B) → C n+1(G, B)}n∈N is a functor from pM od(G) to the category of complexes of abelian groups. Definition 65. Let B ∈ pM od(G) and let n be a positive integer. The map δn is called a coboundary homomorphism. We define the abelian groups Z n(G, B) = ker(δn), Bn(G, B) = im(δn−1). The quotient group H n(G, B) = Z n(G, B)/Bn(G, B) is called the nth cohomology group of G with values in B. We put H 0(G, B) = Z 0(G, B) = ker(δ0). Let G be a groupoid. Denote by I(X)cat the subcategory of BIJcat having as objects the collection (Xe)e∈G0 of abelian semigroups and morphisms EPSILON-STRONGLY GROUPOID GRADED RINGS 29 Xg−1 Xd(g) fg [Xg ], where Xg is a unital ideal of Xc(g) and fg : Xg−1 → Xg, is a Xc(g) monoid isomorphism for all g ∈ G1. The composition in I(X)cat is defined in the same way as in ISOcat, the map ∗ : (I(X)cat)1 → (I(X)cat)1 is also defined by restriction of the map ∗ defined on (BIJcat)1. It follows from Proposition 13 that I(X)cat is an inverse category. Xe, then X ∈ pM od(G), if and only if, there is a partial functor of inverse categories F : G → I(X)cat. Proposition 66. If G is a groupoid and X =Qe∈G0 Proof. First we show the "only if" statement. Let {θg : Xg−1 → Xg}g∈G1 be a partial action of G on X and define F : G → I(X)cat, by F (g) = [Xg ], for g ∈ G1, and Fe = Xe, for any e ∈ G0. Then F is a partial Xc(g) functor of inverse categories. Xg−1 Xd(g) θg of inverse categories. Put F (g) = [Xg Now we show the "if" statement. Let F : G → I(X)cat be a partial functor ], for g ∈ G1. We shall show Xe. It is clear that for each g ∈ G1, Bg is an ideal of Bc(g), Bc(g) is an ideal of B and θg : Bg−1 → Bg is a monoid isomorphism. Now we check (G1), (G2) and (G3) from Definition 38. that the family {θg}g∈G1 gives a partial action of G on X =Qe∈G0 Xg−1 Xd(g) Xc(g) θg (G1): Let e ∈ G0. Then F (e) = hXe Xe on Xe. (θe)Xe Xei , and θe is the identity map (G2)-(G3): Let (g, h) ∈ G2. Then F (g)F (h) = F (gh)F (h)∗F (h), which h ◦ θh = θgh ◦ idXh−1 , which (cid:3) by the definition of F implies θg ◦ θh = θgh ◦ θ−1 in turn implies that θgh is an extension of θg ◦ θh. Let F : G → PICcat be a partial functor of inverse categories. For each e ∈ G0 define the ring Ae by F (e) = Ae. Take g ∈ G1. Put F (g) = Ig−1 Ac(g) (Pg) hIg Ad(g)i. Define Bg = Z(Ig) and B =Qe∈ob(G) Z(Ae). Proposition 67. With the above notation we have B ∈ pM od(G). Proof. By Proposition 31 there is a partial functor of inverse categories L : PICcat → ISOCcat. From Proposition 25 it follows that l = L ◦ F : G → ISOCcat is a partial functor of inverse categories. But l(g) = Bc(g) hBg (γPg ) Bg−1 Bd(g)i , so l : G → I(X)cat and hence we get that B ∈ pM od(G). (cid:3) 10. Proof of the main result In this section, we prove Theorem 3 which was stated in Section 1. For the convenience of the reader, we shall now recall its exact wording. Theorem. If G is a groupoid, F : G → PICcat is a partial functor of inverse categories and f is a partial factor set associated with F , then the map H 2(G, U (Z(A))) → C(A, F ), defined by [q] 7→ qf , is bijective. 30 EPSILON-STRONGLY GROUPOID GRADED RINGS In order to prove the above theorem, we need the following result. Proposition 68. Let f and f ′ be factor sets associated with F . (a) If q ∈ Z 2(G, B), then f q is a factor set associated with F . (b) There is q ∈ Z 2(G, B) such that f ′ = qf . (c) A cocycle q ∈ Z 2(G, B) belongs to B2(G, B) if and only if there is a graded ring isomorphism α from (F, f ) to (F, qf ) such that each for all g ∈ G1 the graded restriction αg to Pg is an Ac(g)-Ad(g)-bimodule isomorphism. (d) The map from Z 2(G, B) to the collection of factor sets associated with F , defined by q 7→ qf , is bijective. Proof. (a) Put f ′′ = f q. We need to verify that (24) commutes for f ′′. Take (g, h, p) ∈ G3, x ∈ Pg, y ∈ Ph and z ∈ Pp. Then (f ′′ g,h ⊗ idPp ))(x ⊗ y ⊗ z) = qgh,pqg,h(fgh,p ◦ (fg,h ⊗ idPp ))(x ⊗ y ⊗ z) gh,p ◦ (f ′′ = (qg,hpγPg (qh,p))(fg,hp ◦ (idPg ⊗ fh,p))(x ⊗ y ⊗ z) = (qg,hpγPg (qh,p))fg,hp(x ⊗ fh,p(y ⊗ z)) = f ′′ = f ′′ = f ′′ = (f ′′ g,hp(γPg (qh,p)x ⊗ fh,p(y ⊗ z))) g,hp(xqh,p ⊗ fh,p(y ⊗ z)) g,hp(x ⊗ f ′′ g,hp ◦ (idPσ ⊗ f ′′ h,p))(x ⊗ y ⊗ x). h,p(y ⊗ z)) g,h ◦f −1 (b) Take (g, h) ∈ G2. Then f ′ g,h)(x) = qσ,τ x, for x ∈ Pgh. By (24) it follows that q ∈ Z 2(G, B). g,h is an Ac(σ)-Ad(σ)-bimodule automor- phism of Pgh. Hence, by Proposition 21, there is qg,h ∈ U (Z(Ic(g))) such that (f ′ g,h ◦ f −1 (c) Suppose now that q ∈ B2(G, A). Then there is c ∈ C 1(G, A) such that for all (g, h) ∈ G2 it follows that qg,h = γPg (ch)cgc−1 gh . Define a map α from (F, qf ) to (F, f ), by α(x) = cgx, for x ∈ Pg. If x ∈ Pg, y ∈ Ph and (g, h) ∈ G2, then α(xy) = cghqg,hfg,h(x ⊗ y) = q−1 g,hγPg (ch)cgqg,hfg,h(x ⊗ y) = fg,h(γPg x ⊗ chy) = α(x)α(y). Clearly, for all g ∈ G1, the map αg is an Ac(g)-Ad(g)-bimodule isomorphism. On the other hand, suppose that there is an isomorphism of graded rings β from (F, qf ) to (F, f ) such that for all g ∈ G1 the map βg is an Ac(g)- Ad(g)-bimodule isomorphism. From Proposition 21 it follows that for each g ∈ G1 there is dg ∈ U (Z(Ic(g))) such that for all x ∈ Pg the equation βg(x) = dgx holds. Therefore, for all x ∈ Pg, y ∈ Ph and all (g, h) ∈ G2, we get that β(xy) = β(x)β(y) ⇔ dghqg,hfg,g(x ⊗ y) = fg,h(dgx ⊗ dhy) ⇔ dghqg,hfg,h(x ⊗ y) = dgγPg (dh)fg,h(x ⊗ y). Thus, q ∈ B2(G, A). (d) This follows from (a), (b) and (c). (cid:3) Definition 69. If f and f ′ are factor sets associated with F , then we write (F, f ) ≈ (F, f ′) if there is an isomorphism of graded rings from (F, f ) to (F, f ′) such that each graded restriction to Pg, g ∈ G1, is an Ac(g)-Ad(g)- bimodule isomorphism. Let C(A, F ) denote the collection of equivalence classes of generalized groupoid epsilon-crossed products (F, f ) modulo ≈, where f runs over all factor sets associated with F . EPSILON-STRONGLY GROUPOID GRADED RINGS 31 Proof of Theorem 3. This follows immediately from Proposition 68. (cid:3) References [1] D. Bagio, D. Flores and A. Paques, Partial actions of ordered groupoids on rings. J. Algebra Appl. 9(3), 501 -- 517 (2010). [2] D. Bagio and A. Paques, Partial Groupoid actions: globalization, Morita theory and Galois theory, Comm. Algebra 40(10), 3658 -- 3678 (2012). [3] H. Bass, Algebraic K-theory, Benjamin, New York (1968). [4] M. Dokuchaev and R. Exel, Associativity of crossed products by partial actions, en- veloping actions and partial representations, Trans. Amer. Math. Soc. 357(5), 1931 -- 1952 (2005). [5] M. Dokuchaev and M. Khrypchenko, Partial cohomology of groups, J. Algebra 427, 142 -- 182 (2015). [6] M. Dokuchaev and B. Novikov, Partial projective representations and partial actions, J. Pure Appl. Algebra 214(3), 251 -- 268 (2010). [7] M. Dokuchaev and B. Novikov, Partial projective representations and partial actions II, J. Pure Appl. Algebra 216(2), 438 -- 455 (2012). [8] M. Dokuchaev, A. Paques and H. Pinedo, Partial Galois cohomology, extensions of the Picard group and related isomorphisms (expanded version), arXiv:1804.03762 [math.RA]. [9] R. Exel, Circle actions on C ∗-algebras, partial automorphisms and generalized Pimsner-Voiculescu exact sequences, J. Funct. Anal. 122(3), 361 -- 401 (1994). [10] D. Gon¸calves and G. Yoneda, Free path groupoid grading on Leavitt path algebras, Internat. J. Algebra Comput. 26(6), 1217 -- 1235 (2016). [11] R. Hazrat, The graded structure of Leavitt path algebras, Israel J. Math. 195(2), 833 -- 895 (2013). [12] T. Kanzaki, On generalized crossed product and Brauer group, Osaka J. Math. 5, 175 -- 188 (1968). [13] P. Lundstrom, The category of groupoid graded modules, Colloq. Math. 100(4), 195 -- 211 (2004). [14] P. Lundstrom, Strongly groupoid graded rings and cohomology, Colloq. Math. 106(1), 1 -- 13 (2006). [15] C. Nastasescu and F. Van Oystaeyen, Graded Ring Theory, North-Holland Publishing Co., Amsterdam-New York (1982). [16] P. Nystedt, Partial category actions on sets and topological spaces, Comm. Algebra 46(2), 671 -- 683 (2018). [17] P. Nystedt, J. Oinert and H. Pinedo, Epsilon-strongly graded rings, separability and semisimplicity, J. Algebra 514, 1 -- 24 (2018). [18] H. Pinedo, Partial projective representations and the partial Schur multiplier: a survey, Bol. Mat. 22(2), 167 -- 175 (2015). [19] J. Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathematics, 793, Springer, Berlin (1980). [20] J. Westman, Groupoid theory in algebra, topology and analysis, University of Califor- nia at Irvine, (1971).
1806.11355	1	1806	2018-06-29T11:16:39	The structured Gerstenhaber problem (II)	[ "math.RA" ]	Let $b$ be a non-degenerate symmetric (respectively, alternating) bilinear form on a finite-dimensional vector space $V$, over a field with characteristic different from $2$. In a previous work, we have determined the maximal possible dimension for a linear subspace of $b$-alternating (respectively, $b$-symmetric) nilpotent endomorphisms of $V$. Here, provided that the cardinality of the underlying field be large enough with respect to the Witt index of $b$, we classify the spaces that have the maximal possible dimension. Our proof is based on a new sufficient condition for the reducibility of a vector space of nilpotent linear operators. To illustrate the power of that new technique, we use it to give a short new proof of the classical Gerstenhaber theorem on large vector spaces of nilpotent matrices (provided, again, that the cardinality of the underlying field be large enough).	math.RA	math	The structured Gerstenhaber problem (II) Cl´ement de Seguins Pazzis∗ June 6, 2021 Abstract Let b be a non-degenerate symmetric (respectively, alternating) bilinear form on a finite-dimensional vector space V , over a field with characteristic different from 2. In a previous work [10], we have determined the maxi- mal possible dimension for a linear subspace of b-alternating (respectively, b-symmetric) nilpotent endomorphisms of V . Here, provided that the car- dinality of the underlying field be large enough with respect to the Witt index of b, we classify the spaces that have the maximal possible dimension. Our proof is based on a new sufficient condition for the reducibility of a vector space of nilpotent linear operators. To illustrate the power of that new technique, we use it to give a short new proof of the classical Ger- stenhaber theorem on large vector spaces of nilpotent matrices (provided, again, that the cardinality of the underlying field be large enough). AMS Classification: 15A30, 15A63, 15A03. Keywords: Symmetric matrices, Skew-symmetric matrices, Nilpotent matrices, Bilinear forms, Dimension, Gerstenhaber theorem. 1 Introduction 1.1 Notation Throughout the article and with the exception of Sections 2 and 3, F denotes a field with characteristic not 2. Let V be a finite-dimensional vector space over F, ∗Universit´e de Versailles Saint-Quentin-en-Yvelines, Laboratoire de Math´ematiques de Ver- sailles, 45 avenue des Etats-Unis, 78035 Versailles cedex, France, [email protected] 1 equipped with a non-degenerate bilinear form b : V × V → F, and assume that b is symmetric or alternating (i.e. skew-symmetric). We say that b is symplectic when it is non-degenerate and alternating. Throughout, orthogonality in V is considered with respect to b, except in Section 3 where b will not be mentioned and dual-orthogonality will be considered instead (of course, no confusion will arise then). A linear subspace X of V is called totally singular for b if X ⊂ X ⊥. The maximal dimension for such a subspace is called the Witt index of b. Finally, we say that b is non-isotropic whenever its Witt index equals zero, or equivalently b(x, x) 6= 0 for all x ∈ V r {0}. An endomorphism u of V is called b-symmetric (respectively, b-alternating) when the bilinear form (x, y) ∈ V 2 7→ b(x, u(y)) is symmetric (respectively, alternating). The set of all b-symmetric endomorphisms is denoted by Sb, while the set of all b-alternating endomorphisms is denoted by Ab. A subset of an F-algebra is called nilpotent when all its elements are nilpo- tent, and the nilindex of a nilpotent element x of such an algebra is denoted by ind(x) (it is the least non-negative integer k such that xk = 0). Given non-negative integers p and n, we denote by Mn,p(F) the vector space of all n by p matrices with entries in F, by Mn(F) the algebra of all square matrices of size n and entries in F (and by 0n and In its zero matrix and its identity matrix, respectively), by Sn(F) the linear subspace of all n by n symmetric matrices, and by An(F) the linear subspace of all n by n skew-symmetric matrices. 1.2 The problem The traditional Gerstenhaber problem consists of the following questions: • What is the maximal possible dimension for a nilpotent linear subspace of End(V )? • What are the spaces with the maximal possible dimension? Remember that a flag F of V is a finite increasing sequence (Fi)0≤i≤p of linear subspaces of V , and such a flag is said to be stable under an endomorphism u if u(Fi) ⊂ Fi for all i ∈ [[0, p]]. The flag (F0, . . . , Fp) is said to be complete if Fp = V and dim Fi = i for all i ∈ [[0, p]]. If F is complete one denotes by NF the set of all endomorphisms u of V such that u(Fi) ⊂ Fi−1 for all i ∈ [[1, p]] (which is equivalent to having u nilpotent and stabilize Fi for all i ∈ [[0, p]]). With that in mind, the Gerstenhaber problem has the following classical answer (with no restriction on the characteristic of the underlying field): 2 Theorem 1.1 (Gerstenhaber (1958), Serezhkin (1982)). Let V be a nilpotent linear subspace of End(V ). Then, dim V ≤ (cid:18)dim V 2 (cid:19), and equality holds if and only if V = NF for a (unique) complete flag F of V . Gerstenhaber [3] proved this result under the additional requirement that F be of cardinality at least dim V ; Serezhkin [11] later lifted that assumption. In this article, we deal with the so-called structured Gerstenhaber problem, which is the equivalent of the traditional one for a space equipped with a non- degenerate symmetric or alternating bilinear form. In short, we are interested in the nilpotent linear subspaces of Sb and Ab. Here, the maximal dimension is connected to the Witt index of b. A flag F = (F0, . . . , Fp) of V is called partially complete whenever dim Fi = i for all i ∈ [[0, p]]. Such a flag is called b-singular whenever Fp is totally singular for b. When it is partially complete and b-singular, it is called maximal if p equals the Witt index of b. It was shown in [10] that if F = (Fi)0≤i≤ν is a maxi- mal partially complete b-singular flag of V , then the set N Sb,F (respectively, the set N Ab,F ) of all nilpotent u ∈ Sb (respectively, u ∈ Ab) that stabilize F (or, equivalently, u(Fi) ⊂ Fi−1 for all i ∈ [[1, ν]]), is a linear subspace with dimension ν(n − ν) (respectively, ν(n − ν − 1)). Moreover, it was proved in [10] that this dimension is optimal for the structured Gerstenhaber problem: Theorem 1.2 (See [10] theorem 1.7). Let V be an n-dimensional vector space equipped with a non-degenerate symmetric or alternating bilinear form b. Denote by ν the Witt index of b. (a) Let V be a nilpotent linear subspace of Sb. Then, dim V ≤ ν(n − ν). (b) Let V be a nilpotent linear subspace of Ab. Then, dim V ≤ ν(n − ν − 1). This theorem generalizes earlier results of Meshulam and Radwan [7] who tackled the case of an algebraically closed field with characteristic 0 with a symmetric form b. In two specific cases (point (a) when b is alternating, point (b) when b is symmetric), the result can also be seen as a special case of a general result of Draisma, Kraft and Kuttler [2] when the underlying field is algebraically closed and with characteristic different from 2. The present article deals with the problem of characterizing the spaces with maximal dimension. It turns out that, in contrast with Theorem 1.2 -- in the 3 proof of which there is no need to discriminate whether b is symmetric or al- ternating -- the search for spaces having the maximal dimension must be split into four subproblems: spaces of symmetric endomorphisms for a symmetric form, spaces of alternating endomorphisms for an alternating form, spaces of alternating endomorphisms for a symmetric form, and spaces of symmetric en- domorphisms for an alternating form. A complete solution to the former two subproblems appears in [10]: Theorem 1.3 (See [10] theorem 1.8). Let V be an n-dimensional vector space equipped with a non-degenerate symmetric bilinear form b, and let V be a nilpo- tent linear subspace of Sb with dimension ν(n − ν). Then, V = N Sb,F for some maximal partially complete b-singular flag F of V . Theorem 1.4 (See [10] theorem 1.9). Let V be an n-dimensional vector space equipped with a symplectic form b, and let V be a nilpotent linear subspace of Ab with dimension ν(n − ν − 1). Then, V = N Ab,F for some maximal partially complete b-singular flag F of V . Theorem 1.3 generalizes a recent result of Bukovsek and Omladic [1], who limited their discussion to complex numbers (their proof can easily be generalized to any quadratically closed field with characteristic not 2). In the proof of the above two theorems, the strategy was the same one as in [1]: by a trace orthogonality argument, one proved that V is stable under squares; one deduced that V is stable under the Jordan product (u, v) 7→ uv + vu; one used the Jacobson triangularization theorem [4, 8] to deduce that V is triangularizable; the conclusion followed with limited effort. However, in the two remaining cases, i.e. alternating endomorphisms for a symmetric form, and symmetric endomorphisms for an alternating form, the above strategy fails. Indeed, in the first case (respectively, the second one) the square of an element of Ab (respectively, of Sb) is an element of Sb (respectively, of Ab), and can belong to Ab (respectively, to Sb) only if it is zero. It is clear though that in general N Ab,F (respectively, N Sb,F ) does not consist solely of square-zero elements. It was found however that any nilpotent subspace of Ab (respectively, of Sb) with the maximal possible dimension is stable under cubes provided that F > 3 (see lemmas 5.2 and 5.3 of [10], reproduced as Lemma 5.4 in Section 5.2 of the present manuscript). However, a nilpotent subspace that is stable under cubes is not necessarily triangularizable. A classical example, in 4 matrix terms, is the space of all 3 by 3 matrices of the form N (x, y) :=   0 x 0 0 y 0 −y x 0   with (x, y) ∈ F2. One checks that every such matrix has cube zero (and hence this vector space is stable under cubes, and even under any odd power). Yet, it is not triangular- izable since it is easily checked that there is no non-zero vector that lies in the kernel of every matrix of the above type. In the above example it can be shown that there is no non-degenerate sym- metric or alternating bilinear form b on F3 for which all the matrices under hand represent b-symmetric (or b-alternating) endomorphisms in the standard basis. However, the example can be used to construct an example of this kind. Simply, let ε ∈ {1, −1}, and consider the bilinear form b whose matrix in the standard basis of F6 is I3 ε I3 03(cid:21) , (cid:20) 03 and the space V of all endomorphisms of F6 with matrix in the standard basis equal to (cid:20)N (x, y) 03 M ε′ N (x, y)T(cid:21) for ε′ := ε (respectively, ε′ := −ε), some pair (x, y) ∈ F2 and some symmetric (respectively, alternating) matrix M ∈ M3(F). Then, one checks that V is a nilpotent subspace of Sb (respectively, of Ab), and that it is stable under cubes (and, more generally, under any odd power). Worse still, the Witt index of b is ν := 3, and V has dimension exactly ν(n − ν) − 1 (respectively, ν(n − ν − 1) − 1) where n := 6, which is just one unit under the critical dimension of Theorem 1.2. Again, it is not hard to check that V is not triangularizable. Hence, the stability under cubes, although a nice property, is clearly insuffi- cient to obtain the triangularizability of the spaces with maximal dimension. The aim of the present article is to give a partial solution in the yet unresolved cases: assuming that the cardinality of the underlying field is large enough with respect to the Witt index of b, we shall prove that the above characterization still holds. Here are our results: Theorem 1.5. Let V be an n-dimensional vector space over F equipped with a non-degenerate symmetric bilinear form b with Witt index ν, and let V be a 5 nilpotent linear subspace of Ab. Assume that F ≥ min(n, 2ν + 1) and dim V = ν(n − ν − 1). Then V = N Ab,F for some maximal partially complete b-singular flag F of V . Theorem 1.6. Let V be an n-dimensional vector space over F equipped with a symplectic form b, and let V be a nilpotent linear subspace of Sb. Set ν := n 2 and assume that F ≥ n and dim V = ν(n − ν). Then V = N Sb,F for some maximal partially complete b-singular flag F of V . The lower bounds on the cardinality of F are relevant because they are the lowest ones so that we can ensure that the nilpotency of a linear subspace of Ab (in Theorem 1.5) or Sb (in Theorem 1.6) be preserved in extending the field of scalars; they are connected to the maximal possible nilindex in Ab and in Sb. See Sections 4.3 and 5.3 for details. At this point, we should note that the above two results are already known in the case when F is algebraically closed. Indeed, in the first case (respectively, the second one) Ab (respectively, Sb) is a Lie subalgebra of End(V ), and it is isomorphic to the Lie algebra of the reductive algebraic group O(b) (respectively, Sp(b)). Moreover, the set of all unipotent orthogonal automorphisms (respec- tively, of all unipotent symplectic automorphisms) that stabilize a given maximal partially complete b-singular flag F is a Borel subgroup of that algebraic group. A general theorem of Draisma, Kraft and Kuttler [2] then yields Theorems 1.5 and 1.6 in that case. However, one must check a tedious condition (named con- dition (C) in [2]), and the proof uses deep results from the theory of algebraic groups. In the special case when b has the maximal possible Witt index (that is, ⌊n/2⌋), one can show through an extension of scalars argument that their result yields ours (this is not entirely straightforward though, we will discuss this issue in Section 5.3). In particular, Theorem 1.6, with its cardinality assumption, is a direct consequence of the general theorem of Draisma et al. However, this still leaves open the problem of proving Theorem 1.5 for small Witt indices. 1.3 Strategy, and structure of the article The aim of the present work is to give a (mostly) self-contained elementary proof of Theorems 1.5 and 1.6. The proof involves a new technique for the study of spaces of nilpotent matrices, which uses elementary algebraic geometry: the idea is that, given a nilpotent subspace V of End(V ) with maximal nilindex p, one looks at the behaviour of the space Im up−1 when u varies in V. This idea is in- spired by the topic of spaces of bounded rank operators, where similar techniques 6 have been successful. Combined with standard trace orthogonality techniques, this yields a very interesting sufficient condition for the reducibility of a nilpo- tent subspace (Lemma 2.5). That condition will be the key to prove the above theorems, and we will illustrate its power by giving a new proof of the classical Gerstenhaber theorem under the mild assumption that the underlying field be of cardinality at least the dimension of the vector space V under consideration. The remainder of the article is organized as follows. In Section 2, we recall the classical trace lemma for nilpotent subspaces of operators (Lemma 2.1), we obtain key results on the spaces Im up−1, when p is the maximal nilindex in a nilpotent subspace V and u varies in V; we conclude the section by obtaining the key Reducibility Lemma (Lemma 2.5). The tools of Section 2 are used in Section 3 to yield an efficient new proof of Gerstenhaber's theorem for fields with large cardinality: there, the novelty resides in the analysis of the spaces with the maximal dimension. In the next two sections, we lay out some groundwork for the structured Ger- stenhaber problem. In Section 4, we collect results on individual b-symmetric or b-alternating endomorphisms: in addition to very basic results and consid- erations of endomorphisms of small rank (the b-symmetric tensors and the b- alternating tensors), we study the maximal nilindex in a space of type N Sb,F or N Ab,F , and we finish with deeper results on the canonical forms for b-symmetric or b-alternating endomorphisms: those results are needed in the most technical parts of the proof of Theorem 1.5. In Section 5, we deal more closely with the structured Gerstenhaber problem: we recall the inductive proof of Theorem 1.2 that already appeared in [10], and as a by-product of that proof we obtain an important property of spaces with the maximal dimension (the Strong Orthogo- nality Lemma, see Lemma 5.1), along with results that allow one to perform an inductive proof. From the stability under cubes that we have already mentioned, we will also derive another stability result (Corollary 5.5) that will turn out to be essential in the proof of Theorem 1.5. We will close Section 5 with a discussion on the extension of scalars: it will be shown in particular that, in order to prove Theorems 1.5 and 1.6, it suffices to do so when the underlying field is infinite. This will allow us to use elementary ideas from algebraic geometry. Having paved the way in Sections 4 and 5, we will then be ready to prove Theorems 1.5 and 1.6. The easier one is by far the latter, so we will start by tackling it (Section 6), and we will finish with the former (Section 7). In both cases, we consider an infinite underlying field, and the proof works by induction on the dimension of the underlying vector space V : the key is to prove the 7 existence of a non-zero isotropic vector x that is annihilated by all the elements of V. 2 General results on spaces of nilpotent matrices Here, V denotes a finite-dimensional vector space over an arbitrary field F (here, there is no restriction of characteristic). 2.1 The trace lemma One of the main keys for analyzing nilpotent spaces of endomorphisms is the so-called trace lemma. It appeared first in [6], where it was proved for fields with characteristic zero. The result was extended in [5], and a variation of the proof was given in [10]. Lemma 2.1 (Trace Lemma). Let V be a nilpotent linear subspace of End(V ). Let u and v belong to V, and let k be a non-negative integer such that k < F. Then, tr(ukv) = 0. 2.2 The generic nilpotency index Here, we assume that V 6= {0}. Definition 1. Let V be a nilpotent subspace of End(V ). The greatest nilindex among the elements of V is called the generic nilindex of V and denoted by ind(V). Then, with p := ind(V), we set V • := [v∈V Im vp−1 and K(V) := span(V •). We say that V is pure when rk up−1 ≤ 1 for all u ∈ V. In other words, V is pure if and only if every element of V with nilindex p has exactly one Jordan cell of size p. Notation 2. For x ∈ V and a linear subspace V of End(V ), we set Vx := {v(x) v ∈ V}, which is a linear subspace of V . 8 Let V be a nilpotent linear subspace of End(V ). Given x ∈ V r{0}, note that Vx is linearly disjoint from Fx: indeed, for all v ∈ V we have v(x) 6∈ Fx r {0} otherwise x would be an eigenvector of v with a non-zero associated eigenvalue. Here is an important observation: Proposition 2.2. Let V be a nilpotent subspace of End(V ) with generic nilindex p. Assume that F ≥ p. Let x ∈ V • and u ∈ V be such that x ∈ Im up−1. Then, Vx ⊂ u(K(V)). Proof. Let v ∈ V. We write x = up−1(y) for some y ∈ V . Let λ ∈ F. We write (u + λv)p−1 = p−1 Xk=0 λkuk where the operators u0, . . . , up−1 belong to End(V ), and more precisely u1 = p−2 Xk=0 ukv up−2−k. Then, for all λ ∈ F, we note that (u + λv)p−1(y) belongs to V •. Now, let ϕ be a linear form that vanishes everywhere on K(V). Hence, ∀λ ∈ F, p−1 Xk=0 λkϕ(uk(y)) = 0. Since F ≥ p, this yields in particular ϕ(u1(y)) = 0. Varying ϕ, we obtain u1(y) ∈ K(V). Finally, since p = ind(V) the (vector-valued) polynomial function λ 7→ (u + λv)p = p−1 Xk=0 λku′ k vanishes everywhere on F, and again we deduce that u′ vup−1, and hence v(x) = v(up−1(y)) = u(−u1(y)) ∈ u(K(V)). 1 = 0. Here, u′ 1 = uu1 + The next results will be used in the most technical parts of the proof of Theorems 1.5 and 1.6. 9 Lemma 2.3. Assume that F is infinite, and let V be a pure nilpotent subspace of End(V ). Then, V • cannot be covered by finitely many proper linear subspaces of K(V). Proof. Assume otherwise, and consider a minimal covering (F1, . . . , FN ) of V • by linear hyperplanes of K(V). Hence, we have linear forms ϕ1, . . . , ϕN on K(V) such that Hk = Ker ϕk for all k ∈ [[1, N ]] and ∀x ∈ V •, N Yk=1 ϕk(x) = 0. Moreover, as this covering is minimal we can find a vector x ∈ V • r {0} that belongs to none of H2, . . . , HN . Choose u ∈ V such that Im up−1 = Fx. Let y ∈ V • r {0}, and choose v ∈ V such that Fy = Im vp−1. Since Ker up−1 and Ker vp−1 are proper linear subspaces of V , they do not cover V and hence we can find a vector z ∈ V r (Ker up−1 ∪ Ker vp−1); then, up−1(z) is non-zero and collinear with x, and vp−1(z) is non-zero and collinear with y. Set The polynomial function γ : λ ∈ F 7→ (cid:0)(1 − λ)u + λv(cid:1)p−1 (z) ∈ V •. λ 7→ N Yk=1 ϕk(cid:0)γ(λ)(cid:1) is identically zero. Since F is infinite, this yields an index k ∈ [[1, p]] such that λ 7→ ϕk(γ(λ)) is identically zero. However, γ(0) is non-zero and collinear with x, and γ(1) is non-zero and collinear with y. It follows from the first point that we must have k = 1, and then we deduce from the second one that y ∈ H1. Varying y yields V • ⊂ H1, and hence K(V) ⊂ H1. This is absurd. The above lemma will be used through the following (essentially equivalent) form: Lemma 2.4 (Linear Density Lemma). Assume that F is infinite, and let V be a pure nilpotent subspace of End(V ). Let F1, . . . , FN be proper linear subspaces of K(V). Then, there exists a basis of K(V) consisting of vectors of V • r Fk. N Sk=1 Proof. Assuming otherwise, some linear hyperplane H of K(V) would include V • r N Sk=1 V •, contradicting Lemma 2.3. Fk, and hence the proper linear subspaces F1, . . . , FN , H would cover 10 2.3 Application to a reducibility lemma for nilpotent subspaces of operators The key to most of the theorems in this article is the following lemma, which relies critically on Proposition 2.2: Lemma 2.5 (Reducibility Lemma). Let V be a finite-dimensional vector space over a field F. Let V be a nilpotent linear subspace of End(V ) such that F ≥ ind(V). Let x ∈ V • r {0} be such that (C) : K(V) ⊂ Fx ⊕ Vx. Then, Vx = {0}. In practice, condition (C) is not always easy to obtain, but we will get it with little effort in the case of the traditional Gerstenhaber theorem. Proof. Set p := ind(V) and choose u ∈ V such that x ∈ Im up−1. For a vector y ∈ V , the height of y (with respect to u) is defined as the supremum of the integers k ≥ 0 such that y ∈ Im uk (hence, it is +∞ if y = 0, and at most p − 1 otherwise). Since K(V) contains the non-zero element x, we can choose an element y ∈ K(V) with minimal height h < +∞. Condition (C) gives an operator v ∈ V and a scalar λ such that y = v(x) + λx. Proposition 2.2 then yields a vector z ∈ K(V) such that v(x) = u(z), whence y = u(z) + λx. Since the height of x is at least p − 1 and the one of u(z) is at least h + 1, we get that the height of y would be at least h + 1 if h < p − 1, which would be a contradiction in that case. Hence, h = p − 1, leading to K(V) ⊂ Im up−1. We conclude, by using Proposition 2.2 once more, that Vx ⊂ u(K(V)) ⊂ u(cid:0)Im up−1(cid:1) = {0}. 3 A new proof of Gerstenhaber's theorem for fields with large cardinality In this section, we give a new proof of Gerstenhaber's theorem for fields whose cardinality is large enough with respect to the dimension of the underlying space. 11 The proof takes advantage of the Reducibility Lemma. Here, F denotes an arbitrary field (possibly of characteristic 2). The proof works by induction. Gerstenhaber's theorem is obviously true for spaces with dimension at most 1, so in the rest of the section we fix an integer n > 1 and we assume that Gerstenhaber's theorem holds whenever the dimension of the underlying space is less than n and less than or equal to the cardinality of F. We let V be an n-dimensional vector space over a field F with at least n elements, and V be a nilpotent linear subspace of End(V ). The first part of the proof, which deals with the inequality, is classical. It can be found in [5]: we have to reproduce it because it is a necessary step towards the second part. 3.1 Proof of the dimension inequality Some preliminary work is required here. Denote by V ⋆ the dual space of V . Let f ∈ V ⋆ and x ∈ V . The endomorphism y ∈ V 7→ f (y) x is denoted by f ⊗ x. Its trace equals f (x); it is nilpotent if and only if f (x) = 0. Now, fix x ∈ V r {0}. We set Through f ∈ V ⋆ 7→ f ⊗ x, the space V x is isomorphic to the linear subspace of V consisting of its elements whose range is included in Fx. V x := (cid:8)f ∈ V ⋆ : f ⊗ x ∈ V(cid:9). Lemma 3.1. Let x ∈ V r {0}. The space V x is dual-orthogonal to Fx ⊕ Vx. Proof. Let f ∈ V x and u ∈ V. Then, (f ⊗ x) ◦ u = (f ◦ u) ⊗ x has trace f (u(x)). The Trace Lemma applied to k = 1 yields f (u(x)) = 0. Besides, f (x) = 0 because f ⊗ x is nilpotent, and the conclusion follows. Now, we fix an arbitrary vector x ∈ V r {0}. Denote by U the kernel of u ∈ V 7→ u(x) ∈ Vx. Then, every element u of U induces a nilpotent endomorphism u of V /Fx, and V mod x := {u u ∈ U} is a nilpotent linear subspace of End(V /Fx). By induction, we have dim(V mod x) ≤ (cid:18)n − 1 2 (cid:19)· 12 Denote by U ′ the kernel of u ∈ U 7→ u ∈ V mod x. Obviously, U ′ is the set of all u ∈ V with range included in Fx, whence U ′ = {f ⊗ x f ∈ V x} and dim U ′ = dim V x. Applying the rank theorem twice, we obtain dim V = dim Vx + dim V x + dim(V mod x). By Lemma 3.1, dim Vx + dim V x ≤ n − 1, and hence dim V ≤ n − 1 +(cid:18)n − 1 2 (cid:19) = (cid:18)n 2(cid:19). 3.2 Spaces with the maximal dimension Assume now that dim V = (cid:18)n 2(cid:19). Then, from the above inequalities, we get, for every x ∈ V r {0}, the two equalities dim Vx + dim V x = n − 1 and dim(V mod x) = (cid:18)n − 1 2 (cid:19). Using Lemma 3.1 once more, we deduce: Claim 3.1. For all x ∈ V r {0}, the space Fx ⊕ Vx is the dual orthogonal of V x. Fix x ∈ V r {0}. Let us assume for a moment that Vx = {0}. Then, U = V. We have just seen that V mod x has dimension (cid:18)n − 1 2 (cid:19), hence by induction there is a complete flag (G0, . . . , Gn−1) of V /Fx that is stable under v for all v ∈ V. For all i ∈ [[1, n]], we denote by Fi the inverse image of Gi−1 under the canonical projection from V to V /Fx, and we deduce that F := ({0}, F1, . . . , Fn) is a complete flag of V that is stable under every element of V. It ensues that V ⊂ NF , and since the dimensions are equal we conclude that V = NF . Hence, in order to conclude it suffices to exhibit a vector x ∈ V r {0} such that Vx = {0}. To obtain such a vector, we start by analyzing the generic nilindex of V, which we denote by p. Claim 3.2. One has p ∈ {n − 1, n} and p ≥ 2. 13 Proof. One has p ≥ 2 since V 6= {0}. Choose x ∈ V r {0}. Then V mod x is a nilpotent linear subspace of End(V /Fx) with dimension (cid:18)n − 1 2 (cid:19). By induction it reads NF ′ for some com- plete flag F ′ of V /Fx, and hence it contains an element v such that vn−2 6= 0. This yields an operator u ∈ V that annihilates x and whose induced endomor- phism of V /Fx equals v. Obviously un−2 6= 0, and hence p > n − 2. It follows in particular that every element of V with nilindex p has exactly one Jordan cell of size p (and one additional Jordan cell of size 1 if p = n − 1). Now, we fix an element u of V with nilindex p, and we choose x ∈ Im up−1 r {0}. Claim 3.3. Let v ∈ V. Then, vk(x) ∈ Fx ⊕ Vx for all k ∈ [[0, n − 1]]. Proof. Let k ∈ [[0, n − 1]]. For all f ∈ V x, the Trace Lemma yields 0 = tr(cid:0)(f ⊗ x) ◦ vk(cid:1) = tr(cid:0)(f ◦ vk) ⊗ x(cid:1) = f (vk(x)). From Claim 3.1, we deduce that vk(x) ∈ Fx ⊕ Vx. Claim 3.4. For all v ∈ V, one has v(x) = 0 or Im vp−1 ⊂ Fx ⊕ Vx. Proof. Let v ∈ V be such that v(x) 6= 0. If the nilindex of v is not p, then Im vp−1 = {0} ⊂ Fx ⊕ Vx. Assume otherwise. Then, Im vp−1 = Im v ∩ Ker v and Im vp−1 has dimension 1: indeed, either v has exactly one Jordan cell, of size n, or it has one Jordan cell of size p and one Jordan cell of size 1. Denoting by k the maximal positive integer such that vk(x) 6= 0, we deduce that vk(x) spans Im vp−1. We conclude by Claim 3.3 that Im vp−1 ⊂ Fx ⊕ Vx. Claim 3.5. One has Vx = {0} or K(V) ⊂ Fx ⊕ Vx. Proof. Assume that Vx 6= {0} and choose v0 ∈ V such that v0(x) 6= 0. We choose a linear form ψ on V such that ψ(v0(x)) 6= 0. Let v ∈ V and z ∈ V . For all (λ, µ) ∈ F2, the vector γ(λ, µ) := (cid:0)λv0 + µv)p−1(z) belongs to K(V). For all (λ, µ) ∈ F2 such that (λv0 + µv)(x) 6= 0, we know from Claim 3.4 that (cid:0)λv0 + µv(cid:1)p−1 (z) ∈ Fx ⊕ Vx. 14 Let ϕ be a linear form that vanishes everywhere on Fx ⊕ Vx. Then, the p- homogeneous polynomial function (λ, µ) 7→ ψ(cid:0)(λv0 + µv)(x)(cid:1) ϕ(cid:0)(λv0 + µv)p−1(z)(cid:1) vanishes everywhere on F2, and since F ≥ p we successively deduce that this function is identically zero, and that one of the polynomial functions (λ, µ) 7→ ψ(cid:0)(λv0 + µv)(x)(cid:1) and (λ, µ) 7→ ϕ(cid:0)(λv0 + µv)p−1(z)(cid:1) is identically zero. The first one is not identically zero since ψ(v0(x)) 6= 0, and hence the second one is identically zero. Taking (λ, µ) = (0, 1) and varying ϕ yields vp−1(z) ∈ Fx ⊕ Vx. Varying v and z finally yields K(V) ⊂ Fx ⊕ Vx. If K(V) ⊂ Fx ⊕ Vx then we deduce from the Reducibility Lemma that Vx = {0}. Otherwise we directly have Vx = {0}. From there, the conclusion follows, as was explained earlier. 4 General results on b-symmetric and b-alternating endomorphisms 4.1 Basic results We start with three basic results that reproduce, respectively, lemmas 1.4, 1.5 and 2.4 from [10]. Lemma 4.1. Let b be a non-degenerate symmetric or alternating bilinear form on V , and let u ∈ Sb ∪ Ab. Then: (a) For every linear subspace W of V that is stable under u, the subspace W ⊥ is stable under u. (b) One has Ker u = (Im u)⊥. Note that, for all u ∈ Sb ∪ Ab, every power of u belongs to Sb ∪ Ab, and in particular Im uk = (Ker uk)⊥ for every integer k ≥ 0. Lemma 4.2. Let b be a non-isotropic symmetric bilinear form on V , and let u ∈ Sb ∪ Ab be nilpotent. Then, u = 0. Lemma 4.3. Let b be a non-degenerate symmetric or alternating bilinear form on V , and let u ∈ Sb ∪ Ab be nilpotent. Then, there exists a maximal partially complete b-singular flag F of V that is stable under u. 15 As a consequence of this last lemma, we find: Lemma 4.4. Let b be a non-degenerate symmetric or alternating bilinear form on V , and let u ∈ Sb ∪ Ab be nilpotent. Denote by ν the Witt index of b. Then, there exists a totally b-singular subspace F of V with dimension ν that is stable under u. Moreover, u maps F ⊥ into F and rk u ≤ 2ν. Proof of Lemma 4.4. By Lemma 4.3, we can find a totally b-singular subspace F of V with dimension ν that is stable under u. Then F ⊂ F ⊥ and F ⊥ is stable under u. The bilinear form b induces a non-degenerate bilinear form b on F ⊥/F that is still b-symmetric or b-alternating. Moreover, since F is maximal among the totally b-singular subspaces of V , the form b is non-isotropic, and in particular it is symmetric (if it is alternating then F ⊥/F = {0} and it is also symmetric). Moreover, u also induces an endomorphism u of F ⊥/F , which is b-symmetric or b-alternating. By Lemma 4.2, we deduce that u = 0, which means that u maps F ⊥ into F . It follows that rk u ≤ codimV (F ⊥) + dim u(F ⊥) ≤ 2 dim F = 2ν. Corollary 4.5. Let b be a non-degenerate symmetric or alternating bilinear form on V (with n := dim V > 0), and let u ∈ Sb ∪ Ab be nilpotent. Denote by ν the Witt index of b. Then, rk u ≤ min(2ν, n − 1) and ind(u) ≤ 2ν + 1. In addition rk u ≤ n − 2 and ind(u) ≤ n − 1 if n = 2ν and u is b-alternating. Proof. We have shown in the previous lemma that rk u ≤ 2ν. Obviously rk u ≤ n − 1 because u is non-injective. Assume finally that u is b-alternating and n = 2ν. Since b is non-degenerate, the rank of u equals the one of the alternating bilinear form (x, y) 7→ b(cid:0)x, u(y)(cid:1), and hence it is even. Since rk u < 2ν we deduce that rk u ≤ 2ν − 2. The statements on the nilindex are then straightforward consequences of the observation that rk u ≥ ind(u) − 1. 4.2 Symmetric and alternating b-tensors Here, b denotes an arbitrary non-degenerate symmetric or alternating bilinear form on a finite-dimensional vector space V . Given vectors x and y of V , we denote by x ⊗b y : z ∈ V 7→ b(y, z) x + b(x, z) y 16 the b-symmetric tensor of x and y, and we note that x ⊗b y belongs to Sb; likewise, x ∧b y : z ∈ V 7→ b(y, z) x − b(x, z) y, which belongs to Ab, is called the b-alternating tensor of x and y. Given a non-zero vector x ∈ V , the mapping y ∈ V 7→ x ⊗b y is linear and injective: the direct image of a linear subspace L of V under it is denoted by x ⊗b L. Given a non-zero vector x ∈ V , the mapping y ∈ V 7→ x ∧b y is linear and its kernel equals Fx: the direct image of a linear subspace L of V under it is denoted by x ∧b L. Now, let x, y be b-orthogonal vectors of V such that b(x, x) = 0. It is easily seen that x⊗b y and x∧b y vanish at x and map {x}⊥ into Fx. Since span(x, y) ⊂ {x}⊥, it easily follows that both maps are nilpotent (with nilindex at most 3). There is a converse statement, and the following result, which was proved in [10] (see proposition 3.1 there), sums everything up: Proposition 4.6. Let x be a non-zero b-isotropic vector of V . Let u ∈ Sb (respectively, u ∈ Ab). The following conditions are equivalent: (i) The endomorphism u is nilpotent, vanishes at x, and maps {x}⊥ into Fx. (ii) There exists y ∈ {x}⊥ such that u = x ⊗b y (respectively, u = x ∧b y). Moreover, the following result is an easy consequence of the Trace Lemma: Proposition 4.7 (Orthogonality Lemma for Tensors). Let x be a non-zero isotropic vector of V . Let V be a nilpotent subspace of Sb (respectively, of Ab). Let v ∈ V, and let y ∈ {x}⊥ be such that x ⊗b y ∈ V (respectively, x ∧b y ∈ V). Then, for every odd positive integer k < F, one has b(cid:0)y, vk(x)(cid:1) = 0. Proof. Let k be an odd positive integer such that F > k. Remembering the notation for tensors from Section 3.1, we note that, with ϕ : z 7→ b(x, z) and ψ : z 7→ b(y, z), we have (x ⊗b y) ◦ vk = (ψ ◦ vk) ⊗ x + (ϕ ◦ vk) ⊗ y and (x ∧b y) ◦ vk = (ψ ◦ vk) ⊗ x − (ϕ ◦ vk) ⊗ y 17 and hence, noting that vk belongs to Sb (respectively, to Ab) because k is odd, tr(cid:0)(x⊗by)◦vk(cid:1) = (ψ◦vk)(x)+(ϕ◦vk)(y) = b(cid:0)y, vk(x)(cid:1)+b(cid:0)x, vk(y)(cid:1) = 2 b(cid:0)y, vk(x)(cid:1) (respectively, tr(cid:0)(x∧by)◦vk(cid:1) = (ψ◦vk)(x)−(ϕ◦vk)(y) = b(cid:0)y, vk(x)(cid:1)−b(cid:0)x, vk(y)(cid:1) = 2 b(cid:0)y, vk(x)(cid:1)). From there, the result follows from the Trace Lemma (Lemma 2.1). 4.3 The maximal nilindex in N Sb,F and N Ab,F Here, given a non-degenerate symmetric or alternating bilinear form b on V , with Witt index ν, and given a maximal partially complete b-singular flag F of V , we investigate the maximal possible nilindex for an element of N Ab,F or of N Sb,F . Those results can be retrieved from the (complicated) classification of pairs consisting of a symmetric bilinear form and an alternating bilinear form under congruence (see e.g. [12]), but we will give elementary proofs for readers that are not well-versed in that theory. Lemma 4.8. Let b be a non-degenerate symmetric bilinear form on V , with Witt index ν. Set n := dim V . Let F be a maximal partially complete b-singular flag of V . If n 6= 2ν, the generic nilindex of N Ab,F is 2ν + 1. If n = 2ν, the generic nilindex of N Ab,F is n − 1. Proof. By Corollary 4.5, it suffices to exhibit an element of N Ab,F with nilindex at least 2ν + 1 if n 6= 2ν, and at least n − 1 otherwise. In any case, we write F = (Fi)0≤i≤ν , and set F := Fν and p := n − 2ν; we take a Witt decomposition V = F ⊕ G ⊕ H in which H is totally b-singular with dimension ν and G = (F ⊕ H)⊥. We choose a basis (f1, . . . , fν) of Fν that is adapted to F (so that Fi = span(f1, . . . , fi) for all i ∈ [[1, ν]]) and then we obtain a basis (h1, . . . , hν ) of H such that b(fi, hj ) = δi,j for all i, j in [[1, ν]]. We choose an orthogonal basis (g1, . . . , gp) of G. The family B := (e1, . . . , eν , g1, . . . , gp, h1, . . . , hν ) is a basis of V in which the matrix of b reads   0ν S :=   [0]ν×p Iν [0]p×ν P [0]p×ν Iν [0]ν×p 0ν where P := (cid:0)b(gi, gj)(cid:1)1≤i,j≤p. We denote by J := (δi+1,j)1≤i,j≤ν ∈ Mν(F) the nilpotent Jordan cell of size ν. 18 Assume first that n = 2ν, and define K = (ki,j) ∈ Aν(F) by ki,j = 0 whenever {i, j} 6= {ν − 1, ν}, and kν,ν−1 = −kν−1,ν = 1. Setting M := (cid:20) J 0ν −J T(cid:21) ∈ Mn(F), K we see that SM is alternating and M is nilpotent. Hence the endomorphism u represented by M in B is b-alternating and nilpotent. Judging from the first ν columns of M , this endomorphism stabilizes the flag F. Finally, we see that uν−1(h1) = (−1)ν−1(−eν + hν), and then uν (h1) = (−1)ν 2eν−1, u2ν−2(h1) = (−1)ν 2e1, whence u2ν−2 6= 0. It follows that ind(u) > 2ν − 2, i.e. ind(u) ≥ n − 1. Assume now that n > 2ν. We define C := (ci,j) ∈ Mp,ν(F) by ci,j = 0 whenever (i, j) 6= (p, ν), and cp,ν = 1. We also set N :=   J −(P C)T [0]p×ν 0ν 0p [0]ν×p 0ν C −J T   ∈ Mn(F), and we see that SN is alternating and N is nilpotent. Hence the endomor- phism u represented by N in B is b-alternating and nilpotent. Once more, we see that u stabilizes F. This time around uν(h1) = (−1)ν−1gp, then uν+1(h1) = (−1)ν b(gp, gp) eν , and then u2ν(h1) = (−1)ν b(gp, gp) e1, to the effect that ind(u) > 2ν. Now, we turn to symmetric endomorphisms for a symplectic form: Lemma 4.9. Let b be a symplectic form on an n-dimensional vector space V . Set n := dim V . Let F be a maximal partially complete b-singular flag of V . Then, the generic nilindex of N Sb,F equals n. Proof. Set ν := n/2. We choose a hyperbolic basis B = (e1, . . . , eν , f1, . . . , fν) of V in which (e1, . . . , eν ) is adapted to F, so that the matrix of b in B reads A := (cid:20) 0ν −Iν 0ν(cid:21) . Iν Here, we consider once more the ν by ν nilpotent Jordan matrix J, and this time around we consider the matrix D ∈ Sν(F) with exactly one non-zero entry, located at the (ν, ν) spot and which equals 1. One defines u as the endomorphism of V whose matrix in the basis B equals M := (cid:20) J 0ν −J T(cid:21) . D 19 Seeing that AM is symmetric, we obtain that u is b-symmetric. Once again, u is nilpotent and stabilizes F, and this time around we see that uν(f1) = (−1)ν−1eν and hence u2ν−1(f1) = (−1)ν−1e1, leading to ind(u) ≥ 2ν. Conversely, every nilpotent element of End(V ) has nilindex at most n = 2ν. Combining the above results with Lemma 4.3, we find: Corollary 4.10. Let b be a non-degenerate symmetric or alternating bilinear form on an n-dimensional vector space, with Witt index ν. (a) If b is alternating then the generic nilindex of Sb equals n. (b) If b is symmetric and n 6= 2ν, then the generic nilindex of Ab equals 2ν + 1. (c) If b is symmetric and n = 2ν, then the generic nilindex of Ab equals n − 1. 4.4 On the indecomposable nilpotent elements in Sb and Ab An endomorphism u of a finite-dimensional vector space U is called a (nilpotent) Jordan cell of size p if dim U = p and if there exists a basis (e1, . . . , ep) of U such that u(ei) = ei−1 for all i ∈ [[2, p]], and u(e1) = 0. Such a basis is then called a Jordan basis of u. Here, we assume that b is non-degenerate. Let u belong to Sb (respectively, to Ab). We say that u is indecomposable when V 6= {0} and there is no ⊥ ⊕ V2 in which V1 decomposition of V into a b-orthogonal direct sum V = V1 and V2 are non-zero linear subspaces that are stable under u. Thanks to the classification theory for pairs of bilinear forms in which one form is symmetric and the other one is alternating (see, e.g. [12]), we have the following information on indecomposable nilpotent endomorphisms: Proposition 4.11. Assume that b is symmetric. Let u ∈ Ab be nilpotent and indecomposable. Then, u is either a Jordan cell of odd size or the direct sum of two Jordan cells of the same even size. Moreover, the Witt index of b equals (cid:4) n 2(cid:5), where n := dim V . We give an elementary proof of this result. In the prospect of that proof, re- member that a square matrix M = (mi,j)1≤i,j≤n is called anti-lower-triangular whenever mi,j = 0 for all (i, j) ∈ [[1, n]]2 with i + j < n + 1; it is called anti- diagonal whenever mi,j = 0 for all (i, j) ∈ [[1, n]]2 with i + j 6= n + 1. The anti-diagonal entries of M are m1,n, m2,n−1, . . . , mn,1. An anti-lower-triangular matrix is invertible if and only if all its anti-diagonal entries are non-zero. 20 Proof. We consider a splitting V = V1 ⊕ · · · ⊕ VN in which every subspace Vi is stable under u and the resulting endomorphism of Vi is a Jordan cell. We denote by q the minimal dimension among the Vi's, and we set U := Ker uq. Assume first that q is odd, and write it q = 2p+1. Assume that b(x, u2p(x)) = 0 for every x ∈ U . Then, the quadratic form x 7→ b(up(x), up(x)) vanishes every- where on U and hence up(U ) -- which equals Ker up+1 because of the definition of q -- is totally b-singular. Yet its orthogonal complement under b is Im up+1. In particular Ker up+1 ⊂ Im up+1, which contradicts the existence of a Jordan cell of size q for u. This yields some x ∈ U such that b(x, u2p(x)) 6= 0. Note that u2p+1(x) = 0. Let (k, l) ∈ N2. Then, b(uk(x), ul(x)) = (−1)kb(x, uk+l(x)) = 0 if k + l > 2p, whereas b(uk(x), ul(x)) 6= 0 if k + l = 2p. It follows that the matrix (cid:0)b(u2p+1−k(x), u2p+1−l(x))(cid:1)1≤k,l≤2p+1 is anti-lower-triangular with all its anti- diagonal entries non-zero. Therefore, the subspace W := span(uk(x))0≤k≤2p has dimension 2p + 1 and is b-regular, and the endomorphism induced by u on W is a Jordan cell. Since W is b-regular, V = W ⊕ W ⊥. As u is indecomposable and W ⊥ is stable under u, we conclude that W = V and that u is a Jordan cell of odd size. Moreover, we see that span(uk(x))p+1≤k≤2p is totally b-singular with dimension p = n−1 2 = (cid:4) n 2(cid:5), and hence the Witt index of b equals (cid:4) n 2(cid:5). Now, we consider the case when q is even, and we write it q = 2p + 2 for some integer p ≥ 0. This time around, we set c : (x, y) ∈ V 2 7→ b(x, u(y)), which If c(x, u2p(y)) = 0 for all (x, y) ∈ U 2, then up(U ) = Ker up+2 is alternating. would be totally singular for c; in other words Ker up+2 would be b-orthogonal to u(up(U )) = Ker up+1. Yet, the b-orthogonal complement of Ker up+1 is Im up+1, and hence we would have Ker up+2 ⊂ Im up+1, contradicting the existence of a Jordan cell of size 2p + 2 for u. This yields two vectors x and y of Ker uq such that b(x, uq−1(y)) 6= 0. Now, we set Wx := span(uk(x))0≤k≤q−1 and Wy := span(uk(y))0≤k≤q−1. We claim that Wx and Wy have dimension q and are linearly disjoint, and that Wx ⊕ Wy is b-regular. To back this up, we consider the family (e1, . . . , e2q) := (cid:0)uq−1(y), uq−1(x), . . . , u(y), u(x), y, x(cid:1). For all non-negative integers k, l such that k+l > q −1, we have b(uk(x), ul(y)) = (−1)kb(x, uk+l(y)) = 0, and likewise b(uk(x), ul(x)) = 0 = b(uk(y), ul(y)). Since c is alternating and q − 1 is odd, b(x, uq−1(x)) = 0 = b(y, uq−1(y)), whence 21 b(uk(x), ul(x)) = 0 = b(uk(y), ul(y)) for all (k, l) ∈ [[0, q − 1]]2 such that k + l = q − 1. Finally, b(uk(x), ul(y)) = (−1)kb(x, uq−1(y)) 6= 0 for all non-negative integers k, l such that k + l = q − 1. It follows that the matrix (cid:0)b(ei, ej)(cid:1)1≤i,j≤2q is anti-lower-triangular with all its anti-diagonal entries non-zero. This shows that e1, . . . , e2q are linearly independent and that their span is b-regular, as claimed. Since u is indecomposable, we deduce that V = Wx ⊕ Wy, and hence u is the direct sum of two Jordan cells of size q. Here, we directly have Ker up+1 = Im up+1, and hence Ker up+1 is totally b-singular with dimension n 2 = q. We conclude that the Witt index of b equals n 2 · Here is the corresponding result for nilpotent symmetric endomorphisms for a symplectic form. We will not need it, so we simply state it and leave the proof (which an easy adaptation of the previous one) to the reader: Proposition 4.12. Assume that b is alternating. Let u ∈ Sb be nilpotent and indecomposable. Then, u is either a Jordan cell of even size or the direct sum of two Jordan cells of the same odd size. We finish with a very simple result: Lemma 4.13. Let u ∈ Sb ∪ Ab be a Jordan cell, and (e1, . . . , en) be a Jordan basis of it, with n odd. Write n = 2p + 1. Then, the radical of the induced bilinear form on span(e1, . . . , ep+1) is span(e1, . . . , ep). Proof. We have Ker up = span(e1, . . . , ep) = Im up+1 and Im up = span(e1, . . . , ep+1), whence the said radical equals Im up ∩ (Im up)⊥ = Im up ∩ Ker up = span(e1, . . . , ep). 5 General results on the structured Gerstenhaber problem In this section, we recall some elements of the study of the structured Gersten- haber problem from [10], and we consider the problem of extending scalars. There are two items we have to review. First (Section 5.1), we need to go back to the inductive proof of Theorem 1.2, and for two reasons: • It will explain how Theorems 1.5 and 1.6 can be proved by induction. 22 • An important partial consequence on the structure of spaces of maximal dimension, called the Strong Orthogonality Lemma (Lemma 5.1), can be drawn from this proof. In Section 5.2, we recall a stability property for spaces with maximal dimension that was mentioned in the introduction, and we draw an important consequence of it. Finally, in Section 5.3, we explain how, in Theorems 1.5 and 1.6, one can reduce the situation to the one where the underlying field is infinite. 5.1 A review of the proof of Theorem 1.2, and some conse- quences on spaces with the maximal dimension Here, we recall the inductive proof of Theorem 1.2 (see section 3 of [10]). Let V be a nilpotent subspace of Sb (respectively, of Ab). If the Witt index ν of b is zero, then Lemma 4.2 shows that V = {0}. Assume now that ν > 0 and let x ∈ V be an arbitrary non-zero isotropic vector. We consider the kernel UV ,x := {u ∈ V : u(x) = 0} of the surjective linear mapping u ∈ V 7→ u(x) ∈ Vx. Any u ∈ UV ,x stabilizes {x}⊥ (because it stabilizes Fx) and hence induces a nilpotent endomorphism u of the quotient space {x}⊥/Fx. Note that b induces a non-degenerate symmetric or alternating bilinear form b on {x}⊥/Fx with Witt index ν − 1. The set V mod x := {u u ∈ UV ,x} is then a nilpotent linear subspace of Sb (respectively, of Ab). Finally, the kernel of the linear mapping ϕ : u ∈ UV ,x 7→ u ∈ V mod x consists of the operators u ∈ V that vanish at x and map {x}⊥ into Fx. Set (respectively, LV ,x := (cid:8)y ∈ V : x ⊗b y ∈ V(cid:9) LV ,x := (cid:8)y ∈ V : x ∧b y ∈ V(cid:9)). 23 By Proposition 4.6, LV ,x is a linear subspace of {x}⊥ (respectively, a linear subspace of {x}⊥ that contains x), the kernel of ϕ reads x ⊗b LV ,x (respec- tively, x ∧b LV ,x), and the dimension of that kernel equals dim LV ,x (respectively, dim LV ,x − 1). Hence, applying the rank theorem twice yields dim V = dim(Vx) + dim LV ,x + dim(V mod x) (respectively, dim V = dim(Vx) + dim LV ,x − 1 + dim(V mod x)). In any case, the Orthogonality Lemma for Tensors (Proposition 4.7) yields that Fx ⊕ Vx is b-orthogonal to LV ,x. Hence, and it follows that (respectively, dim(Vx) + dim(LV ,x) ≤ n − 1, dim V ≤ dim(V mod x) + n − 1 dim V ≤ dim(V mod x) + n − 2). If we do not take Theorem 1.2 for granted, this allows one to prove it by induction on the dimension of V : indeed, by induction dim(V mod x) ≤ (ν − 1)(cid:0)n − 2 − (ν − 1)(cid:1) (respectively, dim(V mod x) ≤ (ν − 1)(cid:0)n − 2 − (ν − 1) − 1(cid:1)), and it follows that dim V ≤ ν(n − ν) (respectively, dim V ≤ ν(n − ν − 1)). Now, assume that dim V = ν(n − ν) (respectively, dim V = ν(n − ν − 1)). Then, the above series of inequalities yields the two equalities dim(Fx ⊕ Vx) + dim LV ,x = n and (respectively, dim(V mod x) = (ν − 1)(cid:0)n − 2 − (ν − 1)(cid:1) dim(V mod x) = (ν − 1)(cid:0)n − 2 − (ν − 1) − 1(cid:1)). Since Fx ⊕ Vx is b-orthogonal to LV ,x, those two equalities yield the following important results: 24 Lemma 5.1 (Strong Orthogonality Lemma). Let x ∈ V be a non-zero b-isotropic vector, and let V be a nilpotent subspace of Sb (respectively, of Ab) with dimen- sion ν(n − ν) (respectively, ν(n − ν − 1)). Then, Fx ⊕ Vx is the b-orthogonal complement of LV ,x. Lemma 5.2 (Preservation of Maximality Lemma). Let x ∈ V be a non-zero b-isotropic vector, and let V be a nilpotent subspace of Sb (respectively, of Ab) with dimension ν(n − ν) (respectively, ν(n − ν − 1)). Denote by b the bilinear form on {x}⊥/Fx induced by b. Then, V mod x is a nilpotent subspace of Sb (respectively, of Ab) with the maximal dimension among such subspaces. Related to the above method is the following basic result: Lemma 5.3 (Lifting Lemma). Let x ∈ V be a non-zero b-isotropic vector, and let V be a nilpotent subspace of Sb (respectively, of Ab) with dimension ν(n − ν) (respectively, ν(n − ν − 1)). Denote by b the bilinear form on {x}⊥/Fx induced by b. Assume that Vx = {0} and that V mod x equals N Sb,G (respectively, N Ab,G) for some maximal partially complete b-singular flag G of {x}⊥/Fx. Then, V = N Sb,F (respectively, V = N Ab,F ) for some maximal partially complete b-singular flag F of V . Proof. Write G = (Gi)0≤i≤ν−1, set F0 := {0} and, for k ∈ [[1, ν]], denote by Fk the inverse image of Gk−1 under the canonical projection of {x}⊥ onto {x}⊥/Fx. Then, one sees that (F0, . . . , Fν ) is a maximal partially complete b-singular flag of V , and every element of V stabilizes it (here UV ,x = V since we have assumed that Vx = {0}). Hence, V ⊂ N Sb,F (respectively V ⊂ N Ab,F ) and since the dimensions of those spaces are equal we conclude that V = N Sb,F (respectively, V = N Ab,F ). 5.2 Stability properties of spaces with the maximal dimension In the course of [10], we have obtained an additional structural result on nilpotent subspaces of Sb or Ab with the maximal possible dimension: Lemma 5.4. Let b be a non-degenerate symmetric or alternating bilinear form on V , with Witt index ν. Let V be a nilpotent linear subspace of Sb (respectively, of Ab) with dimension ν(n − ν) (respectively, ν(n − ν − 1)). Assume furthermore that F > 3. Then, u3 ∈ V for all u ∈ V. 25 See lemmas 5.2 and 5.3 of [10]. We use a simple linearity trick to derive the following corollary, which will turn out to be very useful when b is symmetric: Corollary 5.5. Let b be a non-degenerate symmetric or alternating bilinear form on V , with Witt index ν. Let V be a nilpotent linear subspace of Sb (respectively, of Ab) with dimension ν(n − ν) (respectively, ν(n − ν − 1)). Assume furthermore that F > 3. Then, u2v + uvu + vu2 ∈ V for all u and v in V. Proof. Let u and v belong to V. By Lemma 5.4, the linear subspace V contains (u + v)3 − (u − v)3 − 2v3 = 2(u2v + uvu + vu2) and we conclude by using the fact that the characteristic of F is not 2. 5.3 Extending scalars Let V be a nilpotent subspace of Sb (respectively, of Ab). Let L be a field extension of F. We assume that the extension F − L preserves Witt indices, i.e. it induces an injection from the Witt group of F to the one of L. In particular, this is known to hold if L is purely transcendental over F, e.g. when L = F(t) (see [9] Chapter XV, Section 3). This case is interesting because L then turns out to be infinite. Next, we set VL := L ⊗F V . Given u ∈ End(V ), we define uL as the endomorphism of the L-vector space VL such that uL(1 ⊗ x) = u(x) for all x ∈ V . The mapping u ∈ End(V ) 7→ uL ∈ EndL(VL) is a homomorphism of F-algebras. Given a linear subspace V of End(V ), it is folklore that the span VL of {uL u ∈ V} in the L-vector space EndL(VL) has its dimension over L equal to the one of V over F; better still, for every basis (u1, . . . , uk) of V, the family (cid:0)(u1)L, . . . , (uk)L(cid:1) is a basis of VL. Next, let V be a nilpotent subspace of Sb (respectively, of Ab), and denote by d the generic nilindex of Sb (respectively, of Ab). Assume that F ≥ d. We claim that VL is nilpotent. Indeed, let us take a basis (u1, . . . , up) of V. Then, ∀(λ1, . . . , λp) ∈ Fp, (cid:18) p Xk=1 λk uk(cid:19)d = 0, which, by extending to VL, yields ∀(λ1, . . . , λp) ∈ Fp, (cid:18) p Xk=1 λk (uk)L(cid:19)d = 0. 26 On the left-hand side is a (vector-valued) polynomial function, homogenous with degree d; since we have assumed that F ≥ d, this function vanishes everywhere on Lp, whence every element of VL is nilpotent. Next, b induces a bilinear form bL on the L-vector space VL such that ∀(x, y) ∈ V 2, bL(1 ⊗ x, 1 ⊗ y) = b(x, y). This form is symmetric if b is symmetric, alternating if b is alternating. Finally, it is clear that uL is bL-symmetric (respectively, bL-alternating) for all u ∈ Sb (respectively, u ∈ Ab). Hence, VL is a linear subspace of SbL (respectively, of AbL). Finally, assume that there is a non-zero isotropic vector y of VL at which every element of VL vanishes. There is a unique minimal subspace E of V such that y can be written as a sum of tensors of the form λ ⊗ x with λ ∈ L and x ∈ E. Then, every element of V must vanish at every element of E. Moreover, we claim that E contains a non-zero b-isotropic vector. Indeed, let (e1, . . . , eq) λk ⊗ ek for some (λk)1≤k≤q ∈ Lq. Since be a basis of E. We write y = bL(y, y) = 0, the quadratic form q Pk=1 Q : (µ1, . . . , µq) ∈ Fq 7→ b(cid:18) q Xk=1 µkek, q Xk=1 µkek(cid:19) = X1≤k,l≤q µk µl b(ek, el) becomes isotropic over L. Since we have assumed that the extension L of F preserves Witt indices, we conclude that Q is isotropic, which yields a non-zero b-isotropic vector x of E. We conclude that every element of V vanishes at x. Combining the above remarks with Corollary 4.10, we obtain the following results: Proposition 5.6. Let L be a field extension of F that preserves Witt indices. Let b be a non-degenerate symmetric or alternating bilinear form on an F-vector space V with finite dimension n. Denote by ν the Witt index of b. Let V be a nilpotent linear subspace of Sb or of Ab. (a) If b is alternating, F ≥ n and V ⊂ Sb, then VL is a nilpotent subspace of SbL. (b) If b is symmetric, n 6= 2ν, F ≥ 2ν + 1 and V ⊂ Ab, then VL is a nilpotent subspace of AbL. 27 (c) If b is symmetric, n = 2ν, F ≥ n − 1 and V ⊂ Ab, then VL is a nilpotent subspace of AbL. (d) If there exists a non-zero bL-isotropic vector of VL at which all the elements of VL vanish, then there exists a non-zero b-isotropic vector of V at which all the elements of V vanish. We are now ready to conclude. Proposition 5.7. If Theorem 1.5 holds over any infinite field (with character- istic not 2), then it holds over any field (with characteristic not 2) that satisfes its cardinality requirements. Proof. Assume that Theorem 1.5 holds over any infinite field with characteristic not 2. Then, we prove Theorem 1.5 by induction on the dimension of the vector space V . Let F be a field with characteristic not 2, V be an n-dimensional vector space over F (with n > 0), b be a symmetric bilinear form on V , and V be a nilpotent subspace of Ab with dimension ν(n − ν − 1). Assume that F ≥ 2ν + 1 if n > 2ν, and that F ≥ n − 1 otherwise. Let us choose an infinite extension L of F that preserves Witt indices (e.g. L = F(t)). By Proposition 5.6, VL is a nilpotent subspace of AbL with dimension ν(n − ν − 1). Since L is infinite we deduce that VL = N AbL,F for some maximal partially complete bL-singular flag F of VL. In particular, there is a non-zero bL-isotropic vector of VL at which all the elements of VL vanish. By point (d) of Proposition 5.6, there exists a non-zero isotropic vector x of V at which all the elements of V vanish. Denote by b the symmetric bilinear form induced by b on {x}⊥/Fx. By Lemma 5.2, V mod x is a nilpotent subspace of Ab with dimension (ν − 1)(cid:0)(n − 2)−(ν−1)−1(cid:1). Noting that F ≥ 2ν+1 > 2ν−1 if 2(ν−1) 6= n−2 (i.e. if 2ν 6= n), and F ≥ n − 1 ≥ n − 3 otherwise, we find by induction that V mod x = N Ab,G for some maximal partially complete b-singular flag G of {x}⊥/Fx. Then, the Lifting Lemma (Lemma 5.3) yields that V = N Ab,F for some maximal partially complete b-singular flag F of V . With the same line of reasoning, we also get: Proposition 5.8. If Theorem 1.6 holds over any infinite field (with character- istic not 2), then it holds over any field (with characteristic not 2) that satisfies its cardinality requirements. 28 6 Spaces of symmetric nilpotent endomorphisms for a symplectic form We are now ready for the proofs of Theorems 1.5 and 1.6. In the present section, we deal with the latter, which turns out to be the easier. By Proposition 5.8, Theorem 1.6 requires a proof only for infinite fields. So, we fix an infinite field F (with characteristic not 2) and we prove Theorem 1.6 by induction on the dimension of the vector space V under consideration. The result is obvious for a vector space with dimension 0. Let V be vector space over F with finite dimension n > 0, and let b be a symplectic form on V . Let V be a nilpotent subspace of Sb with dimension ν(n − ν), where ν := n 2 · By induction and the Lifting Lemma (Lemma 5.3), it suffices to show that there exists a non-zero vector x ∈ V such that Vx = {0}. We split the proof of the existence of such a vector into two subcases, whether the generic nilindex of V is at most 2 or not. 6.1 A special case Here, we assume that all the elements of V have square zero. Hence uv + vu = (u + v)2 − u2 − v2 = 0 for all (u, v) ∈ V 2. Classically, any subset A of skew- commuting nilpotent elements of End(V ) vanishes at some non-zero vector. To see this, one can use Jacobson's triangularization theorem [4, 8] for example, but one can also give a very elementary proof: if A ⊂ {0} the result is obvious; otherwise choose u ∈ A r{0} and note that Ker u is stable under all the elements of A. By restricting the elements of A to Ker u, we recover a skew-commuting set of square-zero endomorphisms of Ker u, and by induction there is a non-zero vector x of Ker u at which all the elements of A vanish. 6.2 When the generic nilindex is greater than 2 (part 1) In the rest of the proof, we assume that the generic nilindex of V is greater than 2. Claim 6.1. The generic nilindex of V is greater than or equal to n − 2. Proof. Choose an arbitrary vector x ∈ V r{0}. Denote by b the symplectic form induced by b on {x}⊥/Fx. By Lemma 5.2, the subspace V mod x has dimension (ν − 1)(cid:0)(n − 2) − (ν − 1)(cid:1) and it is a nilpotent subspace of Sb. Hence, by induction V mod x = N Sb,G for some maximal partially complete b-singular flag 29 G of {x}⊥/Fx. Hence, by Lemma 4.9 there is an element v ∈ V mod x with nilindex n − 2. We conclude that ind(u) ≥ n − 2 for some u ∈ V. In the rest of the proof, we denote by p the generic nilindex of V. Remember that we have assumed p ≥ 3. Claim 6.2. The nilpotent subspace V is pure. Proof. Let u ∈ V have nilindex p. Since p ≥ n − 2 and p ≥ 3, we have 2p > n, and hence u has exactly one Jordan cell of size p. Claim 6.3. For all x ∈ V • r {0}, the space V contains x ⊗b x. Proof. Let x ∈ V • r{0} and choose u ∈ V such that x ∈ Im up−1. By Proposition 2.2, Vx ⊂ Im u, whence Fx ⊕ Vx ⊂ Im u. Taking the orthogonal complement and applying the Strong Orthogonality Lemma (Lemma 5.1), we deduce that Ker u = (Im u)⊥ ⊂ LV ,x, and in particular x ⊗b x ∈ V since x ∈ Ker u. Now, we use a reductio ad absurdum, by assuming that Vx 6= {0} for all x ∈ V •. The Reduction Lemma yields (H) : ∀x ∈ V • r {0}, K(V) 6⊂ Fx ⊕ Vx. Claim 6.4. The space K(V) is totally b-singular. Proof. Assume otherwise. Then, K(V) ∩ K(V)⊥ is a proper linear subspace of K(V). Hence, the Linear Density Lemma (Lemma 2.4) yields a basis (e1, . . . , eq) of K(V) made of vectors of V • r K(V)⊥. Applying the Linear Density Lemma once more, we find an additional vector x ∈ V • r {0} such that b(x, ei) 6= 0 for all i ∈ [[1, q]]. Now, let i ∈ [[1, q]]. Then, (ei ⊗b ei)(x) = 2 b(ei, x) ei is non-zero and collinear with ei. Thus, ei ∈ Vx by Claim 6.3. We deduce that K(V) ⊂ Vx, which contradicts property (H). 6.3 When the generic nilindex is greater than 2 (part 2) Now, we are ready for the final contradiction. Choose a basis (x1, . . . , xq) of K(V) made of vectors of V •. Let i ∈ [[1, q]]. By the Strong Orthogonality Lemma, we use the non-inclusion K(V) 6⊂ Fxi ⊕ Vxi to obtain a vector zi ∈ V such that xi ⊗b zi ∈ V and zi 6∈ K(V)⊥. Applying the Linear Density Lemma once more, we find a vector y ∈ 30 V • r {0} such that b(zi, y) 6= 0 for all i ∈ [[1, q]]. Hence, for all i ∈ [[1, q]], we have (xi ⊗b zi)(y) = b(zi, y) xi (because b(xi, y) = 0 by the total singularity of K(V)) and we deduce that xi ∈ Vy. Hence y satisfies condition (C) of the Reducibility Lemma, contradicting property (H). This final contradiction completes the proof. 7 Spaces of nilpotent alternating endomorphisms for a symmetric bilinear form In this ultimate section, we prove Theorem 1.5. By Proposition 5.7, we know that it suffices to do so when the underlying field is infinite. We fix such a field F (with characteristic not 2), and we prove the result by induction on the dimension of the vector space V under consideration. Let V be a finite-dimensional vector space with dimension n > 0, and let b be a non-degenerate symmetric bilinear form on V , whose Witt index we denote by ν. Let V be a nilpotent subspace of Ab with dimension ν(n − ν − 1). If ν = 0 then V = {0}, which is the claimed result. Assume now that ν > 0. Combining the Lifting Lemma and the induction hypothesis, we see that it suffices to prove the existence of a non-zero b-isotropic vector x of V such that Vx = {0}. The overall strategy of the proof is similar to the one of the previous section, but the details are much more difficult. Assume first that the generic nilindex of V equals 2. Then, as in Section 6.1, we find that the elements of V skew-commute pairwise. Moreover, we can take u0 ∈ V with nilindex 2, so that Im u0 is a non-trivial subspace of V , and it is totally b-singular because Im u0 ⊂ Ker u0 = (Im u0)⊥. The elements of V induce nilpotent endomorphisms of Im u0 that skew-commute pairwise, and with the same line of reasoning as in Section 6.1 we deduce that some non-zero vector x of Im u0 satisfies Vx = {0}. The vector x is b-isotropic because it belongs to Im u0, and hence the proof is complete in that case. In the remainder of the proof, we assume that the generic nilindex of V is In the next section, we obtain additional information on the greater than 2. generic nilindex of V. We finish the present one with a basic, yet very useful result: Claim 7.1. Every vector of V • is isotropic. Proof. Denote by p the generic nilindex of V. Let x ∈ V •. Then, for some u ∈ V, we have x ∈ Im up−1, and hence x belongs both to Ker u and to Im u = (Ker u)⊥. This yields the claimed result. 31 7.1 The generic nilindex Claim 7.2. The generic nilindex of V is at least 2ν − 1 if n 6= 2ν, and at least n − 3 otherwise. Proof. Choose an arbitrary isotropic vector x ∈ V r {0}. Denote by b the non- degenerate symmetric bilinear form induced by b on {x}⊥/Fx. Its Witt index equals ν − 1. By Lemma 5.2, the subspace V mod x has dimension (ν − 1)(cid:0)(n − 2) − (ν − 1) − 1(cid:1) and it is a nilpotent subspace of Ab. Hence, by induction V mod x = N Ab,G for some maximal partially complete b-singular flag G of {x}⊥/Fx. By Lemma 4.8 there is an element v ∈ V mod x with nilindex n − 3 if n − 2 = 2(ν − 1), and with nilindex 2ν − 1 otherwise. This proves the claimed statement. From now on, the generic nilindex of V is denoted by p. Claim 7.3. Assume that V is pure, and let u ∈ V be with nilindex p. Then, p is odd and: • Either u has one Jordan cell of size p, one Jordan cell of size 3, and all its other Jordan cells have size 1. • Or u has one Jordan cell of size p, and all its other Jordan cells have size 1. Proof. By Lemma 4.11, the number of Jordan cells of given even size for u is even, and since V is pure we get that p is odd. Set ε := 1 if n = 2ν, otherwise set ε := 0. The combination of Corollary 4.5 and of Claim 7.2 yields 2ν − 1 − 2ε ≤ p ≤ 2ν + 1 − 2ε. Moreover, rk u ≤ 2ν − 2ε. This leaves us with three possibilities: (i) u has exactly one Jordan cell of size p, and all the other Jordan cells have size 1; (ii) u has exactly one Jordan cell of size p, exactly one Jordan cell of size 3, and all its other Jordan cells have size 1; (iii) u has exactly one Jordan cell of size p, exactly two Jordan cells of size 2, and all its other Jordan cells have size 1. 32 We need to discard case (iii). Assume on the contrary that it holds. Then, p + 4 ≤ n, and hence 2ν ≤ p + 3 < n, leading to ε = 0. Besides, Proposition 4.11 ⊥ ⊥ would yield a decomposition V = Vp which all the subspaces Vp, V2 and V1 are stable under u, the endomorphism of Vp induced by u is a Jordan cell of size p, and the endomorphism of V2 induced by u is the direct sum of two Jordan cells of size 2. But then Proposition 4.11 would also yield that the Witt index of b is at least p−1 2 + 2 ≥ ν − 1 + 2 = ν + 1, which is absurd. LV1 into a b-orthogonal direct sum in LV2 Hence, only cases (i) and (ii) are possible. Claim 7.4. Assume that V is not pure. Then, p = 3 and there is an element of V that has exactly two Jordan cells of size 3, and no Jordan cell of size 2. Proof. Choose u ∈ V with nilindex p and at least 2 Jordan cells of size p. Set ε := 1 if n = 2ν, otherwise set ε := 0. Again, we have 2ν −1−2ε ≤ p ≤ 2ν +1−2ε and rk u ≤ 2ν − 2ε. Let u ∈ V have nilindex p and several Jordan cells of size p: denote by k the number of those cells, and by l the number of cells of size less than p and more than 1. Then, rk u ≥ (p − 1) + (k − 1)(p − 1) + l, whence (k − 1)(p − 1) + l ≤ 2. As p ≥ 3 and k − 1 > 0, this yields p = 3, k = 2 and l = 0, which proves the claimed statement. 7.2 A variation of Proposition 2.2 Claim 7.5. Let u ∈ V have nilindex p, and let x ∈ Im up−1 r {0} and v ∈ V. Then, v(x) ∈ u({x}⊥). Proof. Choose y ∈ V such that up−1(y) = x. For λ ∈ F, set γ(λ) := (u + λv)p−1(y) = p−1 Xk=0 λkxk, where x0, . . . , xp−1 all belong to V and x0 = x. We have seen in the proof of Proposition 2.2 that v(x) = u(−x1). For all λ ∈ F, the vector γ(λ) is b-isotropic because it belongs both to the kernel and to the range of the b-alternating endomorphism u + λv; hence b(γ(λ), γ(λ)) = 0. Differentiating this polynomial function at zero yields 2 b(x0, x1) = 0, and it follows that −x1 ∈ {x}⊥, which yields the claimed statement. 33 7.3 Investigating the structure of Fx ⊕ V x Claim 7.6. Let u ∈ V and x ∈ Im up−1. Then, Vx is stable under u2. Proof. Let v ∈ V. By Corollary 5.5, the endomorphism w := u2v + uvu + vu2 belongs to V, and we see that u2(v(x)) = w(x) because x ∈ Ker u. Claim 7.7. Let x ∈ V be isotropic and non-zero. Assume that (Fx ⊕ Vx)⊥ contains a non-isotropic vector. Then, for all v ∈ V, the subspace V contains x ∧b v(x). Proof. Let v ∈ V. By assumption and by the Strong Orthogonality Lemma, there is a non-isotropic vector y of {x}⊥ such that x∧b y ∈ V. Set u := x∧b y and α := b(y, y), the latter of which is non-zero. One computes that u2 = − α 2 x ⊗b x. Another computation shows that v ◦ (x ⊗b x) + (x ⊗b x) ◦ v = −2 x ∧b v(x). Finally, uvu = b(y, v(x)) u. Applying Corollary 5.5, we deduce that the operator α x ∧b v(x) + b(y, v(x)) u belongs to the linear subspace V. Since α 6= 0 and u ∈ V, the desired conclusion follows. Claim 7.8. Let x ∈ V be isotropic and non-zero. Assume that (Fx ⊕ Vx)⊥ contains a non-isotropic vector. Then, vk(x) is orthogonal to x for every v ∈ V and every integer k ≥ 0. Moreover, Fx ⊕ Vx is totally singular. Proof. Let v ∈ V. The first result is already known for odd k because v is b-alternating; on the other hand it is known for k = 0. Now, let k be a non- zero even integer. We write b(x, vk(x)) = −b(v(x), vk−1(x)). By Claim 7.7, x ∧b v(x) belongs to V. By the Orthogonality Lemma for Tensors, we deduce that b(v(x), vk−1(x)) = 0 and the first claimed statement ensues. It follows in particular that b(x, v(x)) = 0, b(v(x), v(x)) = −b(x, v2(x)) = 0 and b(x, x) = 0 for all v ∈ V, which yields that Fx ⊕ Vx is totally singular. From there, we split the study into two subcases, whether V is pure or not. 34 7.4 Case 1: V is pure Here, we assume that V is pure. Claim 7.9. Let u ∈ V have nilindex p, and let x ∈ Im up−1 r {0}. Then, dim(Fx ⊕ Vx) ≤ n 2 , and the inequality is strict if Fx ⊕ Vx is totally singular. Proof. Case 1: u has one Jordan cell of size p and all its other Jordan cells have size 1. Hence Fx = Im up−1, so that {x}⊥ = Ker up−1. Then, there is a basis B = (e1, . . . , ep, f1, . . . , fn−p) of V in which u(ei) = ei−1 for all i ∈ [[2, p]], and u(e1) = u(f1) = · · · = u(fn−p) = 0. We note that Im up−1 = Fx and {x}⊥ = Ker up−1 = span(e1, . . . , ep−1, f1, . . . , fn−p). By Claim 7.5, we have Vx ⊂ u({x}⊥) = span(e1, . . . , ep−2). For i ∈ [[0, p]], set Fi := span(e1, . . . , ei). We claim that Vx ∩ F2i+1 = Vx ∩ F2i for all i ∈ [[0, (p − 1)/2]]. Indeed, assuming otherwise, there would be an index i ∈ [[0, (p − 1)/2]] and a vector y of Vx∩ F2i+1 with a non-zero coefficient α on e2i+1 in the basis B: applying (u2)i would yield that αe1 ∈ Vx (see Claim 7.6), contradicting the fact that Vx is linearly disjoint from Fx. Then, as dim(Vx∩F2i) ≤ dim(Vx∩F2i−1)+1 for all i ∈ [[1, (p − 1)/2]], we obtain by induction that dim(Vx ∩ F2i+1) ≤ i for all i ∈ [[0, (p − 1)/2]]. In particular, and hence dim Vx = dim(Vx ∩ Fp−2) ≤ p − 3 2 , dim(Fx ⊕ Vx) ≤ p − 1 2 < n 2 · Case 2: p > 3 and u has a Jordan cell of size 3. Then, by Claim 7.3 together with Proposition 4.11, we have a decomposition ⊥ ⊕ V1 ⊥ ⊕ V2 into a b-orthogonal direct sum, in which all subspaces V0, V1, V2 V = V0 are stable under u, and the resulting endomorphisms are, respectively, a Jordan cell of size p, a Jordan cell of size 3, and the zero endomorphism of V2. We take a Jordan basis (e1, . . . , ep) for uV0, a Jordan basis (f1, f2, f3) for uV1 and a basis (g1, . . . , gn−p−3) of V2. Hence, B := (e1, . . . , ep, f1, f2, f3, g1, . . . , gn−p−3) is a basis of V . Again, Im up−1 = Fx = Fe1, whence {x}⊥ = Ker up−1 = span(e1, . . . , ep−1, f1, f2, f3, g1, . . . , gn−p−3). 35 Set Fi := span(e1, . . . , ei) for all i ∈ [[0, p]]. We deduce from Claim 7.5 that Fx ⊕ Vx ⊂ Fp−2 ⊕ span(f1, f2). Combining this with Claim 7.6, we find u2(Fx ⊕ Vx) ⊂ Vx ∩ Fp−4. Besides, with the same line of reasoning as in the first case above, we find p − 5 · dim(Vx ∩ Fp−4) ≤ 2 Noting that Ker u2 ∩ (Fx ⊕ Vx) ⊂ span(e1, e2, f1, f2), the rank theorem yields dim(Fx ⊕ Vx) ≤ dim(Vx ∩ Fp−4) + 4 ≤ p + 3 2 ≤ n 2 , and if equality holds then Fx ⊕ Vx includes span(e1, e2, f1, f2), and in particular it contains the vector f2, which is non-isotropic by Lemma 4.13. Claim 7.10. Let u ∈ V have nilindex p, and let x ∈ Im up−1 r {0}. Then, (Fx ⊕ Vx)⊥ contains a non-isotropic vector. Proof. Assume on the contrary that G := (Fx ⊕ Vx)⊥ is totally singular. The space G has dimension at least n 2 , due to Claim 7.9. Therefore, G⊥ = G because G ⊂ G⊥. Then, Fx ⊕ Vx = G⊥ = G would be totally singular. Using Claim 7.9 once more, it would follow that dim G < n 2 , contradicting what we have just found. Claim 7.11. The subspace K(V) is totally singular. Proof. Let x and y belong to V • r {0}. Assume that b(x, y) 6= 0. Since V is pure, we can choose u ∈ V and v ∈ V such that Im up−1 = Fx and Im vp−1 = Fy. Thus, x 6∈ (Im vp−1)⊥ = Ker vp−1, and hence vp−1(x) 6= 0. It follows that y is collinear with vp−1(x). Combining Claims 7.8 and 7.10, we find that x is orthogonal to vp−1(x), which contradicts the fact that x is non-orthogonal to y. We conclude that x is orthogonal to y. Varying x and y yields the claimed result. The remainder of the proof is similar to the one in Section 6.3: the only formal difference is that symmetric tensors must be replaced with alternating tensors. One obtains that Vx = {0} for some x ∈ V • r {0}, which completes the proof. 36 7.5 Case 2: V is not pure Here, we assume that V is not pure. By Claim 7.4, the generic nilindex in V is 3 and there is an element u of V that has exactly two Jordan cells of size 3, with all remaining Jordan cells of size 1. By Proposition 4.11, V splits into ⊥ ⊕ V3 in which u stabilizes V1 and V2 a b-orthogonal direct sum V = V1 and induces Jordan cells of size 3 on those spaces, and V3 ⊂ Ker u. We choose respective Jordan bases (e1, e2, e3) and (f1, f2, f3) of V1 and V2 for u. ⊥ ⊕ V2 Claim 7.12. Every element of V stabilizes Im u2. Proof. Note that Im u2 = span(e1, f1). We shall prove that Ve1 ⊂ span(e1, f1). Noting that {e1}⊥ = span(e1, e2) ⊕ V2 ⊕ V3, we deduce from Claim 7.5 that Fe1 ⊕ Ve1 ⊂ span(e1, f1, f2). Hence, dim(Fe1 ⊕ Ve1) ≤ 3 ≤ n 2 · Now, note that the isotropic vectors of span(e1, f1, f2) are exactly the vectors of span(e1, f1). Indeed, by Lemma 4.13 the vector f2 is non-isotropic, whereas b(e1, e1) = b(f1, f1) = b(e1, f1) = b(e1, f2) = b(f1, f2) = 0 since e1, f1 belong to Ker u and e1, f1, f2 belong to Im u. In particular, if dim(Fe1 ⊕ Ve1) = n 2 , then Fe1 ⊕ Ve1 contains the non- isotropic vector f2. Thus, with the same proof as for Claim 7.10, we deduce that (Fe1 ⊕ Ve1)⊥ contains a non-isotropic vector. It follows from Claim 7.8 that Fe1 ⊕ Ve1 is totally b-singular, and we deduce from the above that Ve1 ⊂ span(e1, f1) = Im u2. Symmetrically Vf1 ⊂ Im u2 and we conclude that every element of V stabi- lizes Im u2. From here, it is easy to conclude. The 2-dimensional totally singular subspace Im u2 is stable under all the elements of V. The induced endomorphisms are nilpotent endomorphisms of Im u2, and by Gerstenhaber's theorem in dimension 2 (or trivially if the resulting space of endomorphisms contains only the zero element) there is a non-zero vector x of Im u2 that is annihilated by all those endomorphisms. Hence, Vx = {0} and x is isotropic: the proof is complete. References [1] D.K. Bukovsek, M. Omladic, Linear spaces of symmetric nilpotent matrices, Linear Algebra Appl. 530 (2017), 384 -- 404. 37 [2] J. Draisma, H. Kraft, J. Kuttler, Nilpotent subspaces of maximal dimension in semisimple Lie algebras, Compos. Math. 142 (2006), 464 -- 476. [3] M. Gerstenhaber, On Nilalgebras and Linear Varieties of Nilpotent Matrices I, Amer. J. Math. 80 (1958), 614 -- 622. [4] N. Jacobson, Lie Algebras, Interscience, New York, 1962. [5] G.W. MacDonald, J.A. MacDougall, L.G. Sweet, On the dimension of linear spaces of nilpotent matrices, Linear Algebra Appl. 437 (2012), 2210 -- 2230. [6] B. Mathes, M. Omladic, H. Radjavi, Linear spaces of nilpotent matrices, Linear Algebra Appl. 149 (1991), 215 -- 225. [7] R. Meshulam, N. Radwan, On linear subspaces of nilpotent elements in a Lie algebra, Linear Algebra Appl. 279 (1998), 195 -- 199. [8] H. Radjavi, The Engel-Jacobson theorem revisited, J. Algebra 111 (1987), 427 -- 430. [9] C. de Seguins Pazzis, Invitation aux formes quadratiques, Calvage & Mounet, Paris, 2011. [10] C. de Seguins Pazzis, The structured Gerstenhaber problem (I), preprint, arXiv: http://arxiv.org/abs/1804.07938 (2018). [11] V.N. Serezhkin, Linear transformations preserving nilpotency (in Russian), Izv. Akad. Nauk BSSR, Ser. Fiz.-Mat. Nauk 125 (1985), 46 -- 50. [12] V.V. Sergeichuk, Classification problems for systems of forms and linear mappings, Math. USSR Izvestiya 31 (3) (1988), 481-501. 38
1504.01986	1	1504	2015-04-08T14:33:55	The Flanders theorem over division rings	[ "math.RA" ]	Let $\mathbb{D}$ be a division ring and $\mathbb{F}$ be a subfield of the center of $\mathbb{D}$ over which $\mathbb{D}$ has finite dimension $d$. Let $n,p,r$ be positive integers and $\mathcal{V}$ be an affine subspace of the $\mathbb{F}$-vector space $M_{n,p}(\mathbb{D})$ in which every matrix has rank less than or equal to $r$. Using a new method, we prove that $\dim_{\mathbb{F}} \mathcal{V} \leq \max(n,p)\,rd$ and we characterize the spaces for which equality holds. This extends a famous theorem of Flanders which was known only for fields.	math.RA	math	The Flanders theorem over division rings Cl´ement de Seguins Pazzis∗† August 20, 2018 Abstract Let D be a division ring and F be a subfield of the center of D over which D has finite dimension d. Let n, p, r be positive integers and V be an affine subspace of the F-vector space Mn,p(D) in which every matrix has rank less than or equal to r. Using a new method, we prove that dimF V ≤ max(n, p) rd and we characterize the spaces for which equality holds. This extends a famous theorem of Flanders which was known only for fields. AMS Classification: 15A03, 15A30. Keywords: Rank, Bounded rank spaces, Flanders's theorem, Dimension, Divi- sion ring. 1 Introduction Throughout the text, we fix a division ring D, that is a non-trivial ring in which every non-zero element is invertible. We let F be a subfield of the center Z(D) of D and we assume that D has finite dimension over F. Let n and p be non-negative integers. We denote by Mn,p(D) the set of all matrices with n rows, p columns, and entries in D. It has a natural structure of F-vector space, which we will consider throughout the text. We denote by Ei,j the matrix of Mn,p(D) in which all the entries equal 0, except the one at the (i, j)-spot, which equals 1. The right D-vector space Dn is naturally identified ∗Universit´e de Versailles Saint-Quentin-en-Yvelines, Laboratoire de Math´ematiques de Ver- sailles, 45 avenue des Etats-Unis, 78035 Versailles cedex, France †e-mail address: [email protected] 1 5 1 0 2 r p A 8 ] . A R h t a m [ 1 v 6 8 9 1 0 . 4 0 5 1 : v i X r a with the space Mn,1(D) of column matrices (with n rows). We naturally identify Mn,p(D) with the set of all linear maps from the right vector space Dp to the right vector space Dn. We have a ring structure on Mn(D) := Mn,n(D) with unity In, and its group of units is denoted by GLn(D). Two matrices M and N of Mn,p(D) are said to be equivalent when there are invertible matrices P ∈ GLn(D) and Q ∈ GLp(D) such that N = P M Q (this means that M and N represent the same linear map between right vector spaces over D under a different choice of bases). This relation is naturally extended to whole subsets of matrices. The rank of a matrix M ∈ Mn,p(D) is the rank of the family of its columns in the right D-vector space Dn, and it is known that it equals the rank of the family of its rows in the left D-vector space M1,p(D): we denote it by rk(M ). Two matrices of the same size have the same rank if and only if they are equivalent. Given a non-negative integer r, a rank-r subset of Mn,p(D) is a subset in which every matrix has rank less than or equal to r. Let s and t be non-negative integers. One defines the compression space R(s, t) := (cid:26)(cid:20)A B [0](n−s)×(p−t)(cid:21) A ∈ Ms,t(D), B ∈ Mn−s,t(D), C ∈ Ms,p−t(D)(cid:27). C It is obviously an F-linear subspace of Mn,p(D) and a rank-s + t subset. More generally, any space that is equivalent to a space of that form is called a com- pression space. A classical theorem of Flanders [4] reads as follows. Theorem 1 (Flanders's theorem). Let F be a field, and n, p, r be positive integers such that n ≥ p > r. Let V be a rank-r linear subspace of Mn,p(F). Then, dim V ≤ nr, and if equality holds then either V is equivalent to R(0, r), or n = p and V is equivalent to R(r, 0). The case when n ≤ p can be obtained effortlessly by transposing. Flanders's theorem has a long history dating back to Dieudonn´e [3], who tackled the case when n = p and r = n − 1 (that is, subspaces of singular matrices). Dieudonn´e was motivated by the study of semi-linear invertibility preservers on square matrices. Flanders came actually second [4] and, due to his use of determinants, he was only able to prove his results over fields with more than r elements (he added the restriction that the field should not be of characteristic 2, but a close examination of his proof reveals that it is unneces- sary). The extension to general fields was achieved more than two decades later 2 by Meshulam [5]. In the meantime, much progress had been made in the classi- fication of rank-r subspaces with dimension close to the critical one, over fields with large cardinality (see [1] for square matrices, and [2] for the generalization to rectangular matrices): the known theorems essentially state that every large enough rank-r linear subspace is a subspace of a compression space. This topic has known a recent revival. First, Flanders's theorem was ex- tended to affine subspaces over all fields [8], and the result was applied to gener- alize Atkinson and Lloyd's classification of large rank-r spaces [9]. Flanders's the- orem has also been shown to yield an explicit description of full-rank-preserving linear maps on matrices without injectivity assumptions [6, 10]. In a recent article, Semrl proved Flanders's theorem in the case of singular matrices over division algebras that are finite-dimensional over their center [11], and he applied the result to classify the linear endomorphisms of a central simple algebra that preserve invertibility. Our aim here is to give the broadest generalization of Flanders's theorem to date. It reads as follows: Theorem 2. Let D be a division ring and F be a subfield of its center such that d := [D : F] is finite. Let n, p, r be non-negative integers such that n ≥ p ≥ r. Let V be an F-affine subspace of Mn,p(D), and assume that it is a rank-r subset. Then, dimF V ≤ dnr. (1) If equality holds in (1), then: (a) Either V is equivalent to R(0, r); (b) Or n = p and V is equivalent to R(r, 0); (c) Or (n, p, r) = (2, 2, 1), #D = 2 and V is equivalent to the affine space U2 := (cid:26)(cid:20)x y x + 1(cid:21) (x, y) ∈ D2(cid:27). 0 The case when n = p, r = n − 1, F = Z(D), D is infinite and V is a linear subspace is Theorem 2.1 of [11]. As in Flanders's theorem, the case n ≤ p can be deduced from our theorem. Beware however that the transposition does not leave the rank invariant! If we 3 denote by Dop the opposite division ring1, then A ∈ Mn,p(D) 7→ AT ∈ Mp,n(Dop) is an F-linear bijection that is rank preserving and that reverses products. Thus, the case n ≥ p over Dop yields the case n ≤ p over D. Taking D as a finite field and F as its prime subfield, we obtain the following corollary on additive subgroups of matrices: Corollary 3. Let F be a finite field with cardinality q. Let n, p, r be positive integers such that n ≥ p ≥ r. Let V be an additive subgroup of Mn,p(F) in which every matrix has rank at most r. Then, #V ≤ qnr, and if equality holds then either V is equivalent to R(0, r), or n = p and V is equivalent to R(r, 0). Broadly speaking, the proof of Theorem 2 will revive some of Dieudonn´e's original ideas from [3] and will incorporate some key innovations. The main idea is to work by induction over all parameters n, p, r, with special focus on the rank 1 matrices in the translation vector space of V. In a subsequent article, this new strategy will be used to improve the classification of large rank-r spaces over fields. The proof of Theorem 2 is laid out as follows: Section 2 consists of a collection of three basic lemmas. The inductive proof of Theorem 2 is then performed in the final section. 2 Basic results 2.1 Extraction lemma Lemma 4 (Extraction lemma). Let M = (cid:20)A C B d(cid:21) be a matrix of Mn,p(D), with A ∈ Mn−1,p−1(D). Assume that rk(M ) ≤ r and rk(M + En,p) ≤ r. Then, rk A ≤ r − 1. Proof. Assume on the contrary that rk A ≥ r. Without loss of generality, we can assume that A = (cid:20) Ir [?](n−1−r)×r [?]r×(p−1−r) [?](n−1−r)×(p−1−r)(cid:21) . 1The ring Dop has the same underlying abelian group, and its multiplication is defined as y := yx. x × op 4 B1 ∈ M1,r(D). Then, by extracting sub-matrices, we find that for all δ ∈ {0, 1}, the matrix [?]1×(p−r−1)(cid:3) with C1 ∈ Mr,1(D) and We write C = (cid:20) C1 [?](n−1−r)(cid:21) and B = (cid:2)B1 Hδ := (cid:20) Ir B1 d + δ(cid:21) C1 has rank less than or equal to r. Multiplying it on the left with the invertible matrix (cid:20) Ir −B1 [0]r×1 1 (cid:21), we deduce that rk(cid:20) Ir ∀δ ∈ {0, 1}, [0]1×r d + δ − B1C1(cid:21) ≤ r. C1 This would yield ∀δ ∈ {0, 1}, d + δ − B1C1 = 0, which is absurd. Thus, rk A < r, as claimed. 2.2 Range-compatible homomorphisms on Mn,p(D) Definition 1. Let U and V be right vector spaces over D, and S be a subset of L(U, V ), the set of all linear maps from U to V . A map F : S → V is called range-compatible whenever ∀s ∈ S, F (s) ∈ Im s. The concept of range-compatibility was introduced and studied in [7]. Here, we shall need the following basic result: Proposition 5. Let n, p be non-negative integers with n ≥ 2, and F : Mn,p(D) → Dn be a range-compatible (group) homomorphism. Then, F : M 7→ M X for some X ∈ Dp. Of course here Mn,p(D) is naturally identified with the group of all linear mappings from Dp to Dn. Proof. We start with the case when p = 1. Then, F is simply an endomorphism of Dn such that F (X) is (right-)collinear to X for all X ∈ Dn. Then, it is well- known that F is a right-multiplication map: we recall the proof for the sake of 5 completeness. For every non-zero vector X ∈ Dn, we have a scalar λX ∈ D such that F (X) = X λX. Given non-collinear vectors X and Y in Dn, we have X λX+Y +Y λX+Y = (X +Y ) λX+Y = F (X +Y ) = F (X)+F (Y ) = X λX +Y λY and hence λX = λX+Y = λY . Given collinear non-zero vectors X and Y of Dn, we can find a vector Z in Dn r (X D) (since n ≥ 2), and it follows from the first step that λX = λZ = λY . Thus, we have a scalar λ ∈ D such that F (X) = Xλ for all X ∈ Dn r {0}, which holds also for X = 0. Now, we extend the result. As F is additive there are group endomorphisms F1, . . . , Fp of Dn such that F : (cid:2)C1 · · · Cp(cid:3) 7→ p Xk=1 Fk(Ck). By applying the range-compatibility assumption to matrices with only one non- zero column, we see that each map Fk is range-compatible. This yields scalars λ1, . . . , λp in D such that F : (cid:2)C1 · · · Cp(cid:3) 7→ p Xk=1 Ck λk. we find that F : M 7→ M X, as claimed. Thus, with X :=   λ1 ... λp   2.3 On the rank 1 matrices in the translation vector space of a rank-r affine subspace Notation 2. Let S be a subset of Mn,p(D) and H be a linear hyperplane of the D-vector space Dp. We define SH := {M ∈ S : H ⊂ Ker M }. Note that SH is an F-linear subspace of S whenever S is an F-linear subspace of Mn,p(D). Let S be an F-affine rank-k space, with translation vector space S. In our proof of Flanders's theorem, we shall need to find a hyperplane H such that the dimension of SH is small. This will be obtained through the following key lemma. 6 Lemma 6. Let V be an F-affine rank-r subspace of Mn,p(D), with p > r > 0. Denote by V its translation vector space. Assume that dimF VH ≥ dr for every linear hyperplane H of the D-vector space Dp. Then, V is equivalent to R(r, 0). Proof. Denote by s the maximal rank among the elements of V. Let us consider a matrix A of V with rank s. Given a linear hyperplane H of Dp that does not include Ker A, let us set TH := XM ∈VH Im M. We claim that TH = Im A. To support this, we lose no generality in assuming that A = Js := (cid:20)Is 0 0(cid:21) 0 and H = Dp−1 × {0}. The elements of VH are the matrices of V whose columns are all zero with the possible exception of the last one. For an arbitrary element N of VH, we must have rk(A + N ) ≤ s as A + N belongs to V. This shows that the last n − s rows of N must equal zero. It follows that dimF VH ≤ ds, and our assumptions show that we must have s = r and dimF VH = dr. In turn, this shows that VH is the set of all matrices of Mn,p(D) with non-zero entries only in the last column and in the first r rows, and it is then obvious that TH = Dr × {0} = Im A. Now, let us get back to the general case. Without loss of generality, we can assume that V contains A = Jr. Then, taking H = Dp−1 × {0} shows that V contains E1,p, . . . , Er,p. Given (i, j) ∈ [[1, r]] × [[1, p − 1]], taking H = {(x1, . . . , xp) ∈ Dp : xj = xp} shows that V contains Ei,j − Ei,p, and as the F-vector space V also contains Ei,p we conclude that it contains Ei,j. It follows that R(r, 0) ⊂ V . As Jr ∈ V, we deduce that 0 ∈ V, and hence V = V . Finally, assume that some matrix N of V has a non-zero row among the last n − r ones: then, we know from R(r, 0) ⊂ V that V contains every matrix of Mn,p(D) with the same last n − r rows as N ; at least one such matrix has rank greater than r, obviously. Thus, V ⊂ R(r, 0) and we conclude that V = R(r, 0). 3 The proof of Theorem 2 We are now ready to prove Theorem 2. We shall do this by induction over n, p, r. The case r = 0 is obvious, and so is the case r = p. 7 Assume now that 1 ≤ r < p. Let V be an F-affine subspace of Mn,p(D) in which every matrix has rank at most r. Denote by V its translation vector space. If V is equivalent to R(r, 0), then it has dimension dpr over F, which is less than or equal to dnr. Moreover, if equality occurs then n = p and we have conclusion (b) in Theorem 2. In the rest of the proof, we assume that V is inequivalent to R(r, 0). Thus, Lemma 6 yields a linear hyperplane H of Dp such that dimF VH < dr. Without loss of generality, we can assume that H = Dp−1 × {0}. From there, we split the discussion into two subcases, whether VH contains a non-zero matrix or not. 3.1 Case 1: VH 6= {0}. Without further loss of generality, we can assume that VH contains En,p. Let us split every matrix M of V into M = (cid:20) K(M ) [?]1×(p−1) [?](n−1)×1 ? (cid:21) with K(M ) ∈ Mn−1,p−1(D). Then, by the extraction lemma, we see that K(V) is an F-affine rank-r − 1 subspace of Mn−1,p−1(D). By induction, dimF K(V) ≤ d(n − 1)(r − 1). On the other hand, by the rank theorem dimF V ≤ dimF K(V) + d(p − 1) + dimF VH . Hence, dimF V < d(n − 1)(r − 1) + d(p − 1) + dr = d(nr + p − n) ≤ dnr. Thus, in this situation we have proved inequality (1), and equality cannot occur. 3.2 Case 2: VH = {0}. Here, we split every matrix M of V into M = (cid:2)A(M ) [?]n×1(cid:3) with A(M ) ∈ Mn,p−1(D). 8 Then, A(V) is an F-affine rank-r subspace of Mn,p−1(D), and as VH = {0} we have By induction we have and hence dimF A(V) = dimF V. dimF A(V) ≤ dnr dimF V ≤ dnr. Assume now that dimF V = dnr, so that dimF A(V) = dnr. As n > p − 1, we know by induction that A(V) is equivalent to R(0, r) (cases (b) and (c) in Theorem 2 are barred). Without loss of generality, we may now assume that A(V) = R(0, r). Then, as VH = {0} we have an F-affine map F : Mn,r(D) → Dn such that V = (cid:26)(cid:2)N [0]n×(p−1−r) F (N )(cid:3) N ∈ Mn,r(D)(cid:27). Claim 1. The map F is range-compatible unless #D = 2 and (n, p, r) = (2, 2, 1). Proof. Throughout the proof, we assume that we are not in the situation where #D = 2, n = p = 2 and r = 1. Set G1 := Dn−1 × {0}. Let us prove that F (N ) ∈ G1 for all N ∈ Mn,r(D) such that Im N ⊂ G1. We have an F-affine mapping γ : Mn−1,r(D) → D such that ∀R ∈ Mn−1,r(D), F (cid:18)(cid:20) R γ(R) (cid:21) [0]1×r(cid:21)(cid:19) = (cid:20)[?](n−1)×1 and we wish to show that γ vanishes everywhere on Mn−1,r(D). Assume that this is not true and choose a non-zero scalar a ∈ D r {0} in the range of γ. Then W := γ−1{a} is an F-affine subspace of Mn−1,r(D) with codimension at most d. Moreover, as V is a rank-r space it is obvious that every matrix in W has rank at most r − 1. Thus, by induction we know that dimF W ≤ d(n − 1)(r − 1). This leads to (n − 1)(r − 1) ≥ (n − 1)r − 1, and hence n ≤ 2. Assume that n = 2, so that p = 2 and r = 1. Then, #D 6= 2. Moreover, we must have dimF W ≤ d(n − 1)(r − 1) = 0 and hence γ is one-to-one. As D has more than 2 elements this yields a non-zero scalar b ∈ D such that γ(b) 6= 0, yielding a rank 2 matrix in V. Therefore, in any case we have found a contradiction. Thus, γ equals 0. We conclude that F (N ) ∈ G1 for all N ∈ Mn,r(D) such that Im N ⊂ G1. Using row operations, we generalize this as follows: for every linear hyperplane 9 G of the right D-vector space Dn and every matrix N ∈ Mn,r(D), we have the implication Im N ⊂ G ⇒ F (N ) ∈ G. Let then N ∈ Mn,r(D). We can find linear hyperplanes G1, . . . , Gk of the right D-vector space Dn such that Im N = Gi, and hence k Ti=1 F (N ) ∈ k Gi = Im N. \i=1 Thus, F is range-compatible. Now we can conclude. Assume first that we are not in the special situa- tion where (n, p, r, #D) = (2, 2, 1, 2). Then, we have just seen that F is range- compatible. In particular this shows that F (0) = 0, and as F is F-affine we obtain that F is a group homomorphism. Proposition 5 yields that F : N 7→ N X for some X ∈ Dr. Setting P :=   Ir [0]r×(p−1−r) −X [0](p−1−r)×r Ip−1−r [0](p−1−r)×1 [0]1×r [0]1×(p−1−r) 1 ,   we see that P is invertible and that V P = R(0, r), which completes the proof in that case. Assume finally that (n, p, r, #D) = (2, 2, 1, 2). Then, F : (cid:20)x y(cid:21) 7→ (cid:20)αx + βy + γ δx + ǫy + η (cid:21) for fixed scalars α, β, γ, δ, ǫ, η. As every matrix in V is singular, we deduce that, for all (x, y) ∈ D2, and hence (cid:12)(cid:12)(cid:12)(cid:12) x αx + βy + γ y δx + ǫy + η (cid:12)(cid:12)(cid:12)(cid:12) = 0, (ǫ + α)xy + (δ + η)x + (β + γ)y = 0. Thus, ǫ = α, δ = η and β = γ. Performing the column operation C2 ← C2 − αC1 on V, we see that no generality is lost in assuming that α = 0. Then, V = (cid:26)(cid:20)x β(y + 1) δ(x + 1)(cid:21) (x, y) ∈ D2(cid:27) y 10 and there are four options to consider: • If β = δ = 0, then V = R(0, 1); • If β = 0 and δ = 1, then V = U2; • If β = 1 and δ = 0, then the row swap L1 ↔ L2 takes V to U2; • Finally, if β = δ = 1, then the column operation C2 ← C2 + C1 followed by the row operation L1 ← L1 + L2 takes V to U2. This completes the proof of Theorem 2. References [1] M. D. Atkinson, S. Lloyd, Large spaces of matrices of bounded rank. Quart. J. Math. Oxford (2) 31 (1980), 253 -- 262. [2] L.B. Beasley, Null spaces of spaces of matrices of bounded rank. In: Current Trends in Matrix Theory, Elsevier, 1987, pp. 45 -- 50. [3] J. Dieudonn´e, Sur une g´en´eralisation du groupe orthogonal `a quatre vari- ables. Arch. Math. 1 (1948), 282 -- 287. [4] H. Flanders, On spaces of linear transformations with bounded rank. J. Lond. Math. Soc. 37 (1962), 10 -- 16. [5] R. Meshulam, On the maximal rank in a subspace of matrices. Quart. J. Math. Oxford (2) 36 (1985), 225 -- 229. [6] L. Rodman, P. Semrl, A localization technique for linear preserver problems. Linear Algebra Appl. 433 (2010), 2257 -- 2268. [7] C. de Seguins Pazzis, Range-compatible homomorphisms on matrix spaces. Preprint, arXiv: http://arxiv.org/abs/1307.3574 [8] C. de Seguins Pazzis, The affine preservers of non-singular matrices. Arch. Math. 95 (2010), 333 -- 342. [9] C. de Seguins Pazzis, The classification of large spaces of matrices with bounded rank. Israel J. Math. (2015), in press. 11 [10] C. de Seguins Pazzis, The singular linear preservers of non-singular matri- ces. Linear Algebra Appl. 433 (2010), 483 -- 490. [11] P. Semrl, Invertibility preservers on central simple algebras. J. Algebra. 408 (2014), 42 -- 60. 12
1712.06480	1	1712	2017-12-18T15:50:29	Leavitt path algebras of Cayley graphs $C_n^j$	[ "math.RA" ]	Let $n$ be a positive integer. For each $0\leq j \leq n-1$ we let $C_n^j$ denote the Cayley graph of the cyclic group $\mathbb{Z}_n$ with respect to the subset $\{1,j\}$. Utilizing the Smith Normal Form process, we give an explicit description of the Grothendieck group of each of the Leavitt path algebras $L_K(C_n^j)$ for any field $K$. Our general method significantly streamlines the approach that was used in previous work to establish this description in the specific case $j=2$. Along the way, we give necessary and sufficient conditions on the pairs $(j,n)$ which yield that this group is infinite. We subsequently focus on the case $j = 3$, where the structure of this group turns out to be related to a Fibonacci-like sequence, called the Narayana's Cows sequence.	math.RA	math	LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO Abstract. Let n be a positive integer. For each 0 ≤ j ≤ n−1 we let C j n denote the Cayley graph of the cyclic group Zn with respect to the subset {1, j}. Utilizing the Smith Normal Form process, we give an explicit description of the Grothendieck group of each of the Leavitt path algebras LK(C j n) for any field K. Our general method significantly streamlines the approach that was used in previous work to establish this description in the specific case j = 2. Along the way, we give necessary and sufficient conditions on the pairs (j, n) which yield that this group is infinite. We subsequently focus on the case j = 3, where the structure of this group turns out to be related to a Fibonacci-like sequence, called the Narayana's Cows sequence. For any finite group G, and any set of generators S of G, the Cayley graph of G with respect to S is the directed graph EG,S having vertex set {vg g ∈ G}, and in which there is an edge from vg to vh in case there exists (a necessarily unique) s ∈ S with h = gs in G. For a positive integer n (n ≥ 3) and j an integer with 0 ≤ j ≤ n − 1, we denote by C j n the Cayley graph EG,S of the cyclic group G = Zn with respect to the generating subset S = {1, j} of Zn. Concretely, the Cayley graph C j n is the directed graph consisting of n vertices {v1, v2, . . . , vn} and 2n edges {e1, e2, . . . , en, f1, f2, . . . , fn} for which s(ei) = vi, r(ei) = vi+1, s(fi) = vi, and r(fi) = vi+j, where indices are interpreted mod n. Intuitively, C j n is the graph with n vertices and 2n edges, where each vertex v emits two edges, one to its "next" neighboring vertex (which we will draw throughout this article in a clockwise direction), and one to the vertex j places clockwise from v. For instance: C 0 4 = •v1 •v4 •v2 v •v3 C 2 4 = •v1 / •v2 C 3 4 = •v1 •v4 •v3 •v4 •v2 + •v3 Leavitt path algebras of Cayley graphs were initially studied in [ASch], where a description is given of the Leavitt path algebras of the form LK(C n−1 ). It is shown in [ASch, Theorem 8] that there are exactly four isomorphism classes represented by the collection {LK(C n−1 ) n ∈ N}. In subsequent work, [AA, Section 2] contains the computation of the two important (and closely related) integers K0(LK(C j ), where A(−) denotes the incidence matrix of a directed graph, At (−) its transpose, and K0(−) denotes the Grothendieck group of a ring (see below for further details). Specifically, it is shown that, for fixed j, the integers K0(LK(C j n)) are (up to sign) the entries in the "jth Haselgrove sequence" ([AA, Definition 2.2]), a sequence investigated in the 1940s by Haselgrove in [H] (in part with an eye towards establishing Fermat's Last Theorem). n)) and det(In − At C j n n n 2010 Mathematics Subject Classification. Primary 16S99 ; Secondary 11B39, 05C99. Key words and phrases. Leavitt path algebra, Cayley graph, Narayana's Cows sequence. 1 / / v 6 6 O O o o R R / x x 8 8 O O [ [ o o + + k k J J + k k J J 2 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO Also in [AA], the following collections of K-algebras are completely described up to isomorphism: - {LK(C 0 n) n ∈ N} there is only one such algebra, - {LK(C 1 n) n ∈ N} there are infinitely many such algebras, each of them subsequently realized in terms of matrices over the standard Leavitt algebras LK(1, m), and - {LK(C 2 n) n ∈ N} there are infinitely many such algebras, each of them subsequently realized up to isomor- phism as the Leavitt path algebra of a graph having at most three vertices; moreover, the groups K0(LK(C 2 n)) are described explicitly in terms of the integers appearing in the second Haselgrove sequence H2(n), together with integers arising from the standard Fibonacci sequence. The descriptions of all these algebras follow from an application of the powerful tool known as the Restricted Algebraic Kirchberg-Phillips Theorem (see Section 4). In the current article we continue the investigation of Leavitt path algebras associated to Cayley graphs. We begin in Section 1 by giving the requisite background information about the topic. In Section 2 we present a method to compute the Grothendieck group of the Leavitt path algebra LK(C j n) for any n ≥ 3 and 0 ≤ j ≤ n − 1. The approach we utilize is completely different from the work done in [AA] for the case C 2 n. Specifically, in Theorem 2.3, we show how to reduce the computation of the Smith Normal Form of the n × n matrix In − At (data which provides C j n a complete description of the K0 group) to that of calculating the Smith Normal Form of the j × j matrix (M n j )t − Ij, where Mj is the companion matrix of a polynomial associated to a recursive sequence which arises naturally in this context. With this result in hand, we then use the Determinant Divisors Theorem to achieve the desired description of the finitely generated abelian group K0(LK(C j n)). With the general result Theorem 2.3 in hand, in Section 3 we focus our attention on the specific case j = 3. This leads us to a number-theoretic analysis which involves some perhaps surprising and apparently nontrivial connections between the third Haselgrove sequence H3 and the Narayana's Cows sequence G (Proposition 3.10). Subsequently, we briefly mention how the result for j = 2 presented in [AA] (which was achieved by using an extremely ad hoc proof that spans more than a dozen journal pages) may be re-established with only a modicum of work. In the final section we show how Theorem 2.3, together with the Restricted Algebraic Kirchberg- Phillips Theorem, allows us to realize any Leavitt path algebra of the form LK(C j n) (specifically, the Leavitt path algebra of a graph with n vertices) up to isomorphism as the Leavitt path algebra of a graph with at most j + 1 vertices. 1. Background information For any field K and directed graph E the Leavitt path algebra LK(E) has been the focus of sustained investigation since 2004. We put off until Section 4 a detailed description of these algebras, choosing instead to focus first on the construction of a monoid which can be carried out for any directed graph. For a directed graph E having n vertices v1, v2, . . . , vn we denote by AE the usual incidence matrix of E, namely, the matrix AE = (ai,j) where, for each pair 1 ≤ i, j ≤ n, the entry ai,j denotes the number of edges e in E for which s(e) = vi and r(e) = vj. Let E be a directed LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 3 graph with vertices v1, v2, . . . , vn and incidence matrix AE = (ai,j). We let Fn denote the free Z+ as monoids). We denote the abelian monoid on the generators v1, v2, . . . , vn (so Fn identity element of this monoid by z. We define a relation ≈ on Fn by setting ∼= Ln i=1 vi ≈ nXj=1 ai,jvj for each non-sink vi, and denote by ∼E the equivalence relation on Fn generated by ≈. For two ∼E equivalence classes [x] and [y] we define [x] + [y] = [x + y]; it is straightforward to show that this gives a well-defined associative binary operation on the set of ∼E equivalence classes, and that [z] acts as an identity element for this operation. We denote the resulting graph monoid Fn/ ∼E by ME. Specifially, Definition. For any n ≥ 1 and 0 ≤ j ≤ n − 1, the graph monoid MC j n Fn generated by [v1], [v2], . . . , [vn], subject to the relations (for all 1 ≤ i ≤ n), where subscripts are interpreted mod n. [vi] = [vi+1] + [vi+j] is the free abelian monoid For examples of the graph monoid MC j , see both [ASch, p. 3] (in which the graph monoid MC2 is shown to consist of the five elements {[z], [v1], [v2], [v3], [v1] + [v2] + [v3]}) and [AA, pp. 3, 4] (where the graph monoid MC2 associated to the graph C 2 4 is explicitly described). n 3 4 We present now a streamlined version of the germane background information which will be utilized throughout the remainder of the article. Much of this discussion appears in [AA]. For a unital K-algebra A, the set of isomorphism classes of finitely generated projective left A-modules is denoted by V(A). We denote the elements of V(A) using brackets; for example, [A] ∈ V(A) represents the isomorphism class of the left regular module AA. V(A) is a monoid, with operation ⊕, and zero element [{0}]. The monoid (V(A), ⊕) is conical: this means that the sum of any two nonzero elements of V(A) is nonzero, or, rephrased, that V(A)∗ = V(A) \ {0} is a semigroup under ⊕. The following striking property of Leavitt path algebras was established in [AMP, Theorem 3.5]. V(LK(E)) ∼= ME as monoids. Moreover, [LK(E)] ↔ Xv∈E0 [v] under this isomorphism. A unital K-algebra A is called purely infinite simple in case A is not a division ring, and A has the property that for every nonzero element x of A there exist b, c ∈ A for which bxc = 1A. It is shown in [AGP, Corollary 2.2] that if A is a unital purely infinite simple K-algebra, then the semigroup (V(A)∗, ⊕) is in fact a group, and, moreover, that V(A)∗ ∼= K0(A), the Grothendieck group of A. Summarizing: when LK(E) is unital purely infinite simple we have the following isomorphisms of groups: K0(LK(E)) ∼= V(LK(E))∗ ∼= M ∗ E. In particular, in this situation we have K0(LK(E)) = M ∗ E. The finite graphs E for which the Leavitt path algebra LK(E) is purely infinite simple have been explicitly described (using only graph-theoretic properties of E) in [AAP2]. This result, together with the preceding discussion, immediately yields 4 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO Proposition. [AA, Proposition 1.3] For each n ≥ 1, and for each 0 ≤ j ≤ n − 1, the K-algebra \ {[z]}, +) is a group, necessarily n) is unital purely infinite simple. In particular, M ∗ C j n = (MC j n LK(C j isomorphic to K0(LK(C j n)). Our goal here is to describe the group M ∗ C j n . Our motivation for doing so is twofold. First, the description turns out to be inherently interesting in its own right (involving the aforementioned Haselgrove sequences). Second, we may utilize the structure of this group, viewed as K0(LK(C j n)), to (quite surprisingly) glean information about the algebra LK(C j n) itself. This is done by invoking the so-called Restricted Algebraic Kirchberg-Phillips Theorem, which we describe fully in Section 4. We now review a useful computational tool. Let M ∈ Mn(Z) and view M as a linear transfor- mation M : Zn → Zn via left multiplication on columns. The cokernel of M is a finitely generated abelian group, having at most n cyclic direct summands; as such, by the invariant factors version of the Fundamental Theorem of Finitely Generated Abelian groups, we have Coker(M) ∼= Zsℓ ⊕ Zsℓ+1 ⊕ · · · ⊕ Zsn for some 1 ≤ ℓ ≤ n, where either n = ℓ and sn = 1 (i.e., Coker(M) is the trivial group), or there are (necessarily unique) nonnegative integers sℓ, sℓ+1, . . . , sn, for which the nonzero values sℓ, sℓ+1, . . . , sr satisfy si ≥ 2 and sisi+1 for ℓ ≤ i ≤ r − 1, and sr+1 = · · · = sn = 0. Coker(M) is a finite group if and only if r = n (i.e., there are no zeros in the sequence sℓ, . . . , sn). In case ℓ > 1, we define s1 = s2 = · · · = sℓ−1 = 1. Clearly then we have Coker(M) ∼= Zs1 ⊕ Zs2 ⊕ · · · ⊕ Zsℓ ⊕ · · · ⊕ Zsn, since any additional direct summands are isomorphic to the trivial group Z1. Remark 1.1. It is straightforward to establish that if P, Q are invertible in Mn(Z), then Coker(M) ∼= Coker(P MQ). (We note that any such P and Q must have determinant ±1.) This means that if N ∈ Mn(Z) is a matrix which is constructed by performing any sequence of Z-elementary row and/or column operations starting with M, then Coker(M) ∼= Coker(N) as abelian groups. Definition 1.2. Let M ∈ Mn(Z), and suppose Coker(M) ∼= Zs1 ⊕ Zs2 ⊕ · · · ⊕ Zsn as described above. The Smith Normal Form of M is the n × n diagonal matrix diag(s1, s2, . . . , sr, 0, . . . , 0). For any matrix M ∈ Mn(Z), the Smith Normal Form of M exists and is unique. If D ∈ Mn(Z) is a diagonal matrix with entries d1, d2, . . . , dn then clearly Coker(D) ∼= Zd1 ⊕ Zd2 ⊕ · · · ⊕ Zdn. In the end we have the following. Proposition 1.3. Let M ∈ Mn(Z), and let S denote the Smith Normal Form of M. Suppose the diagonal entries of S are s1, s2, . . . , sn. Then Coker(M) ∼= Zs1 ⊕ Zs2 ⊕ · · · ⊕ Zsn. In particular, if there are no zero entries in the Smith Normal Form of M, then Coker(M) = s1s2 · · · sn = det(S) = det(M). The key computational device we will utilize to compute the Smith Normal Form of a matrix M is the following. LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 5 Determinant Divisors Theorem. [N, Theorem II.9] Let M ∈ Mn(Z). Define α0 := 1, and, for each 1 ≤ i ≤ n, define the ith determinant divisor of M to be the integer αi := the greatest common divisor of the set of all i × i minors of M. Let s1, s2, ..., sn denote the diagonal entries of the Smith Normal Form of M, and assume that each si is nonzero. Then for each 1 ≤ i ≤ n. si = αi αi−1 Suppose now that E is a finite directed graph having n vertices v1, v2, . . . , vn. Consider the matrix In − At E as a linear transormation Zn → Zn. Invoking the discussion in [AALP, Section 3], in the case LK(E) is purely infinite simple we have that E, where In is the identity n × n matrix. As above, we may view In − At K0(LK(E)) ∼= M ∗ E ∼= Coker(In − At E). Under this isomorphism [vi] 7→ ~bi + Im(In − At E), where ~bi is the element of Zn which is 1 in the ith coordinate and 0 elsewhere. In other words, when LK(E) is purely infinite simple, then K0(LK(E)) is the cokernel of the linear transformation In − At E : Zn → Zn induced by matrix multiplication. Example 1.4. Suppose E = Rm (m ≥ 2), the "rose with m petals" graph having one vertex and m loops. Because m ≥ 2, LK(Rm) is purely infinite simple. Here AE is the 1 × 1 matrix (m), so I1 − At Rm is the 1 × 1 matrix (1 − m), and we have K0(LK(Rm)) ∼= Z1 ∼= Zm−1. 1−m Proposition 1.3 then yields the following. Proposition 1.5. Suppose E is a finite graph with E0 = n, and suppose also that LK(E) is purely infinite simple. Let S be the Smith Normal Form of the matrix In − At E, with diagonal entries s1, s2, . . . , sn. Then K0(LK(E)) ∼= Zs1 ⊕ Zs2 ⊕ · · · ⊕ Zsn. Moreover, if K0(LK(E)) is finite, then an analysis of the Smith Normal Form of the matrix In − At E yields K0(LK(E)) = det(In − At E). (This is immediate, since as noted above any invertible matrix in Mn(Z) has determinant ±1.) Conversely, K0(LK(E)) is infinite if and only if det(In − At E) = 0. In [H], Haselgrove introduces for each pair of positive integers n, k the integer ak(n) := n−1Yℓ=0 (1 − ωℓ − ωk ℓ ), where ωℓ = cos( 2πℓ some elementary number theory.) Subsequently, in [AA, Definition 2.2] the integer n ) in C. (That this expression indeed yields an integer follows from n ) + i sin( 2πℓ and, for fixed k, the sequence ak(n) is denoted by Hk(n), Hk(1), Hk(2), Hk(3), ... 6 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO is referred to as the kth Haselgrove sequence. It is of historical interest to note that Haselgrove's mo- tivation for considering these integers ak(n) was for their potential use in establishing a connection between a resolution of Fermat's Last Theorem (at the time, of course, Fermat's Last Conjecture) and some integers which share properties of the Mersenne numbers. We recall some properties of the integers ak(n) and Hk(n) (established [AA]) which show why these are germane in the current context, and then finish the section with a new result which will be useful in the sequel. Proposition. (See [AA, Section 2]) Let n ∈ N and 0 ≤ k ≤ n − 1. (1) ak(n) = det(In − At (2) det(In − At ) ≤ 0. (This is established by some elementary analysis in C.) ). (This is established using the notion of circulant matrices.) C k n C k n In particular, det(In − At ) = −Hk(n). C k n (3) If Hk(n) > 0, then K0(LK(C k (4) Hk(n) = 0 if and only if K0(LK(C k n)) is infinite. n)) = Hk(n) = Coker(In − At ). C k n Proposition 1.6. Let n ∈ N and 0 ≤ k ≤ n − 1. Then Hk(n) = 0 if and only if k ≡ 5 (mod 6) and n ≡ 0 (mod 6). Proof. By the previous recollections from [AA] we have Hk(n) = − n−1Yℓ=0 (1 − ωℓ − ωk ℓ ), where ωℓ = cos( 2πℓ for some 0 ≤ ℓ ≤ n − 1. In particular, we must have ωℓ + ωk assume 0 ≤ θ < 2π), we have n )+i sin( 2πℓ n ) in C. Then Hk(n) = 0 if and only if one of the factors 1−ωℓ −ωk ℓ = 1. Letting θ = 2πℓ ℓ = 0 n (which we may cos(θ) + cos(kθ) = 1 and sin(θ) + sin(kθ) = 0. The second equation implies that kθ ≡ −θ (mod 2π) or that kθ ≡ 2π − θ (mod 2π). In the first case, which is a contradiction. Hence, the second condition must be true, and cos(θ) + cos(kθ) = cos(θ) + cos(−θ) = 0, Hence, θ = π 3 or θ = 5π π 3 In either case, n ≡ 0 (mod 6). = cos(θ) + cos(kθ) = cos(θ) + cos(2π − θ) = 2 cos(θ) = 1. 3 . Substituting in θ = 2πℓ 2πℓ n n , we see that =⇒ n = 6ℓ, =⇒ 5n = 6ℓ. 2πℓ n 5π 3 or = Similarly, kθ ≡ 2π − θ (mod 2π) implies that (k + 1)θ ≡ 2π ≡ 0 (mod 2π). Hence, for some integer m, (k + 1) π 3 = 2mπ =⇒ k + 1 = 6m, or (k + 1) 5π 3 = 2mπ =⇒ 5(k + 1) = 6m. In either case, k + 1 ≡ 0 (mod 6), or k ≡ 5 (mod 6). LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n Finally, when n ≡ 0 (mod 6) and k ≡ 5 (mod 6), then letting ℓ = n 1 − ωℓ − ωk ℓ = 1 − e 2πi 6 − (e 2πi 6 )k = 1 − e 2πi 6 − (e 6 implies that 6 )−1 = 0 2πi Since one of the factors of Hk(n) is zero, we conclude that Hk(n) = 0. 2. The Smith Normal Form of the matrix In − At C j n 7 (cid:3) In order to investigate the Leavitt path algebras {LK(C j n studied in [AA, Section 4]. The generating relations for M ∗ C j n n) n ∈ N}, we begin in a manner similar are given to that used in the case C 2 by [vi] = [vi+1] + [vi+j] for 1 ≤ i ≤ n, where subscripts are interpreted mod n. We will focus on the element [v1] of M ∗ ; C j n corresponding to any statement established for [v1] in M ∗ , there will be (by the symmetry of the C j n relations) an analogous statement in M ∗ C j n for each [vi], 1 ≤ i ≤ n. The computation of the Smith Normal Form of the n × n matrix In − At C j n is the key tool for determining the K0 of the Leavitt path algebra C j n. We show below that this computation reduces to calculating the Smith Normal Form of a j × j matrix. The authors are quite grateful to M. Iovanov for suggesting this approach. Definition 2.1. Let p(x) = xj + cj−1xj−1 + · · · + c1x + c0 be a degree j monic polynomial with integer coefficients. The companion matrix M(p) of p(x) is the j × j matrix M(p) :=   0 0 0 . . . 1 0 0 . . . 0 1 0 . . . ... . . . 0 0 0 . . . 0 0 0 . . . ... ... 0 −c0 0 −c1 0 −c2 ... ... 0 −cj−2 1 −cj−1   . For j ≥ 2 we define pj(x) = xj − xj−1 − 1 ∈ Z[x]. The companion matrix of pj(x), which we will denote by Mj, is then the j × j matrix Mj := M(pj(x)) =   0 0 0 . . . 1 0 0 . . . 0 1 0 . . . ... . . . 0 0 0 . . . 0 0 0 . . . ... ... 0 1 0 0 0 0 ... ... 0 0 1 1   . Remark 2.2. Clearly the two matrices In − At − In have the same Smith Normal Form C j n (i.e., have isomorphic cokernels). In the sequel we choose to analyze the latter, because it is easier to work with computationally. A similar statement holds for the matrices M j n)t − Ij. We note that in a more general analysis of the structure of Leavitt path algebras than the one carried out here, such an interchange might possibly forfeit some important information. n − Ij and (M j and AC j n 8 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO Theorem 2.3. Let n ≥ j. Then Coker(AC j n − In) ∼= Coker(M n j − Ij). Proof. The proof in the case j ≤ n ≤ 2j is quite similar to the proof which we give here for the case n > 2j, but requires some extra computational and notational energy; we thereby leave the proof of the j ≤ n ≤ 2j case to the interested reader. Thus we assume that n > 2j. By definition, the entry in the kth row and ℓth column of AC j − In is given by n (AC j n − In)kℓ =  −1 1 0 if ℓ = k if ℓ ≡ k + 1 or k + j otherwise. (mod n) Our goal is to obtain a diagonal matrix with integers along the diagonal through elementary row and column operations involving only integer multiples. We first note that the bottom left j × j submatrix of the matrix AC j n with the columns cyclically permuted: − In can be written as Mj   1 0 0 . . . 0 1 0 0 0 1 ... . . . 0 0 0 . . . 1 0 0 . . . 0 0 0 0 0 0 ... 1 0 0 1   = MjP, where P is the j × j permutation matrix P =   0 1 0 . . . 0 0 1 0 0 0 ... . . . 0 0 0 . . . 1 0 0 . . . 0 0 0 0 0 0 ... 0 1 0 0   The first (n − 2j) reduction steps of the Smith Normal Form will result in an (n − 2j) × (n − 2j) identity submatrix in the upper left corner. On the bottom j rows, the ith reduction step adds the ith column to the (i + 1)th column and the (i + j)th column, then zeroes out the ith column. The matrix that accomplishes this reduction step is (cid:2)vi vi+1 . . . vi+j−1(cid:3) ·   1 0 0 . . . 1 0 0 0 1 0 ... 0 0 0 0 0 0 . . . . . . 0 1 0 0 0 0 ... 0 0 1 0   =(cid:2)vi + vi+1 . . . vi+j−1 vi(cid:3) , LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 9 and this matrix is conjugate to the companion matrix Mj via the permutation matrix P :   1 0 0 . . . 1 0 0 0 1 0 ... 0 0 0 0 0 0 . . . . . . 0 1 0 0 0 0 ... 0 0 1 0   = P −1MjP. After i reduction steps, the first j × j submatrix with nonzero column vectors on the bottom j rows will be Denote by Q the j × j matrix MjP · (P −1MjP )i = M i+1 j P. Q = 0 1 0 −1   −1 1 0 −1 0 ... 0 0 0 0 0 0 . . . . . . . . . 0 0 0 0 0 0 ... −1 1 0 −1   . Then after n − 2j reduction steps, we have AC j n − In ∼  In−2j 0(n−2j)×j 0(n−2j)×j 0j×(n−2j) 0j×(n−2j) M n−2j+1 Q MjP Q P j   . Because each reduction step only adds previous columns to the following existing columns, the next j reduction steps results in AC j n − In ∼(cid:20) In−j 0j×(n−j) M n−j+1 j 0(n−j)×j P + Q(cid:21) . Finally, we adjust the bottom right j × j matrix by adding the (n − j + 1)th column through the (n − 1)th column to the nth column, adding the (n − j + 1)th column through the (n − 2)th column to the (n − 1)th column, and so on until adding the (n − j + 1)th column to the (n − j + 2)th column. This procedure is equivalent to multiplying M n−j+1 P + Q on the right by the j × j matrix j R =   1 1 1 . . . 0 1 1 0 0 1 ... 0 0 0 0 0 0 . . . . . . 1 1 1 1 1 1 ... 1 1 0 1   . A straightforward calculation shows that P R = M j−1 j and QR = −Ij, 10 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO resulting in the final bottom right j × j submatrix is (M n−j+1 j P + Q)R = M n j − Ij. Therefore, AC j n − In ∼(cid:20) In−j 0j×(n−j) M n 0(n−j)×j j − Ij(cid:21) and thus by Remark 1.1, we conclude that Coker(AC j n − In) ∼= Coker(M n j − Ij). (cid:3) 3. The case j = 3, and the case j = 2 (briefly) revisited We have shown in Theorem 2.3 that for any n ≥ j, the cokernel of the n × n matrix AC j − In is isomorphic to the cokernel of the j × j matrix M n j − Ij. In this section we investigate in detail the specific situation when j = 3. We do so for two reasons: this case will provide some insight as to how the general case works, and, as it turns out, the j = 3 case provides a sort of "sweet spot" in the general setting. We conclude the section by showing how Theorem 2.3 dramatically simplifies the proof of the corresponding result in the j = 2 case as compared to the proof given in [AA]. n An important role in the j = 3 case will be played in this situation by the elements of the Narayana's Cows sequence G, defined recursively by setting G(1) = 1, G(2) = 1, G(3) = 1, and G(n) = G(n − 1) + G(n − 3) for all n ≥ 4. (We may also define G(0) = 0, G(−1) = 0, G(−2) = 1 and G(−3) = 0 consistent with the given recursion equation). This name is used in the Online Encyclopedia of Integer Sequences [OEIS, Sequence A000930]. (Indeed, this sequence has gained some notoriety in popular culture, including the composition of a musical piece based on it.) The first few terms of the sequence G(n) (n ≥ 1) are: G : 1, 1, 1, 2, 3, 4, 6, 9, 13, 19, 28, 41, 60, 88, 129, . . . By the aforementioned [AA, Proposition 1.3] we have that M ∗ C3 n have is indeed a group. In M ∗ C3 n we [v1] = [v2] + [v4] = ([v3] + [v5]) + [v4] = ([v4] + [v6]) + [v5] + [v4] = 2[v4] + [v5] + [v6] = 2([v5] + [v7]) + [v5] + [v6] = 3[v5] + [v6] + 2[v7] = · · · which by an easy induction gives, for 1 ≤ i ≤ n, [v1] = G(i − 1)[vi−1] + G(i − 3)[vi] + G(i − 2)[vi+1]. Thus is the Narayana's Cows sequence related to the structure of M ∗ C3 n Setting i = n, and using that [vn+1] = [v1] by notational convention, we get in particular that [v1] = G(n − 1)[vn−1] + G(n − 3)[vn] + G(n − 2)[v1], so that 0 = G(n − 1)[vn−1] + G(n − 3)[vn] + (G(n − 2) − 1)[v1] in M ∗ C3 n . It will be quite useful to have an expression for the 3 × 3 matrix M n 3 in terms of the Narayana's Cows sequence, which is the content of the following lemma. M3 =  G(−1) G(0) G(1) G(−2) G(−1) G(0) G(0) G(1) G(2)   . G(n − 3) G(n − 2) G(n − 1) G(n − 4) G(n − 3) G(n − 2) G(n − 2) G(n − 1) G(n)     ·    G(n − 3) G(n − 2) G(n − 1) G(n − 4) G(n − 3) G(n − 2) G(n − 2) G(n − 1) G(n) 0 0 1 1 0 0 0 1 1 G(n − 2) G(n − 1) G(n − 3) + G(n − 1) G(n − 3) G(n − 2) G(n − 4) + G(n − 2) G(n − 1) G(n − 2) + G(n) G(n)     G(n − 2) G(n − 1) G(n − 3) G(n − 2) G(n − 1) G(n + 1) G(n − 1) G(n) G(n) Now suppose that Then M n 3 = M n−1 3 as desired. Corollary 3.2. For all n ≥ 3, M n−1 3 =  · M3 =  =  =  LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 11 Lemma 3.1. Let G denote the Narayana's Cows sequence. Then for each n ∈ N, M n 3 =  G(n − 2) G(n − 1) G(n − 3) G(n − 2) G(n − 1) G(n − 1) G(n + 1) G(n) G(n)   . Proof. The proof is by induction. As mentioned previously, we may extend the G(n) sequence by setting G(0) = 0, G(−1) = 0, and G(−2) = 1. Thus we have that the statement is true for n = 1:   , (cid:3) (cid:3) Coker(AC3 n − In) ∼= Coker  G(n − 2) − 1 G(n − 3) G(n − 1) G(n − 2) − 1 G(n − 1) G(n) G(n) G(n − 1) G(n + 1) − 1 where G's are the Narayana's Cows numbers. Proof. The result follows immediately from Theorem 2.3 and Lemma 3.1. We now analyze the Smith Normal Form for the matrix M n 3 − I3. Definition 3.3. Let n ∈ N. By Lemma 3.1 we have (M n 3 )t − I3 =  G(n − 2) − 1 G(n − 3) G(n − 1) G(n − 2) − 1 G(n − 1) G(n) G(n) G(n − 1) G(n + 1) − 1   . For i = 1, 2, 3 and each n ≥ 1, the corresponding i-th determinant divisors α1(n), α2(n), and α3(n) of (M n 3 )t − I3 have the following values. 12 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO i = 1: α1(n) is the greatest common divisor of the nine 1 × 1 minors (M n 3 )t − I3, i.e., of the nine 3 )t − I3. By eliminating repeated entries, we see that α1(n) is the greatest common entries of (M n divisor of five integers, to wit: α1(n) = gcd{ G(n − 2) − 1, G(n − 3), G(n − 1), G(n), G(n + 1) − 1 }. i = 2: α2(n) is the greatest common divisor of the nine 2 × 2 minors (M n 3 )t − I3, i.e., of the determinants of the nine 2 × 2 submatrices of (M n 3 )t − I3. By doing the standard computation for determinants of 2 × 2 matrices, and then eliminating repeated results, we see that α2(n) is the greatest common divisor of six integers, to wit: α2(n) = gcd{ (G(n − 2) − 1)2 − G(n − 1)G(n − 3), (G(n − 2) − 1)(G(n + 1) − 1) − G(n)G(n − 1), G(n − 1)(G(n + 1) − 1) − G(n)2, G(n − 3)(G(n + 1) − 1) − G(n − 1)2 }. (G(n − 2) − 1)G(n − 1) − G(n)G(n − 3), (G(n − 2) − 1)G(n) − G(n − 1)2, i = 3: α3(n) is the greatest common divisor of the one 3 × 3 minor of (M n 3 )t − I3, in other words, α3(n) = det((M n 3 )t − I3). By Proposition 1.6, since 3 6≡ 5 mod6, we have that Coker(In − At C j n ) is finite, so that by Theorem 3 )t − I3) is also finite, which yields that necessarily none of the entries in the Smith 3 )t − I3 is zero. Therefore, by the Determinant Divisors Theorem, the Smith 2.3 Coker((M n Normal Form of (M n Normal Form of the matrix (M n 3 )t − I3 is given by: SNF((M n 3 )t − I3) =  α1(n) 0 0 0 α2(n) α1(n) 0 0 0 α3(n) α2(n)   , where the values of α1(n), α2(n), and α3(n) are as presented in Definition 3.3. As a consequence, Corollary 3.2 immediately yields the following result. Corollary 3.4. Let n ∈ N. Then K0(LK(C 3 n)) ∼= Zα1(n) ⊕ Z α2(n) α1(n) ⊕ Z α3(n) α2(n) . Corollary 3.4 is the key ingredient we will utilize to prove the main result about the structure of the K0(LK(C 3 n)). Remark 3.5. Our goal for the remainder of this section will be to present a more efficient descrip- tion of the determinant divisors α1(n), α2(n) and α3(n) than those given in Definition 3.3. Our motivation is as follows. In [AA], a description of K0(LK(C 2 n)) is given in terms of greatest common divisors of pairs of integers involving terms of the Fibonacci sequence. Our aim here is to establish the analogous result, by describing K0(LK(C 3 n)) in terms of greatest common divisors of triples of integers involving terms of the Narayana's Cows sequence. For integers a, b, gcd{a, b} denotes the greatest common divisor of a and b. A key role will be played by the following integer. LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 13 Definition 3.6. For any positive integer n we define d3(n) := gcd{ G(n − 1), G(n − 3), G(n − 2) − 1 }. The first few terms of the sequence d3(n) (n ≥ 1) are: d3 : 1, 1, 1, 1, 1, 1, 2, 3, 1, 1, 1, 1, 1, 4, 1, 3, 1, 1, . . . To begin with we show that the first determinant divisor α1(n) coincides with the integer d3(n) given in Definition 3.6. To achieve it recall that we use the following well-known property of greatest common divisors: if bn+1 is a Z-linear combination of the integers b1, b2, . . . , bn, then (∗) gcd{b1, b2, . . . , bn, bn+1} = gcd{b1, b2, . . . , bn}. Lemma 3.7. Let n ∈ N and d3(n) = gcd{G(n − 1), G(n − 3), G(n − 2) − 1}. Then α1(n) = d3(n). Proof. According to Definition 3.3, α1(n) := gcd{G(n − 2) − 1, G(n − 3), G(n − 1), G(n), G(n + 1) − 1}. But G(n + 1) − 1 = G(n) + (G(n − 2) − 1), and in turn G(n) = G(n − 3) + G(n − 1), so applying (∗) twice in order gives α1(n) = gcd{ G(n − 2) − 1, G(n − 3), G(n − 1), G(n), G(n + 1) − 1 } = gcd{ G(n − 2) − 1, G(n − 3), G(n − 1), G(n) } = gcd{ G(n − 2) − 1, G(n − 3), G(n − 1) } = d3(n). (cid:3) We focus now our attention in analyzing the second determinant divisor α2(n) given in Definition 3.3. Thanks to property (∗) we may reduce the number of terms appearing in its expression. Definition 3.8. Let n ∈ N. We define d′ 3(n) = gcd{G(n − 1)G(n − 3) − (G(n − 2) − 1)2, G(n)G(n − 3) − G(n − 1)(G(n − 2) − 1), G(n − 1)2 − G(n)(G(n − 2) − 1)}. The first terms of the d′ 3(n) sequence (n ≥ 1) are: d′ 3 : 1, 1, 1, 1, 1, 1, 4, 9, 1, 1, 1, 1, 1, 16, 1, 9, 1, 1, ... Lemma 3.9. Consider the second determinant divisor α2(n) from Definition 3.3. Then Proof. By definition we have that α2(n) = d′ 3(n). α2(n) := gcd{(G(n − 2) − 1)2 − G(n − 1)G(n − 3), (G(n − 2) − 1)G(n − 1) − G(n)G(n − 3), (G(n − 2) − 1)(G(n + 1) − 1) − G(n)G(n − 1), (G(n − 2) − 1)G(n) − G(n − 1)2, G(n − 1)(G(n + 1) − 1) − G(n)2, G(n − 3)(G(n + 1) − 1) − G(n − 1)2}. 14 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO Taking into account that G(n + 1) − 1 = G(n) + (G(n − 2) − 1) and G(n) = G(n − 3) + G(n − 1), and applying (∗) we have α2(n) = gcd{(G(n − 2) − 1)2 − G(n − 1)G(n − 3), (G(n − 2) − 1)G(n − 1) − G(n)G(n − 3), (G(n − 2) − 1)G(n) − G(n − 1)2, G(n − 1)(G(n + 1) − 1) − G(n)2, G(n − 3)(G(n + 1) − 1) − G(n − 1)2} = gcd{(G(n − 2) − 1)2 − G(n − 1)G(n − 3), (G(n − 2) − 1)G(n − 1) − G(n)G(n − 3), (G(n − 2) − 1)G(n) − G(n − 1)2, G(n − 3)(G(n + 1) − 1) − G(n − 1)2} = gcd{(G(n − 2) − 1)2 − G(n − 1)G(n − 3), (G(n − 2) − 1)G(n − 1) − G(n)G(n − 3), (G(n − 2) − 1)G(n) − G(n − 1)2} = gcd{G(n − 1)G(n − 3) − (G(n − 2) − 1)2, G(n)G(n − 3) − G(n − 1)(G(n − 2) − 1), G(n − 1)2 − G(n)(G(n − 2) − 1)} (since gcd{a, b} = gcd{−a, b} for any integers a, b) = d′ 3(n). (cid:3) Finally, we show that the third determinant divisor α3(n) appearing in Definition 3.3 exactly coincides with H3(n), the nth term of the third Haselgrove sequence. As described above, Hk(n) :=(cid:12)(cid:12)(cid:12)det(In − At C k n ℓ(cid:1)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) )(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) n−1Yℓ=0(cid:0)1 − ωℓ − ωk where ωℓ = e interested in the third Haselgrove sequence H3(n), of which the first few terms (n ≥ 1) are: (0 ≤ ℓ ≤ n − 1) are the n distinct nth roots of unity in C. We are particularly 2πiℓ n H3 : 1, 3, 1, 3, 11, 9, 8, 27, 37, 33, 67, 117, 131, 192, 341, . . . This sequence has many interesting number-theoretic characteristics (e.g., H3(n) is a divisibility sequence); we investigate this and additional properties in [AEG2]. Proposition 3.10. Let n ≥ 1. Consider the 3 × 3 matrix (M n 3 )t − I3 =  G(n − 2) − 1 G(n − 3) G(n − 1) G(n − 2) − 1 G(n − 1) G(n) G(n) G(n − 1) G(n + 1) − 1   , and recall that α3(n) := det((M n 3 )t − I3). Then α3(n) = H3(n) 6= 0. LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 15 Proof. By Proposition 1.6 we have H3(n) 6= 0 for all n ≥ 1. Thus we may invoke the previously cited [AA, Proposition 2.3] and [AA, Proposition 2.5] to get H3(n) = det(In − At ). By Theo- 3 − I3 have isomorphic cokernels. Therefore, using Proposition 1.3 we rem 2.3, AC3 get n − In and M n C3 n det(In − At C3 n ) = det(AC3 n − In) = det(M n 3 − I3) = det((M n 3 )t − I3) := α3(n). We conclude that α3(n) = H3(n) for all n ≥ 1. (cid:3) We are now in a position to give the main result of this section. Applying Corollary 3.4, Lemma 3.7, Lemma 3.9 and Proposition 3.10 finally we get: Theorem 3.11. Let n ∈ N. Then K0(LK(C 3 n)) ∼= Zd3(n) ⊕ Z d′ 3(n) d3(n) ⊕ Z H3(n) . d′ 3 (n) Example 3.12. Using Theorem 3.11, here are explicit descriptions of the Grothendieck groups K0(LK(C 3 n)) which arise for small values of n. n = 3 : K0(LK(C 3 n = 4 : K0(LK(C 3 n = 5 : K0(LK(C 3 n = 6 : K0(LK(C 3 n = 7 : K0(LK(C 3 3 )) ∼= {0} 4 )) ∼= Z3 5 )) ∼= Z11 6 )) ∼= Z9 7 )) ∼= Z2 ⊕ Z2 ⊕ Z2. It is also possible for the K0 group to consist of exactly two nontrivial direct summands; for instance, n = 30: K0(LK(C 3 30)) ∼= Z31 ⊕ Z3069. To finish this section we demonstrate that the results of [AA] follow from Theorem 2.3. When j = 2, the companion matrix is M2 =(cid:20)0 1 1 1(cid:21) . Let F (n) be the nth Fibonacci number. Then a well-known Fibonacci identity states that 2 − I2 =(cid:20)F (n − 1) − 1 F (n) M n F (n + 1) − 1(cid:21) . F (n) So the determinant divisors will be α1 (∗) = gcd(F (n − 1) − 1, F (n)) := d2(n) and α2 = det(M n 2 − I2) = (F (n + 1) − 1)(F (n − 1) − 1) − F (n)2. Another Fibonacci identity states that F (n + 1)F (n − 1) − F (n)2 = (−1)n. 16 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO Putting these two together, α2 = F (n + 1)F (n − 1) − F (n + 1) − F (n − 1) + 1 − F (n)2 = (−1)n − F (n + 1) − F (n − 1) + 1 = −(F (n + 1) + F (n − 1) − 1 − (−1)n), which is negative of the formula for H2(n) in Equation (HtoF) of [AA, Proposition 4.4]. Applying the Smith Normal Form, we obtain the main result of [AA, Theorem 4.13]: K0(LK(C 2 n)) ∼= Zd2(n) ⊕ Z H2(n) d2(n) . 4. Leavitt path algebras of the graphs C 3 n is in its realization as the Grothendieck group of the Leavitt path algebra LK(C j As mentioned in the introductory remarks, the primary motivation for identifying the structure of the group M ∗ n), C j n and subsequent utilization in identifying LK(C j In the final section of the article we bring this program to fruition. We begin by recalling briefly some of the basic ideas; for additional information, see e.g. [AAS]. n) up to isomorphism. Definition of Leavitt path algebra. Let K be a field, and let E = (E0, E1, r, s) be a directed graph with vertex set E0 and edge set E1. The Leavitt path K-algebra LK(E) of E with coefficients in K is the K-algebra generated by a set {v v ∈ E0}, together with a set of variables {e, e∗ e ∈ E1}, which satisfy the following relations: vw = δv,wv for all v, w ∈ E0, s(e)e = er(e) = e for all e ∈ E1, r(e)e∗ = e∗s(e) = e∗ for all e ∈ E1, (V) (E1) (E2) (CK1) e∗e′ = δe,e′r(e) for all e, e′ ∈ E1, (CK2). ✷ (CK2) v =P{e∈E1s(e)=v} ee∗ for every v ∈ E0 for which 0 < s−1(v) < ∞. An alternate description of LK(E) may be given as follows. For any graph E let bE denote the Then LK(E) is the usual path algebra KbE, modulo the ideal generated by the relations (CK1) and "double graph" of E, gotten by adding to E an edge e∗ in a reversed direction for each edge e ∈ E1. It is easy to show that LK(E) is unital if and only if E0 is finite. This is of course the case when n. We now have the necessary background information in hand which allows us to present E = C j the powerful tool which will yield a number of key results. The Restricted Algebraic KP Theorem. [ALPS, Corollary 2.7] Suppose E and F are finite graphs for which the Leavitt path algebras LK(E) and LK(F ) are purely infinite simple. Suppose that there is an isomorphism ϕ : K0(LK(E)) → K0(LK(F )) for which ϕ([LK(E)]) = [LK(F )], and suppose also that the two integers det(IE0 − At F ) have the same sign (i.e., are either both nonnegative, or both nonpositive). Then LK(E) ∼= LK(F ) as K-algebras. E) and det(IF 0 − At The proof of the Restricted Algebraic KP Theorem utilizes deep results and ideas in the theory of symbolic dynamics. The letters K and P in its name derive from E. Kirchberg and N.C. Phillips, who (independently in 2000) proved an analogous result for graph C ∗-algebras. (We note that this analogous result does not include the hypothesis on the signs of the germane determinants; it is not LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 17 known whether this hypothesis is required for the algebraic result, hence the addition of the word 'Restricted' to the name.) We are also in position to apply Algebraic KP Theorem to explicitly realize the algebras LK(C 3 n) as the Leavitt path algebras of graphs having four vertices. The following will be important here: by [AA, Proposition 1.5], for any pair n, j we have that the identity element of the group K0(C j n) is the element [LK(C j Proposition 4.1. Let n ∈ N. The Leavitt path algebra LK(C 3 algebra LK(En), where En is the graph with four vertices given by n) is isomorphic to the Leavitt path n)] =Pv∈E0[v]. •u2 (2+d3(n)) (2) •u1 •u4 (cid:18)2+ H3(n) d′ 3(n) (cid:19) (cid:18)2+ d′ 3(n) d3(n)(cid:19) •u3 (where the numbers in parentheses indicate the number of loops at the indicated vertex). Proof. Using the characterization given in [AAP2], it follows easily that the graph En satisfies the conditions for LK(En) to be unital purely infinite simple. The incidence matrix of En is A straightforward computation yields that the Smith Normal Form of I4 − At En is which immediately yields that K0(LK(En)) is isomorphic to Zd3(n) ⊕ Z d′ 3(n) d3(n) ⊕ Z H3(n) d′ 3(n) . AEn = so that I4 − At En = −       1 2 1 2 + d3(n) 1 1 1 1 1 1 2 + d′ 3(n) d3(n) 1 1 1 1 2 + H3(n) d′ 3(n)   , 1 1 1 1 + d3(n) 1 1 1 1 1 1 1 + d′ 3(n) d3(n) 1 1 1 1 1 + H3(n) d′ 3(n)   . 0 1 0 d3(n) 0 0 0 0 0 0 d′ 3(n) d3(n) 0 0 0 0 H3(n) d3(n)   , w w 7 7 , , L L l l w w [ [ v v 7 7 [ [ L L E E 18 GENE ABRAMS, STEFAN ERICKSON, AND CRIST ´OBAL GIL CANTO Also, it is straightforward to check that det(I4 − At En) = −H3(n) < 0. (We remind the reader that the sign of the determinant of a matrix cannot be gleaned from the Smith Normal Form of the matrix; specifically, one must compute the determinant of I4 − At En directly from the matrix itself.) Finally, by invoking the relation in K0(LK(En)) at u1, we have [u1] + [u2] + [u3] + [u4] = (2[u1] + [u2] + [u3] + [u4]) + [u2] + [u3] + [u4] = 2([u1] + [u2] + [u3] + [u4]), so that σ = [u1] + [u2] + [u3] + [u4] satisfies σ = 2σ in the group K0(LK(En)), which gives that σ = [u1] + [u2] + [u3] + [u4] is the identity element of K0(LK(En)). So the purely infinite simple unital Leavitt path algebras LK(C 3 n) and LK(En) have these prop- erties: (1) K0(LK(C 3 n)) ∼= K0(LK(En)) (as each is isomorphic to Zd3(n) ⊕ Z d′ 3(n) d3(n) ⊕ Z H3(n) ), d′ 3 (n) n)] to [LK(En)] (as each of these is the identity (2) this isomorphism necessarily takes [LK(C 3 element in their respective K0 groups), and ) and det(I4 − At En) are negative. (3) both det(In − At Thus the graphs C 3 C3 n so the desired isomorphism LK(C 3 n and En satisfy the hypotheses of the Restricted Algebraic KP Theorem, and (cid:3) n) ∼= LK(En) follows. Remark 4.2. Although it is relatively easy to produce graphs Fn having three vertices for which K0(LK(Fn)) ∼= K0(LK(C 3 n)), we do not know how to produce such graphs for which [Fn] is the zero element in K0(LK(Fn)), which therefore precludes us from applying the Restricted Algebraic KP Theorem to the Leavitt path algebras of these graphs. Remark 4.3. Using the template afforded by the 4-vertex graph presented in Proposition 4.1, for each pair j, n with 0 ≤ j ≤ n − 1 one can easily construct a graph En(j) having j + 1 vertices, for which the Leavitt path algebras LK(En(j)) and LK(C j n) are isomorphic. A number of intriguing number-theoretic properties of the Haselgrove sequences and group- theoretic properties of the groups K0(LK(C j n)) arose in the context of the investigation presented in this article. For instance, as mentioned previously, the Hk(n) sequences can be shown to be di- visibility sequences. For another example, in the j = 3 case we may give a more explicit description of the integers d′ 3(n), as a product of a power of d3(n) with an "indicator factor". (In retrospect, we see that an analogous statement arose in the proof of the corresponding result in the j = 2 case carried out in [AA], but this indicator factor turned out not to play a role in the invariant factors representation of the abelian group K0(LK(C 2 n)).) However, such an "indicator factor" description does not extend to the cases j ≥ 4; for this reason we refer to the j = 3 case as a "sweet spot" in this setting. These and many additional properties will be presented in [AEG2]. acknowledgements The authors would like to thank G. Aranda Pino and M. Iovanov for fruitful discussions during the preparation of this paper. Some of these results were anticipated and suggested by looking at output from the software package Magma. The authors are grateful to A. Viruel for his valuable help with this software. LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS C j n 19 The first author was partially supported by a Simons Foundation Collaboration Grant #208941. The third author was partially supported by the Spanish MEC and Fondos FEDER through projects MTM2013-41208-P and MTM2016-76327-C3-1-P; by the Junta de Andaluc´ıa and Fondos FEDER, jointly, through project FQM-7156; and by the grant "Ayudas para la realizaci´on de estancias en centros de investigaci´on de calidad" of the "Plan Propio de Investigaci´on y Transferencia" of the University of M´alaga, Spain. Part of this work was carried out during a visit of the third author to the University of Colorado, Colorado Springs, USA. The third author thanks this host institution for its warm hospitality and support. References [AALP] G. Abrams, P.N. ´Anh, A. Louly, E. Pardo, The classification question for Leavitt path algebras, J. Algebra 320(5) (2008), 1983 -- 2026. [AAS] G. Abrams, P. Ara, M. Siles Molina, Leavitt path algebras, Lecture Notes in Mathematics Vol. 2191, Springer Verlag, London, 2017. [AAP2] G. Abrams, G. Aranda Pino, Purely infinite simple Leavitt path algebras, J. Pure App. Alg. 207(3) (2006), 553 -- 563. [AA] G. Abrams, G. Aranda Pino, Leavitt path algebras of generalized Cayley graphs, Mediterranean J. Math 13(1) (2016), 1 -- 27. [AEG2] G. Abrams, S. Erickson, C. Gil Canto, Number-theoretic properties of the Haselgrove sequences, in prepara- tion. [ALPS] G. Abrams, A. Louly, E. Pardo, C. Smith, Flow invariants in the classification of Leavitt path algebras, J. Algebra 333 (2011), 202 -- 231. [ASch] G. Abrams, B. Schoonmaker, Leavitt path algebras of Cayley graphs arising from cyclic groups, American Mathematical Society Contemporary Mathematics Series 634 (2015), 1 -- 10. [AGP] P. Ara, K. Goodearl, E. Pardo, K0 of purely infinite simple regular rings, K-Theory 26 (2002), 69 -- 100. [AMP] P. Ara, M.A. Moreno, E. Pardo, Non-stable K-theory for graph algebras, Algebr. Represent. Theor. 10(2) (2007), 157 -- 178. [H] C.B. Haselgrove, A note on Fermat's Last Theorem and the Mersenne Numbers, Eureka: The Archimedians' Journal 11 (1949), 19 -- 22. [N] M. Newman, Integral Matrices, Monographs and Textbooks in Pure and Applied Mathematics Vol. 45, Academic Press, 1972. [OEIS] Online Encyclopedia of Integer Sequences, http://oeis.org/ Department of Mathematics, University of Colorado, Colorado Springs, CO 80918, USA E-mail address: [email protected] Department of Mathematics, The Colorado College, Colorado Springs, CO 80903, USA E-mail address: [email protected] Departamento de ´Algebra, Geometr´ıa y Topolog´ıa, Universidad de M´alaga, 29071, M´alaga, Spain E-mail address: [email protected] / [email protected]
1302.0379	1	1302	2013-02-02T13:47:53	*-Regular Leavitt path algebras of arbitrary graphs	[ "math.RA" ]	If $K$ is a field with involution and $E$ an arbitrary graph, the involution from $K$ naturally induces an involution of the Leavitt path algebra $L_K(E).$ We show that the involution on $L_K(E)$ is proper if the involution on $K$ is positive definite, even in the case when the graph $E$ is not necessarily finite or row-finite. It has been shown that the Leavitt path algebra $L_K(E)$ is regular if and only if $E$ is acyclic. We give necessary and sufficient conditions for $L_{K}(E)$ to be $^\ast$-regular (i.e. regular with proper involution). This characterization of $^\ast$-regularity of a Leavitt path algebra is given in terms of an algebraic property of $K,$ not just a graph-theoretic property of $E.$ This differs from the known characterizations of various other algebraic properties of a Leavitt path algebra in terms of graph-theoretic properties of $E$ alone. As a corollary, we show that Handelman's conjecture (stating that every $^\ast$-regular ring is unit-regular) holds for Leavitt path algebras. Moreover, its generalized version for rings with local units also continues to hold for Leavitt path algebras over arbitrary graphs.	math.RA	math	∗-REGULAR LEAVITT PATH ALGEBRAS OF ARBITRARY GRAPHS G. ARANDA PINO, K. M. RANGASWAMY, AND L. VAS Abstract. If K is a field with involution and E an arbitrary graph, the involution from K naturally induces an involution of the Leavitt path algebra LK(E). We show that the involution on LK(E) is proper if the involution on K is positive definite, even in the case when the graph E is not necessarily finite or row-finite. It has been shown that the Leavitt path algebra LK(E) is regular if and only if E is acyclic. We give necessary and sufficient conditions for LK(E) to be ∗-regular (i.e. regular with proper involution). This characterization of ∗-regularity of a Leavitt path algebra is given in terms of an algebraic property of K, not just a graph-theoretic property of E. This differs from the known characterizations of various other algebraic properties of a Leavitt path algebra in terms of graph-theoretic properties of E alone. As a corollary, we show that Handelman's conjecture (stating that every ∗-regular ring is unit-regular) holds for Leavitt path algebras. Moreover, its generalized version for rings with local units also continues to hold for Leavitt path algebras over arbitrary graphs. Introduction Leavitt path algebras can be regarded as the algebraic counterparts of the graph C ∗- algebras, the descendants from the algebras investigated by J. Cuntz in [1]. Leavitt path algebras can also be viewed as a broad generalization of the algebras constructed by W. G. Leavitt in [2] to produce rings without the Invariant Basis Number property. The Leavitt path algebra LK(E) was introduced in the papers [3] and [4]. LK(E) was first defined for a row-finite graph E (countable graph such that every vertex emits only a finite number of edges) and a field K. Although their history is very recent, a flurry of activity has followed the papers [3] and [4]. The main directions of research include: characterization of algebraic properties of a Leavitt path algebra LK(E) in terms of graph-theoretic properties of E; study of the modules over LK(E); computation of various substructures (such as the Jacobson radical, the center, the socle and the singular ideal); investigation of the relationship and connections with C ∗(E) and general C ∗-algebras; classification programs; study of the K- theory; and generalization of the constructions and results first from row-finite to countable graphs and finally, from countable to completely arbitrary graphs. For examples of each of these directions see [3, 5, 6, 7, 8, 9, 10, 11]. The base field K is naturally endowed with an involution − (the identity involution can always be considered in the absence of other possibilities). A given involution on K naturally induces an involution of a Leavitt path algebra LK(E). The presence of an involution on a ring yields some favorable features: the ring is isomorphic to its opposite ring and a certain dose of symmetry is present. For example, a left Rickart ∗-ring is also a right Rickart ∗-ring while this is not the case for one-sided Rickart rings. Also, consideration of a complex Leavitt 2000 Mathematics Subject Classification. Primary: 16D70, 16W10, 16S99. Key words and phrases. Leavitt path algebra, ∗-regular, involution, arbitrary graph. 1 2 G. ARANDA PINO, K. M. RANGASWAMY, AND L. VAS path algebra LC(E) as a an algebra with involution (induced from the complex-conjugate involution on C), brings LC(E) a step closer to its analytic counterpart C ∗(E). These facts justify our interest in the study of the involution on a Leavitt path algebra. Iain Raeburn in [12] shows that the involution on LK(E) is proper if the involution on K is positive definite and E is a row-finite countable graph without sinks. We extend this result to arbitrary graphs (Proposition 2.3). We also show that the converse holds: if the induced involution on LK(E) is positive definite for every (equivalently some) graph E, then the involution on K is positive definite (Proposition 2.4). In [13], Gene Abrams and the second named author characterized the (von Neumann) reg- ular Leavitt path algebras LK(E) as precisely those with acyclic underlying graphs E. In light of our consideration of the involution on LK(E), we wonder when is LK(E) ∗-regular (regular with proper involution). In Theorem 3.3, we characterize ∗-regular Leavitt path algebras as exactly those with E acyclic and K that is proper up to a certain extent (determined by the least upper bound of the numbers of all paths that end at any given vertex of E). Note that we do not impose any conditions on the cardinality of E: we work with completely arbitrary graphs. Most existing characterization theorems for Leavitt path algebras have the following form: LK(E) has (ring-theoretic) property (P ) if and only if E has (graph-theoretic) property (P ′). Such theorems have been formulated and proven for a good number of ring-theoretic prop- erties. For example simple, purely infinite simple, exchange, semisimple, regular and other Leavitt path algebras have been characterized. It is interesting that the underlying field K did not play any role in those characterization theorems. Theorem 3.3, however, has a different form: LK(E) has property (P ) if and only if E has property (P ′) and K has property (P ′′). In other words, Theorem 3.3 is the first characterization theorem that involves a ring-theoretic property of the field K as well. Moreover, the characterization of LK(E) that is positive definite (Proposition 2.4) also has the above form that features the field K as well. The paper is organized as follows. In §1 we recall the basic definitions, examples and properties of Leavitt path algebras, whereas in §2 we focus on the involution of LK(E) and prove Propositions 2.3 and 2.4. We devote §3 to the proof of the characterization theorem for the ∗-regular Leavitt path algebras (Theorem 3.3). Finally, in §4 we consider [11, Problem 48, p. 380] that we shall refer to as "Handelman's conjecture". This conjecture is stating that every ∗-regular ring is unit-regular. We prove that Handelman's conjecture holds for Leavitt path algebras. In fact, we formulate a generalized version of the conjecture for rings with local units and show that it holds for Leavitt path algebras over arbitrary graphs. 1. Definitions and preliminaries We recall some graph-theoretic concepts, the definition and standard examples of Leavitt path algebras. A (directed ) graph E = (E0, E1, r, s) consists of two sets E0 and E1 (with no restriction on their cardinals) together with maps r, s : E1 → E0. The elements of E0 are called vertices and the elements of E1 edges. For e ∈ E1, the vertices s(e) and r(e) are called the source and range of e. If s−1(v) is a finite set for every v ∈ E0, then the graph is called row-finite. ∗-REGULAR LEAVITT PATH ALGEBRAS OF ARBITRARY GRAPHS 3 If E0 is finite and E is row-finite, then E1 must necessarily be finite as well; in this case we say simply that E is finite. A vertex which emits (receives) no edges is called a sink (source). A vertex v is called an infinite emitter if s−1(v) is an infinite set. A path µ in a graph E is a finite sequence of edges µ = e1 . . . en such that r(ei) = s(ei+1) for i = 1, . . . , n − 1. In this case, s(µ) = s(e1) and r(µ) = r(en) are the source and range of µ, respectively, and n is the length of µ. We view the elements of E0 as paths of length 0. If µ is a path in E, with v = s(µ) = r(µ) and s(ei) 6= s(ej) for every i 6= j, then µ is called a cycle. A graph which contains no cycles is called acyclic. Let K denote an arbitrary base field and E an arbitrary graph. The Leavitt path K-algebra LK(E) is the K-algebra generated by the set E0 ∪ E1 ∪ {e∗ e ∈ E1} with the following relations: (V) vw = δv,wv for all v, w ∈ E0. (P1) s(e)e = er(e) = e for all e ∈ E1. (P2) r(e)e∗ = e∗s(e) = e∗ for all e ∈ E1. (CK1) e∗f = δe,f r(e) for all e, f ∈ E1. (CK2) v =Pe∈s−1(v) ee∗ for every v ∈ E0 that is neither a sink nor an infinite emitter. The first three relations are the path algebra relations. The last two are the so-called Cuntz-Krieger relations. We let r(e∗) denote s(e), and we let s(e∗) denote r(e). If µ = e1 . . . en is a path in E, we 1 of LK(E). With this notation, the Leavitt path algebra write µ∗ for the element e∗ LK(E) can be viewed as a K-vector space span of {pq∗ p, q are paths in E}. (Recall that the elements of E0 are viewed as paths of length 0, so that this set includes elements of the form v with v ∈ E0.) n . . . e∗ If E is a finite graph, then LK(E) is unital with Pv∈E 0 v = 1LK (E); otherwise, LK(E) is a ring with a set of local units consisting of sums of distinct vertices of the graph. Many well-known algebras can be realized as the Leavitt path algebra of a graph. The most basic graph configurations are shown below (the isomorphisms for the first three can be found in [3], the fourth in [14], and the last one in [15]). Examples 1.1. The ring of Laurent polynomials K[x, x−1] is the Leavitt path algebra of the graph given by a single loop graph. Matrix algebras Mn(K) can be realized by the line graph with n vertices and n − 1 edges. Classical Leavitt algebras L(1, n) for n ≥ 2 can be obtained by the n-rose -- a graph with a single vertex and n loops. Namely, these three graphs are: • • / • / • • / • • e The algebraic counterpart of the Toeplitz algebra T is the Leavitt path algebra of the graph having one loop and one exit: •% / • Combinations of the previous examples are possible. For instance, the Leavitt path algebra of the graph • / • / • • / • e / / / e r r R R % / / / / e r r R R 4 G. ARANDA PINO, K. M. RANGASWAMY, AND L. VAS is Mn(L(1, m)), where n denotes the number of vertices in the graph and m denotes the number of loops. 2. The involution on a Leavitt path algebra We recall some standard definitions first. An involution ∗ on a ring R is an additive map ∗ : R → R that satisfies (ab)∗ = b∗a∗ and (a∗)∗ = a for all a, b ∈ R. For any a ∈ R, the element a∗ is called the adjoint of a. An element p in a ring with involution ∗ is called a projection if p is a self-adjoint (p∗ = p) idempotent (p2 = p). If there is an involution defined on a ring R, R is said to be a ∗-ring. If R is also an algebra over K with an involution −, then R is a ∗-algebra if (ax)∗ = ax∗ for a ∈ K, and x ∈ R. Let n be a positive integer. An involution ∗ on a ring R is said to be n-proper if 1x1 + · · · + x∗ x∗ nxn = 0 implies x1 = · · · = xn = 0 for any n elements x1, . . . , xn in R. A ring with an n-proper involution will be refered to as an n-proper ring. This property is clearly left-right symmetric, since each element xi = a∗ i for some ai ∈ R. 1-proper involution is simply said to be proper. The involution ∗ is said to be positive definite if it is n-proper for every positive integer n. A ∗-ring with a positive definite involution will be referred to as a positive definite ring. A field can have both an n-proper involution and an involution that is not n-proper. For example, consider the field C: it is 2-proper (in fact positive definite) for the conjugate involution (a + ib 7→ a − ib) and not 2-proper for the identity involution. Also, the same involution can be n-proper and not n + 1-proper (identity involution on C for n = 1). Before we turn to Leavitt path algebras, let us recall one last fact about general ∗-rings. Recall that if R is a ring with involution −, then the involution − induces the ∗-transpose involution on the ring Mn(R) of n × n matrices over R given by A = (aij) 7→ A∗ = (aji). We believe that the following lemma is well known but we are not aware of a reference for it. For completeness, we provide a proof here. Lemma 2.1. Let n be a positive integer and let R be a ring with involution −. Then the ∗-transpose involution on Mn(R) is proper if and only if the involution − in R is n-proper. Proof. Assume that the involution − is n-proper in R. Suppose A∗A = 0 for some matrix A = j=1 ajiaji = 0 for every i = 1, . . . , n. Since − is n-proper, aij = 0 for all i, j. Hence A = 0. (aij) ∈ Mn(R). Then the diagonal entries of the product A∗A are zero and soPn Conversely, suppose Pn i=1 aiai = 0 for ai ∈ R. Consider A to be the matrix of Mn(R) that has the elements a1, . . . , an in its first column and zeroes in the rest of its columns. Then A∗A = 0. Since the ∗-transpose involution is proper, A = 0. So ai = 0 for every i. (cid:3) We turn now to Leavitt path algebras. Let K be a field with involution − and let E be an arbitrary graph. Recall that a typical element of the Leavitt path algebra LK(E) can be written asPn the map ∗ given by i=1 kipiq∗ kipiq∗ i!∗ = n Xi=1 kiqip∗ i n Xi=1 ∗-REGULAR LEAVITT PATH ALGEBRAS OF ARBITRARY GRAPHS 5 i where pi and qi are paths and ki ∈ K. It is straightforward to see that defines the involution on LK(E) making it into a ∗-algebra. If K is the field of complex numbers C and we consider the conjugate involution, the ∗- algebra structure of LC(E) agrees with the ∗-algebra structure of LC(E) (that is used for instance to see LC(E) as a dense ∗-subalgebra of C ∗(E) as shown in [10, Theorem 7.3]). As we will see in the next section, the ∗-regularity of a Leavitt path algebra LK(E) is closely related to the condition stating that the involution of LK(E) is n-proper or positive definite. In this section, we characterize the positive definiteness of LK(E). The following proposition can be proved by easily adapting Iain Raeburn's result [12, Lemma 1.3.1] to our notation and context. Proposition 2.2. Let E be a row-finite, countable graph without sinks. If the involution − on K is positive definite, then the involution ∗ on LK(E) is proper. Proof. The proof follows completely [12, Lemma 1.3.1]. One only needs to take into account that the axioms in [12] are given so that a path e1e2 . . . en in [12] corresponds to the path enen−1 . . . e1 here (i.e. the edges in [12] are multiplied so that the action of f precedes the action of e in the product ef, contrary to the action we consider here). Because of this difference, the assumptions of [12, Lemma 1.3.1] that LK(E) is column-finite with no sources, correspond exactly to our assumptions that LK(E) is row-finite with no sinks. (cid:3) It is noted in [12] that the condition that E does not have sinks (sources, in the terminology of that paper) can be avoided by using the so-called Yeend's trick. This assumption can also be avoided using a technique called the Desingularization Process. The added benefit of the desingularization is that it can help us also get rid of the row-finiteness assumption. The desingularization of a graph E is a new graph F obtained by adding a tail (more details can be found in [16] or [17]) to every sink or infinite emitter. The resulting graph F is a row-finite graph without sinks such that LK(E) embeds in LK(F ) via the embedding that is a ∗-algebra homomorphism (see [17, Proposition 5.1] for more details). Finally, the last remaining assumption (that the graph is countable) in Proposition 2.2 can be avoided by means of the Subalgebra Construction (see [13] for more details). We recall here the relevant concept EF used in this construction. We shall use EF in our main theorem too. Let F be a finite set of edges in E. We define s(F ) (resp. r(F )) to be the sets of those vertices in E that appear as the source (resp. range) vertex of at least one element of F . The graph EF is then defined as follows (see [13, Definition 2]): E0 F = F ∪ (r(F ) ∩ s(F ) ∩ s(E1 \ F )) ∪ (r(F ) \ s(F )), E1 F = {(e, f ) ∈ F × E0 F r(e) = s(f )} ∪ [{(e, r(e)) e ∈ F with r(e) ∈ (r(F ) \ s(F ))}], and where s((x, y)) = x and r((x, y)) = y for any (x, y) ∈ E1 F . The graph EF is finite (see comment after [13, Definition 2]). Also, by [13, Proposition 1], there is an algebra homomorphism θ : LK(EF ) → LK(E). Further, [13, Proposition 2] 6 G. ARANDA PINO, K. M. RANGASWAMY, AND L. VAS containing S. The subalgebra B(S) is of the form LK(EF )L(Lm shows that for every finite set of elements S of LK(E), there is a subalgebra B(S) of LK(E) i=1 Kxi) where F is a finite set of edges defined using S (see [13, page 7]) and xi, i = 1, . . . , m is a finite set of vertices (defined in [13, page 8]). By [13, Proposition 2], LK(E) is a directed union of subalgebras B(S), where the S varies over all finite subsets of LK(E). Furthermore, and key to our current discussion, θ preserves the involution by the construction (as it can be seen from the proof of [13, Proposition 1]) so it is a ∗-algebra homomorphism. Proposition 2.3. Let E be an arbitrary graph. If the involution − on K is positive definite, then the involution ∗ on LK(E) is proper. Proof. We prove the claim first for the case when E is a countable. Let F be a desingulariza- tion of E. By the Desingularization Process, F is a row-finite graph without sinks and there is a ∗-algebra monomorphism φ : LK(E) → LK(F ). Now, suppose that a∗a = 0 in LK(E). Apply φ to get that φ(a)∗φ(a) = 0 in LK(F ). Since F is row-finite and does not contain sinks, Proposition 2.2 can be applied and so φ(a) = 0. Then a = 0 since φ is a monomorphism. Now suppose that E is arbitrary and let a ∈ LK(E) be such that a∗a = 0. By the Subalgebra Construction, for the finite set S = {a, a∗}, there is a finite set of edges F and a finite number of vertices xi, i = 1, . . . , m such that the subalgebra B(S) of LK(E) is of i=1 Kxi is a direct summand in the previous equation for B(S), we can actually attach a finite number of isolated vertices v1, . . . , vm 6∈ E0 the form LK(EF )L(Lm F to the graph EF so that we obtain a new finite graph G such that i=1 Kxi) and a, a∗ ∈ B(S). Since Lm Kvi! LK(G) ∼= LK(EF )M m Mi=1 via a ∗-algebra isomorphism. Since B(S) is a subalgebra of LK(E), the equation a∗a = 0 holds in B(S) ∼= LK(G). Apply the previous case to LK(G) in order to deal with possible sinks in G. Then we have that a = 0. This finishes the proof. (cid:3) The last result of this section is a characterization of Leavitt path algebras that have positive definite involutions in terms of the corresponding property in the field K. Proposition 2.4. Let K a field with involution. The following conditions are equivalent. (i) The involution on K is positive definite. (ii) The involution on LK(E) is positive definite for every graph E. (iii) The involution on LK(E) is positive definite for some graph E. Thus, if E is an arbitrary graph, LK(E) is positive definite if and only if K is positive definite. Proof. (i) =⇒ (ii). Given E, let us consider the graph MnE obtained from E by attaching a line of length n − 1 to every vertex of E so that each line ends at the given vertex of the graph (more details in [7]). The graph MnE has the property that Mn(LK(E)) is isomorphic to LK(MnE) as ∗-algebras by [7, Proposition 9.3]. By Proposition 2.3, we know that LK(MnE) is ∗-proper. So, Mn(LK(E)) is ∗-proper. Now apply Lemma 2.1 to get that LK(E) is n-proper. (ii) =⇒ (iii) is a tautology. ∗-REGULAR LEAVITT PATH ALGEBRAS OF ARBITRARY GRAPHS 7 (iii) =⇒ (i). Suppose that Pn 0 = (Pn i=1 kiki = 0 for ki ∈ K. Let E be a graph such that the involution on LK(E) is positive definite. Let v ∈ E0. Since v is a projection then i=1(kiv)∗(kiv) and therefore kiv = 0 for all i by hypothesis. But E0 is a set of linearly independent elements in LK(E) by [11, Lemma 1.5], so that ki = 0 for all i, as needed. (cid:3) i=1 kiki)v = Pn 3. ∗-regular Leavitt path algebras The (von Neumann) regular Leavitt path algebras LK(E) were characterized in [13] as those whose graphs E are acyclic. In light of the consideration of LK(E) as a ring with involution, we wonder which acyclic graphs have LK(E) that is ∗-regular. We provide an answer to this question in this section. Recall that a ring R is (von Neumann) regular if for every a ∈ R there exists b ∈ R such that aba = a, or equivalently [18, Theorem 4.23], every right (resp. left) principal ideal is generated by an idempotent. This statement continues to hold if R is a ring with local units since b ∈ bR (and b ∈ Rb) for all b in R so the principal right (and left) ideals of R have the same form as the principal right (left) ideals of a unital ring. If R is a ∗-ring, the projections take over the role of idempotents. Thus, the concept of regularity for rings corresponds to ∗-regularity for ∗-rings: a ∗-ring R is said to be ∗-regular if every principal right ideal is generated by a projection. This definition naturally extends to rings with local units. Note that the condition of ∗-regularity is left-right symmetric since aR = pR implies that Ra∗ = Rp for any a ∈ R and a projection p ∈ R. So every principal left ideal of R is also generated by a projection in the case when every principal right ideal is. A ∗-ring is ∗-regular if and only if it is regular and the involution ∗ is proper (see [19, Exercise 6A, §3]). In the next proposition we give a proof of this fact for rings with local units. Proposition 3.1. Let R be a ring with local units and with an involution ∗. Then R is ∗-regular if and only if R is regular and ∗ is proper. Proof. If R is ∗-regular then it is also regular because every projection is an idempotent. Now assume that a∗a = 0 for some a ∈ R. Then aR = pR for some projection p so a = pa (a = px for some x ∈ R implies that pa = ppx = px = a) and p = ay for some y ∈ R. So, a∗ = a∗p = a∗ay = 0. Hence a = 0. Thus ∗ is proper. Conversely, suppose that R is regular and ∗ is proper. Since every principal right ideal is generated by an idempotent, it is enough to show that for an arbitrary idempotent x in R, Rx∗ = Rp for some projection p ∈ R. First observe that for any x ∈ R, rR(x) = rR(x∗x) where rR(b) denotes the right annihilator of the element b ∈ R. This is because, x∗xy = 0 implies that (xy)∗xy = y ∗(x∗xy) = 0 so that xy = 0 for any y ∈ R. By the regularity of R, Rx∗x = Rf for some idempotent f ∈ R. Thus rR(x) = rR(x∗x) = rR(f ) and so the left annihilators lR(rR(x)) and lR(rR(f )) are also equal. We claim that Rx = lR(rR(x)). To see this, first note that, since x is an idempotent, rR(x) = {a − xa a ∈ R}. So if y ∈ lR(rR(x)), then y(a − xa) = 0 for all a ∈ R, that is ya = yxa for all a ∈ R. Since R is a ring with local units, there is an idempotent u ∈ R such that yu = y and xu = x. Hence y = yu = yxu = yx ∈ Rx. Thus lR(rR(x)) ⊆ Rx. Since the reverse inclusion is obvious, Rx = lR(rR(x)). 8 G. ARANDA PINO, K. M. RANGASWAMY, AND L. VAS Similarly, Rf = lR(rR(f )). Thus Rx = Rf = Rx∗x. Hence x = ax∗x for some a ∈ R. Let p = ax∗. We claim that p is a projection with Rx∗ = Rp. To see this, note that x = px and so pp∗ = pxa∗ = xa∗ = p∗. Since (pp∗)∗ = pp∗, we get p = p∗. From pp∗ = p∗ and p = p∗ we have p2 = p and so p is a projection. From p = ax∗ and x∗ = x∗p we have that Rx∗ = Rp. (cid:3) We turn to Leavitt path algebras now. For any vertex v in a graph E, let µ(v) denote the cardinality of the set of all the paths α in E with r(α) = v (including the trivial path v). With this notation, we recall the statement of [20, Lemma 3.4 and Proposition 3.5]. Let E be a finite acyclic graph and v a sink in E. The set Iv =(Xi kiαiβ ∗ i αi, βi paths in E with r(αi) = r(βi) = v, ki ∈ K) is an ideal of LK(E) isomorphic to the matrix ring Mµ(v)(K). If {v1, . . . , vm} is the set of all Mµ(vi)(K). Let us denote this isomorphism by φ From [20, Lemma 3.4 and Proposition 3.5] it can be seen that the restriction of φ on a j ) = (kij) ∈ Mµ(v)(K) where i=1 Ivi ∼=Lm i=1 and let us call it a canonical isomorphism. sinks in E, then LK(E) =Lm direct summand Iv, for a vertex v, is given by φ(Pi,j kijαiα∗ canonical isomorphism φ : LK(E) = Lm i, j = 1, . . . , µ(v), αi and αj are paths ending at v and kij ∈ K. i=1 Ivi → Lm Lemma 3.2. Let E be a finite acyclic graph and let {v1, . . . , vm} be all the sinks in E. The Mµ(vi)(K) is a ∗-algebra isomorphism (with the standard involution on LK(E) and the ∗-transpose involution on the matrix algebras). i=1 Proof. Since φ maps direct summands Ivi on direct summands Mµ(vi)(K), it is enough if we prove the statement when E has only one sink v. If α1, . . . , αµ(v) are all the different paths (including the trivial path) ending in v, then a typical element of LK(E) = Iv has the form j for i, j = 1, . . . , µ(v), αi and αj paths ending at v and kij ∈ K. Then we have Pi,j kijαiα∗ kijαiα∗ φ Xi,j j!∗! = φ Xi,j i! = φ Xi,j j(cid:17)(cid:17)∗ This proves the claim since (kij)∗ =(cid:16)φ(cid:16)Pi,j kijαiα∗ kijαjα∗ . kjiαiα∗ j! = (kji) = (kij)∗. (cid:3) We finally have all the ingredients in hand to prove the main result of the paper. Theorem 3.3. Let E be an arbitrary graph, K be a field with involution − and let σ = sup{µ(v) : v ∈ E0} in case the supremum is finite or σ = ω otherwise. The following conditions are equivalent. (i) LK(E) is ∗-regular. (ii) LK(E) is regular and proper. (iii) E is acyclic and K is n-proper for every finite n ≤ σ. Proof. (i) ⇔ (ii) is Proposition 3.1. (ii) ⇔ (iii). By [13, Theorem 1], LK(E) is regular if and only if E is acyclic. So it is enough if we show, under the assumption that LK(E) is regular (equivalently, E is acyclic), that the involution − in K is n-proper for every finite n ≤ σ if and only if the involution ∗ in LK(E) is proper. ∗-REGULAR LEAVITT PATH ALGEBRAS OF ARBITRARY GRAPHS 9 Now [13, Proposition 2 and Theorem 1] also state that, when E is acyclic, LK(E) is a i=1 Kxi) with EF a directed union of subalgebras B(S) where each B(S) finite acyclic graph constructed corresponding to various non-empty finite subsets F of edges in E. Moreover θ is a ∗-algebra isomorphism as we noted before. For a fixed F , EF has a MµEF (vi)(K) finite number of sinks. Let us denote them by v1, . . . , vk. Then LK(EF ) as ∗-algebras by Lemma 3.2. Thus, the involution ∗ in LK(E) is proper if and only if the ∗-transpose involution is proper in MµEF (vi)(K) with vi ∈ EF for all the various graphs EF corresponding to each B(S) in the stated directed system of subalgebras of LK(E). i=1 θ∼= LK(EF )L(Lm ∼=Lk φ We distinguish two situations. Case 1: Suppose σ is infinite. Then either µ(v) is infinite for some vertex v or for every positive integer n there is a vertex vn with µE(vn) an integer larger than n. In either case for each integer n > 1, we can choose a vertex vn and a finite subset Fn of edges that appear in the n − 1 distinct paths (other than the trivial path vn) ending in vn. The vertex vn is a sink in EFn by the definition of the graph EFn. Moreover, e1 . . . ek is a path of length k in E ending in vn if and only if the path in EFn given by (e1, e2)(e2, e3) . . . (ek, vn) has length k and ends in vn. Thus, vn is a sink in EFn with µEFn (vn) = n. The graph EFn is finite acyclic (by [13, Lemma 1]). So, LK(EFn) contains the ideal Ivn ∼= Mn(K). Since this holds for every n, the involution ∗ is proper in each subalgebra B(S) if and only if the ∗-transpose involution is proper in Mn(K) for each positive integer n. This is equivalent, by Lemma 2.1, to the statement that the involution − in K is n-proper for every n, that is, that K is positive definite. Case 2: Suppose that σ is finite, say σ = n for some positive integer n. If n = 1, every algebras are proper so we are done. vertex in E is isolated and LK(E) is isomorphic to Lv∈E 0 Kv where Kv ∼= K. Both of those Suppose n > 1 and let v be a vertex for which µ(v) = n. Let Fv be the non-empty finite set of edges in all the µE(v) − 1 nontrivial paths ending in v. As noted in Case 1, v is a sink (and in this case, the only sink) in the finite acyclic graph EFv and, moreover, µEFv (v) = µE(v). So by Lemma 3.2, we have that LK(EFv) ∼= MµE (v)(K) as ∗-algebras. Moreover, as n is the least upper bound of {µ(v) : v ∈ E0} then all matrices MµEFi (vi)(K) appearing in all other subalgebras B(S) for various other finite subsets of edges Fi and sinks vi will all have order that is less or equal to n. Therefore the involution ∗ is proper in each B(S) if and only if the ∗-transpose involution is proper in MµE (v)(K). Since µ(v) = n, the last statement holds exactly when the involution − in K is n-proper, again by Lemma 2.1. This finishes the proof. (cid:3) It is interesting to point out that the presence of involution gives a more prominent role to the field K than it had in the previous characterization theorems (e.g., simplicity [3], purely infinite simplicity [15], finite-dimensionality [20], just to cite a few). In particular, Theorem 3.3 also contrasts the characterization of regularity from [13] that was independent of the field K. We further illustrate this behavior with an easy example. If E is the graph • / • / 10 G. ARANDA PINO, K. M. RANGASWAMY, AND L. VAS then LK(E) ∼= M2(K) as ∗-algebras for any field K. If K = R with the identity involution, LR(E) is ∗-regular because R is positive definite. However if K = C with the identity involution, then LC(E) ∼= M2(C) is regular but it is not ∗-regular (since the identity involution in C is not 2-proper). Furthermore, if K = C with the conjugate involution, then LC(E) is ∗-regular, because the conjugation of complex numbers is positive definite. Also, since the identity involution on a field of characteristic n > 0 is not n-proper, the properness (thus also ∗-regularity) of a Leavitt path algebra over such field depends on the characteristic of the field. This fact also brings the field characteristic into spotlight. Let us note the following corollary of Theorem 3.3. Corollary 3.4. Let K a field with involution. The following conditions are equivalent. (i) The involution on K is positive definite. (ii) LK(E) is ∗-regular for every acyclic graph E. Proof. (i) =⇒ (ii) follows directly from Theorem 3.3. (ii) =⇒ (i) Let us assume that LK(E) is ∗-regular for every acyclic E. Consider the line of length n − 1 (see second graph in Examples 1.1). The Leavitt path algebra of this graph is isomorphic to Mn(K). From the assumption that this algebra is ∗-regular, we obtain that K is n-proper by Lemma 2.1. Since this holds for every n, K is positive definite. (cid:3) It is also interesting to note that the two equivalences of Corollary 3.4 parallel the first two equivalences of Proposition 2.4. The last equivalence of Proposition 2.4 in the ∗-setting would have the form: " LK(E) is ∗-regular for some acyclic graph E". However, this statement is weaker than the other two equivalences in Corollary 3.4 so we do not have complete analogy with Proposition 2.4. To see this, consider a graph consisting of a single vertex and the complex numbers with the identity involution. The Leavitt path algebra of this graph is ∗-regular but the field is not positive definite. 4. Handelman's Conjecture for Leavitt path algebras We close this paper by pointing out that Handelman's conjecture has a positive answer for the family of Leavitt path algebras of arbitrary graphs. The conjecture can be stated as follows. Conjecture 4.1. (Handelman, [21, Problem 48, p. 380]). Every ∗-regular ring is unit-regular. This conjecture assumes that the ring is unital. First, we note that it is true for unital Leavitt path algebras. Let us assume that a unital LK(E) is ∗-regular. Then E is acyclic by Theorem 3.3. Then we have that LK(E) is unit-regular by [13, Theorem 2]. To prove that the conjecture remains true for Leavitt path algebras of arbitrary graphs, we adapt the notion of unit-regularity for rings with local units, as was done in [13] for instance. Recall that a ring R with identity is said to be unit-regular if for each a ∈ R, there is a unit (an invertible element) u such that aua = a. If R is a ring with local units, then R is called locally unit-regular if for each a ∈ R there is an idempotent (a local unit) v and local inverses u, u′ such that uu′ = v = u′u, va = av = a and aua = a. Clearly, a unit-regular (unital) ring is locally unit-regular (take the idempotent v from the definition of locally unit-regular to be the identity). Conversely, if a ring with identity is locally unit-regular, then it is unit-regular (see also [13, Lemma 3 (1)]). To see this, let ∗-REGULAR LEAVITT PATH ALGEBRAS OF ARBITRARY GRAPHS 11 a ∈ R. Then there is an idempotent v and local inverses u, u′ in vRv such that uu′ = v = u′u, va = av = a and aua = a. Then w = u + (1 − v) and w′ = u′ + (1 − v) satisfy ww′ = 1 = w′w and a = awa. Hence R is unit-regular. Corollary 4.2. Let E be an arbitrary graph and let K be a field with involution. Suppose LK(E) is ∗-regular. Then (i) LK(E) is locally unit-regular. (ii) If LK(E) is a unital ring, then LK(E) is unit-regular. Proof. (i). If LK(E) is ∗-regular, we have that E is acyclic by Theorem 3.3. Then LK(E) is locally unit-regular by [13, Theorem 2]. (ii) is a consequence of the fact that every unital locally unit-regular ring is unit-regular. (cid:3) Let us also note that the converse of Handelman's Conjecture is not true. The examples of unit-regular and not ∗-regular rings can be found in the class of Leavitt path algebras as well. For instance, M2(C) with the identity involution on C is such an example: we know it is not ∗-regular but it is unit-regular (as a semisimple ring, see [21, page 38]). Acknowledgments The authors thank Gene Abrams for his valuable discussions during the preparation of this paper. The first author was partially supported by the Spanish MEC and Fondos FEDER through project MTM2007-60333, and by the Junta de Andaluc´ıa and Fondos FEDER, jointly, through projects FQM-336 and FQM-2467. References [1] J. Cuntz, Simple C ∗-algebras generated by isometries, Comm. Math. Phys. 57 (1977), 173 -- 185. [2] W. G. Leavitt, Modules without invariant basis number, Proc. Amer. Math. Soc. 8 (1957), 322 -- 328. [3] G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2) (2005), 319 -- 334. [4] P. Ara, M.A. Moreno, E. Pardo, Nonstable K-Theory for graph algebras, Algebr. Represent. Theory 10 (2007), 157 -- 178. [5] G. Aranda Pino, E. Pardo, M. Siles Molina, Exchange Leavitt path algebras and stable rank, J. Algebra 305 (2) (2006), 912 -- 936. [6] G. Aranda Pino, D. Mart´ın Barquero, C. Mart´ın Gonz´alez, M. Siles Molina, The socle of a Leavitt path algebra, J. Pure Appl. Algebra 212 (3) (2008), 500 -- 509. [7] G. Abrams, M. Tomforde, Isomorphism and Morita equivalence of graph algebras, Trans. Amer. Math. Soc. 363 (2011), 3733 -- 3767. [8] G. Abrams, P. N. ´Anh, A. Louly, E. Pardo, The classification question for Leavitt algebras, J. Algebra 320 (2008), 1983 -- 2026. [9] P. Ara, M. Brustenga, G. Cortinas, K-theory for Leavitt path algebras, Munster J. of Math 2(2009), 5 -- 33. [10] M. Tomforde, Uniqueness theorems and ideal structure for Leavitt path algebras, J. Algebra 318 (1) (2007), 270 -- 299. [11] K. R. Goodearl, Leavitt path algebras and direct limits, Contemp. Math. 480 (2009), 165 -- 187. [12] I. Raeburn, Chapter in Graph algebras: bridging the gap between analysis and algebra (G. Aranda Pino, F. Perera, M. Siles Molina, eds.), ISBN: 978-84-9747-177-0, University of M´alaga Press, M´alaga, Spain (2007). [13] G. Abrams, K. Rangaswamy, Regularity conditions for arbitrary Leavitt path algebras, Algebr. Rep- resent. Theory 13 (3) (2010), 319 -- 334. 12 G. ARANDA PINO, K. M. RANGASWAMY, AND L. VAS [14] M. Siles Molina, Algebras of quotients of Leavitt path algebras. J. Algebra 319 (12) (2008), 5265 -- 5278. [15] G. Abrams, G. Aranda Pino, Purely infinite simple Leavitt path algebras, J. Pure Appl. Algebra 207 (3) (2006), 553 -- 563. [16] D. Drinen, M. Tomforde, The C ∗-algebras of arbitrary graphs, Rocky Mountain J. Math 35 (1) (2005), 105 -- 135. [17] G. Abrams, G. Aranda Pino, The Leavitt path algebras of arbitrary graphs, Houston J. Math. 34 (2) (2008), 423 -- 442. [18] T.Y. Lam, A First Course on Noncommutative Rings, Springer-Verlag New York (1991). [19] S. K. Berberian, Baer ∗-rings, Die Grundlehren der mathematischen Wissenschaften 195, Springer- Verlag, Berlin-Heidelberg-New York (1972). [20] G. Abrams, G. Aranda Pino, M. Siles Molina, Finite-dimensional Leavitt path algebras, J. Pure Appl. Algebra. 209 (3) (2007), 753 -- 762. [21] K. R. Goodearl, Von Neumann Regular Rings, Second Ed., Krieger, Malabar, FL (1991). Aranda Pino: Departamento de ´Algebra, Geometr´ıa y Topolog´ıa, Universidad de M´alaga, 29071 M´alaga, Spain E-mail address: [email protected] Rangaswamy: Department of Mathematics, University of Colorado, Colorado Springs, CO 80933, USA E-mail address: [email protected] Vas: Department of Mathematics, Physics and Statistics, University of the Sciences in Philadelphia, Philadelphia, PA 19104, USA E-mail address: [email protected]
0911.3696	2	0911	2010-05-24T05:13:27	Hochschild cohomology of group extensions of quantum symmetric algebras	[ "math.RA", "math.QA" ]	Quantum symmetric algebras (or noncommutative polynomial rings) arise in many places in mathematics. In this article we find the multiplicative structure of their Hochschild cohomology when the coefficients are in an arbitrary bimodule algebra. When this bimodule algebra is a finite group extension (under a diagonal action) of a quantum symmetric algebra, we give explicitly the graded vector space structure. This yields a complete description of the Hochschild cohomology ring of the corresponding skew group algebra.	math.RA	math	HOCHSCHILD COHOMOLOGY OF GROUP EXTENSIONS OF QUANTUM SYMMETRIC ALGEBRAS DEEPAK NAIDU, PIYUSH SHROFF, AND SARAH WITHERSPOON Abstract. Quantum symmetric algebras (or noncommutative polynomial rings) arise in many places in mathematics. In this article we find the multiplicative structure of their Hochschild cohomology when the coefficients are in an arbitrary bimodule algebra. When this bimodule algebra is a finite group extension (under a diagonal action) of a quantum symmetric algebra, we give explicitly the graded vector space structure. This yields a complete description of the Hochschild cohomology ring of the corresponding skew group algebra. 0 1 0 2 y a M 4 2 ] . A R h t a m [ 2 v 6 9 6 3 . 1 1 9 0 : v i X r a 1. Introduction The action of a group G on a vector space V induces an action on the corresponding sym- metric algebra S(V ) (a polynomial ring). The resulting skew group algebra S(V ) ⋊ G is a noncommutative ring encoding the group action. Deformations of such skew group algebras are related to deformations of corresponding orbifolds, and some have appeared indepen- dently in several places under different names: graded Hecke algebras, rational Cherednik algebras, and symplectic reflection algebras. Deformations of any algebra are intimately re- lated to its Hochschild cohomology. When G is finite the Hochschild cohomology of S(V )⋊G was computed independently by Farinati [F] and by Ginzburg and Kaledin [GK]. Its algebra structure was first given by Anno [A]. In this paper we replace the symmetric algebra S(V ) with a quantum symmetric algebra and explore its Hochschild cohomology. Our quantum symmetric algebra is a noncommu- tative polynomial ring, denoted Sq(V ), in which the variables commute only up to multi- plication by nonzero scalars (encoded in the vector q). Noncommutative polynomials have been of interest for some time. Our current work incorporates group actions as well and is in part motivated by the recent appearance of two articles: Kirkman, Kuzmanovich, and Zhang [KKZ] prove a version of the classical Shephard-Todd-Chevalley Theorem, namely that the invariant subring of Sq(V ) under a finite group action is again a quantum symmet- ric algebra. In the setting of an ordinary polynomial ring, methods from invariant theory for finding such subrings play a crucial role in computations of Hochschild cohomology (see, for example, [SW]). Around the same time, Bazlov and Berenstein [BB] introduced braided Cherednik algebras, which are deformations of Sq(V ) ⋊ G in some special cases. Knowledge of Hochschild cohomology will provide insight into these and other possible deformations. More specifically, let k be a field of characteristic 0. Let N be a positive integer and for each pair i, j of elements in {1, . . . , N}, let qi,j be a nonzero scalar such that qi,i = 1 and Date: May 14, 2010. 2010 Mathematics Subject Classification. 16E40, 16S35. The second and third authors were partially supported by NSF grant #DMS-0800832 and Advanced Research Program Grant 010366-0046-2007 from the Texas Higher Education Coordinating Board. 1 qj,i = q−1 V be a vector space with basis x1, . . . , xN , and let i,j for all i, j. Denote by q the corresponding tuple of scalars, q := (qi,j)1≤i<j≤N . Let Sq(V ) := khx1, . . . , xN xixj = qi,jxjxi for all 1 ≤ i, j ≤ Ni, the quantum symmetric algebra determined by q. This is a Koszul algebra (see e.g. [AS]), so there is a standard complex K q(Sq(V )) that is a resolution of Sq(V ) as an Sq(V )-bimodule. This complex is given in Section 2.1 below; see [W] for details for more general quantum symmetric algebras arising from braidings. Priddy [P] first introduced Koszul algebras, and the theory was developed further, including such complexes, in [BGS, BG, M]. theorem. Here Vq We first compute cup products, using the resolution in Section 2.1, and obtain the following −1(V ∗) denotes a quantum exterior algebra, defined in (9) below, on the dual vector space V ∗, and B is any algebra with a compatible structure of an Sq(V )- bimodule. The two choices of algebra B in which we are most interested are B = Sq(V ) and B = Sq(V ) ⋊ G, where G is a finite group of graded automorphisms of Sq(V ). Theorem 3.1. The Hochschild cohomology HH q(Sq(V ), B) is a subquotient algebra of the −1(V ∗). tensor product B ⊗Vq That is, the Hochschild cohomology is a vector subquotient of B ⊗Vq product is determined by that of the tensor product of the two algebras B and Vq We next give the graded vector space structure of Hochschild cohomology in the two cases B = Sq(V ) and B = Sq(V ) ⋊ G, when G acts diagonally on the basis x1, . . . , xN of V . We adapt techniques developed in the context of Hochschild homology by Wambst [W]. In the first case this gives the following technical result. The notation will be explained in Section 4. −1(V ∗), and its cup −1(V ∗). Corollary 4.3. For all m ∈ N, HHm(Sq(V )) ∼= Mβ∈{0,1}N β=m Mα∈NN α−β∈C spank{xα ⊗ (x∗)∧β}. This result is a consequence of our more general Theorem 4.1 on Sq(V ) ⋊ G. It should be compared with work of Richard [R], in which there are some results on the Hochschild coho- mology of a related ring of twisted differential operators on quantum affine space. Richard obtains his results by first computing Hochschild homology and then invoking a duality be- tween homology and cohomology. In our setting, such a duality does not hold in general; a comparison of our Example 4.4 below with Wambst's Corollary 6.2 [W] shows that there is no duality in our smallest possible case. Other authors have used various techniques to compute Hochschild homology of generalizations of Sq(V ) [GG, Gu, W]. Since the characteristic of k is 0, the Hochschild cohomology of Sq(V )⋊G is the subalgebra of G-invariant elements of HH q(Sq(V ), Sq(V ) ⋊ G), and the multiplicative structure of this algebra is given by Theorem 3.1. To determine the additive structure precisely, we specialize to diagonal actions of G on the chosen basis x1, . . . , xN of V . This is the case to which Wambst's techniques in [W] may be most easily adapted to express the relevant cochain complex as the direct sum of an acyclic complex and a complex in which all differentials are 0. This leads to our description of the Hochschild cohomology as a graded vector space in the following theorem. The notation will be explained in Section 4. 2 Theorem 4.1. Assume the finite group G acts diagonally on the chosen basis of V . Then for all m ∈ N, HHm(Sq(V ), Sq(V ) ⋊ G) ∼=Mg∈G Mβ∈{0,1}N β=m Mα∈NN α−β∈Cg spank{(xα#g) ⊗ (x∗)∧β}, and HHm(Sq(V ) ⋊ G) is its G-invariant subspace. We plan to address more general group actions, as well as related deformations of the skew group algebra Sq(V ) ⋊ G, in future articles. Organization. This paper is organized as follows. Section 2 contains necessary preliminary information, including the resolution of Sq(V ) that will be used. We also give a chain map from this resolution to the bar resolution of Sq(V ), used in Section 3 to compute cup products. We prove Theorem 3.1 in Section 3. In Section 4 we prove Theorem 4.1 and apply Theo- rem 3.1 to give the cup product on HH q(Sq(V ) ⋊ G). As a special case, when G = {1}, we obtain in Corollary 4.3 the Hochschild cohomology of Sq(V ). 2. Preliminaries All tensor products and exterior powers are taken over the field k of characteristic 0. Let A be an algebra over k, and let M be an A-bimodule. We identify M with a (left) Ae-module, where Ae = A ⊗ Aop; Aop is the algebra A with the opposite multiplication. The Hochschild cohomology of A with coefficients in M is HH q(A, M) := Ext q Ae(A, M), where A is considered to be an Ae-module under left and right multiplication. One useful free Ae-resolution of A is the bar resolution (1) · · · δ3−→ A⊗4 δ2−→ A⊗3 δ1−→ Ae mult−−→ A −→ 0 where δm(a0 ⊗ · · · ⊗ am+1) =Pm i=0(−1)ia0 ⊗ · · · ⊗ aiai+1 ⊗ · · · ⊗ am+1 for all a0, . . . , am+1 ∈ A, and the map from Ae to A is given by multiplication in A. Suppose M = B is an A- bimodule algebra, that is B is an algebra and also an A-bimodule and these two structures are compatible in the sense that a(bb′) = (ab)b′ and (bb′)a = b(b′a) for all a ∈ A and b, b′ ∈ B. Then HH q(A, B) has a cup product defined at the cochain level as follows (e.g. see [G, p. 278]). Let f ∈ HomAe(A⊗(m+2), B), f ′ ∈ HomAe(A(n+2), B). Then f ⌣ f ′ ∈ HomAe(A⊗(m+n+2), B) is determined by f ⌣ f ′(a0⊗a1⊗· · ·⊗am+n⊗am+n+1) = f (a0⊗a1⊗· · ·⊗am⊗1)f ′(1⊗am+1⊗· · ·⊗am+n⊗am+n+1). Let G be a finite group acting on the algebra A by automorphisms. We denote by ga the result of applying g ∈ G to a ∈ A. Then we may form the skew group algebra A ⋊ G: Additively, it is the free A-module with basis G. We write A ⋊ G = ⊕g∈GAg, where Ag = {a#g a ∈ A}, that is for each a ∈ A and g ∈ G we denote by a#g ∈ Ag the a-multiple of g. Multiplication on A ⋊ G is determined by (a#g)(b#h) := a(gb)#gh 3 for all a, b ∈ A, g, h ∈ G. Note that for each g ∈ G, Ag is a (left) Ae-module via the action (a ⊗ b) · (c#g) := (a#1)(c#g)(b#1) = ac(gb)#g for all a, b, c ∈ A, g ∈ G. For convenience in what follows, we will sometimes denote the quantum symmetric algebra Sq(V ) simply by A. 2.1. A free resolution of Sq(V ). By [W, Proposition 4.1(c)], the following is a free Ae- resolution of A = Sq(V ): (2) · · · −→ Ae ⊗V2(V ) d2−→ Ae ⊗V1(V ) d1−→ Ae mult−−→ A −→ 0, that is, for 1 ≤ m ≤ N, the degree m term is Ae ⊗Vm(V ), and dm is defined by dm(1⊗2 ⊗ xj1 ∧ · · · ∧ xjm) qji,js! ⊗ xji# ⊗ xj1 ∧ · · · ∧ xji ∧ · · · ∧ xjm = m Xi=1 (−1)i+1" i Ys=1 qjs,ji! xji ⊗ 1 − m Ys=i whenever 1 ≤ j1 < . . . < jm ≤ N. This is a twisted version of the usual Koszul resolution for a polynomial ring. 1 xα2 2 · · · xαN define xα := xα1 Let us write the above formula for dm in a more convenient form. We first introduce some notation following Wambst [W]. Let NN denote the set of all N-tuples of elements from i=1 αi. For all α ∈ NN , N . For all i ∈ {1, . . . , N}, define [i] ∈ NN by [i]j = δi,j, for all N. For any α ∈ NN , the length of α, denoted α, is the sum PN j ∈ {1, . . . , N}. For any β ∈ {0, 1}N , let x∧β denote the vector xj1 ∧ · · · ∧ xjm ∈ Vm(V ) i,s! ⊗ xi# ⊗ x∧(β−[i]). which is defined by m = β, βjk = 1 for all k ∈ {1, . . . , m}, and j1 < . . . < jm. Then, for any β ∈ {0, 1}N with β = m we have dm(1⊗2 ⊗ x∧β) = δβi,1(−1)Pi−1 qβs s=1 βs" i Ys=1 qβs s,i! xi ⊗ 1 − N Ys=i Xi=1 N 2.2. A chain map into the bar resolution of Sq(V ). We wish to define a chain map from our complex Ae ⊗V q(V ) to the bar resolution (1) for A = Sq(V ). Wambst defined a more general chain map [W, Lemma 5.3 and Theorem 5.4]. Here we introduce notation useful in our setting, and include some details for completeness. For each set of m distinct natural numbers j1, . . . , jm (m ≤ N) and each permutation π ∈ Sm, the scalar qj1,...,jm π is defined by the equation qj1,...,jm π xjπ(1) · · · xjπ(m) = xj1 · · · xjm in A, that is, qj1,...,jm in the product xj1 · · · xjm, using the relations in A to rewrite it. is the scalar arising when one applies the permutation π to the variables π The following lemma is immediate from the definition. Lemma 2.1. If π = στ in Sm, then qj1,...,jm π jτ (1),...,jτ (m) = q σ qj1,...,jm τ . 4 For each m ≥ 1, define the map φm : Ae ⊗Vm(V ) → A⊗(m+2) by (sgn π)qj1,...,jm π ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m) ⊗ 1 (3) φm(1⊗2 ⊗ xj1 ∧ · · · ∧ xjm) = Xπ∈Sm for all distinct xj1, . . . , xjm. Note that φm is injective: The image of a basis of Vm(V ) under φm is linearly independent, as may be seen by comparing the variables involved. Set φ0 and φ−1 to be the identity maps on A ⊗ A and A, respectively. Remark 2.2. By its definition, the image of φm is contained in (4) m−2 (A ⊗ V ⊗i ⊗ R ⊗ V ⊗(m−i−2) ⊗ A), \i=0 where R ⊂ V ⊗ V is the vector subspace spanned by the relations xi ⊗ xj − qi,jxj ⊗ xi. For example, to see that φm(1⊗2 ⊗ xj1 ∧ · · · ∧ xjm) is in A ⊗ R ⊗ V ⊗(m−2) ⊗ A, fix π ∈ Sm. Let (12) denote the permutation transposing 1 and 2. The π- and π(12)-terms of formula (3) above are (writing π(12) = (π(1), π(2))π in Sm and applying the lemma): (sgn π)qj1,...,jm π ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m) ⊗ 1 + (sgn π(12))qj1,...,jm π(12) ⊗ xjπ(2) ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m) ⊗ 1 = (sgn π)qj1,...,jm π ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m) ⊗ 1 −(sgn π)q jπ(1),...,jπ(m) (π(1),π(2)) qj1,...,jm π ⊗ xjπ(2) ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m) ⊗ 1 = (sgn π)qj1,...,jm π ⊗ (xjπ(1) ⊗ xjπ(2) − q jπ(1),...,jπ(m) (π(1),π(2)) xjπ(2) ⊗ xjπ(1)) ⊗ xjπ(3) ⊗ · · · ⊗ xjπ(m) ⊗ 1, which is visibly in A ⊗ R ⊗ V ⊗(m−2) ⊗ A. Lemma 2.3. The map φ defined in equation (3) is a chain map. Proof. We must show that (5) φm−1 ◦ dm(1⊗2 ⊗ xj1 ∧ · · · ∧ xjm) = δm ◦ φm(1⊗2 ⊗ xj1 ∧ · · · ∧ xjm) for all j1, . . . , jm with j1 < · · · < jm, where δm is the differential on the bar complex (1). This may easily be checked when m = 0. By Remark 2.2, the formula (3) for φm and the formula for δm, the right side of (5) is (sgn π)qj1,...,jm π Xπ∈Sm (cid:16)xjπ(1) ⊗ · · · ⊗ xjπ(m) ⊗ 1 + (−1)m ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m)(cid:17) 5 qjs,ji)xji ⊗ 1 − ( m qji,js) ⊗ xji! ⊗ xj1 ∧ · · · ∧ xji ∧ · · · ∧ xjm! Ys=i qji,js) ⊗ xji! m Ys=i for all m ≥ 1. The left side of (5) is qjs,ji)xji ⊗ 1 − ( φm−1 m Xi=1 (−1)i+1 ( Xi=1 = m i i Ys=1 (−1)i+1 ( Ys=1 ⊗  Xπ′∈Si (sgn π′)( m−1 = m Xi=1 (−1)i+1 Xπ′∈Si m−1 i Ys=1 (sgn π′)qj1,..., ji,...,jm π′ xjπ′(1) ⊗ · · · ⊗ xjπ′(m)  qjs,ji)qj1,..., ji,...,jm π′ xji ⊗ xjπ′(1) ⊗ · · · ⊗ xjπ′(m) ⊗ 1 − (sgn π′)( m Ys=i qji,js)qj1,..., ji,...,jm π′ ⊗ xjπ′(1) ⊗ · · · ⊗ xjπ′(m) ⊗ xji! , m−1 denotes the symmetric group on {1, . . . ,i, . . . , m}. In the first set of where for each i, Si summands above, if π := π′(i, i−1, . . . , 1), then sgn π′ = (−1)i+1 sgn π, and we may replace (sgn π′)xji⊗xjπ′(1) ⊗· · ·⊗xjπ′(m) ⊗1 by (−1)i−1(sgn π)xjπ(1) ⊗· · ·⊗xjπ(m) ⊗1. Similarly in the sec- ond set of summands, let π = π′(i, i+1, . . . , m), and replace (sgn π′)⊗xjπ′(1)⊗· · ·⊗xjπ′(m)⊗xji by (−1)m−i(sgn π) ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m). Again, for the first set of summands, notice that s=i qji,js)qj1,..., ji,...,jm qj1,...,jm . π Making all such replacements, we find that the left side of (5) is indeed equal to the right side. (cid:3) Remark 2.4. The image of φm is a free Ae-submodule of A⊗(m+2) that is a direct summand of A⊗(m+2) as an Ae-module: Take a vector space complement in k ⊗ V ⊗m ⊗ k to the image , and for the second set, qj1,...,jm s=1 qjs,ji)qj1,..., ji,...,jm = (Qm = (Qi summand of A⊗(m+2). of k ⊗ k ⊗Vm(V ) under φm, and extend to the required complementary Ae-module direct Ae ⊗V q(V ) for which ψφ is the identity map. It follows that there is a chain map ψ from the bar resolution to 3. The cup product π′ π′ π As before, let A denote the quantum symmetric algebra Sq(V ) and let Ae denote the enveloping algebra A ⊗ Aop. Let B denote an A-bimodule. In this section, we will describe Ae(A, B), when B additionally the cup product on Hochschild cohomology HH q(A, B) = Ext q has a compatible algebra structure. We begin by applying HomAe(·, B) to (2), dropping the term HomAe(A, B) to obtain the complex (6) 0 −→ HomAe(Ae, B) d1−→ HomAe(Ae ⊗V1(V ), B) d2−→ HomAe(Ae ⊗V2(V ), B) −→ · · · , that is, the degree m term is HomAe(Ae ⊗Vm(V ), B), and dm is defined by dm(F ) = F ◦ dm, for all F ∈ HomAe(Ae ⊗Vm−1(V ), B). We identify the spaces HomAe(Ae ⊗Vm(V ), B) and B ⊗Vm(V ∗) via the map Θm : HomAe(Ae ⊗Vm(V ), B) → B ⊗Vm(V ∗) 6 F (1⊗2 ⊗ x∧β) ⊗ (x∗)∧β. F 7→ Xβ∈{0,1}N β=m We thus obtain the following complex, which is equivalent to the complex in (6): d∗ d∗ (7) 0 −→ B 1−→ B ⊗V1(V ∗) that is, the degree m term is B ⊗Vm(V ∗), and d∗ takes b ∈ B to F ∈ HomAe(Ae, B) defined by F (1⊗2) = b. 2−→ B ⊗V2(V ∗) −→ · · · , m is defined by d∗ m explicitly. For any b ∈ B and β ∈ {0, 1}N with β = m − 1 we have m = Θm ◦ dm ◦ Θ−1 m−1, where Θ−1 0 Let us describe d∗ m(b ⊗ (x∗)∧β) d∗ = (Θm ◦ dm ◦ Θ−1 = Θm(Θ−1 m−1(b ⊗ (x∗)∧β) ◦ dm) m−1)(b ⊗ (x∗)∧β) (Θ−1 m−1(b ⊗ (x∗)∧β) ◦ dm)(1⊗2 ⊗ x∧β ′ ) ⊗ (x∗)∧β ′ N = β ′=m = Xβ ′∈{0,1}N Xi=1 Xi=1 = N δβi,0(Θ−1 m−1(b ⊗ (x∗)∧β) ◦ dm)(1⊗2 ⊗ x∧(β+[i])) ⊗ (x∗)∧(β+[i]) δβi,0(Θ−1 m−1(b ⊗ (x∗)∧β)) (−1)Pi s=1 βs" i Ys=1 qβs s,i! xi ⊗ 1 − N Ys=i qβs i,s! ⊗ xi# ⊗ x∧β! ⊗ (x∗)∧(β+[i]). Using the Ae-linearity of Θ−1 be rewritten to show that d∗ m−1(b ⊗ (x∗)∧β) and the definition of Θ, the above expression may m(b ⊗ (x∗)∧β) is equal to (8) δβi,0(−1)Pi N Xi=1 s=1 βs" i Ys=1 qβs s,i! xib − N Ys=i qβs i,s! bxi# ⊗ (x∗)∧(β+[i]). We will use these expressions for the differentials in the sequel. Let q be a tuple of scalars as in the introduction. We define the quantum exterior algebra (9) i = 0 since qii = 1, that is x2 where T (V ) is the tensor algebra on V . Note that the relations corresponding to i = j are 2x2 Vq(V ) = T (V )/(xixj + qi,jxjxi 1 ≤ i, j ≤ N), (graded) vector space is the same as that of the exterior algebra V(V ). i = 0 in Vq(V ). It follows that the dimension of Vq(V ) as a Let q−1 denote the tuple consisting of all inverses of components of q, that is, q−1 = {q−1 i,j }1≤i,j≤N . Using this notation, we have the following cup product theorem. Our main applications are to the cases B = Sq(V ) and B = Sq(V ) ⋊ G for some finite group G in the next section. Theorem 3.1. Let B be an Sq(V )-bimodule algebra. Then the Hochschild cohomology alge- −1(V ∗), that is, the bra HH q(Sq(V ), B) is a subquotient algebra of the tensor product B ⊗Vq cup product on HH q(Sq(V ), B) descends from the product on B ⊗Vq −1(V ∗). 7 such that ψφ is the identity map. (We will not need an explicit formula for ψ.) We will use the cup product as defined on the bar complex, together with the chain maps Proof. Letting A = Sq(V ), by Remark 2.4, there is a chain map ψ : A⊗(m+2) → Ae ⊗Vm(V ) φ and ψ: For two cocycles µ and ν defined on Ae ⊗V(V ), in degrees m and n respectively, their cup product is given by φ∗(ψ∗(µ) ⌣ ψ∗(ν)). This function is defined by its action on all elements of the form 1⊗2 ⊗ xj1 ∧ · · · ∧ xjm+n, which we calculate next. Let Sm,n denote the set of all m, n-shuffles, that is all ρ ∈ Sm+n for which ρ(1) < · · · < ρ(m) and ρ(m + 1) < · · · < ρ(m + n). Note that these shuffles form a set of coset representatives of the subgroup Sm × Sn of Sm+n, where Sm acts on {1, . . . , m} and Sn acts on {m + 1, . . . , m + n}. Thus we may write each π ∈ Sm+n as π = ρστ where ρ ∈ Sm,n, σ ∈ Sm, τ ∈ Sn. By Lemma 2.1, writing ρστ = (ρσρ−1)(ρτ ρ−1)ρ, we have φ∗(ψ∗(µ) ⌣ ψ∗(ν))(1⊗2 ⊗ xj1 ∧ · · · ∧ xjm+n) (sgn π)qj1,...,jm+n π  Xπ∈Sm+n ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m+n) ⊗ 1  (sgn π)qj1,...,jm+n ψ∗(µ)(1 ⊗ xjπ(1) ⊗ · · · ⊗ xjπ(m) ⊗ 1) (sgn ρστ )qj1,...,jm+n ρστ · ψ∗(ν)(1 ⊗ xjπ(m+1) ⊗ · · · ⊗ xjπ(m+n) ⊗ 1) ψ∗(µ)(1 ⊗ xjρστ (1) ⊗ · · · ⊗ xjρστ (m) ⊗ 1) · ψ∗(ν)(1 ⊗ xjρστ (m+1) ⊗ · · · ⊗ xjρστ (m+n) ⊗ 1) = (ψ∗(µ) ⌣ ψ∗(ν)) = Xπ∈Sm+n = Xρ∈Sm,n Xσ∈Sm,τ ∈Sn π = Xρ∈Sm,n = Xρ∈Sm,n (sgn ρ)qj1,...,jm+n ρ (sgn σ)q jρ(1),...,jρ(m) ρσρ−1 Xσ∈Sm · Xτ ∈Sn (sgn τ )q jρ(m+1),...,jρ(m+n) ρτ ρ−1 (sgn ρ)qj1,...,jm+n ρ φ∗ψ∗(µ)(1⊗2 ⊗ xjρ(1) ∧ · · · ∧ xjρ(m)) ψ∗(µ)(1 ⊗ xjρσ(1) ⊗ · · · ⊗ xjρσ(m) ⊗ 1)! ψ∗(ν)(1 ⊗ xjρτ (m+1) ⊗ · · · ⊗ xjρτ (m+n) ⊗ 1)! · φ∗ψ∗(ν)(1⊗2 ⊗ xjρ(m+1) ∧ · · · ∧ xjρ(m+n)). Since ψφ is the identity map, φ∗ψ∗(µ) = µ and φ∗ψ∗(ν) = ν. Now replacing µ by b ⊗ (x∗)∧β and ν by b′ ⊗ (x∗)∧β ′, where b, b′ ∈ B, we see that only one summand in the above can be nonzero, and that is the summand corresponding to ρ for which x∧β = xjρ(1) ∧ · · · ∧ xjρ(m) and x∧β ′ = xjρ(m+1) ∧ · · · ∧ xjρ(m+n). For this summand, we obtain (sgn ρ)qj1,··· ,jm+n ρ bb′. Comparing, we find also that (bb′ ⊗ (x∗)∧β ∧ (x∗)∧β ′ )(1⊗2 ⊗ xj1 ∧ · · · ∧ xjm+n) = (sgn ρ)qj1,··· ,jm+n ρ bb′, 8 by permuting the exterior factors in (x∗)∧β ∧ (x∗)∧β ′ via ρ before applying the function, and jρ(1),...,jρ(m+n) )−1 (a consequence of Lemma 2.1). Therefore by using the identity q ρ−1 the product is as claimed. (cid:3) = (qj1,...,jm+n ρ Remark 3.2. It is not necessary that the characteristic of k be 0 for Theorem 3.1: While Wambst's proof that complex (2) is exact requires characteristic 0, a similar resolution may be obtained in positive characteristic by using a twisted tensor product construction [BO] or general theory for Koszul algebras [BG]. Our proof of Theorem 3.1 uses only resolution (2) and its embedding into the bar resolution. 4. The Hochschild cohomology algebra of Sq(V ) ⋊ G Let G be a finite group acting on A = Sq(V ) by graded algebra automorphisms. We are interested in the cohomology of the skew group algebra A ⋊ G. Since the characteristic of k is zero, this is known to be isomorphic to the G-invariant subalgebra of the cohomology HH q(A, A ⋊ G) of A with coefficients in A ⋊ G. (See for example [S].) In this section we Ae(A, A ⋊ G), in the case when will compute this latter cohomology, HH q(A, A ⋊ G) = Ext q G acts diagonally on the basis x1, . . . , xN of V . Note that each g-component Ag is a (left) Ae-module (see Section 2). Also note that since A ⋊ G =Lg∈G Ag as an Ae-module, Ext q Ae(A, A ⋊ G) ∼=Mg∈G Ext q Ae(A, Ag). We will compute the summands Ext q Ae(A, Ag). Fix g ∈ G. In Section 3, we applied the Hom functor HomAe(·, B), for any A-bimodule B, to the Ae-resolution of A in (2), and made appropriate identification to obtain the com- plex (7). When we specialize this complex to B = Ag, we obtain (10) 0 −→ Ag d∗ where formula (8) yields that d∗ m((a#g) ⊗ (x∗)∧β) is equal to d∗ 1−→ Ag ⊗V1(V ∗) 2−→ Ag ⊗V2(V ∗) −→ · · · , (11) N Xi=1 δβi,0(−1)Pi s=1 βs" i Ys=1 qβs s,i! xia − N Ys=i qβs i,s! a(gxi)! #g# ⊗ (x∗)∧(β+[i]), for all a ∈ A and β ∈ {0, 1}N with β = m − 1. 4.1. Additive structure. Suppose G acts diagonally on the basis x1, . . . , xN of V , that is, there exist scalars λg,i ∈ k such that gxi = λg,ixi, for all g ∈ G, i ∈ {1, . . . , N}. For each g ∈ G, define (12) Cg :=(γ ∈ (N ∪ {−1})N for each i ∈ {1, . . . , N}, qγs i,s = λg,i or γi = −1) . N Ys=1 The following theorem gives the structure of the Hochschild cohomology as a graded vector space. Recall definition (9) of the quantum exterior algebra. 9 Theorem 4.1. If G acts diagonally on the chosen basis of V , then HH q(Sq(V ), Sq(V )g) is the graded vector subspace of Sq(V )g ⊗Vq HHm(Sq(V ), Sq(V )g) ∼= Mβ∈{0,1}N β=m Mα∈NN α−β∈Cg −1(V ∗) given by: spank{(xα#g) ⊗ (x∗)∧β}, for all m ∈ N, g ∈ G. Therefore, HHm(Sq(V ), Sq(V ) ⋊ G) ∼=Mg∈G Mβ∈{0,1}N β=m Mα∈NN α−β∈Cg spank{(xα#g) ⊗ (x∗)∧β}, for all m ∈ N, and HHm(Sq(V ) ⋊ G) is its G-invariant subspace. g := HH q(Sq(V ), Sq(V )g) when N = 2. Example 4.2. Fix g ∈ G. We wish to describe HH q We will work out two special cases. When q1,2 is not a root of unity and λg,1, λg,2 are not both equal to 1, we have HH0 g HH1 g HH2 g ∼= {0} ∼= {0} ∼= spank{(1#g) ⊗ x∗ 1 ∧ x∗ 2}. 1 xα2 1 xα2 ∼= spank{xα1 ∼= spank{(xα1 g,2. Also assume q1,2 6= λg,2 and q−1 For the second case, assume q1,2 is simultaneously a primitive ℓth root of unity, a ℓ1th root of λg,1, and a ℓ2th root of λ−1 HH0 g HH1 g 2 #g α1, α2 ∈ N, ℓ divides both α1 − ℓ2 and α2 − ℓ1} 2 #g) ⊗ x∗ 1 xα2 ∼= spank{(1#g) ⊗ x∗ 1 xα2 2 α1, α2 ∈ N, ℓ divides both α1 − ℓ2 and α2 − ℓ1 − 1} 1 α1, α2 ∈ N, ℓ divides both α2 − ℓ1 and α1 − ℓ2 − 1} 2 #g) ⊗ x∗ 1 ∧ x∗ 2} 2 #g) ⊗ x∗ 1,2 6= λg,1. Then we have 2 α1, α2 ∈ N, ℓ divides α1 − ℓ2 − 1 and α2 − ℓ1 − 1}. 1 ∧ x∗ HH2 g M spank{(xα1 M spank{(xα1 g is the free module over Z(Sq(V )) generated by xℓ2 Thus HH0 module generated by (xℓ2+1 xℓ1 2 #g) ⊗ x∗ sum of the k-linear span of (1#g) ⊗ x∗ (xℓ2+1 xℓ1+1 2 #g) ⊗ x∗ 1 ∧ x∗ 2. 1 1 g is the free Z(Sq(V ))- g is the direct 2 and the free Z(Sq(V ))-module generated by 2 #g) ⊗ x∗ 2, and HH2 1 and (xℓ2 1 ∧ x∗ 1 xℓ1+1 2 #g, HH1 1 xℓ1 Define (13) C :=(γ ∈ (N ∪ {−1})N for each i ∈ {1, . . . , N}, qγs i,s = 1 or γi = −1) . N Ys=1 Taking G to be the trivial group with one element, we immediately obtain the following: Corollary 4.3. HH q(Sq(V )) is the graded vector subspace of Sq(V ) ⊗Vq spank{xα ⊗ (x∗)∧β}, HHm(Sq(V )) ∼= Mβ∈{0,1}N β=m Mα∈NN α−β∈C −1(V ∗) given by: for all m ∈ N. 10 Example 4.4. When N = 2, the description of HH q(Sq(V )) simplifies considerably. In this case, if q1,2 is not a root of unity, then HH0(Sq(V )) ∼= k HH1(Sq(V )) ∼= spank{x1 ⊗ x∗ HH2(Sq(V )) ∼= spank{1 ⊗ x∗ 1, x2 ⊗ x∗ 2} 1 ∧ x∗ and if q1,2 is a primitive ℓth root of unity, ℓ ≥ 2, then 2, x1x2 ⊗ x∗ 1 ⊗ x∗ 2}, α1, α2 ∈ N, ℓ divides both α1 and α2} 1 α1, α2 ∈ N, ℓ divides both α1 − 1 and α2} 2 ⊗ x∗ 2 α1, α2 ∈ N, ℓ divides both α1 and α2 − 1} HH0(Sq(V )) ∼= spank{xα1 HH1(Sq(V )) ∼= spank{xα1 HH2(Sq(V )) ∼= spank{1 ⊗ x∗ 1 xα2 2 2 ⊗ x∗ 1 xα2 1 xα2 1 ∧ x∗ 2} 1 xα2 M spank{xα1 M spank{xα1 2 ⊗ x∗ 1 ∧ x∗ 2 α1, α2 ∈ N, ℓ divides α1 − 1 and α2 − 1}. Note that in the first case, the center of Sq(V ) is Z(Sq(V )) = k, while in the second case, 2. Thus in either case, HH0(Sq(V )) = Z(Sq(V )) as Z(Sq(V )) is generated by xℓ expected, HH1(Sq(V )) is the free Z(Sq(V ))-module generated by x1 ⊗ x∗ 2, and HH2(Sq(V )) is the direct sum of the k-linear span of 1 ⊗ x∗ 2 and the free Z(Sq(V ))- module generated by x1x2 ⊗ x∗ 2. Similar expressions may be obtained when N ≥ 3, but there are more cases to consider due to the many more parameters involved. 1 and x2 ⊗ x∗ 1 and xℓ 1 ∧ x∗ 1 ∧ x∗ We introduce some notation and lemmas before proving Theorem 4.1. Fix g ∈ G. For any α ∈ NN , β ∈ {0, 1}N , and i ∈ {1, . . . , N}, define Ωg(α, β, i) := qαs−βs i,s = λg,i, N 0, if 0, if βi = 1, Ys=1 ε(β, i) i Ys=1   qαs−βs i,s − λg,i qαs−βs s,i ! , otherwise, N Ys=i where ε(β, i) = (−1)Pi s=1 βs. Then, using formula (11) for d∗ m we see that N Xi=1 (14) m((xα#g) ⊗ (x∗)∧β) = d∗ Ωg(α, β, i)(xα+[i]#g) ⊗ (x∗)∧(β+[i]), for all α ∈ NN and β ∈ {0, 1}N with β = m − 1. For any γ ∈ (N ∪ {−1})N and m ∈ N, define g,γ := spank{(xα#g) ⊗ (x∗)∧β α ∈ NN , β ∈ {0, 1}N , β = m, and α − β = γ}. K m g,γ denote the subcomplex of (10) whose mth term is given by K m g,γ. That K q Let K q indeed a subcomplex of (10) follows from (14). We immediately obtain: g,γ is Lemma 4.5. The complex (10) admits a grading by (N ∪ {−1})N . Precisely, the mth term of complex (10) decomposes as Mγ∈(N∪{−1})N K m g,γ, for all m ∈ N. 11 We now separately handle the cases where γ is or is not in the set Cg defined in (12). For the proof of the next lemma, we will need some notation: For any γ ∈ (N ∪ {−1})N , define kγkg := #(i ∈ {1, . . . , N} qγs i,s 6= λg,i and γi 6= −1) . N Ys=1 Lemma 4.6. Let γ ∈ ((N ∪ {−1})N \Cg). Then, the subcomplex K q g,γ of (10) is acyclic. Proof. We will show that the identity chain map of the complex K q define the following scalar. Let α ∈ NN , β ∈ {0, 1}N , and i ∈ {1, . . . N}. Define g,γ is nullhomotopic. First, ωg(α, β, i) := 0, if qαs−βs i,s = λg,i, N Ys=1 0, if αi = 0, 0, if βi = 0, Ωg(α − [i], β − [i], i)−1, otherwise.   hm : K m g,γ → K m−1 g,γ Now, fix m ∈ N and suppose that β = m and α − β = γ. Define by hm((xα#g) ⊗ (x∗)∧β) := Note that kγkg 6= 0, as γ 6∈ Cg. We contend that 1 kγkg N Xi=1 ωg(α, β, i)(xα−[i]#g) ⊗ (x∗)∧(β−[i]). (hm+1 ◦ d∗ m+1 + d∗ m ◦ hm)((xα#g) ⊗ (x∗)∧β) = (xα#g) ⊗ (x∗)∧β. Note that the lemma is proved if this equality is sustained. The left hand side of this equality is equal to 1 kγkg N N Xi=1 Xj=1 Ωg(α, β, i)ωg(α + [i], β + [i], j)(xα+[i]−[j]#g) ⊗ (x∗)∧(β+[i]−[j]) + 1 kγkg N N Xj=1 Xi=1 ωg(α, β, j)Ωg(α − [j], β − [j], i)(xα+[i]−[j]#g) ⊗ (x∗)∧(β+[i]−[j]) N = 1 kγkg Xi=1 (cid:2)Ωg(α, β, i)ωg(α + [i], β + [i], i) + ωg(α, β, i)Ωg(α − [i], β − [i], i)(cid:3)(xα#g) ⊗ (x∗)∧β 1 + kγkg Xi6=j (cid:2)Ωg(α, β, i)ωg(α + [i], β + [i], j) It follows from the definition of Ωg and ωg that + ωg(α, β, j)Ωg(α − [j], β − [j], i)(cid:3)(xα+[i]−[j]#g) ⊗ (x∗)∧(β+[i]−[j]) N Xi=1 Ωg(α, β, i)ωg(α + [i], β + [i], i) + ωg(α, β, i)Ωg(α − [i], β − [i], i) = kγkg. 12 Therefore, it only remains to show that Ωg(α, β, i)ωg(α + [i], β + [i], j) + ωg(α, β, j)Ωg(α − [j], β − [j], i) = 0, whenever i 6= j. To this end, define and Ξg(i, j) := Ωg(α, β, i)ωg(α + [i], β + [i], j) Ξ′ g(i, j) := ωg(α, β, j)Ωg(α − [j], β − [j], i). Suppose i 6= j. Then it is clear from definition of Ωg and ωg that Ξg(i, j) and Ξ′ g(i, j) are simultaneously zero or nonzero, so we may assume that they are nonzero. In this case, we have Ξg(i, j) = ε(β, i)ε(β+[i]−[j], j) i Ys=1 qαs−βs i,s − λg,i N Ys=i and Ξ′ g(i, j) = ε(β−[j], i)ε(β−[j], j) j Ys=1 qαs−βs j,s − λg,j N Ys=j qαs−βs s,i ! j Ys=1 s,j !−1 i Ys=1 qαs−βs qαs−βs j,s − λg,j qαs−βs s,j !−1 N Ys=j qαs−βs i,s − λg,i qαs−βs s,i ! . N Ys=i Therefore, the desired equality Ξg(i, j) = −Ξ′ g(i, j) is equivalent to the equality ε(β − [j], i)ε(β − [j], j) = −ε(β, i)ε(β + [i] − [j], j), which may be easily verified using the definition of ε, the corresponding conditions under which Ωg and ωg are nonzero, and the condition i 6= j. (cid:3) Proof of Theorem 4.1. Observe that the restriction of d∗ theorem now follows immediately from Lemmas 4.5 and 4.6. q to K q g,γ is zero, for all γ ∈ Cg. The (cid:3) 4.2. Cup product. Assume that G acts diagonally on the basis x1, . . . , xN of V . The following is immediate from Theorem 3.1 when we put B = Sq(V ) ⋊ G. Recall the definition −1(V ∗) given by (9). of Cg in (12) and of Vq Theorem 4.7. HH q(Sq(V ), Sq(V ) ⋊ G) is a subquotient algebra of (Sq(V ) ⋊ G) ⊗Vq Thus the cup product on HH q(Sq(V ), Sq(V ) ⋊ G) is given by −1(V ∗). ((xα#g) ⊗ (x∗)∧β) ⌣ ((xα′ #h) ⊗ (x∗)∧β ′ ) = (xα#g)(xα′ #h) ⊗ (x∗)∧β ∧ (x∗)∧β ′ for all g, h ∈ G, α, α′ ∈ NN and β, β′ ∈ {0, 1}N such that α − β ∈ Cg and α′ − β′ ∈ Ch. Moreover, HH q(Sq(V ) ⋊ G) is the G-invariant subalgebra of HH q(Sq(V ), Sq(V ) ⋊ G). Remark 4.8. Note that the above product is zero when the supports of β and β′ intersect nontrivially. Furthermore, we understand the above product to be zero (i.e., a coboundary) if α + α′ − (β + β′) is not in Cgh. Recall the definition of C in (13). Taking G to be the trivial group with one element, we immediately obtain the following: 13 Corollary 4.9. HH q(Sq(V )) is a subquotient algebra of Sq(V ) ⊗Vq product on HH q(Sq(V )) is given by ⊗ (x∗)∧β ∧ (x∗)∧β ′ for all α, α′ ∈ NN and β, β′ ∈ {0, 1}N such that α − β ∈ C and α′ − β′ ∈ C. (xα ⊗ (x∗)∧β) ⌣ (xα′ −1(V ∗). Thus the cup ⊗ (x∗)∧β ′ ) = xαxα′ Remark 4.10. As before, the above product is zero when the supports of β and β′ intersect nontrivially. Furthermore, we understand the above product to be zero (i.e., a coboundary) if α + α′ − (β + β′) is not in C. References [AS] N. Andruskiewitsch and H.-J. Schneider, Pointed Hopf algebras, in: New directions in Hopf algebras, [A] MSRI Publ., 43, Cambridge Univ. Press, Cambridge, 2002. R. Anno, Multiplicative structure on the Hochschild cohomology of crossed product algebras, arXiv:math.QA/0511396. [BB] Y. Bazlov and A. Berenstein, Noncommutative Dunkl operators and braided Cherednik algebras, Sel. Math. New Ser. 14 (2009), no. 3 -- 4, 325 -- 372. [BGS] A. Beilinson, V. Ginzburg and V. Soergel, Koszul duality patterns in representation theory, J. Am. Math. Soc. 9 (1996), no. 2, 473 -- 527. [BO] P. A. Bergh and S. Oppermann, Cohomology of twisted tensor products, J. Algebra 320 (2008), no. 8, 3327 -- 3338. [BG] A. Braverman and D. Gaitsgory, Poincar´e-Birkhoff-Witt Theorem for quadratic algebras of Koszul [F] [G] [GG] type, J. Algebra 181 (1996), no. 2, 315 -- 328. M. Farinati, Hochschild duality, localization, and smash products, J. Algebra 284 (2005), no. 1, 415 -- 434. M. Gerstenhaber, The cohomology structure of an associative ring, Ann. of Math. 78 (1963), no. 2, 267 -- 288. J. A. Guccione and J. J. Guccione, Hochschild and cyclic homology of Ore extensions and some examples of quantum algebras, K-Theory 12 (3) (1997), 259 -- 276. [Gu] A. Guichardet, Homologie de Hochschild des d´eformations quadratiques d'alg´ebres de polynomes, Comm. Algebra 26 (12) (1998), 4309 -- 4330. [GK] V. Ginzburg and D. Kaledin, Poisson deformations of symplectic quotient singularities, Adv. Math. 286 (2004), no. 1, 1 -- 57. [KKZ] E. Kirkman, J. Kuzmanovich, and J. J. Zhang, Shephard-Todd-Chevalley Theorem for skew polyno- [M] [P] [R] mial rings, Alg. Rep. Theory 13 (2010), no. 2, 127 -- 158. Yu. I. Manin, Quantum groups and non-commutative geometry, CRM Universit´e de Montr´eal, 1988. S. Priddy, Koszul resolutions, Trans. Amer. Math. Soc. 152 (1970), 39 -- 60. L. Richard, Hochschild homology and cohomology of some classical and quantum noncommutative polynomial algebras, J. Pure Appl. Algebra 187 (2004), 255 -- 294. [SW] A. V. Shepler and S. Witherspoon, Hochschild cohomology and graded Hecke algebras, Trans. Amer. [S] Math. Soc. 360 (2008), no. 8, 3975 -- 4005. D. S¸tefan, Hochschild cohomology on Hopf Galois extensions, J. Pure Appl. Algebra 103 (1995), 221-233. [W] M. Wambst, Complexes de Koszul quantiques, Ann. Fourier, 43 (1993), no. 4, 1089 -- 1156. E-mail address: [email protected] E-mail address: [email protected] E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, Texas 77843, USA 14
1304.0530	1	1304	2013-04-02T04:48:42	The Ring of Algebraic Functions on Persistence Bar Codes	[ "math.RA" ]	We study the ring of algebraic functions on the space of persistence barcodes, with applications to pattern recognition.	math.RA	math	TheRingofAlgebraicFunctionsonPersistenceBarCodesAaronAdcockErikCarlssonGunnarCarlsson∗May10,20141IntroductionPersistenthomology([3],[13])isafundamentaltoolintheareaofcompu-tationaltopology.Itcanbeusedtoinfertopologicalstructureindatasets(see[1],[4]),butvariationsonthemethodcanbeappliedtostudyaspectsoftheshapeofpointcloudswhicharenotovertlytopological([5],[8]).Themethodologyassignstoanyfinitemetricspace(suchasaretypicallyobtainedinexperimentaldataofvariouskinds)andnon-negativeintegerkabarcode,bywhichwewillmeanafinitecollectionofintervalswithendpointsontherealline.Theintegerkspecifiesadimensionofafeature(zero-dimensionalforacluster,one-dimensionalforaloop,etc.),andanintervalrepresentsafeaturewhichis“born”atthevalueofaparameter(thepersistencepa-rameter)givenbythelefthandendpointoftheinterval,andwhich“dies”atthevaluegivenbytherighthandendpoint.Thesebarcodeshavebeendemonstratedtoidentifystructureinspacesofimagepatchesin[1]and[4],andhavebeendemonstratedtodistinguishbetweenhanddrawnlettersin[8].Becauseoftheunusualstructureoftheinvariant,i.e.asacollectionofintervalsratherthannumericalquantities,themethodcurrentlyrequiressubstantialknowledgeoftopologicalmethods.Itwouldclearlybeusefultoassignandinterpretvariousnumericalquantitiesattachedtobarcodes,sothattheseoutputscouldbeusedasinputtostandardalgorithmswithinma-chinelearning,clusteranalysis,andothermethods.Itisthepurposeofthis∗ResearchsupportedinpartbyNSFDMS-04069921arXiv:1304.0530v1 [math.RA] 2 Apr 2013papertoidentifyanalgebraoffunctionsonthesetofbarcodeswhichisdefinedinaconceptuallycoherentway.Themainideaisthefollowing.Abarcodewithexactlynintervalscanbespecifiedbyavector(x1,y1,x2,y2,...,xn,yn),wherexidenotestheleftendpointofthei-thintervalandyitherightendpoint.However,thisrepresentationismanytoone,inthatthebarcodestructuredoesnotretaintheorderingontheintervals.Infact,thesetofbarcodeswithexactlynintervalscanbeidentifiedwiththesetSpn(R2)then-foldsymmetricproductofR2.ForanysetX,Spn(X)isdefinedtobetheorbitspaceoftheactionofthesymmetricgrouponnlettersontheproductXngivenbypermutingthecoordinates.Ontheotherhand,thespace(R2)nisanalgebraicvarietyoverR([10]).Infact,itisanaffinespaceofdimension2n,andthesymmetricgroupactionmentionedaboveisanalgebraicaction.Itisthenknown(see[12])thattheorbitspaceinheritsthestructureofanalgebraicvariety,andtheelementsofitsaffinecoordinatering([10])arefunctionsonthesetofbarcodeswithexactlynintervals.Theseaffinecoordinateringsarewellknownalgebrasreferredtogenericallyasringsofmultisymmetricpolynomials([9]).Theycanbequitecomplicated,sinceitturnsoutthatanysetofalgebrageneratorsforthemwillsatisfynon-trivialrelationsorsyzygies.Itturnsout,though,thatthereareinclusionsofalgebraicvarietiesSpn(R2)→Spn+1(R2)(1)whichproduceaninversesystemofgradedaffinecoordinaterings···→A[Spn+1(R2)]→A[Spn(R2)]→···whoseinverselimitwewilldenote,byabuseofnotation,byA[Sp∞(R2)].ThenotationA[−]denotestheaffinecoordinatering.Thisalgebraisknowntobefreelygeneratedonasetofminimalalgebragenerators([9]).Theanalysisofthesystem(1)aboveisnotsufficient,though.Thissystemidentifiesapoint((x1,y1),...,(xn,yn))∈Spn(R2)withthepoint((x1,y1),...,(xn,yn),(0,0))∈Spn+1(R2)Inotherwords,asetSofnintervalsisidentifiedwiththesetofn+1intervalsobtainedbyadjoiningtheintervaloflengthzerowhosetwoend-pointsarezero.However,intheparametrizationoftheisomorphismclasses2ofpersistencevectorspacesin[13]bybarcodes,anyintervaloflengthzeroisidentifiedwiththezeromodule.So,wewouldliketodeterminetheringofallalgebraicfunctions(i.e.theelementsofA[Sp∞(R2)])whichhavethepropertythattheytakethesamevalueonanybarcodeasontheresultofadjoininganyintervaloflengthzerotoit.Inthispaper,wewillidentifythissubring,describeitsstructure,anddescribethealgebrageneratorsex-plicitlysothattheycanbeusedeffectivelybythoseinterestedinanalyzingdatabasesofshapes.2TheInd-schemeBWefirstdiscussthesetofbarcodes,withoutanyalgebraicvarietystructures.Foreveryn,wefirstconsiderthesetofbarcodescontainingexactlyninter-vals.Wewillpermitintervalsoflengthzero.ThesetofintervalsIcanbeidentifiedwiththesubsetofR2consistingofpairs(x,y)withx≤y.Abarcodecontainingnintervalsisthereforeidentifiedwiththen-foldsymmetricproductSpn(I),whereforanysetX,Spn(X)isdefinedtobetheorbitspaceoftheactionofthesymmetricgrouponnlettersontheproductXngivenbypermutingthecoordinates.OnecanassemblethesesetsintoadirectedsystemIi1→Sp2(I)i2→Sp3(I)i3→Sp4(I)→···wherethemapsin:Spn(I)→Spn+1(I)aregivenbyin({I1,...,In})={I1,...,In,[0,0]}ThedirectlimitofthissystemwillbedenotedbySp∞(I).WeareinterestedinstudyingfunctionsonSp∞(I).Suchafunctioncanbeidentifiedwithaninfinitevector(f1,f2,f3,...)offunctionsfn:Spn(I)→Rsatisfyingthecompatibilityconditionfn+1·in=fnThesetofallsuchvectorsoffunctionsformsaringRundercoordinate-wiseadditionandmultiplication.Itisnotexactlywhatwewant,however.Thereasonisthatundertheparametrizationofpersistencevectorspacesasdescribedin[3]and[13],intervalsoflengthzerocorrespondtozerovectorspaces,andthereforeallintervalsoflengthzeroshouldbeconsideredequal.ThismeansthatweshouldconsideronlyfunctionsF:Sp∞(I)→Rfor3whichF({I1,I2,...,In,[ξ,ξ]})=F({I1,I2,...,In,[η,η]})forallpossiblevaluesofξandη.ThesetofallsuchfunctionsisasubringR0⊆R.ThissetoffunctionscanbedefinedasthesetofallfunctionsonthesetBdefinedbyB=anSpn(I)/’where’istheequivalencerelationgeneratedbyallrelationsoftheform{I1,I2,In,[ξ,ξ]}’{I1,I2,...,In}.Remark:ThereadermaysuggestthatoneconsiderinsteadonlythesubsetI+consistingofintervalsofpositivelength.Thiswillproduceadisjointunionofsetsofbarcodes,partitionedintothesetscontainingafixedpositivenumberofintervalsofpositivelength.Suchadescriptiondoesnottakeintoaccountthefactthatwewouldliketotopologizethespaceofallbarcodesinsuchawaythatlim→0{I1,I2,...In,[xn+1,xn+1+]}={I1,I2,...,In}Thereasonforthisisthatsmallperturbationstotheinputdatatothepersis-tencealgorithmscanmodifythebarcodesbymodifyinglengthsofintervalsasmallamountandaddintervalsofsmalllength.Thisisthestabilitytheoremforpersistencediagramsprovedin[7].TheringoffunctionsR0istoolargetodealwitheffectively.EventhemuchsmallerringofcontinuousfunctionsonBisstilltoocomplextodescribecompletely.WewillobservethatBisdescribedasacolimitofalgebraicvarieties,andthatitisthereforepossibletodefinetheringofalgebraicfunctionsonB.Itisthisringwewillanalyze.Throughoutthispaper,kwilldenotethefieldR.Allvarietieswillbeoverk.WeconsidertheaffinespaceAn=A(n)ofdimension2n,parametrizedwithcoordinates(x1,y1,x2,y2,...,xn,yn).ItsaffinecoordinateringisthepolynomialringBn=k[x1,y1,...,xn,yn].Thereisanactionofthesym-metricgroupSnonnlettersonAn,andfrom[12]itfollowsthatthesetoforbitsonthesetofpointsofthevarietyisitselfanaffinealgebraicvariety,withaffinecoordinateringequaltotheinvariantsubringBSnn.LetWi⊆Andenotethesubvarietyyi−xi=0.WeletDn⊆BndenotethesubringoffunctionswhoserestrictiontoWiisindependentofxiforalli.Wewishtocharacterizethissubringalgebraically.4Proposition1.TheringDnischaracterizedalgebraicallyasthesubringofallfforwhich(∂∂xi+∂∂yi)f∈(yi−xi)foralli.Proof:Wefixi,andconsiderallthefunctionsfforwhichfWiisindepen-dentofxi(andthereforeyi).Theoperator∂∂xi+∂∂yiinducesadifferentialoperatoronthequotientringQn=Bn/(yi−xi),whichisidentifiedwiththepartialdifferentialoperator2∂∂xiinQn∼=k[x1,y1,...,yi−1,xi,xi+1,...,xn,yn]TherequirementisthattheimagefoffinWiisindependentofxi,andthisisequivalenttothecondition∂∂xi(f)=0.Thisconditionistoholdforeachi,whichgivestheresult.3TheringofalgebraicfunctionsonBWebeginbychangingcoordinatesviatheformulaeξi=xi+yiandηi=yi−xi.ItisclearthatBncanalsobeidentifiedwithk[ξ1,η1,...,ξn,ηn],andthatthesymmetricgroupinthenewcoordinatesystempermutestheξi’sandηi’s.Underthistransformation,theoperator∂∂xi+∂∂yiiscarriedintotheoperator2∂∂ξi.ThismeansthattheringDnisidentifiedwiththesubringoffunctionsf(ξ1,η1,...,ξn,ηn)forwhich∂f∂ξi∈(ηi)foralli.Proposition2.Ak-basisfortheDnisgivenbythesetofmonomialsξa11ξa22···ξannηb11ηb22···ηbnnforwhichai>0impliesbi>0.Proof:Wenotethattheoperator∂/∂ξicarrieseachmonomialtoaconstantmultipleofasinglemonomial,namelythemonomialobtainedbydecreasingaibyone.Moreover,containmentintheideal(ηi)isalsogivenpurelybyconditionsonmonomials,i.e.thatbi>0.WeconcludethatDnisspannedbymonomialslyinginDn.ButitisclearthatamonomialµliesinDnexactlyifitisthecasethatwheneverξidividesµ,thenηialsodividesµ.5Thiscorrespondstotheabovenumericalconditionontheexponentsinthemonomial.ThesymmetricgroupactionclearlypreservesthesubringDn.Moreover,itpreservesthebasisofmonomialswithinDn.Let{µα}α∈AdenoteasetoforbitrepresentativesoftheSn-actiononthesetofmonomialsdefinedinProposition2.Letσαdenotethesumofalltheelementsintheorbitofµα.Proposition3.WeletDSnndenotethesubringofelementsofDnwhichareinvariantundertheactionofSn.Thentheelementsσαformak-basisofDSnn.Proof:ThisresultplainlyholdsforanyalgebraoverafieldofcharacteristiczeroonwhichthereisaG-actionwhichpreservesabasisofmonomials.Wehaverestrictionmapsπn,m:Dn→Dm,whenn≥m,definedbyπn,m(ξi(respηi))=ξi(respηi)fori≤m,andπn,m(ξi)=0fori>m.Themapπn,misSm-equivariant,whereSmactsbypermutingthefirstmpairsofvariables.ItfollowsthatwemayconstructcompositesDSnn,→DSmnπSmn,m→DSmmwhichwedenotebyσn,m,andthereforetheinversesystem···σn+1,n−→DSnnσn,n−1−→DSn−1n−1σn−1,n−2−→···σ2,1−→D1WewilldenotetheinverselimitofthisgradedsystembyD.Wenextrecallsomeofthenotationandbasicfactsaboutmultisymmetricpolynomials,whichcanbefoundinDalbec[9].LetRn,rbethepolynomialringinnrvariables,Rn,r=k[x11,x21,...,xnr],andletΛn,r=RSnn,r,denotetheringofSninvariants,wherethesymmetricgroupactsdiagonally.ThereisaninversesystemparalleltotheoneconstructedaboveinvolvingtheringsΛn,r.Wehaveevaluationmapsπn,m:Rn,r→Rm,r,m≤ndefinedbysettingxir=0ifi>m.Themapπn,misSm-equivariant,whenSm⊆Snisthesubgroupofpermutationsofthefirstmelementsoftheset{1,...,m}.Wehavethecomposites6Λn,r=RSnn,r,→RSmn,rπSmn,m→RSmm,r=Λm,rwhichwedenotebyρn,m.Theinverselimitofthesystem···ρn+1,n−→Λn,rρn,n−1−→Λn−1,rρn−1,n−2−→···ρ2,1−→Λ1,rasgradedringswillbedenotedbyΛr,andreferredtoastheringofr-multisymmetricfunctions.ItsgradingisgivenbyΛr=MkΛkrinducedbythegradingonRn,r.ThereisanevidentembeddingD,→Λ2.WewillusethisembeddingtoidentifythestructureofD.Theringofmultisymmetricfunctionshasseveralinterestingsetsofgen-erators.Givensomenonzerovectorsai=(ai1,...,air)∈Nr\0,wedefinethemultisymmetricmonomialsbyma1,...,ak=Symxa1111···xakrkr∈Λr,whereSymisthesymmetrizationmap.SymappliedtoamonomialyieldsthesumofallmonomialswhichareintheorbitoftheSn-action.TheyformavectorspacebasisofΛn,r,foranyn.ItisknownthatΛn,risgeneratedasanalgebrabythesymmetrizationsofmonomialsinvolvingonly{x11,x12,...,x1r}.Theyaregivenbytheformulaepa=ma=Xixa1i1···xarir,andarecalledthemultisymmetricpowersums.Whiletherearerelationsamongthepowersumsinfinitelymanyvariables,theyfreelygeneratetheinverselimitΛr,makingitapolynomialalgebra.See[9]fordetails.Wewillbeinterestedinthecaser=2.Letussetxi=xi1,yi=xi2,andletAn=Rn,2=k[x1,y1,...,xn,yn].ThesubalgebraD⊆Λ2nowhasthefollowingcharacterization.7Theorem1.AsasubalgebraofΛ2,Disfreelygeneratedbyelementsoftheformpa+1,b−pa,b+1.CheckingthatthesegeneratorsarecontainedinB,andthattherearenorelationsbetweenthemiseasy.TheworkisincalculatingHilbertseries,P(Ω)=Xk≥0dimR(Ωk)tk,withinducedgradingonΩ.Wedothisinthefollowinglemmas.Lemma4.AnR-basisforBnisgivenbythesetofmonomials(x1−y1)a1yb11···(xl−yl)alybllforwhichbi>0impliesai>0.Proof.Makethesubstitutionzi=xi−yi,whichcorrespondstoanisomor-phismA=R[z1,y1,...,zn,yn],andnoticethatBnisexactlythekernelofthedifferentialoperator∂/∂zi.Theoperator∂/∂zicarrieseachmonomialtoaconstanttimesasinglemonomial,namelythemonomialobtainedbydecreasingaibyone.Moreover,contain-mentintheideal(zi)isalsogivenpurelybyconditionsonmonomials,i.e.thatai>0.WeconcludethatBnisspannedbymonomialslyinginBn.ButitisclearthatamonomialµliesinBnexactlyifitisthecasethatwheneveryidividesµ,thenzialsodividesµ.Thisclearlycorrespondstotheabovenumericalconditionontheexponentsinthemonomial.Lemma5.TheHilbertseriesofΩisP(Ω)=Yd≥1(1−tk)−k.Proof.TheabovepropositionshowsthatBnhasabasisofmonomialswhichareinvariantundertheSn-action.Wheneverthisistrue,anysetoforbitrepresentativesconstituteabasisofBSnnoverafieldofcharacteristiczero.Wedefinesuchasetofrepresentativesbythemonomialsoftheformza11yb11···zallybll,ϕ−1(ai,bi)≥ϕ−1(ai+1,bi+1)8wherel≤n,andϕ:N+→N+×Nisthebijection(ϕ1,ϕ2,...)=((1,0),(1,1),(2,0),(1,2),(2,1),(3,0),(1,3),...)ontothesetofpossiblenonzeroexponents.Thedimensionofthek-gradedcomponentofBSnnisjustthenumberofthesemonomialsofdegreek.Letussaythat(a,b)≤(c,d)whenϕ−1(a,b)≤ϕ−1(c,d),andletf(a,b,k)denotethenumberofsequences(a1,b1,...,al,bl)suchthat(a,b)≥(a1,b1)≥···≥(al,bl),(ai,bi)∈N+×Nwithnorestrictionsonl.Itiseasytocheckthatitsatisfiestherecursionrelationf(a,b,k)=X(c,d)≤(a,b)f(c,d,k−c−d),whichleadstotheformulaXk≥0f(a,b,k)tk=(1−ta+b)−aY1≤k≤a+b−1(1−tk)−k.Wethenhavelimn→∞P(BSnn)=lim(a,b)→∞Xk≥1f(a,b,k)tk=Yk≥1(1−tk)−k.Wecannowprovethetheorem.Proof.LetΩ0=R[p10−p01,p11−p20,...].ItissimpletocheckthatΩ0⊂Ω.TherearenorelationsbetweenthesegeneratorsbecausethehomomorphismΛ2→Λ2,pa+1,b7→pa+1,b−pa,b+1,p0,b7→p0,bisanisomorphism.ItremainstoshowthatthetworingshavethesameHilbertseries.ButP(Ω0)obviouslyequalsthegeneratingfunctioninlemma5,becausetherearekgeneratorsindegreek.94MachineLearningonBwithexamples4.1DigitsExampleToillustratetheclassificationpotentialofthistechnique,weapplyittotheMNISTdatabase[11],ofhandwrittendigits.Weemphasizethattheaimisnottooutperformexistingmachinelearningalgorithmsfordigitclassifica-tion,buttopresentanexamplethatdemonstratesonewayofcombiningthistechniquewithexistingmachinelearningtechniques.Whileitisclearthatpuretopologicalclassificationcannotdistinguishbetweenthedigits(therearethreenumbersthatdonothaveanyloops,threethatalwayshaveloops,onethathastwoloopsandthreethathavestyle-dependentloops),wecanusethepowerofpersistenthomologytosiftoutmoreinformation.WebeginbyshowingthefullanalysisofafewdigitsandthengivetheempiricalresultsofapplyingthistechniquetoasubsetoftheMNISTdatabase.4.1.1TopologicalMethodsWebeginbydescribingaparticulargraphconstructiongivenadigitalimage.Wetreatthepixelsasverticesandaddedgesbetweenadjacentpixels(includ-ingdiagonals).Wecannowdefineafiltrationontheverticesofthegraphcorrespondingtotheimagepixels.Anaturalfiltrationcouldbeconstructedusingthepixelintensitiesoftheoriginalimage(seeFigure6,Section4.2).Anotherfiltration,usedin[8],canbeconstructedbythresholding,topro-duceabinaryimage,andadding1-pixelsaswesweepacrosstheimage.Thisaddsspatialinformationintowhatwouldotherwisebeapurelytopologicalmeasurement.Sincetheorientationofthedigitmatters(a6isthesameasa9givena180degreerotation),wechoosethelatterapproachandsweepacrosstherowsandcolumnsofeachdigit.Bytakingintoaccountspatialinformation,wegetaroughviewofthelocationofvarioustopologicalfeatures.Forexample,thougha‘9’and‘6’bothhaveoneconnectedcomponentandasingleloop,theloopwillappearatdifferentlocationsinthetop-downfiltrationforthe‘9’and‘6’.ThedigitsandoneoftheresultingbarcodesareshowninFigures1and2.Usingallfoursweeps,andboththeBetti0andBetti1barcodes,revealsadditionaldifferencesbetweeneachofthedigits.10Figure1:NoLoopDigitswithBetti0barcode,sweeptorightFigure2:LoopDigitswithBetti1barcode,sweeptotop4.1.2FeatureSelectionWecanusethetechniquesdescribedinthispapertocoordinatizethebar-codespaceB.Inmachinelearningterminology,thesecoordinatesarecalledfeatures.Thisallowsustocharacterizethebarcodesgeneratedbyeachdatapointasacompactfeaturevector.Thisalsogivesusgreatflexibilityinse-lectingfeaturesthatworkwellwithourdata.Wecanthenapplyastandard11machinelearningalgorithm,suchasasupportvectormachine(SVM),toclassifythedata.Weselectedasetoffourfeaturesfromtheinvariantsdiscussedinthispaper.Intuitively,theexponentsineachpolynomialwillgivetherelativevalueofsmallbarsorendpointscomparedtolargebarsorendpoints.Forexample,ifcomparingtwobarsoflengthb2andb,thefirstbarwillhavemoreweightinaninvariantlinearpolynomialthaninaninvariantquadraticpolynomial.Indeed,(cid:18)b2(cid:19)2=b24,(cid:18)b2(cid:19)3=b38,(cid:18)b2(cid:19)4=b416,...Weselectedfourfeatures,Xixi(yi−xi)Xi(ymax−yi)(yi−xi)Xix2i(yi−xi)4Xi(ymax−yi)2(yi−xi)4whichwhenappliedtothefoursweeps,eachwitha0-dimensionaland1-dimensionalbarcode,givesafeaturevectoroftotalsize32whichwethenarrangedintoafeaturematrix.Intuitivelyspeaking,thefirsttwofeaturestakeallofthebars,lengthsandendpoints,intoaccount.Thesecondtwofeaturesheavilyfavorthearrangementoflongerbars.Avisualizationofamatrixof10,000digitsusingclassicalmultidimensionalscaling(MDS)isshowninFigure3andthespectrumofthematrixisshowninFigure4.12−1−0.500.511.5−0.8−0.6−0.4−0.200.20.40.60.81 0123456789(a)A2DViewoftheData−1−0.500.511.5−0.8−0.6−0.4−0.200.20.40.60.81−0.8−0.6−0.4−0.200.20.40.60.8 0123456789(b)A3DViewoftheDataFigure3:VisualizationofDatausingTopologicalFeatures130510152025303500.10.20.30.40.50.60.70.80.91Singular Value IndexNormalized Singular ValueFigure4:NormalizedSpectrumofTopologicalFeatureMatrixAsistypicalwhenusingaSVM,wescaledeachcoordinatesuchthatthevalueswerebetween0and1.TheSVMwasimplementedusingsoftwareprovidedby[6].4.1.3ClassificationResultsWeappliedthesemethodsonasubsetof1000digitsfromtheMNISTdatabasetotuneparametersofthealgorithmandtestvariouskernels.Fortheradialbasisfunctione−γu−v2(RBF,alsoknownastheGaussiankernel),weusedγ=8.Forthepolynomialkernel(γ(u∗v)+a)d,weusedd=3withγ=2anda=2.Inbothfunctions,uandvrepresentthecalculatedfeaturevectors.Afterthis,weprogressivelyincreasedthesizeofthesubsetto10,000handwrittendigits.Theclassificationaccuracywasmeasuredbypartitioningthedatasetintoonehundredsubsetsandusingcross-validationsuccessivelyoneachsubset.TheresultsareshowninTable2.14Table1:ClassificationAccuracyoftwoSVMKernelsSVM1000Digits5000Digits10000DigitsGaussian87.70%91.54%92.04%Polynomial88.00%91.62%92.10%Withthepolynomialkernel,anerrorof7.9%isseen.Asmentionedabove,thepurposeofthistestisnottooutperformexistingclassificationalgorithmsbuttodemonstrateoneapplicationofthetopologicalfeatures.Inlinewiththis,weexaminedsomeofthedigitsthatthealgorithmfailedon.Figure5showsafewofthetypicalproblemdigits.(a)StylisticProblems(b)SpuriousTopologicalChangesFigure5:CommonMisclassificationsThemostcommonconfusionisbetweena‘5’anda‘2’writtenwithnoloop.Otherconfusionsoftenoccurbetweentheshownstyleof‘7’andslanted‘3’sandbetweenacertainstyleof‘4’anda‘9’.Theseconfusionsarenotunexpectedsincethesenumbersaretopologicallythesame.Theextraspatialinformationaddedbythedirectionalsweepsissensitivetovariationsintheslantorstyleofhandwritingandavisualinspectionofthesedigitssuggestswhythealgorithmhasdifficultyclassifyingtheseparticularexamples.Othercommonconfusionsoccurwhentopologicalchangesoccurredtothedigit,specificallywhenthewriteraddsorremovesaloop.154.2HepaticLesionClassificationInthisexample,weapplytopologicalfeaturestoclassifyinghepaticlesions.Thedatasetconsistsofcomputedtomography(CT)scansof132hepaticlesionsthatareoutlinedandannotatedbyradiologists.Thereareninedi-agnosesrepresentedinthedata:cysts(45lesions),metastases(45lesions),hemangiomas(18lesions),hepatocellularcarcinomas(HCC,11lesions),focalnodules(5lesions),abscesses(3lesions),neuroendocrineneoplasms(NeN,3lesions),asinglelacerationandasinglefatdeposit.Additionally,therearenocontrolsforthesizeofthelesionandthelesionsvaryfromunder100pixelsto10,000pixels.Becauseoftheunbalancednatureofthedata,wefocusonthesubsetofcysts,metastases,andhemangiomas.Classificationresultsusingthebarcodemetric(matchingmetric)werefirstpresentedin[2],andwefollowthesamemethodsforprocessingandgeneratingbarcodesfromthedata.Wewillbrieflydescribethemethodshere.Foramoredetailedaccount,pleaseread[2].4.2.1TopologicalMethodsAsmentionedabove,anaturalfiltrationforanimageistofilterbythepixelintensity.AnexampleofthisfiltrationisgiveninFigure6.Thevariationinpixelintensityallowsustouseaone-dimensionalfiltrationonthepixelintensity,butastheresultswillshow,theclassificationisimprovedwhengeometricinformationisaddedintothefiltrations.Asthereisnorotationalorientationofthelesions,wecannotaddingeometricinformationusingthesweepsdescribedintheprevioussection.Instead,weusethelesionborderprovidedbytheradiologistandassigneachpixelitsdistancefromtheborder.Then,byusingtwo-dimensionalhomology,weachieveimprovedresults,especiallyinthecaseoftheheman-giomaswhicharecharacterizedbylargedenseregionsontheouterpartofthelesion.Becausetwo-dimensionalfiltrationsarecomputationallyintensive,weapproximatethetwo-dimensionalfiltrationwithone-dimensionalbarcode‘slices’alongtheborderfiltrationaxis.Weuse7slicesperlesionandboththeBetti0andBetti1barcodes.Notethatwecanlookateachfiltrationfromeachdirectionandcatchdifferentfeatures.Theintensityfiltrationcanaddhighintensitypixelsfirstorlowintensitypixelsfirst.Theboundaryfiltrationcanbeginwithpixelsneartheboundaryfirstorpixelsfarfromtheboundaryfirst.Thisyields5616one-dimensionalbarcodesperlesion.121111111111111232222313222223222233222121111111111111232222313222333121111111111111232222313222121111111111111232222313222121111111111111232222313222(a)Simpleimagewithfilteredcomplex(b)β0barcodeforaboveimage(c)β1barcodeforaboveimageFigure6:Constructinganincreasing1D-filtrationonanimage[2]4.2.2FeatureSelectionWeuseaslightlydifferentsetoffourfeaturesascomparedtothedigitsexample.Thesefeaturesareshownbelow.Thetwosetsoffeaturesthatfocusonlongbarsandfeatureswhichtakeintoaccountshorterbarsisusedhere.Inthisapplication,thisisanalogoustofilteringthebarcodetoremovethelargenumberofsmallerbars.Becauseofthevariationsinlesionsize,welookattheaverageovereachbartotryandeliminatetheeffectsoflargevariationsinlesionsize.17nXixi(yi−xi)/nnXi(ymax−yi)(yi−xi)/nnXix2i(yi−xi)4/nnXi(ymax−yi)2(yi−xi)4/nAsmentionedabove,wehave56barcodesperlesion.Withfourfeatures,thisyieldsafeaturevectorof224featuresforeachlesion.4.2.3ClassificationResultsWeapplytheSVMusingonlytheGaussiankernelanduseanexponentialparametersweeptofindoptimalvaluesofγforeachmethod.WeuseLOOCVtocalculatetheclassificationaccuracies.Theresultsareshownbelow.Table2givestheresultsfor1Dand2DfiltrationsforseveraldifferentdatasetswhileTable3showshowwellthealgorithmperformsondifferentlesiontypesforthedifferentfiltrations.Table4demonstratestheeffectofsizeonclassification.Table2:SVMClassificationAccuraciesfor1Dand2DFiltrationsFiltrationFullHcHeCMHeCMCM1D(Intensity)53.03%59.66%65.74%75.56%2D67.42%74.79%81.48%86.67%Using[2],weseethatthattopologicalfeaturesarecomparablewithusingthematchingmetrictogeneratefeatures.TheresultsfromtheHeCMdatasetforthetwomethodsareshownbelow.Theyreflectthecorrectclassificationofasinglelesionusingathetopologicalfeatures,makingthetwomethods18Table3:HeCM%ClassificationAccuracybyLesionTypeFiltration%ofHeCM%ofHeman.%ofCysts%ofMetas.1D65.74%33.33%75.56%68.89%2D81.48%61.11%86.67%84.44%Table4:ClassificationbyLesionSizeofHeCMLesionSizebyArea%Accu.#ofHeman.#ofCysts#ofMetas.All81.48%184545<10000px82.52%184243<5000px84.78%163937<2500px86.25%143234<1250px88.514%82823virtuallythesameforthissubsetofthedata.Comparingwiththeotherresultsin[2]showsthatthetworesultsareverycloseinmostcategories,witheachslightlyoutperformingtheotherincertainsubsetsofthedata.Table5:ClassificationMethodsFiltrationBarcodeFeaturesMatchingMetric1D65.74%63.80%2D81.48%80.56%4.3DiscussionThesetwoexamplesdemonstratetheclassifyingpoweroftopologicalfea-tureswhenappliedtorealworlddatasets.Thiswasdoneusingoff-the-shelfmachinelearningalgorithmsshowingthatthesefeaturescaneasilybecom-binedwithmoretraditionalclassificationmethodsaddingasetofadditionalclassificationfeaturestothemachinelearningtoolbox.19Theseexamplesalsoshowthepowerofcombiningtopologywithgeome-try.Inbothdatasets,thisisanintegralpartoftheclassificationprocedure.Theresultsinthehepaticlesiondatasetprovideanespeciallygoodexampleofthepotentialgainsthatcanbeachievedbycombiningbothfields.Insummary,usingalgebraicgeometryandinvarianttheory,wehaveiden-tifiedafamilyofcoordinatesonthespaceoffinitemetricspaces,orsampledshapes.Thesecoordinatescanserveasamethodfororganizingthecollec-tionofallbarcodes,andthereforeanydatabasewhosemembersproducebarcodes.Ofcourse,wecanalsousevariousmetricsonbarcodespace,suchasthebottleneckorWassersteindistances.Itwouldbeextremelyinterestingtoanalyzetherelationshipbetweenthesedistancesonbarcodespaceswithvariousmorealgebraicnotionsofdistanceonthebarcodecoordinates.Itwouldalsobeveryinterestingtodefineandanalyzeanalogouscoordinatesonspacesofmultidimensionalpersistencemodules,wheretheymightgiveinformationwhichiscurrentlynotaccessibleduetothecomplexityofthealgebraicdescriptionsofmultidimensionalpersistencemodules.References[1]H.AdamsandG.Carlsson,Onthenon-linearstatisticsofrangeimagepatches,SIAMJ.ImagingSci.,(2),1,p.110-117.[2]A.Adcock,D.Rubin,andG.Carlsson,ClassificationofHepaticLesionsusingtheMatchingMetric,arXivpreprintarXiv:1210.0866(2012).[3]G.Carlsson,Topologyanddata,Bull.Amer.Math.Soc.(46),2,2009,pp.255-308.[4]G.Carlsson,T.Ishkhanov,V.deSilva,andA.Zomorodian,Onthelocalbehaviorofspacesofnaturalimages,toappear,InternationalJournalofComputerVision,(76),1,2008,pp.1-12.[5]G.Carlsson,A.Zomorodian,A.Collins,andL.Guibas,Persistencebar-codesforshapes,InternationalJournalofShapeModeling,(11),(2005),pp.149-187[6]Chih-ChungChangandChih-JenLin,LIBSVM:alibraryforsupportvectormachines,ACMTransactionsonIntelligentSys-20temsandTechnology,2:27:1–27:27,2011.Softwareavailableathttp://www.csie.ntu.edu.tw/∼cjlin/libsvm[7]F.Chazal,D.Cohen-Steiner,L.Guibas,F.M´emoli,andS.Oudot,Gromov-Hausdorffstablesignaturesforshapeusingpersistence,Com-puterGraphicsForum,(28),5,2009,pp.1393-1403.[8]A.Collins,A.Zomorodian,G.Carlsson,andL.Guibas,Abarcodeshapedescriptorforcurvepointclouddata,ComputersandGraphics,Volume28,2004,pp.881-894.[9]J.Dalbec,Multisymmetricfunctions,BeitrageAlgebraGeom.,(40),1,1999,pp.27-51.[10]W.FultonandR.Weiss,Algebraiccurves:Anintroductiontoalgebraicgeometry,Addison-Wesley,1989.[11]Y.LeCunandC.Cortes,TheMNISTDatabase,CourantInstituteNYU,Accessed16July2012,http://yann.lecun.com/exdb/mnist/[12]D.Mumford,J.Fogarty,andF.Kirwan,GeometricInvariantThe-ory,SpringerVerlag,2002.[13]A.ZomorodianandG.Carlsson,Computingpersistenthomology,Dis-creteandComputationalGeometry,33(2),2005,pp.247-274.21TheRingofAlgebraicFunctionsonPersistenceBarCodesAaronAdcockErikCarlssonGunnarCarlsson∗May10,20141IntroductionPersistenthomology([3],[13])isafundamentaltoolintheareaofcompu-tationaltopology.Itcanbeusedtoinfertopologicalstructureindatasets(see[1],[4]),butvariationsonthemethodcanbeappliedtostudyaspectsoftheshapeofpointcloudswhicharenotovertlytopological([5],[8]).Themethodologyassignstoanyfinitemetricspace(suchasaretypicallyobtainedinexperimentaldataofvariouskinds)andnon-negativeintegerkabarcode,bywhichwewillmeanafinitecollectionofintervalswithendpointsontherealline.Theintegerkspecifiesadimensionofafeature(zero-dimensionalforacluster,one-dimensionalforaloop,etc.),andanintervalrepresentsafeaturewhichis“born”atthevalueofaparameter(thepersistencepa-rameter)givenbythelefthandendpointoftheinterval,andwhich“dies”atthevaluegivenbytherighthandendpoint.Thesebarcodeshavebeendemonstratedtoidentifystructureinspacesofimagepatchesin[1]and[4],andhavebeendemonstratedtodistinguishbetweenhanddrawnlettersin[8].Becauseoftheunusualstructureoftheinvariant,i.e.asacollectionofintervalsratherthannumericalquantities,themethodcurrentlyrequiressubstantialknowledgeoftopologicalmethods.Itwouldclearlybeusefultoassignandinterpretvariousnumericalquantitiesattachedtobarcodes,sothattheseoutputscouldbeusedasinputtostandardalgorithmswithinma-chinelearning,clusteranalysis,andothermethods.Itisthepurposeofthis∗ResearchsupportedinpartbyNSFDMS-04069921arXiv:1304.0530v1 [math.RA] 2 Apr 2013papertoidentifyanalgebraoffunctionsonthesetofbarcodeswhichisdefinedinaconceptuallycoherentway.Themainideaisthefollowing.Abarcodewithexactlynintervalscanbespecifiedbyavector(x1,y1,x2,y2,...,xn,yn),wherexidenotestheleftendpointofthei-thintervalandyitherightendpoint.However,thisrepresentationismanytoone,inthatthebarcodestructuredoesnotretaintheorderingontheintervals.Infact,thesetofbarcodeswithexactlynintervalscanbeidentifiedwiththesetSpn(R2)then-foldsymmetricproductofR2.ForanysetX,Spn(X)isdefinedtobetheorbitspaceoftheactionofthesymmetricgrouponnlettersontheproductXngivenbypermutingthecoordinates.Ontheotherhand,thespace(R2)nisanalgebraicvarietyoverR([10]).Infact,itisanaffinespaceofdimension2n,andthesymmetricgroupactionmentionedaboveisanalgebraicaction.Itisthenknown(see[12])thattheorbitspaceinheritsthestructureofanalgebraicvariety,andtheelementsofitsaffinecoordinatering([10])arefunctionsonthesetofbarcodeswithexactlynintervals.Theseaffinecoordinateringsarewellknownalgebrasreferredtogenericallyasringsofmultisymmetricpolynomials([9]).Theycanbequitecomplicated,sinceitturnsoutthatanysetofalgebrageneratorsforthemwillsatisfynon-trivialrelationsorsyzygies.Itturnsout,though,thatthereareinclusionsofalgebraicvarietiesSpn(R2)→Spn+1(R2)(1)whichproduceaninversesystemofaffinecoordinaterings···→A[Spn+1(R2)]→A[Spn(R2)]→···whoseinverselimitwewilldenote,byabuseofnotation,byA[Sp∞(R2)].ThenotationA[−]denotestheaffinecoordinatering.Thisalgebraisknowntobefreelygeneratedonasetofminimalalgebragenerators([9]).Theanalysisofthesystem(1)aboveisnotsufficient,though.Thissystemidentifiesapoint((x1,y1),...,(xn,yn))∈Spn(R2)withthepoint((x1,y1),...,(xn,yn),(0,0))∈Spn+1(R2)Inotherwords,asetSofnintervalsisidentifiedwiththesetofn+1intervalsobtainedbyadjoiningtheintervaloflengthzerowhosetwoend-pointsarezero.However,intheparametrizationoftheisomorphismclasses2ofpersistencevectorspacesin[13]bybarcodes,anyintervaloflengthzeroisidentifiedwiththezeromodule.So,wewouldliketodeterminetheringofallalgebraicfunctions(i.e.theelementsofA[Sp∞(R2)])whichhavethepropertythattheytakethesamevalueonanybarcodeasontheresultofadjoininganyintervaloflengthzerotoit.Inthispaper,wewillidentifythissubring,describeitsstructure,anddescribethealgebrageneratorsex-plicitlysothattheycanbeusedeffectivelybythoseinterestedinanalyzingdatabasesofshapes.2TheInd-schemeBWefirstdiscussthesetofbarcodes,withoutanyalgebraicvarietystructures.Foreveryn,wefirstconsiderthesetofbarcodescontainingexactlyninter-vals.Wewillpermitintervalsoflengthzero.ThesetofintervalsIcanbeidentifiedwiththesubsetofR2consistingofpairs(x,y)withx≤y.Abarcodecontainingnintervalsisthereforeidentifiedwiththen-foldsymmetricproductSpn(I),whereforanysetX,Spn(X)isdefinedtobetheorbitspaceoftheactionofthesymmetricgrouponnlettersontheproductXngivenbypermutingthecoordinates.OnecanassemblethesesetsintoadirectedsystemIi1→Sp2(I)i2→Sp3(I)i3→Sp4(I)→···wherethemapsin:Spn(I)→Spn+1(I)aregivenbyin({I1,...,In})={I1,...,In,[0,0]}ThedirectlimitofthissystemwillbedenotedbySp∞(I).WeareinterestedinstudyingfunctionsonSp∞(I).Suchafunctioncanbeidentifiedwithaninfinitevector(f1,f2,f3,...)offunctionsfn:Spn(I)→Rsatisfyingthecompatibilityconditionfn+1·in=fnThesetofallsuchvectorsoffunctionsformsaringRundercoordinate-wiseadditionandmultiplication.Itisnotexactlywhatwewant,however.Thereasonisthatundertheparametrizationofpersistencevectorspacesasdescribedin[3]and[13],intervalsoflengthzerocorrespondtozerovectorspaces,andthereforeallintervalsoflengthzeroshouldbeconsideredequal.ThismeansthatweshouldconsideronlyfunctionsF:Sp∞(I)→Rfor3whichF({I1,I2,...,In,[ξ,ξ]})=F({I1,I2,...,In,[η,η]})forallpossiblevaluesofξandη.ThesetofallsuchfunctionsisasubringR0⊆R.ThissetoffunctionscanbedefinedasthesetofallfunctionsonthesetBdefinedbyB=anSpn(I)/’where’istheequivalencerelationgeneratedbyallrelationsoftheform{I1,I2,In,[ξ,ξ]}’{I1,I2,...,In}.Remark:ThereadermaysuggestthatoneconsiderinsteadonlythesubsetI+consistingofintervalsofpositivelength.Thiswillproduceadisjointunionofsetsofbarcodes,partitionedintothesetscontainingafixedpositivenumberofintervalsofpositivelength.Suchadescriptiondoesnottakeintoaccountthefactthatwewouldliketotopologizethespaceofallbarcodesinsuchawaythatlim→0{I1,I2,...In,[xn+1,xn+1+]}={I1,I2,...,In}Thereasonforthisisthatsmallperturbationstotheinputdatatothepersis-tencealgorithmscanmodifythebarcodesbymodifyinglengthsofintervalsasmallamountandaddintervalsofsmalllength.Thisisthestabilitytheoremforpersistencediagramsprovedin[7].TheringoffunctionsR0istoolargetodealwitheffectively.EventhemuchsmallerringofcontinuousfunctionsonBisstilltoocomplextodescribecompletely.WewillobservethatBisdescribedasacolimitofalgebraicvarieties,andthatitisthereforepossibletodefinetheringofalgebraicfunctionsonB.Itisthisringwewillanalyze.Throughoutthispaper,kwilldenotethefieldR.Allvarietieswillbeoverk.WeconsidertheaffinespaceAn=A(n)ofdimension2n,parametrizedwithcoordinates(x1,y1,x2,y2,...,xn,yn).ItsaffinecoordinateringisthepolynomialringBn=k[x1,y1,...,xn,yn].Thereisanactionofthesym-metricgroupSnonnlettersonAn,andfrom[12]itfollowsthatthesetoforbitsonthesetofpointsofthevarietyisitselfanaffinealgebraicvariety,withaffinecoordinateringequaltotheinvariantsubringBSnn.LetWi⊆Andenotethesubvarietyyi−xi=0.WeletDn⊆BndenotethesubringoffunctionswhoserestrictiontoWiisindependentofxiforalli.Wewishtocharacterizethissubringalgebraically.4Proposition1.TheringDnischaracterizedalgebraicallyasthesubringofallfforwhich(∂∂xi+∂∂yi)f∈(yi−xi)foralli.Proof:Wefixi,andconsiderallthefunctionsfforwhichfWiisindepen-dentofxi(andthereforeyi).Theoperator∂∂xi+∂∂yiinducesadifferentialoperatoronthequotientringQn=Bn/(yi−xi),whichisidentifiedwiththepartialdifferentialoperator2∂∂xiinQn∼=k[x1,y1,...,yi−1,xi,xi+1,...,xn,yn]TherequirementisthattheimagefoffinWiisindependentofxi,andthisisequivalenttothecondition∂∂xi(f)=0.Thisconditionistoholdforeachi,whichgivestheresult.3TheringofalgebraicfunctionsonBWebeginbychangingcoordinatesviatheformulaeξi=xi+yiandηi=yi−xi.ItisclearthatBncanalsobeidentifiedwithk[ξ1,η1,...,ξn,ηn],andthatthesymmetricgroupinthenewcoordinatesystempermutestheξi’sandηi’s.Underthistransformation,theoperator∂∂xi+∂∂yiiscarriedintotheoperator2∂∂ξi.ThismeansthattheringDnisidentifiedwiththesubringoffunctionsf(ξ1,η1,...,ξn,ηn)forwhich∂f∂ξi∈(ηi)foralli.Proposition2.Ak-basisfortheDnisgivenbythesetofmonomialsξa11ξa22···ξannηb11ηb22···ηbnnforwhichai>0impliesbi>0.Proof:Wenotethattheoperator∂/∂ξicarrieseachmonomialtoaconstantmultipleofasinglemonomial,namelythemonomialobtainedbydecreasingaibyone.Moreover,containmentintheideal(ηi)isalsogivenpurelybyconditionsonmonomials,i.e.thatbi>0.WeconcludethatDnisspannedbymonomialslyinginDn.ButitisclearthatamonomialµliesinDnexactlyifitisthecasethatwheneverξidividesµ,thenηialsodividesµ.5Thiscorrespondstotheabovenumericalconditionontheexponentsinthemonomial.ThesymmetricgroupactionclearlypreservesthesubringDn.Moreover,itpreservesthebasisofmonomialswithinDn.Let{µα}α∈AdenoteasetoforbitrepresentativesoftheSn-actiononthesetofmonomialsdefinedinProposition2.Letσαdenotethesumofalltheelementsintheorbitofµα.Proposition3.WeletDSnndenotethesubringofelementsofDnwhichareinvariantundertheactionofSn.Thentheelementsσαformak-basisofDSnn.Proof:ThisresultplainlyholdsforanyalgebraoverafieldofcharacteristiczeroonwhichthereisaG-actionwhichpreservesabasisofmonomials.Wehaverestrictionmapsπn,m:Dn→Dm,whenn≥m,definedbyπn,m(ξi(respηi))=ξi(respηi)fori≤m,andπn,m(ξi)=0fori>m.Themapπn,misSm-equivariant,whereSmactsbypermutingthefirstmpairsofvariables.ItfollowsthatwemayconstructcompositesDSnn,→DSmnπSmn,m→DSmmwhichwedenotebyσn,m,andthereforetheinversesystem···σn+1,n−→DSnnσn,n−1−→DSn−1n−1σn−1,n−2−→···σ2,1−→D1WewilldenotetheinverselimitofthissystembyD.Wenextrecallsomeofthenotationandbasicfactsaboutmultisymmetricpolynomials,whichcanbefoundinDalbec[9].LetRn,rbethepolynomialringinnrvariables,Rn,r=k[xi,j;1≤i≤n,1≤j≤r]WeletthesymmetricgroupSnactonRn,rviatheformulaσ(xij)=xσ(i)j,andletΛn,r=RSnn,r,denotetheringofSninvariants.ThereisaninversesystemparalleltotheoneconstructedaboveinvolvingtheringsΛn,r.Wehaveevaluationmapsπn,m:Rn,r→Rm,r,m≤n6definedbysettingxir=0ifi>m.Themapπn,misSm-equivariant,whenSm⊆Snisthesubgroupofpermutationsofthefirstmelementsoftheset{1,...,m}.WehavethecompositesΛn,r=RSnn,r,→RSmn,rπSmn,m→RSmm,r=Λm,rwhichwedenotebyρn,m.Theinverselimitofthesystem···ρn+1,n−→Λn,rρn,n−1−→Λn−1,rρn−1,n−2−→···ρ2,1−→Λ1,rwillbedenotedbyΛr,andreferredtoastheringofr-multisymmetricfunc-tions.IthasagradingΛr=MkΛkrinducedbythegradingonRn,r.ThereisanevidentembeddingD,→Λ2.WewillusethisembeddingtoidentifythestructureofD.Theringofmultisymmetricfunctionshasseveralinterestingsetsofgen-erators.Givenanarrayofnonnegativeintegersa11a12···a1ra21a22···a2r............ak1ak2···akrwithk≤nandai=(ai1,...,air),wedefinethemultisymmetricmonomialsbyma1,...,ak=Sym(xa1111···xakrkr)=Xσ∈SnYi,jxaijσ(i)j∈Λr,SymappliedtoamonomialyieldsthesumofallmonomialswhichareintheorbitoftheSn-action.TheyformavectorspacebasisofΛn,r,foranyn.ItisknownthatΛn,risgeneratedasanalgebrabythesymmetrizationsofmonomialsinvolvingonly{x11,x12,...,x1r}.Theyaregivenbytheformulaepa=ma=Xixa1i1···xarir,andarecalledthemultisymmetricpowersums.Whiletherearerelationsamongthepowersumsinfinitelymanyvariables,theyfreelygeneratetheinverselimitΛr,makingitapolynomialalgebra.See[9]fordetails.7Ourinterestisinthecaser=2.Letussetξi=xi1,ηi=xi2,andasbeforeBn=Rn,2=k[ξ1,η1,...,ξn,ηn].ThesubalgebraD⊆Λ2nowhasthefollowingcharacterization.Theorem1.AsasubalgebraofΛ2,Disfreelygeneratedbytheset∆ofelementsoftheformpa,bwhereb≥1.Proof:Wefirstconsiderthesubalgebrak[∆]⊆Dgeneratedby∆.Because∆isasubsetofthefreegeneratingsetofΛ2,itisclearthatthecompositek[∆]→D,→Λ2isinjectiveandisomorphicontoapolynomialsubalgebra,andthereforethatk[∆]isitselfapolynomialsubalgebraofD.Inordertoprovethatk[∆]=D,weonlyneedtocountdimensions,andweformulatethecountingintermsoftheHilbertseries.Recallthatforagradedk-vectorspaceV∗,wehavetheHilbertseriesP(V∗)=Xidimk(Vi)tiTheHilbertseriesforapolynomialalgebraonasinglegeneratorxofgradingiis(1−ti)−1.Moreover,ifwearegiventwogradedvectorspacesV∗andW∗,thenP(V∗⊗W∗)=P(V∗)P(W∗).Sincetherearenmonomialsofdegreenin∆,wefindthattheHilbertseriesfork[∆]isP(k[∆])=Yn(1−tn)−nIfwecanshowthattheHilbertseriesforDisequaltothisseries,theproofwillbecomplete.InProposition3,wefoundthatak-basisforDmaybeidentifiedwithasetoforbitrepresentativesoftheSn-actiononthesetofallmonomialswhichhavethepropertythatiftheexponentofξiisnon-zero,thensoistheexponentofηi.Suchasetofrepresentativesisgivenbythesetofmonomialsoftheformξa11ηb11···ξallηbll,ϕ−1(ai,bi)≥ϕ−1(ai+1,bi+1)8wherel≤n,andϕ:N+→N+×Nisthebijection(ϕ1,ϕ2,...)=((1,0),(1,1),(2,0),(1,2),(2,1),(3,0),(1,3),...)ontothesetofpossiblenonzeroexponents.Thedimensionofthek-gradedcomponentofBSnnisjustthenumberofthesemonomialsofdegreek.Letussaythat(a,b)≤(c,d)whenϕ−1(a,b)≤ϕ−1(c,d),andletf(a,b,k)denotethenumberofsequences(a1,b1,...,al,bl)suchthat(a,b)≥(a1,b1)≥···≥(al,bl),(ai,bi)∈N+×NandlX1(ai+bi)=kwithnorestrictionsonl.Itiseasytocheckthatitsatisfiestherecursionrelationf(a,b,k)=X(c,d)≤(a,b)f(c,d,k−c−d),Thiscorrespondstoaruleforthegeneratingfunctionfa,b(t)=Pkf(a,b,k)tk,fa,b(t)=1+X(c,d)≤(a,b)tc+dfc,d(t)=1+X(c,d)<(a,b)tc+dfc,d(t)+ta+bfa,b(t)Solvingforfa,b(t)gives(1−ta+b)fa,b(t)=1+X(c,d)<(a,b)fc,d(t)=fa0,b0(t),where(a0,b0)istheelementimmediatelybelow(a,b)underϕ.Itisreadilyverifiedthattheformulaga,b(t)=(1−ta+b)−aY1≤k≤a+b−1(1−tk)−k,satisfiesthesamerecursionrelation,andthereforethatfa,b(t)=ga,b(t).Tak-inglimitslimn→∞P(BSnn)=lim(a,b)→∞Xk≥1f(a,b,k)tk=Yk≥1(1−tk)−kgivestheresult.94MachineLearningonBwithexamples4.1DigitsExampleToillustratetheclassificationpotentialofthistechnique,weapplyittotheMNISTdatabase[11],ofhandwrittendigits.Weemphasizethattheaimisnottooutperformexistingmachinelearningalgorithmsfordigitclassifica-tion,buttopresentanexamplethatdemonstratesonewayofcombiningthistechniquewithexistingmachinelearningtechniques.Whileitisclearthatpuretopologicalclassificationcannotdistinguishbetweenthedigits(therearethreenumbersthatdonothaveanyloops,threethatalwayshaveloops,onethathastwoloopsandthreethathavestyle-dependentloops),wecanusethepowerofpersistenthomologytosiftoutmoreinformation.WebeginbyshowingthefullanalysisofafewdigitsandthengivetheempiricalresultsofapplyingthistechniquetoasubsetoftheMNISTdatabase.4.1.1TopologicalMethodsWebeginbydescribingaparticulargraphconstructiongivenadigitalimage.Wetreatthepixelsasverticesandaddedgesbetweenadjacentpixels(includ-ingdiagonals).Wecannowdefineafiltrationontheverticesofthegraphcorrespondingtotheimagepixels.Anaturalfiltrationcouldbeconstructedusingthepixelintensitiesoftheoriginalimage(seeFigure6,Section4.2).Anotherfiltration,usedin[8],canbeconstructedbythresholding,topro-duceabinaryimage,andadding1-pixelsaswesweepacrosstheimage.Thisaddsspatialinformationintowhatwouldotherwisebeapurelytopologicalmeasurement.Sincetheorientationofthedigitmatters(a6isthesameasa9givena180degreerotation),wechoosethelatterapproachandsweepacrosstherowsandcolumnsofeachdigit.Bytakingintoaccountspatialinformation,wegetaroughviewofthelocationofvarioustopologicalfeatures.Forexample,thougha‘9’and‘6’bothhaveoneconnectedcomponentandasingleloop,theloopwillappearatdifferentlocationsinthetop-downfiltrationforthe‘9’and‘6’.ThedigitsandoneoftheresultingbarcodesareshowninFigures1and2.Usingallfoursweeps,andboththeBetti0andBetti1barcodes,revealsadditionaldifferencesbetweeneachofthedigits.10Figure1:NoLoopDigitswithBetti0barcode,sweeptorightFigure2:LoopDigitswithBetti1barcode,sweeptotop4.1.2FeatureSelectionWecanusethetechniquesdescribedinthispapertocoordinatizethebar-codespaceB.Inmachinelearningterminology,thesecoordinatesarecalledfeatures.Thisallowsustocharacterizethebarcodesgeneratedbyeachdatapointasacompactfeaturevector.Thisalsogivesusgreatflexibilityinse-lectingfeaturesthatworkwellwithourdata.Wecanthenapplyastandard11machinelearningalgorithm,suchasasupportvectormachine(SVM),toclassifythedata.Weselectedasetoffourfeaturesfromtheinvariantsdiscussedinthispaper.Intuitively,theexponentsineachpolynomialwillgivetherelativevalueofsmallbarsorendpointscomparedtolargebarsorendpoints.Forexample,ifcomparingtwobarsoflengthb2andb,thefirstbarwillhavemoreweightinaninvariantlinearpolynomialthaninaninvariantquadraticpolynomial.Indeed,(cid:18)b2(cid:19)2=b24,(cid:18)b2(cid:19)3=b38,(cid:18)b2(cid:19)4=b416,...Weselectedfourfeatures,Xixi(yi−xi)Xi(ymax−yi)(yi−xi)Xix2i(yi−xi)4Xi(ymax−yi)2(yi−xi)4whichwhenappliedtothefoursweeps,eachwitha0-dimensionaland1-dimensionalbarcode,givesafeaturevectoroftotalsize32whichwethenarrangedintoafeaturematrix.Intuitivelyspeaking,thefirsttwofeaturestakeallofthebars,lengthsandendpoints,intoaccount.Thesecondtwofeaturesheavilyfavorthearrangementoflongerbars.Avisualizationofamatrixof10,000digitsusingclassicalmultidimensionalscaling(MDS)isshowninFigure3andthespectrumofthematrixisshowninFigure4.12−1−0.500.511.5−0.8−0.6−0.4−0.200.20.40.60.81 0123456789(a)A2DViewoftheData−1−0.500.511.5−0.8−0.6−0.4−0.200.20.40.60.81−0.8−0.6−0.4−0.200.20.40.60.8 0123456789(b)A3DViewoftheDataFigure3:VisualizationofDatausingTopologicalFeatures130510152025303500.10.20.30.40.50.60.70.80.91Singular Value IndexNormalized Singular ValueFigure4:NormalizedSpectrumofTopologicalFeatureMatrixAsistypicalwhenusingaSVM,wescaledeachcoordinatesuchthatthevalueswerebetween0and1.TheSVMwasimplementedusingsoftwareprovidedby[6].4.1.3ClassificationResultsWeappliedthesemethodsonasubsetof1000digitsfromtheMNISTdatabasetotuneparametersofthealgorithmandtestvariouskernels.Fortheradialbasisfunctione−γu−v2(RBF,alsoknownastheGaussiankernel),weusedγ=8.Forthepolynomialkernel(γ(u∗v)+a)d,weusedd=3withγ=2anda=2.Inbothfunctions,uandvrepresentthecalculatedfeaturevectors.Afterthis,weprogressivelyincreasedthesizeofthesubsetto10,000handwrittendigits.Theclassificationaccuracywasmeasuredbypartitioningthedatasetintoonehundredsubsetsandusingcross-validationsuccessivelyoneachsubset.TheresultsareshowninTable2.14Table1:ClassificationAccuracyoftwoSVMKernelsSVM1000Digits5000Digits10000DigitsGaussian87.70%91.54%92.04%Polynomial88.00%91.62%92.10%Withthepolynomialkernel,anerrorof7.9%isseen.Asmentionedabove,thepurposeofthistestisnottooutperformexistingclassificationalgorithmsbuttodemonstrateoneapplicationofthetopologicalfeatures.Inlinewiththis,weexaminedsomeofthedigitsthatthealgorithmfailedon.Figure5showsafewofthetypicalproblemdigits.(a)StylisticProblems(b)SpuriousTopologicalChangesFigure5:CommonMisclassificationsThemostcommonconfusionisbetweena‘5’anda‘2’writtenwithnoloop.Otherconfusionsoftenoccurbetweentheshownstyleof‘7’andslanted‘3’sandbetweenacertainstyleof‘4’anda‘9’.Theseconfusionsarenotunexpectedsincethesenumbersaretopologicallythesame.Theextraspatialinformationaddedbythedirectionalsweepsissensitivetovariationsintheslantorstyleofhandwritingandavisualinspectionofthesedigitssuggestswhythealgorithmhasdifficultyclassifyingtheseparticularexamples.Othercommonconfusionsoccurwhentopologicalchangesoccurredtothedigit,specificallywhenthewriteraddsorremovesaloop.154.2HepaticLesionClassificationInthisexample,weapplytopologicalfeaturestoclassifyinghepaticlesions.Thedatasetconsistsofcomputedtomography(CT)scansof132hepaticlesionsthatareoutlinedandannotatedbyradiologists.Thereareninedi-agnosesrepresentedinthedata:cysts(45lesions),metastases(45lesions),hemangiomas(18lesions),hepatocellularcarcinomas(HCC,11lesions),focalnodules(5lesions),abscesses(3lesions),neuroendocrineneoplasms(NeN,3lesions),asinglelacerationandasinglefatdeposit.Additionally,therearenocontrolsforthesizeofthelesionandthelesionsvaryfromunder100pixelsto10,000pixels.Becauseoftheunbalancednatureofthedata,wefocusonthesubsetofcysts,metastases,andhemangiomas.Classificationresultsusingthebarcodemetric(matchingmetric)werefirstpresentedin[2],andwefollowthesamemethodsforprocessingandgeneratingbarcodesfromthedata.Wewillbrieflydescribethemethodshere.Foramoredetailedaccount,pleaseread[2].4.2.1TopologicalMethodsAsmentionedabove,anaturalfiltrationforanimageistofilterbythepixelintensity.AnexampleofthisfiltrationisgiveninFigure6.Thevariationinpixelintensityallowsustouseaone-dimensionalfiltrationonthepixelintensity,butastheresultswillshow,theclassificationisimprovedwhengeometricinformationisaddedintothefiltrations.Asthereisnorotationalorientationofthelesions,wecannotaddingeometricinformationusingthesweepsdescribedintheprevioussection.Instead,weusethelesionborderprovidedbytheradiologistandassigneachpixelitsdistancefromtheborder.Then,byusingtwo-dimensionalhomology,weachieveimprovedresults,especiallyinthecaseoftheheman-giomaswhicharecharacterizedbylargedenseregionsontheouterpartofthelesion.Becausetwo-dimensionalfiltrationsarecomputationallyintensive,weapproximatethetwo-dimensionalfiltrationwithone-dimensionalbarcode‘slices’alongtheborderfiltrationaxis.Weuse7slicesperlesionandboththeBetti0andBetti1barcodes.Notethatwecanlookateachfiltrationfromeachdirectionandcatchdifferentfeatures.Theintensityfiltrationcanaddhighintensitypixelsfirstorlowintensitypixelsfirst.Theboundaryfiltrationcanbeginwithpixelsneartheboundaryfirstorpixelsfarfromtheboundaryfirst.Thisyields5616one-dimensionalbarcodesperlesion.121111111111111232222313222223222233222121111111111111232222313222333121111111111111232222313222121111111111111232222313222121111111111111232222313222(a)Simpleimagewithfilteredcomplex(b)β0barcodeforaboveimage(c)β1barcodeforaboveimageFigure6:Constructinganincreasing1D-filtrationonanimage[2]4.2.2FeatureSelectionWeuseaslightlydifferentsetoffourfeaturesascomparedtothedigitsexample.Thesefeaturesareshownbelow.Thetwosetsoffeaturesthatfocusonlongbarsandfeatureswhichtakeintoaccountshorterbarsisusedhere.Inthisapplication,thisisanalogoustofilteringthebarcodetoremovethelargenumberofsmallerbars.Becauseofthevariationsinlesionsize,welookattheaverageovereachbartotryandeliminatetheeffectsoflargevariationsinlesionsize.17nXixi(yi−xi)/nnXi(ymax−yi)(yi−xi)/nnXix2i(yi−xi)4/nnXi(ymax−yi)2(yi−xi)4/nAsmentionedabove,wehave56barcodesperlesion.Withfourfeatures,thisyieldsafeaturevectorof224featuresforeachlesion.4.2.3ClassificationResultsWeapplytheSVMusingonlytheGaussiankernelanduseanexponentialparametersweeptofindoptimalvaluesofγforeachmethod.WeuseLOOCVtocalculatetheclassificationaccuracies.Theresultsareshownbelow.Table2givestheresultsfor1Dand2DfiltrationsforseveraldifferentdatasetswhileTable3showshowwellthealgorithmperformsondifferentlesiontypesforthedifferentfiltrations.Table4demonstratestheeffectofsizeonclassification.Table2:SVMClassificationAccuraciesfor1Dand2DFiltrationsFiltrationFullHcHeCMHeCMCM1D(Intensity)53.03%59.66%65.74%75.56%2D67.42%74.79%81.48%86.67%Using[2],weseethatthattopologicalfeaturesarecomparablewithusingthematchingmetrictogeneratefeatures.TheresultsfromtheHeCMdatasetforthetwomethodsareshownbelow.Theyreflectthecorrectclassificationofasinglelesionusingathetopologicalfeatures,makingthetwomethods18Table3:HeCM%ClassificationAccuracybyLesionTypeFiltration%ofHeCM%ofHeman.%ofCysts%ofMetas.1D65.74%33.33%75.56%68.89%2D81.48%61.11%86.67%84.44%Table4:ClassificationbyLesionSizeofHeCMLesionSizebyArea%Accu.#ofHeman.#ofCysts#ofMetas.All81.48%184545<10000px82.52%184243<5000px84.78%163937<2500px86.25%143234<1250px88.514%82823virtuallythesameforthissubsetofthedata.Comparingwiththeotherresultsin[2]showsthatthetworesultsareverycloseinmostcategories,witheachslightlyoutperformingtheotherincertainsubsetsofthedata.Table5:ClassificationMethodsFiltrationBarcodeFeaturesMatchingMetric1D65.74%63.80%2D81.48%80.56%4.3DiscussionThesetwoexamplesdemonstratetheclassifyingpoweroftopologicalfea-tureswhenappliedtorealworlddatasets.Thiswasdoneusingoff-the-shelfmachinelearningalgorithmsshowingthatthesefeaturescaneasilybecom-binedwithmoretraditionalclassificationmethodsaddingasetofadditionalclassificationfeaturestothemachinelearningtoolbox.19Theseexamplesalsoshowthepowerofcombiningtopologywithgeome-try.Inbothdatasets,thisisanintegralpartoftheclassificationprocedure.Theresultsinthehepaticlesiondatasetprovideanespeciallygoodexampleofthepotentialgainsthatcanbeachievedbycombiningbothfields.Insummary,usingalgebraicgeometryandinvarianttheory,wehaveiden-tifiedafamilyofcoordinatesonthespaceoffinitemetricspaces,orsampledshapes.Thesecoordinatescanserveasamethodfororganizingthecollec-tionofallbarcodes,andthereforeanydatabasewhosemembersproducebarcodes.Ofcourse,wecanalsousevariousmetricsonbarcodespace,suchasthebottleneckorWassersteindistances.Itwouldbeextremelyinterestingtoanalyzetherelationshipbetweenthesedistancesonbarcodespaceswithvariousmorealgebraicnotionsofdistanceonthebarcodecoordinates.Itwouldalsobeveryinterestingtodefineandanalyzeanalogouscoordinatesonspacesofmultidimensionalpersistencemodules,wheretheymightgiveinformationwhichiscurrentlynotaccessibleduetothecomplexityofthealgebraicdescriptionsofmultidimensionalpersistencemodules.References[1]H.AdamsandG.Carlsson,Onthenon-linearstatisticsofrangeimagepatches,SIAMJ.ImagingSci.,(2),1,p.110-117.[2]A.Adcock,D.Rubin,andG.Carlsson,ClassificationofHepaticLesionsusingtheMatchingMetric,arXivpreprintarXiv:1210.0866(2012).[3]G.Carlsson,Topologyanddata,Bull.Amer.Math.Soc.(46),2,2009,pp.255-308.[4]G.Carlsson,T.Ishkhanov,V.deSilva,andA.Zomorodian,Onthelocalbehaviorofspacesofnaturalimages,toappear,InternationalJournalofComputerVision,(76),1,2008,pp.1-12.[5]G.Carlsson,A.Zomorodian,A.Collins,andL.Guibas,Persistencebar-codesforshapes,InternationalJournalofShapeModeling,(11),(2005),pp.149-187[6]Chih-ChungChangandChih-JenLin,LIBSVM:alibraryforsupportvectormachines,ACMTransactionsonIntelligentSys-20temsandTechnology,2:27:1–27:27,2011.Softwareavailableathttp://www.csie.ntu.edu.tw/∼cjlin/libsvm[7]F.Chazal,D.Cohen-Steiner,L.Guibas,F.M´emoli,andS.Oudot,Gromov-Hausdorffstablesignaturesforshapeusingpersistence,Com-puterGraphicsForum,(28),5,2009,pp.1393-1403.[8]A.Collins,A.Zomorodian,G.Carlsson,andL.Guibas,Abarcodeshapedescriptorforcurvepointclouddata,ComputersandGraphics,Volume28,2004,pp.881-894.[9]J.Dalbec,Multisymmetricfunctions,BeitrageAlgebraGeom.,(40),1,1999,pp.27-51.[10]W.FultonandR.Weiss,Algebraiccurves:Anintroductiontoalgebraicgeometry,Addison-Wesley,1989.[11]Y.LeCunandC.Cortes,TheMNISTDatabase,CourantInstituteNYU,Accessed16July2012,http://yann.lecun.com/exdb/mnist/[12]D.Mumford,J.Fogarty,andF.Kirwan,GeometricInvariantThe-ory,SpringerVerlag,2002.[13]A.ZomorodianandG.Carlsson,Computingpersistenthomology,Dis-creteandComputationalGeometry,33(2),2005,pp.247-274.21
1511.08870	1	1511	2015-11-28T06:04:47	On Symmetric Polynomials	[ "math.RA" ]	In this paper, we study structure theorems of algebras of symmetric functions. Based on a certain relation on elementary symmetric polynomials generating such algebras, we consider perturbation in the algebras. In particular, we understand generators of the algebras as perturbations. From such perturbations, define injective maps on generators, which induce algebra-monomorphisms (or embeddings) on the algebras. They provide inductive structure theorems on algebras of symmetric polynomials. As application, we give a computer algorithm, written in JAVA v. 8, for finding quantities from elementary symmetric polynomials.	math.RA	math	On Symmetric Polynomials Ryan Golden Ilwoo Cho∗† February 2, 2021 Abstract In this paper, we study structure theorems of algebras of symmetric functions. Based on a certain relation on elementary symmetric polyno- mials generating such algebras, we consider perturbation in the algebras. In particular, we understand generators of the algebras as perturbations. From such perturbations, define injective maps on generators, which in- duce algebra-monomorphisms (or embeddings) on the algebras. They provide inductive structure theorems on algebras of symmetric polynomi- als. As application, we give a computer algorithm, written in JAVA v. 8, for finding quantities from elementary symmetric polynomials. Symmetric Functions, Elementary Symmetric Polynomials, Symmetric Sub- algebras, Perturbations. 05A19, 30E50, 37E99, 44A60. 1 Introduction In this paper, we study structure theorems of algebras generated by symmetric polynomials with commutative multi-variables. By establishing certain recur- rence relations on symmetric polynomials, we prove our structure theorems. As application, we consider how to construct a degree-(n + 1) single-variable poly- nomial ft0(z) from a given degree-n polynomial f (z) by adding a zero t0, and we provide a computer algorithm, written in the computer language JAVA version 8; for finding quantities obtained from elementary symmetric polynomials. Throughout this paper, fix n ∈ N, with additional condition: n > 1, and let x1, ..., xn be arbitrary commutative variables (or indeterminants), for n ∈ N. Then one can have an algebra (1.1) Ax1,...,xn = C [{x1, ..., xn}] , ∗Saint Ambrose Univ., Dept. of Math., 518 W. Locust St., Davenport, Iowa, 52803, USA †This paper is one of the results in the undergraduate research program of St. Ambrose University, Department of Mathematics and Statistics. The second-named author thanks the first-named author, for his hard works and endless passion. 1 consisting of all n-variable polynomials in x1, ..., xn. We call Ax1,...,xn of (1.1), the n-variable polynomial algebra. i.e., if f ∈ Ax1,...,xn, then it is expressed by f = f (x1, ..., xn) = t0+ Pk j=1 (r1,...,rj)∈{1,...,n}j P t(r1,...,rj) (cid:18) j Π i=1 xri(cid:19) , with t0, t(r1,...,rj) ∈ C, for k ∈ N. Let X be a finite set. The symmetric group SX on X is a group under the usual functional composition consisting of all bijective maps, called permu- tations, on X. If X = {1, ..., n}, then we denote SX simply by Sn, for n ∈ N. Definition 1.1 An element f of Ax1,...,xn of (1.1) is said to be symmetric, if f (x1, x2, ..., xn) = f(cid:0)xσ(1), xσ(2), ..., xσ(n)(cid:1) , for all σ ∈ Sn, where Sn is the symmetric group on {1, ..., n}. Define now the subset Sx1,...,xn of Ax1,...,xn by (1.2) Sx1,...,xn = {f ∈ Ax1,...,xn : f is symmetric}. We call Sx1,...,xn, the symmetric subalgebra of Ax1,...,xn. It is not difficult to check that, (1.3) (1.4) f1, f2 ∈ Sx1,...,xn =⇒ f1 + f2 ∈ Sx1,...,xn, f ∈ Sx1,...,xn and t ∈ C =⇒ tf ∈ Sx1,...,xn, and hence, the subset Sx1,...,xn of (1.2) forms a well-defined vector subspace of Ax1,...,xn over C, by (1.3) and (1.4). Moreover, one has (1.5) f1, f2 ∈ Sx1,...,xn =⇒ f1f2 ∈ Sx1,...,xn, where f1f2 means the usual functional multiplication of f1 and f2 in Ax1,...,xn. Since f1 (f2f3) = (f1f2)f3 ∈ Sx1,...,xn, f1 (f2 + f3) = f1f2 + f1f3 ∈ Sx1,...,xn, and (f1 + f2)f3 = f1f3 + f2f3 ∈ Sx1,...,xn, by (1.5), whenever f1, f2, f3 ∈ Sx1,...,xn, the subspace Sx1,...,xn indeed forms a well-determined subalgebra of Ax1,...,xn. Let's define the following functions (1.6) 2 εk(x1, ..., xn) = i1<i2<...<ik∈{1,...,n} (cid:18) k Π l=1 P xil(cid:19) in Ax1,...,xn, for all k = 1, ..., n. Then they are symmetric in Ax1,...,xn. i.e., ε1(x1, ..., xn) = Pn ε2(x1, ..., xn) = j=1 xj, ε3(x1, ..., xn) = Pi1<i2∈{1,...,n} xi1 xi2 , xi1 xi2 xi3 , i1<i2<i3∈{1,...,n} P εn(x1, ..., xn) = n Π j=1 xj · · ·, are elements of Sx1,...,xn. Definition 1.2 We call such polynomials εk(x1, ..., xn) of (1.6), the k-th ele- mentary symmetric polynomials of Sx1,...,xn, for all k = 1, ..., n. Notation 1.1 In the rest of this paper, we denote εk(x1, ..., xn) simply by ε1,...,n k , for all k = 1, ..., n. Also, for convenience, define ε1,...,n 0 = 1 and ε1,...,n n+i = 0, as constant functions in Sx1,...,xn, for all i ∈ N. Whenever we want to em- , for k = 1, precisely, we denote them by εx1,...,xn k phasize the variables of ε1,...,n ..., n. (cid:3) k The following proposition is well-known under its name: Fundamental The- orem of Symmetric Functions. Proposition 1.1 (See [3]) Let Sx1,...,xn be the symmetric subalgebra (1.2) of the n-variable polynomial algebra Ax1,...,xn of (1.1). Then (1.7) Sx1,...,xn Alg = Ch{ε1,...,n 1 , ε1,...,n 2 , ..., ε1,...,n n }i , where " are the elementary symmetric polynomials in the sense of (1.6), for all k = 1, ..., n. (cid:3) Alg = " means "being algebra-isomorphic," and where ε1,...,n k The above structure theorem (1.7) shows that all symmetric functions in Ax1,...,xn are generated by the elementary symmetric polynomials {ε1,...,n of (1.6). k }n k=1 For more about symmetric functions and related studies in mathematics, see e.g., [2], [3], and cited papers therein. 3 2 A Certain Relation on Sx1,...,xn In this section, we establish a recurrence relation on the elementary symmetric polynomials {ε1,...,n }n k=1 generating the symmetric subalgebra Sx1,...,xn of the n-variable polynomial algebra Ax1,...,xn. k As in Notation 1.1, for any k ∈ {1, ..., n}, ε1,...,n k = εk(x1, ..., xn). , for ij ∈ {1, ..., n}, j = 1, ..., t, with t ≤ n in N, then So, if we write εi1,...,it k it means (2.1) εi1,...,it k = (cid:26) εk(xi1 , ...xit ) 0 if k ≤ t if k > t, Moreover, nonzero new elementary symmetric polynomials εi1,...,it 1 , ..., εi1,...,it t are understood as the symmetric functions generating Sxi1 ,...,xit For example, if we write ε1,...,n−1 in Axi1 ,...,xit . , for k = 1, ..., n − 1, then they are the elementary symmetric polynomials generating Sx1,...,xn−1 in Ax1,...,xn−1; if we write εt 1, then it is a single-variable function xt in Axt = C[xt]. k Theorem 2.1 Let ε1,...,n the symmetric subalgebra Sx1,...,xn. Then k be the elementary symmetric polynomials generating (2.2) ε1,...,n k = ε1,...,n−1 k + ε1,...,n−1 k−1 εn 1 , where the summands and the factors of the right-hand side of (2.2) are in the sense of (2.1). Proof: Assume first that k = 1. Then ε1,...,n 1 = Pn j=1 xj = (cid:16)Pn−1 j=1 xj(cid:17) + xn = ε1,...,n−1 1 + 1· εn 1 = ε1,...,n−1 1 + ε1,...,n−1 0 εn 1 , by Notation 1.1 and (2.1). So, if k = 1, then the relation (2.2) holds. Suppose now that k = n. Then ε1,...,n n = n Π j=1 xj = 0 +(cid:18)n−1 Π j=1 xj(cid:19) (xn) = ε1,...,n−1 n + ε1,...,n−1 n−1 εn 1 , 4 by Notation 1.1 and (2.1). Thus, if k = n, then the relation (2.2) holds. Now, take k ∈ {2, ..., n − 1}, and define a set T 1,...,n k = {(i1, ..., ik) ∈ {1, ..., n}k : i1 < i2 < ... < ik}. Define also a set T 1,...,n−1 k = {(j1, ..., jk) ∈ {1, ..., n − 1}k : j1 < ... < jk}, and similarly, T 1,...,n−1 k−1 = {(l1, ..., lk−1) ∈ {1, ..., n − 1}k−1 : l1 < ... < lk−1}. By the very construction, two sets T 1,...,n−1 k subsets of T 1,...,n , for all k = 2, 3, ..., n − 1. Depending on the above sets, construct k and T 1,...,n−1 k−1 are understood as X 1,...,n k = (cid:26) k Π l=1 xil : (i1, ..., ik) ∈ T 1,...,n k (cid:27) , X 1,...,n−1 k = (cid:26) k Π l=1 xil : (i1, ..., ik) ∈ T 1,...,n−1 k (cid:27) , and X 1,...,n−1 k−1 = (cid:26)k−1 Π l=1 xil : (i1, ..., ik−1) ∈ T 1,...,n−1 k−1 (cid:27) , respectively. Now, let's define Y 1,...,n−1 k−1 = (cid:26)(cid:18)k−1 Π l=1 xil(cid:19) xn : k−1 Π l=1 xil ∈ X 1,...,n−1 k−1 (cid:27) , i.e., Y 1,...,n−1 k−1 = X 1,...,n−1 k−1 xn, where Xx = {yx : y ∈ X}. Notice that if k = 2, ..., n − 1, then (2.3) X 1,...,n k = X 1,...,n−1 k ⊔ Y 1,...,n−1 k−1 , 5 set-theoretically, by the very definitions, where ⊔ means the disjoint union. Observe now that, whenever k = 2, ..., n − 1, we have ε1,...,n k = εk (x1, ..., xn) = (xi1 ,...,xik )∈X 1,...,n k xil(cid:19) Π l=1 (cid:18) k xil(cid:19) (xi1 ,...,xik−1 ,xn)∈Y 1,...,n−1 k−1 (cid:18)(cid:18)k−1 Π l=1 xil(cid:19) (xn)(cid:19) by (2.3) = = = P Π l=1 (cid:18) k xil(cid:19) P xil(cid:19) P (xi1 ,...,xik )∈X 1,...,n−1 k P ⊔Y 1,...,n−1 k−1 (xi1 ,...,xik )∈X 1,...,n−1 k P (cid:18) k Π l=1 + (xi1 ,...,xik )∈X 1,...,n−1 k (cid:18) k Π l=1 P +   (xi1 ,...,xik−1 ,xn)∈X 1,...,n−1 k−1 (cid:18)k−1 Π l=1 xil(cid:19)  (xn) = ε1,...,n−1 k + ε1,...,n−1 k−1 εn 1 . Therefore, the relations (2.2) hold, for all k = 2, ..., n − 1. So, the relations (2.2) hold, for all k = 1, 2, ..., n. The above relation (2.2) shows the relations between the generators QED {ε1,...,n 1 , ..., ε1,...,n n } of Sx1,...,xn and those {ε1,...,n−1 1 , ..., ε1,...,n−1 n−1 } of Sx1,...,xn−1 and {εn 1 } of Sxn = Axn Also, it shows that all generators of Sx1,...,xn are induced by the generators of Sx1,...,xn−1 and Sxn , by (2.1). Motivated by (2.2), one can obtain the following generalized result. Theorem 2.2 Let {ε1,...,n }n k=1 be the elementary symmetric polynomials gen- erating the symmetric subalgebra Sx1,...,xn. Then, for any fixed i0 ∈ {1, ..., n}, we have (2.4) k 6 ε1,...,n k = ε1,...,i0−1, i0+1,...,n k + ε1,...,i0−1,i0+1,...,n k−1 εi0 1 , for all k = 1, 2, ..., n, where εi1,...,im l are in the sense of (2.1). to i0. Proof: The proof of (2.4) is similarly done by that of (2.3), by replacing n QED So, one can realize that the generators of symmetric subalgebra Sx1,...,xn are induced by the generators of Sx1,...,xi0−1,xi0+1,...,n and Sxi0 = C[xi0 = εi0 1 ], by (2.4). Example 2.1 Suppose we have the symmetric subalgebra Sx1,x2,x3,x4 whose generators are the elementary symmetric polynomials ε1,...,4 , for j = 1, ..., 4. i.e., j ε1,2,3,4 1 ε1,2,3,4 2 ε1,2,3,4 3 = x1 + x2 + x3 + x4, = x1x2 + x1x3 + x1x4 + x2x3 + x2x4 + x3x4, = x1x2x3 + x1x3x4 + x2x3x4, and ε1,2,3,4 4 = x1x2x3x4 in Sx1,x2,x3,x4 . Then ε1,2,3,4 1 ε1,2,3,4 2 ε1,2,3,4 3 and ε1,2,3,4 4 = ε1,2,3 = ε1,2,3 = ε1,2,3 1 + ε1,2,3 2 + ε1,2,3 3 + ε1,2,3 1 0 2 = ε1,2,3 4 + ε1,2,3 3 Similarly, one has ε1,2,3,4 1 ε1,2,3,4 2 ε1,2,3,4 3 ε1,2,3,4 4 and etc. = ε1,2,4 = ε1,2,4 = ε1,2,4 1 + ε1,2,4 2 + ε1,2,4 3 + ε1,2,4 1 0 2 = ε1,2,4 4 + ε1,2,4 3 + ε4 1, 1 1 = ε1,2,3 ε4 ε4 1, ε4 1, 1 = ε1,2,3 ε4 3 ε4 1. 1 + ε3 1, 1 = ε1,2,4 ε3 ε3 1, ε3 1, 1 = ε1,2,4 ε3 3 ε3 1, Acknowledgment Before submitting this paper, the authors realized that the relation (2.2) was already known in psychology; psychological tests (e.g, see [1]). According to [1], Dr. Fischer proved the relation (2.2) in his book "Einfuhrung in die Theorie Psychologischer Tests." However, we could not find the sources, including the book, containing Fischer's proof. So, we provided our own proofs of (2.2) and (2.4) above. (cid:3) 7 3 Perturbations on S and Shifts on Sx1,...,xn In this section, we define the collection S of all symmetric subalgebras of finitely- many commutative variables. And construct certain perturbation processes on S, by understanding each f of Sx1,...,xn ∈ S as multiplication from Sy1,...,yk to Sx1,...,xn,y1,...,yk in S under additional axiomatization (See Section 3.1). Also, we consider a shifting process on a fixed symmetric subalgebra Sx1,...,xn ∈ S by reformulating indexes of generators of Sx1,...,xn, in Section 3.2. In Section 3.3, we apply our perturbation of Section 3.1 and shifting process of Section 3.2 to the inductive construction processes of S. 3.1 Perturbation on S Let S be the collection of symmetric subalgebras Sx1,...,xn in commutative vari- ables {x1, ..., xn}, for all n ∈ N. Let's fix n0 ∈ N, and take Sx1,...,xn0 in S. Definition 3.1 Define now perturbations of Sx1,...,xn0 (3.1.1) on S by f : h ∈ Sy1,...,yn 7−→ f h ∈ Sx1,...,xn0 ,y1,...,yn in S, for f ∈ Sx1,..,xn0 , for all Sy1,...,yn ∈ S, with identification: if ∂ = {x1, ..., xn0 } ∩ {y1, ..., yn} is non-empty, then (3.1.1)′ Sx1,...,xn0 ,y1,...,yn = S({x1,..,xn0 }\∂)⊔ ∂ ⊔ ({y1,...,yn}\∂), in S. Since every symmetric subalgebra is generated by elementary symmetric on S is characterized on polynomials, the perturbations (3.1.1) of Sx1,...,xn0 generators, i.e., (3.1.2) ε x1,...,xn0 k : εy1,...,yn l ∈ Sy1,...,yn 7−→ ε x1,...,xn0 k εy1,...,yn l ∈ Sx1,..,xn0 ,y1,...,yn in S, satisfying the identification; (3.1.1)′. Now, let Sy = C[y] = Ay in S, and assume that y 6= xj, for j = 1, ..., n0. Consider the perturbation of Sx1,...,xn0 on S acting on Sy, i.e., ε x1,...,xn0 k : εy 1 = y 7−→ ε x1,...,xn0 k εy 1. Proposition 3.1 The perturbations of Sx1,...,xn0 rial functor on S. on S is a well-defined catego- Proof: The proof is from the very definition (3.1.1), with identification QED (3.1.1)′, with help of (3.1.2). 8 3.2 Shifts on Sx1,...,xn Let Sx1,...,xn ∈ S be a symmetric subalgebra. Define a shift U on Sx1,...,xn by a linear multiplicative transformation on Sx1,...,xn satisfying (3.2.1) U : ε1,...,n k 7−→ ε1,...,n k−1 on Sx1,...,xn, for all k = 1, ..., n, with additional axiomatization; (3.2.1)′ making all constant functions of C in Sx1,...,xn be zero, i.e., U (C) = 0, for U(cid:16)ε1,...,n 0 (cid:17) = U (1) = 0, all C ∈ C in Sx1,...,xn. Definition 3.2 The morphism U on Sx1,...,xn of (3.2.1) is called the shift on Sx1,...,xn. More generally, the shift U on Sx1,...,xn satisfies (3.2.2) t=0 U t0 +Pk = Pk P P under (3.2.1)′, where ti1,...,ik ∈ C. (i1,...,ik)∈{1,...,n}k ti1,..,ik k Π j=1 ε1,...,n ij ! t=0 (i1,...,ik)∈{1,...,n}k ti1,..,ik k Π j=1 ε1,...,n ij −1 , Proposition 3.2 The shift U of (3.2.1), satisfying (3.2.2), is a well-defined algebra-homomorphism on Sx1,...,xn, for all Sx1,...,xn ∈ S. Proof: Let U be the shift (3.1.1) on Sx1,...,xn ∈ S satisfying (3.1.1)′. By the very construction, the morphism U is a linear transformation which is mul- tiplicative. So, it is automatically an algebra-homomorphism. Clearly, by the linearity, one has U (t1f1 + t2f2) = t1U (f1) + t2U (f2) , for all t1, t2 ∈ C and f1, f2 ∈ Sx1,...,xn. Also, by the multiplicativity of U, U(cid:16)(cid:16)ε1,...,n k1 kl k1 (cid:17)m1(cid:17) ...U(cid:16)(cid:16)ε1,...,n (cid:17)m1 (cid:17)ml(cid:17) = U(cid:16)(cid:16)ε1,...,n ...(cid:16)ε1,...,n (cid:17)(cid:17)m1 (cid:17)(cid:17)ml ...(cid:16)U(cid:16)ε1,...,n = (cid:16)U(cid:16)ε1,...,n k1−1(cid:17)m1 = (cid:16)ε1,...,n ...(cid:16)ε1,...,n (cid:17)ml k1 kl kl , kl (cid:17)ml(cid:17) by (3.2.2), for all k1, ..., kl = 0, 1, ..., n (with (3.2.1)′), for all m1, ..., ml ∈ N, for all l ∈ N. So, for any f1, f2 ∈ Sx1,...,xn, we have 9 U (f1f2) = U (f1) U (f2) in Sx1,...,xn. i.e., the morphism U is indeed an algebra-homomorphism on Sx1,...,xn. QED 3.3 From Sx1,...,xn to Sx1 ,...,xn,y in S in S. Also, fix Now, let's fix n0 ∈ N, and the symmetric subalgebra Sx1,...,xn0 a symmetric subalgebra Sy = C[y] = Ay in S, where y 6= xj, for all j = 1, ..., n0. Let Ex1,...,xn0 = {ε x1,...,xn0 k }n0 k=1 be the generator set of Sx1,...,xn0 . If we understand generators of Ex1,...,xn0 as the perturbations (3.1.1) on S, then they act (3.3.1) ε x1,...,xn0 k (εy 1) = ε x1,...,xn0 k εy 1 ∈ Sx1,...,xn0 ,y on Sy, for all k = 0, 1, ..., n0, with identity: ε On the perturbations Ex1,...,xn0 (3.3.2) of (3.3.1), consider x1,...,xn0 0 = 1. U(cid:0)ε (cid:1) = ε where U is the shift on Sx1,...,xn0 Define now a morphism x1,...,xn0 k x1,...,xn0 k−1 , for all k = 1, ..., n0, of (3.2.1) satisfying (3.2.1)′. α : Ex1,...,xn0 → Sx1,...,xn0 ,y, by a function satisfying (3.3.3) x1,...,xn0 k x1,...,xn0 k x1,...,xn0 k α(cid:0)ε (cid:1) = ε + U(cid:0)ε 1. (cid:1) εy Theorem 3.3 The function α of (3.3.3) is a well-defined injective function into Sx1,...,xn0 ,y. Furthermore, this function α of (3.3.3) is injec- from Ex1,...,xn0 into the generator set Ex1,...,xn0 ,y tive from the generator set Ex1,...,xn0 of Sx1,...,xn0 ,y. In particular, one has of Sx1,...,xn0 (3.3.4) Ex1,...,xn0 ,y = α(cid:0)Ex1,...,xn0(cid:1) ⊔ {ε x1,...,xn0 n0 εy 1}. Proof: By the very definition (3.3.3), the function α has its domain Ex1,...,xn0 whose range is contained in Sx1,...,xn0 ,y. It is not difficult to check that α is in- jective. Indeed, whenever k1 6= k2 in {1, ..., n0}, , x1,...,xn0 k1 α(cid:0)ε (cid:1) = ε 6= ε x1,...,xn0 k1 + ε x1,...,xn0 k1−1 εy 1 x1,...,xn0 k2 + ε x1,...,xn0 k2 10 x1,...,xn0 k2 εy 1 = α(cid:0)ε (cid:1) , in Sx1,...,xn0 ,y, by (1.6). Again by (3.3.3), the range α(cid:0)Ex1,...,xn0(cid:1) of this map α is contained in the generator set Ex1,...,xn0 ,y of Sx1,...,xn0 ,y. Indeed, the symmetric subalgebra Sx1,...,xn0 ,y is generated by the elementary symmetric polynomials, ε x1,...,xn0 ,y l = εl (x1, ..., xn0 , y) , for l = 1, 2, ..., n0 + 1, satisfying by (2.2), for all l = 1, 2, ..., n0. Therefore, + ε x1,...,xn0 l−1 εy 1 x1,...,xn0 l ε x1,...,xn0 ,y l = ε = ε x1,...,xn0 l x1,...,xn0 l + U(cid:0)ε Ex1,...,xn0 ,y = α(cid:0)Ex1,...,xn0(cid:1) ⊔ {ε x1,...,xn0 l (cid:1) εy 1 = α(cid:0)ε (cid:1) , x1,...,xn0 ,y n0+1 = ε x1,...,xn0 n0 εy 1}. QED The above theorem with the relation (3.3.4) illustrates the embedding prop- into the generator set Ex1,...,xn0 ,y of Sx1,...,xn0 erty of the generator set Ex1,...,xn0 of Sx1,...,xn0 ,y. Define now a linear multiplicative morphism Φ : Sx1,...,xn0 → Sx1,...,xn0 ,y by (3.3.5) Φ t0 +Pk j=1 (i1,...,ik)∈{1,...,n0}k P ti1,...,ik k Π l=1 ε x1,...,xn0 il ! = t0 +Pk j=1 (i1,...,ik)∈{1,...,n0}k P ti1,...,ik k Π l=1 α(cid:0)ε x1,...,xn0 il (cid:1) , for all k ∈ N, where t0, ti1,...,ik ∈ C. The above linear multiplicative morphism Φ of (3.3.5) is well-defined because the function α of (3.3.3) is well-defined, and it preserves the generators Ex1,...,xn0 injectively, by (3.3.4), into Ex1,...,xn0 ,y of Sx1,...,xn0 ,y. Corollary 3.4 The symmetric subalgebra Sx1,...,xn0 the symmetric subalgebra Sx1,...,xn0 ,y, for fixed variables x1, ..., xn0 , i.e., is algebra-monomorphic to (3.3.6) Sx1,...,xn0 Alg ֒→ Sx1,...,xn0 ,y, where " Alg ֒→" means "being embedded in." Proof: By the algebra-monomorphism Φ of (3.3.5), Sx1,...,xn0 monomorphic to Sx1,...,xn0 ,y, equivalently, Sx1,...,xn0 Sx1,...,xn0 ,y. is algebra- is naturally embedded in QED More precise to (3.3.6), we obtain the following structure theorem. 11 Theorem 3.5 Let Φ be the algebra-monomorphism (or the embedding) (3.3.5) of Sx1,...,xn0 into Sx1,...,xn0 ,y. Then (3.3.7) Sx1,...,xn0 ,y Alg = Φ(cid:0)Sx1,...,xn0(cid:1) ⊕ C(cid:2){ε x1,...,xn0 n0 εy 1}(cid:3) , where C[X] mean the algebras generated by sets X, and ⊕ means the (pure- algebraic) direct product on algebras. Proof: Note that by (1.7) by (3.3.4) by construction by (3.3.5) x1,...,xn0 n0 εy 1}(cid:3) x1,...,xn0 n0 εy Sx1,...,xn0 ,y = C(cid:2)Ex1,...,xn0 ,y(cid:3) = C(cid:2)α(cid:0)Ex1,...,xn0(cid:1) ⊔ {ε = C(cid:2)C(cid:2)α(cid:0)Ex1,...,xn0(cid:1)(cid:3) ⊔ C(cid:2)ε = C(cid:2)Φ(cid:0)Sx1,...,xn0(cid:1)(cid:3) ⊕ C(cid:2)ε = Φ(cid:0)Sx1,...,xn0(cid:1) ⊕ C(cid:2)ε x1,...,xn0 n0 εy x1,...,xn0 n0 εy 1(cid:3) , 1(cid:3)(cid:3) 1(cid:3) since Φ is an embedding. Therefore, the isomorphism theorem (3.3.7) is QED obtained. By (3.3.7) and (2.4), we obtain the following theorem, too. Theorem 3.6 Let n ∈ N, and let Sx1,...,xn+1 be the symmetric algebra in {x1, ..., xn, xn+1}. Then (3.3.8) Sx1,...,xn+1 Alg = Φ(cid:0)Sx1,...,xj−1,xj+1,...,xn+1(cid:1) ⊕ C(cid:2)ε where Φ is in the sense of (3.3.5). x1,...,xj−1,xj+1,...,xn n ε xj 1 (cid:3) , Proof: The proof of the structure theorem (3.3.8) is done by that of (3.3.7) QED in terms of (2.4). We finish this section with the following example. Example 3.1 Let Sx1,x2,x3 be the symmetric subalgebra in {x1, x2, x3}, with its generator set Ex1,x2,x3 = {εx1,x2,x3 1 , εx1,x2,x3 2 , εx1,x2,x3 3 }, where and εx1,x2,x3 1 εx1,x2,x3 2 = x1 + x2 + x3, = x1x2 + x1x3 + x2x3, εx1,x2,x3 3 = x1x2x3. For the injective map α of (3.3.3), we have 12 α(εx1,x2,x3 α(εx1,x2,x3 1 ) = εx1,x2,x3 ) = εx1,x2,x3 1 2 2 + εx1,x2,x3 + εx1,x2,x3 0 1 εy 1 = εx1,x2,x3,y εy 1 = εx1,x2,x3,y , 1 2 and inducing So, α(εx1,x2,x3 3 ) = εx1,x2,x3 3 + εx1,x2,x3 2 εy 1 = εx1,x2,x3,y 3 , α(Ex1,x2,x3) ⊔ {εx1,x2,x3,y 4 = x1x2x3y} = Ex1,x2,x3,y. Sx1,x2,x3,y = C [α(Ex1,x2,x3) ⊔ {x1x2x3y}] = Φ (Sx1,x2,x3) ⊕ C[x1x2x3y]. 4 Applications In this section, we consider applications of (3.3.7) based on (2.2). 4.1 Zeroes of Single-Variable Polynomials Let f (z) be a degree-n single-variable C-polynomial, i.e., f ∈ Az = C[z]. By the fundamental theorem of algebra, f (z) has its zeroes λ1, ..., λn (without considering multiplicities), i.e., f (z) = n Π j=1 (z − λj) . xj = −λj, for j = 1, ..., n. For convenience, let Then (4.1.1) f (z) = n Π j=1 k=0 εx1,...,xn k zn−k (z + xj ) = Pn = 1, and in Az, where εx1,...,xn (4.1.2) 0 εx1,...,xn k = εk(x1, ..., xn) = (−1)kεk (λ1, ..., λn) = (−1)kελ1,...,λn , k 13 for k = 1, ..., n. Now, suppose f (z) is arbitrarily given (without knowing zeroes of f (z)) in Az. Then one can construct the polynomial fλn+1(z) whose zeroes are the zeroes of f (z) and λn+1 in C. Proposition 4.1 Let f (z) be a degree-n polynomial Pn let λn+1 ∈ C. Then there exists a polynomial k=0 tkzn−k in Az, and whose zeroes are the zeroes of f (z) and λn+1 in C. In particular, fλn+1(z) = Pn+1 k=0 sk z(n+1)−k ∈ Az s0 = 1, sk = tk − tk−1λn+1, for all k = 1, ..., n, with additional identity: t−1 = 0, and sn+1 = −tnλn+1, in C. Proof: Assume f (z) is a degree-n polynomial in Az and λn+1 ∈ C are arbitrarily fixed, and suppose k=0 tk zn−k in Az. f (z) = Pn By the fundamental theorem of algebra, f (z) has its n-zeroes λ1, ..., λn (without considering the multiplicities). If we let xj = −λj in C, for j = 1, ..., n, then (4.1.3) k=0 tk zn−k with tk = εx1,...,xn k , f (z) = Pn by (4.1.1) satisfying (4.1.2). So, one can construct (4.1.4) εx1,...,xn,xn+1 0 εx1,...,xn,xn+1 k = 1, = α (εx1,...,xn k where α is in the sense of (3.3.3), and ) , for k = 1, ..., n, εx1,...,xn,xn+1 n+1 = εx1,...,xn n xn+1 = x1x2...xnxn+1, by (3.3.4) (and by (3.3.7)), for xn+1 = − λn+1 in C. Then, by (4.1.3), the quantities (4.1.4) are (4.1.5) s0 = 1, sk = α(tk) = tk + tk−1xn+1, for k = 1, ..., n, and sn+1 = tnxn+1, where xn+1 = −λn+1. In other words, one can construct the degree-(n + 1) polynomial 14 fλn+1(z) = Pn+1 k=0 sk z(n+1)−k, whose constant term s0 and coefficients sj satisfy (4.1.5) in Az, such that the zeroes of fλn+1(z) are the zeroes of f (z) and a given C-quantity λn+1. QED The above proposition is illustrated in the following example. Example 4.1 Let f (z) = z4 − 2z2 + z + 3 in Az, and let i ∈ C. One can let t0 = 1, t1 = −2, t2 = 0, t3 = 1, and t4 = 3, in C. Then, for the fixed C-quantity i, we have s0 = 1, s1 = t1+ t0i = −2 + i, s2 = t2 + t1i = 0 + (−2)i = −2i, s3= t3 + t2i = 1 + 0i = 1, s4 = t4 + t3i = 3 + i, and s5 = t4i = 3i, in C, inducing a new degree-4 polynomial fi(z) fi(z) = P5 in Az. Then this new degree-4 polynomial k=0 skz5−k = z5 + (−2 + i)z4 − 2iz3 + z2 + i z + 3i. z5 + (−2 + i)z4 − 2iz3 + z2 + i z + 3i has its zeroes i and all zeroes of f (z). The above proposition allows us to construct a degree-(n + 1) polynomial fλ(z) whose zeroes are the zeroes of f (z) and λ, even though we do not know the zeroes of a fixed degree-n polynomial f (z). 4.2 JAVA Algorithm for C-Quantities from Elementary Symmetric Polynomials In this section, as an application of (3.3.7) and (2.2), we establish a computer algorithm for finding quantities from elementary symmetric polynomials. This computer algorithm is constructed by the program language, JAVA version 8. JAVA (v.8) Program: Computing C-quantities from Elementary Symmet- ric Polynomials. import java.util.Scanner; public class ComplexRecursiveAlgo { public static void main(String[] args) { Scanner in = new Scanner(System.in); //Prompt for number of generators, n. 15 System.out.print("Enter number of generating variables: "); int n = in.nextInt(); System.out.println(); //Prompt for the n known generators, vector X. System.out.print("Enter the values of the n generators as a white space separating list: "); long[][] X = new long[n][2]; for(int i=0; i <n; i++){ X[i][0] = in.nextInt(); X[i][1] = in.nextInt(); } System.out.println(); //Recursively solve for the values of the elementary symmetric functions and store in 2D array long[][][] epsilon = new long[n][][]; for(int i=0; i <n; i++) epsilon[i] = new long[i+1][2]; epsilon[0][0] = X[0]; for(int i=1; i <n; i++){ for(int k=0; k <epsilon[i].length; k++){ if(k-1 <0) epsilon[i][k] = add(epsilon[i-1][k], X[i]); //Since epsilon[i-1][k] = 1 for k <0 else if(k >i-1) epsilon[i][k] = multiply(epsilon[i-1][k-1],X[i]); //Since epsilon[i-1][k] = 0 for k >i-1 else epsilon[i][k] = add(epsilon[i-1][k], multiply(epsilon[i-1][k-1],X[i])); } } //Provide values of epsilon to user System.out.print("Enter the values of n and k for the desired iteration: "); int N = in.nextInt() - 1; int K = in.nextInt() - 1; System.out.println(); System.out.println("epsilon[" + (N+1) + "][" + (K+1) + "] = (" + ep- silon[N][K][0] + "," + epsilon[N][K][1] + ")"); } public static long[] add(long[] z1, long[] z2){ long[] w = new long[2]; w[0] = z1[0] + z2[0]; w[1] = z1[1] + z2[1]; return w; } public static long[] multiply(long[] z1, long[] z2){ long[] w = new long[2]; w[0] = z1[0]z2[0] - z1[1]z2[1]; 16 w[1] = z1[0]z2[1] + z1[1]z2[0]; return w; } } return w; } } References [1] Baker, F. B. and Harwell, M. R., Computing Elementary Symmetric Functions and Their Derivatives: A Didactic, Appl. Psycho. Measurement, Vol 20, no. 2, (1996) 169 - 192. [2] David, F. N., Kendall, M. G., and Heliotrope, B., "Symmetric Func- tions and Allied Tables", (1966) Cambridge Univ. Press. [3] Zhou, J., Introductions to Symmetric Polynomials and Symmetric Func- tions, Avialable at http://faculty.math.tsinghua.edu.cn/jzhou, (2003) Lec- ture Note. Mr. Ryan Golden is an undergraduate at Saint Ambrose University. He has worked under Dr. Ilwoo Cho for the past two years and attributes his decision to pursue a PhD in applied math largely to this partnership. His particular interests include dynamical systems theory, stochastic processes, and statistical learning theory. Ph.D. Ilwoo Cho has been a faculty member of the department of math- ematics and statistics at Saint Ambrose University since 2005. His research interests include free probability, operator algebra and theory, combinatorics, and groupoid dynamical systems. Ryan Golden, 536 Carlsbad Tr., Roselle, IL 60172, USA [email protected] Ilwoo Cho, Dept. of Mathematics, Saint Ambrose University, 518 w. Locust St., Davenport, IA 52803, USA [email protected] 17
1303.6533	2	1303	2013-09-25T14:27:54	Simple Rings and Degree Maps	[ "math.RA" ]	For an extension A/B of neither necessarily associative nor necessarily unital rings, we investigate the connection between simplicity of A with a property that we call A-simplicity of B. By this we mean that there is no non-trivial ideal I of B being A-invariant, that is satisfying AI \subseteq IA. We show that A-simplicity of B is a necessary condition for simplicity of A for a large class of ring extensions when B is a direct summand of A. To obtain sufficient conditions for simplicity of A, we introduce the concept of a degree map for A/B. By this we mean a map d from A to the set of non-negative integers satisfying the following two conditions (d1) if a \in A, then d(a)=0 if and only if a=0; (d2) there is a subset X of B generating B as a ring such that for each non-zero ideal I of A and each non-zero a \in I there is a non-zero a' \in I with d(a') \leq d(a) and d(a'b - ba') < d(a) for all b \in X. We show that if the centralizer C of B in A is an A-simple ring, every intersection of C with an ideal of A is A-invariant, ACA=A and there is a degree map for A/B, then A is simple. We apply these results to various types of graded and filtered rings, such as skew group rings, Ore extensions and Cayley-Dickson doublings.	math.RA	math	Simple Rings and Degree Maps Patrik Nystedt∗ Johan Öinert† September 10, 2018 For an extension A/B of neither necessarily associative nor necessarily unital rings, we investigate the connection between simplicity of A with a property that we call A-simplicity of B. By this we mean that there is no non-trivial ideal I of B being A-invariant, that is satisfying AI ⊆ IA. We show that A-simplicity of B is a necessary condition for simplicity of A for a large class of ring extensions when B is a direct summand of A. To obtain sufficient conditions for simplicity of A, we introduce the concept of a degree map for A/B. By this we mean a map d from A to the set of non-negative integers satisfying the following two conditions (d1) if a ∈ A, then d(a) = 0 if and only if a = 0; (d2) there is a subset X of B generating B as a ring such that for each non-zero ideal I of A and each non-zero a ∈ I there is a non-zero a′ ∈ I with d(a′) ≤ d(a) and d(a′b − ba′) < d(a) for all b ∈ X. We show that if the centralizer C of B in A is an A-simple ring, every intersection of C with an ideal of A is A-invariant, ACA = A and there is a degree map for A/B, then A is simple. We apply these results to various types of graded and filtered rings, such as skew group rings, Ore extensions and Cayley-Dickson doublings. 1 Introduction Let A/B be a ring extension. By this we mean that A and B are rings that are neither necessarily associative nor necessarily unital with B contained in A. For general ring extensions, simplicity of A is, of course, neither a necessary nor a sufficient condition for simplicity of B. However, in many cases where A is graded or filtered, and B sits in A as a direct summand, simplicity of A is connected to weaker simplicity conditions for B. In particular, this often holds if there is an action on the ring B induced by the ring structure on A. The aim of this article is to investigate both necessary (see Theorem ∗Address: University West, Department of Engineering Science, SE-46186 Trollhättan, Sweden, E-mail: [email protected] †Address: Centre for Mathematical Sciences, Lund University, P.O. Box 118, SE-22100 Lund, Sweden, E-mail: [email protected] 1 1.3) and sufficient (see Theorem 1.4) conditions on B for simplicity of A. The impetus for our approach is threefold. The first part of our motivation comes from group graded ring theory and topological dynamics. A lot of attention has been given the connection between properties of topo- logical dynamical systems (X, s) and the ideal structure of the corresponding C ∗-algebras C ∗(X, s) (see e.g. [3], [14], [16], [17] and [18]). In particular, Power [14] has shown that if X is a compact Hausdorff space of infinite cardinality, then C ∗(X, s) is simple if and only if (X, s) is minimal. Inspired by this result, Öinert [8, 13] has shown that there is an analogous result for skew group algebras defined by general group topological dynamical systems. In fact Öinert loc. cit. shows this by first establishing the following result for general skew group rings (for more details concerning topological dynamics, see Section 3.6). Theorem 1.1 (Öinert [8, 13]). Suppose that B is an associative, unital ring and A = B ⋊σ G is a skew group ring. (a) If B is commutative, then A is simple if and only if B is G-simple and B is a maximal commutative subring of A; (b) If G is abelian, then A is simple if and only if B is G-simple and Z(A) is a field. The second part of our motivation comes from the filtered class of rings called Ore extensions A = B[x; σ, δ]. A lot of work has been devoted to the study of the connection between the ideal structure of A and B in this situation (see e.g. [2], [4], [5], [6] and [12]). In [12] Öinert, Richter and Silvestrov show that there are simplicity results for differential polynomial rings that are almost completely analogous to the skew group ring situation (for more details, see Section 3.7). Theorem 1.2 (Öinert, Richter and Silvestrov [12]). Suppose that B is an associative and unital ring and A = B[x; δ] is a differential polynomial ring. (a) If B is a δ-simple and maximal commutative subring of the differential polynomial ring A, then A is simple; (b) The ring A is simple if and only if B is δ-simple and Z(A) is a field. The third part of our motivation comes from non-associative ring theory. Namely, starting with the real numbers, using the classical Cayley-Dickson doubling procedure (see Section 3.3), we get successively, the complex numbers, Hamilton's quaternions, the octonions, the sedenions and so on. Although none of these rings, from the octonions on, are associative, it is well known that they are all simple. Furthermore, matrix algebras over such rings are also simple but, of course, in general not associative. In this article, we unify all of the three types of simplicity results for ring extensions A/B mentioned above, by introducing the notion of A-simplicity of B (see Definition 2.1). It turns out that for crossed product algebras and Ore extensions, A-simplicity coincides with respectively G-simplicity (see Proposition 3.12) and σ-δ-simplicity (see Proposition 3.30). We also introduce a weak form of associativity that we call ideal associativity (see Definition 2.2). We show that A-simplicity of B is often a necessary condition for simplicity of A. Theorem 1.3. If A/B is an ideal associative ring extension satisfying (i) B is a direct summand of A as a left (right) B-module, (ii) every ideal of B has the identity property as a right (left) B-module and (iii) A is simple, then B is A-simple. 2 Many of the proofs of the sufficient conditions for simplicity of A are often based on some kind of reduction argument on the number of homogeneous components that are present in elements of ideals. We show that many of these arguments can be treated using a notion that we call a degree map (see Definition 2.3) for the extension A/B. Theorem 1.4. If A/B is a ring extension satisfying (i) CA(B) is a ring which is A- simple, (ii) every intersection of CA(B) with an ideal of A is A-invariant, (iii) ACA(B)A = A and (iv) there is a degree map for A/B, then A is simple. In Section 2, we define the relevant notions concerning ring extensions that will be used throughout this article and we show Theorem 1.3 and Theorem 1.4. In Section 3.1, we apply Theorem 1.3 and Theorem 1.4 to category graded rings. In particular, we show that there are versions of Theorem 1.1 that hold for strongly groupoid graded rings. In Section 3.2, we utilize results from Section 3.1 in order to analyze simplicity of category crossed products. In Sections 3.3 -- 3.7, we apply the results from the previous sections to obtain simplicity results for, respectively, Cayley extensions, twisted and skew matrix algebras over non-associative rings, skew group rings over non-associative algebras associated with topological dynamical systems, and Ore extensions. 2 Preliminaries and Proofs of the Main Results In this section, we fix the relevant notions concerning ring extensions that we will use in the sequel and we also prove Theorem 1.3 and Theorem 1.4. To this end, we show three results (see Proposition 2.4, Proposition 2.5 and Lemma 2.7) concerning ideals of ring extensions. Let A/B be a ring extension and suppose that X, Y and Z are subsets of A. We let XY denote the set of all finite sums of elements of the form xy, for x ∈ X and y ∈ Y . Furthermore, we let XY Z denote the set of all finite sums of elements of the form x(yz) + (x′y′)z′ for x, x′ ∈ X, y, y′ ∈ Y and z, z′ ∈ Z. If A is unital, then the multiplicative identity element of A is denoted by 1A. Recall that the centralizer of B in A, denoted by CA(B), is the set of elements in A that commute with every element in B. If CA(B) = B, then B is said to be a maximal commutative subring of A. The set CA(A) of A is called the center of A and is denoted by Z(A). We say that a left (or right) B-module X has the identity property if BX = X (or XB = X). By an ideal we always mean a two-sided ideal. The pre-ordered set of ideals of A is denoted by ideal(A). Definition 2.1. We say that an ideal I of B is A-invariant if AI ⊆ IA. We say that A/B is A-simple if there is no non-trivial A-invariant ideal I of B. We let the collection of A-invariant ideals of B be denoted by idealA(B). Definition 2.2. We say that A/B is ideal associative if any ideal I of B associates with any finite collection of copies of A. So for instance using two copies of A we get that (IA)A = I(AA), (AI)A = A(IA) and (AA)I = A(AI). Clearly, if A is associative then A/B is necessarily ideal associative. 3 Definition 2.3. We say that a function d from A to the set of non-negative integers is a degree map for A/B if it satisfies the following two conditions: (d1) if a ∈ A, then d(a) = 0 if and only if a = 0; (d2) there is a subset X of B generating B as a ring, such that for each non-zero ideal I of A and each non-zero a ∈ I there is a non-zero a′ ∈ I satisfying d(a′) ≤ d(a) and d(a′b − ba′) < d(a) for all b ∈ X. Proposition 2.4. If A/B is an ideal associative ring extension, then there is a map of pre-ordered sets i : idealA(B) → ideal(A), defined by i(I) = IA, for all I ∈ idealA(B). Proof. All that we need to show is that i is well defined. Take an A-invariant ideal I of B. Then A(IA) = (AI)A ⊆ (IA)A = I(AA) ⊆ IA and (IA)A = I(AA) ⊆ IA. Therefore, IA is an ideal of A. Proposition 2.5. If A/B is an ideal associative ring extension satisfying (i) B is a direct summand of A as a left (right) B-module and (ii) every ideal of B has the identity property as a right (left) B-module, then i : idealA(B) → ideal(A) is an injective map of pre-ordered sets. Proof. We show the left-right part of the claim. The right-left part of the claim is shown in a completely analogous way and is therefore left to the reader. Let the set of additive subgroups of A be denoted by S(A). Let p denote the map S(A) ∋ Y 7→ B ∩ Y ∈ S(B). We will show that p ◦ i = ididealA(B), which, in particular, implies injectivity of i. To this end, take an A-invariant ideal I of B and a left B-bimodule X such that A = B ⊕ X as left B-modules. Since I has the identity property as a right B-module we get that (p ◦ i)(I) = B ∩ IA = B ∩ (IB ⊕ IX) = B ∩ (I ⊕ IX) = I. Proof of Theorem 1.3. This follows from Proposition 2.5. Definition 2.6 ([10]). A/B is said to have the ideal intersection property if every non- zero ideal of A has non-zero intersection with B. Lemma 2.7. If A/B is a ring extension equipped with a degree map, then A/CA(B) has the ideal intersection property. Proof. Let I be a non-zero ideal of A. Take a ∈ I with d(a) = min[d(I \ {0})]. By (d2) there is a non-zero a′ ∈ I with d(a′) ≤ d(a) and d(a′b − ba′) < d(a) for all b ∈ X. Since a′b − ba′ ∈ I it follows from the assumptions on d(a) that a′b = ba′ for all b ∈ X. Since X generates B as a ring, we can conclude that a′ ∈ CA(B). Proof of Theorem 1.4. Suppose that A/B is a ring extension satisfying (i)-(iv) in the formulation of Theorem 1.4. Let I be a non-zero ideal of A. From Lemma 2.7 it follows that I ∩ CA(B) is a non-zero A-invariant ideal of CA(B), which, from A- simplicity of CA(B), implies that I ∩ CA(B) = CA(B), i.e. that CA(B) ⊆ I. Therefore A = ACA(B)A ⊆ AIA ⊆ I and hence A = I. This shows that A is simple. 4 3 Applications In the Sections 3.1-3.7 below, we demonstrate how the notion of a degree map can be useful in the study of various classes of ring extensions. 3.1 Category Graded Rings In this section, we apply Theorem 1.3 and Theorem 1.4 to category graded rings. Thereby, we obtain both necessary (see Proposition 3.3) and sufficient (see Proposition 3.10) con- ditions for such rings to be simple. Definition 3.1. Let G be a category. The family of objects of G is denoted by ob(G). We will often identify an object of G with its associated identity morphism. The family of morphisms in G is denoted by mor(G). Throughout this article G is assumed to be small, that is with the property that mor(G) is a set. The domain and codomain of a morphism g in G is denoted by d(g) and c(g) respectively. We let G(2) denote the collection of composable pairs of morphisms in G, that is all (g, h) in mor(G) × mor(G) satisfying d(g) = c(h). We will often view a group as a one-object category. Definition 3.2. We follow the notation from [7], [10] and [11]. Let A be a ring. A G-filter on A is a set of additive subgroups, {Ag}g∈mor(G), of A such that for all g, h ∈ mor(G), we have AgAh ⊆ Agh, if (g, h) ∈ G(2), and AgAh = {0} otherwise. The ring A is called G-graded if there is a G-filter, {Ag}g∈mor(G) on A such that A = ⊕g∈mor(G)Ag. Let A be a ring which is graded by a category G. We say that an ideal I of A is graded if I = ⊕g∈mor(G)(I ∩ Ag). In that case we put Ig = I ∩ Ag, for g ∈ mor(G). We say that A is locally unital if for every e ∈ ob(G) the ring Ae is non-zero and unital, making every Ag, for g ∈ mor(G), a unital Ac(g)-Ad(g)-bimodule. Each a ∈ A can be written as a = Pg∈G rg in a unique way using rg ∈ Ag, for g ∈ G, of which all but finitely many are zero. For g ∈ G, we let (a)g denote rg. The G-gradation on A is said to be right non-degenerate (respectively left non-degenerate) if to each isomorphism g ∈ mor(G) and each non-zero x ∈ Ag, the set xAg−1 (respectively Ag−1x) is non-zero. If a ∈ A and a =Pg∈mor(G) ag with ag ∈ Ag, for g ∈ mor(G), then we let Supp(a) denote the (finite) set of all g ∈ mor(G) with ag 6= 0. For a subset H of mor(G), let AH denote the direct sum ⊕h∈HAh. Let A0 denote Aob(G). Proposition 3.3. If A is a simple ring graded by a category G such that A/A0 is ideal associative and A is locally unital, then A0 is A-simple. Proof. This follows immediately from Proposition 2.5. In fact, if we put B = A0 and C = ⊕g∈mor(G)\ob(G)Ag, then A = B ⊕ C as left B-modules and, since A is locally unital, every ideal of B has the right identity property. Proposition 3.4. If A is a ring graded by a groupoid G and the gradation on A is left or right non-degenerate, then A/Z(A0) has a degree map. 5 Proof. Suppose that the gradation on A is right non-degenerate. Since the claim holds trivially if A0 = {0}, we can assume that A0 6= {0}. Put B = Z(A0). For a ∈ A, define d(a) to be the cardinality of Supp(a). Condition (d1) holds trivially. Now we show condition (d2). Take a non-zero ideal I of A and a non-zero a ∈ I. Take b ∈ B. We now consider two cases. Case 1: there is e ∈ ob(G) with ae 6= 0. Put a′ = a. By the definition of the gradation on A, we get that (a′b − ba′)e = (ab − ba)e = aeb − bae = 0. Therefore d(a′b − ba′) < d(a). Case 2: ae = 0 for all e ∈ ob(G). Take a non-identity g ∈ G with ag 6= 0. By right non-degeneracy of the gradation, there is c ∈ Ag−1 with (ac)c(g) 6= 0. Since d(ac) ≤ d(a) we can put a′ = ac and use the argument from Case 1. The left non-degeneracy case is treated analogously. Definition 3.5. If A is a ring which is graded by a category G, then we say that A/A0 is graded ideal associative if every ideal of A0 associates with any combination of graded components of A. It is clear that graded ideal associativity of A/A0 implies ideal associativity of A/A0. At present, it is not clear to the authors whether or not the converse holds. Proposition 3.6. Suppose that A is a ring graded by a category G such that A is locally unital. If I is an ideal of A0, then (a) I is A-invariant if and only if AgId(g) ⊆ Ic(g)Ag, for all g ∈ mor(G); (b) if A is strongly graded by a groupoid G and A/A0 is graded ideal associative, then I is A-invariant if and only if AgId(g)Ag−1 ⊆ Ic(g), for all g ∈ mor(G). Proof. (a) I is A-invariant if and only if AI ⊆ IA. Since A is graded, this inclusion holds if and only if AgId(g) ⊆ Ic(g)Ag for all g ∈ mor(G). (b) We use (a). Take g ∈ mor(G). First we show the "only if" statement. Suppose that AgId(g) ⊆ Ic(g)Ag. Then AgId(g)Ag−1 ⊆ Ic(g)AgAg−1 = Ic(g)Ac(g) ⊆ Ic(g). Now we show the "if" statement. Suppose that AgId(g)Ag−1 ⊆ Ic(g). Then AgId(g) ⊆ AgId(g)Ad(g) = AgId(g)Ag−1Ag ⊆ Ic(g)Ag. Proposition 3.7. If A is a unital ring which is strongly graded by an abelian group G such that A/Ae is graded ideal associative, Ae is A-simple, then A/A has a degree map. Proof. Put B := A. For a ∈ A, define d(a) to be the cardinality of Supp(a). Let X = ∪g∈GAg, i.e. we let X be the set of homogeneous elements of A. It is clear that X generates A as a ring. Condition (d1) holds trivially. Now we show condition (d2). Take a non-zero ideal I of A and a non-zero a ∈ I and b ∈ X. We now consider two cases. Case 1: ae 6= 0. Let hai denote the ideal in A generated by a. Now put J = {c ∈ hai Supp(c) ⊆ Supp(a)}e. It is clear that J is a non-zero ideal of Ae. Now we show that J is A-invariant. By Proposition 3.6(b) it is enough to show that AgJAg−1 ⊆ J for all g ∈ G. Take g ∈ G and c ∈ hai with Supp(c) ⊆ Supp(a). Then hai ⊇ AgcAg−1 = Ph∈G AgchAg−1. Since 6 G is abelian, we get that AgchAg−1 ⊆ AghAg−1 + AgAhg−1 ⊆ Aghg−1 = Ah for any h ∈ G. Hence Supp(AgcAg−1) ⊆ Supp(a) and therefore AgceAg−1 ⊆ J. Hence, we get that AgJAg−1 ⊆ J. Since Ae is A-simple we get that J = Ae and hence there is a′ ∈ hai with a′ e = 1A, G is abelian and b is homogeneous, we get that d(a′b − ba′) ≤ d(a′) − 1 < d(a′) ≤ d(a). e = 1A. Since a′ Case 2: ae = 0. Take a non-identity g ∈ G with ag 6= 0. By right non-degeneracy of the strong gradation, there is c ∈ Ag−1 with (ac)e 6= 0. Since d(ac) ≤ d(a) we can use the element ac and proceed as in Case 1. Lemma 3.8. Suppose that A is a locally unital ring graded by a category G. If I is an ideal of A, then (a) I = A if and only if 1Ae ∈ I for all e ∈ ob(G); (b) if G is a connected groupoid and A is strongly graded, then I = A if and only if there is e ∈ ob(G) such that 1Ae ∈ I. Proof. The "only if" statements are trivial. Therefore we only need to show the "if" statements. It is enough to show that I ⊇ Ag for all g ∈ mor(G). Take g ∈ mor(G). (a) Since 1Ad(g) ∈ I, we get that Ag = Ag1Ad(g) ⊆ I. (b) Suppose that there is e ∈ ob(G) such that 1Ae ∈ I. Take f ∈ ob(G). By (a) we are done if we can show that 1Af ∈ I. Since G is connected, there is g ∈ mor(G) such that d(g) = e and c(g) = f . Hence, by the strong gradation, we get that I ⊇ Ag1AeAg−1 = AgAg−1 = Af ∋ 1Af . Proposition 3.9. If A is a ring which is locally unital and strongly graded by a connected groupoid G with AGe simple for all e ∈ ob(G), then A is simple. Proof. Take a non-zero ideal I of A and take a non-zero a ∈ I. By local unitality the strong gradation is both left and right non-degenerate. Hence we can always choose a such that ae 6= 0, for some e ∈ ob(G). But then J = 1AeI1Ae is a non-zero ideal of AGe. Since AGe is simple, we get that J = AGe. This implies that 1Ae ∈ J ⊆ I. The claim now follows from Lemma 3.8(b). Proposition 3.10. Suppose that A is a locally unital groupoid graded ring with A0 A- simple. (a) If the gradation is left or right non-degenerate, A0 is a maximal commutative subring of A and every intersection of A0 with an ideal of A is an A-invariant ideal of A0, then A is simple; (b) If A is strongly graded by a locally abelian connected groupoid such that A/A0 is graded ideal associative and for each e ∈ ob(G) the ring Z(AGe) is simple, then A is simple. Proof. (a) This follows immediately from Theorem 1.4 and Proposition 3.4. (b) Take e ∈ ob(G). We claim that the A-simplicity of A0 implies that Ae is AGe- simple. If we assume that the claim holds, then it follows by Proposition 3.7 and Theorem 1.4, that AGe is simple. Hence, by Proposition 3.9, we get that A is simple. Now we show the claim. Suppose that Ie is a non-zero AGe-invariant ideal of Ae. Define the set I as the sum of the additive groups AhIeAh−1, for h ∈ mor(G) with d(h) = e. Clearly, I is an 7 ideal of A0. By Proposition 3.6(b) it is clear that I is A-invariant. Since Ie is non-zero it follows that I is non-zero, and hence we get that I = A0. Thus, I ∩ Ae = Ae. But I ∩ Ae equals the sum of the sets AhIeAh−1, for h ∈ Ge. Since Ie is AGe-invariant we thus get, by Proposition 3.6(b), that Ae = I ∩ Ae ⊆ Ie. This shows that Ae is AGe-simple. 3.2 Category Crossed Product Algebras In this section, we use the results from the previous section to obtain both necessary (see Proposition 3.14) and sufficient (see Proposition 3.15) conditions for category crossed products to be simple. Note that the definition that is being used here, differs slightly from the one in [9]. Definition 3.11. Let Ring denote the category of unital rings and ring homomorphisms that respect multiplicative identity elements. By a crossed system we mean a pair (σ, α) where σ is a functor G → Ring and α is a map from G(2) to the disjoint union of the sets σe, for e ∈ ob(G), satisfying αg,h ∈ σd(g), for (g, h) ∈ G(2). By abuse of notation σe, for e ∈ ob(G), is the image of the object e, or the image of its associated identity morphism, under the functor σ. Hence, σe will denote either an object (ring) in Ring or a morphism (ring morphism) in Ring. It will be clear from the context, how to interpret σe. In this article, we suppose that each αg,h, for g, h ∈ mor(G) with d(g) = c(h), is a unit in Bc(g) and associates and commutes with elements in Bc(g), i.e. such that the equalities αg,h(bc) = (bαg,h)c = b(αg,hc) = (bc)αg,h hold for all b, c ∈ Bc(g). We will use the notation Be = σe, for e ∈ ob(G), and B = ⊕e∈ob(G)Be. Let A = B ⋊σ bg ∈ Bc(g), g ∈ mor(G), are chosen so that all but finitely many of them are zero. Define addition on A pointwise α G denote the collection of formal sums Pg∈G bgug, where Xg∈G agug +Xg∈G bgug = Xg∈G (ag + bg) ug and let the multiplication be defined by the bilinear extension of the relation (agug)(bhuh) = agσg(bh)αg,hugh (1) for g, h ∈ mor(G) with d(g) = c(h) and agugbhuh = 0 when d(g) 6= c(h). We call A the crossed product algebra defined by the crossed system (σ, α). We also suppose that αc(g),g = αg,d(g) = 1Ac(g) for all g ∈ mor(G). If we put Ag = Bc(g)ug, for g ∈ mor(G), then it is clear that this defines a gradation on the ring A which makes it locally unital with 1Ae = 1Be ue, for e ∈ ob(G). We often identify B with A0. Given a crossed product A = B ⋊σ α G, we say that an ideal I of B is G-invariant if σg(Id(g)) ⊆ Ic(g), for all g ∈ mor(G). We say that B is G-simple if there is no non-trivial G-invariant ideal of B. Proposition 3.12. If A = B ⋊σ if and only if it is A-invariant. α G is a crossed product, then an ideal of B is G-invariant 8 Proof. Let I be an ideal of B. First we show the "if" statement. Take g ∈ mor(G). Suppose that I is A-invariant. By Proposition 3.6(a), we get that AgId(g) ⊆ Ic(g)Ag. By the definition of the product in A, we therefore get that Bc(g)σg(Id(g)) ⊆ Ic(g)Bc(g). Since Bc(g) is unital and Ic(g) is an ideal of Bc(g), we get that σg(Id(g)) ⊆ Bc(g)σg(Id(g)) ⊆ Ic(g)Bc(g) ⊆ Ic(g). Hence I is G-invariant. Now we show the "only if" statement. Suppose that I is G-invariant. Take g ∈ mor(G). Then, since Ic(g) is an ideal of Bc(g) and Bc(g) is unital, we get that AgId(g) = (Bc(g)ug)Id(g) = Bc(g)σg(Id(g))ug ⊆ Bc(g)Ic(g)ug ⊆ Ic(g)ug ⊆ Ic(g)Bc(g)ug = Ic(g)Ag. By Proposition 3.6(a), we get that I is A-invariant. Proposition 3.13. If A = B ⋊σ surjective, then A/B is graded ideal associative. α G is a crossed product with every σg, for g ∈ mor(G), Proof. Let I be an ideal of B. Since each σg, for g ∈ mor(G), is surjective, we get that images of ideals by these maps are again ideals. Also, since each αg,h, for (g, h) ∈ G(2), is a unit in Bc(g), the product of αg,h with an ideal of Bc(g) equals the ideal. Therefore, if m and n are non-negative integers and gi ∈ mor(G), for i ∈ {1, . . . , (m + n)}, satisfy d(gi) = c(gi+1) for i ∈ {1, . . . , (m + n − 1)}, then a straightforward induction over m and n shows that all the products Ag1 · · · AgmIAgm+1 · · · Agm+n, performed in any prescribed order by inserting parentheses between the factors, equals σg1···gm(Ic(gm+1))ug1···gm+n. Proposition 3.14. If G is a category and A = B ⋊σ every σg, for g ∈ mor(G), surjective, then B is G-simple. α G is a simple crossed product with Proof. This follows immediately from Proposition 3.3, Proposition 3.12 and Proposition 3.13. Proposition 3.15. Suppose that G is a groupoid and A = B ⋊σ α G is a crossed product where B is a G-simple ring. (a) If B is a maximal commutative subring of A, then A is simple; (b) If G is locally abelian and connected, and for each e ∈ ob(G), the ring Z(AGe) is simple, then A is simple. Proof. First note that since G is a groupoid, every σg, for g ∈ mor(G), is an isomorphism and hence, in particular, a surjection. Indeed, σgσg−1 = σc(g) = idBc(g) and σg−1σg = σd(g) = idBd(g). (a) Let I be an ideal of A. By Proposition 3.10 and Proposition 3.12 we are done if we g,g−1 = can show that I∩B is G-invariant. If g ∈ mor(G), then σg(Id(g)) = σg(Id(g))αg,g−1α−1 ugId(g)ug−1α−1 g,g−1 ⊆ I ∩ Bc(g) = Ic(g). (b) This follows immediately from Proposition 3.10, Proposition 3.12 and Proposition 3.13. Remark 3.16. In the definition of crossed product algebras one may loosen the definition of the product (1) in the following way. For each (g, h) ∈ G(2) we let ∗g,h denote either the ordinary multiplication on Bc(g) or the opposite multiplication on Bc(g). Then we define (aug)(buh) = a ∗g,h σg(b)αg,hugh, for (g, h) ∈ G(2), a ∈ Bc(g) and b ∈ Bc(h). One may also 9 start with a functor σ from G to the larger category Ring′ with unital rings as objects and as morphisms functions that are ring homomorphisms or ring anti-homomorphisms. It is straightforward to verify that Proposition 3.14 and Proposition 3.15 also hold in these cases. Example 3.17. There are group graded rings that are not in a natural way crossed products by groups. This serves as motivation for the results in Section 3.1. To exemplify this, let B be an associative, commutative and unital ring. Then the ring A = M3(B) can be strongly graded by Z2 = {0, 1} by putting A0 =  B B 0 B B 0 0 B 0   and A1 =  0 0 B 0 0 B B B 0   . With respect to this gradation, A is not a crossed product. Indeed, if there were u and v in A1 satisfying uv = 1A, then we would get a contradiction, since the determinant of every element in A1 is zero. Furthermore, it is clear that every ideal of A0 is of the form IJ,K =  J J J J 0 0 0 0 K   for ideals J and K of B. By Proposition 3.6(b) we get that IJ,K is A-invariant if and only if A1IJ,KA1 ⊆ IJ,K. It is easy to check that this holds precisely when J = K. So A0 is A-simple if and only if B is simple. Therefore, by Proposition 3.3 and Proposition 3.10, we get that A is simple precisely when A0 is A-simple. 3.3 Cayley-Dickson Doublings In this section, we analyze simplicity of Cayley-Dickson doublings (see Proposition 3.19). We also show how to construct an infinite chain of simple rings using this procedure (see Example 3.20). Definition 3.18. By a Cayley-Dickson doubling we mean a crossed product algebra A = B ⋊σ α G, where G is a group of order two and B is a unital ring. We let the identity element of G be denoted by e and suppose that g is a generator for G. Suppose that σ is a functor from G (considered as a one-object category) to Ring′. We let σg be denoted by σ and we let Be be denoted by B. Then σ : B → B is either a ring automorphism or a ring anti-automorphism satisfying σ2 = idB. We let α denote αg,g and we suppose that α is a unit in B. We let 1 denote 1Bue. By Remark 3.16, the product in A is given by (a + bug)(c + dug) = a ∗e,e c + b ∗g,g σ(d)α + (a ∗e,g d + b ∗g,e σ(c))ug for a, b, c, d ∈ B. An ideal I of B is called σ-invariant if σ(I) ⊆ I. The ring B is called σ-simple if there is no non-trivial σ-invariant ideal of B. By a classical Cayley-Dickson 10 doubling we mean a doubling where σ is a ring anti-automorphism of B and a ∗e,e b = ab, a∗e,g b = ba, a∗g,e b = ab and a∗g,g b = ba for all a, b ∈ B. Recall that in this case σ can be extended to a to a ring anti-automorphism of A by the relation σ(a + bug) = σ(a) − bug for all a, b ∈ B. Proposition 3.19. Suppose that A is a (classical) Cayley-Dickson doubling of B. (a) If A is simple (σ-simple), then B is σ-simple. (b) If B is σ-simple and Z(A) is simple, then A is simple. Proof. (a) follows from Proposition 3.14. Indeed, if I is a σ-invariant ideal of B, then IA is also a σ-invariant ideal of A, since σ(IA) = σ(I + Iug) = σ(I) − Iug = I + Iug = IA. (b) follows from Proposition 3.15. Example 3.20. Now, suppose that we are given an infinite tower of ring extensions B0 ⊆ B1 ⊆ . . . and units αi ∈ Bi, for i ≥ 0, such that αi commutes and associates with any two elements of Bi. Furthermore, suppose that we for each i ≥ 0 are given a ring anti-automorphism σi of Bi and that the rings B0, B1, B2, . . . are defined recursively by the relations B0 = B and Bi+1 equals the classical Cayley-Dickson doubling of Bi defined by σi and αi. By Proposition 3.19(b) we get: • If B0 is σ0-simple and Z(Bi) is simple for each i ≥ 1, then Bi is simple for each i ≥ 1. Starting with the real numbers B0 = R equipped with the trivial action of σ and putting αi = −1, for i ≥ 0, we get successively, B1 the complex numbers C, B2 Hamilton's quaternions H, B3 the octonions O, B4 the sedenions S and so on. It is easy to check that the center of each of these rings, except for B1 = C, equals the real numbers and is hence a simple ring. We therefore, as a special case, get the following well-known result. • B1 = C is simple and all of the rings B0 = R, B2 = H, B3 = O, B4 = S, etc. are central simple R-algebras. The question of when a twisted Cayley-Dickson doubling of a quaternion algebra is a division algebra has been studied before (see e.g. [15]). This of course implies simplic- ity, but it seems to the authors of the present article that the above result concerning simplicity of the whole chain of Cayley-Dickson doublings is new. 3.4 Twisted Group Rings In this section, we obtain a simplicity result (see Proposition 3.22) for twisted group rings. This provides us with a different proof of simplicity of the Cayley-Dickson doubling algebras from Example 3.20. Definition 3.21. Let G be a group with identity element e. By a twisted group ring we mean a crossed product A = B ⋊σ α G where B is a unital, not necessarily associative, ring, σg = idB, for g ∈ G, and α is a map from G × G to the units of B such that αe,g = αg,e = 1B, for g ∈ G, and each αg,h, for g, h ∈ G, associates and commutes with all elements of B. In that case we write A = B ⋊α G. 11 Proposition 3.22. Let A = B ⋊α G be a twisted group ring, where G is abelian, and B and Z(B) are simple. If for each non-identity g ∈ G, there is a non-identity h ∈ G such that αg,h − αh,g is a unit in B, then A is simple. Proof. By Proposition 3.15(b) it is enough to show that Z(A) = Z(B). The inclusion Z(A) ⊇ Z(B) is trivial. Now we show the inclusion Z(A) ⊆ Z(B). To this end, suppose that x = Pp∈G bpup belongs to Z(A), where bp ∈ B is zero for all but finitely many p ∈ G. Fix a non-identity g ∈ G. Choose a non-identity h ∈ G such that αg,h − αh,g is a unit in B. From the relation xuh = uhx and the fact that G is abelian, we get that bgαg,h = bgαh,g, which, since αg,h − αh,g is a unit in B, implies that bg = 0. Therefore x ∈ B. But since B ⊆ A and x ∈ Z(A), we get that x ∈ Z(B). Example 3.23. Now we can use Proposition 3.22 to deduce simplicity of the algebras R, C, H, O, S etc. (see Example 3.20) in a different way. In fact, Bales [1] has shown that the nth Cayley-Dickson doubling of R is a twisted group algebra over the group Z2n 2 with α : Z2n 2 → {1, −1} defined recursively by the following five relations 2 × Z2n α(p, 0) = α(0, p) = 1, α(2p, 2q) = α(p, q), α(1, 2q + 1) = −1, for p 6= 0, and α(2p + 1, 2q + 1) = α(q, p), α(2p, 2q + 1) = −α(p, q), for integers p and q. In particular, this implies that α is anti-commutative, in the sense that αp,q = −αq,p holds for all non-zero p, q ∈ Z2n 2 , we get that αp,q − αq,p equals either 2 or −2, which, in either case, is a real unit. 2 . Hence, for all non-zero p, q ∈ Z2n 3.5 Matrix Algebras In this section, we obtain a simplicity result for skew and twisted matrix algebras (see Proposition 3.25). Definition 3.24. Suppose that I is a (possibly infinite) set and let G be the groupoid with ob(G) = I and mor(G) = I × I. For i, j ∈ I, we put d(i, j) = j and c(i, j) = i so we can write (i, j) : j → i. If i, j, k ∈ I, then (i, j)(j, k) is defined to be (i, k). Suppose that we are given unital rings Bi, for i ∈ I, and ring isomorphisms σij : Bj → Bi, for i, j ∈ I, such that σii = idBi, for i ∈ I, and σijσjk = σik for all i, j, k ∈ I. Suppose also that we, for each triple (i, j, k) ∈ I × I × I, are given a unit αijk ∈ Bi, with αijk = 1Bi whenever i = j or j = k, that commute and associate with any two elements of Bi. For such a triple, we let ∗ijk denote either the ordinary multiplication on Bi or the opposite multiplication on Bi. Then the crossed product A = B ⋊σ α G can be interpreted as the set of matrices, over the index set I with entries in the rings Bi, with a product which is both "skew" and "twisted" by the αijk's, the ∗ijk's and the σij's and we write A = MI (B). Proposition 3.25. With the above notation, the ring MI (B) is simple if and only if for each i ∈ I, the ring Bi is simple. In particular, all of the rings MI (R), MI (C), MI (H), MI (O), MI (S) etc. are simple. 12 Proof. Note that since each each Ge, for e ∈ ob(G), is a trivial group, we get that B is G-simple precisely when each Bi, for i ∈ I, is simple. Therefore, the first part of the result follows from Proposition 3.9 and Proposition 3.14. The second part of the result follows from the first part and Example 3.20. 3.6 Topological Dynamics In this section, we specialize the simplicity results for crossed products obtained earlier to show a simplicity result for skew group rings, over non-associtive rings, associated with topological dynamical systems (see Proposition 3.28). Definition 3.26. Let G be a group with identity element e. By a skew group ring we mean a crossed product A = B ⋊σ α G where B is a unital, not necessarily associative, ring, σg is a ring automorphism of B, for g ∈ G, and αg,h = 1B, for g, h ∈ G. In that case we write A = B ⋊σ G. Definition 3.27. Let G be a group and suppose that X is a compact Hausdorff space. Let Aut(X) denote the group of homeomorphisms of X. By a topological dynamical system we mean a group homomorphism s : G → Aut(X). A subset Y of X is called G-invariant if s(g)(Y ) ⊆ Y holds for all g ∈ G. The topological dynamical system s is said to be minimal if there is no closed non-empty proper invariant subset of X. Furthermore, s is called faithful if there to each non-identity g ∈ G, is some x ∈ X such that s(g)(x) 6= x. For every non-negative integer i, let Ci denote the ith Cayley-Dickson doubling starting with C0 = R and taking α = −1 at each step. Then we get C1 = C, C2 = H, C3 = O, C4 = S and so on. By abuse of notation we let the norm (respectively involution) on each Ci, for i ≥ 0, be denoted by · (respectively ·). For each i ∈ I, let Bi denote the algebra of Ci-valued continuous functions on X. The addition and multiplication on Bi are defined pointwise and the elements of Ci are identified with the constant functions from X to Ci. We define a norm k · k on each Bi, for i ≥ 0, by kf k = supx∈X f (x), for f ∈ Bi. For each i ≥ 0, the group homomorphism s induces a group homomorphism σ : G → Aut(Bi) by σg(f )(x) = f (s(g−1)(x)), for g ∈ G, f ∈ Bi and x ∈ X. This allows us to define the skew group rings Bi ⋊σ G, for i ≥ 0. Öinert [8, 13] has shown that the following result holds in the case i = 1. Proposition 3.28. If G is abelian, then for each i ≥ 0, the skew group ring Bi ⋊σ G is simple if and only if s is minimal and faithful. Proof. Fix i ≥ 0 and put Ai = Bi ⋊σ G. Suppose first that Ai is simple. By Proposition 3.14, we get that Bi is G-simple. Then s is minimal. In fact, suppose that Y is a non-empty closed G-invariant subset of X. Let IY denote the ideal of Bi consisting of all f ∈ Bi that vanish on Y . Since every compact Hausdorff space is regular, it follows that IY is non-zero. Since Y is G-invariant it follows that IY is G-invariant, and hence by the G-simplicity of Bi, we get that IY = Bi. Thus, Y = X. Now we show that s is faithful. It is easy to see that this is equivalent to 13 σ being injective (see e.g. [13]). Seeking a contradiction, suppose that there is a non- identity morphism g ∈ G with σg = idBi. Let I be the ideal of Ai consisting of finite sums of elements of the form b(uhh′ − uhgh′) for b ∈ Bi and h, h′ ∈ G. Then I is non-zero (since g is not an identity morphism) and proper (since the sum of the coefficients of each element in I equals zero) giving us a contradiction. Therefore s is faithful. Now suppose that s is both minimal and faithful. We show that Ai is simple for i 6= 1 (the case i = 1 was already treated in [8, 13]). Again faithfulness is equivalent to injectivity of σ. Since s is minimal we get that Bi is G-simple. In fact, seeking a contradiction, suppose that there is a non-trivial G-invariant ideal I of Bi. For a subset J of Bi, let NJ denote the set Tf ∈J f −1({0}). We claim that NI is a closed, non-empty proper G-invariant subset of X. If we assume that the claim holds, then s is not minimal and we have reached a contradiction. Now we show the claim. Since I is G-invariant the same is true for NI. Since I is non-zero it follows that NI is a proper subset of X. Since each set f −1({0}), for f ∈ I, is closed, the same is true for NI. Seeking another contradiction, suppose that NI is empty. Since X is compact, there is a finite subset J of I such that NJ = NI. Then the function F = Pf ∈J f 2 = Pf ∈J f f belongs to I and, since NJ is empty, it has the property that F (x) 6= 0 for all x ∈ X. Therefore 1 = F · 1 F ∈ I. This implies that I = Bi which is a contradiction. Therefore NI is non-empty. To show that Ai is simple it is, by Proposition 3.15(b), enough to show that Z(Ai) is simple. We will in fact show that Z(Ai) equals the field R which in particular is a simple ring. It is clear that Z(Ai) ⊇ R. Now we show the reversed inclusion. Take Pg∈G fgug ∈ Z(Ai) where fg ∈ Bi, for g ∈ G, and fg = 0 for all but finitely many g ∈ G. For every h ∈ G, the equality uh(cid:16)Pg∈G fgug(cid:17) = (cid:16)Pg∈G fgug(cid:17) uh holds. From the fact that G is abelian, we get that fg(s(h−1)(x)) = fg(x), for g, h ∈ G and x ∈ X. For every g ∈ G choose a number zg in the image of fg. Since fg ◦ s(h−1) = fg, for each h ∈ G, it follows that the set f −1 g (zg) is a non-empty G-invariant closed subset of X. Since s is minimal it follows that f −1 g (zg) = X and hence that fg = zg, for g ∈ G. Take a non-identity h ∈ G. From the fact that s is faithful, we get that there is a ∈ X such that s(h−1)(a) 6= a. Since X is a compact Hausdorff space, it follows by Urysohn's lemma that there is a p ∈ Bi, taking its values in R, such that p(a) 6= p(s(h−1)(a)). Since Pg∈G zgug commutes with p, we get that zg(p(x) − σ(g)(p)(x)) = 0 for all x ∈ X and g ∈ G. By specializing this equality to x = a and g = h, we get that zh(p(a) − p(s(h−1)(a)) = 0 which in turn implies that zh = 0. Therefore Z(Ai) ⊆ R. 3.7 Ore Extensions In this section, we show that Theorem 1.2, which recently appeared in [12], follows from Theorem 1.4. Definition 3.29. Let B be an associative and unital ring equipped with a ring endomor- phism σ : B → B, and a σ-derivation δ, i.e. an additive endomorphism of B satisfying δ(bc) = σ(b)δ(c) + δ(b)c, for b, c ∈ B. The Ore extension A of B is defined to be polyno- mials over B in the indeterminate x, subject to the relation xb = σ(b)x + δ(b), for b ∈ B. For a non-zero element a ∈ B[x; σ, δ], we let deg(a) denote its degree as a polynomial 14 in x. An ideal I of B is called σ-δ-invariant if σ(I) ⊆ I and δ(I) ⊆ I hold. The ring B is called σ-δ-simple if there is no non-trivial σ-δ-invariant ideal of B. If σ = idB, then A is called a differential polynomial ring in which case σ-δ-simplicity coincides with δ-simplicity. Proposition 3.30. If A = B[x; σ, δ] is an Ore extension, then an ideal of B is σ-δ- invariant if and only if it is A-invariant. Proof. Let I be an ideal of B. First we show the "if" statement. Suppose that I is A-invariant. Take b ∈ I. Then σ(b)x + δ(b) = xb ∈ xI ⊆ IA. Since IB = I and A is a free left B-module with the non-negative powers of x as a basis, this implies that σ(b) ∈ I and δ(b) ∈ I. Hence σ(I) ⊆ I and δ(I) ⊆ I. Therefore I is σ-δ-invariant. Now we show the "only if" statement. Suppose that I is σ-δ-invariant. We claim that if n is a non-negative integer, then the inclusion BxnI ⊆ IA holds. If we assume that the claim holds, then, since A is a free left B-module with the non-negative powers of x as a basis, we get that AI ⊆ IA and hence that I is an A-invariant ideal. Now we show the claim by induction over n. Since Bx0I = B1BI = BI = I = IB ⊆ IA, the claim holds for n = 0. The claim also holds for n = 1 since BxI ⊆ Bσ(I)x + Bδ(I) ⊆ BIx + BI = Ix + I ⊆ IA. Suppose now that the claim holds for an integer n ≥ 1. By the induction hypothesis and the case n = 1, we get that Bxn+1I = BxxnI ⊆ BxBxnI ⊆ BxIA ⊆ IAA = IA, which completes the proof. The following result appeared in [12]. We show that it follows from the results in Section 2. Proposition 3.31. If A = B[x; σ, δ] is a simple Ore extension, then B is σ-δ-simple. Proof. This follows immediately from Theorem 1.3 and Proposition 3.30. Proposition 3.32. If A = B[x; δ] is a differential polynomial ring, then (a) A/Z(B) has a degree map; (b) if B is δ-simple, then A/A has a degree map and thereby A/Z(A) has the ideal intersection property. Proof. For a non-zero p ∈ A, define d(p) = deg(p) + 1. Let d(0) = 0. (a) Let X = Z(B). Condition (d1) is trivially satisfied. Now we check condition (d2). Take a non-zero ideal I of A and a non-zero a ∈ I and b ∈ X. If d(a) = 1, then, trivially, ab − ba = 0 and, hence, d(ab − ba) = d(0) = 0 < d(a). Now suppose that d(a) = n + 1 ≥ 2, i.e. deg(a) = n ≥ 1. For each non-negative integer i, we may j=0 si,i−j(b)xj using some elements si,0(b), si,1(b), . . . , si,i(b) of B. Let i=0 bixi, for {b0, b1, . . . , bn} ⊆ B, with bn 6= 0. Then, since b ∈ Z(B), we get that write xib = Pi a =Pn ab − ba =(cid:0)Pn i=0 bixi(cid:1) b − b(cid:0)Pn i=0(cid:16)(cid:16)Pi =Pn i=0 bixi(cid:1) =Pn j=0 bisi,i−j(b)xj(cid:17) − bbixi(cid:17) . i=0(bixib − bbixi) (2) 15 Since b ∈ Z(B), we get that the coefficient of the term of degree n equals bnsn,0(b)−bbn = bnb − bbn = 0. Therefore, we get that d(ab − ba) ≤ n < n + 1 = d(a). So we can put a′ = a. This shows (d2). (b) Let X = B∪{x}. Condition (d1) is trivially satisfied. Now we check condition (d2). Take a non-zero ideal I of A and a non-zero a ∈ I and b ∈ X. By an argument similar to the graded case (Proposition 3.7) we may assume that the highest degree coefficient i=0 bixi where bn = 1. We now consider two cases. Case 1: Take b ∈ B. By the calculation in (2), we get that d(ab − ba) < d(a). Case 2: Let b = x. Then, since δ(1) = 0, we get that of a is 1 (see Lemma 4.13 in [12]), i.e. a = Pn ax − xa = n Xi=0 Xi=0 = n (bixi+1 − (bix + δ(bi))xi) δ(bi)xi = − δ(bi)xi. n−1 Xi=0 bixi! x − x n Xi=0 bixi! = (bixi+1 − bixi+1 − δ(bi)xi) = − n Xi=0 Xi=0 n Therefore, we get that d(ax − xa) ≤ n < n + 1 = d(a). So we can put a′ = a. This shows (d2). The final conclusion now follows from Lemma 2.7. Proof of Theorem 1.2. (a) This follows immediately from Theorem 1.4, Proposition 3.30 and Proposition 3.32(a), if we can show that every intersection of B with an ideal of A is δ-invariant. Take an ideal I of A. If b ∈ I ∩ B, then δ(b) = bx + δ(b) − bx = xb − bx ∈ I ∩ B. (b) The necessary part follows immediately from Proposition 3.31 and the fact that the center of any associative unital ring is a field. The sufficient part follows immediately from Theorem 1.4 and Proposition 3.32(b). Acknowledgements The second author was partially supported by The Swedish Research Council (postdoc- toral fellowship no. 2010-918 and repatriation grant no. 2012-6113) and The Danish National Research Foundation (DNRF) through the Centre for Symmetry and Deforma- tion. References [1] J. W. Bales, A Tree for Computing the Cayley-Dickson Twist, Missouri J. Math. Sci. 21 (2009), no. 2, 83 -- 93. [2] J. Cozzens, C. Faith, Simple Noetherian rings. Cambridge Tracts in Mathematics, No. 69. Cambridge University Press, Cambridge-New York-Melbourne, 1975. [3] K. R. Davidson, C ∗-algebras by Example, Fields Institute Monographs, 6. American Mathematical Society, Providence, RI, 1996. 16 [4] K. R. Goodearl, R. B. Warfield, Primitivity in differential operator rings, Math. Z. 180 (1982), no. 4, 503 -- 523. [5] N. Jacobson, Pseudo-linear transformations, Ann. of Math. (2) 38 (1937), 484 -- 507. [6] D. A. Jordan, Ore extensions and Jacobson rings, Ph.D Thesis, University of Leeds, 1975. [7] P. Lundström, Separable Groupoid Rings, Comm. Algebra 34 (2006), 3029 -- 3041. [8] J. Öinert, Simple Group Graded Rings and Maximal Commutativity, Operator structures and dynamical systems, 159 -- 175, Contemp. Math., 503, Amer. Math. Soc., Providence, RI, 2009. [9] J. Öinert, P. Lundström, Commutativity and Ideals in Category Crossed Products, Proc. Est. Acad. Sci. 59 (2010), no. 4, 338 -- 346. [10] J. Öinert, P. Lundström, The Ideal Intersection Property for Groupoid Graded Rings, Comm. Algebra 40 (2012), 1860 -- 1871. [11] J. Öinert, P. Lundström, Miyashita Action in Strongly Groupoid Graded Rings, Int. Electron. J. Algebra 11 (2012), 46 -- 63. [12] J. Öinert, J. Richter, S. D. Silvestrov, Maximal commutative subalgebras and sim- plicity of Ore extensions, J. Algebra Appl. 12 (2013), 1250192, 16 pp. [13] J. Öinert, Simplicity of Skew Group Rings of Abelian Groups. To appear in Com- munications in Algebra. arXiv:1111.7214v2 [math.RA]. [14] S. C. Power, Simplicity of C ∗-algebras of minimal dynamical systems, J. London Math. Soc. 18 (1978), 534 -- 538. [15] S. Pumplün, How to Obtain Division Algebras from Twisted Cayley-Dickson Dou- blings, Comm. Algebra 40 (2012), no. 8, 2989 -- 3009. [16] J. Tomiyama, Invitation to C ∗-algebras and topological dynamics, World Sci., Sin- gapore, New Jersey, Hong Kong, 1987. [17] J. Tomiyama, The interplay between topological dynamics and theory of C ∗-algebras. Lecture Notes Series, 2. Global Anal. Research Center, Seoul, 1992. [18] D. P. Williams, Crossed Products of C ∗-algebras, Mathematical Surveys and Mono- graphs, American Mathematical Society, Province, R. I., 2007. 17
1408.0522	4	1408	2015-08-13T00:09:27	Witt's Extension Theorem for Quadratic Spaces over Semiperfect Rings	[ "math.RA" ]	We prove that every isometry of between (not-necessarily orthogonal) summands of a unimodular quadratic space over a semiperfect ring can be extended an isometry of the whole quadratic space. The same result was proved by Reiter for the broader class of semilocal rings, but with certain restrictions on the base modules, which cannot be removed in general. Our result implies that unimodular quadratic spaces over semiperfect rings cancel from orthogonal sums. This improves a cancellation result of Quebbemann, Scharlau and Schulte, which applies to quadratic spaces over hermitian categories. Combining this with other known results yields further cancellation theorems. For instance, we prove cancellation of (1) systems of sesquilinear forms over henselian local rings, and (2) non-unimodular hermitian forms over (arbitrary) valuation rings. Finally, we determine the group generated by the reflections of a unimodular quadratic space over a semiperfect ring.	math.RA	math	WITT'S EXTENSION THEOREM FOR QUADRATIC SPACES OVER SEMIPERFECT RINGS URIYA A. FIRST Abstract. We prove that every isometry of between (not-necessarily orthog- onal) summands of a unimodular quadratic space over a semiperfect ring can be extended an isometry of the whole quadratic space. The same result was proved by Reiter for the broader class of semilocal rings, but with certain restrictions on the base modules, which cannot be removed in general. Our result implies that unimodular quadratic spaces over semiperfect rings cancel from orthogonal sums. This improves a cancellation result of Quebbe- mann, Scharlau and Schulte, which applies to quadratic spaces over hermitian categories. Combining this with other known results yields further cancellation theorems. For instance, we prove cancellation of (1) systems of sesquilinear forms over henselian local rings, and (2) non-unimodular hermitian forms over (arbitrary) valuation rings. Finally, we determine the group generated by the reflections of a unimod- ular quadratic space over a semiperfect ring. 1. Introduction Let F be a field of characteristic not 2 and let (V, q) be a nondegenerate quadratic space over F . The following theorem, known as Witt's Theorem or Witt's Extension Theorem, is fundamental in the theory of quadratic forms. Theorem 1.1 (Witt). Let V1, V2 be subspaces of V and let ψ : qV1 → qV2 be an isometry. Then ψ extends to an isometry ψ′ of q. Furthermore, ψ′ is a product of reflections. Among the theorem's consequences are cancellation of nondegenerate quadratic spaces and the fact that O(V, q), the isometry group of (V, q), acts transitively on maximal totally isotropic subspaces of V . The works of Bak [2], Wall [26] and others have led to defining a notion of qua- dratic forms over arbitrary (non-commutative) rings, and also to an appropriate definition of reflections (see [12], [14], [20]). In this context, Witt's Extension The- orem was generalized by Reiter [20] to semilocal rings, but with certain restrictions on the quadratic spaces (see also [12] and [14] for earlier results). Cancellation of unimodular quadratic spaces was likewise generalized to various families of semilo- cal rings A including the cases where A is commutative ([22], [12] or [11, §3.4.3]), local ([11, Rm. 3.4.2], for instance), or A = lim←−{A/ Jac(A)n}n∈N ([19, §3.4]; this case includes all one-sided artinian rings). However, despite the previous evidence, Date: September 7, 2018. Key words and phrases. quadratic form, Witt's Theorem, reflection, isometry group, semiper- fect ring, semilocal ring, Dickson's invariant, hermitian category, sesquilinear form. The author was supported by an ERC grant #226135 and an SNFS grant #IZK0Z2 151061. 1 2 URIYA A. FIRST Keller [11] has demonstrated that cancellation fails over arbitrary semilocal rings, implying that the restrictions in Reiter's Theorem cannot be removed in general. (Most of the results mentioned here can also be found in [15, Ch. VI].) In this paper, we restrict our attention to a family of semilocal rings called semiperfect rings, and study to what extent Witt's Extension Theorem holds in this setting. Recall that a ring A is called semiperfect if it is semilocal and its Jacobson radical, Jac(A), is idempotent lifting. For example, local rings and semilocal rings satisfying A = lim←−{A/ Jac(A)n}n∈N are semiperfect. See §2.5 below for further examples and details. Let (P, [β]) be a unimodular quadratic space over a semiperfect ring (the defini- tion is recalled below) and let Q, S be summands of P . Our main results are: (1) Every isometry (Q, [βQ]) → (S, [βS]) extends to an isometry of (P, [β]) (2) Under mild assumptions, every isometry of (P, [β]) is a product of quasi- (Corollary 4.9, Theorem 4.8). reflections (Corollary 4.11, Theorem 4.5). (3) We determine the subgroup of O(P, [β]) generated by reflections (Theo- rem 5.8). Apart from an obvious exception in which there are no reflections, this subgroup is always of finite index in O(P, [β]). The proofs are based on Reiter's ideas with certain improvements. In particular, we generalize Reiter's e-reflections (see [20]). We also stress that (1) -- (3) hold without assuming 2 is invertible. Our results imply that unimodular quadratic spaces over semiperfect rings cancel from orthogonal sums. (Note that the base ring in Keller's counterexample [11, §2] is semilocal but not semiperfect.) This in turn leads to other cancellation theorems as follows: In [5], [7] and [6], it was shown that systems of (not-necessarily unimodular) sesquilinear forms can be treated as (single) unimodular hermitian forms over a different base ring. Thus, cancellation holds when this base ring is semiperfect. Using this, we show that cancellation holds for (a) arbitrary (i.e. not-necessarily unimodular) hermitian forms over involutary valuation rings (Corollary 4.18), (b) systems of sesquilinear forms over involutary henselian valuation rings (Corol- lary 4.17). We also strengthen a cancellation theorem of Quebbeman, Scharlau and Schulte [19, §3.4] which applies to quadratic spaces over hermitian categories (Corollary 4.14). Specifically, the cancellation of [19, §3.4] assumes that the underlying hermitian category satisfies: (i) all idempotents split, (ii) every object is the direct sum of objects with local endomorphism ring, and (iii) if A is the endomorphism ring of an object, then A = lim←−{A/ Jac(A)n}n∈N. We show that cancellation holds even without assuming condition (iii). The paper is organized as follows: In section 2, we recall the definitions of quadratic forms over rings and several results to be used throughout. Section 3 introduces quasi-reflections and reflections. In section 4, we prove our version of Witt's Extension Theorem and discuss its applications. Finally, in section 5, we describe the group spanned by the reflections of a unimodular quadratic space (over a semiperfect ring). WITT'S EXTENSION THEOREM 3 2. Preliminaries This section collects several preliminary topics that will be used throughout the paper: We recall quadratic forms over unitary rings, several facts concerning them, a notion of orthogonality for unitary rings, and several facts about semiperfect rings. §2.1. Quadratic Forms. We start with recalling quadratic forms. The definitions go back to Bak [2] and Wall [26]. See [3], [23, Ch. 7], or [15] for an extensive discussion. Let A be a ring. An anti-structure on A consists of a pair (σ, u) such that σ : A → A is an anti-automorphism (written exponentially) and u ∈ A× satisfies uσu = 1 and aσσ = uau−1 for all a ∈ A. Denote by P(A) the category of finitely generated projective right A-modules. As usual, a sesquilinear space is a pair (P, β) such that P ∈ P(A) and β : P × P → A is a biadditive map satisfying β(xa, yb) = aσβ(x, y)b for all x, y ∈ P , a, b ∈ A. In this case, we call β a sesquilinear form. The form β is called u-hermitian if it also satisfies β(x, y) = β(y, x)σu. We say that (P, β) is unimodular if the map Lβ : P → P ∗ := HomA(P, A) given by sending x ∈ P to [y 7→ β(x, y)] ∈ P ∗ is an isomorphism. Note that P ∗ can be made into a right A-module by setting (f a)x = aσ(f x) for all f ∈ P ∗, a ∈ A, x ∈ P . This makes Lβ is a homomorphism of A-modules. There is a natural isomorphism ωP : P → P ∗∗ give by (ωP f )x = (f x)σu for all f ∈ P ∗, x ∈ P . Next, set Λmin = {a − aσu a ∈ A} and Λmax = {a ∈ A : aσu = −a}. A form parameter (for (A, σ, u)) consists of an additive group Λ such that ∀ a ∈ A . Λmin ⊆ Λ ⊆ Λmax aσΛa ⊆ Λ and In this case, the quartet (A, σ, u, Λ) is called a unitary ring. For P ∈ P(A), let SP denote the abelian group of sesquilinear forms on P , and let ΛP denote the subgroup consisting of sesquilinear forms γ ∈ SP satisfying γ(x, y) = −γ(y, x)σu and γ(x, x) ∈ Λ for all x, y ∈ P . The image of β ∈ SP in SP /ΛP is denoted [β]. A quadratic space (over (A, σ, u, Λ)) is a pair (P, [β]) with P ∈ P(A) and [β] ∈ SP /ΛP . Associated with [β] are the u-hermitian form hβ(x, y) = β(x, y) + β(y, x)σu and the quadratic map β : P → A/Λ given by β(x) = β(x, x) + Λ . Both hβ and β are determined by the class [β], and conversely, [β] is determined by hβ and β. We also have (2.1) β(x + y) = β(x) + β(y) + hβ(x, y) ∀ x, y ∈ P . We say that (P, [β]) is unimodular if hβ is a unimodular u-hermitian form, i.e. if L[β] := Lhβ is an isomorphism (β itself may be non-unimodular). Isometries between quadratic (resp. sesquilinear, u-hermitian) spaces are defined in the standard way (cf. [15, §I.2.2, §I.5.2]). We let O(P, [β]) denote the group of isometries of (P, [β]). The category of unimodular quadratic spaces over (A, σ, u, Λ) (with isometries as morphisms) is denoted by UQu,Λ(A, σ). 4 URIYA A. FIRST Remark 2.1. When 2 ∈ A×, we have Λmin = Λmax, so there is only one form parameter Λ. Furthermore, in this case, UQu,Λ(A, σ) is isomorphic to the the cat- egory of unimodular u-hermitian forms over (A, σ, u). Indeed, [β] can be recovered from hβ via [β] = [ 1 2 hβ]. Remark 2.2. The "classical" notion of a quadratic space over a commutative ring A (see [22], for instance) occurs in the special case σ = idA, u = 1 and Λ = 0. Then, the quadratic map β determines hβ by (2.1). Remark 2.3. The triple (P(A),∗,{ωP}P ∈P(A)) is a hermitian category and the pair (1, Λ := {ΛP}P ∈P(A)) is a form parameter (see [19] or [15, §II.2] for the definitions). There is a one-to-one correspondence between quadratic spaces over (A, σ, u, Λ) and quadratic spaces over (P(A), 1, Λ) given by (P, [β]) 7→ (P, [Lβ]). §2.2. Conjugation and Transfer. We now introduce two well-known procedures that we refer to as conjugation and e-transfer. They allow one to alter a unitary ring (A, σ, u, Λ) while maintaining data about isometries between quadratic forms (isometry groups in particular). We shall use these manipulations several times in the sequel. Proposition 2.4 ("Conjugation"). Let v ∈ A×. Define (σ′, u′, Λ′) by aσ′ = vaσv−1, u′ = v(vσ)−1u, Λ′ = vΛ . (A, σ′). We call (σ′, u′, Λ′) the conjugation of (σ, u, Λ) by v. Then (A, σ′, u′, Λ′) is a unitary ring and UQu,Λ(A, σ) ∼= UQu′,Λ′ Proof. This proposition is essentially [20, Lm. 1.6]; everything follows by straight- forward computation. The categorical equivalence is constructed as follows: For any P ∈ P(A) and β ∈ SP , define vβ : P × P → A by (vβ)(x, y) = v · β(x, y) . Then (P, vβ) is s sesquilinear form over (A, σ′), and the assignment (P, [β]) 7→ (P, [vβ]) defines the required categorical isomorphism (isometries are mapped to themselves). (cid:3) Proposition 2.5 ("e-transfer"). Let e ∈ A be an idempotent satisfying eσ = e and AeA = A. For every sesquilinear form β : P × P → A, denote by βe the restriction of β to P e × P e. Then: (i) (B, τ, v, Γ) := (eAe, σeAe, eu, eΛe) is a unitary ring. (ii) (P e, [βe]) is a quadratic space over (B, τ, v, Γ). It is unimodular if and only if (P, [β]) is unimodular. (iii) The map [β] 7→ [βe] : SP /ΛP → SP e/ΛP e is an abelian group isomorphism. (iv) The assignment (P, [β]) 7→ (P e, [βe]), called e-transfer, gives rise to an In particular, [β] = [0] ⇐⇒ [βe] = [0]. equivalence of categories UQu,Λ(A, σ) ∼ UQv,Γ(B, τ ). Proof (sketch). Part (i) is straightforward. For parts (ii), (iii) and (iv), view P(A) and P(B) as hermitian categories with a form parameter as in Remark 2.3. Let F : P(A) → P(B) be the functor given by F P = P ⊗A Ae ∼= P e. By Morita Theory (see [18, §18D] for instance), F is an equivalence of categories. In addition, there is a natural isomorphism φP from F (P ∗) = P ∗e to (F P )∗ = HomeAe(P e, eAe) given by φ(f ) 7→ fP e (check this for WITT'S EXTENSION THEOREM 5 P = AA, the general case follows by additivity). It is routine to check that (F, φ) is strictly duality preserving functor from P(A) to P(B) (see [19, §2]). This means parts (ii) and (iii) hold tautologically, and part (iv) follows from [19, Lm. 2.1] (for instance). (cid:3) §2.3. Simple Unitary Rings. A unitary ring (A, σ, u, Λ) is called simple if the only ideals of A which are invariant under σ are 0 and A. It is not hard to show that in this case, A is either simple, or A ∼= B × Bop, where B is a simple ring, and σ is given by (a, bop)σ = (tbt−1, aop) for some t ∈ B×. Assume (A, σ, u, Λ) is simple and A is artinian. Then the Artin-Wedderburn Theorem implies that A ∼= Mn(D) where D is a division ring or a product of a division ring and its opposite. Identifying A with Mn(D), we say that such (A, σ, u, Λ) is standard or in standard form if: (1) σ is of the form (dij )i,j 7→ (dτ ji)i,j for some involution τ : D → D. (In particular, σ is an involution.) (2) When D ∼= E × Eop with E a division ring, τ is the exchange involution (a, bop) 7→ (b, aop). (3) u = 1 if τ 6= idD. In this case, we have u ∈ {±1} (because when τ = idD, we have u2 = uσu = 1). It is not true that any simple artinian unitary ring is isomorphic to a unitary ring in standard form. However, this is true after applying a suitable conjugation in the sense of Proposition 2.4, and conjugation does not essentially change the category of quadratic spaces. Proposition 2.6. After a suitable conjugation (cf. Proposition 2.4), any simple artinian unitary ring (A, σ, u, Λ) is isomorphic to a unitary ring in standard form. Proof. Apart from a small difference in condition (3), this proposition is [20, Pr. 2.1]. We have included here a full proof of the sake of completeness. Let A = Mn(D) be as above. By [8, Th. 7.8], σ is conjugate to some σ′ of the form (dij )i,j 7→ (dτ ji)i,j where τ : D → D is an anti-automorphism, so assume σ is in this form. This implies that u commutes with the standard matrix units {eij} (because they satisfy eσσ ij = eij), hence we may view u as an element of D (embedded diagonally in A = Mn(D)) which satisfies dτ τ = udu−1 for all d ∈ D. Now, it is enough to show that (D, τ ) can be made standard by conjugation. Assume that there exists v ∈ D× with u−1vτ = v. Then u′ := v(vτ )−1u = 1, so by Proposition 2.4 (or by computation), τ ′ : d 7→ vdτ v−1 is an involution. Observe that d 7→ u−1dτ is an involutary additive map, hence v = d + u−1dτ always satisfies u−1vτ = v. E ) to get v ∈ D×. Otherwise, any d with v = d + u−1dτ 6= 0 will do. If such d does not exist, then dτ = −ud for all d ∈ D. Taking d = 1 implies u = −1 and hence τ = idD. Thus, either (u, τ ) can be conjugated to (1, τ ′) with τ ′ : D → D an involution, or τ = idD and u = −1 (in which case D is a field). This implies (1) and (3). Indeed, when D is not a division ring, there is an isomorphism D ∼= E × Eop and under that isomorphism τ is given by (a, bop)τ = (tbt−1, aop) for some t ∈ E. Since τ is an involution, it must be the exchange involution. If D = E × Eop with E a division ring, take d = (1E, 0op It remains to check (2). Remark 2.7. Proposition 2.6 is the reason why many authors require σ to be an involution in the definition of unitary rings. The author does not know if there (cid:3) 6 URIYA A. FIRST exists a similar result for semilocal rings (i.e. a statement guaranteeing that σ can always be conjugated into an involution). See [9, Rm. 7.7] for further discussion. Proposition 2.8. Let (A, σ, u, Λ) be a unitary ring such that A is a semisimple (artinian) ring. Then (A, σ, u, Λ) factors into a product (A, σ, u, Λ) ∼= (Ai, σi, ui, Λi) := (cid:16)Y tY i=1 Ai,Y i σ, (ui)i,Y i Λi(cid:17) i with each (Ai, σi, ui, Λi) simple artinian. Proof. See [20, p. 486], for instance. (cid:3) §2.4. Orthogonality. We now define a notion of orthogonality for simple artinian unitary rings which will be used later in the text (compare with the orthogonality defined in [3, Ch. 4, §2] in the commutative case). This notion is used implicitly and repeatedly in [20]. Definition 2.9. A simple artinian unitary ring (A, σ, u, Λ) is called orthogonal if: (1) A is simple and of finite dimension over its center, denoted K, (2) σK = idK, (3) Λ is a K-vector space and dimK Λ = 1 2 n(n − 1) where n = √dimK A. If in addition A ∼= Mn(K) (i.e. A is split as a central simple K-algebra), then we say that (A, σ, u, Λ) is split-orthogonal. Remark 2.10. We use the term "orthogonal" because isometry groups of unimod- ular quadratic forms over an orthogonal unitary ring (A, σ, u, Λ) are forms of the the orthogonal group Om(K), when viewed as algebraic groups over K := Cent(A). This follows from the discussion in §5.1 below. (A symplectic unitary ring can like- wise be defined by replacing 1 Example 2.11. If (A, σ, u, Λ) is simple artinian and in standard from (see §2.3), then it is split-orthogonal if and only if A ∼= Mn(K) for a field K, σ is the matrix transposition, u = 1, and Λ = Λmin. 2 n(n + 1) in condition (3).) 2 n(n − 1) with 1 Generalizing the example, let (A, σ, u, Λ) be a unitary ring such that σ is an involution. If (A, σ, u, Λ) satisfies conditions (1) and (2), then A is a central simple algebra over its center K (see [16, §1]) and σ is an involution of the first kind (i.e. it fixes Cent(A)). This easily implies u ∈ {±1}. By [16, Pr. 2.6], when char K 6= 2, there is ε ∈ {±1} such that dimK{a ± aσ a ∈ A} = n(n ± ε) 1 2 where n = deg A := √dimK A. When ε = 1 (resp. ε = −1) σ is called orthogonal (resp. symplectic). Furthermore, when char K = 2, we always have dimK Λmin = dimK{a − aσ a ∈ A} = 1 2 n(n − 1) . Thus, when σ is an involution, condition (3) is equivalent to having one of the following: (3a) char K 6= 2, σ is orthogonal and u = 1, (3b) char K 6= 2, σ is symplectic and u = −1, (3c) char K = 2 and Λ = Λmin. WITT'S EXTENSION THEOREM 7 See [16, §2] for further details about orthogonal and symplectic involutions. We further recall that the index of a central simple K-algebra A admitting an involution of the first kind is a power of 2 ([16, Cr. 2.8]). Thus, if deg A is odd, then A is split (i.e. ind A = 1). Proposition 2.12. Orthogonality (resp. split-orthogonality) of simple artinian uni- tary rings is preserved under conjugation (see Proposition 2.4). Furthermore, if e ∈ A is an idempotent satisfying eσ = e, then (A, σ, u, Λ) is orthogonal (resp. split- orthogonal) if and only if (eAe, σeAe, ue, eΛe) is orthogonal (resp. split-orthogonal). Proof. That orthogonality (resp. split-orthogonality) is invariant under conjugation is clear from the definitions, so we turn to prove the second statement. Note that since (A, σ, u, Λ) is simple and eσ = e, we have AeA = A (because (AeA)σ = AeA). Morita Theory (see [18, §18D], for instance) now implies that A is simple if and only if eAe is simple, and Cent(eAe) = e Cent(A). Writing K = Cent(A), it follows that A is a (split) central simple K-algebra if and only if eAe is. Furthermore, in this case, it is easy to see that σ is of the first kind if and only if σeAe is. Therefore, we may assume A is a central simple K-algebra and σ is of the first kind. We claim that Λ is a K-vector space if and only if eΛe is a K-vector space. (In fact, this is clear when char K 6= 2 because Λ = Λmin and σ is of the first kind.) One direction is evident so we turn to show the other. Assume eΛe is a K-vector space and let a ∈ Λ and k ∈ K. Write 1A = Pi xieyi for {xi, yi}i ⊆ A. Then a = (Pi xieyi)σa(Pj xjeyj) = Pi,j aij where aij = yσ i axjeyj. Observe that since aσu = −a, aji = −aσ i axie ∈ eΛe ⊆ Λ i axie)y ∈ Λ. Now, ka = Pi,j aij = kPi<j(aij − for all i, hence k · aii = yσ iju) + kPi aii = Pi<j (kaij − (kaij)σu) +Pi(kaii) ∈ Λ, as required. aσ Assume Λ is a K-vector space. It is left to show that dimK Λ = 1 2 n(n − 1) if and only if dimK eΛe = 1 2 m(m − 1), where n = deg A and m = deg eAe. Observe that by the above discussion, when σ is an involution, we always have dimK Λ = 2 n(n ± 1) and dimK eΛe = 1 1 2 m(m ± 1). Thus, by Proposition 2.6, the same holds for arbitrary σ. Let f = 1 − e. There is nothing to prove if f = 0. Otherwise, deg f Af = n− m, hence dimK f Λf = 1 2 (n− m)(n− m± 1). It is easy to check that dim Λ = dim eΛe + dim f Λf + dim eAf = dim eΛe + dim f Λf + m(n− m), and this implies dim Λ = 1 §2.5. Semiperfect Rings. We finish this section with recalling several facts about semiperfect rings. Proofs and additional details can be found in [21, §2.7 -- §2.9]. iju. Since eΛe is a K-vector space, k · exσ i (kexσ 2 n(n − 1) if and only if dim eΛe = 1 2 m(m − 1). i exσ (cid:3) A ring A is called semiperfect if it satisfies the following equivalent conditions: (a) A is is semilocal and Jac(A) is idempotent lifting. (b) All finitely generated right (or left) A-modules have a projective cover (see [21, Df. 2.8.31]). (c) There exists orthogonal idempotents e1, . . . , en ∈ A with Pi ei = 1 and such that eiAei is local for all i. In this case, any system of orthogonal idempotents in A/ Jac(A) can be lifted to a system of orthogonal idempotents in A. Furthermore, eAe is semiperfect for any idempotent e ∈ A. gerenally, all semilocal rings A with A = lim←−{A/ Jac(A)n}n∈N. Further examples Examples of semiperfect rings include all one-sided artinian rings, and more that will be used later can be obtained from the following proposition. 8 URIYA A. FIRST Proposition 2.13. Let R be a henselian local (commutative) ring, and let A be an R-algebra. Then A is semiperfect if one of the following holds: (1) R is noetherian and A is finitely generated as an R-module. (2) R is a valuation ring and A is R-torsion-free and of finite rank over R.1 Proof. When (1) holds, this follows from [1, Th. 22] or [25, Lm. 12]. When (2) holds, A is semilocal by [27, Th. 5.4]. Let e ∈ A be an idempotent such that eAe has no idempotents other than e and 0. The proof of [25, Lm. 14] then implies that eAe is local. Replacing A with (1 − e)A(1 − e) and repeating this procedure yields a (finite) system of orthogonal idempotents e = e1, e2 . . . , et ∈ A with Pi ei = 1 and such that eiAei local for all i, so A is semiperfect. (cid:3) Let A be a semiperfect ring. Then, up to isomorphism, there exist finitely i=1, and every P ∈ P(A) can be i=1 uniquely determined. If A = A/ Jac(A), i with (ni)t many indecomposable projective A-modules, {Pi}t written as P ∼= Lt i=1 P ni then {P i := Pi/Pi Jac(A)}t simple artinian rings Qt The modules P1, . . . , Pt can be constructed as follows: Write A as a product of i=1 Ai, let εi be a primitive idempotent in Ai, and let ei be a lifting of εi to A. Then e1A, . . . , etA are the indecomposable projective right A-modules, up to isomorphism. i=1 are the simple A-modules, up to isomorphism. 3. Reflections and Quasi-Reflections In this section we introduce and study quasi-reflections, which slightly extend a notion of reflections used by Reiter [20]. Throughout, (A, σ, u, Λ) is a unitary ring, and (P, [β]) be a quadratic space over (A, σ, u, Λ). Let e, f ∈ A be idempotents. An element a ∈ eAf is called (e, f )-invertible if there exists a′ ∈ f Ae such that aa′ = e and a′a = f . It is easy to see that a′ is unique and has a as its (f, e)-inverse. We hence write a′ = (eaf )◦, or just a′ = a◦ when e, f are understood from the context. Notice that there exists an (e, f )- invertible element if and only eA ∼= f A, in which case we write e ∼ f . Indeed, left multiplication by an (e, f )-invertible element gives an isomorphism from f A to eA, and any isomorphism f A → eA is easily seen to be of this form. Lemma 3.1. Let e, f ∈ A be idempotents, and set x := x + Jac(A) ∈ A/ Jac(A) for all x ∈ A. Then a ∈ eAf is (e, f )-invertible if and only if a is (e, f )-invertible. Proof. We only show the non-trivial direction. Assume a has an (e, f )-inverse b with b ∈ f Ae. Then ab ∈ (eAe)×, hence ab ∈ eAe×. (This follows from the easy fact that Jac(A) ∩ eAe = e Jac(A)e = Jac(eAe).) Likewise, ba ∈ f Af ×. Let c be the inverse of ab in eAe, and let d be the inverse of ba in f Af . Then a(bc) = e, (db)a = f , and db = (db)e = (db)a(bc) = f (bc) = bc. Thus, a′ = bc = db is an (e, f )-inverse of a. (cid:3) Let e ∈ A be an idempotent, let y ∈ P e, and let c ∈ eσ β(y)e = β(y, y) + eσΛe be (eσ, e)-invertible. We define sy,e,c : P → P by sy,e,c(x) = x − y · c◦ · hβ(y, x) . 1 For an integral domain R, the rank an R-module M is dimK M ⊗R K, where K is the fraction field of R. WITT'S EXTENSION THEOREM 9 Observe that sy,e,c is completely determined by the class [β]. Following [20], we call sy,e,c an e-reflection. We will also use the name quasi-reflection, which does not restrict us to a particular idempotent e. A reflection of (P, [β]) is a 1-reflection. When we want to stress the quadratic form [β], we shall write s[β] y,e,c instead of sy,e,c. Remark 3.2. Reiter's definition of e-reflections ([20, Df. 1.2]) is essentially the same, except that he assumes e = eσ (in which case c◦ is just the inverse of c in eAe). The generalization defined here will play a crucial role later in the text. Proposition 3.3. In the previous setting, sy,e,c is an isometry of (P, [β]). inverse is sy,e,cσu. Its Proof. This is similar to the proof of [20, Pr. 1.3]; replace the usual inverses with (eσ, e)-inverses. (cid:3) Lemma 3.4. Let e, f ∈ A be idempotents. (i) If e ∼ f , then e-reflections and f -reflections coincide. (ii) If ef = f e = 0, then the composition of an e-reflection and an f -reflection is an (e + f )-reflection. Specifically, sy,e,csz,f,d = sy+z,e+f,c+d+h(y,z). Proof. (i) Let a ∈ eAf be an (e, f )-invertible element. computation to verify that sy,e,c = sya,f,aσca, which proves the claim. It is a straightforward (ii) Throughout the proof, we shall make repeated implicit usage of the fact that yf = ze = 0 and f c◦ = c◦f σ = ed◦ = d◦eσ = 0, which easily follows from ef = f e = 0. Observe first that c + d + h(y, z) ∈ eσ β(y)e + f σ β(z)f + eσh(y, z)f ⊆ (e + f )σ( β(z) + β(y) + h(y, z))(e + f ) = (e + f )σ β(z + y)(e + f ) and that c+d+h(y, z) is ((e+f )σ, e+f )-invertible with inverse c◦ +d◦−c◦h(y, z)d◦. Thus, r := sy+z,e+f,c+d+h(y,z) is an (e + f )-reflection. Now, for all x ∈ P , we have rx = x − (y + z)(c◦ + d◦ − c◦h(y, z)d◦)h(y + z, x) = x − (yc◦ + zd◦ − yc◦h(y, z)d◦)(h(y, x) + h(z, x)) = x − zd◦h(z, x) − yc◦h(y, x) + yc◦h(y, z)d◦h(z, x) = (x − zd◦h(z, x)) − yc◦h(y, x − zd◦h(z, x)) = sy,e,c(sz,f,dx) , as required. (cid:3) Lemma 3.5. Let e ∈ A be an idempotent and let x, y ∈ P e. (i) If c := hβ(x − y, x) is (eσ, e)-invertible, then sx−y,e,c(x) = y. (ii) If there exist z ∈ P e and (eσ, e)-invertible c ∈ β(z) such that d := hβ(y, w) is (eσ, e)-invertible for w = y − sz,e,c(x) = y − x + zc◦hβ(z, x), then sw,e,dsz,e,c(x) = y. Proof. This is essentially the same as the proofs of Lemma 1.4 and Lemma 1.5 in [20]; replace the usual inverses with (eσ, e)-inverses. (cid:3) 10 URIYA A. FIRST y,e,c = s[vβ] y,e,vc (note that if aσ′ Remark 3.6. (i) When applying conjugation (see Proposition 2.4) with respect to v ∈ A×, e-reflections remain e-reflections. Indeed, it is straightforward to check that s[β] (In Reiter's setting, which assumes e = eσ, e-reflections are preserved only when e commutes with v, for otherwise, e is not invariant under the conjugation of σ by v.) := vaσv−1, then (eσ′ vce)◦ = c◦v−1). (ii) Let e, f ∈ A be idempotents with e = eσ and f ∈ eAe. Then e-transfer (see Proposition 2.5) sends f -reflections of (P, [β]) to f -reflections of (P e, [βe]). Indeed, y,e,cP e = s[βe] s[β] y,e,c. 4. Witt's Extension Theorem Using methods of Reiter [20] and the notion of quasi-reflections above, we now show that every isometry between subspaces of a unimodular quadratic space over a semiperfect ring can be extended to an isometry of the whole quadratic space. Furthermore, with small exception, the resulting isometry is a product of quasi- reflections. We compare our results with those of Reiter in Remark 4.12. §4.1. General Setting. We set some general notation that will be used through- out. Let (A, σ, u, Λ) be a semiperfect unitary ring. For P ∈ P(A), set P := P/P Jac(A). In particular, A = A/ Jac(A). We shall occasionally view P as a right A-module. The image of x ∈ P in P will be denoted by x. Note that P, Q ∈ P(A) are isomorphic if and only if P ∼= Q, because P is a projective cover of P and projective covers are unique up to isomorphism. Let Λ = {λ λ ∈ Λ} and let σ be the map induced by σ on A. Then (A, σ, u, Λ) is a semisimple unitary ring, hence, by Proposition 2.8, it factors into a product (A, σ, u, Λ) ∼= ℓY i=1 (Ai, σi, ui, Λi) with each (Ai, σi, ui, Λi) a simple artinian unitary ring (see §2.3). We write Ai = Mni(Di) with Di a division ring, or a product of a division ring and its opposite. By Proposition 2.6, for every 1 ≤ i ≤ ℓ, there exists vi ∈ A× such that the conjugation (see Proposition 2.4) of (σi, ui, Λi) by vi is standard (see §2.3). Choose v ∈ A× whose image in Ai is vi. Then, by conjugating (σ, u, Λ) with v, we may assume that (Ai, σi, ui, Λi) is in standard form for all i. Note that conjugation preserves quasi-reflections by Remark 3.6(i), so this is allowed if our goal is to prove that certain isometries extend to a product of quasi-reflections. i Now, let εi denote the standard matrix unit e11 in Ai. Then ε1, . . . , εℓ are orthogonal σ-invariant idempotents. Since Jac(A) is idempotent lifting, we can lift ε1, . . . , εℓ to orthogonal idempotents e1, . . . , eℓ ∈ A. The idempotents ei may not be invariant under σ, but we have eiA ∼= eσ i A as right A-modules (because eiA = eσ i A), and hence ei ∼ eσ i . Next, we set (A(i), σ(i), u(i), Λ(i)) = (εiAiεi, σiεiAiεi , εiui, εiΛiεi) . Note that by Proposition 2.12, (Ai, σi, ui, Λi) is split-orthogonal if and only if (A(i), σ(i), u(i), Λ(i)) is split-orthogonal, and in this case, A(i) is a field, σ(i) = idA(i) , u(i) = 1 and Λ(i) = 0. Also note that A(i) ∼= Di. WITT'S EXTENSION THEOREM 11 Let (P, [β]) be a quadratic space over (A, σ, u, Λ). Then (P, [β]) gives rise to a quadratic space (P , [β]) over (A, σ, u, Λ); the map β is defined by β(x, y) = β(x, y) ∀ x, y ∈ P . Since (A, σ, u, Λ) factors into a product of unitary rings, the datum of (P , [β]) is equivalent to the datum of quadratic spaces (Pi, [βi])ℓ i=1. Specifically, if we write P = Qℓ i=1 Pi with Pi a right Ai-module, then βi is just the restriction of β to the copy of Pi in P . We further set P(i) = Piεi ∈ P(A(i)) and let β(i) = (βi)εi := βiP(i)×P(i) . Recall from Proposition 2.5 that (P(i), [β(i)]) is a quadratic space over (A(i), σ(i), u(i), Λ(i)), which is the εi-transfer of (Pi, βi). As a result, we have i=1 over (Ai, σi, ui, Λi)ℓ Next, let [βi] = [0] ⇐⇒ [β(i)] = [0] . πi : P → Pi denote the map sending x ∈ P to its image in Pi. We clearly have πi(P ei) = P(i) and πiβ(x, y) = βi(πix, πiy). Keeping the previous setting, let Q ⊆ P be a submodule. Then hβ induces a map LQ = L[β],Q : P → Q∗ := HomA(Q, A) sending x ∈ P to [y 7→ hβ(x, y)] ∈ Q∗. This map is evidently onto when (P, [β]) is unimodular and Q is a summand of P . Lastly, set Q⊥ = Q⊥[β] = {x ∈ P : hβ(x, P ) = 0} , i , we have and let βQ denote the restriction of β to Q × Q. §4.2. Lemmas. Before giving the main result, we collect several lemmas. Lemma 4.1. Let P ∈ P(A). Then P ∼= P ∗ ⇐⇒ P is isomorphic to a direct sum of copies of the modules e1A, . . . , eℓA. (See §2.1 for the definition of P ∗.) Proof. It is straightforward to check that (eiA)∗ ∼= eσ i A, and since ei ∼ eσ i A ∼= eiA. This settles the direction "⇐=", so we turn to show the converse. eσ Without loss of generality, Di is a division ring for i ≤ t, and a product of a divi- sion ring and its opposite otherwise. In the latter case, εi can be written as a sum of t+1, . . . , δσ two orthogonal idempotents εi = δi + δσ ℓ is a system of orthogonal idempotents in A, hence it can be lifted to a system of orthogonal idempotents f1, . . . , f2ℓ−t in A. Moreover, f1A1, . . . , f2ℓ−tA are the indecomposable projective right A-modules, up to isomorphism (see §2.5). Assume that P ∼= P ∗. Then P ∼= L fiAni i with (ni)2ℓ−t i=1 uniquely determined. It is straightforward to check that (fiAi)∗ ∼= f σ i Ai. By the way we have constructed f1, . . . , f2ℓ−t, we have f σ i ∼ fi+(ℓ−t) if t < i ≤ ℓ. Thus, ni = ni+(ℓ−t) for all t < i ≤ ℓ. Since eiA ∼= fiA ⊕ fi+(ℓ−t)A for t < i ≤ ℓ, it follows that P ∼= Lℓ Lemma 4.2. Let 1 ≤ i ≤ ℓ, let (P, [β]) be a quadratic space, and let Q, V ⊆ P summands of P . Then: i ∼ fi if i ≤ t, and f σ i . Now, ε1, . . . , εt, δt+1, . . . , δℓ, δσ i=1(eiA)ni , as required. (cid:3) (i) L[β],Q(V ) = Q∗ =⇒ L[β(i)],Q(i) (V(i)) = Q∗ (i). (ii) [β] is unimodular =⇒ [βi] and [β(i)] are unimodular. (iii) V ⊥[βi] and V ⊥[βi] = V ⊥[βi] εi = V . ⊥[β(i)] (i) i (i) i 12 URIYA A. FIRST (Here, Q(i) = πi(Q)εi and V(i) = πi(V )εi.) Proof. (i) Let ψ ∈ HomA(i) (Q(i), A(i)) = HomεiAiεi (Qiεi, εiAεi). As explained in the proof of Proposition 2.5, the functor U 7→ U εi : P(Ai) → P(εiAiεi) is an equivalence, hence there exists unique ψ′ ∈ HomAi(Qi, εiAi) with ψ′Q(i) = ψ. Since Q is projective, there is ϕ ∈ HomA(Q, eσ i A) with πi ◦ ϕ = ψ′ ◦ πi. By assumption, there is x ∈ V such that L[β],Q(x) = ϕ, and by replacing x with xei we may assume x = xei ∈ Viei. Now, for all y ∈ Qei, we have L[β(i)],Q(i) (πix)(πiy) = β(i)(πix, πiy) = βi(πix, πiy) = πi(β(x, y)) = πi(L[β],Q(x)(y)) = πi(ϕy) = ψ(πiy), so L[β(i)],Q(i) (πix) = ψ. (ii) Putting Q = V = P in (i) implies that L[β(i)] is onto. Since P ∗ (i) and P(i) have the same length, L[β(i)] is also injective, so [β(i)] is unimodular. By Proposition 2.5, [β] is also unimodular. (iii) This is routine. Use the fact that V(i)Ai = ViAiεiAi = ViAi = Vi. (cid:3) Lemma 4.3. Let P ∈ P(A), and assume P ∗ = U ⊕ U ′. Then P factors as a direct sum Q ⊕ Q′ such that U = {f ∈ P ∗ : f (Q′) = 0} and U ′ = {f ∈ P ∗ : f (Q) = 0}. Proof. Let ω = ωP : P → P ∗∗ (see §2.1 for the definition). We identify P ∗∗ with U ∗ ⊕ U ′∗ via g 7→ (gU , gU ′). Let Q = ω−1(U ∗) and Q′ = ω−1(U ′∗). We clearly have P = Q ⊕ Q′. Furthermore, for f ∈ P ∗, we have f ∈ U ⇐⇒ ψf = 0 for all ψ ∈ U ′∗ ⇐⇒ (ωx)f = 0 for all x ∈ Q′ ⇐⇒ (f x)σu = 0 for all x ∈ Q′ ⇐⇒ f x = 0 for all x ∈ Q′. Thus, U = {f ∈ P ∗ : f (Q′) = 0}, and likewise, U ′ = {f ∈ P ∗ : f (Q) = 0}. Lemma 4.4. Assume A ∼= F2 × F2 and σ is the exchange involution (this forces u = 1 and Λ = {0, 1}). Let (P, [β]) be a quadratic space over A, let V ⊆ P be a submodule, and let x ∈ P , z ∈ V be such that hβ(x, z) = 1. Then the following conditions are equivalent: (cid:3) (a) There is no z′ ∈ V with hβ(x, z′) = 1 and β(z′) = Λ = {0, 1}. (b) hβ(z, z) = 1 and [βz⊥∩V ] = [0]. In this case, LV (V ) ∼= A. Proof. For the equivalence (a) ⇐⇒ (b), see [20, Lm. 3.8c]. (In Reiter's notation, Hf = H ∩ f ⊥ and N stands for the zero form [0].) When (b) holds, we have V = zA⊕ (z⊥∩ V ) since x = (zhβ(z, x)) + (x− zhβ(z, x)) for all x ∈ V . This means that LV (V ) = LV (zA), which is easily seen to imply LV (V ) ∼= (zA)∗ ∼= A∗ ∼= A. (cid:3) §4.3. Witt's Extension Theorem. We are now in position to phrase and prove an analogue of Witt's Extension Theorem (Theorems 4.5 and 4.8). Following are several immediate (and less technical) corollaries. We compare our results with those of Reiter [20] in Remark 4.12 below. Theorem 4.5. Let (P, [β]) be a quadratic space, let Q, S, V be summands of P , and let ψ : (Q, [βQ]) → (S, [βS]) be an isometry. Assume that the following holds: (1a) LQ(V ) = Q∗ and LS(V ) = S∗. (1b) ψx − x ∈ V for all x ∈ Q. (1c) Q ∼= Q∗ and S ∼= S∗. (2a) If (Ai, σi, ui, Λi) is split-orthogonal, then Qi = 0 or [βiVi] 6= 0. (2b) If Di ∼= F2 and (Ai, σi, ui, Λi) is split-orthogonal, then [βiVi∩V ⊥ i ] 6= [0]. WITT'S EXTENSION THEOREM 13 and [βiz⊥∩Vi] = [0]. (2c) If Di ∼= F2 × F2, then there is no z ∈ V(i) = Viei satisfying hβi(z, z) = εi Then there is ϕ ∈ O(P, [β]) such that ϕ is a product of quasi-reflections taken with respect to elements of V , and ϕQ = ψ. (The former implies ϕx − x ∈ V for all x ∈ P .) We first prove the following special case. Lemma 4.6. Theorem 4.5 holds when Q ∼= eiA for some 1 ≤ i ≤ ℓ. In fact, in this case, ϕ can be taken to be a product of ei-reflections. Proof. The proof is based on reduction to the proof of [20, Th. 4.1], with certain modifications (particularly in the case Di ∼= F2 × F2). Throughout, h = hβ. Write e = ei. Since Q ∼= eA, there is x ∈ Q with xe = x and Q = xA. Let y = ψx. It is enough to find a product of e-reflections of (P, [β]), taken with respect to elements of V , that sends x to y. We shall prove this by applying Lemma 3.5, except maybe in the case Di ∼= F2 × F2. If h(x−y, x) is (eσ, e)-invertible, then sx−y,e,h(x−y,x)(x) = y (and x−y = x−ψx ∈ V by (1b)). If not, then it is enough to find z ∈ V ei and (eσ, e)-invertible c ∈ β(z) such that Φ(z, c) := h(y, y − x + zc◦h(z, x)) = h(y, y − x) + h(y, z)c◦h(z, x) is (eσ, e)-invertible, in which case we shall have sw,e,Φ(z,c)sz,e,c(x) = y for w = (x − y) + zc◦h(z, x) ∈ V ei. Note that h(x − y, x) = h(x, x) − h(y, x) = h(y, y) − h(y, x) = h(y, y − x) , so we may henceforth assume that h(y, y − x) is not (eσ, e)-invertible. to find z ∈ V(i) = V ei and c ∈ β(i)(z) ∩ A× where Reducing everything modulo Jac(A) (using Lemma 3.1), we see that it is enough (i) such that Φ(z, c) is invertible in A(i), Φ(z, c) = hβ(i) (y, y − x) + hβ(i) (y, z)c◦hβ(i) (z, x) . (Use Lemma 4.2 and Proposition 2.12.) We are now reduced to the setting of steps 2 -- 5 in the proof of [20, Th. 4.1] (applied to (P(i), [β(i)]) and (A(i), σ(i), u(i), Λ(i))), in which the existence of z and c as above is shown, except maybe in the case A(i) ∼= F2 × F2. Assume henceforth that A(i) ∼= F2× F2 (which means σ(i) is the exchange involu- tion). Let ε := εi = 1A(i) . It is shown in cases 2 and 4 of the proof of [20, Th. 4.1], that either there are z and c as above, or there are f ′, g′ ∈ V(i) with ε ∈ β(i)(f ′) and [β(i)] ε ∈ β(i)(g′) such that s g′,ε,ε(x) = y. In the former case we are done. In the latter case, let f, g ∈ V ei be such that f = f ′ and g = g′, and choose c ∈ eσ β(f )e [β(i)] f ′,ε,εs Now, conditions (1a) and (2a) -- (2c) imply (respectively) that: We may assume that hβ(i) (y, y − x) is not invertible in A(i). (1a-i) LQ(i) (V(i)) = Q∗ (2a-i) If (A(i), σ(i), u(i), Λ(i)) is split-orthogonal, then [β(i)V ] 6= 0. (2b-i) If A(i) ∼= F2 and (A(i), σ(i), u(i), Λ(i)) is split-orthogonal, then [β(i)V(i)∩V ⊥ ] 6= (2c-i) If A(i) ∼= F2 × F2, then there is no z ∈ V(i) with hβ(i) (z, z) = 1 and (i) and LS(i) (V(i)) = S∗ (i). [0]. (i) [β(i)z⊥∩V(i) ] = [0] . 14 URIYA A. FIRST f,e,cs[β] f,e,cs[β] g,e,d(x) = πiy. Replacing x with s[β] and d ∈ eσ β(g)e with c = d = ε. Then c and d are (eσ, e)-invertible (Lemma 3.1), and πis[β] g,e,d(x), we may assume x = y. Under this new assumption, we establish the existence of z and c as above. Indeed, this amounts to finding z ∈ V(i) with hβ(i) (x, z) = 1 and β(i)(z) = Λ(i) = {0, 1}. By [20, Lm. 3.5], there is z′ ∈ V(i) with h(x, z′) = 1. Condition (2c-i) and Lemma 4.4 now give there required z ∈ V(i). (cid:3) Proof of Theorem 4.5. The proof is based on the inductive step of [20, Th. 4.1] with certain modifications. Step 1. If Q = 0 there is nothing to prove. Otherwise, by Lemma 4.1, eiA is isomorphic to a summand of Q for some i. Since LQ : V → Q∗ is onto and Q∗ is a projective right A-module, we can write V = W ⊕ R such that LQW is an isomorphism and LQ(R) = 0. Evidently, (eiA)∗ ∼= eσ i A ∼= eiA is also isomorphic to a summand V ′ of W . If (Ai, σi, ui, Λi) is not split-orthogonal or Di ≇ F2 × F2, we choose V ′ arbitrarily. Otherwise, we claim that V ′ can be chosen such that condition (2a), (2b) and (2c) are satisfied when V is replaced with V ′ ⊕ R. Indeed, assume (Ai, σi, ui, Λi) is split-orthogonal and Di ≇ F2. By condition (2a), we have [β(i)] 6= [0]. Choose some x ∈ V(i) with β(i)(x) 6= Λ(i) = 0, and If w = 0, then [β(i)R(i) ] 6= [0], as write x = w + r with w ∈ W(i), r ∈ R(i). required. Otherwise, wA is a summand of W (because A is semisimple) having eiA as projective cover. Thus, W has a summand V ′ ∼= eiA whose image in W is wA. That V ′ satisfies [β(i)V ′ The same strategy also works when Di ∼= F2 by Lemma 4.2(iii); start with x ∈ V ⊥ When Di ∼= F2× F2, by Lemma 4.4 and conditions (2c) and (1a), there is z ∈ V(i) and x ∈ Q(i) such that hβ(i) (x, z) = εi and β(i)(z) = Λ(i) = {0, εi}. Write z = w + r with w ∈ W(i), r ∈ R(i), and choose V ′ as above. (We must have wA ∼= εiA. Otherwise, there is an idempotent δ ∈ A(i) ∼= F2 × F2 such that δ 6= εi and wδ = w, which implies εi = hβ(i) (x, z) = hβ(i) (w + r, z) = hβ(i) (w, z) = δσhβ(i) (w, z) ∈ (i) ⊕ R(i) such that β(i)(z) = {0, εi} δσA(i), a contradiction.) Now, we have z ∈ V ′ and hβ(i) (x, z) = εi, so condition (2c) holds for V ′ ⊕ R by Lemma 4.4. This settles our claim about the choice of V ′. (i)⊕R(i) ] 6= [0], as required. i ∩ Vi)εi such that β(i)(x) 6= 0. (i) ∩ V(i) = (V ⊥ Step 2. Write W = V ′ ⊕ V ′′, and let U ′ = LQ(V ′) and U ′′ = LQ(V ′′). Then Q∗ = U ′ ⊕ U ′, so by Lemma 4.3, we have a decomposition Q = Q′ ⊕ Q′′ such that LQ(V ′) = {f ∈ Q∗ : f (Q′′) = 0} = Q′∗ and LQ(V ′′) = {f ∈ Q∗ : f (Q′) = 0} = Q′′∗. Now, Q′∗ ∼= V ′ ∼= eiA, hence Q′ ∼= Q′∗∗ ∼= (eiA)∗ ∼= eiA, so Q′ ∼= Q′∗. Thus, Q′ ⊕ Q′′∗ ∼= (Q′ ⊕ Q′′)∗ = Q∗ ∼= Q ∼= Q′ ⊕ Q′′, which implies Q′′∗ ∼= Q′′ (see the discussion at the end of §2.5). Thus, we may apply induction to (Q′′, [βQ′′ ]) and ψQ′′ . This yields ϕ ∈ O(P, [β]), a product of reflections taken with respect to elements of V , such that ϕQ′′ = ψQ′′ . Step 3. We now claim that conditions (1a) -- (1c) and (2a) -- (2c) hold for the isometry ϕ−1ψQ : Q′ → ϕ−1ψ(Q′) and the module V ′⊕ R (in place of V ). Indeed, (2a) -- (2c) hold by our choice of V ′ (cf. Step 1), and (1c) was verified in Step 2. WITT'S EXTENSION THEOREM 15 Next, notice that by construction, an element z ∈ V lives in V ′ ⊕ R if and only if LQ(z) ∈ Q′∗, i.e. 0 = LQ(z)(Q′′) = h(z, Q′′). Now, for all y ∈ Q′′, we have h(ϕ−1ψx − x, y) = h(ϕ−1ψx, y) − h(x, y) = h(ψx, ϕy) − h(x, y) = h(ψx, ψy) − h(x, y) = h(x, y) − h(x, y) = 0 . This implies that ϕ−1ψx − x ∈ V ′ ⊕ R for all x ∈ Q′, so (1b) holds. Finally, we have LQ′ (V ′ ⊕ R) = Q′∗ by construction, so to show (1a) amounts to showing Lϕ−1ψQ′ (V ′ ⊕ R) = (ϕ−1ψQ′)∗. Let f ∈ (ϕ−1ψQ′)∗. We extend f to ϕ−1S = ϕ−1ψQ = ϕ−1ψQ′ ⊕ ϕ−1ψQ′′ = ϕ−1ψQ′ ⊕ Q′′ by setting it to be 0 on Q′′. View f ϕ−1 as an element of S∗. By (1a) (for Q, S, V ), there is x ∈ V with LS(x) = f ϕ−1. Now, for all y ∈ ϕ−1ψQ, we have h(ϕ−1x, y) = h(x, ϕy) = LS(x)(ϕy) = f ϕ−1ϕy = f y, so Lϕ−1ψQ′ (ϕ−1x) = f . Since ϕ is a product of reflections taken with respect to element of V , ϕ(V ) = V , hence ϕ−1x ∈ V . In addition, when y ∈ Q′′, we have h(ϕ−1x, y) = f y = 0, so ϕ−1x ∈ V ′ ⊕ R. This shows that f ∈ Lϕ−1ψQ′(V ′ ⊕ R). Step 4. To finish, we apply Lemma 4.6 to ϕ−1ψQ : Q′ → ϕ−1ψ(Q′) with V ′ ⊕ R in place of V to get a product η of ei-reflections, taken with respect to elements of V ′ ⊕ R, such that ηQ′ = ϕ−1ψ. Since LQ′′ (V ′ ⊕ R) = 0, reflections taken with respect to elements of V ′ ⊕ R fix Q′′. Thus, ηQ′′ = idQ′′ , so ηQ = ϕ−1ψ. This means ψ = ϕηQ and we are done. Remark 4.7. Theorem 4.5 actually holds when (2c) is replaced with the milder conditions: (2c′) If Di ∼= F2 × F2 and ni = 1, then there is no z ∈ V(i) = Vi satisfying (2c′′) If Di ∼= F2 × F2 and ni > 1, then Qi ≇ εiAi. The idea is to extend Lemma 4.6 to the case Di ∼= F2 × F2, ni > 1, and Q ∼= e′ iA, where e′ i = e11 + e22 ∈ Ai. The proof requires using both ei-reflections (to get x = y) and e′ i-reflections (to find z and c; use [20, In case Lm. 3.6]). The inductive step of Theorem 4.5 should also be modified. Di ∼= F2 × F2 and ni > 1, one chooses V ′ isomorphic to e′ iA rather than eiA. If this results in Q′′ iA into two copies of eiA and apply the induction hypothesis to the direct sum of Q′′ and one of these copies. The full and detailed argument would overload the document while not benefiting any result beside Theorem 4.5 and Corollary 4.9 below, so we chose to omit it. i ∼= εiAi, then one should split Q′ ∼= e′ hβi(z, z) = εi = 1Ai and [βiz⊥∩Vi] = [0]. i is a lifting of the idempotent ε′ (cid:3) If we do not require in Theorem 4.5 that the extension of ψ will be a product of quasi-reflections, then conditions (1c), (2a), (2b), (2c) can be dropped. Theorem 4.8. Let (P, [β]) be a quadratic space, let Q, S, V be summands of P , and let ψ : (Q, [βQ]) → (S, [βS]) be an isometry. Assume that (1a) LQ(V ) = Q∗ and LS(V ) = S∗, and (1b) ψx − x ∈ V for all x ∈ V . Then ψ extends to an isometry ϕ ∈ O(P, [β]) such that ϕx − x ∈ V for all x ∈ P . Proof. The proof is essentially the same as the proof of [20, Th. 6.2]. However, since there are some differences in the conditions to be checked, we recall the argument. Assume first that (1c) holds. Let T be a free A-module with basis {z, w}. We choose a ∈ A such that: 16 URIYA A. FIRST • ai := πia = 0 if Di ≇ F2 × F2, or D ∼= F2 × F2 and L[β(i)],V(i) (V(i)) ≇ A(i). • ai := πia ∈ A(i) \ {0, εi} if Di ∼= F2 × F2 and L[β(i)],V(i) (V(i)) ∼= A(i). We define a sesquilinear form γ : T × T → A by linearly extending γ(z, z) =0 γ(z, w) =1 γ(w, z) =0 γ(w, w) =a Let (P ′, [β′]) = (P, [β]) ⊥ (T, [γ]), Q′ = Q ⊕ zA, S′ = S ⊕ zA, V ′ = V ⊕ (z + w)A and define ψ′ : Q′ → S′ by ψ′(x⊕ zc) = ψx⊕ zc (c ∈ A). It is easy to check that ψ′ is an isometry from [β′Q′ ] to [β′S ′ ]. Furthermore, the conditions of Theorem 4.5 hold for (P ′, [β′]), Q′, S′, V ′, ψ′: (1a) holds since hβ ′(0 ⊕ (z + w), 0 ⊕ z) = u ∈ A×, (1b) is straightforward, (1c) holds since Q′∗ ∼= Q∗ ⊕ A∗ ∼= Q ⊕ A ∼= Q′, (2a) holds since γ(i)(πi(zei + wei)) = εi + ai + Λ(i), (i) ∩ V ′⊥ (2b) holds since πi(0 ⊕ (zei + wei)) ∈ V ′ (2c) holds by Lemma 4.4, because if Di ∼= F2 × F2, then L[β ′ hγi(πi(zei + wei), πi(zei + wei)) = 2εi = 0), and Indeed, we have L[β ′ hγ(i) (πi(zei + wei), πi(zei + wei)) = ai + aσ i . (i)],V ′ (i) (i) when Di ∼= F2 (because then (V ′ (i)) ≇ A(i). (i)) ∼= L[β(i)],V(i) (V(i)) ⊕ (ai + aσ (V ′ i )A because (i)],V ′ (i) Thus, ψ′ extends to ϕ′ ∈ O(P ′, [β′]) with ϕ′x − x ∈ V ′ for all x ∈ P ′. We finish by showing that ϕ′(P ⊕ 0) = P ⊕ 0. Indeed, for x ⊕ 0 ∈ P ⊕ 0, we can write ϕ′(x⊕ 0) = y⊕ (z + w)c for some y ∈ P , c ∈ A. This implies 0 = hβ ′(0⊕ z, x⊕ 0) = hβ ′(ϕ′(0⊕z), ϕ′(x⊕0)) = hβ ′(0⊕z, y⊕(z +w)c) = c, so c = 0 and ψ′(x⊕0) ∈ P ⊕0. When (1c) does not hold, choose U ∈ P(A) such that (Q ⊕ U )∗ ∼= Q ⊕ U . Define a sesquilinear form γ on U ⊕ U ∗ by γ(x ⊕ f, y ⊕ g) = f y (x, y ∈ U , f, g ∈ U ∗; (U ⊕ U ∗, [γ]) is the hyperbolic quadratic space associated with U ). Now, let (P ′, [β′]) = (P, [β])⊕ (U ⊕ U ∗, [γ]), Q′ = Q⊕ U ⊕ 0, S′ = S⊕ U ⊕ 0, V ′ = V ⊕ 0⊕ U ∗, and define ψ′ : Q′ → S′ by ψ′(x ⊕ z ⊕ 0) = ψx ⊕ z ⊕ 0 (x ∈ Q, z ∈ U ). Conditions (1a), (1b), (1c) are easily seen to hold for (P ′, [β′]), Q′, S′, V ′, ψ′, so by what we have shown above, ψ′ extends to an isometry ϕ′ ∈ O(P ′, [β′]) with ϕ′x − x ∈ V ′. Similar arguing implies that ϕ′ maps P into itself. (cid:3) The following corollaries can be viewed as analogues of Witt's Extension Theo- rem and Witt's Cancellation Theorem. Corollary 4.9. Let (P, [β]) be a quadratic space and let Q, S be summands of P . Assume that (P, [β]) or (Q, [βQ]) is unimodular. Then any isometry of ψ : (Q, [βQ]) → (S, [βS]) extends to an isometry ϕ ∈ O(P, [β]). Proof. Take V = P in Theorem 4.8. Condition (1b) is clear, and condition (1a) follows from the unimodularity assumption. (cid:3) Corollary 4.10. Let (P, [β]), (Q1, [γ1]), (Q2, [γ2]) be quadratic spaces such that (P, [β]) is unimodular and (P, [β]) ⊥ (Q1, [γ1]) ∼= (P, [β]) ⊥ (Q2, [γ2]). Then (Q1, [γ1]) ∼= (Q2, [γ2]). Proof. Identify (Z, [ζ]) := (P, [β]) ⊥ (Q1, [γ1]) with (P, [β]) ⊥ (Q2, [γ2]). Corol- lary 4.9 implies that any isometry between the two copies of (P, [β]) in (Z, [ζ]) extends to an isometry of (Z, [ζ]). This isometry must map (Q1, [γ1]) isometrically onto (Q2, [γ2]). (cid:3) WITT'S EXTENSION THEOREM 17 Corollary 4.11. Let (P, [β]) be a unimodular quadratic space. Assume that (1) if (Ai, σi, ui, Λi) is split-orthogonal, then Di ≇ F2, and (2) if Di ∼= F2 × F2, then Pi ≇ εiAi. Then O(P, [β]) is generated by quasi-reflections. Proof. It is enough to show that the conditions of Theorem 4.5 hold when we take V, Q, S to be P . Indeed, conditions (1a) and (1b) hold as in the proof of Corollary 4.9. Condition (1c) holds because L[β] : P → P ∗ is an isomorphism. To see (2a), observe that by Lemma 4.2(ii), LP(i) : P(i) → P ∗ (i) is an isomorphism, and hence either Pi = 0 or [βi] 6= 0. Condition (2b) follows from assumption (1). Finally, assumption (2) and the unimodularity of [β] imply, LV(i) (V(i)) = LP(i) (P(i)) = P ∗ (cid:3) (i) ∼= P(i) ≇ A(i), so condition (2c) holds by Lemma 4.4. Remark 4.12. Theorems 4.5 and 4.8 should be compared with Theorems 6.1 and 6.2 of [20] (and also Theorems 4.1, and 5.1 in [20]). Theorem 6.1 in [20] is similar to Theorem 4.5, but it applies to the broader class of semilocal rings, and it derives the stronger conclusion that ϕ is is a product of reflections (rather than quasi-reflections). However, it assumes the stronger assumptions (1c′) Q and S are free, (2a′) if (Ai, σi, ui, Λi) is split-orthogonal, then [βiQ⊥ in place of (1c) and (2a), and condition (2c) is replaced with condition (2c′) of Remark 4.7 (note that (1c′) and (2c′) imply (2c′′)). In section 5, we shall give a precise description of which isometries are products of reflection (in case A is semiperfect) from which it will follow that condition (2a′) cannot be completely removed in general. i ∩Vi] 6= [0], Theorem 6.1 in [20] resembles Theorem 4.8. It applies to semilocal unitary rings, but it assumes condition (2a′) in addition to (1a) and (1b). This additional assumption does not allow one to derive Corollaries 4.9 and 4.10 from [20, Th. 6.2], and in fact, these corollaries are false for semilocal rings, as demonstrated by Keller [11, §2]. Nevertheless, [20, Th. 6.2] still implies Corollary 4.10 in case [(γ1)i] 6= 0 whenever (Ai, σi, ui, Λi) is split-orthogonal. See also [11, Th. 3.4] for different proof and strengthening of this statement. Remark 4.13. Condition (2b) in Theorem 4.5 (and also [20, Th. 6.1]) cannot be completely removed, even when (P, [β]) is unimodular. Indeed, over F2, there is a unimodular quadratic form of dimension 4 whose isometry group is not generated by quasi-reflections (which are just reflections in this case). However, up to isomor- phism, this is the only exception over F2 (see [24, Cr. 11.42], for instance). We do not know if condition (2c) can be removed in general. §4.4. Applications. We now use the previous results to derive several conse- quences to hermitian categories and systems of sesquilinear forms. Hermitian cat- egories are a purely categorical framework to work with quadratic forms. We refer the reader to [19, §1], [23, Ch. 7] or [15, Ch. II] for the relevant definitions. Let (C ,∗, ω) be a hermitian category with form parameter (ε, Λ). In [19, §3.4], Quebbemann, Scharlau and Schulte prove that unimodular (ε, Λ)-quadratic forms over C cancel from orthogonal sums, provided the following assumptions hold: (A) All idempotent morphisms in C split. 18 URIYA A. FIRST (B) Every object of C is a direct sum of finitely many objects with local endo- morphism ring (or, alternatively, the endomorphism ring of every object in C is semiperfect). (C) For every M ∈ C , E := EndC (M ) is complete in the Jac(E)-adic topology. After proving this consequence, the authors comment that condition (C) is in fact unnecessary since one can apply transfer (see [19, Pr. 2.4]) to translate the prob- lem to cancellation of quadratic forms over semiperfect unitary rings, and then apply Reiter's Theorem 6.2 in [20]. This also implies that condition (A) is unnec- essary. However, the authors seem unaware that Reiter's Theorem does not imply cancellation in general (see Remark 4.12), and hence this argument implies cancel- lation only in certain cases (e.g. when all three (ε, Λ)-quadratic spaces involved are unimodular and their base objects are progenerators of C ). Nevertheless, by replacing [20, Th. 6.2] with Theorem 4.8 (or Corollary 4.10), we see that condition (C) can indeed be dropped. We have therefore obtained: Corollary 4.14. Assume condition (B) is satisfied. Then unimodular (ε, Λ)- quadratic forms over C cancel from orthogonal sums. By the same methods (i.e. via transfer in the sense of [19, Pr. 2.4]), Corollary 4.9 implies: Corollary 4.15. Assume condition (B) is satisfied, let (P, [β]) be a unimodular (ε, Λ)-quadratic space, and let Q, S be summands of P . Then any isometry ψ : (Q, [βQ]) → (S, [βS]) extends to an isometry ϕ ∈ O(P, [β]). We now combine Corollary 4.14 with results from [5], [7] and [6] to obtain can- cellation results for systems of (not-necessarily unimodular) sesquilinear forms. Henceforth, let R be a commutative ring, and let C be an R-category equipped with R-linear hermitian structures {∗i, ωi}i∈I ; see [6, §2.4, §4] for the definitions. Corollary 4.16. Assume that R is local and henselian, 2 ∈ R×, and at least one of the following holds: (1) R is noetherian and EndC (M ) is finitely generated as an R-module for all (2) R is a valuation ring, and EndC (M ) is R-torsion-free and has finite rank M ∈ C . for all M ∈ C . Then systems of sesquilinear forms over (C ,{∗i, ωi}i∈I ) cancel from orthogonal sums. Proof. This is similar to the proof of [6, Th. 5.2]. By [6, Th. 4.1], the category of sesquilinear forms over (C ,{∗i, ωi}i∈I ) is equivalent to the category of unimod- ular 1-hermitian forms over another hermitian category (D,∗, ω), where D is the category Aer2I (C ) constructed in [6, §4]. It therefore enough to prove that uni- modular 1-hermitian forms over (D,∗, ω) cancel from orthogonal sums. Note that since 2 ∈ R×, there is only one form parameter (1, Λ) for (D,∗, ω), and 1-hermitian forms and (1, Λ)-quadratic forms are essentially the same. By construction, the en- domorphism ring of an object Z ∈ D is a sub-R-algebra of EndC (M )× EndC (N )op for some M, N ∈ C , so by Corollary 4.14, it is enough to check that such rings are semiperfect. This is indeed the case by Proposition 2.13. (cid:3) As a special case of Corollary 4.16, we get: WITT'S EXTENSION THEOREM 19 Corollary 4.17. Let A be an R-algebra and let {σi}i∈I be a system of R-involutions on A. Assume that is R local and henselian, 2 ∈ R×, and at least one of the following holds: (1) R is noetherian and A is finitely generated as an R-module. (2) R is a valuation ring, and A is R-torsion-free of finite rank. Then systems of sesquilinear forms over (A,{σi}) (that are defined on f.g. projective A-modules) cancel from orthogonal sums. Finally, we use [5] to show that non-unimodular hermitian forms over non- henselian valuation rings cancel from orthogonal sums. Here, the prefix "non-" stands for "not-necessarily" and not for absolute negation. Corollary 4.18. Let (R, σ) be an arbitrary valuation ring with involution, and let u ∈ R be such that uσu = 1. Suppose that there exists r ∈ R with r + rσu ∈ R× (e.g. if 1 + u ∈ R×). Then non-unimodular hermitian forms over (R, σ) cancel from orthogonal sums. Proof. Write v = r + rσu. Then v ∈ R× and vσu = v. By conjugating with v−1 (see Proposition 2.4) and replacing r with rv−1, we may assume that u = 1 and r + rσ = 1. By [5, Th. 4.1], the category of arbitrary 1-hermitian forms over (R, σ) is equiv- alent to the category of unimodular 1-hermitian forms over a hermitian category (D,∗, ω) (see [5, §3] for its definition). The category D is the category of morphisms in P(R), denoted Mor(P(R)). (One can also use [6, Th. 3.2] to get essentially the same result.) Notice that the Hom-sets of D are R-modules and it not hard to check that (f · a)∗ = f · aσ for every morphism f in D and a ∈ R. Since r + rσ = 1, the map (Q, h) 7→ (Q, [rh]) taking 1-hermitian forms over D to (1, Λmin)-quadratic spaces over D is an isomorphism of categories (and there is only one form param- eter (1, Λ) for (D,∗, ω)). Now, by Corollary 4.14, it is enough to prove that every object of Mor(P(R)) is the direct sum of objects with local endomorphism rings. Recall that an object of Mor(P(R)) consists of a triple (P, f, P ′) such that P, P ′ ∈ P(R) and f ∈ HomR(P, P ′). The endomorphism ring of (P, f, P ′) is the set of pairs (g, g′) ∈ EndR(P ) × EndR(P ′) such that g′f = f g. Using the fact that R is a valuation ring, one can show that every (P, f, P ′) is the direct sum of objects of the form (R, 0, 0), (0, 0, R) and (R, t, R) with t 6= 0: The proof essentially boils down to showing that for every matrix f = (fij ) ∈ Mn×m(R) there are invertible matrices w ∈ Mn(R) and w′ ∈ Mm(R) such that (aij ) := wf w′ is diagonal (meaning that aij = 0 for i 6= j), and this is well-known. The endomorphism rings of (R, 0, 0), (0, 0, R) and (R, t, R) are easily seen to be R, which is local, so we are done. (cid:3) 5. Generation by Reflections Let (P, [β]) be a unimodular quadratic space defined over a semiperfect unitary ring. Denote by O′(P, [β]) the subgroup of O(P, [β]) generated by reflections. In this section, we describe O′(P, [β]) and show that, apart from an obvious exception, it is a subgroup of finite index in O(P, [β]), and that index can be determined in terms of A and P . §5.1. Dickson's Invariant. Let (A, σ, u, Λ) be an orthogonal simple artinian uni- tary ring (see §2.4), let K = Cent(A), and let (P, [β]) be a unimodular qua- dratic space. We now recall Dickson's Invariant (also called pseudodeterminant 20 URIYA A. FIRST or quasideterminant ), which is a group homomorphism ∆ = ∆[β] : O(P, [β]) → Z/2Z . This homomorphism will be used in describing what is O′(P, [β]). The map ∆ is related to the reduced norm map on E := EndA(P ) via ∀ψ ∈ O(P, [β]) , NrdE/K(ψ) = (−1)∆(ψ) (5.1) so one can use NrdE/K to define ∆ in case char K 6= 2. Unfortunately, there seems to be no reference defining Dickson's invariant di- rectly for quadratic forms over noncommutative central simple algebras (but see [24, p. 160] for the case A = K, and [4], [15, §IV.5] for arbitrary commutative rings). We therefore satisfy with giving an ad-hoc definition that suits our needs, but may seem unnatural. In the end, we shall briefly explain how to obtain a more intrinsic definition by using the Dickson invariant of quadratic pairs introduced in [16, §12C]. We start by recalling Dickson's invariant in the case A = K (the orthogonality then forces σ = idK, u = 1 and Λ = 0). In this case, (P, [β]) is just a quadratic space in the classical sense. Dickson's invariant can then be defined by ∆(ψ) = dimK im(1 − ψ) + 2Z ∀ ψ ∈ O(P, [β]) . The map ∆ is indeed a group homomorphism ([24, Th. 11.43]), and moreover, it is a morphism of algebraic groups over K ([4]). See also [16, Df. 12.12] and [10, p. 129] for alternative definitions in characteristic 2. (The equivalence of the definitions, and the identity (5.1), can be verified using the references specified. In particular, one should only verify that the definitions coincide on reflections by [24, Th. 11.39].) Assume now that A is arbitrary and let E = EndA(P ). Then E is a central simple K-algebra. We define ∆ by ∆(ψ) = dimK(1 − ψ)E deg E + 2Z ∀ φ ∈ O(P, [β]) It is easy to see that this definition agrees with the definition in the case A = K. However, we have to check that ∆ is indeed a homomorphism of groups. Proposition 5.1. If A is non-split, then ∆ ≡ 0. Otherwise, ∆ is a surjective group homomorphism, and furthermore, it is a homomorphism of algebraic groups defined over K. Proof. Assume A is non-split. If ∆(ψ) = 1 + 2Z, then deg E must be odd (because it divides dimK(1 − ψ)E). On the other hand, E is Brauer equivalent to A, which has an involution of the first kind (Proposition 2.6), and hence ind E = ind A is a power of two ([16, Cr. 2.6]). Thus, ind A = 1, i.e. A is split. (Compare with [13, Lm. 2.6.1].) Assume A is split. Observe that the definition of ∆ depends only on the isomor- phism class of the endomorphism ring E = EndA(P ), which does not change under conjugation and e-transfer (see Propositions 2.4 and 2.5). These transitions also do not affect split-orthogonality by Proposition 2.12, so we may freely apply them. Now, by conjugation, we may assume (A, σ, u, Λ) is standard (Proposition 2.6), i.e. u = 1 and σ is the matrix transpose involution. Let e be the matrix unit e11. By applying e-transfer, we may assume A = K. But in this case it is known that ∆ is a surjective group homomorphism, and also a homomorphism of algebraic groups defined over K. (cid:3) WITT'S EXTENSION THEOREM 21 We now compute the Dickson invariant of reflections. Proposition 5.2. Let (A, σ, u, Λ) be an orthogonal simple artinian unitary ring, let e ∈ A be an idempotent, and let (P, [β]) be a unimodular quadratic space over (A, σ, u, Λ) with P 6= 0. Then for every e-reflection s ∈ O(P, [β]) we have ∆(s) = deg eAe + 2Z . i=1 eii for some r. By Lemma 3.4, me replace e with Pr Proof. We assume e 6= 0; the proposition becomes trivial otherwise. Also, if A is non-split, then ind eAe = ind A is even, so we are done by Proposition 5.1. Assume A is split, i.e. A = Mn(K) where K is a field. By conjugation, we may assume that σ is the transpose involution and u = 1 (Proposition 2.6, Re- mark 3.6(i)). Let {eij} be the standard matrix units of A. Then e is equivalent to Pr i=1 eii to assume eσ = e. Now, by applying e-transfer, we may assume e = 1 (Proposition 2.5, Re- mark 3.6(ii)). We further tensor (A, σ, u, Λ) and (P, [β]) with an algebraic closure F of K, to assume K is algebraically closed. (Namely, we replace (A, σ, u, Λ) with (A ⊗K F, σ ⊗K idF , u ⊗ 1, Λ ⊗K F ), and (P, [β]) with (P ⊗K F, [βF ]), where βF is defined by βF (x⊗ a, x′ ⊗ a′) = β(x, x′)⊗ aa′ for all x, x′ ∈ P , a, a′ ∈ A. The details are left to the reader.) Let S be the set of pairs (y, λ) ∈ P × Λmin satisfying β(y, y) + λ ∈ GLn(K). Then S is open in the Zaritzki topology of P × Λ (viewed as an affine space over K), and hence connected. Since ∆ : O(P, [β]) → Z/2Z is a morphism of algebraic groups (Proposition 5.1), the map (y, λ) 7→ ∆(sy,1,β(y,y)+λ) : S → Z/2Z is continuous in the Zaritzki topology. Therefore, since S is connected, it is enough to prove that ∆(s) = deg A + 2Z = n + 2Z for some reflection s. Let {eij}i,j be the standard matrix units of A = Mn(K). By Lemma 3.4(i), every e11-reflection is an eii-reflection for any i, so by Lemma 3.4(ii), the product of n e11-reflections is a reflection. Since ∆ is a group homomorphism, it is therefore enough to prove that there is an e11-reflection s with ∆(s) = 1 + 2Z. Now, note that e11Ae11 ∼= K. Applying e11-transfer (together with Remark 3.6(ii)), we are reduced to show that when A = K, there exists a reflection s with ∆(s) = 1 + 2Z. But this well-established; see [16, Ex. 12.13], for instance. (cid:3) Remark 5.3. Another way to define ∆[β] when A is an arbitrary central simple K-algebra is by relating (P, [β]) with a quadratic pair in the sense of [16, §5B]: By [17, Pr. 4.2],2 (P, [β]) induces a quadratic pair (τ, f ) on the central simple K-algebra E := EndA(P ), and one can show that O(P, [β]) is canonically isomorphic to O(τ, f ) := {x ∈ E : xτ x = 1 and f (x−1ax) = f (a) ∀ a ∈ E} . The map ∆ : O(P, [β]) → Z/2Z can now be defined via the Dickson invariant of similitudes of quadratic pairs introduced in [16, §12C]. To check the equivalence of this definition with the one previously given, it is enough to show that conjugation and e-transfer do not affect the quadratic pair (τ, f ) constructed in [17, Pr. 4.2], and that both definitions are compatible with extending to a splitting field. Provided that, it is enough to check that the definitions coincide when A = K, and this 2 Pr. 4.2 of [17] is phrased in the case A is a division ring, but the proof works for any central simple K-algebra. 22 URIYA A. FIRST follows from [16, Ex. 12.31]. We leave all the details (which seem to be missing in the literature) to the reader. §5.2. Generation by Reflections. We now use the Dickson invariant to describe the group O′(P, [β]). Throughout, (A, σ, u, Λ) is a semiperfect (and at one point semilocal) ring. We use the same setting as in §4.1 and set some further notation. Let (P, [β]) be a quadratic space. Then any ψ ∈ O(P, [β]) gives rise to ψi ∈ O(Pi, [βi]) given by ψi(πix) = πi(ψix). Observe that if ψ is an e-reflection of (P, [β]) (with e an idempotent of A), then ψi is a πie-reflection of (Pi, [βi]). In particular, if ψ is a reflection, then so is ψi. Conversely, when the image of e in A lives in Ai, every πie-reflection of (Pi, [βi]) is induced from an e-reflection. This fact, which easily follows from Lemma 3.1, will be used repeatedly henceforth. When (Ai, σi, ui, Λi) split-orthogonal and (P, [β]) is unimodular, we define ∆i = ∆i,[β] : O(P, [β]) → Z/2Z by ∆i(ψ) = ∆[βi](ψi) (Lemma 4.2(ii) implies that (Pi, [βi]) is unimodular). It is clear that ∆i is a group homomorphism. We shall call (Ai, σi, ui, Λi) odd (resp. even) split-orthogonal if (Ai, σi, ui, Λi) is split-orthogonal and ni is odd (resp. even; recall that Ai = Mni (Di)). i Finally, let ε(j) denote the matrix unit ejj in Ai. Then {ε(j) i }i,j is a complete system of orthogonal idempotents in A, hence it can be lifted to a complete system i }i,j in A. We choose this lifting to have ei = e(1) of orthogonal idempotents {e(j) (where ei is defined as in §4.1), and set fi = Pj e(j) (so f i = 1Ai). i i Lemma 5.4. Let (Pi, [βi]) be a unimodular quadratic space over (Ai, σi, ui, Λi). (i) If (Ai, σi, ui, Λi) is not split-orthogonal, then idP is an ei-reflection of (P, [β]). (ii) If Pi 6= 0, then (Pi, [βi]) admits an εi-reflection. Proof. (i) If (Ai, σi, ui, Λi) is not split-orthogonal, then so is (A(i), σ(i), u(i), Λ(i)) (i) 6= ∅. Indeed, if A(i) ∼= E × Eop with E (Proposition 2.12). We claim that Λ(i)∩ A× a division ring, then (1E, 0op (i)u(i) ∈ Λ(i)∩A× (i). If A(i) is a division ring and Λ(i) ∩ A× (i) 6= ∅, then Λ(i) = 0. This implies, aσu(i) = a for all a ∈ A(i), and by taking a = 1, we get u(i) = 1 and σ(i) = idA(i) . But this means (A(i), σ(i), u(i), Λ(i)) is split-orthogonal, a contradiction. E )−(1E, 0op E )σ (i). Then a is (eσ i Λei such that πia ∈ Λ(i) ∩ A× Now, choose a ∈ eσ i , ei)-invertible (Lemma 3.1), hence s0,ei,a is an ei-reflection, and it is easy to see that s0,ei,a = idP . (ii) If (Ai, σi, ui, Λi) is not split-orthogonal, take the projection to Pi of the ei- reflection constructed in (i). Otherwise, A(i) is a field and Λ(i) = 0, so we need to find y ∈ P(i) := Piεi with β(i)(y) 6= 0. Indeed, [β] is unimodular, hence [βi] is unimodular (Lemma 4.2(ii)). Since Pi 6= 0, this means [βi] 6= 0, and hence [β(i)] 6= 0. As Λ(i) = 0, there must be y ∈ P(i) with β(i)(y) 6= 0, as required. Proposition 5.5. Let (P, [β]) be a unimodular quadratic space. Then the following conditions are equivalent: (cid:3) (a) (P, [β]) admits a reflection. (b) For all i, if (Ai, σi, ui, Λi) is odd split-orthogonal, then Pi 6= 0. In this case, there exist fi-reflections for all i. WITT'S EXTENSION THEOREM 23 × Proof. Observe that (P, [β]) has a reflection ⇐⇒ (P , [β]) has a reflection ⇐⇒ there exists x ∈ P such that β(x) ∩ A 6= ∅ ⇐⇒ for all i, there exists x ∈ Pi such that βi(x) ∩ A× 6= ∅ ⇐⇒ (Pi, [βi]) admits a reflection ⇐⇒ (P, [β]) has an i fi-reflection for all i. Assume (b) holds. By Lemma 5.4, if (Ai, σi, ui, Λi) is not split-orthogonal or Pi 6= 0, then (Pi, [βi]) admits an εi-reflection s. By Lemma 3.4(i), s is also an ε(j) - reflection for all j, hence by Lemma 3.4(ii), sni is a reflection of (Pi, [βi]) (because 1Ai = Pni i ). If (Ai, σi, ui, Λi) is even split-orthogonal, then Di is a field, ni is even, σi is the matrix transpose involution, and u = 1. It is then easy to see that Λi contains a unit a ∈ A× Assume (b) is false. Then there is i such that (Ai, σi, ui, Λi) is odd split- orthogonal and Pi = 0. Thus, (Pi, [βi]) has a reflection if and only if Λi contains units. Since Λi consists of ni × ni alternating matrices and ni is odd, Λi does not contain any units and hence (Pi, [βi]) has no reflections. i , so s0,1,a is a reflection of (Pi, [βi]). j=1 ε(j) (cid:3) i Remark 5.6. Proposition 5.5 can also be proved using the more general result [20, Lm. 3.6]. Lemma 5.7. Let (P, [β]) be a unimodular quadratic space such that Pk 6= 0 when- ever (Ak, σk, uk, Λk) is odd split-orthogonal. Then the product of two ei-reflections of (P, [β]) equals a product of two reflections. Proof. By Proposition 5.5, there exist an fk-reflection sk for all 1 ≤ k ≤ ℓ. Let t and t′ be ei-reflections. By Lemma 3.4(i), t and t−1 are also e(j) -reflections for all 1 ≤ j ≤ ni. Thus, by Lemma 3.4(ii), ψ = tni s1 . . . si . . . sk and ϕ = s−1 k . . . s−1 1 (t−1)ni−1t′ are reflections (here, means omission). But tt′ = ψϕ, so we are done. (cid:3) . . . s−1 i i Recall that O′(P, [β]) denotes the subgroup of O(P, [β]) generated by reflections. Theorem 5.8. Let (P, [β]) be a unimodular quadratic space, and assume that (1) if (Ai, σi, ui, Λi) is split-orthogonal, then Di ≇ F2, and (2) if Di ∼= F2 × F2, then Pi ≇ εiAi. Let I be the set of indices 1 ≤ i ≤ ℓ such that Pi 6= 0 and (Ai, σi, ui, Λi) is split-orthogonal, and let ξ := (ni + 2Z)i∈I ∈ (Z/2Z)I. Denote by ∆I the group homomorphism Then: ψ 7→ (∆i(ψ))i∈I : O(P, [β]) → (Z/2Z)I . (i) ∆I is onto (even without assuming (1) and (2)). (ii) If Pi 6= 0 whenever (Ai, σi, ui, Λi) is odd split-orthogonal, then O′(P, [β]) = (ii) If there is i such that Pi = 0 and (Ai, σi, ui, Λ) is odd split-orthogonal, then ∆−1 I ({0, ξ}). O′(P, [β]) = 1. Proof. (i) Let j ∈ I. By Proposition 5.2, ∆I of an ej-reflection is (δij )i∈I (where δij = 1 if i = j and 0 otherwise), and by Lemma 5.4(ii), ej-reflections exist for all i ∈ I, so ∆I is onto. (ii) By Proposition 5.5, (P, [β]) admits a reflection s, and by Proposition 5.2, ∆I(s) = ξ. This also implies that O′(P, [β]) ⊆ ∆−1 I ({0, ξ}). Now, to prove the other inclusion, it enough to show that ker ∆I ⊆ O′(P, [β]). Let ψ ∈ ker ∆I . By 24 URIYA A. FIRST Corollary 4.11, ψ is a product of quasi-reflections. Moreover, from the proof of Theorem 4.5 it follows that ψ can written as a product of e1-reflections followed by a product of e2-reflection etc., and if Pi = 0, then the product includes no ei- reflections. For 1 ≤ i ≤ ℓ with Pi 6= 0, let mi denote the number of ei-reflection used to express ψ. We claim that mi can be taken to be even. By Lemma 5.7, this will imply ψ ∈ O′(P, [β]). Indeed, if (Ai, σi, ui, Λi) is split-orthogonal (and Pi 6= 0), then by Proposition 5.2, mi + 2Z = ∆i(ψ) = 0, so mi is even. When (Ai, σi, ui, Λi) is not split-orthogonal, we can freely increase mi by inserting idP , which is an ei-reflection by Lemma 5.4(i), into the product, so again, mi can be made even. (iii) This follows from Proposition 5.5. (cid:3) Corollary 5.9. Let n (resp. m) denote the number of i-s such that (Ai, σi, ui, Λi) is odd (resp. even) split-orthogonal. Then for any unimodular quadratic space (P, [β]) admitting a reflection (see Proposition 5.5), we have [O(P, [β]) : O′(P, [β])] = 2m+max{n−1,0} . We believe that Theorem 5.8(ii) should also be true when A is semilocal and P is a progenerator (i.e. A is a summand of P n for some A, or equivalently Pi 6= 0 for all i). Indeed, we have the following. Theorem 5.10. Provided P is free, part (ii) of Theorem 5.8 holds under the milder assumption that A is semilocal. Proof. By Proposition 5.5, (P , [β]) admits a reflection, which can lifted to a re- flection s of (P, [β]). That reflection satisfies ∆I(s) = ξ by Proposition 5.2. This proposition also implies that O′(P, [β]) ⊆ ∆−1 I ({0, ξ}), so again, it is left to prove that ker ∆I ⊆ O′(P, [β]). Let ψ ∈ ker ∆I, and let ψ be the isometry it induces on (P , [β]) (namely, ψ(x) = ψx). Arguing as in the proof of Theorem 5.8, we see that ψ is a product of reflections of (P , [β]). These reflections can be lifted to reflec- tions of (P, [β]), and their product is an isometry ψ′ ∈ O(P, [β]) such that ψ = ψ′. Replacing ψ with ψ′−1ψ, we may assume ψ = idP , or rather, ψx − x ∈ P Jac(A) for all x ∈ P . Now, it is shown in the proof of [20, Th. 6.2] that such ψ is a product of reflections (here we need P to be free), so we are done. (Notice that [20, Th. 6.2] assumes that [βi] 6= 0 whenever (Ai, σi, ui, Λi) is split-orthogonal, but the argument that we have counted on only needs [βi] 6= 0 when (Ai, σi, ui, Λi) is odd split-orthogonal. Also note that [βi] 6= 0 ⇐⇒ Pi 6= 0 because (Pi, [βi]) is unimodular.) (cid:3) Example 5.11. Part (i) of Theorem 5.8 may fail when A is only assumed to be semilocal. For example, take A to be a non-local semilocal commutative domain with 2 ∈ A×, set σ = idA, u = 1, Λ = 0, and define β : A× A → A by β(x, y) = xy. It is easy to check that O(A, [β]) = {± idA}. However, I > 1 because A is not a field, and hence ∆I cannot be onto. References [1] Goro Azumaya. On maximally central algebras. Nagoya Math. J., 2:119 -- 150, 1951. [2] Anthony Bak. On modules with quadratic forms. In Algebraic K-Theory and its Geometric Applications (Conf., Hull, 1969), pages 55 -- 66. Springer, Berlin, 1969. WITT'S EXTENSION THEOREM 25 [3] Hyman Bass. Unitary algebraic K-theory. In Algebraic K-theory, III: Hermitian K-theory and geometric applications (Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972), pages 57 -- 265. Lecture Notes in Math., Vol. 343. Springer, Berlin, 1973. [4] Hyman Bass. Clifford algebras and spinor norms over a commutative ring. Amer. J. Math., 96:156 -- 206, 1974. [5] Eva Bayer-Fluckiger and Laura Fainsilber. Non-unimodular Hermitian forms. Invent. Math., 123(2):233 -- 240, 1996. [6] Eva Bayer-Fluckiger, Uriya A. First, and Daniel A. Moldovan. Hermitian categories, extension of scalars and systems of sesquilinear forms. Pacific J. Math., 270(1):1 -- 26, 2014. [7] Eva Bayer-Fluckiger and Daniel Arnold Moldovan. Sesquilinear forms over rings with invo- lution. J. Pure Appl. Algebra, 218(3):417 -- 423, 2014. [8] Uriya A. First. General bilinear forms. Israel J. Math., 205(1):145 -- 183, 2015. [9] Uriya A. First. Rings that are Morita equivalent to their opposites. J. Algebra, 430:26 -- 61, 2015. [10] Larry C. Grove. Classical groups and geometric algebra, volume 39 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2002. [11] Bernhard Keller. A remark on quadratic spaces over noncommutative semilocal rings. Math. Z., 198(1):63 -- 71, 1988. [12] Manfred Knebusch. Isometrien uber semilokalen Ringen. Math. Z., 108:255 -- 268, 1969. [13] M. Kneser. Lectures on Galois cohomology of classical groups. Tata Institute of Fundamental Research, Bombay, 1969. With an appendix by T. A. Springer, Notes by P. Jothilingam, Tata Institute of Fundamental Research Lectures on Mathematics, No. 47. [14] Martin Kneser. Witts Satz uber quadratische Formen und die Erzeugung orthogonaler Grup- pen durch Spiegelungen. Math.-Phys. Semesterber., 17:33 -- 45, 1970. [15] Max-Albert Knus. Quadratic and Hermitian forms over rings, volume 294 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 1991. With a foreword by I. Bertuccioni. [16] Max-Albert Knus, Alexander Merkurjev, Markus Rost, and Jean-Pierre Tignol. The book of involutions, volume 44 of American Mathematical Society Colloquium Publications. American Mathematical Society, Providence, RI, 1998. With a preface in French by J. Tits. [17] Max-Albert Knus and Oliver Villa. Quadratic quaternion forms, involutions and triality. In Proceedings of the Conference on Quadratic Forms and Related Topics (Baton Rouge, LA, 2001), number Extra Vol., pages 201 -- 218 (electronic), 2001. [18] T. Y. Lam. Lectures on modules and rings, volume 189 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1999. [19] H.-G. Quebbemann, W. Scharlau, and M. Schulte. Quadratic and Hermitian forms in additive and abelian categories. J. Algebra, 59(2):264 -- 289, 1979. [20] H. Reiter. Witt's theorem for noncommutative semilocal rings. J. Algebra, 35:483 -- 499, 1975. [21] Louis H. Rowen. Ring theory. Vol. I, volume 127 of Pure and Applied Mathematics. Academic Press Inc., Boston, MA, 1988. [22] Amit Roy. Cancellation of quadratic form over commutative rings. J. Algebra, 10:286 -- 298, 1968. [23] Winfried Scharlau. Quadratic and Hermitian forms, volume 270 of Grundlehren der Math- ematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer- Verlag, Berlin, 1985. [24] Donald E. Taylor. The geometry of the classical groups, volume 9 of Sigma Series in Pure Mathematics. Heldermann Verlag, Berlin, 1992. [25] Peter V´amos. Decomposition problems for modules over valuation domains. J. London Math. Soc. (2), 41(1):10 -- 26, 1990. [26] C. T. C. Wall. On the axiomatic foundations of the theory of Hermitian forms. Proc. Cam- bridge Philos. Soc., 67:243 -- 250, 1970. [27] R. B. Warfield, Jr. Cancellation of modules and groups and stable range of endomorphism rings. Pacific J. Math., 91(2):457 -- 485, 1980. Hebrew University of Jerusalem, Israel.
1803.08870	3	1803	2018-12-25T22:26:25	High order Bellman equations and weakly chained diagonally dominant tensors	[ "math.RA", "math.NA" ]	We introduce high order Bellman equations, extending classical Bellman equations to the tensor setting. We introduce weakly chained diagonally dominant (w.c.d.d.) tensors and show that a sufficient condition for the existence and uniqueness of a positive solution to a high order Bellman equation is that the tensors appearing in the equation are w.c.d.d. M-tensors. In this case, we give a policy iteration algorithm to compute this solution. We also prove that a weakly diagonally dominant Z-tensor with nonnegative diagonals is a strong M-tensor if and only if it is w.c.d.d. This last point is analogous to a corresponding result in the matrix setting and tightens a result from [L. Zhang, L. Qi, and G. Zhou. "M-tensors and some applications." SIAM Journal on Matrix Analysis and Applications (2014)]. We apply our results to obtain a provably convergent numerical scheme for an optimal control problem using an "optimize then discretize" approach which outperforms (in both computation time and accuracy) a classical "discretize then optimize" approach. To the best of our knowledge, a link between M-tensors and optimal control has not been previously established.	math.RA	math	High order Bellman equations and weakly chained diagonally dominant tensors Parsiad Azimzadeh∗ Erhan Bayraktar∗†. Abstract We introduce high order Bellman equations, extending classical Bellman equations to the tensor setting. We introduce weakly chained diagonally dominant (w.c.d.d.) tensors and show that a sufficient condition for the existence and uniqueness of a positive solution to a high order Bellman equation is that the tensors appearing in the equation are w.c.d.d. M-tensors. In this case, we give a policy iteration algorithm to compute this solution. We also prove that a weakly diagonally dominant Z-tensor with nonnegative diagonals is a strong M-tensor if and only if it is w.c.d.d. This last point is analogous to a corresponding result in the matrix setting and tightens a result from [L. Zhang, L. Qi, and G. Zhou. "M-tensors and some applications." SIAM Journal on Matrix Analysis and Applications (2014)]. We apply our results to obtain a provably convergent numerical scheme for an optimal control problem using an "optimize then discretize" approach which outperforms (in both computation time and accuracy) a classical "discretize then optimize" approach. To the best of our knowledge, a link between M-tensors and optimal control has not been previously established. Introduction 1 In this work, we introduce and study the nonlinear problem n A(P)um−1 − b(P)o = 0 find u ∈ Rn such that min P∈P (1) where A(P) is an m-order and n-dimensional real tensor, b(P) is a real vector, P is a nonempty compact set, and the minimum is taken with respect to the coordinatewise order on Rn (see (iii) in Section 5). If m = 2, then A(P) is a square matrix and A(P)um−1 ≡ A(P)u is the ordinary matrix- vector product. In this case, (1) is the celebrated Bellman equation for optimal decision making on a Markov chain. Aside from Markov chains, (1) also arises from discretizations of differential equations from optimal control [14]. ∗Department of Mathematics, University of Michigan ([email protected], [email protected]). †E. Bayraktar is supported in part by the National Science Foundation under grant DMS-1613170. 1 If m > 2, then A(P)um−1 defines a vector whose i-th component is a multivariate polynomial in the entries of u: (A(P)um−1)i = X i2,...,im (A(P))ii2···imui2 ··· uim (see also (2)). As such, we refer to (1) as a Bellman equation of order m. We are motivated to study this equation since, as we will see in the sequel, it arises from a so-called "optimize then discretize" [4] scheme for a differential equation. Our main goal is to characterize the existence and uniqueness of a solution u to (1) and to obtain a fast and provably convergent algorithm for computing it. If m = 2, existence and uniqueness is guaranteed when A(P) is a nonsingular M-matrix for each P [5, Theorem 2.1] (along with some other mild conditions on the functions A and b). We obtain an analogous result for the m > 2 case when A(P) is a strictly diagonally dominant (and hence nonsingular1) M-tensor for each P (Lemma 23 and Lemma 24). M-tensors, a generalization of M-matrices, were introduced in [20, 9] in order to test the positive definiteness of multivariate polynomials. However, strict diagonal dominance is a rather strong condition. In order to generalize our results, we extend the notion of weakly chained diagonal dominance from matrices to tensors (Definition 15). By restricting our attention to the case in which P is finite, we establish existence and uniqueness of a solution u to (1) under the weaker requirement that A(P) is a weakly chained diagonally dominant M-tensor (Lemma 26) and give a policy iteration algorithm to compute the solution (Algorithm 1). Analogously to the m = 2 case, the assumed finitude of P is sufficient for practical applications, though whether this assumption can be dropped remains an interesting open theoretical question (Remark 27). We also establish the following result, which should be of broader interest to the M-tensor community: Theorem 1. Let A be a weakly diagonally dominant Z-tensor with nonnegative diagonals. Then, the following are equivalent: (i) A is a strong M-tensor. (ii) Zero is not an eigenvalue of A. (iii) A is weakly chained diagonally dominant (Definition 15). An analogous equivalence result for matrices was recently proved in [1]. Since a weakly irreducibly diagonally dominant tensor is a weakly chained diagonally dominant tensor (Lemma 17), the following is an immediate consequence: Corollary 2. Let A be a Z-tensor with nonnegative diagonals. If A is weakly irreducibly diagonally dominant, then A is a strong M-tensor. Since an irreducible tensor is weakly irreducible (Corollary 10), Theorem 1 and Corollary 2 can be thought of as tightening the following result: 1The terms "nonsingular M-tensor" and "strong M-tensor" are synonymous in the literature. 2 Proposition 3 ([20, Theorem 3.15]). Let A be a Z-tensor with nonnegative diagonals. If A is strictly or irreducibly diagonally dominant, then A is a strong M-tensor. Moreover, Theorem 1 yields a graph-theoretic characterization of weakly diagonally dominant strong M-tensors and a fast algorithm to determine if an arbitrary weakly diagonally dominant tensor is a strong M-tensor (Remark 16). This work is organized as follows. In Section 2, we recall some standard definitions and results for tensors. In Section 3, we introduce the notion of a weakly chained diagonally dominant tensor. A proof of Theorem 1 is given in Section 4. In Section 5, we study high order (m > 2) Bellman equations. In Section 6, we use our results to study numerically some problems from optimal stochastic control. 2 Preliminaries For the convenience of the reader, we gather in this section some definitions and well-known results (cf. [17]) concerning tensors of the form ai1···im ∈ R, 1 (cid:54) i1, . . . , im (cid:54) n. A = (ai1···im), Such tensors are called m-order and n-dimensional real tensors, though we will simply refer to them as "tensors" in this work. We call a1···1, a2···2, etc. the diagonal entries of A. All other entries are referred to as off-diagonal. Definition 4 ([16]). Let x = (x1, . . . , xn) be a vector in Cn and A = (ai1···im) be a tensor. n) the coordinatewise power of x and by Axm−1 the vector in We denote by x[α] = (xα1 , . . . , xα Cn whose i-th entry is (2) aii2···imxi2 ··· xim. X We call λ in C an eigenvalue of A if we can find a vector x in Cn \ {0} such that i2,...,im Axm−1 = λx[m−1]. The vector x is called an eigenvector associated with λ. The spectrum σ(A) of A is the set of all eigenvalues of A. The spectral radius of A is ρ(A) = maxλ∈σ(A) λ. Z and M-tensors, defined below, are natural extensions of Z and M-matrices. Definition 5 ([20, Pg. 440]). A Z-tensor is a real tensor whose off-diagonal entries are nonpositive. Definition 6 ([20, Definition 3.1]). A tensor A is an M-tensor if there exists a nonnegative tensor B (i.e., a tensor with nonnegative entries) and a real number s (cid:62) ρ(B) such that A = sI − B where I is the identity tensor (i.e., the tensor with ones on its diagonal and zeros elsewhere). If s > ρ(B), then A is called a strong M-tensor. 3 Unlike the matrix setting, there are two distinct notions of irreducibility for tensors, introduced in [12]. Both are given below. Definition 7 ([12, Pg. 739]). A tensor A = (ai1···im) is reducible if there exists a nonempty proper index subset Λ (cid:40) {1, . . . , n} such that aii2···im = 0 Otherwise, we say A is irreducible. Definition 8 ([13, Definition 2.2]). Let A = (ai1···im) be a tensor and R(A), the representa- tion of A, denote the n × n matrix whose (i, j)-th entry is given by if i ∈ Λ and i2, . . . , im /∈ Λ. X aii2···im 1{i2,...,im}(j) where 1U is the indicator function of the set U. We say A is weakly reducible if R(A) is a reducible matrix. Otherwise, we say A is weakly irreducible. i2,...,im In [12, Lemma 3.1], the authors show that irreducibility is a stronger requirement than weak irreducibility. We summarize this below, including what we believe to be a simpler proof for the reader's convenience. Proposition 9. A tensor that is weakly reducible is reducible. Proof. Let A = (ai1···im) be a weakly reducible tensor so that the n × n matrix R(A) = (rij) is reducible. Then, there exists a nonempty proper index subset Λ (cid:40) {1, . . . , n} such that rij = X aii2···im 1{i2,...,im}(j) = 0 if i ∈ Λ and j /∈ Λ. i2,...,im Therefore, aii2···im = 0 if i ∈ Λ and ik /∈ Λ for some 2 (cid:54) k (cid:54) m and hence A is reducible. Corollary 10. An irreducible tensor is weakly irreducible. We close this section with the notion of diagonal dominance. Definition 11 ([20, Definition 3.14]). Let A = (ai1···im) be a tensor. We say that the i-th row of A is strictly diagonally dominant (s.d.d.) if ai···i > aii2···im . (3) X (i2,...im)6=(i,...,i) We say A is s.d.d. if all of its rows are s.d.d. Weakly diagonally dominant (w.d.d.) is defined with weak inequality ((cid:62)) instead. We use J(A) = {1 (cid:54) i (cid:54) n: i satisfies (3)} to denote the set of s.d.d. rows of A. Definition 12 ([20, Definition 3.14]). We say a tensor A is (weakly) irreducibly diagonally dominant if it is (weakly) irreducible, w.d.d., and J(A) is nonempty. 4 3 Weakly chained diagonally dominant tensors Before we introduce the notion of weakly chained diagonally dominant (w.c.d.d.) tensors, we define the directed graph associated with a tensor. Definition 13. Let A = (ai1···im) be a tensor. (i) The directed graph of A, denoted graph A, is a tuple (V, E) consisting of the vertex set V = {1, . . . , n} and edge set E ⊂ V × V satisfying (i, j) ∈ E if and only if aii2···im 6= 0 for some (i2, . . . , im) such that j ∈ {i2, . . . , im}. (ii) A walk in graph A is a nonempty finite sequence (i1, i2), (i2, i3), . . . , (ik−1, ik) of "adja- cent" edges in E. The proof of the next result, being a trivial consequence of the above definition, is omitted. Lemma 14. graph A = graph R(A) for any tensor A. Since each vertex in the directed graph of a tensor corresponds to a row i, we use the terms row and vertex interchangeably. To simplify matters, we hereafter denote edges by i → j instead of (i, j) and walks by i1 → i2 → ··· → ik instead of (i1, i2), . . . , (ik−1, ik). We are now ready to define w.c.d.d. Definition 15. A tensor A is w.c.d.d. if all of the following are satisfied: (i) A is w.d.d. (ii) J(A) is nonempty. (iii) For each i1 /∈ J(A), there exists a walk i1 → i2 → ··· → ik in graph A such that ik ∈ J(A). If the tensor is of order m = 2 (i.e., the tensor is a matrix), then the above becomes the usual definition of w.c.d.d. for matrices [1, Definition 2.20]. Remark 16. A trivial extension of the algorithm in [1] allows us to test if a weakly diagonally dominant tensor is weakly chained diagonally dominant with O(nm) effort if the tensor is dense. Less computational effort is required if the tensor is sparse (we refer to [1] for details). Lemma 17. A weakly irreducibly diagonally dominant tensor is a w.c.d.d. tensor. Proof. If A is weakly irreducibly diagonally dominant, then J(A) is nonempty and graph R(A) is strongly connected (i.e., for any pair of vertices (i, j), there is a walk starting at i and ending at j). The result then follows by Lemma 14. 5 4 Proof of Theorem 1 In the following, we denote by Re z the real part of a complex number z. We say a vector is positive if it lies in the positive orthant Rn++ = (0,∞)n. Similarly, any element of Rn+ = [0,∞)n is called a nonnegative vector. Our proof relies on the following results: Proposition 18 ([20, Theorem 3.3]). minλ∈σ(A) Re λ is nonnegative (resp. positive) whenever A is an M-tensor (resp. strong M-tensor). Proposition 19 ([20, Theorem 3.15]). If A is a w.d.d. Z-tensor with nonnegative diagonals, then A is an M-tensor. Our proof also relies on a corollary to the following result: Proposition 20 ([10, Pg. 697]). Let A be a strong M-tensor. For each positive vector b (of compatible size), there exists a unique positive vector x which solves the tensor equation ++ : Rn++ → Rn++ the mapping from positive right hand sides b to Axm−1 = b. Denoting by A−1 positive solutions x, A−1 ++ is nondecreasing with respect to the coordinatewise order. The corollary, which is of independent interest, establishes existence (but not uniqueness) of nonnegative solution x to the tensor equation Axm−1 = b when b is a nonnegative vector and A is a strong M-tensor. Corollary 21. Let A be a strong M-tensor. Then, there exists a nondecreasing map A−1 + : Rn+ → Rn+ which associates to each nonnegative right hand side b a nonnegative solution x of the tensor equation Axm−1 = b. Proof. For k (cid:62) 1, define x(k) = A−1 ++(b + 1/k) where b + 1/k is the vector obtained by adding 1/k to each entry of b. Since A−1 ++ is nondecreasing, the sequence (x(k))k is nonincreasing with respect to the coordinatewise order. Moreover, since each x(k) is a nonnegative vector, this sequence is bounded below by the zero vector and hence has a limit x. Taking k → ∞ in Axm−1 (k) = b + 1/k and employing the continuity of the map y 7→ Aym−1, we obtain Axm−1 = b. The above implies that the map A−1 + : Rn+ → Rn+ k→∞ A−1 A−1 + (b) = lim ++(b + 1/k) is well-defined and associates to each nonnegative vector b a nonnegative solution x of the tensor equation Axm−1 = b. That A−1 + is nondecreasing is an immediate consequence of A−1 ++ being nondecreasing. We require one last intermediate result, which captures the invariance of spectra under permutation. It can be thought of as a generalization of the fact that for any square matrix A and permutation matrix P of compatible size, σ(A) = σ(P AP (cid:124)). 6 Lemma 22. Let A = (ai1···im) be a tensor, π be a permutation of {1, . . . , n}, and A0 = (a0 be the tensor with entries i1···im) a0 i1···im = aπ(i1)···π(im). Then, σ(A) = σ(A0). Proof. Let λ be an eigenvalue of A with corresponding eigenvector x. Let y denote the vector whose entries are yi = xπ(i). Then, for each i, λ(yi)m−1 = λ(xπ(i))m−1 = X = X = X i2,...,im i2,...,im i2,...,im aπ(i)i2···imxi2 ··· xim aπ(i)π(i2)···π(im)xπ(i2) ··· xπ(im) ii2···imyi2 ··· yim. a0 We are now ready to prove Theorem 1. We split the proof into parts. In each part, we use A = (ai1···im) to denote a w.d.d. Z-tensor with nonnegative diagonals. Proof of (i) implies (ii). This follows directly from Proposition 18. Proof of (iii) implies (i). Suppose A is w.c.d.d. Therefore, A is w.d.d. by definition, and hence A is an M-tensor by Proposition 19. Let λ and x be an eigenvalue-eigenvector pair of A. We may, without loss of generality, assume kxk∞ = 1 (otherwise, define a new vector y = x/kxk∞ and note that Aym−1 = λy[m−1]). Since A is an M-tensor, Re λ (cid:62) 0 by Proposition 18. In order to arrive at a contradiction, suppose Re λ = 0. Now, fix i such that xi = kxk∞ = maxj xj. It follows that X λ − ai···i (cid:54) (i2,...,im)6=(i,...,i) aii2···imxi2···xim (cid:54) aii2···im . (i2,...,im)6=(i,...,i) X Since Re λ = 0 and ai···i is real, Combining the above inequalities, ai···i (cid:54) ai···i = Re λ − ai···i (cid:54) λ − ai···i . X aii2···imxi2···xim (cid:54) X (i2,...,im)6=(i,...,i) (i2,...,im)6=(i,...,i) aii2···im . Since A is w.d.d., the above chain of inequalities holds with equality so that i /∈ J(A) and xi2 = ··· = xim = 1 whenever aii2···im 6= 0 for some (i2, . . . , im). Since A is w.c.d.d., we may pick a walk i1 → i2 → ··· → ik starting at some row i1 for which xi1 = 1 and ending at some row ik ∈ J(A). Setting i = i1 in the previous paragraph, we get xi2 = 1 and therefore also i2 /∈ J(A). Applying this reasoning inductively, we conclude that ik /∈ J(A), a contradiction. 7 Proof of (ii) implies (iii). Suppose A is not w.c.d.d. Proceeding by contrapositive, it is sufficient to show that λ = 0 is an eigenvalue of A. Let i1 /∈ J(A): there exists a walk i1 → i2 → ··· → ik such that ik ∈ J(A)o W(A) =n . Let R(A) = {1, . . . , n} \ (J(A) ∪ W(A)). Since A is not w.c.d.d., R(A) is nonempty. By Lemma 22, we may assume that R(A) = {1, . . . , r} for some 1 (cid:54) r (cid:54) n (otherwise, permute the indices appropriately). For the remainder of the proof, let e = (1, . . . , 1)(cid:124) be the vector of ones in Rr. If r = n, it follows that J(A) is empty and hence every row of A is not s.d.d. Since A is a w.d.d. Z-tensor, we have, for each row i, ai···i = − X (i2,...,im)6=(i,...,i) aii2···im. (4) In other words, Aem−1 = 0, and hence λ = 0 is an eigenvalue of A. If r < n, the adjacency graph has the structure shown in Figure 1. In particular, there are no edges from vertices i ∈ R(A) to vertices j /∈ R(A) since if there were, i would not be a member of R(A) by definition. This implies that aii2···im = 0 if i ∈ R(A) and ik /∈ R(A) for some 2 (cid:54) k (cid:54) m. Equivalently, aii2···im = 0 if i (cid:54) r and max{i2, . . . , im} > r. Define the m-order and n-dimensional tensor B = (bi1···im) by By (5), it follows that B is a w.d.d. Z-tensor with no s.d.d. rows and hence similarly to (4), we can establish that for any vector w in Rn−r, (5) (6) bi1···im = 0 ai1···im if 1 (cid:54) i1, . . . , im (cid:54) r ! otherwise. Bxm−1 = 0 where x = e w . 1 if 1 (cid:54) i1 (cid:54) r and (i2, . . . , im) = (i1, . . . , i1) 0 otherwise. 0 di1···im = if 1 (cid:54) i1 (cid:54) r ai1···im otherwise. Define the m-order and n-dimensional tensors C = (ci1···im) and D = (di1···im) by ci1···im = and By construction, C + D is a w.c.d.d. Z-tensor with nonnegative diagonals. Since we have already proven (iii) implies (i) in Theorem 1, we conclude that C + D is a strong M-tensor. By Corollary 21, we can find a nonzero vector y such that . (7) ! e 0 (C + D)ym−1 = 8 (cii2···im + dii2···im)yi2 ··· yim = ym−1 i = 1. ! e w y = Note that if 1 (cid:54) i (cid:54) r, the above implies X Therefore, i2···im for some vector w in Rn−r. Since A = B + (C + D) − C, (6) and (7) imply ! ! Aym−1 = Bym−1 + (C + D)ym−1 − Cym−1 = 0 + − = 0, e 0 e 0 so that λ = 0 is an eigenvalue of A. 1 ... r r + 1 ... n R(A) (R(A))c = J(A) ∪ W(A) Figure 1: Adjacency graph in the proof of Theorem 1 5 High order Bellman equations We now return to the high order Bellman equation (1), repeated below for the reader's convenience: n A(P)um−1 − b(P)o = 0. min P∈P In the above, A(P) = (ai1···im(P)) is an m-order and n-dimensional real tensor and b(P) = (b1(P), . . . , bn(P)) is a vector in Rn. It is understood that (cf. [2]) (i) P = P1 × ··· × Pn is a finite product of nonempty sets. That is, each P = (P1, . . . , Pn) in P is an n-tuple with Pi in Pi. (ii) Policies are "row-decoupled". That is, for any two policies P and P 0 in P, aii2···im(P) = i. In other words, the i-th row of A(P) aii2···im(P 0) and bi(P) = bi(P 0) whenever Pi = P 0 and b(P) are determined solely by Pi. (iii) Infimums (and other extrema) are taken with respect to the coordinatewise order. That . For example, is, for {y(α)}α ⊂ Rn, inf α y(α) is a vector whose i-th entry is inf α y min{( 1 (α) i 1 )} = ( 0 0 ). 0 ), ( 0 9 We require the following assumptions to study the problem: (H1) b(P) is a positive vector for each P in P. (H2) P is a compact topological space and A : P → Rnm and b : P → Rn are continuous functions. In practice, P is usually finite (Remark 27) in which case (H2) is trivially satisfied. 5.1 Existence and uniqueness We now establish existence and uniqueness of positive solutions to (1). Lemma 23 (Uniqueness). Suppose (H1), (H2), and A(P) is a strong M-tensor for each P in P. Then, there is at most one positive solution u of (1). Proof. Let u and w be two positive solutions of (1). By the compactness of P and continuity of A and b, we can find P ∗ such that A(P ∗)um−1 − b(P ∗) = 0 = min P∈P n A(P)wm−1 − b(P)o (cid:54) A(P ∗)wm−1 − b(P ∗). Therefore, 0 < A(P ∗)um−1 (cid:54) A(P ∗)wm−1. ++ is nondecreasing, applying the function (A(P ∗))−1 Using the fact that (A(P ∗))−1 inequality yields u (cid:54) w. Reversing the roles of u and w, we obtain the reverse inequality. Lemma 24 (Existence I). Suppose (H1), (H2), and A(P) is an s.d.d. M-tensor for each P in P. Then, there exists a positive solution u of (1). ++ to the above A close examination of the proof below reveals that we can relax the requirement that "b(P) is positive" in (H1) to "b(P) is nonnegative". In this case, the arguments establish the existence of a nonnegative solution u. Proof. We claim that it is sufficient to consider the case in which 1 − ai···i(P) = 0 for all i (we will come back to this claim later). Note that u is a solution of (1) if and only if it is a fixed point of the map F defined by n(I − A(P))um−1 + b(P)o[1/(m−1)] . F(u) = max P∈P Since the diagonals of I − A(P) are zero, the off-diagonals of A(P) are nonpositive, and b(P) is positive, it follows that F maps nonnegative vectors to positive vectors (i.e., F(Rn+) ⊂ Rn++). Next, we prove that F is continuous on Rn+. In order to do so, it is sufficient to show that the function G defined by G(u) = (F(u))[m−1] is locally Lipschitz on Rn+. Indeed, for nonnegative vectors u and w, (cid:13)(cid:13)(cid:13)(I − A(P))um−1 − (I − A(P)) wm−1(cid:13)(cid:13)(cid:13)∞ kG(u) − G(w)k∞ (cid:54) max o mX X P∈P ∞ ,kwkm−2 (cid:54) const. maxnkukm−2 ∞ ∞ ,kwkm−2 ∞ aii2···im(P) maxnkukm−2 P∈P max (i2,...,im)6=(i,...,i) (cid:54) max (cid:12)(cid:12)(cid:12)uij − wij (cid:12)(cid:12)(cid:12) oku − wk∞ j=2 i 10 where const. is a positive constant which does not depend on u or w. Note that in the m = 2 case, kukm−2∞ = kwkm−2∞ = 1, and hence this argument establishes global Lipschitzness. Next, we derive some bounds on F. The triangle inequality yields kF(u)k∞ (cid:54) max P∈P ∞ max i aii2···im(P) + kb(P)k∞ X (i2,...,im)6=(i,...,i) kukm−1  X 1/(m−1) ai∗i2···im(P ∗) + kb(P ∗)k1/(m−1) ∞ . 1/(m−1) . Therefore, there exist i∗ and P ∗ such that kF(u)k∞ (cid:54) kuk∞ (i2,...,im)6=(i∗,...,i∗) Since A(P ∗) is s.d.d., X (i2,...,im)6=(i∗,...,i∗) ai∗i2···im(P ∗) < ai∗···i∗(P ∗) = 1. In other words, there exist positive constants C1 < 1 and C2 such that for any u, kF(u)k∞ (cid:54) C1 kuk∞ + C2. Now, let M = C2/(1 − C1) and K = {u ∈ Rn+ : kuk∞ (cid:54) M}. By the above, F(K) ⊂ K. By the Schauder fixed point theorem, F admits a fixed point in K. Moreover, since F(K) ⊂ F(Rn+) ⊂ Rn++, this fixed point must be positive. Now, let us return to the unproven claim in the previous paragraph. Let D(P) = diag(a1···1(P), . . . , am···m(P)) be the diagonal matrix obtained from the diagonal entries of A(P). Note, in particular, that the i-th entry of the vector D(P)−1(A(P)um−1) is aii2···im(P) ai···i(P) ui2 ··· uim. A(P)um−1 − b(P)(cid:17)o = 0, Therefore, to establish the claim, it is sufficient to show that if u satisfies (8) X n D(P)−1(cid:16) i2,...,im min P∈P then u is a solution of (1) (while the converse is also true, we do not require it). Indeed, if u satisfies (8), then D(P ∗)−1(A(P ∗)um−1 − b(P ∗)) = 0 for some P ∗. Multiplying both sides of this equation by D(P ∗), we get A(P ∗)um−1 − b(P ∗) = 0 and hence n A(P)um−1 − b(P)o (cid:54) 0. min P∈P To establish the reverse inequality, we proceed by contradiction, assuming that we can find P∗ such that the vector A(P∗)um−1 − b(P∗) has a strictly negative entry. In this case, D(P∗)−1(A(P∗)um−1 − b(P∗)) also has a strictly negative entry, contradicting (8). Therefore, u is a solution of (1), as desired. 11 We would like to extend the above existence result to w.c.d.d. M-tensors. In order to do so, we require the following intermediate result, which is of independent interest. Lemma 25. Let A be a strong M-tensor and b be a positive vector (of compatible size). Then, the set n(A + I)−1 ++(b): (cid:62) 0o is bounded. Proof. Since A is a strong M-tensor there exists a positive vector z such that Azm−1 is positive [9, Theorem 3]. This in turn implies that (A + I)zm−1 = Azm−1 + z[m−1] (cid:62) Azm−1 > 0. Now, defining γ = max i bi ((A + I) zm−1)i (cid:54) max i bi (Azm−1)i , the arguments in the proof of [10, Theorem 3.2] imply that the unique positive solution x of (A + I)xm−1 = b satisfies x (cid:54) γ1/(m−1)z, giving us an upper bound that is independent of . Lemma 26 (Existence II). Suppose (H1), P is finite, and A(P) is a w.c.d.d. M-tensor for each P in P. Then, there exists a positive solution u of (1). Proof. As in the proof of Lemma 24, it is sufficient to consider the case in which ai···i(P) = 1. Now, let k be a positive integer. Since A(P) is w.d.d., it follows that A(P) + k−1I is s.d.d. Therefore, by Lemma 24, we can find a positive vector u(k) and a policy P k such that n(cid:16) min P∈P (cid:17) (k) − b(P)o =(cid:16) um−1 A(P) + k−1I (cid:17) A(P k) + k−1I (k) − b(P k) = 0. um−1 Since the sequence (P k)k has finite range (due to the finitude of P), the pigeonhole principle affords us the existence of an increasing sequence (k')' of positive integers and a n(cid:16) policy P ∗ such that for all ', A(P) + k−1 ' I (k') − b(P)o =(cid:16) (k') − b(P ∗) = 0. um−1 A(P ∗) + k−1 ' I um−1 (9) (cid:17) (cid:17) min P∈P For brevity, let A = A(P ∗) and b = b(P ∗). Since u(k') = (A + k−1 ' I)−1 ++(b), Lemma 25 implies that the sequence (u(k'))' is contained in a compact set and thereby admits a convergent subsequence with limit u(∞). (k') → 0 as ' → ∞. Now, we show that u(∞) is a solution of (1). First, note that k−1 Therefore, it is sufficient to establish that the function H defined by ' Ium−1 n A(P)um−1 − b(P)o H(u) = min P∈P is continuous on Rn+ and take limits in (9) to arrive at the desired result. This follows immediately from the fact that H(u) = u[m−1] − G(u) where G is the locally Lipschitz (and hence continuous) function defined in the proof of Lemma 24. 12 Remark 27. The proof of Lemma 26 uses a pigeonhole principle which relies on the assumed finitude of the policy set P. Whether this assumption can be dropped remains an interesting open question. From a practical perspective, it is important to note that classical m = 2 Bellman equations appear almost exclusively with finite policy sets P in applications. For example, in the context of discretizations of differential equations from optimal control, the policy set is always chosen to be a discretization of the corresponding control set so that the problem can be made amenable to numerical computation (see, e.g., [5, Section 5.2]). Analogously, we do not expect the finiteness assumption to be particularly obstructive in the m > 2 case. Indeed, the applications studied in Section 6 involve finite policy sets. 5.2 Policy iteration In the classical m = 2 setting, a popular computational procedure to solve (1) is policy iteration. We give a brief sketch of the algorithm and refer to [5] for details. At the k-th iteration, the algorithm picks a policy P k in P and solves the system A(P k)u(k) = b(P k). The policy P k is picked to ensure u(k−1) (cid:54) u(k) so that u = limk u(k) exists. Using continuity arguments, it can be shown that this limit is a solution of (1). When P is finite, policy iteration takes at most P iterations before achieving the limit (i.e., u(P) = u(P+1) = ··· = u). Analogously to the simplex algorithm, whose worst case complexity is determined by the number of vertices in the feasible polytope, policy iteration generally terminates in far fewer iterations. Continuing our analogy, we call the map which associates to each iteration k a policy P k a pivot rule. Below, we present an obvious extension of policy iteration to the case of m > 2. In the statement of the algorithm, we allow for some freedom in the choice of pivot rule. Unlike the m = 2 case, it is not clear if there exists a pivot rule which ensures u(k−1) (cid:54) u(k). The resulting algorithm is below. Algorithm 1 A policy iteration algorithm for (1) 1: procedure Policy-Iteration(A, b) 2: 3: 4: 5: 6: 7: 8: 9: 10: end procedure Pick P k in P \ {P 1, . . . , P k−1} according to some pivot rule Solve the tensor equation A(P k)um−1 (k) = b(P k) for u(k) in Rn++ if minP∈P{A(P)um−1 end if (k) − b(P)} = 0 then for k ← 1, . . . ,P do return u(k) end for error no solution found Remark 28. The terminating condition on line 5, while convenient for a theoretical dis- cussion, is unsuitable for a practical implementation. Such an implementation should use instead a condition on the relative error between iterates u(k−1) and u(k) (see, e.g., (22)) or a condition on the norm of minP∈P{A(P)um−1 (k) − b(P)}. 13 Theorem 29. Suppose (H1), P is finite, and A(P) is a w.c.d.d. M-tensor for each P in P. Then, Policy-Iteration returns the unique positive solution of (1). As usual, we can relax the requirement that "b(P) is positive" in (H1) to "b(P) is nonnegative" by replacing Rn++ with Rn+ on line 4 of the algorithm. In this case, the algorithm returns a (possibly nonunique) nonnegative solution u. Proof. This is a direct consequence of Lemma 23 and Lemma 26 along with the fact that the algorithm iterates over all policies. Taking w = u(k) and P 0 = P k in the result below establishes that the solution u described in Theorem 29 dominates the iterates u(k) generated by the algorithm. Lemma 30. Suppose (H1), (H2), and A(P) is a strong M-tensor for each P in P. If u is a positive solution of (1) and w is a positive vector satisfying A(P 0)wm−1 = b(P 0) for some P 0 in P, then u (cid:62) w. Proof. Since n A(P)um−1 − b(P)o = 0 = A(P 0)wm−1 − b(P 0), A(P 0)um−1 − b(P 0) (cid:62) min P∈P it follows that A(P 0)um−1 (cid:62) A(P 0)wm−1 > 0. The desired result follows by applying the function (A(P 0))−1 ++ to this inequality. 5.3 Locally optimal policy The pivot rule which we employ in the numerical tests appearing in the sequel is inspired by the classical (m = 2) policy iteration algorithm. The idea behind the pivot rule is simple: at the k-th iteration, let n A(P)um−1 (k−1) − b(P)o O = arg min P∈P be the set of policies that are "locally optimal" for u(k−1) where, for convenience, we define u(0) = 0. Let H = P \ {P 1, . . . , P k−1} be the set of policies the algorithm has not yet considered. If O ∩ H 6= ∅, we pick P k in O ∩ H. Otherwise, we pick P k in H. It is readily verified that if m = 2, we retrieve the classical policy iteration algorithm [5, Algorithm Ho-1]. 5.4 Incorporating lower order tensors The results of the previous sections can also be applied to the following more general higher order Bellman equation n A(P)um−1 − Bm−1(P)um−2 − Bm−2(P)um−3 − ··· − B2(P)u − b(P)o = 0 (10) min P∈P where each Bp(P) is a row-decoupled (see (ii) at the beginning of Section 5) p-order n- dimensional nonnegative tensor. This is possible since if A(P) is an m-order n-dimensional 14 strong M-tensor for each P, then u is a solution of (10) if and only if w = (cid:16) u 1(cid:17)(cid:124) is a solution of ( A0(P)wm−1 − min P∈P !) b(P) 1 = 0 where for each P, A0(P) is an appropriately chosen m-order (n + 1)-dimensional strong M-tensor whose construction is detailed in the proof of the next lemma. Lemma 31. Let A = (ai1···im) be an m-order n-dimensional strong M-tensor and Bp = (bi1···ip) be a p-order n-dimensional nonnegative tensor for 2 (cid:54) p < m. Then, there exists an m-order (n + 1)-dimensional strong M-tensor A0 such that Axm−1 − Bm−1xm−2 − Bm−2xm−3 − ··· − B2x ! for all x ∈ Rn. (11) !m−1 = A0 x 1 1 Proof. We claim that we can construct an m-order (n + 1)-dimensional Z-tensor A0 = (a0 i1···im) satisfying (11). Indeed, if this is the case, since A is a strong M-tensor, we can find a positive vector v such that Avm−1 − Bm−1vm−2 − Bm−2vm−3 − ··· − B2v > 0 (see the proof of [10, Theorem 3.6] for details) and hence !m−1 A0 v 1 > 0. Therefore, A0 is semi-positive and hence a strong M-tensor [9, Theorem 3]. Returning to the claim above, we give the construction in the case of m = 3, from which the general m (cid:62) 3 case should be evident. Indeed, in the case of m = 3, we can take the nonzero entries of A0 to be a0 i,j,k = ai,j,k, i,n+1,j = −bi,j/2, a0 i,j,n+1 = a0 if 1 (cid:54) i, j, k (cid:54) n if 1 (cid:54) i, j (cid:54) n a0 n+1,n+1,n+1 = 1. Clearly, A0 is a Z-tensor. Now, let x in Rn be arbitrary and y =(cid:16) (Ax2 − Bx)i = X i,j,n+1 (xj · 1) + X = X ai,j,kxjxk − X i,j,k (xjxk) + X 1(cid:54)j,k(cid:54)n a0 bi,jxj a0 1(cid:54)j(cid:54)n 1(cid:54)j,k(cid:54)n 1(cid:54)j(cid:54)n x 1(cid:17)(cid:124) . Then, for 1 (cid:54) i (cid:54) n, i,n+1,j (1 · xj) a0 1(cid:54)j(cid:54)n = X 1(cid:54)j,k(cid:54)n+1 a0 i,j,kyjyk = (A0y2)i so that (11) is satisfied. 15 6 Application to optimal stochastic control In this section, we apply our results to solve numerically the differential equation o = 0, on Ω n LλU − ηU − 1 2αγ2U + βγ − max U = g, (γ,λ)∈(Γ,Λ) on ∂Ω (12) where Lλ is the (possibly degenerate) elliptic operator LλU(x) = 1 2σ(x, λ)2U00(x) + µ(x, λ)U0(x). and Ω = (0, 1), Γ = [0,∞), We require the following assumptions: (A1) Letting dH denote the Hausdorff metric, to each ∆x > 0, we can associate a finite Λ is a compact metric space. and subset Λ∆x of Λ such that dH(Λ∆x, Λ) → 0 as ∆x ↓ 0. (A2) σ, µ (resp. η, α, β, g) are real (resp. positive) maps with dom(σ) = dom(µ) = Ω × Λ (resp. dom(η) = dom(α) = dom(β) = dom(g) = Ω). (A3) supx β(x)2/(α(x)η(x)) < ∞. Note that (A1) simply says that we can approximate Λ by finite subsets. Now, there are two ways to discretize (12): a "discretize then optimize" (DO) approach and an "optimize then discretize" (OD) approach. In the DO approach, we first replace the unbounded control set Γ by a partition of the interval Γ0 = [0, γmax] (for some γmax > 0 chosen large enough). Next, we replace the various quantities Uxx, Ux, and U by their discrete approximations. The resulting system is a classical m = 2 Bellman equation, which can be solved by policy iteration. Since the DO approach is well-understood, we present its derivation in Appendix A. In the OD approach, we first find the point γ at which the maximum (cid:26) max γ∈Γ −1 2αγ2U + βγ (cid:27) (13) is attained. Substituting this back into (12), we discretize the resulting differential equation. The OD approach results in a scheme with lower truncation error. In general, applying an OD approach to an elliptic differential equation may result in a scheme which is nonmonotone and/or hard to solve (see the discussion in [11]). This is problematic, since it is well-known that nonmonotone schemes are not guaranteed to converge [15]. In our case, the resulting OD system ends up being a higher order Bellman equation involving a w.c.d.d. M-tensor, making it both monotone and easy to solve by policy iteration (recall Theorem 29). 16 Z t 0 Z t 0 Remark 32 (Connection to optimal stochastic control). Let W = (Wt)t(cid:62)0 be a standard Brownian motion on a filtered probability space satisfying the usual conditions. It is well-known that (under some mild conditions), the value function "Z τ 0 (cid:18)Z t 0 v(x) = sup γ,λ E exp −η(X λ s ) − 1 2 α(X λ s )γ2 s ds (cid:19) t )γtdt β(X λ (cid:18)Z τ + exp 0 −η(X λ s ) − 1 2α(X λ s )γ2 s ds (cid:19) # τ ) g(X λ is a viscosity solution of (12) [19]. In the above, the supremum is over all progressively measurable processes γ = (γt)t(cid:62)0 and λ = (λt)t(cid:62)0 taking values in Γ and Λ, respectively, and X λ t = x + µ(Xs, λs)ds + σ(Xs, λs)dWs and τ = inf{t (cid:62) 0: X λ t /∈ Ω}. To ensure that the process X λ is well-defined, one should impose some additional assumptions (e.g., σ(·, λ) and µ(·, λ) are Lipschitz uniformly in λ). 6.1 Optimize then discretize scheme In this subsection, we derive the OD scheme and prove that it converges to the solution of (12). Let ∆x = 1/M for some positive integer M. We write ui ≈ U(i∆x) for the numerical solution at i∆x and let u = (u0, . . . , uM). We denote by + µi(λ) 1 ∆x 2σi(λ)2 ui−1 − 2ui + ui+1 ui+1 − ui, if µi(λ) (cid:62) 0 if µi(λ) < 0 ∆xu)i = 1 (Lλ ui − ui−1, (∆x)2 a standard upwind discretization of LλU(i∆x) where, for brevity, we have defined σi(λ) = σ(i∆x, λ) and µi(λ) = µ(i∆x, λ). First, note that a solution U of (12) must be everywhere positive since otherwise (13) is (cid:26) unbounded. By virtue of this, the maximum in (13) is = 1 2 ) −1 2 αγ2U + βγ max (cid:27) γ ( (Lλ∆xu)i − ηiui + 1 2 β2 i αi 1 ui − max λ∈Λ∆x ui = gi, This suggests approximating (12) by the M + 1 "discrete" equations β2 α 1 U . = 0, for 0 < i < M for i = 0, M (14) where we have defined αi = α(i∆x), βi = β(i∆x), ηi = η(i∆x), and gi = g(i∆x). The difficulty in the OD approach is that (14) cannot be written as a classical m = 2 Bellman equation due to the term 1/ui. We resolve this by writing (14) as a Bellman equation 17 of order m = 3 instead. In order to do so, we first note that a positive vector u = (u0, . . . , uM) satisfies (14) if and only if it satisfies ( ui(Lλ∆xu)i − ηiu2 − max λ∈Λ∆x i = g2 u2 i , ) i + 1 2 β2 i αi = 0, for 0 < i < M for i = 0, M. (15) To see this, multiply each equation in (14) by ui (conversely, divide each equation in (15) by ui). Next, define the Cartesian product Λ∆x = (Λ∆x)M+1 and denote by λ = (λ0, . . . , λM) an element of Λ∆x with λi ∈ Λ∆x. Define A(λ) = (aijk(λ)) as the order m = 3 tensor whose only nonzero entries are ai,i−1,i(λ) = ai,i,i−1(λ), 2ai,i,i−1(λ) = −1 2σi(λi)2 ai,i,i(λ) = +1 2σi(λi)2 2ai,i,i+1(λ) = −1 2σi(λi)2 ai,i+1,i(λ) = ai,i,i+1(λ), 1 (∆x)2 + µi(λi) 1 ∆x (∆x)2 + µi(λi) 1 2 ∆x (∆x)2 − µi(λi) 1 1 ∆x 1(−∞,0)(µi(λi)), + ηi, if 0 < i < M (16a) 1(0,+∞)(µi(λi)), and ai,i,i(λ) = 1, Lastly, define the vector b = (b0, . . . , bM) by β2 i 2 αi g2 i , 1 bi = , if i = 0, M. if 0 < i < M if i = 0, M. Then, (15) is equivalent to the order m = 3 Bellman equation min λ∈Λ∆x A(λ)u2 = b. (16b) (17) (18) We would like to apply policy iteration to compute a positive solution of (18). According to Theorem 29, this requires A(λ) to be an s.d.d. strong M-tensor. We establish this by showing that A(λ) is an s.d.d. Z-tensor with positive diagonals and applying Proposition 3. Indeed, that A(λ) is a Z-tensor with positive diagonals is clear from its definition. Next, note that by (16a) and (16b), ai···i(λ) + X (i2,...,im)6=(i,...,i) aii2···im(λ) = if 0 < i < M if i = 0, M. (19) ηi, 1, Since ηi > 0 by (A2), the above implies that A(λ) is s.d.d. While the above establishes that the numerical solution is well-defined for each ∆x > 0, we have yet to relate its limit as ∆x ↓ 0 back to the differential equation (12). In order to do so, we use the framework of viscosity solutions [8]: 18 Theorem 33. Suppose (A1), (A2), (A3), and that the differential equation (12) satisfies a strong comparison principle in the sense of Barles and Souganidis [3]. For each ∆x > 0, define the function u∆x : Ω → R by u∆x(x) = ui1[xi−∆x/2,xi+∆x/2)(x) i=0 MX vuut1  2  < ∞ where u = (u0, . . . , uM) is the unique positive solution of (18). Then, u∆x (cid:54) sup x max β2(x) α(x)η(x), g(x) (20) and, as ∆x ↓ 0, u∆x converges uniformly to the viscosity solution U of (12). Proof. We first prove the bound (20). Choose j such that uj = maxi ui. If 0 < j < M, it is straightforward to show that, (Lλ∆xu)j (cid:54) 0. In this case, ) 1 uj ∆xu)j − ηjuj + 1 (Lλ 2 (cid:62) ηjuj − 1 2 0 = − max λ∈Λ∆x β2 j αj β2 j αj 1 uj ( and hence by (A2), uj (cid:54) vuut1 2 vuut1 2 β2 j αjηj (cid:54) sup x β2(x) α(x)η(x). If instead j = 0 or j = M, then uj = gj (cid:54) max{g(0), g(1)} by (A2). Therefore, vuut1  2 uj (cid:54) sup x max β2(x) α(x)η(x), g(x)  , which is finite due to (A3). The remainder of the proof relies on the standard machinery of Barles and Souganidis [3], so we simply sketch the ideas. The discrete equations (14) define a scheme that is monotone and consistent in the sense of [3]. Moreover, u∆x is bounded independently of ∆x by (20). Therefore, by [3, Theorem 2.1], u∆x converges locally uniformly to U. Since Ω is compact, this convergence is uniform. 6.2 Numerical results In this subsection, we apply the OD and DO schemes (described in the previous subsection and Appendix A, respectively) to compute a numerical solution of (12) under the parameterization σ(x, λ) = 0.2 µ(x, λ) = 0.04λ α(x) = 2 − x β(x) = 1 + x η(x) = 0.04 g(x) = 1 (21) where λ ∈ Λ = {−1, 1}. Since Λ is finite, we take Λ∆x = Λ. For the DO scheme, we take γmax = 2 and discretize Γ0 = [0, γmax] by a uniform partition 0 = γ0 < ··· < γK = γmax (see Appendix A for details). 19 M 32 64 128 256 512 1024 Value Rel. err. Ratio 2.8093 2.8278 2.8367 2.8411 2.8433 2.8444 1.3e−2 6.0e−3 2.9e−3 1.3e−3 5.7e−4 1.9e−4 2.07 2.04 2.02 2.00 Its. 5 5 5 5 5 5 Inner its. 7 7 7 7 7 7 (a) Optimize then discretize M K Value Rel. err. Ratio 32 64 128 256 512 1024 5.9e−1 3.3e−1 4.5e−2 2.2e−2 5.0e−3 9.8e−4 1.1783 1.9179 2.7161 2.7825 2.8306 2.8421 0.93 12.0 1.38 4.19 1 2 4 8 16 32 Its. 7 7 7 7 7 7 Time 4.6e−2 4.8e−2 6.5e−2 1.2e−1 1.8e−1 3.5e−1 Time 3.3e−2 5.4e−2 1.4e−1 4.5e−1 1.4e+0 5.3e+0 (b) Discretize then optimize Table 1: Convergence results (parameters used: (21)) (cid:13)(cid:13)(cid:13)u(k) − u(k−1) In our implementation of Policy-Iteration, instead of using the terminating condition on line 5 of the algorithm, we terminate the algorithm when it meets the relative error tolerance (22) In the case of the DO scheme, m = 2 and A(P) is a tridiagonal matrix (see (25a) and (25b) of Appendix A). Therefore, we use a tridiagonal solver to solve A(P)x = b(P). As for the OD scheme, m = 3 and we use the Newton's method described in [10, Section 4] to solve A(P)x2 = b(P). Denoting by x(k) the iterates produced by Newton's method, we terminate the algorithm when it meets the error tolerance (cid:54) 10−12 + 10−6(cid:13)(cid:13)(cid:13)u(k) (cid:13)(cid:13)(cid:13)∞ . (cid:13)(cid:13)(cid:13)∞ (cid:13)(cid:13)(cid:13)x(k) − x(k−1) (cid:13)(cid:13)(cid:13)∞ (cid:54) 10−24 + 10−12(cid:13)(cid:13)(cid:13)x(k) (cid:13)(cid:13)(cid:13)∞ . Convergence results are given in Table 1, in which we report a representative value of the numerical solution (Value), the relative error (Rel. err.), ratio of errors (Ratio), number of policy iterations (Its.), average number of iterations (Inner its.) to solve the system A(P)xm−1 = b(P) if applicable, and total time elapsed in seconds (Time). The representative value of the numerical solution is u∆x(x0) where x0 = 1/2 is the midpoint of Ω. The relative error is given by (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)u∆x(x0) − U(x0) U(x0) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) where U is the exact solution. Since the exact solution is generally unavailable, we replace U by the solution computed by the OD scheme at a level of refinement higher than that which is shown in the table. The ratio of errors is given by u∆x/2(x0) − u∆x(x0) u∆x/4(x0) − u∆x/2(x0) 20 M 32 64 128 256 512 1024 2048 Value Rel. err. Ratio 3.0703 3.3567 3.5114 3.5917 3.6327 3.6534 3.6638 1.6e−1 8.5e−2 4.3e−2 2.1e−2 9.9e−3 4.3e−3 1.4e−3 1.85 1.92 1.96 1.98 1.99 Its. 3 3 3 3 3 3 3 Inner. its. 6.67 6.67 6.67 6.67 6.67 6.67 7.67 (a) Optimize then discretize M K Value Rel. err. Ratio 32 64 128 256 512 1024 2048 7.5e−1 4.9e−1 1.2e−1 3.6e−2 1.4e−2 5.5e−3 1.7e−3 0.9273 1.8839 3.2430 3.5376 3.6163 3.6490 3.6629 0.70 4.61 3.74 2.41 2.35 1 2 4 8 16 32 64 Its. 3 5 7 9 14 15 7 Time 4.3e−2 3.7e−2 5.4e−2 9.3e−2 1.7e−1 3.3e−1 6.5e−1 Time 1.7e−2 3.9e−2 1.4e−1 6.4e−1 3.5e+0 1.4e+1 2.3e+1 (b) Discretize then optimize Table 2: Convergence results (parameters used: (23)) so that the base-2 logarithm of this quantity gives us an estimate on the order of convergence (e.g., Ratio ≈ 2 suggests linear convergence, Ratio ≈ 4 suggests quadratic, etc.). Plots of the solution and optimal controls are given in Figure 2. The OD scheme is faster and more accurate than the DO scheme. Since both schemes require roughly the same number of policy iterations, it follows that the DO scheme loses most of its time on the pivot step on line 3 of Policy-Iteration. As K → ∞, this effect becomes more pronounced. Note also that the OD scheme exhibits a fairly stable linear order of convergence while that of the DO scheme is erratic. 6.3 A problem which is neither s.d.d. nor weakly irreducibly di- agonally dominant We now turn to a parameterization of (12) whose corresponding "discretization tensor" A(λ) defined by (16a) and (16b) is neither s.d.d. nor weakly irreducibly diagonally dominant. We are motivated by an analogous phenomenon that occurs for classical discretizations of degenerate elliptic differential equations first studied in [6] in which the matrix arising from the discretization is neither s.d.d. nor irreducibly diagonally dominant (cf. [18, 2, 7]). The parameterization we study is σ(x, λ) = 0.3(1 − λ) µ(x, λ) = 0.04λ (23) where λ ∈ Λ = {0, 1}. Note that this parameterization does not satisfy (A2) or (A3) since η is allowed to be zero. Therefore, the argument used to establish that A(λ) is a strong α(x) = 1 β(x) = 1 η(x) = 1{x(cid:54)0.5}(x) g(x) = 1 21 M-tensor in Section 6.1 fails (see, in particular, (19)). Letting ω be any function which maps R++ to itself such that limt→0 ω(t) = 0, one way to get around the above issue is to replace A(λ) by A(λ) + Iω(∆x) in (18), since the latter is trivially s.d.d. The obvious downside of this approach is that it introduces additional discretization error. Fortunately, it turns out that we can directly establish that A(λ) is a strong M-tensor by relying on the theory of w.c.d.d. tensors. In particular, by (19), (i) A(λ) is a w.d.d. (not s.d.d.) Z-tensor since ηi (cid:62) 0 and (ii) 0, M ∈ J(A(λ)) are s.d.d. rows. The directed graph of A(λ) is shown in Figure 4 where we have (i) ignored self-loops of the form i → i and (ii) used a dashed line to indicate an edge that is present only when λi = 0. Note that for any i /∈ J(A(λ)), we can form the walk i → i + 1 → ··· → M ending at the s.d.d. vertex M ∈ J(A(λ)). Therefore, A(λ) is w.c.d.d. and hence a strong M-tensor by Theorem 1. Now, by Theorem 29, we can compute a solution of the OD scheme as applied to the parameterization (23) by policy iteration. Convergence results and plots are shown in Table 2 and Figure 3, respectively. The C1 discontinuity in the solution is due to the discontinuity in η. 7 Summary In this work, we introduced the high order Bellman equation (1), extending classical Bellman equations to the tensor setting. We also introduced w.c.d.d. tensors (Definition 15), also extending the notion of w.c.d.d. matrices to the tensor setting. We established a relationship between w.c.d.d. tensors and M-tensors (Theorem 1), analogous to the relationship between w.c.d.d. matrices and M-matrices [1]. We proved that a sufficient condition to ensure the existence and uniqueness of a positive solution to a high order Bellman equation is that the tensors appearing in the equation are s.d.d. M-tensors (Lemma 23 and Lemma 24). We also showed that the s.d.d. requirement can be relaxed to the weaker requirement of w.c.d.d. so long as we restrict ourselves to a finite set of policies (Lemma 26). In this case, the solution of (1) can be computed by a policy iteration algorithm (Theorem 29). The question of whether or not the assumption of finitude can be removed remains open (Remark 27). We applied our findings to create a so-called "optimize then discretize" scheme for an optimal stochastic control problem which outperforms (in both computation time and accuracy) a classical "discretize then optimize" approach (Section 6). ifclassloadedsiamart1116 22 (a) M = 128 (b) M = 512 Figure 2: Solution and controls computed by both schemes (parameters used: (21)) 23 0.00.20.40.60.81.0x1.01.52.02.53.0U(x)Optimize then discretizeDiscretize then optimize0.00.20.40.60.81.0x1.00.50.00.51.0λ(x)0.00.20.40.60.81.0x0.00.51.01.52.0γ(x)0.00.20.40.60.81.0x1.01.52.02.53.0U(x)Optimize then discretizeDiscretize then optimize0.00.20.40.60.81.0x1.00.50.00.51.0λ(x)0.00.20.40.60.81.0x0.00.51.01.52.0γ(x)Figure 3: Solution and controls computed (parameters used: (23); OD scheme only) 0 1 ··· i−1 i i+1 ··· M−1 M Figure 4: Directed graph of A defined by (16a) and (16b) with σ and µ given by (23) 24 0.00.20.40.60.81.0x1.01.52.02.53.03.54.0U(x)0.00.20.40.60.81.0x0.00.20.40.60.81.0λ(x)0.00.20.40.60.81.0x0.20.30.40.50.60.70.80.91.0γ(x)References [1] P. Azimzadeh, A fast and stable test to check if a weakly diagonally dominant matrix is a nonsingular M-matrix, Math. Comp., (2018). [2] P. Azimzadeh and P. A. Forsyth, Weakly chained matrices, policy iteration, and impulse control, SIAM J. Numer. Anal., 54 (2016), pp. 1341 -- 1364, https://doi.org/ 10.1137/15M1043431. [3] G. Barles and P. E. Souganidis, Convergence of approximation schemes for fully nonlinear second order equations, Asymptotic Anal., 4 (1991), pp. 271 -- 283. [4] J. T. Betts and S. L. Campbell, Discretize then optimize, in Mathematics for industry: challenges and frontiers, SIAM, Philadelphia, PA, 2005, pp. 140 -- 157. [5] O. Bokanowski, S. Maroso, and H. Zidani, Some convergence results for Howard's algorithm, SIAM J. Numer. Anal., 47 (2009), pp. 3001 -- 3026, https://doi.org/10. 1137/08073041X. [6] J. H. Bramble and B. E. Hubbard, On a finite difference analogue of an elliptic boundary problem which is neither diagonally dominant nor of non-negative type, J. Math. and Phys., 43 (1964), pp. 117 -- 132. [7] Y. Chen, J. W. L. Wan, and J. Lin, Monotone mixed finite difference scheme for Monge-ampère equation, Journal of Scientific Computing, (2018). [8] M. G. Crandall, H. Ishii, and P.-L. Lions, User's guide to viscosity solutions of second order partial differential equations, Bull. Amer. Math. Soc. (N.S.), 27 (1992), pp. 1 -- 67, https://doi.org/10.1090/S0273-0979-1992-00266-5. [9] W. Ding, L. Qi, and Y. Wei, M-tensors and nonsingular M-tensors, Linear Algebra Appl., 439 (2013), pp. 3264 -- 3278, https://doi.org/10.1016/j.laa.2013.08.038. [10] W. Ding and Y. Wei, Solving multi-linear systems with M-tensors, J. Sci. Comput., 68 (2016), pp. 689 -- 715, https://doi.org/10.1007/s10915-015-0156-7. [11] P. A. Forsyth and G. Labahn, Numerical methods for controlled Hamilton-Jacobi- Bellman PDEs in finance, Journal of Computational Finance, 11 (2007). [12] S. Friedland, S. Gaubert, and L. Han, Perron-Frobenius theorem for nonnegative multilinear forms and extensions, Linear Algebra Appl., 438 (2013), pp. 738 -- 749, https: //doi.org/10.1016/j.laa.2011.02.042. [13] S. Hu, Z. Huang, and L. Qi, Strictly nonnegative tensors and nonnegative ten- sor partition, Sci. China Math., 57 (2014), pp. 181 -- 195, https://doi.org/10.1007/ s11425-013-4752-4, https://doi.org/10.1007/s11425-013-4752-4. [14] H. J. Kushner and P. G. Dupuis, Numerical methods for stochastic control problems in continuous time, vol. 24 of Applications of Mathematics (New York), Springer-Verlag, New York, 1992, https://doi.org/10.1007/978-1-4684-0441-8. [15] A. M. Oberman, Convergent difference schemes for degenerate elliptic and parabolic equations: Hamilton-Jacobi equations and free boundary problems, SIAM J. Numer. Anal., 44 (2006), pp. 879 -- 895, https://doi.org/10.1137/S0036142903435235. [16] L. Qi, Eigenvalues of a real supersymmetric tensor, J. Symbolic Comput., 40 (2005), 25 pp. 1302 -- 1324, https://doi.org/10.1016/j.jsc.2005.05.007. [17] L. Qi and Z. Luo, Tensor analysis, Society for Industrial and Applied Mathematics, Philadelphia, PA, 2017, https://doi.org/10.1137/1.9781611974751.ch1. Spectral theory and special tensors. [18] P. N. Shivakumar, J. J. Williams, Q. Ye, and C. A. Marinov, On two- sided bounds related to weakly diagonally dominant M-matrices with application to digital circuit dynamics, SIAM J. Matrix Anal. Appl., 17 (1996), pp. 298 -- 312, https: //doi.org/10.1137/S0895479894276370. [19] N. Touzi, Optimal stochastic control, stochastic target problems, and backward SDE, vol. 29 of Fields Institute Monographs, Springer, New York; Fields Institute for Re- search in Mathematical Sciences, Toronto, ON, 2013, https://doi.org/10.1007/ 978-1-4614-4286-8. With Chapter 13 by Angès Tourin. [20] L. Zhang, L. Qi, and G. Zhou, M-tensors and some applications, SIAM J. Matrix Anal. Appl., 35 (2014), pp. 437 -- 452, https://doi.org/10.1137/130915339. A Discretize then optimize scheme In this appendix, we derive the DO scheme for the differential equation (12). Since the set Γ = [0,∞) is unbounded, we restrict our attention to controls γ in some bounded interval Γ0 = [0, γmax]. To ensure the consistency of the resulting scheme, γmax should be chosen sufficiently large. For each ∆x > 0, we let Γ∆x denote a finite subset of Γ0 such that dH(Γ∆x, Γ0) → 0 as ∆x ↓ 0. The DO discretization is given by the M + 1 equations for 0 < i < M for i = 0, M (24) − ui = gi max (γ,λ)∈Γ∆x×Λ∆x n(Lλ∆xu)i − ηiui − 1 2αiγ2ui + βiγ o = 0, where the various quantities Lλ∆x, ηi, etc. are defined in Section 6.1 (compare (24) with the OD discretization (14)). We can transform (24) into a classical m = 2 Bellman equation as follows. Define the Cartesian product Γ∆x = (Γ∆x)M+1 and denote by γ = (γ0, . . . , γM) an element of Γ∆x with γi ∈ Γ∆x. Define Λ∆x and λ = (λ0, . . . , λM) similarly. Let A(γ, λ) = (aij(γ, λ)) be the tridiagonal matrix whose only nonzero entries are (∆x)2 + µi(λi) 1 1 ∆x 2 (∆x)2 + µi(λi) 1 ∆x 1 (∆x)2 − µi(λi) 1 ∆x ai,i−1(γ, λ) = −1 ai,i(γ, λ) = +1 ai,i+1(γ, λ) = −1 1(−∞,0)(µi(λi)), + ηi + 1 2 αiγ2 i , 1(0,+∞)(µi(λi)), 2 σi(λi)2 2 σi(λi)2 2 σi(λi)2 if 0 < i < M (25a) and ai,i(γ, λ) = 1, if i = 0, M. (25b) 26 Lastly, define the vector b(γ) = (b0(γ), . . . , bM(γ)) by βiγi, gi, bi(γ) = if 0 < i < M if i = 0, M. Then, the discrete equations (24) are equivalent to the classical Bellman equation min (γ,λ)∈Γ∆x×Λ∆x {A(γ, λ)u − b(γ)} = 0. The downside of the above approach is twofold: (i) The size of the policy set is P = Γ∆xΛ∆x. In the OD approach, the size of the policy set is P = Λ∆x (recall that the policy iteration algorithm takes, in the worst case, P iterations). (ii) Assuming γmax is chosen large enough, the approximation of Γ0 by Γ∆x introduces O(dH(Γ∆x, Γ0)) local truncation error. 27
1307.7339	2	1307	2013-08-29T18:53:25	Uniquely Strongly Clean Triangular Matrix Rings	[ "math.RA" ]	A ring $R$ is uniquely (strongly) clean provided that for any $a\in R$ there exists a unique idempotent $e\in R \big(\in comm(a)\big)$ such that $a-e\in U(R)$. Let $R$ be a uniquely bleached ring. We prove, in this note, that $R$ is uniquely clean if and only if $R$ is abelian, and $T_n(R)$ is uniquely strongly clean for all $n\geq 1$, if and only if $R$ is abelian, $T_n(R)$ is uniquely strongly clean for some $n\geq 1$. In the commutative case, the more explicit results are obtained. These also generalize the main theorems in [6] and [7], and provide many new class of such rings.	math.RA	math	Uniquely Strongly Clean Triangular Matrix Rings Huanyin Chen Department of Mathematics, Hangzhou Normal University Hangzhou 310036, China, [email protected] O. Gurgun Department of Mathematics, Ankara University 06100 Ankara, Turkey, [email protected] H. Kose Department of Mathematics, Ahi Evran University Kirsehir, Turkey, [email protected] Abstract A ring R is uniquely (strongly) clean provided that for any a ∈ R there exists a unique idempotent e ∈ R(cid:0) ∈ comm(a)(cid:1) such that a − e ∈ U (R). Let R be a uniquely bleached ring. We prove, in this note, that R is uniquely clean if and only if R is abelian, and Tn(R) is uniquely strongly clean for all n ≥ 1, if and only if R is abelian, Tn(R) is uniquely strongly clean for some n ≥ 1. In the commutative case, the more explicit results are obtained. These also generalize the main theorems in [6] and [7], and provide many new class of such rings. Keywords: uniquely strongly clean ring, uniquely bleached ring, triangular matrix ring. 1 1 Introduction The commutant of an element a in a ring R is defined by comm(a) = {x ∈ R xa = ax}. A ring R is strongly clean provided that for any a ∈ R there exists an idempotent e ∈ comm(a) such that a − e ∈ U (R). Strongly clean triangular matrix rings are extensively studied by many authors, e.g., [1] and [3]. A ring R is called uniquely clean provided that for any a ∈ R there exists a unique idempotent e ∈ R such that a − e ∈ U (R). Many characterizations of such rings are studied in [2-4] and [8]. Following J. Chen; Z. Wang and Y. Zhou [6], a ring R is called uniquely strongly clean provided that for any a ∈ R there exists a unique idempotent e ∈ comm(a) such that a − e ∈ U (R). Uniquely strongly cleanness behaves very different from the properties of uniquely clean rings (cf. [6]). In general, matrix rings have not such properties (cf. [10, Proposition 11.8]). Thus, it is attractive to investigate uniquely strongly cleanness of triangular matrix rings. Chen et al. proved that if R is commutative then R is uniquely clean if and only if Tn(R) is uniquely strongly clean for all n ≥ 1, if and only if Tn(R) is uniquely strongly clean for some n ≥ 1. [6, Question 12] and [10, Question 11.13] asked that if "commutative" in the preceding result can be replaced by "abelian". The motivation of this note article is to explore this problem. Recall that a ring R is uniquely bleached provided that for any a ∈ J(R), b ∈ U (R), 2 Huanyin Chen, O. Gurgun and H. Kose la − rb and lb − ra are isomorphism. Let R be a uniquely bleached ring. We prove, in this note, that R is uniquely clean if and only if R is abelian, and Tn(R) is uniquely strongly clean for all n ≥ 1, if and only if R is abelian, Tn(R) is uniquely strongly clean for some n ≥ 1. In the commutative case, the more explicit results are obtained. These also generalize the main theorems in [6] and [7], and provide many new class of such rings. We write U (R) for the set of all invertible elements in R and J(R) for the Jacobson radical of R. Tn(R) stand for the rings of all n × n triangular matrices over a ring R. Let a, b ∈ R. We denote the map from R to R : x 7→ ax − xb by la − rb. 2 The Main Results Clearly,we have { uniquely clean rings } ( { abelian clean rings } ( { uniquely strongly clean rings }. We begin with Theorem 1. Let R be a uniquely bleached ring. Then the following are equivalent: (1) R is uniquely clean. (2) R is abelian, and Tn(R) is uniquely strongly clean for all n ≥ 1. (3) R is abelian, Tn(R) is uniquely strongly clean for some n ≥ 1. Proof. (1) ⇒ (2) In view of [8, Theorem 20], R is abelian. Clearly, the result holds for n = 1. Assume that the result holds for n(n ≥ 1). Let A = (cid:18) a11 α A1 (cid:19) ∈ Tn+1(R) where a11 ∈ R, α ∈ M1×n(R) and A1 ∈ Tn(R). By hypothesis, we can find a unique idempotent e11 ∈ R such that u11 := a11 − e11 ∈ U (R) and a11e11 = e11a11. Furthermore, we have a unique idempotent E1 ∈ Tn(R) such that U1 := A1 − E1 ∈ U(cid:0)Tn(R)(cid:1) and U1 (cid:19), A1E1 = E1A1; hence, U1E1 = E1U1. Let E = (cid:18) e11 E1 (cid:19) and U = (cid:18) u11 α − x x where x ∈ M1×n(R). Observing that E2 = E ⇔ e11x + xE1 = x; U E = EU ⇔ u11x + (α − x)E1 = e11(α − x) + xU1, and then (u11 + 2e11 − 1)x − xU1 = e11α − αE1 (∗). In view of [8, Theorem 20], R/J(R) is Boolean, and so 2 ∈ J(R). Furthermore, u11 ∈ 1 + J(R). This shows that u11 + 2e11 − 1 ∈ J(R). Write · · · c11 xn (cid:1) , e11α − αE1 = (cid:0) v1 · · · vn (cid:1) , , where each cii ∈ U (R). x = (cid:0) x1   U1 = c12 c22 · · · · · · . . . c1n c2n cnn   Then (u11 + 2e11 − 1)x1 − x1c11 = v1; (u11 + 2e11 − 1)x2 − x2c22 = v1 + x1c12; ... (u11 + 2e11 − 1)xn − xncnn = vn + x1c1n + · · · + xn−1c(n−1)n. Uniquely Strongly Clean Triangular Matrix Rings 3 By hypothesis, we have a unique x such that (∗) holds. As E1U1 = U1E1, we get (u11 + 2e11 − 1)x(e11In + E1) − x(e11In + E1)U1 = α(e11In − E1)(e11In + E1) = α(e11In − E1) = (u11 + 2e11 − 1)x − xU1. Set y = x(e11In + E1). Then (u11 + 2e11 − 1)(y − x) − (y − x)U1 = 0. Write y − x = (cid:0) z1 · · · zn (cid:1). Then (u11 + 2e11 − 1)z1 − z1c11 = 0; (u11 + 2e11 − 1)z2 − z2c22 = z1c12; ... (u11 + 2e11 − 1)zn − zncnn = z1c1n + · · · + zn−1c(n−1)n. By hypothesis, we get each zi = 0, and so y = x. This implies that e11x + xE1 = x. Furthermore, u11x − (α − x)E1 = e11(α − x) + xU1. Therefore, we have a uniquely strongly clean expression A = (cid:18) e11 x E1 (cid:19) +(cid:18) a11 − e11 α − x A1 − E1 (cid:19) . By induction, Tn(R) is uniquely strongly clean for all n ∈ N. (2) ⇒ (3) is trivial. (3) ⇒ (1) In light of [6, Example 5], R is uniquely strongly clean, hence the result by [8, ✷ Theorem 20]. Corollary 2. Let R be a ring with nil Jacobson radical. Then the following are equivalent: (1) R is uniquely clean. (2) R is abelian, and Tn(R) is uniquely strongly clean for all n ≥ 1. (3) R is abelian, and Tn(R) is uniquely strongly clean for some n ≥ 1. Proof. Let a ∈ U (R), b ∈ J(R). Write bn = 0. Choose ϕ = la−1 + la−2rb + · · · + la−n rbn−1 : R → R. One easily checks that (cid:0)la − rb(cid:1)ϕ = ϕ(cid:0)la − rb(cid:1) = 1R. Thus, la − rb : R → R are isomorphism. Similarly, lb − ra is isomorphic. Therefore R is uniquely bleached, and the result follows by Theorem 1. ✷ Corollary 3. Let R be a ring for which some power of each element in J(R) is central. Then the following are equivalent: (1) R is uniquely clean. (2) R is abelian, and Tn(R) is uniquely strongly clean for all n ≥ 1. (3) R is abelian, and Tn(R) is uniquely strongly clean for some n ≥ 1. Proof. Let a ∈ U (R), b ∈ J(R). Write bn ∈ C(R). Choose ϕ = la−1 + la−2rb + · · · + la−n rbn−1 : R → R. Then (cid:0)la − rb(cid:1)ϕ = l1 − la−n rbn = l1−a−nbn . Further, we check that ϕ(cid:0)la − rb(cid:1) = l1−a−nbn . Thus, la − rb : R → R is isomorphic. Likewise, lb − ra is isomorphic. That is, R is uniquely bleached. Therefore we complete the proof by Theorem 1. ✷ Immediately, we show that every triangular matrix ring over a Boolean ring is uniquely strongly clean. For instance, Tn(cid:0)Z2(cid:1)(n ≥ 2) is uniquely strongly clean, while it is not uniquely clean. 4 Huanyin Chen, O. Gurgun and H. Kose Corollary 4 [6, Theorem 10]. Let R be a commutative ring. Then the following are equivalent: (1) R is uniquely clean. (2) Tn(R) is uniquely strongly clean for all n ≥ 1. (3) Tn(R) is uniquely strongly clean for some n ≥ 1. Proof. It is clear as every commutative ring is uniquely bleached and abelian. ✷ Lemma 5. Let R be a ring. If T2(R) is uniquely strongly clean, then R is uniquely bleached. Proof. In view of [6, Example 5], R is uniquely strongly clean. Let a ∈ J(R) and b ∈ U (R), idempotents. Further, a−e11 ∈ U (R) and b−e22 ∈ U (R). As a−0 ∈ U (R) and b−1 ∈ U (R), and let r ∈ R. Choose A = (cid:18) a −r b (cid:19) ∈ T2(R). Then there exists a unique idempotent E = (eij) ∈ T2(R) such that A − E ∈ U(cid:0)T2(R)(cid:1) and EA = AE. Clearly, e11, e22 ∈ R are by the unique strong cleanness of R, we get e11 = 0 and e22 = 1. Thus, E = (cid:18) 0 x 1 (cid:19) for we have an idempotent F = (cid:18) 0 y 1 (cid:19) such that A − F ∈ U(cid:0)T2(R)(cid:1) and AF = F A. By some x ∈ R. It follows from EA = AE that ax − xb = r. Assume that ay − yb = r. Then the uniqueness of E, we get x = y. Therefore, la − rb : R → R is an isomorphism. Likewise, lb − ra : R → R is an isomorphism. Accordingly, R is uniquely bleached, as asserted. ✷ Theorem 6. Let R be an abelian ring. Then the following are equivalent: (1) R is uniquely bleached, uniquely clean. (2) Tn(R) is uniquely strongly clean for all n ≥ 1. (3) Tn(R) is uniquely strongly clean for some n ≥ 2. Proof. (1) ⇒ (2) is proved by Theorem 1. (2) ⇒ (3) is trivial. (3) ⇒ (1) In view of [6, Example 5], T2(R) is uniquely strongly clean. Further, R is uniquely strongly clean. As R is abelian, R is uniquely clean. Therefore the proof is complete by Lemma 5. ✷ Let Lat(R) be the lattice of all right ideals of a ring R. A ring R said to be a D-ring if Lat(R) is distribute, i.e., AT(B + C) = (AT B) + (AT C) for all A, B, C ∈ Lat(R). Corollary 7. Let R be a D-ring. Then the following are equivalent: (1) R is uniquely bleached, uniquely clean. (2) Tn(R) is uniquely strongly clean for all n ≥ 1. (3) Tn(R) is uniquely strongly clean for some n ≥ 1. Proof. Let e ∈ R be an idempotent, and let a ∈ R. Choose f = e − (1 − e)ae. Clearly, eR, f R ∈ Lat(R). Then eR = (cid:0)eRT f R(cid:1) + (cid:0)eRT(1 − f )R(cid:1); hence, f eR ⊆ eR. Thus, ef e = f e = f . This implies that ae = eae. Similarly, we show that ea = eae. Therefore R is abelian, and the proof is complete by Theorem 6. ✷ Corollary 8 ([7, Theorem 1]). Let R be a local ring. Then the following are equivalent: Uniquely Strongly Clean Triangular Matrix Rings 5 (1) R is uniquely bleached and R/J(R) ∼= Z2. (2) Tn(R) is uniquely strongly clean for all n ≥ 1. (3) Tn(R) is uniquely strongly clean for some n ≥ 2. Proof. As R is local, R is uniquely clean if and only if R/J(R) ∼= Z2, by [8, Theorem 15]. Therefore we complete the proof from Theorem 6. ✷ The double commutant of an element a in a ring R is defined by comm2(a) = {x ∈ R xy = yx for all y ∈ comm(a)}. Clearly, comm2(a) ⊆ comm(a). We end this note by a more explicit result than [6, Theorem 10]. Theorem 9. Let R be a commutative ring, and let n ∈ N. Then the following are equivalent: (1) R is uniquely clean. (2) For any A ∈ Tn(R), there exists a unique idempotent E ∈ comm2(A) such that A−E ∈ U(cid:0)Tn(R)(cid:1). Proof. (1) ⇒ (2) For any A ∈ Tn(R), we claim that there exists an idempotent E ∈ α result holds for n = 1, by [8, Theorem 20]. Suppose that the result holds for n − 1(n ≥ 2). Then there is an idempotent e11 ∈ R such that w11 := a11 − e11 ∈ J(R). By hypothesis, comm2(A) such that A − E ∈ J(cid:0)T2(R)(cid:1). As R is a commutative uniquely clean ring, the A1 (cid:19) ∈ Tn(R) where a11 ∈ R, α ∈ M1×(n−1)(R) and A1 ∈ Tn−1(R). Let A = (cid:18) a11 there is an idempotent E1 ∈ Tn−1(R) such that W1 := A1 − E1 ∈ J(cid:0)Tn−1(R)(cid:1) and E1 ∈ comm2(A1). E1 (cid:19), W1 + (cid:0)1 − 2e11 − w11(cid:1)In−1 ∈ In−1 + J(cid:0)Tn−1(R)(cid:1) ⊆ U(cid:0)Tn−1(R)(cid:1). Let E = (cid:18) e11 where β = α(E1 − e11In−1)(cid:0)W1 + (1 − 2e11 − w11)In−1(cid:1)−1 . Then A − E ∈ J(cid:0)Tn(R)(cid:1). As e11β + βE1 = β(E1 + e11In−1) = α(E1 − e11In−1)(E1 + e11In−1)(cid:0)W1 + (1 − 2e11 − w11)In−1(cid:1)−1 X1 (cid:19) ∈ comm(A), we have α(X1 − x11In−1) = γ(A1 − a11In−1). In view of [8, Theorem 20], R/J(R) is Boolean, and so 2 ∈ J(R). Thus, β = β, we see that E = E2. For any X = (cid:18) x11 γ Since E1 ∈ comm2(A1), as in the proof of [4, Theorem 4.11], we check that γ(A1 − a11In−1)(E1 − e11In−1) = α(X1 − x11In−1)(E1 − e11In−1) = α(E1 − e11In−1)(X1 − x11In−1) = β(cid:0)W1 + (1 − 2e11 − w11)In−1(cid:1)(X1 − x11In−1) = β(X1 − x11In−1)(cid:0)W1 + (1 − 2e11 − w11)In−1(cid:1). γ(A1 − a11In−1)(E1 − e11In−1) Moreover, = γ(E1 − e11In−1)(cid:0)E1 + W1 − (e11 + w11)In−1(cid:1) = γ(E1 − e11In−1)(cid:0)E1 + e11In−1 + (W1 − 2e11 − w11)In−1(cid:1) = γ(cid:0)E1 − e11In−1 + (E1 − e11In−1)(W1 − 2e11 − w11)In−1(cid:1) = γ(E1 − e11In−1)(cid:0)W1 + (1 − 2e11 − w11)In−1(cid:1). This shows that γ(E1 −e11In−1) = β(X1 −x11In−1). Thus, we get e11γ +βX1 = x11β +γE1; hence, EX = XE. That is, E ∈ comm2(A), as claimed. 6 Huanyin Chen, O. Gurgun and H. Kose Therefore, for any A ∈ Tn(R), there exists an idempotent E ∈ comm2(A) such that = In, we W := A − E ∈ J(cid:0)T2(R)(cid:1). Hence, A = (cid:0)In − E(cid:1) +(cid:0)2E − In + W(cid:1). As (cid:0)2E − In(cid:1)2 see that 2E − In ∈ U(cid:0)Tn(R)(cid:1). Additionally, In − E ∈ comm2(A). Suppose that there exists an idempotent F ∈ comm2(A) such that A − F ∈ U(cid:0)T2(R)(cid:1). Then In − E, F ∈ comm(A). In view of Corollary 4, Tn(R) is uniquely strongly clean. This implies that F = In − E, proving (1). (2) ⇒ (1) Let a ∈ R. Then A = diag(a, a, · · · , a) ∈ Tn(R). Then we can find a unique is an idempotent and a − e11 ∈ U (R). Suppose that a − e ∈ U (R) with an idempotent e ∈ R. Then F = diag(e, e, · · · , e) ∈ Tn(R) is an idempotent. Further, F ∈ comm2(A), and that idempotent E = (eij ) ∈ comm2(A) such that A − E ∈ U(cid:0)Tn(R)(cid:1). This implies that e11 ∈ R A − F ∈ U(cid:0)Tn(R)(cid:1). By the uniqueness, we get E = F , and then e = e11. Therefore R is uniquely clean, as asserted. ✷ Corollary 10. Let R be a commutative uniquely clean ring, and let A ∈ Tn(R). Then for any A ∈ Tn(R), there exists a unique idempotent E ∈ comm(A) such that A − E, A + E ∈ Proof. In view of Theorem 9, we have a unique idempotent E ∈ comm2(A2) such that U(cid:0)Tn(R)(cid:1). U := A2 − E ∈ GLn(R). Explicit, we may assume that U ∈ −In + J(cid:0)Tn(R)(cid:1). Thus, (A − E)(A + E) ∈ U(cid:0)Tn(R)(cid:1). Therefore, A − E, A + E ∈ U(cid:0)Tn(R)(cid:1). Assume that there is an idempotent F ∈ comm(A) such that A − F, A + F ∈ U(cid:0)Tn(R)(cid:1). Then A2 − F = (A − F )(A + F ) ∈ U(cid:0)Tn(R)(cid:1). Hence, (In − E) + F = (A2 + In − E) − (A2 − F ) ∈ U(cid:0)Tn(R)(cid:1). Clearly, (In − E)F = F (In − E), and so (cid:0)(In − E) − F(cid:1)(cid:0)(In − E) + F − In(cid:1) = 0. This shows that F = In − E, as desired. ✷ Corollary 11. Let R be a Boolean ring, and let A ∈ Tn(R). (1) There exists a unique idempotent E ∈ comm2(A) such that A − E ∈ U(cid:0)Tn(R)(cid:1). (2) There exists a unique idempotent E ∈ comm2(A) such that A − E, A + E ∈ U(cid:0)Tn(R)(cid:1). Proof. Clearly, R is a commutative uniquely clean ring, hence the result by Theorem 9 and Corollary 10. ✷ Acknowledgements This research was supported by the Natural Science Foundation of Zhejiang Province (LY13A0 10019) and the Scientific and Technological Research Council of Turkey (2221 Visiting Scientists Fellowship Programme). References [1] G. Borooah; A.J. Diesl and T.J. Dorsey, Strongly clean triangular matrix rings over local rings, J. Algebra, 312(2007), 773 -- 797. [2] H. Chen, On uniquely clean rings, Comm. Algebra, 39(2011), 189 -- 198. [3] H. Chen, Rings Related Stable Range Conditions, Series in Algebra 11, World Scientific, Hackensack, NJ, 2011. Uniquely Strongly Clean Triangular Matrix Rings 7 [4] H. Chen; S. Halicioglu and H. Kose, On perferctly clean rings, preprint, arXiv:1307.6087 [math.RA]. [5] J. Chen; W.K. Nicholson and Y. Zhou, Group rings in which every element is uniquely the sum of a unit and an idempotent, J. Algebra, 306(2006), 453 -- 460. [6] J. Chen; Z. Wang and Y. Zhou, Rings in which elements are uniquely the sum of an idempotent and a unit that commutate, J. Pure Appl. Algebra, 213(2009), 215 -- 223. [7] J. Cui and J. Chen, Uniquely strongly clean triangular matrix rings, J. Southeast Univ. (English Edition), 27(2011), 463 -- 465. [8] W.K. Nicholson and Y. Zhou, Rings in which elements are uniquely the sum of an idempotent and a unit, Glasgow Math. J., 46(2004), 227 -- 236. [9] X.L. Wang, Uniquely strongly clean group rings, Commu. Math. Research, 28(2012), 17 -- 25. [10] X. Yang, A survey of strongly clean rings, Acta Appl. Math., 108(2009), 157 -- 173.
1608.01183	3	1608	2017-11-02T15:59:16	Solvability and uniqueness criteria for generalized Sylvester-type equations	[ "math.RA", "math.FA", "math.NA" ]	We provide necessary and sufficient conditions for the generalized $\star$-Sylvester matrix equation, $AXB + CX^\star D = E$, to have exactly one solution for any right-hand side E. These conditions are given for arbitrary coefficient matrices $A, B, C, D$ (either square or rectangular) and generalize existing results for the same equation with square coefficients. We also review the known results regarding the existence and uniqueness of solution for generalized Sylvester and $\star$-Sylvester equations.	math.RA	math	Solvability and uniqueness criteria for generalized Sylvester-type equations∗ Fernando De Ter´an1, Bruno Iannazzo2, Federico Poloni3, and Leonardo Robol4 1 2 3 Departamento de Matem´aticas, Universidad Carlos III de Madrid, Avda. Universidad 30, 28911 Legan´es, Spain. [email protected]. Corresponding author. Dipartimento di Matematica e Informatica, Universit`a di Perugia, Via Vanvitelli 1, 06123 Perugia, Italy. [email protected]. Dipartimento di Informatica, Universit`a di Pisa, Largo B. Pontecorvo 3, 56127 Pisa, Italy. [email protected]. 4 Dept. Computerwetenschappen, KU Leuven, Celestijnenlaan 200A, 3001 Heverlee (Leuven), Belgium. [email protected]. March 22, 2018 Abstract We provide necessary and sufficient conditions for the generalized ⋆- Sylvester matrix equation, AXB + CX ⋆D = E, to have exactly one solu- tion for any right-hand side E. These conditions are given for arbitrary coefficient matrices A, B, C, D (either square or rectangular) and gener- alize existing results for the same equation with square coefficients. We also review the known results regarding the existence and uniqueness of solution for generalized Sylvester and ⋆-Sylvester equations. Keywords. Sylvester equation, eigenvalues, matrix pencil, matrix equation AMS classification: 15A22, 15A24, 65F15 1 Introduction We consider the generalized ⋆-Sylvester equation AXB + CX ⋆D = E (1) ∗ This work was partially supported by the Ministerio de Econom´ıa y Competitividad of Spain through grants MTM2015-68805-REDT, and MTM2015-65798-P (F. De Ter´an), and by an INdAM/GNCS Research Project 2016 (B. Iannazzo, F. Poloni, and L. Robol). Part of this work was done during a visit of the first author to the Universit`a di Perugia as a Visiting Researcher. 1 for the unknown X ∈ Cm×n, with ⋆ being either the transpose (⊤) or the con- jugate transpose (∗), and A, B, C, D, E being matrices with appropriate sizes. We are interested in the most general situation, where both the coefficients and the unknown are allowed to be rectangular. This equation is closely related to the generalized Sylvester equation AXB − CXD = E, (2) and equations (1) and (2) are natural extensions of the ⋆-Sylvester equation and the Sylvester equation, AX + X ⋆D = E and AX − XD = E, respectively. Sylvester-like equations are among the most popular matrix equations, and they arise in many applications (see, for instance, [1–3, 16] and the recent re- view [20]). In particular, equations with rectangular coefficients arise in several eigenvalue perturbation and updating problems [15, 21]. As with every class of equations, the two most natural questions regard- ing (1) and (2) are: S (solvability). Does the equation have a solution, for given A, B, C, D, E? US (unique solvability). Does the equation have exactly one solution, for given A, B, C, D, E? Moreover, three additional questions arise naturally for linear equations, due to their peculiar structure: SR (solvability for any right-hand side). Given A, B, C, D, does the equa- tion have at least one solution for any choice of the right-hand side E? OR (at most one solution for any right-hand side). Given A, B, C, D, does the equation have at most one solution for any choice of the right-hand side E? UR (unique solvability for any right-hand side). Given A, B, C, D, does the equation have exactly one solution for any choice of the right-hand side E? Though the contribution of this paper is mainly restricted to question UR, we present here an overview of the known results on all these problems for equations (1) and (2). The main results regarding these closely related problems are scattered within the literature, so we gather all them in Table 1 (the contents of this table are explained below). Using vectorizations, equations (1) and (2) can be transformed into a system of linear equations (either over the real or the complex field) of the form M x = e, where M depends only on the coefficients A, B, C, D, and e depends only on the right-hand side E. This is clear for (2), where M = B⊤ ⊗ A − D⊤ ⊗ C (see e.g. [12, Section 4.3]), and can be established with a little more effort for (1) (see Section 1.2). In this setting, question SR is equivalent to asking whether M has full row rank, and question OR is equivalent to asking whether M has full column rank. Question OR is also equivalent to asking whether the 2 square coefficients AXB − CXD = E general AXB + CX ⋆D = E square general coefficients coefficients coefficients [8, Th. 6.1] [4, Th. 1] S US SR same as US OR same as US UR same as US [8, Th. 6.1], [13, Th. 1] [13, Th. 1] Th. 10 (using [13]) [14, Cor. 5],Th. 10 (using [13]) Th. 10 (using [13]) [8, Th. 6.1] [6, Th. 15] same as US same as US same as US [8, Th. 6.1] open open open Th. 3 Table 1: Existing solvability and uniqueness results for Equations (1) and (2) in terms of "small-size" matrices and pencils. homogeneous equation (E = 0) has only the trivial solution X = 0. In order for the answer of question UR to be affirmative, M must be square and, in this case, all US, SR, and OR are equivalent to UR. Instead of the conditions on the matrix M , in many cases it is possible to find conditions of a different kind, related to the spectral properties of smaller matrix pencils. For instance, let us consider question UR in the case in which all matrices are square and have the same size m = n: for Equation (2), it has a positive answer if and only if the n× n matrix pencils A− λC and D − λB are regular and do not have common eigenvalues [4, Th. 1]; and for Equation (1) the answer depends on spectral properties of the 2n × 2n matrix pencil Q(λ) =(cid:20) λD⋆ B⋆ λC (cid:21) A (3) (see Theorem 2 for a precise statement, or [6, Th. 15] for more details). By con- trast, M has size n2 or 2n2. Even for equations with rectangular coefficients, the picture is the same: the characterizations that do not involve vectorization lead to matrix pencils with smaller size, of the same order as the coefficient matrices, while approaches based on Kronecker products lead to larger dimensions. Several authors have given conditions of this kind: in Table 1, we show an overview of these results. In this table we consider separately the cases where the coefficient matrices A, B, C, D are square, and the more general case where they have arbitrary size (as long as the product is well-defined). This distinction fits with the historical flow of the problem, as can be seen in the table. Note that, in both (1) and (2), if the coefficient matrices A, B, C, D are square then the coefficient matrix M of the associated linear system is square as well. However, there is a difference between (1) and (2) regarding this issue: whereas in (2) it may happen that all A, B, C, D are square but A and B have different size, in (1) if all coefficient matrices are square, then they must all have the same size in order for the products to be well-defined. The solution of Equation (2), allowing for rectangular coefficients, had been considered in [17], where an approach through the Kronecker canonical form of A − λC and D − λB was proposed. However, no explicit characterization 3 of the uniqueness of solution was given in that reference. Also, [15, Th. 3.2] presents a computation of the solution space of (2) with B = I, depending on the Kronecker canonical form of A − λC, but restricted to the case where this canonical form does not contain right singular blocks. In [18], the author identifies the correct line of attack of the problem, while a complete analysis of the solution space of Equation (2) is given by Kosir in [13]; the same author gives an explicit answer to question SR in [14]. Answers to OR and UR follow from Kosir's work [13], but they are not explicitly stated there; for completeness, we formulate them in Section 4. Instead, for the ⋆-Sylvester equation (1), to the best of our knowledge, several problems are still open; i.e., only characterizations based on vectorization and Kronecker products are known. The main goal of this paper is to give a solution to UR in the most general case of rectangular coefficients: we give necessary and sufficient conditions for the unique solvability of (1) in terms of the pencil Q(λ) in (3). Our focus on question UR is motivated by the fact that this is the only case where the operator X 7→ AXB + CX ⋆D is invertible. Moreover, this is the only case where the solution of the equation is a well-posed problem, since the (unique) solution depends continuously on the (entries of the) coefficient matrices. This is no longer true in the remaining cases. We emphasize that the approach followed in [6] for square coefficients cannot be applied in a straightforward manner to the rectangular case. Indeed, that approach is based on the characterization of the uniqueness of solution of the ⋆-Sylvester equation AX + X ⋆D = 0 provided in [3, 16], which is valid only for A, D, X ∈ Cn×n. We follow a different approach that allows us to extend that characterization to the case of rectangular coefficients. The most interesting feature of our characterization is the appearance of an additional invertibility constraint that is not present in the square case. Indeed, the unique solvability of (1) cannot be characterized completely in terms of the eigenvalues of Q(λ) only, as we show with a counterexample in Section 2.3. 1.1 The main result In this section we state the main result of this paper, namely Theorem 3. The rest of the paper is devoted to prove this result, but, before its statement, we introduce some notation and tools. Throughout the paper we denote by I the identity matrix of appropriate size, and by M −⋆ we denote the inverse of the matrix M ⋆, for an invertible matrix M . rectangular or q(λ) := det(cid:0)P(λ)(cid:1) is identically zero. If P(λ) is not singular, and A matrix pencil P(λ) = λM + N is said to be singular if either P(λ) is so M, N are n × n matrices, then it is said to be regular and the set of roots of q(λ), complemented with ∞ if the degree of q(λ) is less than n, is the spectrum of P, denoted by Λ(P). With mλ(P) we denote the algebraic multiplicity of the eigenvalue λ in P. 4 We shall deal with certain matrices and matrix pencils that always have m − n zero or infinite eigenvalues which are dimension-induced, that is, they are present simply because of the sizes of the coefficient matrices they are con- structed from (see [19]). Hence we define a variant of the spectrum in which these eigenvalues are omitted: bΛ(P) :=(cid:26) eΛ(P) :=(cid:26) Λ(P), Λ(P) \ {∞}, Λ(P), Λ(P) \ {0}, if m∞(P) > m − n, if m∞(P) = m − n, if m0(P) > m − n, if m0(P) = m − n. values. Following [19], we refer to the eigenvalues in either bΛ(P) oreΛ(P) as core eigen- The reversal pencil of the matrix pencil P(λ) = λM + N is the pencil rev P(λ) := λN + M . The pencil P(λ) has an infinite eigenvalue if and only if rev P(λ) has the zero eigenvalue. The multiplicity of the infinite eigenvalue in P(λ) is the multiplicity of the zero eigenvalue in rev P(λ), thus eΛ(rev P) =nλ−1 : λ ∈bΛ(P)o , (4) with 0−1 = ∞ and ∞−1 = 0. conjugate of λ. By λ⋆ we denote either λ, if ⋆ = ⊤, or λ, if ⋆ = ∗, with λ being the complex If M is a square matrix, by Λ(M ) and mλ(M ) we denote, respectively, the spectrum of M and the algebraic multiplicity of λ as an eigenvalue of M . Furthermore, we use eΛ(M ) to denote eΛ(λI − M ). We also recall the following notion, which plays a central role in Theorem 3. Definition 1. (Reciprocal free and ∗-reciprocal free set) [3, 16]. Let S be a subset of C ∪ {∞}. We say that S is (a) reciprocal free if λ 6= µ−1, for all λ, µ ∈ S; (b) ∗-reciprocal free if λ 6= (µ)−1, for all λ, µ ∈ S. This definition includes the values λ = 0,∞, with the customary assumption λ−1 = (λ)−1 = ∞, 0, respectively. Before stating the characterization of the uniqueness of solution in the gen- eral case, we recall here the main result in [6], namely the characterization of the uniqueness of solution of (1) when all coefficients are square and have the same size. Theorem 2. Then the equation AXB +CX ⋆D = E has a unique solution, for any right-hand side E, if and only if Q(λ) is regular and: [6, Th. 15] Let A, B, C, D ∈ Cn×n and let Q(λ) = (cid:2) λD A λC(cid:3). B ⋆ ⋆ • If ⋆ = ⊤, Λ(Q) \ {±1} is reciprocal free and m1(Q) = m−1(Q) 6 1. 5 • If ⋆ = ∗, Λ(Q) is ∗-reciprocal free. If we allow for rectangular coefficient matrices, that is A ∈ Cp×m, B ∈ Cn×q, C ∈ Cp×n, D ∈ Cm×q, then several subtleties arise, and in the end, they will result in additional restrictions on the pencil Q(λ). More precisely, we will prove that the only case in which unique solvability arises is when (p, q) ∈ {(m, n), (n, m)}. In the case where p = m, that is, A ∈ Cm×m, B ∈ Cn×n, C ∈ Cm×n, D ∈ Cm×n, the spectrum of the matrix pencil (3) contains the infinite eigenvalue, with multiplicity at least m − n. Then, in this case we denote by bΛ(Q) the following set obtained from Λ(Q): Λ(Q), Λ(Q) \ {∞}, if m∞(Q) > m − n, if m∞(Q) = m − n. bΛ(Q) :=(cid:26) eΛ(Q) :=(cid:26) If p = n, that is A ∈ Cn×m, B ∈ Cn×m, C ∈ Cm×m, D ∈ Cn×n, the spectrum of the matrix pencil (3) contains the zero eigenvalue, with multiplicity at least m − n. Then, in this case we denote by eΛ(Q) the following set obtained from Λ(Q): Λ(Q), Λ(Q) \ {0}, if m0(Q) > m − n, if m0(Q) = m − n. The presence of these additional zero/infinity eigenvalues of Q(λ) in (3) is due to the "rectangularity" of either the diagonal blocks C, D or the anti- diagonal blocks A and B. Following [10], based on the theory developed in [19], these extra zero/infinity eigenvalues are called dimension induced eigenvalues. With these considerations in mind, we can state the main result of this paper, The sets bΛ(Q) and eΛ(Q) are referred to as the set of core eigenvalues. which is an extension of Theorem 2 and which will be proved in Section 2. Theorem 3. Let A ∈ Cp×m, B ∈ Cn×q, C ∈ Cp×n, and D ∈ Cm×q and set Q(λ) :=(cid:2) λD A λC(cid:3). The equation AXB + CX ⋆D = E B ⋆ ⋆ has a unique solution, for any right-hand side E, if and only if Q(λ) is regular and one of the following situations holds: (1) p = m 6= n = q, either m < n and A is invertible or m > n and B is invertible, and invertible, and (2) p = n 6= m = q, either m > n and C is invertible or m < n and D is – If ⋆ = ⊤, bΛ(Q) \ {±1} is reciprocal free and m1(Q) = m−1(Q) 6 1. – If ⋆ = ∗, bΛ(Q) is ∗-reciprocal free. – If ⋆ = ⊤, eΛ(Q) \ {±1} is reciprocal free and m1(Q) = m−1(Q) 6 1. – If ⋆ = ∗, eΛ(Q) is ∗-reciprocal free. 6 (3) p = m = n = q, and – If ⋆ = ⊤, Λ(Q) \ {±1} is reciprocal free and m1(Q) = m−1(Q) 6 1. – If ⋆ = ∗, Λ(Q) is ∗-reciprocal free. 1.2 Vectorization Equation (1) can be considered as a linear system in the entries of the unknown matrix X. The natural approach to get such system is applying the vectorization (vec) operator [12, §4.3]. Set X ∈ Cm×n, and let A ∈ Cp×m, B ∈ Cn×q, C ∈ Cp×n, D ∈ Cm×q, E ∈ Cp×q. In the case ⋆ = ⊤, after applying the vec operator we obtain a linear equation M vec(X) = vec(E), with M ∈ C(pq)×(mn) given by M = B⊤ ⊗ A + (D⊤ ⊗ C)Π, (5) where Π is a permutation matrix associated with the transposition [12, Equa- tion 4.3.9b]. a linear system over R of size (2pq) × (2mn) in Y = vec((cid:2) re(X) In the case ⋆ = ∗, some more care is needed, since the system obtained by vectorization is not linear over C, due to the presence of conjugations. Never- theless, we can separate real and imaginary parts as in [6, §1.1] and write it as The fact that Equation (1) is equivalent to a linear system has two important consequences. The first one is that (1) can have a unique solution, for any right- hand-side, only if the coefficient matrix of the linear system is square, that is, mn = pq. The second one is that, provided that pq = mn, the uniqueness of solution does not depend on the right-hand side: Equation (1) has a unique solution for any E if and only if the corresponding homogeneous equation im(X) (cid:3)). AXB + CX ⋆D = 0 (6) has only the trivial solution X = 0. Hence, from now on, we assume mn = pq and we focus on Equation (6) instead of Equation (1). A different reformulation of the ⋆ = ∗ case as a linear system (in the homo- geneous case) is the following. Lemma 4. Equation (6), with ⋆ = ∗, has a unique solution if and only if the linear system of equations AXB + CY D = 0, D∗XC ∗ + B∗Y A∗ = 0, (7) has a unique solution. Proof. Let us first assume that (6) has a nonzero solution X. Then this gives a nonzero solution (X, X ∗) of (7). To prove the converse, let (X, Y ) be a nonzero solution of (7). Then, the matrix X + Y ∗ is a solution of (6). If X + Y ∗ is zero, then Y = −X ∗, and in this case iX is a nonzero solution of (6), with i := √−1. 7 The matrix associated to (7) after applying the vec operator is M =(cid:20) B⊤ ⊗ A D⊤ ⊗ C C ⊗ D∗ A ⊗ B∗ (cid:21) . (8) 2 Proof of Theorem 3 Here we provide a proof of Theorem 3, which gives a complete characterization of the uniqueness of solution of (1) for any right-hand side. We split the proof in four cases: (C1) mn 6= pq; (C2) mn = pq and p 6∈ {m, n}; (C3) mn = pq and p ∈ {m, n}, with m 6= n; (C4) mn = pq and p ∈ {m, n}, with m = n. In Case (C4) all coefficients are square and Theorem 3 holds, because it reduces to Theorem 2. In Case (C1) the theorem is true because Equation (1) fails to have a unique solution for any right-hand side E and the conditions (1)–(3) cannot be fulfilled. Indeed, in order for (1) to have a unique solution, for any right-hand side E, the coefficient matrix associated with Equation (1) must be square, and this implies mn = pq, as explained in Section 1.2. On the other hand, the conditions in each case (1)–(3) in the statement, imply mn = pq. Case (C2), in which every coefficient is non-square, namely p /∈ {m, n}, will be treated in Section 2.1, and Case (C3), in which two coefficients are square and two are non-square, namely p ∈ {m, n}, with m 6= n, will be treated in Section 2.2. 2.1 The case mn = pq and p 6∈ {m, n} In this section we show that Theorem 3 holds if p /∈ {m, n}, with mn = pq. In particular we will show that, in this case, the pencil Q(λ) is singular and Equation (1) fails to have a unique solution for the right hand side E = 0. Note that, because of the restriction mn = pq, this also implies that q /∈ {m, n}, so this situation covers all instances of Theorem 3 where none of the coefficient matrices are square. We first show that Q(λ) is non-square and thus singular. Note that Q(λ) has size (p + q) × (m + n). If p + q = m + n this fact, together with the identity mn = pq, would imply {m, n} = {p, q}, since both m, n and p, q are the roots of the same quadratic polynomial, namely x2 − (m + n)x + mn. Then, we show that Equation (6) admits a non-zero solution. Lemma 5. Let A ∈ Cp×m, B ∈ Cn×q, C ∈ Cp×n, D ∈ Cm×q. If mn = pq and p /∈ {m, n} then AXB + CX ⋆D = 0 has a nonzero solution. 8 Proof. We consider separately four cases, depending on whether p is smaller or larger than m and n. 1. p < min{m, n}. There are two nonzero vectors u, v such that Au = 0 and Cv = 0, because of the dimensions of these two matrices. Then X = uv⋆ is a nonzero solution of (6). 2. If p > max{m, n}, the identity mn = pq implies q < min{m, n}. Then, there are two nonzero vectors u, v such that v⋆B = 0, u⋆D = 0, and X = uv⋆ is a nonzero solution of (6). 3. m < p < n. In this case, and because of the identity mn = pq, we have m < q < n as well. Therefore, m < min{p, q}. In particular, there exist nonzero vectors u, v such that u⊤A = 0, v⊤D⊤ = 0. Now we consider the cases: (a) ⋆ = ⊤. As argued in Section 1.2, Equation (6) is equivalent to the linear system M vec X = 0, with the matrix M ∈ C(mn)×(mn) as in (5). Then, (v⊤ ⊗ u⊤)M = 0, so M is singular and (6) has a nonzero solution. (b) ⋆ = ∗. As a consequence of Lemma 4, Equation (6) has a nonzero solution if and only if the (square) matrix (8) is singular. It is easy to verify thathv⊤ ⊗ u⊤ u∗ ⊗ v∗i M = 0, so M is indeed singular. 4. n < p < m. By setting Y = X ⋆, Equation (6) is equivalent to CY D + AY ⋆B = 0, so we use the result for the previous case on this equation. 2.2 The case mn = pq and p ∈ {m, n}, with m 6= n In this section we show that Theorem 3 holds for p ∈ {m, n}, with mn = pq and m 6= n. Since for mn = pq the matrix associated with Equation (1) is square, the unique solvability of the generalized ⋆-Sylvester equation (1), for any right-hand side, is equivalent to the existence of a unique solution of the homogeneous equation (6). Then, it is sufficient to prove the equivalence of conditions on the pencil Q(λ) with the uniqueness of solution of the homogeneous equation (6). We have either p = m or p = n, which imply q = n and q = m, respectively, due to the constraint mn = pq. We will prove first the case in which p = m > n = q; then we will reduce the remaining cases: p = m < n = q, and p = n 6= m, to this one. Let us assume that p = m > n = q, so that D has more rows than columns, and there is some u 6= 0 such that u⋆D = 0. If B is singular, then there is some v 6= 0 such that v⋆B = 0. Therefore X = uv⋆ is a nontrivial solution of (6). Assume now that (6) has a unique solution. Then, B is guaranteed to be nonsingular, and AXB + CX ⋆D = 0 has a unique solution if and only if 9 AX +CX ⋆DB−1 = 0 has a unique solution. Moreover, we can find an invertible matrix Q ∈ Cm×m such that QDB−1 =(cid:20) D1 0 (cid:21) , (9) (10) (11) with D1 ∈ Cn×n. This allows us to rewrite (6), after multiplying on the right by B−1, and setting Y = Q−⋆X, in the equivalent form Y ⋆ 1 If we partition AQ⋆ 0 (cid:21) = 0, AQ⋆(cid:20) Y1 eA22i has full column with eA11 ∈ Cn×n, eA22 ∈ C(m−n)×(m−n), then the blockh eA12 rank. If that was not the case we could find Y2 6= 0 such that where Y1 has size n × n and Y2 has size (m − n) × n. conformally as 2 (cid:3)(cid:20) D1 Y2 (cid:21) + C(cid:2) Y ⋆ eA22 # , AQ⋆ =" eA11 eA12 eA21 eA22# Y2 = 0, "eA12 2(cid:3)(cid:20)D1 0 (cid:21) = 0, Y2(cid:21) + C(cid:2)0 Y ⋆ eA22 # =" bA11 bA22 # , eA12 bA21 If we set U C = h bC1 bC2i, with bC1 ∈ with bA22 ∈ C(m−n)×(m−n) nonsingular. Cn×n,bC2 ∈ C(m−n)×n then, after multiplying on the left by U , (10) is equivalent AQ⋆(cid:20) 0 so equation (10) would have a nontrivial solution. Then, there is an invertible matrix U ∈ Cm×m such that U AQ⋆ = U" eA11 eA21 and this would imply to the system 0 bA11Y1 + bC1Y ⋆ bA22Y2 1 D1 = 0, = −(bA21Y1 + bC2Y ⋆ 1 D1). Since bA22 is nonsingular, the above system has a unique solution if and only if the first equation 1 D1 = 0 (12) has a unique solution. bA11Y1 + bC1Y ⋆ 10 We are now ready to relate the uniqueness of solution of (6) to the spectral properties of the pencil Q(λ) in the statement of the theorem. We perform the following left and right invertible transformations to Q(λ): 0 0 λD⋆ 1 (13) 0 I (cid:20) B−⋆ 0 U (cid:21)(cid:20) λD⋆ B⋆ A Set  . λbC1 bA22 λbC2 0 0 I (cid:21) = λC (cid:21)(cid:20) Q⋆ bA11 bA21 λbC1 (cid:21) . bQ(λ) =(cid:20) λD⋆ bA11 I 1 well) and is a nonzero constant. Since all coefficient matrices in (12) are square and have the same size, Then, by (13), det Q(λ) = α det bQ(λ), where α = ±(det bA22 det B⋆)/(det U det Q⋆) namely n × n, Theorem 2 implies that bQ(λ) is regular (so Q(λ) is regular as • If ⋆ = ⊤, Λ(bQ) \ {±1} is reciprocal free and m1(bQ) = m−1(bQ) 6 1. • If ⋆ = ∗, Λ(bQ) is ∗-reciprocal free. Note that (13) implies that bΛ(Q) = Λ(bQ), since bQ is obtained by deflating m − n infinite eigenvalues from the pencil in the right hand side of (13). So the previous two conditions are equivalent to the conditions on the spectrum of Q(λ) in the statement of the theorem. To prove the converse, let us assume that B is invertible, and that Q(λ) is regular and its spectrum satisfies the conditions in the statement of the theorem. Then we can define the matrix Q as in (9) and we arrive at (11). Again, the I (cid:21)(cid:20) λD⋆ B⋆ singular. This is an immediate consequence of the identity: block h eA12 eA22i has full column rank since, otherwise, the pencil Q(λ) would be (cid:20) B−⋆ whereh eC1 eC2i = C. fact that bΛ(Q) = Λ(bQ) and applying Theorem 2, with the hypotheses on Q(λ), to Equation (12), we conclude that the latter has a unique solution, and this implies that (6) has a unique solution. Proceeding as before, we conclude, that (6) is equivalent to (12). Using the eA12 λeC1 eA22 λeC2 I (cid:21) = λC (cid:21)(cid:20) Q⋆ eA11 eA21  , λD⋆ 1 A 0 0 0 0 0 I Now that we have proved the case p = m > n = q, we consider the case p = m < n = q and then the case p = n 6= m. Let us assume that p = m < n = q. After applying the ⋆ operator in (6) and setting Y = X ⋆, we arrive at the equivalent equation B⋆Y A⋆ + D⋆Y ⋆C ⋆ = 0. This equation is of the form (6), with the coefficients of the first summand being 11 square, B⋆ ∈ Cn×n, A⋆ ∈ Cm×m and n > m, so we are in the same conditions as before. Applying the result just proved for this case, we get that the unique solvability is equivalent to requiring that A is invertible and that the pencil satisfies the conditions in the statement of the theorem. But, since A eQ(λ) =(cid:20) λC eQ(λ) =(cid:20) 0 B⋆ λD⋆ (cid:21) 0 (cid:21) Q(λ)(cid:20) 0 I I I I 0 (cid:21) , this is equivalent to requiring that Q(λ) satisfies these conditions as well. For the case p = n 6= m, we apply the ⋆ operator in (6), and we arrive at the equivalent equation D⋆XC ⋆ + B⋆X ⋆A⋆ = 0, whose coefficients are in the conditions of the previous case. The pencil associated to this last equation is This pencil is the reversal of the pencil: C eQ(λ) =(cid:20) λA (cid:20) 0 D⋆ λB⋆ (cid:21) . 0 (cid:21) Q(λ), I I so Λ(eQ) = Λ−1(Q) := {λ−1 : λ ∈ Λ(Q)}, including multiplicities. In particular, the conditions on being (∗-)reciprocal free in the statement are the same for both pencils, and the roles of the zero and the infinite eigenvalue are exchanged. Remark 6. The conditions n = q in part 1, and m = q in part 2 in Theorem 3 are redundant, but we have included them for emphasis. These conditions are a consequence of the fact that Q(λ) in (3) is regular and the other conditions on the size, namely p = m and p = n, respectively. As indicated in the proof of Theorem 3, since Q(λ) has size (p + q) × (m + n), if it is regular, it must be, in particular, square, and this implies m + n = p + q. 2.3 Necessity of the invertibility conditions The characterization of the uniqueness of solution of (1) in Theorem 3 involves, in cases 1 and 2, the invertibility of some of the coefficient matrices. One might wonder if these conditions are really needed, or whether they could be stated in terms of spectral properties of the pencil Q(λ). However, the following example shows that the uniqueness of solution does not depend solely on the eigenvalues of Q(λ). Consider the following generalized ⊤-Sylvester equations (the same example works for the ⋆ = ∗ case): 0 (cid:20)1 (cid:20)0 0 0 1(cid:21)(cid:20)x y(cid:21)(cid:2)0(cid:3) +(cid:20)1 1(cid:21)(cid:20)x y(cid:21)(cid:2)1(cid:3) +(cid:20)1 0(cid:21)(cid:2)x y(cid:3)(cid:20)1 0(cid:21)(cid:2)x y(cid:3)(cid:20)1 0(cid:21) = 0, 0(cid:21) = 0. 0 12 (14) (15) The above equations have associated pencils defined as follows: Q1(λ) = λ 0 1 0 0 0 λ 0 1  , Q2(λ) = λ 0 0 0 1 0 λ 0 1  . The above pencils are the same up to row and column permutations, so they have not just the same eigenvalues, but also the same Kronecker canonical form. However, the corresponding generalized Sylvester equations (14)–(15) can be rewritten, respectively, as x = 0, and x = y = 0. Then (14) has infinitely many solutions, while (15) has a unique solution. 3 Some corollaries The characterization given in Theorem 3 depends on spectral properties of the pencil Q(λ) in (3), which has twice the size of the coefficient matrices of Equation (1). With some additional effort, we can provide a characterization in terms of pencils with exactly the same size. Corollary 7. Let A ∈ Cp×m, B ∈ Cn×q, C ∈ Cp×n, and D ∈ Cm×q. Then the equation AXB + CX ⋆D = E has a unique solution, for any right-hand side E, if and only if one of the following situations holds: (a) p = m 6 n = q, A is invertible, the pencil P1(λ) := B⋆ − λD⋆A−1C is regular and (b) p = m > n = q, B is invertible, the pencil P2(λ) := A⋆ − λDB−1C ⋆ is regular and (c) p = n 6 m = q, C is invertible, the pencil P3(λ) := D⋆ − λB⋆C −1A is regular and – If ⋆ = ⊤, bΛ(P1) \ {1} is reciprocal free and m1(P1) 6 1. – If ⋆ = ∗, bΛ(P1) is ∗-reciprocal free. – If ⋆ = ⊤, bΛ(P2) \ {1} is reciprocal free and m1(P2) 6 1. – If ⋆ = ∗, bΛ(P2) is ∗-reciprocal free. – If ⋆ = ⊤, bΛ(P3) \ {1} is reciprocal free and m1(P3) 6 1. – If ⋆ = ∗, bΛ(P3) is ∗-reciprocal free. – If ⋆ = ⊤, bΛ(P4) \ {1} is reciprocal free and m1(P4) 6 1. 13 (d) p = n > m = q, D is invertible, the pencil P4(λ) := C ⋆ − λBD−1A⋆ is regular and – If ⋆ = ∗, bΛ(P4) is ∗-reciprocal free. Proof. Let us assume first that (1) has a unique solution, for any right-hand side E. Then [7, Th. 3] implies that at least one of the following situations holds: (C1) p = m < n = q and A is invertible, (C2) p = m > n = q and B is invertible, (C3) p = n < m = q and C is invertible, (C4) p = n > m = q and D is invertible, or (C5) p = m = n = q. Let us first assume that case (C1) holds. We can perform the following unimodular equivalence on Q(λ): (cid:20)I −λD⋆A−1 0 I (cid:21)(cid:20)λD⋆ B⋆ λC(cid:21) =(cid:20) 0 B⋆ − λ2D⋆A−1C λC A A (cid:21) . (16) Taking determinants in (16) we arrive at det(Q(λ)) = ± det(A) det(P1(λ2)). (17) This shows that P1 is regular. Note that D⋆A−1C has rank at most m < n, hence det(P1(λ)) has degree at most m and n − m dimension-induced infinite eigenvalues are present in Λ(P1). Similarly, Q(λ) has n− m dimension-induced infinite eigenvalues. The left- and right-hand sides of Equation (17) are nonzero [7, Th. 3] implies that part (a) in the statement holds. polynomials in λ with degree at most 2m; therefore we havebΛ(Q) =qbΛ(P1) := (cid:8)µ : µ2 ∈ bΛ(P1)(cid:9), including multiplicities and core infinite eigenvalues. Then If case (C4) holds, then we apply the ⋆ operator in (1) and the previous arguments to the new equation and its corresponding pencil C − λAD−⋆B⋆, namely(cid:0)P4(λ⋆)(cid:1)⋆ If case (C3) holds, then after introducing the change of variables Y = X ⋆, the roles of A, B and C, D are exchanged, so we apply the same arguments as in case (C1) to the corresponding pencil, P3(λ) and we get part (c). In case (C2), we apply the ⋆ operator in (1) and introduce the change of variables Y = X ⋆. Then we apply the same arguments as for case (C1) to the , new equation and its corresponding pencil A − λCB−⋆D⋆, namely (cid:0)P2(λ⋆)(cid:1)⋆ , and part (d) of the statement follows. Finally, if we are in case (C5), [6, Cor. 12] guarantees that at least one of A, B, C, D is invertible and thus at least one of (a)–(d) in the statement holds, and we are done. and part (b) of the statement follows. To prove the converse, let us assume that any of (a)–(d) in the statement holds. Then, reversing the previous arguments and using (4), we can conclude that at least one of the situations (i)–(iii) in the statement of [7, Th. 3] occurs, and [7, Th. 3] implies that (1) has a unique solution, for any right-hand side. As another consequence of Theorem 3 we get an extension of [3, Lemma 5.10] and [16, Lemma 8] for the ⋆-Sylvester equation AX + X ⋆D = E (see also [5, Th. 10, Th. 11]) to the case of rectangular coefficients, showing that when the coefficients are non-square the equation cannot be uniquely solvable. 14 Corollary 8. Let A ∈ Cn×m and D ∈ Cm×n. Then the equation AX + X ⋆D = E has a unique solution, for any right-hand side E, if and only if the matrix pencil P(λ) = A − λD⋆ is regular and: • If ⋆ = ⊤, Λ(P) \ {1} is reciprocal free and m1(P) 6 1. • If ⋆ = ∗, Λ(P) is ∗-reciprocal free. In particular, if the coefficients are non-square, then the equation cannot have a unique solution, for any right-hand side E. A more interesting situation involves the ⋆-Stein equation AXB + X ⋆ = E, for which we get the following result that generalizes Theorem 10 of [6] (see also the references in [6]). Corollary 9. Let A, B ∈ Cn×m. Then the equation AXB + X ⋆ = E has a unique solution, for any right-hand side E, if and only if the following conditions hold: • If ⋆ = ⊤, Λ(AB⊤) \ {1} is reciprocal free and m1(AB⊤) 6 1. • If ⋆ = ∗, Λ(AB∗) is ∗-reciprocal free. Proof. It is sufficient to observe that the condition in Corollary 7 (taking C = I, D = I) is equivalent to the condition stated on the spectrum of AB⋆, for each of the cases in Corollary 7. If m > n, we are in case (c), with P3 = I − λB⋆A. The eigenvalues of P3 are the reciprocals of the eigenvalues of B⋆A. Note that B⋆A has m − n dimension-induced zero eigenvalues and eΛ(B⋆A) = Λ(AB⋆) (this equality follows from [11, Theorem 1.3.20]). Hence the set Λ(AB⋆) is the reciprocal of bΛ(P3), so one of the two is (∗-)reciprocal-free if and only if the Similarly, if m < n, we can take P4(λ) = I − λBA⋆; then bΛ(P4) is the reciprocal ofeΛ(BA⋆), or, applying the ⋆ operator, the (∗-)reciprocal ofeΛ(AB⋆). Since a matrix never has ∞ as an eigenvalue, eΛ(AB⋆) is a (∗-)reciprocal-free set if and only if Λ(AB⋆) is so, regardless of the additional zero eigenvalues. other is, while the multiplicity of 1 is the same in both spectra. The cases with m = n can be proved in a similar way. Comparing Corollaries 8 and 9, we see an interesting difference between the unique solvability of the ⋆-Sylvester equation AX + X ⋆D = E and the ⋆-Stein equation AXB + X ⋆ = E. In the first case, the equation can not have a unique solution, for any right-hand side E, unless the coefficient matrices are square. However, the ⋆-Stein equation can have unique solution, for any right-hand side E, for rectangular coefficient matrices A, B ∈ Cn×m. As an elementary example, consider the matrices A =(cid:2) 1 2 (cid:3) , with n = 1, m = 2. The matrix AB⊤ = 3 satisfies the conditions in the statement of Corollary 9 (for both ⋆ = ⊤,∗), so AXB + X ⋆ = E has a unique solution, for any right-hand side E. Indeed, the equation, in the case ⋆ = ⊤, is x2 (cid:21)(cid:2) 1 (cid:2) 1 1 (cid:3)(cid:20) x1 2 (cid:3) +(cid:2) x1 x2 (cid:3) =(cid:2) e1 1 (cid:3) and B =(cid:2) 1 e2 (cid:3) , 15 which has a unique solution for any e1, e2 ∈ C. More precisely, this solution is given by x1 = 3e1 − e2 4 , x2 = e2 − e1 . 2 We leave to the reader to check that the system also has a unique solution, for any e1, e2 ∈ C, in the case ⋆ = ∗. 4 Explicit characterization for the generalized Sylvester equation In this section, we provide an explicit solution of problems SR, OR, and UR for Equation (2). These characterizations follow from the results and lemmas in [13], but we state them explicitly in Theorem 10. Unlike the characterization given in Theorem 3 of UR for Equation (1), the characterizations for Equation (2) depend on further constraints on the Kronecker canonical form (KCF) of the matrix pencils A − λC and D⊤ − λB⊤, and not just on their spectrum. Though the KCF is a standard canonical form that can be found in most of the basic references on matrix pencils, we refer the reader to [5, Th. 2], since we follow the notation in that paper. In particular, J(α) denotes a Jordan block associated with the eigenvalue α, including α = ∞ (which is denoted by N in [5]), Lε denotes a right singular block of size ε× (ε + 1), and L⊤ η denotes a left singular block of size (η + 1) × η. We denote by Z+ the set of positive integers. Theorem 10. Let A, C ∈ Cp×m and B, D ∈ Cn×q. SR Equation (2) has at least one solution, for any right-hand side E, if and only if the following two conditions are satisfied: eigenvalues, and • the (possibly singular) pencils A−λC and D⊤−λB⊤ have no common • if the KCF of either A− λC or D⊤ − λB⊤ contains a block L⊤ η , then the KCF of the other pencil is a direct sum of blocks Lεi with εi 6 η. OR Equation (2) has at most one solution, for any right-hand side E, if and only if the following two conditions are satisfied: eigenvalues, and • the (possibly singular) pencils A−λC and D⊤−λB⊤ have no common • if the KCF of either A − λC or D⊤ − λB⊤ contains a block Lε, then ηi with ηi 6 ε. the KCF of the other pencil is a direct sum of blocks L⊤ UR Equation (2) has exactly one solution, for any right-hand side E, if and only if one of the following situations hold: • the pencils A − λC and D⊤ − λB⊤ are regular and have no common eigenvalues, or 16 • there is some s ∈ Z+ such that the KCF of either A−λC or B⊤−λD⊤ is a direct sum of blocks Ls, and the KCF of the other pencil is a direct sum of blocks L⊤ s . Proof. Let P1(A − λC)Q1 = bA − λbC and P ⊤ 2 = bD⊤ − λbB⊤ be the KCFs of A− λC and D⊤ − λB⊤, respectively. Equation (2) is equivalent to bAbXbB − bCbXbD = bE, with bX = Q−1 bE = P1EP2. (18) Partitioning the matrices conformably with the (possibly rectangular) blocks in the KCFs, we get 2 (D⊤ − λB⊤)Q⊤ 1 XQ−1 2 ,  bA11  bC11 − bA22 bC22 . . . . . . ...   bX11 bX21 bApp bXp1   bX11 bX21 bXp1 bCpp ... . . . . . . . . . . . . . . . . . . . . . . . . ... bX12 bX22 bXp2 bX12 bX22 bXp2 ... ... ...   bX1q bB11 bX2q bXpq   bX1q bD11 bX2q bXpq = bE11 bE21 bEp1 ... . . . . . . . . . . . . . . . . . . bBqq   bDqq  bE1q bE2q bEpq ... , bB22 bD22 bE12 bE22 bEp2 ... and (18) is equivalent to the system of pq independent equations i = 1, 2, . . . , p, j = 1, 2, . . . , q. (19) bAiibXijbBjj − bCiibXijbDjj = bEij , In particular, the existence (resp. uniqueness) of solution X of (2), for any right- hand side E, is equivalent to the simultaneous existence (resp. uniqueness) of the 1, 2, . . . , q. solution bXij of (19), for any right-hand side bEij , and for any i = 1, 2, . . . , p, j = Solvability and uniqueness conditions for the systems (19), where bAii − λbCii and bD⊤ jj −λbB⊤ jj are blocks from the KCF, are provided in [13]. We recall them in Table 2. One can check, using this table, that each equation of the system (19) has at least one solution if and only if the SR conditions in the statement of the theorem hold; similarly, each equation has at most one solution if and only if the OR conditions hold, and it has exactly one solution if and only if the UR conditions hold. Theorem 10 shows that the characterization of SR, OR, and UR for the generalized Sylvester equation (2) depends on the KCF of the pencils A−λC and 17 A − λC J(α) J(α) L⊤ η Lε L⊤ η1 Lε1 J(α) J(α) L⊤ η Lε At least one solution? At most one solution? D⊤ − λB⊤ J(α) No J(β) (β 6= α) Yes Lε L⊤ η L⊤ η2 Lε2 Lε L⊤ η J(α) J(α) Only when η > ε Only when η > ε No Yes Yes No No Yes No Yes Only when η 6 ε Only when η 6 ε Yes No No Yes Yes No Table 2: Summary of the results on existence and uniqueness of solutions for the equation AXB − CXD = E when A − λC and D are Kronecker blocks, obtained from the lemmas in [13, Sec. 4 and 5]. The Jordan blocks include α = ∞. − λB ⊤ ⊤ D⊤−λB⊤. It is natural to expect a characterization of UR for the generalized ⋆- Sylvester equation (1) depending also on the KCF of the pencil Q(λ) in Theorem 3. However, the example provided in Section 2.3 shows that this is not the case. This example presents two different equations with different behavior (Equation (15) has a unique solution, for any right-hand side, whereas Equation (14) has not). However, the associated pencils, Q2(λ) and Q1(λ), respectively, have the same KCF. 4.1 An alternative characterization of UR The criterion for unique solvability for each right-hand side (UR) for the gen- eralized Sylvester equation (2) gives a peculiar restriction that can be expressed in terms of the ratios between the two dimensions of the involved pencils. Corollary 11. Let A, C ∈ Cp×m and B, D ∈ Cn×q. If the generalized Sylvester equation (2) has a unique solution for any right-hand side, then p/m = n/q = d, for some d ∈ {1} ∪(cid:26) s s + 1 : s ∈ Z+(cid:27) ∪(cid:26) s + 1 s : s ∈ Z+(cid:27) . Proof. By Theorem 10, there are two possibilities in order for (2) to have a unique solution, for any right-hand side: either the pencils A−λC and D⊤−λB⊤ are regular with no common eigenvalues or one of the two pencils is a direct sum of blocks Ls and the other is a direct sum of blocks L⊤ s . In the former case, A and B must be square (p = m and n = q) and then d = 1 in the statement. In the latter case, if A− λC is a direct sum of k right singular blocks Ls and D⊤− λB⊤ is a direct sum of ℓ left singular blocks L⊤ s , then A has size (ks) × (k(s + 1)), while B has size (ℓs)× (ℓ(s + 1)), and then d = p/m = n/q = s/(s + 1), while if 18 A − λC is a direct sum of k left singular blocks L⊤ sum of ℓ right singular blocks L⊤ s , we have d = (s + 1)/s. s and D⊤ − λB⊤ is a direct The necessary condition given in Corollary 11 is not sufficient. In order to give a sufficient condition, we need the following result. Lemma 12. Let A, C ∈ Rp×m, and ε ∈ Z+. Then, the KCF of the pencil A − λC consists only of blocks of the form Lε if and only if the p(ε + 1) × mε matrix Mε(A, C) := A C A  C . . . . . . A C  is square and invertible. Similarly, the KCF of A − λC is composed uniquely of blocks of the form L⊤ η if and only if Mη(A⊤, C ⊤) is square and invertible. Proof. Let us first suppose that Mε(A, C) is square and invertible. Let p1 × m1, p2 × m2, . . . , pk × mk be the sizes of the blocks in the KCF of A − λC. It follows from the arguments in [9, Section XII.3] that Mε(A, C) has a nontrivial kernel if and only if there is a polynomial vector x(λ) of degree strictly smaller than ε such that (A − λC)x(λ) = 0, or, equivalently, if and only if the KCF of A − λC contains a block Lε2 with ε2 < ε. Hence, if Mε(A, C) is invertible, then the KCF of A − λC contains only Jordan blocks, together with blocks L⊤ η , and blocks Lε2 with ε2 > ε. For each of these blocks, one can check that mi/pi 6 (ε + 1)/ε, for i = 1, . . . , k, and the equality holds only for blocks of type Lε. In particular, we have p(ε + 1) = pi(ε + 1) > kXi=1 miε = mε. kXi=1 Since Mε(A, C) is square, equality must hold for all i = 1, . . . , k, which means that all blocks are of type Lε. Now let us prove the other direction: suppose that the KCF of A − λC consists only of blocks Lε. Then, p = kε and m = k(ε + 1), so p(ε + 1) = mε, and Mε(A, C) is square. Moreover, again by the same reasonings in [9, Section XII.3] as above, Mε(A, C) has trivial kernel, so it is invertible. The second claim in the statement follows by applying the first statement to A⊤ − λC ⊤. Using Lemma 12 we can give an alternative version of the UR characteri- zation in Theorem 10 that does not involve the KCF explicitly. Theorem 13. Let A, C ∈ Cp×m and B, D ∈ Cn×q. Equation (2) has exactly one solution, for any right-hand side E, if and only if one of the following situations hold: 19 • p = m, q = n, the pencils A − λC and D⊤ − λB⊤ are regular and have no common eigenvalues, or • p < m, n < q, s = p/(m − p) = n/(q − n) is a positive integer, and the square matrices Ms(A, C) and Ms(B, D) are both invertible, or • p > m, n > q, s = m/(p − m) = q/(n − q) is a positive integer, and the square matrices Ms(A⊤, C ⊤) and Ms(B⊤, D⊤) are both invertible. Proof. The proof is an immediate consequence of Theorem 10, part UR, and Lemma 12, just taking into account that if the KCF of an m× n pencil (respec- tively, the KCF of an n × m pencil) is a direct sum of k blocks Ls (resp., L⊤ s ), for some fixed s, then, as we have seen in the proof of Lemma 12, it must be p = ks, m = k(s + 1) (resp., p = k(s + 1), m = ks), which implies m > p and s = p/(m − p) (resp., m < p and s = m/(p − m)). 5 Conclusions and open problems We have provided necessary and sufficient conditions for the generalized ⋆- Sylvester equation (1) to have a unique solution for any right-hand side E (UR). In particular, the coefficient matrix of the associated linear system must be square, which is equivalent to the condition mn = pq, and the problem becomes equivalent to characterizing the uniqueness of solution of the homo- geneous equation (6). The characterization that we have obtained extends the recent one in [6] for the case of square coefficients. We have also reviewed the solution of problems SR (solvability for any right-hand side), OR (at most one solution for any right-hand side), and UR (unique solvability for any right-hand side) for the generalized Sylvester equation (2). It is interesting to compare the conditions for unique solvability (UR) for the two equations (1) and (2), given in Theorems 3 and 10 since, in the case of rectangular coefficients, there are more significant differences than those in the case of square coefficients. For the generalized Sylvester equation (2), the only additional case with unique solution is when the KCF of the associated pencils A− λC and D⊤− λB⊤ contains only certain singular blocks; for the generalized ⋆-Sylvester equation, the spectral properties and Kronecker invariants are not sufficient to determine the answer, and it is necessary to check the invertibility of one of the coefficients. To our knowledge, small-pencil characterizations for questions US, SR, and OR for the generalized ⋆-Sylvester equation are still not present in the literature and arise as a natural open problem to approach in the future. Acknowledgments. We wish to thank Emre Mengi for pointing out refer- ence [15] and, as a consequence, references [13, 14] as well. We also thank two anonymous referees for valuable comments that allowed to improve the original version of the paper, as well as one of the referees for pointing out reference [18]. 20 References [1] M. Baumann and U. Helmke. Singular value decomposition of time-varying matrices. Future Generation Computer Systems, 19(3):353 – 361, 2003. Special Issue on Geometric Numerical Algorithms. [2] H. W. Braden. The equations AT X ± X T A = B. SIAM J. Matrix Anal. Appl., 20(2):295–302, 1999. [3] R. Byers and D. Kressner. Structured condition numbers for invariant subspaces. SIAM J. Matrix Anal. Appl., 28(2):326–347, 2006. [4] K.-W. E. Chu. The solution of the matrix equations AXB − CXD = E and (Y A − DZ, Y C − BZ) = (E, F ). Linear Algebra Appl., 93:93–105, 1987. [5] F. De Ter´an, F. M. Dopico, N. Guillery, D. Montealegre, and N. Z. Reyes. The solution of the equation AX + X ⋆B = 0. Linear Algebra Appl., 438:2817–2860, 2011. [6] F. De Ter´an and B. Iannazzo. Uniqueness of solution of a generalized ⋆-Sylvester matrix equation. Linear Algebra Appl., 493:323–335, 2016. [7] F. De Ter´an, B. Iannazzo, F. Poloni, for generalized and L. Robol. Solv- Sylvester-type In press. Available and ability equations. on https://doi.org/10.1016/j.laa.2017.07.010. uniqueness Linear Algebra Appl., criteria 2017. [8] A. Dmytryshyn and B. Kagstrom. Coupled sylvester-type matrix equations and block diagonalization. SIAM J. Matrix Anal. Appl., 36(2):580–593, 2016. [9] F. R. Gantmacher. The Theory of Matrices. Vols. 1, 2. Chelsea Publishing Co., New York, 1959. [10] R. Granat, B. Kagstrom, and D. Kressner. Matlab tools for solving periodic eigenvalue problems. IFAC Proceedings Volumes, 40(14):169–174, 2007. [11] R. A. Horn and C. R. Johnson. Matrix Analysis. Cambridge University Press, 1985. [12] R. A. Horn and C. R. Johnson. Topics in Matrix Analysis. Cambridge University Press, 1994. [13] T. Kosir. The matrix equation AXDT − BXC T = E. Technical Report (Research paper 737), Dept. of Mathematics and Statistics, Univ. of Cal- gary, 1992. [14] T. Kosir. Kronecker bases for linear matrix equations, with application to two-parameter eigenvalue problems. Linear Algebra Appl., 249:259–288, 1996. 21 [15] D. Kressner, E. Mengi, I. Naki´c, and N. Truhar. Generalized eigenvalue problems with specified eigenvalues. IMA J. Numer. Anal., 34:480–501, 2014. [16] D. Kressner, C. Schroder, and D. S. Watkins. Implicit QR algorithms for palindromic and even eigenvalue problems. Numer. Algorithms, 51(2):209– 238, 2009. [17] S. K. Mitra. The matrix equation AXB + CXD = E. SIAM J. Appl. Math., 32(4):823–825, 1977. [18] P. R´ozsa. Lineare Matrizengleichungen und Kroneckersche Produkte. Z. Angew. Math. Mech., 58(7):T395–T397, 1978. [19] V. V. Sergeichuk. Computation of canonical matrices for chains and cycles of linear mappings. Linear Algebra Appl., 376:235–263, 2004. [20] V. Simoncini. Computational methods for linear matrix equations. SIAM Rev., 58(3):377–441, 2016. [21] Y. Yuan and H. Dai. The direct updating of damping and gyroscopic matrices. J. Comput. Appl. Math., 231(1):255–261, 2009. 22
1104.3774	1	1104	2011-04-19T15:21:17	Complements of Intervals and Prefrattini Subalgebras of Solvable Lie Algebras	[ "math.RA", "math.GR" ]	In this paper we study a Lie-theoretic analogue of a generalisation of the prefrattini subgroups introduced by W. Gasch\"utz. The approach follows that of P. Hauck and H. Kurtzweil for groups, by first considering complements in subalgebra intervals. Conjugacy of these subalgebras is established for a large class of solvable lie algebras.	math.RA	math	COMPLEMENTS OF INTERVALS AND PREFRATTINI SUBALGEBRAS OF SOLVABLE LIE ALGEBRAS DAVID A. TOWERS Department of Mathematics and Statistics Lancaster University Lancaster LA1 4YF England [email protected] Abstract In this paper we study a Lie-theoretic analogue of a generalisa- tion of the prefrattini subgroups introduced by W. Gaschutz. The ap- proach follows that of P. Hauck and H. Kurtzweil for groups, by first considering complements in subalgebra intervals. Conjugacy of these subalgebras is established for a large class of solvable lie algebras. Mathematics Subject Classification 2000: 17B05, 17B20, 17B30, 17B50. Key Words and Phrases: Lie algebras, complemented, solvable, Frat- tini ideal, prefrattini subalgebra, residual. 1 Complements of subalgebra intervals Throughout, L will denote a solvable Lie algebra over a field F . For a subalgebra U of L we denote by [U : L] the set of all subalgebras S of L with U ⊆ S ⊆ L. We say that [U : L] is complemented if, for any S ∈ [U : L] there is a T ∈ [U : L] such that S ∩ T = U and < S, T >= L. Our objective is to study the set Ω(U, L) = {S ∈ [U : L] : [S : L] is complemented}; 1 in particular, to show that, for a large class of solvable Lie algebras L, the minimal elements of this set, Ω(U, L)min, are conjugate in L. The develop- ment initially follows closely that of [3]. We denote by [U : L]max the set of maximal subalgebras in [U : L]; that is, the set of maximal subalgebras of L containing U . If L = A + B where A and B are subalgebras of L and A ∩ B = 0 we will write L = A ⊕ B. Lemma 1.1 If S ∈ Ω(U, L), S 6= L then S = T{M : M ∈ [S : L]max}. Proof. Put T = T{M : M ∈ [S : L]max}. Then [S : L] is complemented, since S ∈ Ω(U, L), and so T has a complement C in [S : L]. If C 6= L then C ⊆ M for some M ∈ [S : L]max. But then < T, C >= M , contradicting the fact that C is a complement of T in [S : L]. Hence C = L and S = T ∩ C = T ∩ L = T , as required. (cid:3) The Frattini subalgebra of L, φ(L), is the intersection of the maximal subalgebras of L. When L is solvable this is always an ideal of L, by [1, Lemma 3.4]. Extending this notion slightly we put φ(S, L) = T{M : M ∈ [S : L]max}; clearly, φ(0, L) = φ(L). The above lemma shows that φ(U, L) ⊆ S for all S ∈ Ω(U, L). Lemma 1.2 If I is an ideal of L and S ∈ Ω(U, L), then S + I ∈ Ω(U, L). Proof. Let B ∈ [S + I : L] ⊆ [S : L]. Since S ∈ Ω(U, L), B has a complement D in [S : L]; that is B ∩ D = S and < B, D >= L. Put C = D+I. Then < B, C >= L and B ∩C = B ∩(D+I) = B ∩D+I = S +I, whence C is a complement for B in [S + I : L] and S + I ∈ Ω(U, L). (cid:3) Lemma 1.3 Let A be a minimal ideal of L and let M be a complement of A in L containing U . Then Ω(U, M ) = {S ∈ Ω(U, L) : S ⊆ M }. In particular Ω(U, M )min = {S ∈ Ω(U, L)min : S ⊆ M }. Proof. Note that since L is solvable, M is a maximal subalgebra of L and L = A ⊕ M . Suppose first that S ∈ Ω(U, L) with S ⊆ M . Then S + A ∈ Ω(U, L) by Lemma 1.2. The interval [S : M ] is lattice isomorphic to [S + I : L] and so is complemented. Hence S ∈ Ω(U, M ). Conversely, let S ∈ Ω(U, M ). Then [S : M ] is complemented. We need to show that S ∈ Ω(U, L); that is, that [S : L] is complemented. Let B ∈ [S : L]. Then B ∩ M ∈ [S : M ], so there is a subalgebra D ∈ [S : M ] such that < B ∩ M, D >= M and B ∩ D = B ∩ M ∩ D = S. 2 If B 6⊆ M then M is a proper subalgebra of < B, D >. But M is a maximal subalgebra of L, and so < B, D >= L and D is a complement of B in [S : L]. Hence [S : L] is complemented. If B ⊆ M , put C = D + A. Then L = A ⊕ M ⊆ < B, A > + < B, D > ⊆ < B, D + A > = < B, C >, so < B, D + A >= L. Also B ∩ C = B ∩ (D + A) = B ∩ M ∩ (D + A) = B ∩ (D + M ∩ A) = B ∩ D = S, yielding that C is a complement of B in [S : L] and [S : L] is complemented. (cid:3) Lemma 1.4 Let A be a minimal ideal of L and let S ∈ Ω(U, L)min with A 6⊆ S. Then there is an M ∈ [S : L]max such that A 6⊆ M . Proof. This follows easily from Lemma 1.1. (cid:3) Lemma 1.5 Let A be a minimal ideal of L. Then the following are equiv- alent: (i) A 6⊆ S for some S ∈ Ω(U, L)min; (ii) A 6⊆ M for some M ∈ [U : L]max; and (iii) for every S ∈ Ω(U, L)min there is a complement of A in L containing S. Proof. (i) ⇒ (ii): This follows from Lemma 1.4. (ii) ⇒ (iii): Suppose that A 6⊆ M for some M ∈ [U : L]max. Then L = A ⊕ M . Let S ∈ Ω(U, L)min. Suppose first that A ⊆ S. Then S = A ⊕ M ∩ S and M ∩ S ∼= S/A, so the interval [S : L] is lattice isomorphic to [M ∩ S : M ]. It follows that M ∩ S ∈ Ω(U, M ). But Lemma 1.3 now gives that M ∩ S ∈ Ω(U, L), contradicting the minimality of S. Hence A 6⊆ S and Lemma 1.4 gives a complement of A containing S. (iii) ⇒ (i): This is trivial. (cid:3) Lemma 1.6 If A is an ideal of L and S ∈ Ω(U, L)min then S + A ∈ Ω(U + A, L)min. 3 Proof. It suffices to show that (S + A)/A ∈ Ω((U + A)/A, L/A)min and so we may suppose that A is a minimal ideal of L. The result is clear if A ⊆ S, since then U + A ⊆ S. So suppose that A 6⊆ S. Then there is a complement M of A in L containing S, by Lemma 1.5, and L = A⊕M . Moreover, S+A ∈ Ω(U +A, L). Choose C ∈ Ω(U +A, L)min such that C ⊆ S + A. Then U ⊆ M ∩ C ⊆ S ⊆ M and the interval [M ∩C : M ] is lattice isomorphic to [C : L]. It follows that M ∩C ∈ Ω(U, M ) and so M ∩ C ∈ Ω(U, L), by Lemma 1.3. But S ∈ Ω(U, L)min, which yields that M ∩ C = S; that is, C = S + A. (cid:3) At this point the theory starts to diverge from that for groups. We say that L is completely solvable if L2 is nilpotent. For these algebras Ω(U, L)min takes on a particularly simple form. Theorem 1.7 Let L be completely solvable and let U be a subalgebra of L. Then Ω(U, L)min = {φ(U, L)}. In particular, if U = 0 then Ω(U, L)min = {φ(L)}. Proof. Put B = Ω(U, L)min, C = φ(U, L). Then φ(U, L) ⊆ B and so C ⊆ B, by Lemma 1.1. We now use induction on the dimension of L. Suppose first that there is a minimal ideal A of L with A ⊆ C. Then B/A ∈ Ω((U + A)/A, L/A)min, by Lemma 1.6, and so B/A = φ((U + A)/A, L/A), by the inductive hypothesis. From this it is clear that B = C. So suppose now that no such minimal ideal exists. Then L is φ-free and so L is complemented, by [4, Theorem 1]. Thus there is a subalgebra V such that hC, V i = L and C ∩ V = 0. It follows that hC, U + V i = L and C ∩ (U + V ) = U + C ∩ V = U , whence C ∈ [U : L] and [C : L] is complemented. Thus C ∈ Ω(U, L) and the minimality of B yields that B = C. (cid:3) If L is not completely solvable then Ω(U, L)min can contain more than one element as we shall see in the next section. However, we do have a con- jugacy result in some cases. First we need to consider inner automorphisms of L. Let x ∈ L and let ad x be the corresponding inner derivation of L. If F has characteristic zero, suppose that (ad x)n = 0 for some n; if F has characteristic p, suppose that x ∈ I where I is a nilpotent ideal of L of class less than p. Put exp(ad x) = Xr=0 Then exp(ad x) is an automorphism of L. 4 ∞ 1 r! (ad x)r. [x, B] = 0}. We define the nilpotent residual to be L∞ = T∞ If B is a subalgebra of L, the centraliser of B in L is CL(B) = {x ∈ L : i=1 Li, where Li are the terms of the lower central series for L. Then we have conjugacy for the following metanilpotent Lie algebras. Theorem 1.8 Suppose that L is a solvable Lie algebra over a field F of characteristic p, and suppose further that L∞ has nilpotency class less than p. Let U be a subalgebra of L. Then the elements of Ω(U, L)min are conjugate under I(L : L∞). Proof. We use induction on the dimension of L. It is clearly true if L has dimension one, so suppose it holds for such algebras with dimension smaller than that of L. We can assume that L∞ 6= 0. Let S1, S2 ∈ Ω(U, L)min and let A be a minimal ideal of L with A ⊆ L∞. Then (S1 + A)/A, (S2 + A)/A ∈ Ω((U + A)/A, L/A)min, by Lemma 1.6, and so (S1 + A)/A and (S2 + A)/A are conjugate under I(L/A : L∞/A), by the inductive hypothesis. If A ⊆ S1 then A ⊆ S2, by Lemma 1.5, and there is an x ∈ L∞ such that S1 exp(ad x) = S2; that is, S1 and S2 are conjugate under I(L : L∞). So suppose that A 6⊆ S1. Then there are complements M1 and M2 of A in L with S1 ⊆ M1 and S2 ⊆ M2, by Lemma 1.5. Put C = CM1(A), which is an ideal of L. If C = 0 then CL(A) = A and there is a ∈ A such that M2 exp(ad a) = M1, by [2, Theorem 1.1], whence S2 exp(ad a) ⊆ M2 exp(ad a) = M1. If C 6= 0, then (S1 + C)/C and (S2 + C)/C are conjugate under I(L/C : (L∞ + C)/C), by the inductive hypothesis. It follows that there is an x ∈ L∞ such that S2 exp(ad x + C) ⊆ S1 + C exp(ad a) ⊆ M1, which gives S2 exp(ad x) ⊆ M1. Now L = A ⊕ M1, so L∞ ⊆ A ⊕ M ∞ 1 . Moreover, [A, L∞] = 0 since L∞ is nilpotent, so M ∞ is an ideal of L. Put x = a + b, 1 where a ∈ A, b ∈ M ∞ 1 . Then, for each s2 ∈ S2, we have s2 + s2 ad x+ . . . + s2 (ad x)n ∈ M1, which gives s2 + s2 ad a ∈ M1. Thus, again we have that S2 exp(ad a) ⊆ M1 for some a ∈ A. So S1, S2 exp(ad a) ⊆ M1 for some a ∈ A. Now U ⊆ S1 ⊆ M1 and U exp(ad a) ⊆ S2 exp(ad a) ⊆ M1, so, for each u ∈ U , u + [a, u] ∈ M1 which gives [a, u] ∈ A ∩ M1 = 0; that is, a ∈ CL(U ) and U exp(ad a) = U . Thus S2 exp(ad a) ∈ Ω(U exp(ad a), L)min = Ω(U, L)min. But now Lemma 1.3 yields that S1, S2 exp(ad a) ∈ Ω(U, M1)min and the required conjugacy of S1 and S2 follows from the inductive hypothesis. (cid:3) 5 2 U-prefrattini subalgebras Let 0 = A0 ⊂ A1 ⊂ . . . ⊂ An = L (1) be a fixed chief series for L. We say that Ai/Ai−1 is a Frattini chief factor if Ai/Ai−1 ⊆ φ(L/Ai−1); it is complemented if there is a maximal subalgebra M of L such that L = Ai + M and Ai ∩ M = Ai−1. When L is solvable it is easy to see that a chief factor is Frattini if and only if it is not complemented. This can be generalised as follows. The factor algebra Ai/Ai−1 is called a U -Frattini chief factor if Ai ⊆ φ(U + Ai−1, L) or if U + Ai−1 = L. If Ai/Ai−1 is not a U -Frattini chief factor there is an M ∈ [U + Ai−1 : L]max for which Ai 6⊆ M ; that is, M is a complement of the chief factor Ai/Ai−1. We have a sharpened form of the Jordan-Holder Theorem in which U -Frattini chief factors correspond. First we need a lemma. Lemma 2.1 Let A1, A2 be distinct minimal ideals of the solvable Lie algebra L. Then there is a bijection θ : {A1, (A1 + A2)/A1} → {A2, (A1 + A2)/A2} such that corresponding chief factors have the same dimension and U -Frattini chief factors correspond to one another. Proof. Clearly we can assume that U 6= L. Put A = A1 ⊕ A2. Suppose first that A1 is a U -Frattini chief factor. Then A1 ⊆ φ(U, L). Thus A ⊆ φ(U + A2, L) and A/A2 is a U -Frattini chief factor. If A/A1 is also a U - Frattini chief factor, then A ⊆ φ(U + A1, L), which yields that A ⊆ φ(U, L), and all four factors are U -Frattini. In this case we can choose θ so that θ(A1) = A/A2 and θ(A/A1) = A2. If A/A1 is not a U -Frattini chief factor, then nor is A2, by the same argument as above, and so the same choice of θ suffices; likewise if none of the factors are U -Frattini chief factors. The remaining case is where A1 and A2 are not U -Frattini chief factors but A/A2 is. Then A1 6⊆ φ(U, L), A2 6⊆ φ(U, L) and either A ⊆ φ(U + A2, L) or U + A2 = L. Thus there exists M ∈ [U, L]max such that A1 6⊆ M , giving L = A1 ⊕ M . Put A3 = M ∩ A. Then A3 ⊕ A1 = M ∩ A ⊕ A1 = (M + A1) ∩ A = A, and so A3 ∼= A/A1 ∼= A2. If A3 = A2 then U + A2 ⊆ M which gives A ⊆ M : a contradiction. Hence A3 6= A2, A = A3 ⊕ A2 and 6 A3 ∼= A/A2 ∼= A1. dimension. It follows that all of the chief factors have the same If U + A1 = L, then A/A1 is a U -Frattini chief factor, so we can choose If U + A1 6= L, let N ∈ θ so that θ(A1) = A2 and θ(A/A1) = A/A2. [U + A1, L]max. If A2 6⊆ N then L = A2 ⊕ N and N ∩ A = A1. But A ⊆ φ(U +A2, L) implies that A ⊆ φ(U, L)+A2, whence A+A2+φ(U, L)∩A. It follows that φ(U, L) ∩ A ⊆ N ∩ A = A1, giving φ(U, L) ∩ A = A1. But now A1 ⊆ φ(U, L): a contradiction. We must, therefore, have A2 ⊆ N and so A ⊆ N . Thus A ⊆ φ(U + A1, L); that is, A/A1 is a U -Frattini chief factor. In this case we can again choose θ so that θ(A1) = A2 and θ(A/A1) = A/A2. (cid:3) Theorem 2.2 Let 0 < A1 < . . . < An = L 0 < B1 < . . . < Bn = L (1) (2) be chief series for the solvable Lie algebra L. Then there is a bijection between the chief factors of these two series such that corresponding factors have the same dimension and such that the U -Frattini chief factors in the two series correspond. Proof. These two series have the same length by a version of the Jordan Holder Theorem. We induction on n. The result is clearly true if n = 1. So let n > 1 and suppose that the result holds for all solvable Lie algebras with chief series of length ≤ n − 1. If A1 = B1, then applying the inductive hypothesis to L/A1 gives a suitable bijection between the factors above A1, and then we can map A1 to B1 and we have the result. So suppose that A1 and B1 are distinct and put A = A1 ⊕ B1. Then A/A1 and A/B1 are chief factors of L and there are chief series of the form 0 < A1 < A < C3 < . . . < Cn = L 0 < B1 < A < C3 < . . . < Cn = L (3) (4) Define an equivalence relation on the chief series of L by saying that two such series are equivalent if there is a bijection between their chief factors satisfying the requirements of the theorem. Since series (1) and (3) have a minimal ideal in common, they are equivalent. Similarly, sereis (2) and (4) are equivalent. Moreover, since series (3) and (4) coincide above A they are also equivalent, by Lemma 2.1. Hence the series (1) and (2) are equivalent, as required. (cid:3) 7 We define the set I by i ∈ I if and only if Ai/Ai−1 is not a U -Frattini chief factor of L. For each i ∈ I put Mi = {M ∈ [U + Ai−1, L]max : Ai 6⊆ M }. Then B is a U -prefrattini subalgebra of L if B = \i∈I Mi for some Mi ∈ Mi. If U = 0 we will refer to B simply as a prefrattini subalgebra of L. The subalgebra B avoids Ai/Ai−1 if B ∩ Ai = B ∩ Ai−1; likewise, B covers Ai/Ai−1 if B + Ai = B + Ai−1. Then we have the following important property of U -prefrattini subalgebras of L. Lemma 2.3 If B is a U -prefrattini subalgebra of L then it covers all U - Frattini chief factors of L in (1) and avoids the rest. Proof. Let B be a U -prefrattini subalgebra of L and let Ai/Ai−1 be a chief factor of L. If it is a U -Frattini chief factor then either Ai ⊆ φ(U + Ai−1, L) or else U +Ai−1 = L. In the former case, every maximal subalgebra of L that contains U + Ai−1 also contains Ai, and so Ai ⊆ B. In either case, therefore, B covers Ai/Ai−1. If it is not a U -Frattini chief factor we have B ⊆ Mi where L = Ai + Mi and Ai ∩ Mi = Ai−1. Hence B ∩ Ai = B ∩ Mi ∩ Ai = B ∩ Ai − 1 ⊆ B ∩ Ai, and so B avoids Ai/Ai−1. (cid:3) The next four results are dedicated to showing how the U -prefrattini subalgebras relate to the material in the previous section. The first lemma is helpful when trying to calculate U -prefrattini subalgebras. Lemma 2.4 Let B be a U -prefrattini subalgebra of L. Then dim B = Xi /∈I (dim Ai − dim Ai−1); in particular, all U -prefrattini subalgebras of L have the same dimension. Proof. We use induction on dim L. The result is clear if L is abelian, so suppose it holds for Lie algebras of smaller dimension than L. It is easy to check that (B + A1)/A1 is a ((U + A1)/A1)-prefrattini subalgebra of L/A1 and so dim(cid:18) B + A1 A1 (cid:19) = Xi∈I,i6=1 (dim Ai − dim Ai−1), 8 by the inductive hypothesis. If A1/A0 is a U -Frattini chief factor of L, then B covers A1/A0, whence B = B + A1 and dim B = dim A1 + dim(cid:18)B + A1 A1 (cid:19) = Xi∈I (dim Ai − dim Ai−1). If A1/A0 is not a U -Frattini chief factor of L, then B avoids A1/A0, whence B ∩ A1 = 0 and dim B = dim(cid:18) B + A1 A1 (cid:19) = Xi∈I (dim Ai − dim Ai−1). (cid:3) Let Π(U, L) be the set of U -prefrattini subalgebras of L. Lemma 2.5 Π(U, L) ⊆ Ω(U, L). Proof. (i) We use induction on dim L. The result is clear if L is abelian, so suppose it holds for Lie algebras of dimension less than that of L. Let B ∈ Π(U, L). Then B + A1 A1 ∈ Π(cid:18) U + A1 A1 , L A1(cid:19) ⊆ Ω(cid:18) U + A1 A1 , L A1(cid:19) , whence B + A1 ∈ Ω(U, L). If A1 ⊆ B we have B ∈ Ω(U, L). So suppose that A1 6⊆ B. Then B does not cover A1/A0, so A1/A0 is not a U -Frattini chief factor of L. It follows that 1 ∈ I, and so there is a maximal subalgebra M of L with B ⊆ M and A1 6⊆ M . But now L = A1 ⊕ M and the intervals [B + A1 : L] and [B : M ] are lattice isomorphic, which yields that [B : M ] is complemented. It follows from Lemma 1.3 that B ∈ Ω(U, L) again. (cid:3) Lemma 2.6 Ω(U, L)min ⊆ Π(U, L). Proof. Let B ∈ Ω(U, L)min and let Ai/Ai−1 be a chief factor of L. By Lemma 1.6, (cid:18) B + Ai−1 Ai−1 (cid:19) ∈ Ω(cid:18) U + Ai−1 Ai−1 , L Ai−1(cid:19)min . If We now apply Lemma 1.5 to the minimal ideal Ai/Ai−1 of L/Ai−1. Ai/Ai−1 is a U -Frattini chief factor then it doesn't have a complement in L/Ai−1 and Lemma 1.5 gives that Ai ⊆ B +Ai−1, whence Ai +B = Ai−1 +B and B covers Ai/Ai−1. 9 If Ai/Ai−1 is not a U -Frattini chief factor then it has a complement Mi/Ai−1 in L/Ai−1 and Lemma 1.5 gives that it has such a complement containing (B + Ai−1)/Ai−1; that is L = Mi + Ai, Mi ∩ Ai = Ai−1 and B + Ai−1 ⊆ Mi. But now B ∩ Ai ⊆ B ∩ Ai + Ai−1 = (B + Ai−1) ∩ Ai ⊆ Mi ∩ Ai = Ai−1. It follows that B ∩ Ai = B ∩ Ai−1 and B avoids Ai/Ai−1. Clearly Mi ∈ Mi and B ⊆ C = Ti∈I Mi ∈ Π(U, L). But B covers or avoids the same chief factors of (1) as C, so the proof of Lemma 2.4 shows that dim B = dim C. It follows that B = C ∈ Π(U, L). (cid:3) Putting the previous three Lemmas together yields the following result. Theorem 2.7 Ω(U, L)min = Π(U, L). Notice that, in particular, the above result shows that the definition of U -prefrattini subalgebras does not depend on the choice of chief series. Corollary 2.8 If A is an ideal of L and S ∈ Π(U, L) then (S + A)/A ∈ Π((U + A)/A, L/A). Proof. This follows from Theorem 2.7 and Lemma 1.6. (cid:3) Corollary 2.9 For every solvable Lie algebra L, φ(U, L) = \B∈Π(U,L) B. Proof. Put P = TB∈Π(U,L) B. Then φ(U, L) ⊆ P , by Theorem 2.7 and Lemma 1.1. Let M ∈ [U, L]max. There is an i such that Ai−1 ⊆ M but Ai 6⊆ M (1 ≤ i ≤ n). Then Ai/Ai−1 is not a U -Frattini chief factor of L, so i ∈ I and M ∈ Mi. Thus there is B ∈ Π(U, L) such that B ⊆ M , whence P ⊆ M . Hence P ⊆ φ(U, L). (cid:3) Corollary 2.10 Let L be completely solvable and let U be a subalgebra of L. Then Π(U, L) = {φ(U, L)}. In particular, Π(0, L) = {φ(L)}. Proof. This follows from Theorem 2.7 and Theorem 1.7. (cid:3) Corollary 2.11 Suppose that L is a solvable Lie algebra over a field F of characteristic p, and suppose further that L∞ has nilpotency class less than p. Let U be a subalgebra of L. Then the elements of Π(U, L) are conjugate under I(L : L∞). 10 Proof. This follows from Theorem 2.7 and Theorem 1.8. (cid:3) If L2 is not nilpotent then Π(U, L) can contain more than one element, as the following example shows. Example 2.1 Let F be a field of characteristic p (perfect if p = 2), and L = (⊕p−1 i=0 F ei) ⊕ F c ⊕ F s ⊕ F x with [ei, c] = ei, [ei, s] = ei+1 for i = 0, . . . , p − 2, [ep−1, s] = 0, [ei, x] = iei−1 for i = 0, . . . , p − 1 and e−1 = 0, [s, x] = c, and all other products zero. Put A0 = 0, A1 = ⊕p−1 i=0 F ei, A2 = A1 ⊕ F c, A3 = A2 ⊕ F s, A4 = L. Then 0 = A0 ⊂ A1 ⊂ A2 ⊂ A3 ⊂ A4 = L is a chief series for L in which A2/A1 is the only Frattini chief factor. It is, therefore, straightforward to see that the prefrattini subalgebras of L are the one-dimensional subalgebras F (αc + a) where a ∈ A1 = L∞, α ∈ F . Note that these are all conjugate under inner automorphisms of the form 1+ ad a. This is not always the case, however. For, if B is a faithful completely reducible L-module and we form X = B⊕L, where B2 = 0 and L acts on B under the given L-module action, then the prefrattini subalgebras are still of the form F (αc + a) where a ∈ A1. However, B is the unique minimal ideal of L and these subalgebras are not conjugate under inner automorphisms of the form 1+ ad b, b ∈ B. Since B is the nilradical of X, defining other inner automorphisms is problematic. Note that X∞ = B +A1 which is not nilpotent. References [1] D.W. Barnes and H.M. Gastineau-Hills, 'On the theory of soluble Lie algebras', Math. Z. 106 (1968), 343 -- 354. [2] D.W. Barnes and M.L. Newell, 'Some Theorems on saturated hom- nomorphs of soluble Lie algebras', Math. Z. 115 (1970), 179 -- 187. [3] P. Hauck and H. Kurtzweil, 'A lattice-theoretic characterization of prefrattini subgroups', Manuscripta Math. 1990, 295 -- 301. [4] D.A. Towers, 'On complemented Lie algebras', J. London Math. Soc. (2) 22 (1980), 63 -- 65. 11
1808.10194	2	1808	2018-10-09T16:37:47	On Pr\"ufer-Like Properties of Leavitt Path Algebras	[ "math.RA" ]	Pr\"{u}fer domains and subclasses of integral domains such as Dedekind domains admit characterizations by means of the properties of their ideal lattices. Interestingly, a Leavitt path algebra $L$, in spite of being non-commutative and possessing plenty of zero divisors, seems to have its ideal lattices possess the characterizing properties of these special domains. In [8] it was shown that the ideals of $L$ satisfy the distributive law, a property of Pr\"{u}fer domains and that $L$ is a multiplication ring, a property of Dedekind domains. In this paper, we first show that $L$ satisfies two more characterizing properties of Pr\"{u}fer domains which are the ideal versions of two theorems in Elementary Number Theory, namely, for positive integers $a,b,c$, $\gcd(a,b)\cdot\operatorname{lcm}(a,b)=a\cdot b$ and $a\cdot \operatorname{gcd}(b,c)=\operatorname{gcd}(ab,ac)$. We also show that $L$ satisfies a characterizing property of almost Dedekind domains in terms of the ideals whose radicals are prime ideals. Finally, we give necessary and sufficient conditions under which $L$ satisfies another important characterizing property of almost Dedekind domains, namely the cancellative property of its non-zero ideals.	math.RA	math	On Prufer-like Properties of Leavitt Path Algebras S. Esin, M. Kanuni, A. Ko¸c, K. Radler, K. M. Rangaswamy October 10, 2018 Abstract Prufer domains and subclasses of integral domains such as Dedekind domains admit characterizations by means of the properties of their ideal lattices. Interestingly, a Leavitt path algebra L, in spite of being non- commutative and possessing plenty of zero divisors, seems to have its ideal lattices possess the characterizing properties of these special domains. In [8] it was shown that the ideals of L satisfy the distributive law, a property of Prufer domains and that L is a multiplication ring, a property of Dedekind domains. In this paper, we first show that L satisfies two more characterizing properties of Prufer domains which are the ideal versions of two theorems in Elementary Number Theory, namely, for positive integers a, b, c, gcd(a, b) · lcm(a, b) = a · b and a · gcd(b, c) = gcd(ab, ac). We also show that L satisfies a characterizing property of almost Dedekind domains in terms of the ideals whose radicals are prime ideals. Finally, we give necessary and sufficient conditions under which L satisfies another important characterizing property of almost Dedekind domains, namely the cancellative property of its non-zero ideals. 1 Introduction This paper is devoted to investigating some of the Prufer-like properties of the ideals in a Leavitt path algebra L := LK(E) of an arbitrary directed graph over a field K. Recall that an integral domain D is called a valuation domain if its ideals are totally ordered by set inclusion. D is called a Prufer domain if all its localizations at maximal ideals are valuation domains. D is called an almost Dedekind domain if all its localizations at maximal ideals are noetherian valuation domains and D is called a Dedekind domain if it is a noetherian domain and all its localizations at maximal ideals are noetherian valuation domains (see [5]). There are many equivalent characterizations of Prufer domains that are widely studied in the literature. Some of the characterizations of Prufer domains can be listed as given in [5, Theorem 6.6]: 02010 Mathematics Subject Classification: 16D25; 13F05 Key words and phrases: Leavitt path algebras, Prufer domain 1 1. R is a Prufer domain. 2. If AB = AC, where A, B, C are ideals of R and A is finitely generated and nonzero, then B = C. 3. A(B ∩ C) = AB ∩ AC for all ideals A, B, C of R. 4. (A + B)(A ∩ B) = AB for all ideals A, B of R. 5. A ∩ (B + C) = (A ∩ B) + (A ∩ C) for all ideals A, B, C of R. 6. If A and C are ideals of R, with C finitely generated, and if A ⊆ C, then there is an ideal B of R such that A = BC. Although a Leavitt path algebra L is non-commutative in nature and has plenty of zero divisors, it is somewhat intriguing and certainly interesting that the ideals of such a highly non-commutative algebra share many of the proper- ties of the ideals of various types of (commutative) integral domains. To start with, the multiplication of ideals in L is commutative ([1], [9, Theorem 3.4]), L satisfies the property of a B´ezout domain, namely, all the finitely generated ide- als of L are principal ([8]), every ideal of L is projective, a property of Dedekind domains and the ideal lattice of L is distributive ([9]) which characterizes Prufer domains among integral domains. In this paper, we consider characterizing properties of Prufer domains which are ideal versions of well-known theorems in elementary number theory. Re- call that if a, b, c are positive integers, then gcd(a, b) · lcm(a.b) = a · b and a · gcd(b, c) = gcd(ab, ac) and a · lcm(b, c) = lcm(ab, ac). The first two results when expressed in terms of its ideals lead to a characterization of Prufer do- mains: An integral domain D is a Prufer domain if and only if, for any two non-zero ideals A, B of D, (A ∩ B)(A + B) = AB and, if and only if, for any three ideals A, B, C in D, we have A(B ∩ C) = AB ∩ AC (see [5]). We will show that every Leavitt path algebra satisfies these two characterizing proper- ties. (Note that the ideal version of the result a.lcm(b, c) = lcm(ab, ac), namely, A(B + C) = AB + AC holds for ideals A, B, C in any ring R.) We also in- vestigate whether a Leavitt path algebra L possesses any of the properties of almost Dedekind domains which form a subclass of the class of Prufer domains. It is known (see [5]) that an integral domain D is an almost Dedekind domain if and only if every ideal in D, whose radical is prime, is a power of a prime ideal. We show that every Leavitt path algebra also possesses this characteriz- ing property. As a corollary, we show that two more properties of a Dedekind domain are satisfied by Leavitt path algebras. Almost Dedekind domains D are also characterized among integral domains by the property that every non-zero ideal A of D is cancellative, that is, AB = AC implies B = C for any two ideals B, C of D. While not every Leavitt path algebra satisfies this property, we give necessary and sufficient conditions on the graph E under which every non-zero ideal of LK(E) is cancellative. Various graphical constructions illustrate our results. 2 2 Preliminaries In this section, we will mention some of the needed basic concepts and results in Leavitt path algebras. A directed graph E = (E0, E1, r, s) consists of two countable sets E0, E1 and functions r, s : E1 → E0. The elements E0 and E1 are called vertices and edges, respectively. For each e ∈ E1, s(e) is the source of e and r(e) is the range of e. If s(e) = v and r(e) = w, then we say that v emits e and that w receives e. A vertex which does not receive any edges is called a source, and a vertex which emits no edges is called a sink. A vertex v is called a regular vertex if it emits a non-empty finite set of edges. A vertex is called an infinite emitter if it emits infinitely many edges. A graph is called row-finite if s−1(v) is a finite set for each vertex v. For a row-finite graph the edge set E1 of E is finite if its set of vertices E0 is finite. Thus, a row-finite graph is finite if E0 is a finite set. A path in a graph E is a sequence of edges µ = e1 . . . en such that r(ei) = s(ei+1) for i = 1, . . . , n − 1. In such a case, s(µ) := s(e1) is the source of µ and r(µ) := r(en) is the range of µ, and n is the length of µ, i.e., l(µ) = n. If s(µ) = r(µ) and s(ei) 6= s(ej) for every i 6= j, then µ is called a cycle. An exit for a path µ = e1 . . . en is an edge e such that s(e) = s(ei) for some i and e 6= ei. A graph E is said to satisfy condition (L) if every cycle in E has an exit. A subset D of vertices is said to be downward directed if for any u, v ∈ D, there exists a w ∈ D such that u ≥ w and v ≥ w. A subset H of E0 is called hereditary if whenever v ∈ H and w ∈ E0 for which v ≥ w, then w ∈ H. H is saturated if whenever a regular vertex v has the property that {r(e)e ∈ E1, s(e) = v} ⊆ H, then v ∈ H. The path K-algebra over E is defined as the free K-algebra K[E0∪ E1] with the relations: (1) vivj = δijvi for every vi, vj ∈ E0. (2) ei = eir(ei) = s(ei)ei for every ei ∈ E1. This algebra is denoted by KE. Given a graph E, define the extended graph i ei ∈ E1} of E as the new graph bE = (E0, E1 ∪ (E1)∗, r′, s′) where (E1)∗ = {e∗ and the functions r′ and s′ are defined as r′E 1 = r, s′E 1 = s, r′(e∗ i ) = s(ei) and s′(e∗ i ) = r(ei). The Leavitt path algebra of E with coefficients in K is defined as the path (CK1) e∗ algebra over the extended graph bE, with relations: (CK2) vi =P{ej ∈E 1 s(ej )=vi} eje∗ i ej = δij r(ej ) for every ej ∈ E1 and e∗ j i ∈ (E1)∗. for every regular vertex vi ∈ E0. This algebra is denoted by LK(E). The conditions (CK1) and (CK2) are called the Cuntz-Krieger relations. 3 A ring R is called a ring with local units, if for every non-empty finite subset X of R, there is a non-zero idempotent u ∈ R such that ux = x = xu for v. all x ∈ X. When E0 is finite, LK(E) is a ring with unit element 1 = Xv∈E 0 Otherwise, LK(E) is not a unital ring, but is a ring with local units consisting of sums of distinct elements of E0. a = kiαiβ∗ nPi=1 A useful observation is that every element a of LK(E) can be written as i , where ki ∈ K, αi, βi are paths in E and n is a suitable integer. vLK(E). Another useful fact is that if p∗q 6= 0, where p, q are paths, then either p = qr or q = ps where r, s are suitable paths in E. Moreover, LK(E) = Lv∈E 0 LK(E)v = Lv∈E 0 One of the most important properties of Leavitt path algebras is that each Ln induced by defining, LK(E) is a Z-graded K-algebra. That is, LK(E) =Mn∈Z deg(v) = 0 for all v ∈ E0 and deg(e) = 1, deg(e∗) = −1 for all e ∈ E1. Further, the homogeneous component Ln for each n ∈ Z is given by i ∈ L : Ln = {Pkiαiβ∗ An ideal I of LK(E) is said to be a graded ideal if I =Mn∈Z l(αi) − l(βi) = n} where ki ∈ K, αi, βi ∈ P ath(E). (I ∩ Ln). We shall be using the following concepts and results from [10]. A vertex v is called a breaking vertex of a hereditary subset H if v belongs to the set BH := {v ∈ E0\H v is an infinite emitter and 0 < s−1(v)∩r−1(E0\H) < ∞}. In words, BH consists of those vertices of E which are infinite emitters, which do not belong to H, and for which the ranges of the edges they emit are all, except for a finite (but nonzero) number, inside H. For v ∈ BH , the element vH of LK(E) is defined by vH := v − Xe∈s−1(v)∩r−1(E 0\H) ee∗. We note that any such vH is homogeneous of degree 0 in the standard Z-grading on LK(E). Given a hereditary saturated subset H and a subset S ⊆ BH , (H, S) is called an admissible pair. The ideal generated by H ∪{vH : v ∈ S} is denoted by I(H, S) where (H, S) is an admissible pair. The quotient graph E\(H, S) of E by an admissible pair (H, S) is defined as follows: (E\(H, S))0 = (E0\H) ∪ {v′v ∈ BH\S}, 4 (E\(H, S))1 = {e ∈ E1r(e) /∈ H} ∪ {e′e ∈ E1 and r(e) ∈ BH\S}, and range and source maps in E\(H, S) are defined by extending the range and source maps in E when appropriate, and in addition setting s(e′) = s(e) and r(e′) = r(e)′. It was shown in [10] that LK(E)/I(H, S) ∼= LK(E\(H, S)). Let Λ be an arbitrary non-empty set. Given a ring R, MΛ(R) denotes the ring of matrices with entries from R, all but finitely many of which are non-zero and where the rows and columns are indexed by elements of Λ. We will be using an important result about the ideals of the ring MΛ(R). This result was proved in Theorem 3.1 of [4] when R is a ring with identity 1 and when Λ is finite. We need this result for rings with local units and when Λ is an arbitrary non-empty set. As far as we know, this generalized statement has not appeared in print and so we give the general statement and its proof. Thus Proposition 1(a) below is a generalization of Theorem 3.1 of [4] to rings with local units. We thank Zak Mesyan for help in writing the proof of Proposition 1(b). For rings R with identity, this Proposition is also obtained by using the Morita equivalence of R and MΛ(R) (see [2]). Proposition 1. Suppose R is a ring with local units and Λ is an arbitrary non- empty set. (a) Every ideal of MΛ(R) is of the form MΛ(A) for some ideal A of R. The map A 7−→ MΛ(A) defines a lattice isomorphism between the lattice of ideals of R and the lattice of ideals of MΛ(R). (b) For any two ideals A, B of R, MΛ(AB) = MΛ(A)MΛ(B). Proof. If A is an ideal of R, then it is easy to see that MΛ(A) is an ideal of MΛ(R). Also, if the ideals A 6= B, then MΛ(A) 6= MΛ(B). Let I be a non-zero ideal of MΛ(R). We wish to show that I = MΛ(A) for some ideal A of R. Let A be the set of all the entries at the (first row - first column) position in all the matrices belonging to I. A is clearly an ideal in R. We wish to show that I = MΛ(A). Let U denote the set of all local units in R. Let 0 6= M = (mij) ∈ I. Corresponding to all the finitely many non-zero entries mij in M , choose a local unit u ∈ U satisfying umij = mij = miju for all i, j. For any k, l, we have the identity Eij uM Eklu = mjkEilu (∗) where, for every i, j, Eij u denotes the Λ × Λ matrix having u at the ith row In particular, E1juM Ek1u = and jth column entry and 0 at every other entry. mjkE11u ∈ I and thus mjk ∈ A for all j, k ∈ Λ. Hence M ∈ MΛ(A). Conversely, let N = (aij ) ∈ MΛ(A). Since ail ∈ A, there is a matrix M = (mij ) ∈ I such that m11 = ail. Let v ∈ U be a local unit satisfying vaij = aij = aijv and also vmij = mij = mijv for all the entries aij in N . It is enough to show that 5 ailEilv ∈ I for all i, l. Applying the identity (∗), we have ailEilv = m11Eilv = Ei1vM E1l ∈ I, for all i, l. This proves that I = MΛ(A). form P aibj for some ai ∈ A and bj ∈ B and hence an element of AB. Thus It is straightforward to verify that, for any two ideals A, B of R, MΛ(A + B) = MΛ(A) + MΛ(B) and MΛ(A ∩ B) = MΛ(A) ∩ MΛ(B). This shows that map A 7−→ MΛ(A) defines a lattice isomorphism. We next show that MΛ(AB) = MΛ(A)MΛ(B) for any two ideals A, B of R. Given I ∈ MΛ(A) and J ∈ MΛ(B), every entry of IJ is a finite sum of the MΛ(A)MΛ(B) ⊆ MΛ(AB). To prove the reverse inclusion, first note that every element M ∈ MΛ(AB) can be written as a finite sum of elements of the form abEiju where a ∈ A, b ∈ B and u is a local unit corresponding to the finitely many non-zero entries of M and also satisfying ua = a = au and ub = b = bu. Since MΛ(A)MΛ(B) is an ideal of MΛ(R), it is enough to show that abEiju ∈ MΛ(A)MΛ(B), for all a ∈ A, b ∈ B and i, j ∈ Λ. Now aEilu ∈ MΛ(A) and bElju ∈ MΛ(B) and so abEiju = abEiluElju = (aEilu)(bElju) ∈ MΛ(A)MΛ(B). Thus MΛ(AB) = MΛ(A)MΛ(B) for any two ideals A, B of R. Throughout the following, L will denote the Leavitt path algebra LK(E) of an arbitrary graph E over a field K. 3 Prufer-like properties satisfied in Leavitt path algebras In this section, we shall describe how the ideals of every Leavitt path algebra L satisfy two of the characterizing properties of a Prufer domain mentioned in the introduction. In this connection, the graded ideals of L seem to be well-behaved and some extra efforts are needed in dealing with the non-graded ideals of L. The following theorem consists of the results in [8] and [7] which will be used in the sequel. Theorem 2. Let I be a non-graded ideal of L = LK(E) with H = I ∩ E0 and S = {u ∈ BH : uH ∈ I}. Then (i) ([8, Theorem 4]) I = I(H, S) +Pt∈Thft(ct)i where T is some index set, for each t ∈ T, ct is a cycle without exits in E\(H, S), c0 s = ∅ for t 6= s and ft(x) ∈ K[x] is a polynomial with its constant term non-zero and is of the smallest degree such that ft(ct) ∈ I. t ∩ c0 6 (ii) ([7, Lemma 3.6]) I(H, S) is the largest graded ideal inside I. We shall denote I(H, S) by gr(I) and call it the graded part of the ideal I. Before proving the main theorem, we consider the case of graded ideals which are easy to handle. A useful property of graded ideals of a Leavitt path algebra L (see [9, Lemma 3.1]) is that if A is a graded ideal of L, then for any ideal B, AB = A ∩ B. Lemma 3. Let A, B, C be three ideals of a Leavitt path algebra L. If one of them is a graded ideal then Proof. Case 1: Suppose A is a graded ideal. Then by [9, Lemma 3.1 (i)], A(B ∩ C) = AB ∩ AC. A(B ∩ C) = A ∩ (B ∩ C) = (A ∩ B) ∩ (A ∩ C) = AB ∩ AC Case 2: Suppose B or C is a graded ideal, say, B is a graded ideal. Then AB ∩ AC = (A ∩ B) ∩ AC = B ∩ AC = B(AC) since B is graded since AC ⊆ A since B is graded = ABC = A(B ∩ C) since AB = BA by [9, Theorem 3.4] since B is graded. Next, we consider the case when all ideals are non-graded in the next two lemmas. In the proofs, we shall be using Theorem 4.3 of [9], namely, A ∩ (B + C) = (A ∩ B) + (A ∩ C) for any three ideals A, B, C in L. We shall also using the fact that for a graded ideal I, IJ = I ∩ J for any ideal J in L. Lemma 4. Let A, B, and C be non-graded ideals. If A ⊆ gr(A) + gr(B ∩ C) or (B ⊆ gr(A)+gr(B∩C) and C ⊆ gr(A)+gr(B∩C)), then A(B∩C) = AB∩AC. Proof. We want to show that AB ∩ AC ⊆ A(B ∩ C) since the other inclusion is always true. Suppose that A ⊆ gr(A) + gr(B ∩ C). By the Modular Law, A = A ∩ (gr(A) + gr(B ∩ C)) = gr(A) + (A ∩ gr(B ∩ C)). Then AB = (gr(A) + (A ∩ gr(B ∩ C)))B = gr(A)B + (A ∩ gr(B ∩ C))B = gr(A)B + (A ∩ gr(B ∩ C)) 7 by [9, Lemma 3.2] and similarly, AC = gr(A)C + (A ∩ gr(B ∩ C)). Hence AB ∩ AC = (gr(A)B + (A ∩ gr(B ∩ C))) ∩ (gr(A)C + (A ∩ gr(B ∩ C))) By [9, Theorem 4.3], AB ∩ AC = [(gr(A)B + (A ∩ gr(B ∩ C))) ∩ gr(A)C] +[(gr(A)B + (A ∩ gr(B ∩ C))) ∩ (A ∩ gr(B ∩ C)))] Now, by [9, Theorem 4.3, Lemma 3.1 (i)], AB ∩ AC = gr(A) ∩ (B ∩ C) + gr(A) ∩ gr(B ∩ C) ∩ C +A ∩ gr(B ∩ C) = gr(A)(B ∩ C) + gr(A)gr(B ∩ C) +Agr(B ∩ C) ⊆ A(B ∩ C) by using [9, Lemma 3.1 (i)] Now, suppose that B ⊆ gr(A) + gr(B ∩ C) and C ⊆ gr(A) + gr(B ∩ C). Then, by the Modular Law, B = B ∩ (gr(A) + gr(B ∩ C)) = gr(B ∩ C) + (gr(A) ∩ B) and similarly, C = gr(B ∩ C) + (gr(A) ∩ C). Hence, AB = Agr(B ∩ C) + A(gr(A) ∩ B) = Agr(B ∩ C) + (gr(A) ∩ B) by [9, Lemma 3.2] and similarly, AC = Agr(B ∩ C) + (gr(A) ∩ C). Therefore, by using [9, Theorem 4.3, Lemma 3.1 (i)], AB ∩ AC = [Agr(B ∩ C) + (gr(A) ∩ B)] ∩ [Agr(B ∩ C) + (gr(A) ∩ C)] = Agr(B ∩ C) + gr(A)gr(B ∩ C) + gr(A)gr(B ∩ C) + gr(A)(B ∩ C) ⊆ A(B ∩ C) In proving the next lemma, we shall be using the easy-to-see statement that, for any two ideals B, C of L, gr(B ∩ C) = gr(B) ∩ gr(C). Lemma 5. Let A, B, and C be non-graded ideals of L. If A * gr(A)+gr(B∩C) and (B * gr(A) + gr(B ∩ C) or C * gr(A) + gr(B ∩ C)), then A(B ∩ C) = AB ∩ AC. Proof. Without loss of generality, we may assume A * gr(A) + gr(B ∩ C) and B * gr(A) + gr(B ∩ C). Let I = I(H, S) = gr(A) + gr(B ∩ C). By Theorem 2 (i), A = I(H1, S1) +Xi∈X C = I(H3, S3) +Xk∈Z hfi(ci)i , B = I(H2, S2) +Xj∈Y hgj(cj )i , and hhk(ck)i 8 where X, Y, and Z are some index sets, I(H1, S1) = gr(A), I(H2, S2) = gr(B), I(H3, S3) = gr(C) and for all i ∈ X, j ∈ Y, and k ∈ Z, fi(x), gj (x), hk(x) ∈ K[x] and ci, cj, and ck are cycles without exits in E\(H1, S1), E\(H2, S2), and E\(H3, S3), respectively. In L = L/I ∼= LK(E\(H, S)), A = (A + I)/I is an epimorphic image of A/gr(A) and let B = (B +I)/I and C = (C +I)/I. Hence, A = Xi∈X ′ hfi(ci)i , B = (gr(A) + gr(B) + I)/I +Xj∈Y ′ hgj(cj)i + I /I, and C = (gr(A) + gr(C) + I)/I +"Xk∈Z ′ hhk(ck)i + I# /I, where X ′, Y ′, Z ′ are subsets of the sets X, Y , Z respectively. For the sake of convenience, we shall write B = (gr(A) + gr(B) + I)/I + [P2 +I] /I, where P2 =Pj∈Y ′ hgj(cj)i and C = (gr(A) + gr(C) + I)/I + [P3 +I] /I, where P3 =Pk∈Z ′ hhk(ck)i. B ∩ C = {(gr(A) + gr(B) + I)/I + [P2 +I] /I} ∩{(gr(A) + gr(C) + I)/I + [P3 +I] /I} = (gr(A) + gr(B) + I)/I ∩ [P3 +I] /I +(gr(A) + gr(C) + I)/I ∩ [P2 +I] /I + [P2 +I] /I ∩ [P3 +I] /I, noting that [(gr(A) + gr(B) + I)/I] ∩ [(gr(A) + gr(C) + I)/I simplifies to 0. ¯A( ¯B ∩ ¯C) =(cid:2) ¯A(gr(A) + gr(B) + I)/I(cid:3) ∩(cid:2) ¯AP3 +I(cid:3) /I +(cid:2) ¯A(gr(A) + gr(C) + I)/I(cid:3) ∩(cid:2) ¯AP2 +I(cid:3) /I + ¯A ([(P3 +I) /I] ∩ [(P2 +I) /I]) (1) On the other hand, ¯A ¯B =(cid:2) ¯A(gr(A) + gr(B) + I)/I(cid:3) +(cid:2) ¯AP2 +I(cid:3) /I 9 and ¯A ¯C =(cid:2) ¯A(gr(A) + gr(C) + I)/I(cid:3) +(cid:2) ¯AP3 +I(cid:3) /I Hence, by [9, Theorem 4.3] A B ∩ A C = (cid:2) ¯A(gr(A) + gr(B) + I)/I(cid:3) ∩(cid:2) ¯A(gr(A) + gr(C) + I)/I(cid:3) } =0 since (gr(B)+gr(A)+I)/I∩(gr(C)+gr(A)+I)/I=0 {z + ¯A(gr(A) + gr(B) + I)/I ∩(cid:2) ¯AP3 +I(cid:3) /I +(cid:2) ¯AP2 +I(cid:3) /I ∩ ¯A(gr(A) + gr(C) + I)/I +(cid:2) ¯AP2 +I(cid:3) /I ∩(cid:2) ¯AP3 +I(cid:3) /I (2) Let G be the graded ideal of ¯L generated by the vertices on all the cycles ci where i belongs to X ′. Since the ci are cycles without exits in E\(H, S), G suitable index sets by [1, Theorem 2.7.3]. Note that ¯A is contained in G and so ¯AG = ¯A, by [9, Lemma 3.2]. Then We wish to show that ¯A ¯B ∩ ¯A ¯C = ¯A( ¯B ∩ ¯C). Now comparing (1) and (2), all we need is to show that ¯A ([(P2 +I) /I] ∩ [(P3 +I) /I]) =(cid:2) ¯A (P2 +I) /I(cid:3) ∩(cid:2) ¯A (P3 +I) /I(cid:3) . is isomorphic to the ring direct sum Li∈X ′ MΛi(K[x, x−1]) where the Λi are ¯A([(P2 +I)/I] ∩ [(P3 +I)/I]) = ¯AG([P2 +I)/I] ∩ [(P3 +I)/I]) = ¯A(G[(P2 +I)/I] ∩ [G(P3 +I)/I]) by Lemma 3 = ¯A([G ∩ (P2 +I)/I] ∩ [G ∩ (P3 +I)/I]) as G is graded. phic to the ring direct sum Li∈X ′ MΛi (K[x, x−1]). Moreover, by Proposition 1, there is an isomorphism between the ideal lattices of MΛi(K[x, x−1]) and K[x, x−1] which preserves multiplication. Since K[x, x−1] is a Prufer domain, H(K ∩ L) = HK ∩ HL holds for any three ideals of K[x, x−1] and consequently, any three ideals of G also satisfy this property. We observe that ¯A([G∩(P2 +I)/I]∩[G∩(P3 +I)/I]) = ( ¯A[G∩(P2 +I)/I]∩ ¯A[G∩(P3 +I)/I]) = ¯AG(P2 +I)/I ∩ ¯AG(P3 +I)/I as G is graded = ¯A(P2 +I)/I ∩ ¯A(P3 +I)/I, by [9, Lemma 3.2]. Now all three ideals in the preceding equation are ideals of G and G is isomor- 10 We thus conclude that ¯A ¯B ∩ ¯A ¯C = ¯A( ¯B ∩ ¯C). Then AB ∩ AC = A(B ∩ C) + (gr(A) + gr(B ∩ C)). Now A ∩ B ∩ C contains both AB ∩ AC and A(B ∩ C) and so using modular law, we have AB ∩ AC = (AB ∩ AC) ∩ (A ∩ B ∩ C) = [A(B ∩ C) + (gr(A) + gr(B ∩ C))] ∩ (A ∩ B ∩ C) = A(B ∩ C) + (gr(A) + gr(B ∩ C)) ∩ (A ∩ B ∩ C) gr(A)∩(A∩B∩C)+gr(B∩C)∩(A∩B∩C) = A(B ∩ C) + gr(A)(B ∩ C) + Agr(B ∩ C) ⊆ A(B ∩ C). {z } Since A(B ∩ C) ⊆ AB ∩ AC is always true, we get A(B ∩ C) = AB ∩ AC. Hence, we can state the main result. Theorem 6. If A, B, C are any ideals of a Leavitt path algebra L of an arbitrary graph E, then Proof. By using Lemmas 3, 4 and 5, the result follows. A(B ∩ C) = AB ∩ AC. In elementary number theory, it is well-known that for any two positive integers a, b, we have gcd(a, b) · lcm(a, b) = ab. This property can be stated for ideals as: for any ideals A, B, (A + B)(A ∩ B) = AB. This equality holds for ideals in a Dedekind domain. If this equality holds for finitely generated ideals A, B, then the integral domain is a Prufer domain. The next theorem shows that any Leavitt path algebra satisfies this characterizing property. We shall prove this by using Theorem 6. Theorem 7. For any two ideals A, B of a Leavitt path algebra L, (A + B)(A ∩ B) = AB. Proof. Now, by Theorem 6 and [9, Theorem 3.4], (A + B)(A ∩ B) = (A + B)A ∩ (A + B)B ⊇ BA ∩ AB = AB ∩ AB = AB. The converse inclusion is always true since (A + B)(A ∩ B) = A(A ∩ B) + B(A ∩ B) ⊆ AB+BA = AB+AB = AB Thus we obtain (A + B)(A ∩ B) = AB. 11 by [9, Theorem 3.4]. 4 Almost Dedekind domains and Leavitt path algebras As noted in the Introduction, an almost Dedekind domain D is a Prufer domain with the property that all its localizations with respect to maximal ideals are noetherian valuation domains. In this section, we investigate whether a Leavitt path algebra L satisfies any of the other characterizing properties of almost Dedekind domains. Recall that the radical √I of an ideal I in a ring R is It is known (see the intersection of all the prime ideals of R containing I. [5]) that an integral domain D is an almost Dedekind domain if and only if every non-zero ideal I with its radical √I a prime ideal is a power of a prime ideal. It turns out that every Leavitt path algebra L satisfies this property. As a corollary, we show that if P is a non-zero prime ideal in a Leavitt path algebra L, then all the P -primary ideals of L form a well-ordered chain under set inclusion and that there are no ideals of L strictly between P n and P n+1, properties satisfied by Dedekind domains. We also consider the property of non- zero ideals being cancellative, an important property that characterizes almost Dedekind domains among integral domains. Recall that a non-zero ideal A in a ring R is cancellative if, for any two ideals B, C of R, AB = AC implies that B = C. Examples indicate that not all Leavitt path algebras have this property. We show that, in a Leavitt path algebra L := LK(E), every non-zero ideal of L is cancellative if and only if either (a) there is a cycle c without exits in E based at a vertex v such that u ≥ v for every u ∈ E0 and HE= {H} where H is the hereditary saturated closure of {v} and BH = ∅, or (b) E satisfies Condition (K), HE ≤ 2, for any two X, Y ∈ HE with X 6= Y , X ∩ Y = ∅ and, for each H ∈ HE, BH is empty and H is the saturated closure of each u ∈ H. Here HE denotes the set of all non-empty proper hereditary saturated subsets of vertices in the graph E. Equivalent conditions on L are, either (a) L contains a graded ideal M which contains every proper ideal of L and M ∼= MΛ(K[x, x−1]) where Λ is an arbitrary finite or infinite index set or (b) L has at most two non-zero ideals each of which is graded and is a principal ideal. We begin by showing that every Leavitt path algebra satisfies the first men- tioned property of almost Dedekind domains. Theorem 8. Let I be a non-zero ideal of L. If its radical √I is a prime ideal, say P , then I = P n, for some n ≥ 1. Proof. If I is a graded ideal, then I = √I by [3, Lemma 2.1]. So if √I is prime, then trivially I is a prime power. Suppose now that I is a non-graded ideal such that its radical √I is a prime ideal. By [8, Theorem 4], I = I(H, S)+Xi∈X hfi(ci)i, where the ci are distinct cycles without exits in E\(H, S), fi(x) are polynomials in K[x] with non-zero constant terms. Now, by [9, Lemma 5.4], gr(√I) = gr(I) = I(H, S). By [7, Theorem 3.12], the prime ideal √I = I(H, BH ) + hp(c)i where p(x) ∈ K[x] is an irreducible polynomial with non-zero constant term, c is a cycle without exists in E\(H, BH ) and, moreover, E\(H, BH ) is 12 downward directed. The downward directness of E\(H, BH ) implies that c is the only cycle without exits in E\(H, BH ). Thus I must be of the form I = I(H, BH ) + hf (c)i where f (x) ∈ K[x] has its constant term non-zero. We claim f (x) = pn(x) for some integer n ≥ 1. Suppose, on the contrary, f (x) = q(x)g(x) where q(x) 6= p(x) is an irreducible polynomial with non- zero constant term. Then I = I(H, BH ) + hf (c)i ⊆ I(H, BH ) + hq(c)i. Since E\(H, BH ) is downward directed, I(H, BH )+hq(c)i is a prime ideal ([7, Theorem 3.12]) and so it contains the radical of I, namely I(H, BH ) + hp(c)i. Then, in L/I(H, BH) ∼= LK(E\(H, BH )), hp(c)i ( hq(c)i ⊆ M , the ideal generated by c0. Now, by [1, Theorem 2.7.1], M ∼= MΛ(K[x, x−1]) for a suitable set Λ and by Proposition 1, the ideal lattices of MΛ(K[x, x−1]) and K[x, x−1] are isomorphic. Now, both hp(x)i and hq(x)i are maximal ideals in K[x, x−1] and so, identifying M with MΛ(K[x, x−1]), both hp(c)i and hq(c)i are maximal ideals of M . This is a contradiction since hp(c)i ( hq(c)i. Hence f (x) = pn(x) for some n ≥ 1. Then I = I(H, BH ) + hpn(c)i = (I(H, BH ) + hp(c)i)n is a power of the prime ideal I(H, BH ) + hp(c)i. Recall that, given a non-zero prime ideal P , an ideal I is called P -primary, if its radical √I = P . It is known (see Theorem 6.20, [5]) a noetherian domain D is a Dedekind domain if and only if for any non-zero prime ideal P of D, the set of all the P -primary ideals of D is a totally ordered set under inclusion and, if and only if there are no ideals of D strictly between P and P 2. As an easy corollary to Theorem 8, we show that a Leavitt path algebra L satisfies both these properties of a Dedekind domain. Corollary 9. Let L be a Leavitt path algebra and let P be a non-zero prime ideal of L. Then (i) the set of all the P -primary ideals is totally ordered under set inclusion; (ii) there is no ideal A of L satisfying P 2 ( A ( P . Proof. (i) Note that if I is a P -primary ideal of L, then √I = P . By Theorem 8, I = P k for some k ≥ 1. Thus the set of P -primary ideals of L is the set {P n : n ≥ 1} which is a totally ordered (countable) set under inclusion. (ii) We shall actually show that there is no ideal A satisfying P n+1 ( A ( P n for any n ≥ 1. Note that if there is an ideal A of L satisfying P n+1 ⊆ A ⊆ P n for some n ≥ 1, then √A = P and so, by Theorem 8, A is a power of P and hence A = P n+1 or P n. Next, we consider the cancellative property of the ideals of an almost Dedekind domains. This property does not seem to hold in arbitrary Leavitt path algebras as the next two examples show. Example 10. Consider the following graph E: •v1 ∞ •v2 e1 / •v3 e4 •v6 ∞ / •v7 •v4 =③ ③ e3 ③ ③ ③ e2 ③ ③ ③ / •v5 ∞ 13 o o / o o / = / O O from v to w, that is v is an infinite emitter. ) / •w means that there are infinitely many edges emitted Given the ideals A = hv1i, B = hv5i and C = hv7i of LK(E). Clearly B 6= C (In a graph •v ∞ but AB = 0 = AC. Example 11. Consider the graph E c •u / •v •w / •z Then H = {v} is a hereditary saturated subset. Let A = hHi, be the principal ideal generated by H. Clearly c has no exits in E\H. Let B be the nongraded ideal A + hp(c)i, where p(x) is a polynomial in K[x]. Clearly gr(B) = A. Since A is a graded ideal, we apply [9, Lemma 3.1(i)], to conclude that AB = A∩ B = A = A2 = AA. However, A 6= B. The next theorem gives necessary and sufficient conditions (both graphical and algebraic) under which non-zero finitely generated ideals of L is cancella- tive. Interestingly, in this case, every non-zero ideal of L also turns out to be cancellative. We prove a useful lemma and in its proof we shall again use the result [8, Lemma 3.1] that if A is a graded ideal of L, then for any other ideal B, A · B = A ∩ B and that A2 = A. Lemma 12. If the cancellation property for finitely generated ideals holds in L, then there cannot be two ideals A, B with A ( B and A graded and non-zero. In particular, gr(B) = 0. Proof. By [10], A = I(H, S), where H = A ∩ E0 and S ⊆ BH . Since A 6= 0, H 6= ∅. Then, for a vertex u ∈ H, C = hui is a finitely generated graded ideal. If B is an ideal such that A ( B then we get C·B = C∩B = C = C·C, but B 6= C, a contradiction to the cancellation property. This proves the first statement. By taking A = gr(B), the second statement follows from the first. Corollary 13. If the cancellation property for finitely generated ideals holds in L, then every non-empty hereditary saturated subset H ( E0 is the hereditary saturated closure of each single vertex u ∈ H and, moreover, BH = ∅. In particular, every non-zero graded ideal of L is a principal ideal, being generated by a single vertex. Proof. If there is a vertex u ∈ H such that the hereditary saturated closure X of {u} is not equal to H, then the non-zero graded ideal A = hXi satisfies A ( hHi, contradicting Lemma 12. Likewise, if BH 6= ∅, then again we have the non-zero graded ideal I(H,∅) ( I(H, BH ), contradiction Lemma 12. If A = I(H, S) is a non-zero graded ideal of L, then since BH = ∅, S = ∅ and since H is the hereditary saturated closure of any vertex u ∈ H, A = hHi = h{u}i is a principal ideal. 14 / 8 8 / o o / Theorem 14. Let E be an arbitrary graph. The following conditions are equiv- alent for L: (i) The cancellation property holds for all non-zero ideals in L; (ii) The cancellation property holds for all non-zero finitely generated ideals of L; (iii) Either (a) there is a cycle c without exits in E based at a vertex v such that u ≥ v for every u ∈ E0 and HE= {H} where H is the hereditary saturated closure of {v} and BH = ∅, or (b) E satisfies Condition (K), HE ≤ 2, for any two X, Y ∈ HE with X 6= Y , X ∩ Y = ∅ and, for each H ∈ HE, BH is empty and H is the saturated closure of each u ∈ H. (iv) Either (a) L contains a graded ideal M which contains every proper ideal of L and M ∼= MΛ(K[x, x−1]) where Λ is an arbitrary finite or infinite index set or (b) L has at most two non-zero ideals each of which is graded and is a principal ideal and is a principal ideal. Proof. Clearly (i)⇒(ii). Assume (ii). Case (a): Suppose L has a non-graded prime ideal P with P ∩ E0 = H ′. By [7, Theorem 3.12], P = I(H ′, BH ′ ) + hp(c)i, where c is a cycle without exits based at a vertex v in E\(H ′, B′ H), p(x) is an irreducible polynomial in K[x, x−1] such that p(c) ∈ P and u ≥ v for all u ∈ E0\H ′. By Lemma 12, I(H ′, BH ′ ) = 0 and hence both H ′ and BH ′ are empty. This means that E contains a unique cycle c without exits based at a vertex v and u ≥ v for every vertex u ∈ E0. Let H be the hereditary saturated closure of {v}. Observe that there cannot be two members X, Y ∈ HE with one of them properly containing the other, say X ( Y . Because, we will then have two non-zero graded ideals A = hXi and B = hY i with A ( B and this is not possible by Lemma 12. We claim that HE= {H}. Suppose, on the contrary, there is another element Z ∈ HE. As noted above, Z * H and H * Z. So Z ∩ H ( H and is further non-empty since v ∈ Z ∩ H. This then gives rise to two non-empty hereditary saturated subsets with one containing the other, a contradiction. Also BH = ∅, by Corollary 13. This proves (iii)(a). Case (b): Suppose every prime ideal of L is graded. Then, by [7, Corollary 3.13], E satisfies Condition (K) and so every ideal of L is graded. If HE = 0, then HE = ∅ and we are done. Assume HE 6= 0. Suppose X, Y ∈ HE with X 6= Y . We claim X ∩ Y = ∅. Because if a vertex u ∈ X ∩ Y , then, by Corollary 13, both X and Y are saturated closures of {u} and hence X = Y , a contradiction. We claim that HE ≤ 2. Indeed if there are three distinct members X, Y, Z ∈ HE, then the three distinct non-zero graded ideals A = hXi, B = hY i, C = hZi are principal by Corollary 13 and satisfy A·B = A∩B = 0 = A ∩ C = A · C, but B 6= C. This contradiction shows that HE ≤ 2. By Corollary 13, each H ∈ HE is the hereditary saturated closure of each u ∈ H and the corresponding BH = ∅. This proves (iii)(b). Assume (iii) (a). Let M be the ideal of L generated by the hereditary saturated closure H of {v}, so M is a graded ideal of L. Since c is a cycle without 15 exits, by [1, Lemma 2.7.1], M ∼= MΛ(K[x, x−1]) where Λ is an arbitrary finite or infinite index, being the set of all paths in E that end at v, but do not include the entire cycle c. Since HE = {H} and since BH = ∅, M is the only non-zero graded ideal of L. We claim that every non-zero ideal N of L is contained in M (and non-graded). Suppose N 6= M . Now N must be a non-graded ideal, since M is the only non-zero graded ideal of L. Then, by Lemma 12, gr(N ) = 0 and so N does not contain any vertices. As E0 is downward directed, [7, Lemma 3.5] then implies that N = hf (c′)i where f (x) ∈ K[x] and c′ is a cycle without exits in E. Since u ≥ v for every u ∈ E0, c is the only cycle without exits in E and so c′ = c. Then N = hf (c)i ⊂ M , as c ∈ M . This proves (iv)(a). Assume (iii)(b). Since Condition (K) holds, every ideal of L is graded [1]. If HE= ∅, then the only hereditary saturated subsets of E0 are E0 and ∅ and so L contains no non-zero proper ideals. Suppose HE= {H}, with BH = ∅. Then I = hH,∅i is the only proper non-zero ideal of L. Suppose HE= {H1, H2}. Since BH1 = ∅ = BH2 , A = hH1i and B = hH2i are the only proper non-zero ideals of L and they both are principal ideals as H1, H2 are saturated closures of single vertices v1 ∈ H1 and v2 ∈ H2. This proves (iv)(b). Assume (iv)(a) so that L contains a graded ideal M ∼= MΛ(K[x, x−1]) where Λ is an arbitrary finite or infinite index set and that M contains every other proper ideal of L. Now, by Proposition 1, the ideals of MΛ(K[x, x−1]) are of the form MΛ(I) where I are the ideals of K[x, x−1], the map I 7−→ MΛ(I) is an isomorphism of the ideal lattices of K[x, x−1] and MΛ(K[x, x−1]) and, further, MΛ(I · J) = MΛ(I)· MΛ(J) for any two ideals of K[x, x−1]. Since the cancellation for non-zero ideals hold in the principal ideal domain K[x, x−1], we conclude that the cancellation property holds for non-zero ideals in M . Now the graded ideal M possesses local units, being isomorphic to a Leavitt path algebra of a suitable graph (see [11]). From this, it easy to show that the ideals of M are also the ideals of L. Since M contains every other ideal A of L and is graded (so M A = M ∩ A), we conclude that M is also cancellative. Thus all the non-zero ideals of L have the cancellation property. This proves (i). Now (iv)(b)⇒ (i) is immediate since L has at most two distinct non-zero proper ideals which are graded and thus the cancellation property holds trivially in L. We conclude the paper by illustrating some examples of the graphs that satisfy the conditions of part (iii) of Theorem 14. Example 15. Consider the graph E1 •u / •v c which satisfies the conditions of part (iii) a). So, H = {v} is the only non- zero proper hereditary saturated subset and M = hHi ∼= K[x, x−1] contains every proper ideal of LK(E1). 16 8 8 / f f Consider the graph R2 •u which satisfies the conditions of part (iii) b) and there are no non-zero proper hereditary saturated subsets. Actually LK(R2) is the simple Leavitt algebra of type (1,2). Consider the graph E3 •u / •v proper hereditary saturated subset. which satisfies the conditions of part (iii) b) and H = {v} is the only non-zero Consider the graph F •w •u / •v which satisfies the conditions of part (iii) b) and H1 = {v} and H2 = {w} are the only two non-zero proper hereditary saturated subsets. Acknowledgement The first three authors are deeply thankful to Nesin Math Village, S¸irince, Izmir for providing an excellent research environment where this work has been developed. References [1] G. Abrams, P. Ara and M. Siles Molina, Leavitt path algebras, Lecture Notes in Mathematics, vol. 2191, Spinger (2017). [2] F.W. Anderson and K.R. Fuller, Rings and categories of modules, Graduate Texts in Mathematics, vol. 13, Springer-Verlag, NewYork - Heidelberg - Berlin (1974). [3] S. Esin, M. Kanuni and K.M. Rangaswamy, On intersections of two-sided ideals of Leavitt path algebras, J. Pure and Applied Algebra, vol. 221 (2017), 632 - 644. [4] T.Y. Lam, A First Course in Noncommutative Rings, Grad. Texts in Math., vol.131, Springer-Verlag, (2001). [5] M.D. Larsen, P.J. McCarthy, Multiplicative Theory of Ideals, Academic Press, New York-London,1971. 17 8 8 f f Z Z / Z Z o o Z Z / Z Z [6] C. Nastasescu, F. van Oystaeyen, Graded Ring Theory, North-Holland, Am- sterdam, (1982). [7] K.M. Rangaswamy, The theory of prime ideals of Leavitt path algebras over arbitrary graphs, J. Algebra 375 (2013), 73 - 96. [8] K.M. Rangaswamy, On generators of two-sided ideals of Leavitt path alge- bras over arbitrary graphs, Communications in Algebra 42 (2014), 2859 - 2868. [9] K.M. Rangaswamy, The multiplicative ideal theory of Leavitt path algebras, Journal of Algebra 487 (2017) 173 - 199. [10] M. Tomforde, Uniqueness theorems and ideal structure of Leavitt path algebras, J. Algebra 318 (2007) 270 - 299. [11] E. Ruiz, M. Tomforde, Ideals in graph algebras, Algebr. Represent. Theory 17 (2014) 849-861. [12] O. Zariski, P. Samuel, Commutative algebra, vol. 1, van Nostrand Company (1969). Songul Esin 19 Mayıs mah. No:10A/25 Kadıkoy, Istanbul, Turkey. [email protected] Muge Kanuni Department of Mathematics, Duzce University, Konuralp 81620 Duzce, Turkey. [email protected] Ayten Ko¸c Department of Mathematics and Computer Science, Istanbul Kultur University, Istanbul, Turkey. [email protected] Katherine Radler Department of Mathematics, Saint Louis University, St. Louis, Missouri, 63103, USA. [email protected] Kulumani M. Rangaswamy Department of Mathematics, University of Colorado. Colorado Springs, CO 80918, USA. [email protected] 18
1604.06676	2	1604	2017-06-06T13:03:20	On free Gelfand-Dorfman-Novikov-Poisson algebras and a PBW theorem	[ "math.RA" ]	In 1997, X. Xu \cite{Xiaoping Xu Poisson} invented a concept of Novikov-Poisson algebras (we call them Gelfand-Dorfman-Novikov-Poisson (GDN-Poisson) algebras). We construct a linear basis of a free GDN-Poisson algebra. We define a notion of a special GDN-Poisson admissible algebra, based on X. Xu's definition and an S.I. Gelfand's observation (see \cite{Gelfand}). It is a differential algebra with two commutative associative products and some extra identities. We prove that any GDN-Poisson algebra is embeddable into its universal enveloping special GDN-Poisson admissible algebra. Also we prove that any GDN-Poisson algebra with the identity $x\circ(y\cdot z)=(x\circ y )\cdot z +(x\circ z) \cdot y$ is isomorphic to a commutative associative differential algebra.	math.RA	math	On free Gelfand-Dorfman-Novikov-Poisson algebras and a PBW theorem∗ L. A. Bokut† School of Mathematical Sciences, South China Normal University Guangzhou 510631, P. R. China Sobolev Institute of mathematics, Novosibirsk, 630090 Novosibirsk State University, Novosibirsk 630090, Russia [email protected] Yuqun Chen‡ and Zerui Zhang§ School of Mathematical Sciences, South China Normal University Guangzhou 510631, P. R. China [email protected] [email protected] Abstract: In 1997, X. Xu [18, 19] invented a concept of Novikov-Poisson algebras (we call them Gelfand-Dorfman-Novikov-Poisson (GDN-Poisson) algebras). We construct a linear basis of a free GDN-Poisson algebra. We define a notion of a special GDN-Poisson admissible algebra, based on X. Xu's definition and an S.I. Gelfand's observation (see [9]). It is a differential algebra with two commutative associative products and some ∗Supported by the NNSF of China (11171118, 11571121). †Supported by Russian Science Foundation (project 14-21-00065). ‡Corresponding author. §Supported by the Innovation Project of Graduate School of South China Normal University. 1 extra identities. We prove that any GDN-Poisson algebra is embeddable into its universal enveloping special GDN-Poisson admissible algebra. Also we prove that any GDN-Poisson algebra with the identity x ◦ (y · z) = (x ◦ y) · z + (x ◦ z) · y is isomorphic to a commutative associative differential algebra. Key words: Poincar´e-Birkhoff-Witt theorem; GDN-Poisson algebra; special GDN- Poisson admissible algebra. AMS Mathematics Subject Classification (2000): 16S10, 17A30, 17A50. 1 Introduction I.M. Gelfand and I.Ya. Dorfman [9], in connection with Hamiltonian operators in the formal calculus of variations, invented a new class of nonassociative algebras, defined by identities x ◦ (y ◦ z) − (x ◦ y) ◦ z = y ◦ (x ◦ z) − (y ◦ x) ◦ z (left symmetry), (x ◦ y) ◦ z = (x ◦ z) ◦ y (right commutativity) (1) S.I. Gelfand (see [9]) introduced an important subclass of new algebras (1). Namely, any associative commutative differential algebra (C, ·) is an algebra (1) under a new multiplication x ◦ y = x(Dy). Independently, S.P. Novikov ([3, 11]) invented the same algebras in connection with linear Poisson brackets of hydrodynamic type. J.M. Osborn [12 -- 15] gave the name Novikov algebra (he knew both papers [3, 9]) to this kind of algebras and began to classify simple Novikov algebras as well as irreducible modules. Considering the contributions of Gelfand and Dorfman to Novikov algebras, in this paper we call Novikov algebras as Gelfand-Dorfman-Novikov algebras (GDN algebras for short), Novikov-Poisson algebras as Gelfand-Dorfman-Novikov-Poisson algebras (GDN- Poisson algebras for short). As it was pointed out in [11], E.I. Zelmanov answered to a Novikov's question about simple finite dimensional GDN algebras over a field of characteristic 0. He proved that there are no such algebras, see [23]. V.T. Filippov [8] found first examples of simple infinite dimensional GDN algebras of characteristic p ≥ 0 and simple finite dimensional 2 GDN algebras of characteristic p > 0. J.M. Osborn and E.I. Zelmanov [16] classified sim- ple GDN algebras A over an algebraically closed field of characteristic 0 with a maximal subalgebra H such that A/H has a finite dimensional irreducible H-submodule (V. Gulli- men's type of condition) (up to isomorphisms of B. Weisfeiler associated graded algebras). A. Dzhumadil'daev and C. Lofwall [7] constructed a linear basis of a free GDN algebra using trees and a free commutative associative differential algebra. There were also quite a few papers on the structure theory (for example, X. Xu [17 -- 19], C. Bai and D. Meng [1, 2], L. Chen, Y. Niu and D. Meng [6], D. Burde and K. Dekimpe [5]) and combinatorial theory of GDN algebras, and irreducible modules over GDN algebras, with applications to mathematics and mathematical physics. In [18 -- 20], X. Xu introduced a concept of GDN-Poisson algebras in order to study the tensor theory of GDN algebras. Then he classified GDN-Poisson algebras whose GDN algebras are simple with an idempotent element. Moreover, he showed that a class of simple GDN algebras without idempotent elements can be constructed through GDN-Poisson algebras. Certain new simple Lie superalgebras induced by GDN-Poisson algebras are also introduced by him. X. Xu connected certain GDN-Poisson algebras with Hamiltonian superoperators and proved that they induce Lie superalgebras, which are natural generalizations of the super-Virasoro algebras in special cases. Since then, there have been several papers on GDN-Poisson algebras, see [22, 24]. There are also papers about embedding on GDN-Poisson algebras, for example [21]; papers on GDN-Poisson algebras and associative commutative derivation algebras [25]. The paper is organized as follows. In section 2 we first introduce the concept of a special GDN-Poisson admissible algebra and then construct a linear basis of a free GDN- Poisson algebra over an arbitrary field. In section 3 we prove a Poincar´e-Birkhoff-Witt (PBW for short) theorem: any GDN-Poisson algebra is embeddable into its universal enveloping special GDN-Poisson admissible algebra. In section 4 we show that any GDN- Poisson algebra (with unit e with respective to ·) satisfying the identity x ◦ (y · z) = (x ◦ y) · z + (x ◦ z) · y is isomorphic to a commutative associative differential algebra both as GDN-Poisson algebra and as commutative associative differential algebra. 3 2 Free GDN-Poisson algebras A GDN algebra ([3]) A is a vector space with a bilinear operation ◦ satisfying the two identities x ◦ (y ◦ z) − (x ◦ y) ◦ z = y ◦ (x ◦ z) − (y ◦ x) ◦ z (left symmetry), (x ◦ y) ◦ z = (x ◦ z) ◦ y (right commutativity). A GDN-Poisson algebra ([18]) A is a vector space with two bilinear operations " · " and " ◦ " such that (A, ·) forms a commutative associative algebra and (A, ◦) forms a GDN algebra with the compatibility conditions: (x · y) ◦ z = x · (y ◦ z), (x ◦ y) · z − x ◦ (y · z) = (y ◦ x) · z − y ◦ (x · z), x, y, z ∈ A. Throughout the paper, we only consider GDN-Poisson algebras with unit e with re- spect to ·. As usual, let F (Ω, X) be a free Ω-algebra generated by X, where Ω is a given set of operator symbols and X is a non-vacuous set. The set of words in F (Ω, X) is denoted by W (Ω, X) (for definitions of Ω-algebra and words, see [4] and Chapter 2 in [10]). For example, let GDNP (X) be a free GDN-Poisson algebra over a field k generated by X. Then (i) x, e are words in GDNP (X) for any x ∈ X; (ii) If u, v are words in GDNP (X), then ·uv and ◦uv are words in GDNP (X). For convenience, we will write (uδv) instead of δuv if δ is a binary operator. Moreover, we always omit the leftmost "(" and the rightmost ")" of a word. For example, ((x1 ◦ x1) · (x1 ◦ e)) ◦ x2 is a word in GDNP (X). For any words w1, w2, . . . , wn ∈ W (Ω, X) and binary operators δ1, δ2, . . . , δn−1 ∈ Ω, define [w1δ1w2δ2 · · · δn−1wn]L := (. . . ((w1δ1w2)δ2w3)δ3 · · · δn−2wn−1)δn−1wn (left-normed bracketing) and [w1δ1w2δ2 · · · δn−1wn]R := w1δ1(w2δ2 · · · δn−3(wn−2δn−2(wn−1δn−1wn)) . . . ) (right-normed bracketing). 4 For any word T in GDNP (X), define T ◦ := the number of " ◦ " that appears in T, T X := the number of elements in X that appears in T . For example, let T = (x1 ◦ x1) · (x1 ◦ e). Then T X = 3, T ◦ = 2. We call w a GDN tableau ([7]) over a well-ordered set X, if w = (. . . ((a1,r1+1 ◦ A1) ◦ A2) ◦ · · · ◦ An) (left-normed bracketing), where Ai = (ai,ri ◦ · · · ◦ (ai,3 ◦ (ai,2 ◦ ai,1)) . . . ) (right-normed bracketing), 1 ≤ i ≤ n, ai,j ∈ X, satisfying that (i) ri ≥ ri+1, (ii) ai,1 ≥ ai+1,1 if ri = ri+1, i ≥ 1, (iii) a1,r1+1 ≥ a1,r1 ≥ · · · ≥ a1,3 ≥ a1,2 ≥ a2,r2 ≥ · · · ≥ a2,3 ≥ a2,2 ≥ an,rn ≥ · · · ≥ an,3 ≥ an,2. Let [X] be the commutative monoid generated by X with unit e. We call T = u · w a GDN-Poisson tableau over X, if u = b1 · · · bm ∈ [X] (each bi ∈ X) and w = (· · · ((a1,r1+1 ◦ A1) ◦ A2) ◦ · · · ◦ An) is a GDN tableau over X ∪ {e} (e < x for any x ∈ X) satisfying that (i) an,2 ≥ b1 ≥ · · · ≥ bm, (ii) if an,2 = e, then m = 0, i.e., T = w. We call [a1 ◦ a2 ◦ · · · ◦ an]R a row in GDNP (X) if ai ∈ [X], for all 1 ≤ i ≤ n. We will prove that the set of the GDN-Poisson tableaux over X is a linear basis of GDNP (X). Unless otherwise stated, throughout the paper, we use a, b, c, . . . to denote elements in X ∪ {e}, α, β, γ, . . . elements in the field k, u, v, . . . words in the commutative associative monoid [X], T word in GDNP (X), m, n, p, q, t, l, i, j, . . . nonnegative integer numbers. Moreover, we may omit ·, if · is a commutative and associative bilinear operation. For example, T1 · T2 · T3 (with any bracketing) in GDNP (X) is simply denoted by T1T2T3. Definition 2.1. A special GDN-Poisson admissible algebra (A, ·, ∗, D) is a vector space with bilinear operations ·, ∗ and a linear operation D such that (A, ·) forms a commuta- tive associative algebra with unit e, (A, ∗) forms a commutative associative algebra and "·, ∗, D" are compatible in the sense that the following identities hold: (x · y) ∗ z = x · (y ∗ z), D(x ∗ y) = (Dx) ∗ y + x ∗ (Dy), D(x · y) = (Dx) · y + x · (Dy) − x · y · (De). 5 Let (A, ·, ∗, D) be a special GDN-Poisson admissible algebra. Then for any x, y, z ∈ A, we have (x · y) ∗ z = x · (y ∗ z), (x ∗ y) · z = (y ∗ x) · z = z · (y ∗ x) = (z · y) ∗ x = x ∗ (z · y) = x ∗ (y · z). Therefore, for any y1, . . . , yn ∈ A, δ1, . . . , δn−1 ∈ {·, ∗}, we have [y1δ1y2δ2 · · · δn−1yn]L = (y1δ1y2δ2 · · · δn−1yn) (with any bracketing), and thus y1δ1y2δ2 · · · δn−1yn makes sense. Lemma 2.2. Let (A, ·, ∗, D) be a special GDN-Poisson admissible algebra. The operation ◦ on A is defined by x ◦ y := x ∗ Dy, x, y ∈ A. Then (A, ·, ◦) forms a GDN-Poisson algebra. Proof. The proof is straightforward. For example, (x ◦ y) · z − x ◦ (y · z) = (x ∗ Dy) · z − x ∗ (Dy · z) − x ∗ (y · Dz) + x ∗ (y · z · De) = −x ∗ y · Dz + x ∗ y · z · De = (y ◦ x) · z − y ◦ (x · z). The other identities can be proved similarly. Define D0a := a, a ∈ X ∪ {e}. Let Y = {Dia, ·, ∗ i ∈ Z≥0, a ∈ X ∪ {e}} and Y + be the free semigroup generated by Y , C[X] := {Di1a1 ∗ · · · ∗ Dij aj · Dij+1aj+1 · Dij+2aj+2 · · · Dinan ∈ Y + n ≥ 1, 1 ≤ j ≤ n, (i1, a1) ≥ · · · ≥ (in, an) lexicographically, and if j < n, then (in, an) 6= (0, e)}, where Z≥0 is the set of nonnegative integer numbers. For any w1 = Di1a1 ∗ · · · ∗ Dinan · Din+1an+1 · · · Din+man+m, w2 = Dj1b1 ∗ · · · ∗ Djpbp · Djp+1bp+1 · · · Djp+qbp+q ∈ C[X], define w1 = w2 if and only if n = p, m = q, (il, al) = (jl, bl) for all 1 ≤ l ≤ n + m. Let kC[X] be the linear space with a k-basis C[X] and w1, w2 be as above. Let Dl1c1, . . . , Dln+m+p+qcn+m+p+q be a reordering of Di1a1, . . . , Din+man+m, Dj1b1, . . . , Djp+qbp+q, 6 such that (l1, c1) ≥ · · · ≥ (ln+m+p+q, cn+m+p+q) lexicographically. For convenience, we de- fine (ln+m+p+q+1, cn+m+p+q+1) = (0, e). We define the bilinear operations ·, ∗ and the linear operation D on kC[X] as follows: (i) w1 · w2 := Dl1c1 ∗ · · · ∗ Dln+p−1cn+p−1 · Dln+pcn+p · · · Dltct, where t = n + p − 1 if (ln+p, cn+p) = (0, e); otherwise, n+p ≤ t ≤ n+m+p+q, (lt, ct) > (0, e) = (lt+1, ct+1). (ii) w1 ∗ w2 := Dl1c1 ∗ · · · ∗ Dln+pcn+p · Dln+p+1cn+p+1 · · · Dltct, where t = n + p if (ln+p+1, cn+p+1) = (0, e); otherwise, n + p + 1 ≤ t ≤ n + m + p + q, (lt, ct) > (0, e) = (lt+1, ct+1). (iii) D(w1) := P1≤t≤n+m[Di1a1δ1Di2a2δ2 · · · δt−1Dit+1atδt · · · δn+m−1Din+man+m]L−mw1· De, where δ1 = · · · = δn−1 = ∗ and δn = · · · = δn+m−1 = ·. With notations as above, we have Lemma 2.3. (kC[X], ·, ∗, D) is a special GDN-Poisson admissible algebra. Proof. By definitions of · and ∗, we have [Di1a1∗· · ·∗Dinan·Din+1an+1 · · · Din+man+m·e]L = w1 = w1·e = e·w1 and w1·w2 = [Dl1c1∗· · ·∗Dln+p−1cn+p−1·Dln+pcn+p · · · Dln+m+p+qcn+m+p+q]L. It follows immediately that · is associative and commutative. Similarly, ∗ is associative and commutative. Moreover, the identity x · (y ∗ z) = (x · y) ∗ z follows by the above arguments. Let Din+m+1an+m+1 = Din+m+2an+m+2 = · · · = Din+m+lan+m+l = e. Then D([w1 · Din+m+1an+m+1 · Din+m+2an+m+2 · · · Din+m+lan+m+l]L) = D(w1) = [D(w1) · Din+m+1an+m+1 · Din+m+2an+m+2 · · · Din+m+lan+m+l]L + X n+m+1≤t≤n+m+l [w1 · Din+m+1an+m+1 · · · Dit+1at · · · Din+m+lan+m+l]L − lw1 · De = X 1≤t≤n+m+l [Di1a1δ1 · · · δt−1Dit+1atδt · · · δn+m+l−1Din+m+lan+m+l]L −(m + l)[w1 · Din+m+1an+m+1 · Din+m+2an+m+2 · · · Din+m+lan+m+l]L · De, 7 where δ1 = · · · = δn−1 = ∗ and δn = · · · = δn+m+l−1 = ·. Therefore, D(w1 · w2) = D([Dl1c1 ∗ · · · ∗ Dln+p−1cn+p−1 · Dln+pcn+p · · · Dln+m+p+qcn+m+p+q]L) = X [Dl1c1δ1Dl2c2δ2 · · · δt−1Dlt+1ctδt · · · δn+m+p+q−1Dln+m+p+qcn+m+p+q]L 1≤t≤n+m+p+q −(m + q + 1)w1 · w2 · De = X [Di1a1η1Di2a2η2 · · · ηt−1Dit+1atηt · · · ηn+m−1Din+man+m]L · w2 − mw1 · De · w2 1≤t≤n+m + X 1≤t≤p+q [Dj1b1ν1Dj2b2ν2 · · · νt−1Djt+1btνt · · · νp+q−1Djp+qbp+q]L · w1 − qw2 · De · w1 −w1 · w2 · De = D(w1) · w2 + D(w2) · w1 − w1 · w2 · De, where δ1 = · · · = δn+p−2 = η1 = · · · = ηn−1 = ν1 = · · · = νp−1 = ∗ , δn+p−1 = · · · = δn+m+p+q−1 = ηn = · · · = ηn+m−1 = νp = · · · = νp+q−1 = ·. D(w1 ∗ w2) = D([Dl1c1 ∗ · · · ∗ Dln+pcn+p · Dln+p+1cn+p+1 · · · Dln+m+p+qcn+m+p+q]L) = X [Dl1c1δ1Dl2c2δ2 · · · δt−1Dlt+1ctδt · · · δn+m+p+q−1Dln+m+p+qcn+m+p+q]L 1≤t≤n+m+p+q −(m + q)w1 · w2 · De, = X [Di1a1η1Di2a2η2 · · · ηt−1Dit+1atηt · · · ηn+m−1Din+man+m]L ∗ w2 − mw1 · De ∗ w2 1≤t≤n+m + X 1≤t≤p+q [Dj1b1ν1Dj2b2ν2 · · · νt−1Djt+1btνt · · · νp+q−1Djp+qbp+q]L ∗ w1 − qw2 · De ∗ w1 = D(w1) ∗ w2 + D(w2) ∗ w1, where δ1 = · · · = δn+p−1 = η1 = · · · = ηn−1 = ν1 = · · · = νp−1 = ∗ , δn+p = · · · = δn+m+p+q−1 = ηn = · · · = ηn+m−1 = νp = · · · = νp+q−1 = ·. It follows that (kC[X], ·, ∗, D) is a special GDN-Poisson admissible algebra. Let C(X) be a free special GDN-Poisson admissible algebra generated by X and C[X]′ := {[w]L w ∈ C[X]}. Then we have Lemma 2.4. The set C[X]′ is a linear generating set of C(X). 8 Proof. Given a word w ∈ C(X), we can apply D(x ∗ y) = Dx ∗ y + x ∗ Dy and D(x·y) = Dx·y +x·Dy −x·y ·De (if any) to rewrite w as a linear combination of words of the form (Di1a1δ1Di2a2δ2 · · · δn−1Dinan) (with some bracketing), where n ≥ 1, a1, . . . , an ∈ X ∪{e}, i1, . . . in ∈ Z≥0. Noting that for any x, y, z ∈ C(X), (x·y)∗z = x·(y∗z) and (x∗y)· z = x ∗ (y · z), we have [Di1a1δ1Di2a2δ2 · · · δn−1Dinan]L = (Di1a1δ1Di2a2δ2 · · · δn−1Dinan) (with any bracketing). Since [x · y ∗ z]L = (x · y) ∗ z = z ∗ (x · y) = (z ∗ x) · y = (x ∗ z) · y = x ∗ (z · y) = x ∗ (y · z) = (x ∗ y) · z = [x ∗ y · z]L, we may assume that, in [Di1a1δ1Di2a2δ2 · · · δn−1Dinan]L, δ1 = δ2 = · · · = δl = ∗, δl+1 = δl+2 = · · · = δn−1 = · for some 0 ≤ l ≤ n − 1. Finally, by applying the commutativity of · and ∗, we get that any element of C(X) can be written as a linear combination of elements of C[X]′. Lemma 2.5. The algebra (kC[X], ·, ∗, D) is isomorphic to C(X). Proof. Define a special GDN-Poisson admissible algebra homomorphism η : C(X) −→ kC[X] induced by η(a) = a for any a ∈ X. Since C[X]′ is a linear generating set of C(X) and η([w]L) = w for any w ∈ C[X] and C[X] is a linear basis of kC[X], we have that C[X]′ is a linear basis of C(X). It follows that η is an isomorphism. From now on, we identify C[X]′ with C[X]. Let w = Di1a1 ∗ · · · ∗ Dij aj · Dij+1aj+1 · Dij+2aj+2 · · · Dinan ∈ C[X]. Define wt(w) := ( X w∗ := j − 1, ord(w) := (w∗, i1, a1, . . . , in, an, −1). it) − (j − 1), 1≤t≤n We order the set of C[X] as follows: w1 < w2 ⇔ ord(w1) < ord(w2) lexicographically. The proof of the following lemma is straightforward. Lemma 2.6. (i) For any w1, w2, w3 ∈ C[X], we have w1 < w2 ⇒ w1 ∗ w3 < w2 ∗ w3, w1 · w3 < w2 · w3. (ii) Let w = Di1a1 ∗ · · · ∗ Dij aj · Dij+1aj+1 · Dij+2aj+2 · · · Dinan. If (i1, a1) > (i2, a2), then D(w) = Di1+1a1 ∗ · · · ∗ Dij aj · Dij+1aj+1 · Dij+2aj+2 · · · Dinan, where f is the leading word of f ∈ kC[X]. 9 Define GDNP0(X) := spank{w ∈ C[X] wt(w) = 0}, which is the subspace of C(X) spanned by the set {w ∈ C[X] wt(w) = 0}. Then it is clear that (GDNP0(X), ·, ◦) is a GDN-Poisson subalgebra of (C(X), ·, ◦). We will prove that GDNP0(X) is a free GDN-Poisson algebra generated by X, i.e., GDNP (X) ∼= GDNP0(X), see Lemma 3.1. Definition 2.7. Let Ω = {·, ◦}, where ·, ◦ are binary operator symbols. A root function from the set of words of a free Ω-algebra F (Ω, X ∪ {e}) to Z≥0 is defined as follows: for any words T1, T2, (i) r(a) = 0 for any a ∈ X ∪ {e}; (ii) r(T1 · T2) = r(T1) + r(T2); (iii) r(T1 ◦ T2) = r(T1) + 1 if T2◦ = 0, and r(T1 ◦ T2) = r(T1) + r(T2) if T2◦ ≥ 1. We call r(T ) the root number of T . It is clear that r(T1 ◦ T2) = r(T1) + max{1, r(T2)}. For any word T ∈ GDNP (X), we can take T as a word in F (Ω, X ∪ {e}). Then r(T ) also makes sense. Example 2.8. For any ui ∈ [X], i ≥ 1, n ≥ 1, we have r([u1 ◦ u2 ◦ · · · ◦ un]L) = n − 1. By definition, we have the following lemmas. Lemma 2.9. For any words x, y, z ∈ F (Ω, X ∪ {e}), we have r((x ◦ y) ◦ z) = r((x ◦ z) ◦ y) and r((x · y) ◦ z) = r(x · (y ◦ z)). Lemma 2.10. For any words T1, T2, T ∈ F (Ω, X ∪ {e}), if r(T1) > r(T2), then we have (i) r(T1 ◦ T ) > r(T2 ◦ T ); (ii) r(T ◦ T1) > r(T ◦ T2) if r(T1) > 1, r(T ◦ T1) = r(T ◦ T2) if r(T1) = 1. Lemma 2.11. For any word T ∈ GDNP (X), if T ◦ ≥ 1, then T = T1 ◦ T2 for some words T1, T2 ∈ GDNP (X). 10 Proof. If T = T1 ◦ T2, the result follows. Otherwise, we may assume that T = T1T2 · · · Tl, where T1◦ ≥ · · · ≥ Tl◦ and T1 = T11 ◦ T12. Then T = (T11T2 · · · Tl) ◦ T12. Lemma 2.12. For any word T ∈ GDNP (X) with T ◦ = n − 1, we have r(T ) ≤ T ◦ and the equality holds if and only if T = [u1 ◦ u2 · · · ◦ un]L for some u1, u2, . . . , un ∈ [X]. Proof. By Example 2.8, we just need to prove that (i) r(T ) ≤ T ◦, (ii) if r(T ) = T ◦, then T = [u1 ◦ u2 · · · ◦ un]L, where each ui ∈ [X]. Induction on T ◦. If T ◦ ≤ 1, the result follows by Lemma 2.11. Let T ◦ = n > 1 and T = T1 ◦ T2. If T2◦ ≥ 1, then by induction we have r(T ) = r(T1) + r(T2) ≤ T1◦ + T2◦ = T ◦ − 1 < T ◦. If T2◦ = 0, then by induction we have r(T ) = r(T1) + 1 ≤ T1◦ + 1 = T ◦. Moreover, r(T ) = T ◦ implies that T2 = un and T1 = [u1 ◦ u2 ◦ · · · ◦ un−1]L by induction. For any words T, T ′ ∈ GDNP (X), we define T → T ′ if T = T ′ + X αiTi 1≤i≤n for some n ∈ Z≥0, words Ti ∈ GDNP (X), αi ∈ k, such that for any 1 ≤ i ≤ n, T ◦ = T ′◦ = Ti◦, T X = T ′X = TiX , r(T ) = r(T ′) < r(Ti). It is clear that T → T ′ → T ′′ ⇒ T → T ′′. For any A = [ur ◦ · · · ◦ u2 ◦ u1]R ∈ GDNP (X), u1, . . . , ur, vj ∈ [X], we define Aui := [ur ◦ · · · ◦ ui+1 ◦ ui−1 ◦ · · · ◦ u2 ◦ u1]R, Aui7→vj := [ur ◦ · · · ◦ ui+1 ◦ vj ◦ ui−1 ◦ · · · ◦ u2 ◦ u1]R, Aui↔uj := [ur ◦ · · · ◦ ui+1 ◦ uj ◦ ui−1 ◦ · · · ◦ uj+1 ◦ ui ◦ uj−1 ◦ · · · ◦ u1]R. Lemma 2.13. For any A = [ur ◦ · · · ◦ u2 ◦ u1]R ∈ GDNP (X), r ≥ 3, u1, . . . , ur ∈ [X], we have A → Aui↔uj and A → uj ◦ Auj for any r ≥ i > j ≥ 2. Proof. Since A = [ur ◦ · · · ◦ (uj+1 ◦ uj) ◦ uj−1 ◦ · · · ◦ u1]R + [ur ◦ · · · ◦ uj ◦ uj+1 ◦ uj−1 ◦ · · · ◦ u1]R −[ur ◦ · · · ◦ (uj ◦ uj+1) ◦ uj−1 ◦ · · · ◦ u1]R, 11 we have A → [ur ◦ · · · ◦ uj ◦ uj+1 ◦ uj−1 ◦ · · · ◦ u1]R. By a similar argument, we have A → [ur ◦ · · · ◦ uj ◦ uj+1 ◦ uj−1 ◦ · · · ◦ u1]R → · · · → [ur ◦ · · · ◦ ui+1 ◦ uj ◦ ui ◦ ui−1 ◦ · · · uj+1 ◦ uj−1 ◦ · · · ◦ u1]R → [ur ◦ · · · ◦ ui+1 ◦ uj ◦ ui−1 ◦ ui ◦ · · · uj+1 ◦ uj−1 ◦ · · · ◦ u1]R → · · · → [ur ◦ · · · ◦ ui+1 ◦ uj ◦ ui−1 ◦ · · · uj+1 ◦ ui ◦ uj−1 ◦ · · · ◦ u1]R. Similarly, we have A → uj ◦ Auj for any j ≥ 2. Lemma 2.14. Let A = [ur+1 ◦ · · · ◦ u2 ◦ u1]R, B = [vm ◦ · · · ◦ v2 ◦ v1]R ∈ GDNP (X). Then A ◦ B → Aui7→vj ◦ Bvj 7→ui. Moreover, let T = b1 · · · bm · [u1,r1+1 ◦ A1 ◦ A2 ◦ · · · ◦ An]L, where each Ai = [ui,ri ◦ · · · ◦ ui,2 ◦ ui,1]R, 1 ≤ i ≤ n. Then T → (T )ui,j↔up,l, j, l ≥ 2 and T → (T )bt↔ui,j , j ≥ 2, 1 ≤ t ≤ m. Proof. By Lemma 2.13, we get A◦B → (ui◦Aui)◦B → (ui◦B)◦Aui → (vj◦Bvj 7→ui)◦Aui → (vj◦Aui)◦Bvj 7→ui → Aui7→vj ◦Bvj 7→ui. Since T = b1 · · · bm · [u1,r1+1 ◦ A1 ◦ A2 ◦ · · · ◦ An]L = b1 · · · bm · [u1,r1+1 ◦ Ai ◦ Ap ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ Ap ◦ · · · ◦ An]L and T = b1 · · · bt · · · bmu1,r1+1 · [bt ◦ A1 ◦ A2 ◦ · · · ◦ An]L (if (i, j) = (1, r1 + 1), we are done.) → b1 · · · bt · · · bmu1,r1+1 · [bt ◦ Ai ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L → b1 · · · bt · · · bmu1,r1+1 · [ui,j ◦ (Ai)ui,j 7→bt ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L → b1 · · · bt · · · bmui,,j · [u1,r1+1 ◦ (Ai)ui,j 7→bt ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L → b1 · · · bt · · · bmui,,j · [u1,r1+1 ◦ A1 ◦ · · · ◦ (Ai)ui,j 7→bt ◦ · · · ◦ An]L, the result follows immediately. Define the deg-lex order on [X] as follows: for any u = x1 · · · xn, v = y1 · · · ym ∈ [X], x1 ≥ · · · ≥ xn, y1 ≥ · · · ≥ ym, where each xi, yj ∈ X, u > v if and only if n > m, or n = m and for some p ≤ n, xp > yp, xi = yi for all 1 ≤ i ≤ p − 1. 12 Lemma 2.15. For any word T ∈ GDNP (X), we have T = P αlTl, where each Tl is of the following form: Tl = [u1,r1+1◦A1◦A2◦· · ·◦An]L, where each Ai = [ui,ri ◦· · ·◦ui,2◦ui,1]R, 1 ≤ i ≤ n, ui,j ∈ [X], such that the following conditions hold: (i) r(Tl) ≥ r(T ), Tl◦ = T ◦, TlX = T X ; (ii) r1 ≥ · · · ≥ rn; (iii) If ri = ri+1, then ui,1 ≥ ui+1,1 by deg-lex order on [X]; (iv) u1,r1+1 ≥ · · · ≥ u1,2 ≥ u2,r2 ≥ · · · ≥ u2,2 ≥ · · · ≥ un,rn ≥ · · · ≥ un,2 by deg-lex order on [X]. Proof. Induction on T ◦. If T ◦ = 0 or 1, then the result follows by Lemma 2.11. Suppose that the result holds for any T with T ◦ < t. Let T ◦ = t ≥ 2. Induction on T ◦ − r(T ). If T ◦ −r(T ) = 0, then r(T ) = t ≥ 2. By Lemma 2.12, we get T = [u1 ◦u2 ◦· · ·◦ut+1]L. By right commutativity, the result follows. If T ◦ − r(T ) ≥ 1, then by Lemma 2.11, we have T = T1 ◦ T2 for some words T1, T2 ∈ GDNP (X). By induction, we have T1 = P αiTi, T2 = P βjTj, where αi, βj ∈ k and Ti, Tj satisfy the claimed conditions. Then r(Ti ◦ Tj) ≥ r(Ti ◦ T2) ≥ r(T1 ◦ T2) = r(T ) by Lemma 2.10. Say Ti = [u ◦ A1 ◦ A2 ◦ · · · ◦ Ap]L, Tj = [v ◦ B1 ◦ B2 ◦ · · · ◦ Bq]L, where p, q ≥ 0, and each Am, Bm′ is a row. If Ti◦ ≥ 1, then p ≥ 1. Induction on q. If q = 0 or 1, then by right commutativity, Lemmas 2.9, 2.13 and 2.14 and induction hypothesis, the result follows. If q > 1, then Ti ◦ Tj = [u ◦ A1 ◦ A2 ◦ · · · ◦ Ap−1 ◦ Tj ◦ Ap]L. By induction, [u ◦ A1 ◦ A2 ◦ · · · ◦ Ap−1 ◦ Tj]L = P α′ l ◦ ≥ 1 and r(Ap) = 1, the result follows by the above argument. l ◦ Ap) ≥ p + q = r(Ti ◦ Tj) ≥ r(T ). Since T ′ l ) ≥ p − 1 + q. So r(T ′ lT ′ l , r(T ′ If Ti◦ = 0, then Tj◦ ≥ 1, q ≥ 1. If q = 1, by Lemma 2.13 and induction hypothesis, the result follows. If q > 1, then Ti ◦ Tj = (Ti ◦ [v ◦ B1 ◦ B2 ◦ · · · ◦ Bq−1]L) ◦ Bq + [v ◦ B1 ◦ B2 ◦ · · · ◦ Bq−1]L ◦ (Ti ◦ Bq) − [v ◦ B1 ◦ B2 ◦ · · · ◦ Bq−1 ◦ Ti]L ◦ Bq. By induction and the above argument for the case Ti◦ ≥ 1, the result follows. 13 Lemma 2.16. For any u ∈ [X], a1, a2, . . . , an ∈ X, n ≥ 1, we have u ◦ (a1 · · · an) = X 1≤i≤n (ua1 · · · ai · · · an) ◦ ai − (n − 1)(ua1 · · · an) ◦ e. Proof. Induction on n. If n = 1, the result follows. If n > 1, then u ◦ (a1 · · · an) = u · (e ◦ (a1 · · · an)) = u · ((e ◦ (a1 · · · an−1)) · an) + u · ((a1 · · · an−1) ◦ (e · an)) − u · (((a1 · · · an−1) ◦ e) · an) = uan ◦ (a1 · · · an−1) + (ua1 · · · an−1) ◦ an − (ua1 · · · an−1an) ◦ e = X (ua1 · · · ai · · · an) ◦ ai − (n − 1)(ua1 · · · an) ◦ e + (ua1 · · · an−1) ◦ an 1≤i≤n−1 = X 1≤i≤n (ua1 · · · ai · · · an) ◦ ai − (n − 1)(ua1 · · · an) ◦ e. Lemma 2.17. For any T = [u1,2 ◦ u1,1 ◦ u2,1 ◦ · · · ◦ un,1]L ∈ GDNP (X) where u1,2, u1,1, . . . , un,1 ∈ [X], n ≥ 1, we have T = P αlTl where each Tl is a GDN-Poisson tableau with r(Tl) = r(T ) = Tl◦ = T ◦, TlX = T X . Proof. We may assume that u1,1 ≥ · · · ≥ un,1 by the deg-lex order on [X] and u1,2 = b1 · · · bm, b1 ≥ · · · ≥ bm, b1, . . . , bm ∈ X, m ≥ 0. Let t = max{i ui,1X > 1, 1 ≤ i ≤ n}. Induction on t. If t = 0 and m ≤ 1, then T is a GDN-Poisson tableau. If t = 0 and m > 1, then T = b2 · · · bm[b1 ◦ u1,1 ◦ u2,1 ◦ · · · ◦ un,1]L is a GDN-Poisson tableau. If t > 1 and u1,1 = a1 · · · aq ∈ [X], then T = [u1,2 ◦ (a1 · · · aq) ◦ u2,1 ◦ · · · ◦ un,1]L = P1≤i≤q[(u1,2a1 · · · ai · · · aq)◦ai ◦u2,1 ◦· · ·◦un,1]L −(q −1)[(u1,2a1 · · · an)◦e◦u2,1 ◦· · ·◦un,1]L. By induction, the result follows. Lemma 2.18. The set of the GDN-Poisson tableaux over X forms a linear generating set of GDNP (X). Proof. Induction on T ◦. If T ◦ = 0, then it is obvious. Suppose that the result holds for any T with T ◦ < t. Let T ◦ = t ≥ 1. Induction on T ◦ − r(T ). If T ◦ − r(T ) = 0, then by Lemmas 2.12 and 2.17, the result follows. 14 Suppose that the result holds for any T with T ◦ < t, or T ◦ = t, T ◦ − r(T ) < p. Let T ◦ = t, T ◦ − r(T ) = p. By Lemma 2.15, we may assume that T = [u1,r1+1 ◦ A1 ◦ A2 ◦ · · · ◦ An]L, where Ai = [ui,ri ◦ · · · ◦ ui,2 ◦ ui,1]R for all 1 ≤ i ≤ n. Suppose ui,1 = c1 · · · cq, q ≥ 2 for some 1 ≤ i ≤ n. If ri = 1, then by Lemma 2.16, we have T = [u1,r1+1 ◦ A1 ◦ A2 ◦ · · · ◦ An]L = [u1,r1+1 ◦ Ai ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L = X 1≤l≤q [(u1,r1+1 · c1 · · · cl · · · cq) ◦ cl)]R ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L −(q − 1)[(u1,r1+1c1 · · · cq) ◦ e)]R ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L. If ri > 1, then by Lemma 2.16, we have T = [u1,r1+1 ◦ A1 ◦ A2 ◦ · · · ◦ An]L = [u1,r1+1 ◦ Ai ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L = X 1≤l≤q [u1,r1+1 ◦ [ui,ri ◦ · · · ◦ ui,3 ◦ (ui,2 · c1 · · · cl · · · cq) ◦ cl)]R ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L −(q − 1)[u1,r1+1 ◦ [ui,ri ◦ · · · ◦ ui,3 ◦ (ui,2c1 · · · cq) ◦ e)]R ◦ A1 ◦ · · · ◦ Ai ◦ · · · ◦ An]L. Repeating this process if there is some other j such that uj,1X > 1. So by induction hypothesis and Lemma 2.14, we may assume that ui,1X ≤ 1 for any 1 ≤ i ≤ n. Induction on T X . If T X = 0, then T is a GDN-Poisson tableau over {e}. Suppose that the result holds for any T with T ◦ < t, or T ◦ = t, T ◦ − r(T ) < p, or T ◦ = t, T ◦ − r(T ) = p, T X < q. Let T ◦ = t, T ◦ − r(T ) = p, T X = q. If u1,r1+1X ≤ 1, then T is already a GDN-Poisson tableau. If u1,r1+1X > 1, say u1,r1+1 = b1 · · · bm, then T = b2 · · · bm · [b1 ◦ A1 ◦ A2 ◦ · · · ◦ An]L. By induction and Lemma 2.14, the result follows. Theorem 2.19. The set of the GDN-Poisson tableaux over X forms a linear basis of GDNP (X). Proof. By Lemma 2.18, it is enough to show that the set of the GDN-Poisson tableaux over X is linear independent. By Lemma 2.2, (C(X), ·, ◦) is a GDN-Poisson algebra. Now we show that the GDN-Poisson algebra homomorphism ϕ : (GDNP (X), ·, ◦) −→ (C(X), ·, ◦) induced by ϕ(a) = a, a ∈ X, is injective. Given any GDN-Poisson tableau T = b1 · · · bm[a1,r1+1 ◦ A1 ◦ · · · ◦ An]L, where Ai = [ai,ri ◦ · · · ◦ ai,2 ◦ ai,1]R, 1 ≤ i ≤ n, by 15 Lemma 2.6, we have ϕ(T ) = Dr1a1,1 ∗ Dr2a2,1 ∗ · · · ∗ Drnan,1 ∗ a1,r1+1 ∗ · · · ∗ a1,2 ∗ a2,r2 ∗ · · · ∗ a2,2 ∗ · · ·∗ an,rn ∗ · · ·∗ an,2 · b1 · · · bm. Suppose that P1≤i≤t αiTi = 0, where each αi ∈ k, Ti is a GDN-Poisson tableau and Ti 6= Tj for any i 6= j, 1 ≤ i, j ≤ t. Then P1≤i≤t αiϕ(Ti) = 0. Since Ti 6= Tj ⇒ ϕ(Ti) 6= ϕ(Tj), we have αi = 0 for any 1 ≤ i ≤ t. 3 A Poincar´e-Birkhoff-Witt theorem For any S ⊆ C(X), let us denote by Id[S] the ideal of C(X) generated by S and C(XS) := C(X)/Id[S] the special GDN-Poisson admissible algebra generated by X with defining relations S. Noting that w ∗ Dts = w ∗ Dts · e = w ∗ e · Dts, we have Id[S] = spank{w · Dts w ∈ C[X], j, t ∈ Z≥0, s ∈ S}. By Lemma 2.2, (C(X), ·, ◦) becomes a GDN-Poisson algebra with respect to f ◦ g = f ∗ Dg, for any f, g ∈ C(X). Recall that GDNP0(X) = spank{w ∈ C[X] wt(w) = 0} and GDNP0(X) is a subalgebra of (C(X), ·, ◦) (as GDN-Poisson algebra). Lemma 3.1. Let ϕ : (GDNP (X), ·, ◦) −→ (GDNP0(X), ·, ◦) be the GDN-Poisson alge- bra homomorphism induced by ϕ(a) = a, a ∈ X. Then ϕ is an isomorphism. Proof. It is clear that ϕ(GDNP (X)) ⊆ GDNP0(X). To show that GDNP (X) ∼= GDNP0(X), we only need to show that for any word w ∈ C[X] with wt(w) = 0, we have w ∈ ϕ(GDNP (X)). Since wt(w) = 0, we have w = Dr1a1,1 ∗ · · · ∗ Drnan,1 ∗ c ∗ a1,r1 ∗ · · · ∗ a1,2 ∗ · · · ∗ an,rn ∗ · · · ∗ an,2 · b1 · · · bm ∈ C[X] for some n ≥ 0, m ≥ 0, ri ≥ 1, c, ai,j ∈ X ∪{e}, bl ∈ X, 1 ≤ i ≤ n, 1 ≤ j ≤ ri, 1 ≤ l ≤ m. Let T = b1 · · · bm[c ◦ A1 ◦ A2 ◦ · · · ◦ An]L, where Ai = [ai,ri ◦ · · · ◦ ai,2 ◦ ai,1]R, 1 ≤ i ≤ n. Then ϕ(T ) = w. Induction on w. If w = e, then w = ϕ(e) ∈ ϕ(GDNP (X)). Suppose that the result holds for any word < w. Since ϕ(T ) = w, by induction we have w−ϕ(T ) ∈ ϕ(GDNP (X)). Thus w ∈ ϕ(GDNP (X)). 16 Lemma 3.2. Let S be a nonempty subset of GDNP0(X) and Id(S) be the ideal of (GDNP0(X), ·, ◦) generated by S. Then Id(S) = spank{w · Dts w ∈ C[X], t ∈ Z≥0, s ∈ S, wt(w · Dts) = 0}. Proof. It is clear that the right part is a GDN-Poisson algebra ideal that contains S. We just need to show that w · Dts ∈ Id(S) whenever wt(w · Dts) = 0. Since wt(w · Dts) = 0, we have w · Dts = Dts ∗ d1 ∗ · · · ∗ dt ∗ Dr1c1 ∗ · · · ∗ Drncn ∗ a1 ∗ · · · ∗ am · b1 · · · bl, where w = Dr1c1 ∗ · · · ∗ Drncn ∗ a1 ∗ · · · ∗ am ∗ d1 ∗ · · · ∗ dt ∗ b1 · b2 · · · bl, n ≥ 0, m ≥ 0, t ≥ 0, m = r1 + · · · + rn − n and rn ≥ rn−1 ≥ · · · ≥ r1 ≥ 1. So the lemma will be clear if we show (i) Dts ∗ d1 ∗ · · · ∗ dt ∈ Id(S) whenever s ∈ S; (ii) f ∗ Drc ∗ a1 ∗ · · · ∗ ar−1 ∈ Id(S) whenever f ∈ Id(S). To prove (i), we use induction on t. If t = 0, it is clear. Suppose that it holds for any t ≤ p. Then Dp+1s ∗ d1 ∗ · · · ∗ dp+1 = dp+1 ∗ Dp+1s ∗ d1 ∗ · · · ∗ dp = dp+1 ◦ (Dps ∗ d1 ∗ · · · ∗ dp) − dp+1 ∗ X 1≤i≤p Dps ∗ d1 ∗ · · · ∗ Ddi ∗ · · · ∗ dp = dp+1 ◦ (Dps ∗ d1 ∗ · · · ∗ dp) − X (Dps ∗ d1 ∗ · · · ∗ di−1 ∗ di+1 · · · dp+1) ◦ di ∈ Id(S). 1≤i≤p To prove (ii), we use induction on r. If r = 1, it is clear. Suppose that it holds for any r ≤ p. Then f ∗ Dp+1c ∗ a1 ∗ · · · ∗ ap = f ◦ (Dpc ∗ a1 ∗ · · · ∗ ap) − f ∗ X Dpc ∗ a1 ∗ · · · ∗ Dai ∗ · · · ∗ ap 1≤i≤p = f ◦ (Dpc ∗ a1 ∗ · · · ∗ ap) − X (f ∗ Dpc ∗ a1 ∗ · · · ∗ ai−1 ∗ ai+1 ∗ · · · ∗ ap) ◦ ai ∈ Id(S). 1≤i≤p So Id(S) = spank{w · Dts w ∈ C[X], t ∈ Z≥0, s ∈ S, wt(w · Dts) = 0}. The following theorem is a Poincar´e-Birkhoff-Witt theorem for GDN-Poisson algebras. 17 Theorem 3.3. Any GDN-Poisson algebra GDNP (XS) can be embedded into its uni- versal enveloping special GDN-Poisson admissible algebra C(XS). Proof. By Lemmas 3.1 and 3.2, we get GDNP (X) ∼= GDNP0(X) and GDNP (X) GDNP0(X) GDNP0(X) GDNP0(X) + Id[ϕS] Id(S) Id(ϕS) GDNP0(X) ∩ Id[ϕS] Id[ϕS] ∼= = ∼= ≤ C(XϕS), where ϕ is defined in Lemma 3.1. 4 Differential GDN-Poisson algebras Definition 4.1. A differential GDN-Poisson algebra (A, ·, ◦) is a GDN-Poisson algebra satisfying the following identity: x ◦ (y · z) = (x ◦ y) · z + (x ◦ z) · y (♦) Let DGDNP (X) denote a free differential GDN-Poisson algebra generated by a well- ordered set X. It is clear that DGDNP (X) = GDNP (X♦) := GDNP (X) Id(♦) , where Id(♦) is the ideal of GDNP (X) generated by (♦). So for any GDN-Poisson al- gebra A generated by X, if A satisfies (♦), then A ∼= DGDNP (XR) for some R ⊆ DGDNP (X). For any x ∈ DGDNP (X), since x ◦ e = x ◦ (e · e) = x ◦ e + x ◦ e, we have x ◦ e = 0. For any word T in F (Ω, X ∪ {e}), the number of e that appears in T will be denoted by T e. For any word T ∈ DGDNP (X), we consider T as a word in F (Ω, X ∪{e}). Then T e also makes sense. For example, let a, b ∈ X, T = ((((e · e) ◦ a) ◦ (a ◦ (b ◦ e))) · e) · e. Then T X = 3, T e = 5. Define [e ◦ e ◦ · · · ◦ a]i := [e ◦ e ◦ · · · ◦ e } {z i times ◦a]R, i ≥ 0, a ∈ X. We call T a normal word if T = [e ◦ e ◦ · · · ◦ a1]i1 · [e ◦ e ◦ · · · ◦ a2]i2 · · · [e ◦ e ◦ · · · ◦ an]in (T = e if n = 0), where a1, . . . , an ∈ X, (i1, a1) ≥ · · · ≥ (in, an), n ∈ Z≥0. 18 Lemma 4.2. For any normal word T ∈ DGDNP (X), we have e ◦ T = Pi αiTi, where each αi ∈ k, Ti is normal with TiX = T X . Proof. Let T = [e ◦ e ◦ · · · ◦ a1]i1 · · · [e ◦ e ◦ · · · ◦ an]in, n ∈ Z≥0. Induction on n. If n = 1, the result follows. Let n = t ≥ 2. Then e ◦ T = (e ◦ [e ◦ e ◦ · · · ◦ a1]i1) · [e ◦ e ◦ · · · ◦ a2]i2 · · · [e ◦ e ◦ · · · ◦ at]it + (e ◦ ([e ◦ e ◦ · · · ◦ a2]i2 · · · [e ◦ e ◦ · · · ◦ at]it)) · [e ◦ e ◦ · · · ◦ a1]i1. The result follows by induction. Lemma 4.3. For any word T ∈ DGDNP (X), we have T = Pi αiTi, where each αi ∈ k and Ti is normal. Proof. Induction on T X . If T X = 0, then it is clear that T = 0 or e. Let T X = n ≥ 1. Now, induction on T e. If T e = 0, then T = T1 · T2 or T = T1 ◦ T2 = (T1 · e) ◦ T2 = T1 · (e◦ T2), where T1X < n, e◦ T2X = T2X < n. By induction hypothesis, the result follows. Suppose that the result holds for any T with T X < n, or T X = n, T e < t. Let T X = n, T e = t ≥ 1, T = T1·T2 or T = T1 ◦ T2 = T1 · (e ◦ T2). If T1X < n and e◦T2X = T2X < n, then the result follows by induction. If T1X = 0 and T2X = n, then T1 = 0 or e, and T2e < t. By induction, we may assume that T2 is normal. Then T1 · T2 = 0 or T1 · T2 = T2; T1 ◦ T2 = 0 or T1 ◦ T2 = e ◦ T2. By Lemma 4.2, the result follows. If T1X = n and T2X = 0, then T2 = 0 or e, and T1e < t. Then T1 · T2 = 0 or T1 · T2 = T1; T1 ◦ T2 = 0. By induction, the result follows. Let k{X} be a free commutative associative differential algebra generated by a well- ordered set X (De = 0) with unit e and one linear derivation D. Then it is clear that the set [DωX] of a linear basis of k{X} consists of T = Dr1a1 · · · Drnan (T = e if n = 0), where a1, . . . , an ∈ X, each ri ≥ 0, (r1, a1) ≥ · · · ≥ (rn, an). Let (A, ·, D) be a commutative associative differential algebra with one derivation D. Define the operation ◦ on A by x ◦ y := xDy, x, y ∈ A. Then (A, ◦) forms a GDN- Poisson algebra ([19]). Moreover, for any x, y, z ∈ A, we have x ◦ (yz) = xD(yz) = x(Dy)z + xyDz = (x ◦ y)z + (x ◦ z)y. 19 Lemma 4.4. The set of normal words forms a linear basis of DGDNP (X). In partic- ular, (DGDNP (X), ·, ◦) ∼= (k{X}, ·, ◦) as GDN-Poisson algebras, where ◦ in k{X} is defined as f ◦ g = f Dg for any f, g ∈ k{X}, and (DGDNP (X), ·, ∂) ∼= (k{X}, ·, D) as commutative associative differential algebras, where ∂ in DGDNP (X) is defined as ∂(f ) = e ◦ f for any f ∈ DGDNP (X). Proof. Let θ : DGDNP (X) −→ k{X} be a GDN-Poisson algebra homomorphism induced by θ(a) = a, a ∈ X. Then it is clear that θ is an isomorphism. So the set θ−1([DωX]) of normal words forms a linear basis of DGDNP (X). It is clear that (DGDNP (X), ·, ∂) forms a commutative associative differential alge- bra. Moreover, θ(∂(uv)) = θ(e◦(uv)) = θe◦θ(uv) = e◦(uv) = D(uv) = D(u)v +uD(v) = θ((e ◦ u)v) + θ((e ◦ v)u) = θ((∂u)v + (∂v)u), so θ is a commutative associative differential algebra isomorphism. Theorem 4.5. Let θ : DGDNP (X) −→ k{X} be a GDN-Poisson algebra homomor- phism induced by θ(a) = a, a ∈ X. Then any GDN-Poisson algebra GDNP (XS) satis- fying (♦) is isomorphic to k{Xθ(S)} both as GDN-Poisson algebras and as commutative associative differential algebras. Proof. Let Id(θ(S)) be the ideal of (k{X}, ·, ◦) as GDN-Poisson algebra and Id[θ(S)] the ideal of (k{X}, ·, D) as commutative associative differential algebra. It is clear that Id[θ(S)] = {P αiuiDri(θsi) αi ∈ k, ui ∈ k{X}, si ∈ S}. Since Drg = [e ◦ e ◦ · · · ◦ e } for any g ∈ k{X}, it is straightforward to show that Id(θ(S)) = Id[θ(S)]. By Lemma 4.4, ◦g]R {z r times we have GDNP (XS) = GDNP (X) Id(S) ∼= θ(DGDNP (X)) Id(θ(S)) = k{X} Id[θ(S)] = k{Xθ(S)}. Corollary 4.6. (♦) holds in a GDN-Poisson algebra A if and only if A is embeddable into a commutative associative differential algebra B with one derivation D, where ◦ is defined as f ◦ g = f Dg for any f, g ∈ B. Proof. If (♦) holds in A, then by Theorem 4.5, A can be embedded into a commuta- tive associative differential algebra. Conversely, assume that θ : A −→ B is an embedding. 20 Then for any x, y, z ∈ A, we have θ(x ◦ (yz)) = θ(x) ◦ (θ(y)θ(z)) = θ(x)D(θ(y)θ(z)) = (θ(x)Dθ(y))θ(z) + (θ(x)Dθ(z))θ(y) = (θ(x) ◦ θ(y))θ(z) + (θ(x) ◦ θ(z))θ(y) = θ((x ◦ y)z + (x ◦ z)y). Since θ is injective, (♦) holds in A. Acknowledgment We would like to thank the referee for valuable suggestions. References [1] C. Bai and D. Meng, The classification of Novikov algebras in low dimensions, J. Phys. A 34(8) (2001) 1581-1594. [2] C. Bai and D. Meng, Transitive Novikov algebras on four-dimensional nilpotent Lie algebras, Internat. J. Theoret. Phys. 40(10) (2001) 1761-1768. [3] A.A. Balinskii and S.P. Novikov, Poisson brackets of hydrodynamics type. Frobenius algebras and Lie algebras (Russian), Dokl. Akad. Nauk SSSR 283(5) (1985) 1036- 1039. [4] L.A. Bokut, Yuqun Chen and Jiapeng Huang, Grobner-Shirshov bases for L-algebras, Internat. J. Algebra Comput. 23(3) (2013) 547-571. [5] D. Burde and K. Dekimpe, Novikov structures on solvable Lie algebras, J. Geom. Phys. 56(9) (2006) 1837-1855. [6] L. Chen, Y. Niu and D. Meng, Two kinds of Novikov algebras and their realizations, J. Pure Appl. Algebra 212(4) (2008) 902-909. [7] A.S. Dzhumadil'daev and Clas Lofwall, Trees, free right-symmetric algebras, free Novikov algebras and identities, Homology Homotopy Appl. 4(2) (2002) 165-190. [8] V.T. Filippov, A class of simple nonassociative algebras, Mat. Zametki 45(1) (1989) 101-105; translation in Math. Notes 45(1-2) (1989) 68-71. 21 [9] I.M. Gelfand and I.Ya. Dorfman, Hamiltonian operators and algebraic structures related to them, Funktsional. Anal. i Prilozhen. 13(4) (1979) 13-30. [10] N. Jacobson, Basic Algebra II, second edition, W.H. Freeman and Company, New York, 1989. [11] S.P. Novikov, Geometry of conservative systems of hydrodynamic type. The averag- ing method for eld-theoretic systems, International conference on current problems in algebra and analysis (Moscow-Leningrad, 1984), Russian Mathematical Surveys, 40(4) (1985) 85-98. [12] J.M. Osborn, Novikov algebras, Nova J. Algebra Geom. 1(1) (1992) 1-13. [13] J.M. Osborn, Simple Novikov algebras with an idempotent, Comm. Algebra 20(9) (1992) 2729-2753. [14] J.M. Osborn, Infinite dimensional Novikov algebras of characteristic 0, J. Algebra 167(1) (1994) 146-167. [15] J.M. Osborn, Modules over Novikov algebras of characteristic 0, Comm. Algebra 23(10) (1995) 3627-3640. [16] J.M. Osborn and E.I. Zelmanov, Nonassociative algebras related to Hamiltonian operators in the formal calculus of variations, J. Pure Appl. Algebra 101(3) (1995) 335-352. [17] X. Xu, Hamiltonian operators and associative algebras with a derivation, Lett. Math. Phys. 33(1) (1995) 1-6. [18] X. Xu, On simple Novikov algebras and their irreducible modules, J. Algebra 185(3) (1996) 905-934. [19] X. Xu, Novikov-Poisson algebras, J. Algebra 190(2) (1997) 253-279. [20] X. Xu, Quadratic Conformal Superalgebras, J. Algebra 231(1) (2000) 1-38. 22 [21] A.S. Zakharov, Embedding of Novikov-Poisson algebras in Novikov-Poisson algebras of vector type, Algebra Logika 52(3) (2013) 352-369; translation in Algebra Logic 52(3) (2013) 236-249. [22] A.S. Zakharov, Novikov-Poisson algebras and superalgebras of Jordan brackets, Comm. Algebra 42(5) (2014) 2285-2298. [23] E.I. Zelmanov, On a class of local translation invariant Lie algebras, Soviet Math. Dokl. 35(1) (1987) 216-218. [24] Y. Zhao and C. Bai, Some results on Novikov-Poisson algebras, Internat. J. Theoret. Phys. 43(2) (2004) 519-528. [25] V.N. Zhelyabin and A.S. Tikhov, Novikov-Poisson algebras and associative commuta- tive derivation algebras (Russian), Algebra Logika 47(2) (2008) 186-202; translation in Algebra Logic 47(2) (2008) 107-117. 23
1703.10601	3	1703	2019-06-19T14:04:19	Group gradations on Leavitt path algebras	[ "math.RA" ]	Given a directed graph $E$ and an associative unital ring $R$ one may define the Leavitt path algebra with coefficients in $R$, denoted by $L_R(E)$. For an arbitrary group $G$, $L_R(E)$ can be viewed as a $G$-graded ring. In this article, we show that $L_R(E)$ is always nearly epsilon-strongly $G$-graded. We also show that if $E$ is finite, then $L_R(E)$ is epsilon-strongly $G$-graded. We present a new proof of Hazrat's characterization of strongly $\mathbb{Z}$-graded Leavitt path algebras, when $E$ is finite. Moreover, if $E$ is row-finite and has no source, then we show that $L_R(E)$ is strongly $\mathbb{Z}$-graded if and only if $E$ has no sink. We also use a result concerning Frobenius epsilon-strongly $G$-graded rings, where $G$ is finite, to obtain criteria which ensure that $L_R(E)$ is Frobenius over its identity component.	math.RA	math	GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS Department of Engineering Science, University West, SE-46186 Trollhättan, Sweden PATRIK NYSTEDT JOHAN ÖINERT Department of Mathematics and Natural Sciences, Blekinge Institute of Technology, SE-37179 Karlskrona, Sweden Abstract. Given a directed graph E and an associative unital ring R one may define the Leavitt path algebra with coefficients in R, denoted by LR(E). For an arbitrary group G, LR(E) can be viewed as a G-graded ring. In this article, we show that LR(E) is always nearly epsilon-strongly G-graded. We also show that if E is finite, then LR(E) is epsilon-strongly G-graded. We present a new proof of Hazrat's characterization of strongly Z-graded Leavitt path algebras, when E is finite. Moreover, if E is row-finite and has no source, then we show that LR(E) is strongly Z-graded if and only if E has no sink. We also use a result concerning Frobenius epsilon-strongly G-graded rings, where G is finite, to obtain criteria which ensure that LR(E) is Frobenius over its identity component. 1. Introduction A Leavitt path algebra associates with a directed graph E and an associative unital ring R, an R-algebra LR(E). These algebras are algebraic analogues of graph C ∗-algebras and are also natural generalizations of Leavitt algebras of type (1, n) constructed in [8]. Leavitt path algebras were introduced, in the case when R is a field, by Ara, Moreno and Pardo [3], and almost simultaneously by Abrams and Aranda Pino [2], using a different approach. In the case when R is a commutative ring these structures were defined by Tomforde in [13]. Throughout this article we will follow Hazrat [6] and consider Leavitt path algebras with coefficients in an arbitrary associative unital ring R (see Definition 18). Each Leavitt path algebra may, in a canonical way, be viewed as a Z-graded ring (see Section 4). The characterization of Leavitt path algebras (e.g. simple, purely infinite simple, semisimple, prime, finite dimensional, locally finite, exchange, artinian, noetherian) in terms of properties of the underlying graph has been the subject of many studies, see Abrams' extensive survey [1], and the references therein. Many of these investigations have, however, been carried out without taking into account the Z-graded structure of LR(E). Although there are previous studies (see e.g. Ara, Moreno and E-mail addresses: [email protected], [email protected]. Date: June 20, 2019. 2010 Mathematics Subject Classification. 16S99, 16W50. Key words and phrases. s-unital ring, strongly graded ring, epsilon-strongly graded ring, Leavitt path algebra. 1 2 GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS Pardo [3], Tomforde [12]) where the Z-graded structure of LR(E) has been taken into account, it seems to the authors of the present article that Hazrat [6] was first to utilize this philosophy systematically, and to e.g. study properties of the Z-gradation itself. In loc. cit. he proves, among many other things, the following result. Theorem 1 (Hazrat [6]). If E is a finite directed graph and R is an associative unital ring, then the Leavitt path algebra LR(E) is strongly Z-graded if and only if E has no sink. To motivate the approach taken in this article, consider the following directed graphs. A : • • B : • / • C : (∞) / • • According to Theorem 1, the graph A produces a strongly Z-graded Leavitt path algebra, but the graphs B and C do not. In an attempt to single out a class of well-behaved group graded rings which naturally includes all crossed products by unital twisted partial actions, Nystedt, Öinert and Pinedo [10] recently introduced a generalization of unital strongly graded rings called epsilon-strongly graded rings (see Definition 6). In this article, we establish the following result which shows that LR(B) is epsilon-strongly Z-graded. Theorem 2. If E is a finite directed graph and R is an associative unital ring, then the Leavitt path algebra LR(E) is epsilon-strongly Z-graded. It turns out that LR(C) is not epsilon-strongly Z-graded (see Example 32). To capture Leavitt path algebras corresponding to the graph C we introduce the class of nearly epsilon- strongly graded rings (see Definition 10), which is a further weakening of the notion of a strongly graded ring, and show the following result. Theorem 3. If E is a (possibly non-finite) directed graph and R is an associative unital ring, then the Leavitt path algebra LR(E) is nearly epsilon-strongly Z-graded. Here is a detailed outline of this article. In Section 2, we recall some notions concerning unital and s-unital modules that we need in the sequel. In Section 3, we provide the necessary background on group graded rings for use in subse- quent sections. We define the following notions: symmetrically graded ring, epsilon-strongly graded ring, nearly epsilon-strongly graded ring, non-degenerately graded ring and strongly non-degenerately graded ring (see Definitions 5, 6, 10, 12 resp. 13) and we show a result connecting these notions (see Proposition 14). In Section 4, we accomodate the Leavitt path algebra framework that we need in the sequel; in particular standard G-gradations on Leavitt path algebras, for an arbitrary group G. In the same section, we prove G-graded versions of Theorem 2 and Theorem 3 (see Theorem 28 and Theorem 30) through a series of propositions. At the end of the section, we apply a result, by Pinedo and the authors of the present article, concerning epsilon-strongly G-graded rings, to obtain criteria which ensure that a Leavitt path algebra is Frobenius over its identity component, when the group G is finite (see Theorem 38). In Section 5, we provide a useful characterization of strongly Z-graded rings (see Proposi- tion 39) and several corollaries. In Section 6, we obtain a new proof of Hazrat's Theorem 1 (see Proposition 45) by using our results from Section 4 and Section 5. We also characterize strongly Z-graded Leavitt path algebras over row-finite graphs without sources (see Theorem 47). o o / / GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS 3 2. s-unital modules In this section, we recall some notions concerning unital and s-unital modules that we need in the sequel. Throughout this article all rings are associative but not necessarily unital. For the rest of this section T and T ′ denote rings. Recall the following definitions. If M is a left T -module (right T ′-module), then M is called left (right) unital if there is t ∈ T (t′ ∈ T ′) such that, for all m ∈ M, the relation tm = m (mt′ = m) holds. In that case t (t′) is said to be a left (right) identity for M. If M is a T -- T ′-bimodule, then M is called unital if it is unital both as a left T -module and as a right T ′-module. The ring T is said to be left (right) unital if it is left (right) unital as a left (right) module over itself. The ring T is called unital if it is unital as a bimodule over itself. In this article, we will use the notion of s-unitality, introduced in [14]. Namely, if M is a left T -module (right T ′-module), then M is said to be s-unital if for each m ∈ M the relation m ∈ T m (m ∈ mT ′) holds. If M is a T -- T ′-bimodule, then M is said to be s-unital if it is s-unital both as a left T -module and as a right T ′-module. The ring T is said to be left (right) s-unital if it is left (right) s-unital as a left (right) module over itself. The ring T is said to be s-unital if it is s-unital as a bimodule over itself. Remark 4. From [14, Theorem 1] it follows that if M is a left T -module (right T ′-module), then M is s-unital if and only if for all n ∈ N and all m1, . . . , mn ∈ M there exists t ∈ T (t′ ∈ T ) such that, for each i ∈ {1, . . . , n}, the equality tmi = mi (mit′ = mi) holds. Let M be a left T -module (right T ′-module). Recall that M is called torsion-free if for 6= {0}) holds. If M is a T -- T ′-bimodule all non-zero m ∈ M the relation T m 6= {0} (mT ′ which is torsion-free both as a left T -module and as a right T ′-module, then M is said to be torsion-free. It is clear that the following chain of implications hold: M is unital ⇒ M is s-unital ⇒ M is torsion-free. (1) 3. Graded rings In this section, we provide the necessary background on group graded rings. In particular, we define the notion of a symmetrically graded ring, an epsilon-strongly graded ring, a nearly epsilon-strongly graded ring, a non-degenerately graded ring resp. a strongly non-degenerately graded ring (see Definitions 5, 6, 10, 12 resp. 13) and show a result connecting these notions (see Proposition 14). Throughout this section, G denotes a group with identity element e and S denotes an associative ring which is G-graded (or graded by G). Recall that this means that for each g ∈ G, there is an additive subgroup Sg of S such that S = ⊕g∈GSg and for all g, h ∈ G the inclusion SgSh ⊆ Sgh holds. If, in addition, for all g, h ∈ G the equality SgSh = Sgh holds, then S is said to be strongly G-graded (or strongly graded by G). For an overview of the theory of group graded rings, we refer the reader to [9]. Following [4, Definition 4.5] we make the following definition. Definition 5. The ring S is said to be symmetrically G-graded (or symmetrically graded by G) if for each g ∈ G the relation SgSg−1Sg = Sg holds. We now recall from [10] the following weakening of the notion of a unital strongly graded ring. Definition 6. The ring S is said to be epsilon-strongly G-graded (or epsilon-strongly graded by G) if for each g ∈ G the SgSg−1 -- Sg−1Sg-bimodule Sg is unital. 4 GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS The following characterization of epsilon-strongly graded rings more or less follows from [10, Proposition 7]. For the convenience of the reader we include a proof. Proposition 7. The following assertions are equivalent: (i) The ring S is epsilon-strongly G-graded; (ii) The ring S is symmetrically G-graded and for each g ∈ G the ring SgSg−1 is unital; (iii) For each g ∈ G there exists ǫg ∈ SgSg−1 such that for all s ∈ Sg the equalities ǫgs = sǫg−1 = s hold. Proof. (i)⇒(ii): Suppose that S is epsilon-strongly G-graded. Take g ∈ G. First we show that S is symmetrically G-graded. It is clear that SgSg−1Sg ⊆ Sg. Now we show the reversed inclusion. Take s ∈ Sg. From Definition 6 it follows that there exists ǫg ∈ SgSg−1 such that In particular, it follows that s ∈ ǫgSg ⊆ SgSg−1Sg. Now we show that the ring ǫgs = s. SgSg−1 is unital. By Definition 6, Sg is a unital left SgSg−1-module and Sg−1 is a unital right SgSg−1-module. Thus, SgSg−1 is unital as an SgSg−1-bimodule. (ii)⇒(iii): Suppose that S is symmetrically G-graded and for each g ∈ G the ring SgSg−1 is unital. Take g ∈ G. Let ǫg denote the multiplicative identity element of SgSg−1. Take s ∈ Sg. Using that S is symmetrically G-graded, there exist n ∈ N, ai, ci ∈ Sg and bi ∈ Sg−1, for i ∈ {1, . . . , n}, such that s = Pn i=1 aibici. Take i ∈ {1, . . . , n}. Since aibi ∈ SgSg−1 we get i=1 ǫgaibici = Pn that ǫgaibi = aibi and thus ǫgs = Pn i=1 aibici = s. Similarly, sǫg−1 = s. (iii)⇒(i): This is immediate. (cid:3) Remark 8. If S is epsilon-strongly G-graded, then the element ǫg from Proposition 7(iii) is a multiplicative identity element of the Se-ideal SgSg−1 and ǫg ∈ Z(Se). Moreover, ǫe is a multiplicative identity element of both Se and S, making S a unital ring. Proposition 9. If S is epsilon-strongly G-graded, then the following assertions are equivalent: (i) S is strongly G-graded; (ii) for each g ∈ G the equality SgSg−1 = Se holds; (iii) for each g ∈ G the equality ǫg = 1 holds. Proof. The equivalence (i)⇔(ii) is [9, Proposition 1.1.1(3)] and the equivalence (ii)⇔(iii) fol- lows from Remark 8. (cid:3) Now we introduce the following weakening of epsilon-strongly graded rings. Definition 10. The ring S is said to be nearly epsilon-strongly G-graded (or nearly epsilon- strongly graded by G) if for each g ∈ G the SgSg−1 -- Sg−1Sg-bimodule Sg is s-unital. Now we show an "s-unital version" of Proposition 7. Proposition 11. The following assertions are equivalent: (i) The ring S is nearly epsilon-strongly G-graded; (ii) The ring S is symmetrically G-graded and for each g ∈ G the ring SgSg−1 is s-unital; (iii) For each g ∈ G and each s ∈ Sg there exist ǫg(s) ∈ SgSg−1 and ǫ′ g(s) ∈ Sg−1Sg such that the equalities ǫg(s)s = sǫ′ g(s) = s hold. If (i), (ii) and (iii) are satisfied, then for all g ∈ G, the Se-ideal SgSg−1 is generated by {ǫg(s) s ∈ Sg} ∪ {ǫ′ g−1(s) s ∈ Sg}. GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS 5 Proof. (i)⇒(ii): Suppose that S is nearly epsilon-strongly G-graded. The symmetrical part is shown in the same way as for Proposition 7. Take g ∈ G. By Definition 10, Sg is an s-unital left SgSg−1-module and Sg−1 is an s-unital right SgSg−1-module. Thus, the ring SgSg−1 is s-unital. (ii)⇒(iii): Suppose that the ring S is symmetrically G-graded and for each g ∈ G the ring SgSg−1 is s-unital. Take g ∈ G and s ∈ Sg. Using that S is symmetrically G-graded, there exist n ∈ N, ai, ci ∈ Sg and bi ∈ Sg−1, for i ∈ {1, . . . , n}, such that s = Pn i=1 aibici. Since aibi ∈ SgSg−1 and bici ∈ Sg−1Sg, for i ∈ {1, . . . , n}, it follows from s-unitality and Remark 4 that there exist ǫg(s) ∈ SgSg−1 and ǫ′ g(s) ∈ Sg−1Sg such that the equalities ǫg(s)aibi = aibi and biciǫ′ g(s) = bici hold, for i ∈ {1, . . . , n}. Thus, ǫg(s)s = sǫ′ g(s) = s. (iii)⇒(i): This is immediate. (cid:3) Following Cohen and Montgomery [5], and Lundström and Öinert [11], we make the follow- ing definition. Definition 12. The gradation on S is said to be right non-degenerate (resp. left non- degenerate) if for each g ∈ G and each non-zero s ∈ Sg, the relation sSg−1 6= {0} (resp. Sg−1s 6= {0}) holds. If the gradation on S is right (resp. left) non-degenerate, then S is said to be a right (resp. left) non-degenerately G-graded ring. Now we define a strengthening of non-degenerate gradations. Definition 13. The gradation on S is said to be strongly right non-degenerate (resp. strongly left non-degenerate) if for each g ∈ G the left SgSg−1-module (right Sg−1Sg-module) Sg is torsion-free. If the gradation on S is strongly right (resp. left) non-degenerate, then S is said to be a strongly right (resp. left) non-degenerately G-graded ring. Proposition 14. If S is nearly epsilon-strongly G-graded, then S is symmetrically G-graded, strongly right non-degenerately G-graded and strongly left non-degenerately G-graded. Proof. This follows immediately from (1) and Proposition 11(ii). Definition 15. An involution (·)∗ : S → S is said to be anti-graded if for all g ∈ G the equality (Sg)∗ = Sg−1 holds. Proposition 16. If S is equipped with an anti-graded involution, then the following assertions are equivalent: (cid:3) (i) S is epsilon-strongly G-graded; (ii) for each g ∈ G, there exists ǫg ∈ SgSg−1 such that ǫ∗ g = ǫg and for all s ∈ Sg the equality ǫgs = s holds; (iii) for each g ∈ G, there exists ǫg ∈ SgSg−1 such that ǫ∗ g = ǫg and for all s ∈ Sg the equality sǫg−1 = s holds. Proof. (i)⇒(ii): Suppose that S is epsilon-strongly G-graded. Take g ∈ G and an element ǫg ∈ SgSg−1 satisfying the conditions in Proposition 7(iii). Then ǫ∗ g = g)∗ = ǫg. SgSg−1 and thus, from Remark 8, it follows that ǫ∗ g = (ǫ∗ Thus, (ii) holds. g ∈ (SgSg−1)∗ = (Sg−1)∗S∗ g)∗ǫ∗ g = (ǫgǫ∗ g)∗ = (ǫ∗ g = ǫgǫ∗ s ∈ Sg. From the fact that s∗ ∈ Sg−1 it follows that sǫg−1 = (ǫ∗ (ii)⇒(i): Suppose that (ii) holds for some elements ǫg ∈ SgSg−1, for g ∈ G. Take g ∈ G and g−1s∗)∗ = (ǫg−1s∗)∗ = (s∗)∗ = s. The proof of the equivalence (i)⇔(iii) is analogous and is therefore left to the reader. (cid:3) Proposition 17. Suppose that S is equipped with an anti-graded involution. 6 GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS (a) If for each g ∈ G and all s ∈ Sg, there exists ǫg(s) ∈ SgSg−1 such that ǫg(s)∗ = ǫg(s) and the equality ǫg(s)s = s holds, then S is nearly epsilon-strongly G-graded. In that case, for each g ∈ G, the Se-ideal SgSg−1 is generated by {ǫg(s) s ∈ Sg}. g(s) ∈ Sg−1Sg such that ǫ′ (b) If for each g ∈ G and all s ∈ Sg, there exists ǫ′ g(s)∗ = ǫ′ and the equality sǫ′ case, for each g ∈ G, the Se-ideal SgSg−1 is generated by {ǫ′ g(s) g(s) = s holds, then S is nearly epsilon-strongly G-graded. In that g−1(s) s ∈ Sg}. Proof. Analogous to the proofs of (ii)⇒(i) resp. (iii)⇒(i) in Proposition 16. (cid:3) 4. Leavitt path algebras In this section, we accomodate the Leavitt path algebra framework that we need; in par- ticular standard G-gradations on Leavitt path algebras, for an arbitrary group G. We prove G-graded versions of Theorem 2 and Theorem 3 (see Theorem 28 and Theorem 30). At the end of this section we establish criteria which ensure that a Leavitt path algebra is Frobenius over its identity component when the group G is finite (see Theorem 38). Let R be an associative unital ring and let E = (E0, E1, r, s) be a directed graph. Recall that r (range) and s (source) are maps E1 → E0. The elements of E0 are called vertices and the elements of E1 are called edges. A vertex v for which s−1(v) is empty is called a sink. A vertex v for which r−1(v) is empty is called a source. If s−1(v) is a finite set for every v ∈ E0, then E is called row-finite. If both E0 and E1 are finite sets, then we say that E is finite. A path µ in E is a sequence of edges µ = µ1 . . . µn such that r(µi) = s(µi+1) for i ∈ {1, . . . , n − 1}. In such a case, s(µ) := s(µ1) is the source of µ, r(µ) := r(µn) is the range of µ, and l(µ) := n is the length of µ. For any vertex v ∈ E0 we put s(v) := v and r(v) := v. The elements of E1 are called real edges, while for f ∈ E1 we call f ∗ a ghost edge. The set {f ∗ f ∈ E1} will be denoted by (E1)∗. We let r(f ∗) denote s(f ), and we let s(f ∗) denote r(f ). Following Hazrat [6] we make the following definition. Definition 18. The Leavitt path algebra of E with coefficients in R, denoted by LR(E), is the algebra generated by the sets {v v ∈ E0}, {f f ∈ E1} and {f ∗ f ∈ E1} with the coefficients in R, subject to the relations: (1) uv = δu,vv for all u, v ∈ E0; (2) s(f )f = f r(f ) = f and r(f )f ∗ = f ∗s(f ) = f ∗, for all f ∈ E1; (3) f ∗f ′ = δf,f ′r(f ), for all f, f ′ ∈ E1; (4) Pf ∈E 1,s(f )=v f f ∗ = v, for every v ∈ E0 for which s−1(v) is non-empty and finite. Here the ring R commutes with the generators. There are many ways to assign a group gradation to LR(E). Indeed, let G be an arbitrary group. Put deg(v) = e, for each v ∈ E0. For each f ∈ E1, choose some g ∈ G and put deg(f ) = g and deg(f ∗) = g−1. By assigning degrees to the generators of the free algebra FR(E) = Rhv, f, f ∗ v ∈ E0, f ∈ E1i in this way, we are ensured that the ideal coming from relations (1) -- (4) in Definition 18 is graded, i.e. homogeneous. Using this it is easy to see that the natural G-gradation on FR(E) carries over to a G-gradation on the quotient algebra LR(E). We will refer to such a gradation as a standard G-gradation on LR(E). If µ = µ1 . . . µn is a path, then µ can be viewed as an element in LR(E), and we denote by µ∗ the element µ∗ 1 of LR(E). This gives rise to an anti-graded involution on LR(E). For n ≥ 1 we define En to be the set of paths of length n, and let E∗ = ∪n≥0En be the set of all paths (and vertices). It is clear that the map deg : E0 ∪ E1 ∪ (E1)∗ → G, described above, n · · · µ∗ GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS 7 can be extended to a map deg : E∗ ∪ {µ∗ µ ∈ E∗} → G. One important instance of such a gradation is coming from the case G = Z, when we put deg(v) = 0, for v ∈ E0, deg(f ) = 1 and deg(f ∗) = −1, for f ∈ E1. We will refer to this gradation as the canonical Z-gradation on LR(E). Assumption: Unless otherwise stated, throughout the rest of this section E will denote an arbitrary directed graph, R will denote an arbitrary associative unital ring, G will denote an arbitrary group and we will assume that LR(E) is equipped with an arbitrary standard G-gradation, S = LR(E) = ⊕g∈GSg. We wish to introduce a partial order related to Leavitt path algebras. To do that we need some concepts and a well-known result which we include for the convenience of the reader. Definition 19. Let ≤ be a relation on a set X. Recall that ≤ is called a preorder if it is reflexive and transitive. If ≤ is a preorder, then it is called a partial order if it is antisymmetric. Proposition 20. Suppose that ≤ is a preorder on a set X. If we, for any x, y ∈ X, put x ∼ y, whenever x ≤ y and y ≤ x, then ∼ is an equivalence relation on X which makes the quotient relation (cid:22) on X/ ∼, induced by ≤, a partial order. Proof. This is well-known and straightforward to check. (cid:3) Definition 21. Let X denote the set of all formal expressions of the form αβ∗ where α, β ∈ E∗ and r(α) = r(β). Take g ∈ G and put Xg = {x ∈ X deg(x) = g}. Suppose that α, β, γ, δ ∈ E∗ satisfy αβ∗, γδ∗ ∈ Xg. We put αβ∗ ≤ γδ∗ if α is an initial subpath of γ. We put αβ∗ ∼ γδ∗ whenever αβ∗ ≤ γδ∗ and γδ∗ ≤ αβ∗, that is if α = γ. Remark 22. (a) In the above definition of X, it is fully possible to let α or β be a vertex. (b) For each g ∈ G, the homogeneous component Sg in the G-gradation of LR(E) is the R-linear span of Xg. Proposition 23. Let g ∈ G be arbitrary. The following assertions hold: (a) The relation ≤ is a preorder on Xg. (b) The relation ∼ is an equivalence relation on Xg. (c) The quotient relation (cid:22) on Xg/ ∼ induced by ≤ is a partial order. 1 ≤ α1β∗ Proof. (a) Take αi, βi ∈ E∗, such that αiβ∗ 1, since α1 is an initial subpath of itself. Thus, ≤ is reflexive. Now we show that ≤ is transitive. 1 ≤ α2β∗ Suppose that α1β∗ 3. Then α1 must be an initial subpath of α3. 1 ≤ α3β∗ Thus, α1β∗ 3. (b) This is clear. (c) This follows from Proposition 20. i ∈ Xg, for i ∈ {1, 2, 3}. It is clear that α1β∗ (cid:3) 2 and α2β∗ 2 ≤ α3β∗ Proposition 24. Let g ∈ G be arbitrary. The map Ng : Xg/ ∼ ∋ [x] 7→ Ng(x) := xx∗ ∈ LR(E)e is well-defined. Proof. Suppose that [x] = [y]. Then x ∼ y which implies that x ≤ y and y ≤ x. Then x = αβ∗ and y = αγ∗ for some α, β, γ ∈ E∗. Thus, Ng(x) = xx∗ = αβ∗βα∗ = αα∗ = αγ∗γα∗ = yy∗ = Ng(y). (cid:3) Lemma 25. If α, β ∈ E∗ are chosen so that α∗β 6= 0 in LR(E), then α is an initial subpath of β or vice versa. 8 GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS Proof. Put m = l(α) and n = l(β). Suppose that α = f1 . . . fm and β = f ′ 1 . . . f ′ n. Case 1: m ≤ n. Then 0 6= α∗β = f ∗ m · · · f ∗ 1 f ′ 1 · · · f ′ n = δf1,f ′ 1 · · · δfm,f ′ mf ′ m+1 · · · f ′ n. This implies that fi = f ′ Case 2: m > n. Then i for i ∈ {1, . . . , m}. Thus, α is an initial subpath of β. 0 6= α∗β = f ∗ m · · · f ∗ 1 f ′ 1 · · · f ′ n = δf1,f ′ 1 · · · δfn,f ′ nf ∗ m · · · f ∗ n+1. Hence, fi = f ′ i for i ∈ {1, . . . , n}. Thus, β is an initial subpath of α. (cid:3) Proposition 26. Let g ∈ G and x, y ∈ Xg be arbitrary. The following assertions hold: (a) If [x] (cid:22) [y], then Ng(x)y = y. (b) If [x] (cid:14) [y] and [y] (cid:14) [x], then Ng(x)y = 0. Proof. Put x = αβ∗ and y = γδ∗. (a) Suppose that [x] (cid:22) [y]. Then α is an initial subpath of γ. Hence, γ = αα′ for some α′ ∈ E∗ such that r(α) = s(α′). Therefore, Ng(x)y = αβ∗βα∗γδ∗ = αα∗αα′δ∗ = αα′δ∗ = γδ∗ = y. (cid:3) (b) From Lemma 25, we get that Ng(x)y = αβ∗βα∗γδ∗ = αα∗γδ∗ = α0δ∗ = 0. Definition 27. Suppose that E is a finite directed graph and let g ∈ G. Since E is finite, we can, on account of Proposition 23, choose ng ∈ N and m1, . . . , mng ∈ Xg such that {[m1], . . . , [mng ]} equals the set of minimal elements of Xg/ ∼, with respect to (cid:22). We define ǫg = ng X i=1 Ng(mi) in LR(E). Notice that we will always get ǫe = Pv∈E 0 v. The following result allows us establish Theorem 2, in particular. Theorem 28. If E is a finite directed graph and R is an associative unital ring, then LR(E) is epsilon-strongly G-graded. Proof. Let g ∈ G be arbitrary and put ǫg = Png i=1 Ng(mi) in LR(E). Clearly, ǫg ∈ SgSg−1. Take γδ∗ ∈ Xg. There is a unique j ∈ {1, . . . , ng} such that [mj] (cid:22) [γδ∗]. From Proposition 26 it follows that Ng(mj)γδ∗ = γδ∗ and if i 6= j, then Ng(mi)γδ∗ = 0. Thus, ǫgγδ∗ = γδ∗. The claim now follows from Proposition 16 and Remark 22(b). (cid:3) Remark 29. Consider the directed graph B from the introduction and suppose that LR(B) is equipped with a standard G-gradation. From Theorem 28 it follows that LR(B) is epsilon- strongly G-graded. The following result allows us establish Theorem 3, in particular. Theorem 30. If E is a directed graph and R is an associative unital ring, then LR(E) is nearly epsilon-strongly G-graded. Proof. Take g ∈ G and s ∈ Sg. There is a finite subset I of N such that s = Pi∈I riαiβ∗ i for suitable ri ∈ R and αiβ∗ j ]}j∈J equals the set of minimal elements in {[αiβ∗ j ) ∈ SgSg−1. For each i ∈ I, there is a unique j(i) ∈ J such that [αj(i)β∗ i ]. From Proposition 26 it follows that Ng(αj(i)β∗ i = 0. Thus, i ∈ Xg. Take a subset J of I such that {[αjβ∗ i ]}i∈I . Put ǫg(s) = Pj∈J Ng(αjβ∗ j(i)] (cid:22) [αiβ∗ i = αiβ∗ i and if j 6= j(i), then Ng(αjβ∗ j )αiβ∗ j(i))αiβ∗ GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS 9 ǫg(s)s = Pi∈I Pj∈J riNg(αjβ∗ claim now follows from Proposition 17. j )αiβ∗ i = Pi∈I riNg(αj(i)β∗ j(i))αiβ∗ i = Pi∈I riαiβ∗ i = s. The (cid:3) Example 31. Consider the following finite directed graph E: •v1 f1 •v2 f2 / •v3 f3 •v4 f4 •v5 Equip S = LR(E) with its canonical Z-gradation, and view it as a Z-graded ring. Notice that v1 and v3 are sinks. Hence, by Theorem 1, S is not strongly Z-graded. However, S is epsilon-strongly Z-graded, by Theorem 28. We shall now, for each n ∈ Z, explicitly describe the homogeneous component Sn and the element ǫn. A quick calculation yields the following: S0 = spanR{v1, v2, v3, v4, v5, f1f ∗ 1 , f2f ∗ S1 = spanR{f1, f2, f3, f4, f4f3f ∗ 2 }, S2 = spanR{f4f3}, S−2 = spanR{f ∗ 2 , f3f ∗ 3 , f2f ∗ S−1 = spanR{f ∗ 3 f ∗ 2 }, 4 }, 1 , f ∗ 4 }, and Sn = {0}, for n > 2. 4 , f2f ∗ 2 , f ∗ 3 , f ∗ 3 f ∗ Following the proof of Theorem 28 we shall now choose: • ǫ0 = v1 + v2 + v3 + v4 + v5 ∈ S0; • ǫ1 = v2 + v4 + v5 = (f2f ∗ 2 + f1f ∗ • ǫ−1 = f2f ∗ 2 + v1 + v3 + v4 = f2f ∗ • ǫ2 = v5 = f4f3f ∗ 3 f ∗ • ǫ−2 = v3 = f ∗ • ǫn = 0, if n > 2. 3 f ∗ 4 ∈ S2S−2; 4 f4f3 ∈ S−2S2; 1 ) + f3f ∗ 3 f ∗ 4 f4f3f ∗ 3 + f4f ∗ 2 + f ∗ 4 ∈ S1S−1; 1 f1 + f ∗ 3 f3 + f ∗ 4 f4 ∈ S−1S1; It is not difficult to see that, for any n ∈ Z, ǫn satisfies the conditions in Proposition 7(iii). The next example shows that Theorem 28 can not be generalized to arbitrary (possibly non-finite) directed graphs E. / •v2. For this graph Example 32. Consider the graph from the introduction C : •v1 C 1 = {f1, f2, f3, . . .} is infinite. Is LR(C) epsilon-strongly Z-graded? We begin by writing out some of the homogeneous components of the canonical Z-gradation on S = LR(C). (∞) • S0 = spanR{v1, v2, f1f ∗ • S1 = spanR{f1, f2, f3, . . .} • S−1 = spanR{f ∗ 2 , f ∗ 1 , f ∗ 3 , . . .} 1 , f1f ∗ 2 , . . . , fif ∗ j , . . .} If S1 had only contained a finite number of elements, then we could have put ǫ1 = Pf ∈C 1 f f ∗. However, in the current situation that is not possible. The only option left for us is to put ǫ1 = v1. We need to check whether the conditions in Definition 6 are satisfied, and in particular whether v1 ∈ S1S−1. Seeking a contradiction, suppose that v1 = Pi∈F rifpif ∗ qi, where pi, qi ∈ N+, for some finite set F . For any k ∈ N+ we get 0 6= v2 = f ∗ k v1fk = X i∈F rif ∗ k fpif ∗ qifk which in particular yields that there is some i ∈ F such that k = pi = qi. Thus, F is not finite, which is a contradiction. This shows that ǫ1 /∈ S1S−1 and hence S = LR(C) is not epsilon-strongly Z-graded. o o / o o o o / 10 GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS Remark 33. By Example 31 and Example 32 we have established the following chain of strict inclusions: {unital strongly Z-graded Leavitt path algebras} ( {epsilon-strongly Z-graded Leavitt path algebras} ( {nearly epsilon-strongly Z-graded Leavitt path algebras}. Proposition 34. If E is a directed graph and R is an associative unital ring, then the Leavitt path algebra LR(E) is symmetrically G-graded, strongly right non-degenerately G-graded and strongly left non-degenerately G-graded. Proof. This follows immediately from Proposition 14 and Theorem 30. (cid:3) Remark 35. The "symmetric" part of Proposition 34 can be shown directly without the use of Theorem 30. Indeed, take g ∈ G. If Sg = {0}, then clearly SgSg−1Sg = Sg holds. Now, suppose that Sg 6= {0}. It follows immediately from the G-gradation that SgSg−1Sg ⊆ Sg holds. To show the reversed inclusion, take an arbitrary non-zero monomial αβ∗ ∈ Sg. Then r(α) = r(β) and βα∗ ∈ Sg−1. We get αβ∗ = αr(β)r(α)β∗ = α(β∗β)(α∗α)β∗ = (αβ∗)(βα∗)(αβ∗) ∈ SgSg−1Sg. This shows that Sg ⊆ SgSg−1Sg. We end this section with an application to Frobenius extensions. To this end, recall the following. Definition 36. Suppose that S/T is a ring extension. By this we mean that T ⊆ S and both S and T are unital with a common multiplicative identity. The ring extension S/T is called Frobenius if there is a finite set J, xj, yj ∈ S, for j ∈ J, and a T -bimodule map E : S → T such that, for every s ∈ S, the equalities s = Pj∈J xjE(yjs) = Pj∈J E(sxj)yj hold. In that case, (E, xj , yj) is called a Frobenius system. In [10, Theorem 25] Nystedt, Öinert and Pinedo have shown the following result. Theorem 37. Suppose that S is a ring which is epsilon-strongly graded by a finite group G. If we put T = Se, then S/T is a Frobenius extension. As a direct consequence of the above result, in combination with Theorem 28, we get the following result. Theorem 38. Let E be a finite directed graph and R an associative unital ring. If G is a finite group and we equip the Leavitt path algebra LR(E) with a standard G-gradation, then the ring extension LR(E)/LR(E)e is Frobenius. 5. Strongly Z-graded rings Throughout this section S denotes a, not necessarily unital, Z-graded ring. Proposition 39. The ring S is strongly Z-graded if and only if for every n ∈ Z the S0- bimodule Sn is unital, and the equalities S1S−1 = S−1S1 = S0 hold. Proof. The "only if" statement is immediate. Now we show the "if" statement. Take positive integers m and n. First we show by induction that Sm = (S1)m. The base case m = 1 is clear. Next suppose that Sm = (S1)m. Then we get that Sm+1 = S0Sm+1 = S1S−1Sm+1 ⊆ S1S−1+m+1 = S1Sm = S1(S1)m = (S1)m+1 ⊆ Sm+1. Next we show by induction that S−n = (S−1)n. The base case n = 1 is clear. Next suppose that S−n = (S−1)n. Then we get GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS 11 that S−n−1 = S−n−1S0 = S−n−1S1S−1 ⊆ S−n−1+1S−1 = S−nS−1 = (S−1)nS−1 = (S−1)n+1 ⊆ S−n−1. Case 1: SmSn = (S1)m(S1)n = (S1)m+n = Sm+n. Case 2: S−mS−n = (S−1)m(S−1)n = (S−1)m+n = S−m−n. Case 3: Now we show that SmS−n = Sm−n. We get that SmS−n = (S1)m(S−1)n. By repeated application of the equality S1S−1 = S0, we get that (S1)m(S−1)n = (S1)m−n = Sm−n, if m ≥ n, or (S1)m(S−1)n = (S−1)n−m = Sn−m, otherwise. Case 4: S−mSn = Sn−m. This is shown in a similar fashion to Case 3, using the equality (cid:3) S−1S1 = S0, and is therefore left to the reader. Corollary 40. If for every n ∈ Z, the S0-bimodule Sn is s-unital, then S is strongly Z-graded if and only if the equalities S1S−1 = S−1S1 = S0 hold. Corollary 41. If S is symmetrically Z-graded, then S is strongly Z-graded if and only if the equalities S1S−1 = S−1S1 = S0 hold. Corollary 42. If S is epsilon-strongly Z-graded, then S is strongly Z-graded if and only if ǫ1 = ǫ−1 = 1. 6. Strongly Z-graded Leavitt path algebras Throughout this section, E denotes a directed graph, R denotes an associative unital ring and we equip the associated Leavitt path algebra S = LR(E) with its canonical Z-gradation. Recall that if E is finite, then LR(E) is epsilon-strongly Z-graded by Theorem 28. In that case, we will use the notation established in Section 4 without further mention. Lemma 43. If E is a finite directed graph, then ǫ1 = Pv∈E 0\{sinks} v. Proof. From the definition of the relation (cid:22) it follows that {[α] α ∈ E1} is a set of minimal elements of X1/ ∼. Thus ǫ1 = X α∈E 1 N1(α) = X α∈E 1 αα∗ = X X αα∗ = X v. v∈E 0\{sinks} α∈E1 s(α)=v v∈E 0\{sinks} (cid:3) Lemma 44. If E is a finite directed graph which has no sink, then for every v ∈ E0 there exist n(v) ∈ N and m1,v, . . . , mn(v),v ∈ X−1 such that no pair of elements from [m1,v], . . . , [mn(v),v] has a common lesser element in X−1/ ∼, with respect to (cid:22), and v = Pn(v) i=1 N−1(mi,v) ∈ S−1S1. Proof. Take v ∈ E0. Since E is finite and has no sink, it is clear that we can make repeated use of property (4) in Definition 18 to find n(v) ∈ N and αi,v ∈ E∗, for i ∈ {1, . . . , n(v)}, such that r(αi,v) is a vertex belonging to a cycle Ci in E, and v = Pn(v) i,v. For each i ∈ {1, . . . , n(v)}, choose a path βi in Ci such that r(βi) = r(αi,v) and αi,vβ∗ i ∈ X−1. Then βiα∗ i,v ∈ X1. For each i ∈ {1, . . . , n(v)}, put mi = αi,vβ∗ i=1 αi,vα∗ αi,vα∗ i,v = n(v) X i=1 αi,vr(αi,v)α∗ i,v = i . Then n(v) X i=1 αi,vr(βi)α∗ i,v v = = n(v) X i=1 n(v) X i=1 αi,vβ∗ i βiα∗ i,v = n(v) X i=1 mi,vm∗ i,v = n(v) X i=1 N−1(mi,v) ∈ S−1S1. 12 GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS Notice that the paths α1,v, . . . , αi,n(v) are all distinct. Thus, no pair of elements from [m1,v], . . . , [mn(v),v] has a common lesser element in X−1/ ∼. (cid:3) We now prove Theorem 1 using an approach different from Hazrat's [6]. Proposition 45. If E is a finite directed graph, then LR(E) is strongly Z-graded if and only if E has no sink. Proof. The "only if" statement follows from Lemma 43, and also from Lemma 46. Now we show the "if" statement. We claim that for all v ∈ E0 the equality ǫ−1v = v holds. If we assume that the claim holds, then we get that ǫ−1 = ǫ−11 = Pv∈E 0 ǫ−1v = Pv∈E 0 v = 1. From Lemma 43 we get that ǫ1 = 1. Thus, by Corollary 42, we get that LR(E) is strongly Z-graded. Now we show the claim. Take v ∈ E0. By Lemma 44 we may write v = Pn(v) suitable n(v) ∈ N and m1,v, . . . , mn(v),v ∈ X−1 ⊆ S−1. Thus, ǫ−1v = ǫ−1 Pn(v) Pn(v) Lemma 46. Let E be an arbitrary directed graph. The following assertions hold: i=1 ǫ−1mi,vm∗ i,v = v. i,v = Pn(v) i=1 mi,vm∗ i=1 N−1(mi,v) for i=1 N−1(mi,v) = (cid:3) (a) If LR(E) is strongly Z-graded, then E has no sink. (b) If E is row-finite and has no sink, then S1S−1 = S0. (c) If E has no source, then S−1S1 = S0. Proof. (a) Suppose that S = LR(E) is strongly Z-graded. Seeking a contradiction, suppose that there is a sink v in E. Then v ∈ S0 = S1S−1. Using that v is a sink, we get that v = v2 ∈ vS1S−1 = {0}. This is a contradiction. (b) It suffices to show that E0 ⊆ S1S−1. Take v ∈ E0. Then v = Ps(f )=v f f ∗ ∈ S1S−1. (c) It suffices to show that E0 ⊆ S−1S1. Take v ∈ E0. Choose some f ∈ E1 such that (cid:3) v = r(f ) = f ∗f ∈ S−1S1. The following results generalizes [7, Theorem 1.6.15]. Theorem 47. Let E be a row-finite directed graph which has no source. Then LR(E) is strongly Z-graded if and only if E has no sink. Proof. The "only if" statement follows from Lemma 46(a). The "if" statement follows from Lemma 46(b)-(c), Proposition 34 and Corollary 41. (cid:3) References [1] G. Abrams, Leavitt path algebras: the first decade, Bull. Math. Sci. 5(1), 59 -- 120 (2015). [2] G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293(2), 319 -- 334 (2005). [3] P. Ara, M. A. Moreno and E. Pardo, Nonstable K-theory for graph algebras, Algebr. Represent. Theory 10(2), 157 -- 178 (2007). [4] L. O. Clark, R. Exel and E. Pardo, A generalized uniqueness theorem and the graded ideal structure of Steinberg algebras, Forum Math. 30(3), 533 -- 552 (2018). [5] M. Cohen and S. Montgomery, Group-graded rings, smash products and group actions, Trans. Amer. Math. Soc. 282(1), 237 -- 258 (1984). [6] R. Hazrat, The graded structure of Leavitt path algebras, Israel J. Math. 195(2), 833 -- 895 (2013). [7] R. Hazrat, Graded rings and graded Grothendieck groups, London Mathematical Society Lecture Note Series, 435. Cambridge University Press, Cambridge, (2016). [8] W. G. Leavitt, The module type of a ring, Trans. Amer. Math. Soc. 103, 113 -- 130 (1962). [9] C. Nastasescu and F. Van Oystaeyen, Methods of graded rings, Lecture Notes in Mathematics, 1836. Springer-Verlag, Berlin (2004). GROUP GRADATIONS ON LEAVITT PATH ALGEBRAS 13 [10] P. Nystedt, J. Öinert and H. Pinedo, Epsilon-strongly graded rings, separability and semisimplicity, J. Algebra 514, 1 -- 24 (2018). [11] J. Öinert and P. Lundström, The ideal intersection property for groupoid graded rings, Comm. Algebra 40(5), 1860 -- 1871 (2012). [12] M. Tomforde, Uniqueness theorems and ideal structure for Leavitt path algebras, J. Algebra 318(1), 270 -- 299 (2007). [13] M. Tomforde, Leavitt path algebras with coefficients in a commutative ring, J. Pure Appl. Algebra 215(4), 471 -- 484 (2011). [14] H. Tominaga, On s-unital rings, Math. J. Okayama Univ. 18(2), 117 -- 134 (1975/76).
1709.08465	1	1709	2017-09-25T13:00:23	Wedderburn principal theorem for Jordan superalgebras I	[ "math.RA" ]	We consider finite dimensional Jordan superalgebras $\jor$ over an algebraically closed field of characteristic 0, with solvable radical $\rad$ such that $\radd=0$ and $\jor/\rad$ is a simple Jordan superalgebra of one of the following types: Kac $\kac$, Kaplansky $\mathcal{K}_3$ superform or $\algdt$. We prove that an analogue of the Wedderburn Principal Theorem (WPT) holds if certain restrictions on the types of irreducible subsuperbimodules of $N$ are imposed, where $N$ is considered as a $J$-superbimodule, and $J$ is a simple Jordan superalgebra. Using counterexamples, it is shown that the imposed restrictions are essential.	math.RA	math	WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. G ´OMEZ-GONZ ´ALEZ F.A. Abstract. We consider finite dimensional Jordan superalgebras J over an algebraically closed field of characteristic 0, with solvable radical N such that N 2 = 0 and J/N is a simple Jordan superalgebra of one of the following types: Kac K10, Kaplansky K3 superform or Dt. We prove that an analogue of the Wedderburn Principal Theorem (WPT) holds if certain restrictions on the types of irreducible subsuperbimodules of N are imposed, where N is considered as a J/N -superbimodule. Using coun- terexamples, it is shown that the imposed restrictions are essential. Key Words: Jordan superalgebras, Wedderburn, Decomposition, Semisimple. 2010 Mathematics subject Classification: 17A15, 17A70, 17C50. 1. Introduction In 1892, T. Mollien [1] proved that for any finite-dimensional associative algebra A with nilpotent radical N over the complex field there exists a subalgebra S ⊆ A such that S ∼= A/N and A = S ⊕ N . This result was generalized in 1905 by J. H. Maclagan-Wedderburn [3] for all finite dimensional associative algebras over an arbitrary field. This result is known as the Wedderburn's Principal Theorem (WPT). Analogues of the WPT were proved for finite-dimensional alternative alge- bras by R. D. Schafer [4], and for finite-dimensional Jordan algebras by A. Albert, Penico, Askinuze, and Taft [5, 7, 8, 6]. Thus it is natural to try to extend this result to superalgebras. In the case of finite dimensional alternative superalgebras A over a field of cha- racteristic zero, Pisarenko [9] proved an analogue to the WPT. He proved that the theorem holds if some restrictions are imposed over summands in the semisimple superalgebra A/N . It was also shown with counter-examples that the restrictions are essential. In the current paper, we consider finite dimensional Jordan superalgebras A over a field of characteristic zero with radical N such that N 2 = 0 and A/N is a The author is thankful to Prof. Ivan Shestakov for his suggestion to solve the problem consi- dered in this paper as a part of the author's Doctoral Thesis and for other valuable advises. The author was partially supported by CAPES/CNPq IEL Nacional, Brazil, and Universidad de Antioquia, Colombia. 1 2 G ´OMEZ-GONZ ´ALEZ F.A. simple Jordan superalgebra of one of the following types: Kac K10, Kaplansky K3, superform or Dt. The cases of simple quotients of the types K10, superform, Dt, K3 are considered. It's proved that a Wedderburn decomposition is possible with certain essential restrictions This paper is organized as follows. In Section 2, the basic examples of Jordan superalgebras are given. Sections 3-6 contain the proof of the Main Theorem. In Section 3, the necessary reductions are done. Sections 4-6 are devoted to prove of theorem. Section 7, the main theorem is enounced. Note that the cases Mnm(F)(+) and Jospn2m(F), are considered in [19] and [20] respectively. The other cases, when J/N is isomorphic to JPn(F), Qn(F)(+), K3 ⊕ K3 ⊕ · · · ⊕ K3 ⊕ F · 1, and Kantor superalgebra, are to be considered in the next paper. We also stress that the Main Theorem implies that the second cohomology group H 2(J, N ) is not trivial for some simple Jordan superalgebra J and some irreducible J-superbimodule N . This gives one more subject of interest to be considered in future papers. 2. Jordan superalgebras, definition and some examples Throughout the paper, all algebras are considered over an algebraically closed field of characteristic zero F. Recall that an algebra A is said to be a superalgebra if it is a direct sum A = A0 ∔ A1 of vector spaces satisfying the relation AiAj ⊆ Ai+j(mod 2), i.e. A is a Z2-graded algebra. For an element a ∈ Ai, i = 0, 1, the number a = i denotes a parity of a. Let Γ = alg h 1, ei, i ∈ Z+eiej + ejei = 0 i be the Grassmann algebra. Then Γ = Γ0 ∔ Γ1, where Γ0 and Γ1 are the spans of all monomials of even and odd lengths, respectively. It is not difficult to see that Γ has a superalgebra structure. For a superalgebra A = A0 ∔ A1, we define the Grassmann envelope of A as follows: Γ(A) = Γ0 ⊗ A0 ∔ Γ1 ⊗ A1. Assuming that M is a homogeneous variety of algebras. The superalgebra A is said to be an M-superalgebra if the Grassmann envelope Γ(A) lies in M. Following this definition, one can consider associative, alternative, Lie, Jordan, etc. superalgebras. We recall that an algebra J is said a Jordan algebra if its multiplication satisfies the identity ab = ba of commutativity and the Jordan identity (a2b)a = a2(ba). In this paper, we consider algebras over a field characteristic zero. Thus, the Jordan identity is equivalent to its complete linearization ((ac)b)d + ((ad)b)c + ((cd)b)a = (ac)(bd) + (ad)(bc) + (cd)(ba). WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 3 An associative superalgebra is just a Z2-graded associative algebra, but it is not the case in general terms. It is easy to see that a Jordan superalgebra it is not a Jordan algebra. One can verify that a superalgebra J = J0 ∔ J1 is a Jordan superalgebra iff it satisfies the superidentities (2.1) (2.2) aiaj = (−1)ijajai, ((aiaj)ak)al + (−1)l(k+j)+kj ((aial)ak)aj + (−1)i(j+k+l)+kl((ajal)ak)ai = = (aiaj)(akal) + (−1)l(k+j)(aial)(aj ak) + (−1)jk(aiak)(ajal) for homogeneous elements at ∈ Jt, t ∈ {i, j, k, l}. We stress that, in view of the restriction on the characteristic of ground field, superidentity (2.1) yields that the Jordan superalgebra J = J0 ∔ J1 is a (Z2-graded) Jordan algebra iff (J1)2 = 0. Throughout the paper, we denote by ∔ a direct sum of vector space, by + denote a sum of vector space and by ⊕ we denote a direct sum of superalgebras. Some examples of Jordan Superalgebras. Let A be an associative superalgebra with multiplication ab. We define on the vector space A a new multiplication a ◦ b = 1 2 (ab + (−1)abba) for a, b ∈ A0 ∪ A1. It is not hard to verify that A gains a structure of Jordan superalgebra with respect to the defined multiplication. We denote this superalgebra by A(+). C.T.C Wall [14] proved that every associative simple finite-dimensional super- algebra over an algebraically closed field F is isomorphic to one of the following associative superalgebras: (i) A = Mnm(F), A0 =n(cid:16) a 0 (ii) A = Qn(F) = Q(n), A0 =n(cid:16) a 0 0 d (cid:17)o, A1 =n(cid:16) 0 0 (cid:17)o, 0 a (cid:17)o, A1 =n(cid:16) 0 h h 0 (cid:17)o. b c where a, h ∈ Mn(F), d ∈ Mm(F), b ∈ Mn×m(F), c ∈ Mm×n(F). (I) Applying the multiplication "◦" to the associative superalgebras Qn(F) and Mnm(F), we get the Jordan superalgebras Qn(F)(+) and Mnm(F)(+) respectively. (II) Let A be an associative superalgebra. A graded linear mapping ∗ : A −→ A is called superinvolution if (a∗)∗ = a and (ab)∗ = (−1)abb∗a∗. By H(A, ∗) denote the set of symmetric elements of A relative to ∗. Then H(A, ∗) is a Jordan superalgebra such that H(A, ∗) ⊆ A(+). Let In, Im be the identity matrices of order n and m respectively, t be the trans- position and U = −U t = −U −1 =(cid:16) 0 Im 0 −Im (cid:17). 4 G ´OMEZ-GONZ ´ALEZ F.A. Consider linear mappings Osp : Mn2m(F) −→ Mn2m(F) and σ : Qn(F) −→ Qn(F) given by (2.3) c d (cid:17)Osp (cid:16) a b (cid:16) a b d (cid:17)σ c 0 0 =(cid:16) In =(cid:16) dt −bt at ct (cid:17). U (cid:17)(cid:16) at −ct dt bt (cid:17)(cid:16) In 0 0 U −1 (cid:17), It is easy to check that Osp and σ are superinvolutions and its Jordan superalge- bras are H(Mn2m(F), Osp) and H(Qn(F), σ). We denote these superalgebras by Jospn2m(F) and JPn(F) respectively. One also may consider the following Jordan superalgebras. (III) The 4-dimensional 1-parametric family Dt = (F · e1 + F · e2) ∔ (F · x + F · y), i = ei, eix = xei = 1 with nonzero products given by e2 2 y, xy = −yx = e1 + te2. The superalgebra Dt is simple for t 6= 0. 2 x, eiy = yei = 1 (IV) The non unital 3-dimensional Kaplansky superalgebra K3 = F·e∔(F·x+F·y), with nonzero products ex = xe = 1 2 y, xy = −yx = e. The superalgebra K3 is simple. 2 x, ey = ye = 1 (V) Let V = V0 ⊕ V1 be a vector superspace. We say that a bilinear mapping f : V × V −→ F is a superform if f is symmetric over V0, skew-symmetric over V1, and satisfies f (V0, V1) = 0. Consider a superalgebra J = (F · 1 ⊕ V0) ∔ V1 with the unit 1 and the multiplication v · w = f (v, w) · 1, (v, w ∈ V ). If f is a non-degenerate superform and dim V0 > 1, then J is a simple Jordan superalgebra. (VI) The introduced by Kac 10-dimensional superalgebra K10 is a simple Jordan superalgebra. A detailed description of K10 is given in Section 4. (VII) I. Kantor [11] defined a simple Jordan superalgebra structure in the finite- dimensional Grassmann algebra generated by e1, . . . , en. V. Kac [10], proved that every simple finite-dimensional Jordan superalgebra over F is isomorphic to one of the superalgebras Mnm(F)(+), Qn(F)(+), Jospn2m(F), JPn(F), Dt, K3, K10, a superalgebra of superform or a Kantor superalgebra. A J-superbimodule M = M0 ∔ M1 is called a Jordan superbimodule if the corresponding split null extension E = J ⊕ M is a Jordan superalgebra [?]. Re- calling that the split null extension is a direct sum J ⊕ M of vector spaces with a multiplication that extends the multiplication in J through the action of J on M, while the product of two arbitrary elements in M is zero. Let M be a J-superbimodule. The opposite superbimodule Mop = Mop 0 ∔ Mop 1 1 = M0, and by the following action is defined by the conditions Mop 0 = M1, Mop WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 5 of J over Mop: a · mop = (−1)a(am)op, mop · a = (ma)op for all a ∈ J0 ∪ J1, m ∈ Mop 1 . Whenever M is a Jordan J-superbimule, Mop is a Jordan one as well. 0 ∪Mop Let A = J as a vector superspace and let am, ma with m ∈ J, a ∈ A be the products as defined in the superalgebra J. It is easy to see that A has a natural structure of J-superbimodule. We call A a regular superbimodule. The irreducible superbimodules over the Jordan superalgebras of superform, Jospn2m(F), JPn(F), Mnm(F)(+), were classified by C. Martinez and E. Zelmanov [17]. E. Zelmanov, C. Martinez and I. Shestakov [18], classified the irreducible su- perbimodules for Jordan superalgebras Qn(F)(+). Irreducible superbimodules for Jordan superalgebra Dt and K3 were classified by C. Martinez and E. Zelmanov [16] and independently by M. Trushina in [15]. C. Martinez and I. Shestakov [?], classified the irreducible superbimodules over the Jordan superalgebra M11(F)(+) and Shtern classified the irreducible superbimodules over Jordan superalgebras of type K10, and Kantor superalgebra γ(e1, . . . , en), n ≥ 4 [13]. The Peirce decomposition Recall, that if J is a Jordan (super)algebra with unity 1, and {e1, . . . , en} is a set of pairwise orthogonal idempotents such that i=1 ei, then J admits Peirce decomposition [17], it is 1 =Pn J =(cid:18) nMi=1 Jii(cid:19)M(cid:18)Mi<j Jij(cid:19), where Jii = { x ∈ J : x }, if i 6= j are the Peirce components of J relative to the idempotents ei, and ej, moreover the following relations hold when i 6= k, l; j 6= k, l eix = x, } and Jij = { x ∈ J : ejx = 1 2 eix = x, 1 2 J2 ij ⊆ Jii + Jjj , Jij · Jjk ⊆ Jik, Jij · Jkl = 0. 3. Preliminary Reductions for WPT As in the case of Jordan algebras, we can make some restrictions before the main proof. To start we prove the following proposition. Proposition 3.1. Let J be a Jordan superalgebra without 1 and with radical N . If the WPT is valid for J#, then it is also valid for J. Proof. Let J be a Jordan superalgebra without 1 and radical N . Consider J# = J ⊕ F · 1. It is clear that N (J) = N (J#) = N and J#/N = (J/N )# = J/N ⊕ F · ¯1. By the condition, there exists S1 ⊆ J#, S1 ∼= J#/N ∼= (J/N )#, S1 ∩ N = (0), J# = S1 ⊕ N . Denote S = S1 ∩ J, then S ∩ N = (0). Let us show that S ⊕ N = J. Take a ∈ J, then a = s1 + n, s1 ∈ S1, n ∈ N . But s1 = a − n ∈ J. Hence, s1 ∈ J ∩ S1 = S and a ∈ S ⊕ N . Finally, S ∼= S/(S ∩ N ) ∼= (S ⊕ N )/N ∼= J/N . (cid:3) 6 G ´OMEZ-GONZ ´ALEZ F.A. Let J be a unital Jordan superalgebra of dim J = n. Assume that for any unital Jordan superalgebra of dimension less that n the WPT is true. A base for induction is dimF J = 1, J = F · 1. Proposition 3.2. Let J/N = J1 ⊕ · · · ⊕ Jk, where Ji are unital simple Jordan superalgebras with N (Ji) = 0. If k > 1, then the WPT is true for J. Proof. Denote by ei the identity elements in Ji. Then (by Jordan algebras results) there are orthogonal idempotents fi ∈ J such that ei = fi + N , i = 1, 2, . . . , k. Consider J1(fi) = {fi, J, fi}, then J1(fi)/(J1(fi)∩N ) ∼= Ji. By virtue of N (Ji) = 0, we have the inclusion N (J1(fi)) ⊆ J1(fi)∩N . Since the inverse inclusion is obvious, we have the equality N (J1(fi)) = J1(fi)∩N . If k > 1, then dim J1(fi) < dim J and by the inductive hypothesis, there exists Si ⊆ J1(fi), Si ∼= Ji/(N ∩ Ji). Note that Si · Sj = 0. Further, S = S1 ⊕ · · · ⊕ Sk is a direct sum and S ∼= J1 ⊕ · · · ⊕ Jk. (cid:3) Now by the Zelmanov Theorem [12], in the case of characteristic zero, it is sufficient to prove the WPT for unital finite dimensional Jordan superalgebras J satisfying one of the following conditions: (1) J/N is simple unital; (2) J/N = (K3 ⊕K3 ⊕· · ·⊕K3)⊕ F·1, where K3 is the Kaplansky superalgebra. Theorem 3.3. Let J be a finite dimensional semisimple Jordan superalgebra, i.e N (J) = 0, where N is the solvable radical. Let M(J) be a class of finite dimensional Jordan J-superbimodules N such that M(J) is closed with respect to subsuperbimod- ules and homomorphic images. Denote by K(M, J) the class of finite dimensional Jordan superalgebras A that satisfy the following conditions: (1) A/N (A) ∼= J, (2) N (A)2 = 0, (3) N (A) considered as a J-superbimodule, in M(J). Then if WPT is true for all superalgebras B ∈ K(M, J) with the restriction that the radical N (B) is an irreducible J-superbimodule, then it is true for all superalgebras A from K(M, J). Proof. We use the induction on dim A. The base of induction is provide by the case dim A = dim J, so A = J, N (A) = 0. Assume that the theorem is true for all Jordan superalgebras B ∈ K(M, J) with dim B ≤ dim A. Let us set by N = N (A). If N is an irreducible J-superbimodule, then the theorem is true by the conjecture. Suppose that N is not irreducible, then let us take a minimal J-superbimodule M contained in N . Since that A is unital JM = AM = M, therefore N is irreducible. Observe that N /M 6= 0, otherwise N = M would be irreducible. We see that A/M N /M ∼= A/N ∼= J. Since A/N is semisimple, we have that N (A/M) ⊆ N /M. But (N /M)2 = 0. Thus N (A/M) = N /M. Observe that A/M ∈ K(M, J) and dim A/M ≤ dim A. WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 7 ∼= A/N Therefore there exists a subsuperalgebra S ⊆ A/M such that S ∼= A/M N /M and A/M = S ⊕ N /M. By the main theorems on homomorphisms, there is a subsuperalgebra S ⊆ A such that M ⊆ S and S/M ∼= S ∼= A/N ∼= J. We observe that S ∈ K(M, J) and N (S) = M is an irreducible J-superbimodule. By the assumption, WPT is true for S, hence there is a subsuperalgebra S1 ⊆ S ⊆ A, such that S1 ∼= S/M ∼= A/N . Since S1 is semisimple, N ∩S ⊆ N (S1) = 0. Furthermore, dim S1 = dim A − dim N . Hence, dim(N + S1) = dim A and A = N ⊕ S1. (cid:3) Let V1, . . . , Vk be irreducible J-superbimodules, and J be a simple Jordan super- algebra. Let M(J; V1, . . . , Vk) = {V / V is a J-superbimodule, doesn't containing amoung its irreducible subsuperbimodule any copy isomorphic to one of the su- perbimodules V1, . . . , Vk} It is clear that M is closed with respect to taking of subsuperbimodules and homomorphic images. Thus it satisfies the conditions of Theorem 3.3. In each section, we assume that A is a finite dimensional Jordan superalgebra over F, with radical N and such that N 2 = 0, A/N ∼= J, where J is a simple Jordan superalgebra and N is an irreducible J-superbimodule. Moreover, if b1, b2, . . . , bn J1, we can assume that ¯v1, . . . , ¯vk is an additive base of A1/N1, and we shall find is an additive base of J0, then we assume thateb1,eb2, . . . ,ebn is an additive base of A0, moreover,ebi ·ebj = gbibj. If A1/N1 ∼= J1 and v1, . . . , vk is an additive base of ev1, . . . ,evk additive base of A1 such that evi ·evj = gvivj and evi ·ebj = gvibj. In each case we can assume thatea · n = an, whereea ∈ A0 ∪A1, a ∈ J0 ∪J1, n ∈ N0 ∪N1 4. Kac superalgebra In this section, we consider the 10-dimensional Kac superalgebra K10 = J0 ∔ J1, 4Xi=1 where J0 = (F · e + F · vi) ⊕ F · f , J1 = F · x1 + F · x2 + F · y1 + F · y2, and all nonzero products of the basis elements are the following (4.1) (4.2) (4.3) e2 = e, e · vi = vi, f 2 = f, v1 · v2 = v3 · v4 = 2e. f · xj = 1 2 xj, y1 · v1 = x2, x2 · v3 = x1, 2 yj, e · xj = 1 2 xj , f · yj = 1 2 yj, y2 · v1 = −x1, x1 · v2 = −y2, x2 · v2 = y1, y1 · v3 = y2, y2 · v4 = y1. x1 · v4 = x2, e · yj = 1 x1 · x2 = v1, x1 · y2 = v3, x2 · y1 = v4, y1 · y2 = v2, xi · yi = e − 3f. The zero characteristic of the ground field implies that K10 is a simple Jordan superalgebra. Consider the regular superbimodule over K10, Reg (K10) and assume that a ↔ e, b ↔ f , ui ↔ vi, mj ↔ xj, and nj ↔ yj for i = 1, 2, 3, 4, j = 1, 2, thus (Reg K10)0 = (F · a + F · u1 + F · u2 + F · u3 + F · u4) ⊕ F · b, (Reg K10)1 = F · m1 + F · m2 + F · n1 + F · n2 8 G ´OMEZ-GONZ ´ALEZ F.A. i=1 (4.4) λx1e Lemma 4.1. in (2.2), we get m1 m1 + λx1e m2 m2 + λx1e e = 1 2 Λxi It is easy to see that Λxi F·evi ⊕F·ef , and (K10)0 ∼= S0 and A1/N1 = F · ¯x1 + F · ¯x2 + F · ¯y1 + F · ¯y2. ef ·exj = 1 2exj, ey1 ·ev1 =ex2, ex2 ·ev3 =ex1, Proof. Firts prove eeex1 = 1 s1 s1, such thateeex1 = 1 Let A0 = (S0 ⊕N0) and A1/N1 ∼= (K10)1. Assume that S0 = F·ee+P4 ef ·eyj = 1 ee ·eyj = 1 ee ·exj = 1 2eyj, 2eyj, 2exj , ey2 ·ev1 = −ex1, ex1 ·ev2 = −ey2, ex2 ·ev2 =ey1, ey1 ·ev3 =ey2, ey2 ·ev4 =ey1, ex1 ·ev4 =ex2, 2ex1. To start, we can assume that there exist scalars 2ex1 + λx1e 2ex1 + Λx1 e ·ee = 1 e . Substituting ai =ex1 and aj = ak = al =ee 2((ex1 ·ee) ·ee) ·ee +ex1 ·ee = 3(ex1 ·ee) ·ee. 2ex1 + Λx1 2ex2, exi · ef = 1 e = 0 and ex1 ·ee = 1 2ex1. 2exi, eyi ·ee = 1 2eyi and Combining the above equality with ex1 ·ee = 1 Similarly one can prove the equalities ex2 ·ee = 1 eyi · ef = 1 2eyi. part in the productexi ·evj where Λij s ·evj ) ·evj = 0 for s = x or s = y. We set ai =ey1 and aj = ak = al =ev1 in (2.2). Sinceev2 Now we shall prove that others equalities in (4.4) hold. Let Λij m2 , λijx Firts note that (Λij x be the radical n2 n2 for some n2 = 0. Thus, Λx1 n2 . (Similarly, Λij e , we have 5 i = 0, we have therefore, λx1e n1 n1 + λx1e m1 m1 + λijx m2 m2 + λijx n1 and λijx n1 n1 + λijx e = 3Λx1 e , m1 = λx1e n1 = λx1e m2 = λx1e scalars λijx x = λijx m1 , λijx n2 n2. 2 Λx1 y .) 0 = ((ey1 ·ev1) ·ev1) ·ev1 = ((ex2 + Λ11 y ) ·ev1) ·ev1 = (ex2 ·ev1) ·ev1 = Λ21 n1 m2 − λ21x n2 = 0 and therefore Λ21 m2 m2, Λ12 n2 m1 = 0. The linear independence of m1 and m2 implies m2 m2. Similarly one can prove n2 n2, Λ22 n2 n2, n2 n2 and m1 m1 + λ21x n1 n1 + λ12y x = λ21x y = λ12y m1 m1 + λ11x m1 m1 + λ24x m2 m2 + λ23y n1 n1 + λ22y n1 n1, Λ24 x = λ24x y = λ22y y = λ23y x ·ev1. x = λ11x m2 m2 + λ13x m1 m1 + λ14y n1 n1, Λ23 n2 n2. Thus, λ21x n1 = λ21x λ21x that Λ11 Λ13 Λ14 x = λ13x y = λ14y y = λ12y n1 n1 + λ12y Observe that Λ12 n2 n2, therefore Λ12 (4.5), we obtain the equality Substituing ai =ey1, aj =ev1 and ak = al =ev2 in (2.2), we have, (4.5) ((ey1 ·ev1) ·ev2) ·ev2 + ((ey1 ·ev2) ·ev2) ·ev1 + ((ev1 ·ev2) ·ev2) ·ey1 = 2(ev1 ·ev2) · (ey1 ·ev2) y ·ev2 = 0. Recall thatev1 ·ev2 = 2ee, ee ·evi =evi andee · (ey1 ·ev2) = 1 2ey1 ·ev2. Thus, combining the above observation with x ·ev2 = Λ12 Taking ai =ey1, aj =ev1 and ak = al =ev3 in (2.2) we have, 0 = ((ey1 ·ev1) ·ev3) ·ev3 + ((ey1 ·ev3) ·ev3) ·ev1. 0 =((ey1 ·ev1) ·ev2) ·ev2 = Λ12 therefore (λ12y linearly independent, we obtain λ12y m1 ) n2 = 0. Using the fact that n1 and n2 are x ·ev2, m2 ) n1 + (λ12y m2 and λ12y n1 = −λ22x n2 = λ22x m1 . n1 + λ22x n2 − λ22x y + Λ22 y + Λ22 WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 9 Thus we obtain λ13x n2 = −λ23x n1 . m2 = −λ22y λ12x Using 0 = ((ey2 ·ev1) ·ev2) ·ev2 + ((ey2 ·ev2) ·ev2) ·ev1, we obtain λ12x n1 . Since, 0 = ((ey2 ·ev1) ·ev4) ·ev4 + ((ey2 ·ev4) ·ev4) ·ev1, then λ24x Similarly, we obtain λ21x m2 = −λ13x λ23x −λ21y Thus, we have Λ21 m1 , λ12y m2 = −λ24y n1 = −λ12x m1 , λ14y x = 0, Λ12y = Λ22y = λ12y m1 = λ11y m2 , λ12y n1 = −λ24y n2 , λ21x n2 = λ12x n2 , λ22y m2 = −λ11y m1 , λ24x n2 = −λ12x n1 , λ23y m2 = −λ14x m1 , λ13x n1 n1 and Λ11 m1 = −λ13y m1 , λ11x m1 = −λ23x y = λ11y m1 m1 + λ11y n2 , λ14y m2 , λ23y m2 = λ21y m2 , λ11y n2 = −λ13y n1 , n1 , λ11x m1 = n1 = 0. n2 = λ11y m1 = λ22y n2 and n1 = −λ14x n2 . m2 m2. Let ai = ey1, aj = al = ev1 and ak = ev2 in (2.2). Then we have ey1 ·ev1 = ((ey1 ·ev1) ·ev2) ·ev1, therefore, λ22x n2 = 0, λ12x m2 and λ22x n1 = −λ11y n1 = λ14x n2 = λ11y n2 , λ11x n1 = λ12x m2 , λ24y m1 . Similarly, we can obtain m1 = −λ21y n1 , n1 = −λ24y m2 , n1 = −λ22x n2 , n2 = 0, λ13x m1 = 0, λ12y n2 , λ14y n1 = −λ13y m2 = −λ23x n2 , λ21y m1 = −λ22x n1 , λ24x m1 = λ21y m2 , λ23y m2 = 0, λ23y n2 = 0, λ14x m2 = −λ23x n1 = λ12x n1 = λ21y m1 , λ24y m1 = λ23x λ11x n2 = 0, λ14x λ13x n1 = −λ13x λ22y m1 = λ24y λ11y n2 . Thus, we have Λ12 m2 , λ13y x = Λ13 x = Λ24 x = Λ12 y = Λ22 y = Λ23 y = 0. Setting ai =ey1, aj =ev1, ak =ev3 and al =ev2 in (2.2), we obtain ((ey1 ·ev1) ·ev3) ·ev2 + ((ey1 ·ev2) ·ev3) ·ev1 +ey1 ·ev3 = 0. n2 . Similarly one can prove the equalities λ11x y = Λ24 Therefore we have λ11y n1 = 0. Thus Λ11 λ22x m2 = λ13y x = Λ22 x = Λ23 y = Λ13 y = Λ14 y = Λ21 x = Λ14 x = Λ11 m1 = λ23x y = 0. m1 = (cid:3) Lemma 4.2. There exist α ∈ F such that (4.6) x , Λ12 Proof. We can assume that there exist Λ12 (i) ex1 ·ex2 =eu1 + αu1, (iii) ex1 ·ey2 =ey3 + αu3, (v) ex1 ·ey1 =ee − 3ef + αa − 3αb x , ey1 ·ey2 =ev2 + Λ12 xy and ex2 ·ey2 =ee − 3ef + Λ22 (ii) ey1 ·ey2 =ey2 + αu2 (iv) ex2 ·ey1 =ey4 + αu4 (vi) ex2 ·ey2 =ee − 3ef + αa − 3αb xy, ex2 ·ey1 =ev4 + Λ21 x , ex1 ·ey2 =ev3 + Λ12 We assume that there exist ηtij t = ηtij that ex1 ·ex2 =ev1 + Λ12 ex1 ·ey1 =ee − 3ef + Λ11 Replacing ai =ex1, aj =ex2, ak = al =ev1 in the equation (2.2) and using (4.4); we have ((ex1 ·ex2) ·ev1) ·ev1 = 0, thus u1 , ηtij u4 ∈ F such that u4 u4 for i, j ∈ {1, 2} and t ∈ {x, y, xy}. xy ∈ N0 such xy, b , ηtij u3 u3+ηtij u2 u2 + ηx12 u1 u1 + ηx12 u3 u3 + ηx12 a a + ηx12 u3 and ηtij xy and Λ22 u1 u1+ηtij u2 u2+ηtij a a+ηtij b b+ηtij a , ηtij u2 , ηtij b + ηx12 xy, Λ11 xy, Λ21 y , Λ12 Λij xy. b u4 u4) ·ev1) ·ev1 u2 = 0. In the same way one can prove that ηx12 u4 = ηx12 u3 = 0, thus 0 =((ev1 + ηx12 =(ηx12 (4.7) u2 u1. Therefore, ηx12 a u1 + 2ηx12 u2 a) ·ev1 = 2ηx12 ex1 ·ex2 =ev1 + ηx12 Similarly, we obtain that ey1 ·ey2 =ev2 + ηy12 a a + ηx12 b b + ηx12 u1 u1. a a + ηy12 b b + ηy12 u2 u2. 10 G ´OMEZ-GONZ ´ALEZ F.A. b a b . u3 (4.8) b + ηy12 = ηxy21 u3 = 0. and ηxy21 a = ηx12 a = ηy12 u1 = ηxy21 u2 = ηxy21 a a + ηy12 u4 = ηxy12 u1 = ηxy12 b = 0 and ηx12 u1 u2 = u2 . If we u2 u2), therefore ηy12 above case we obtain ηxy21 u1 = ηy12 b = 0. u2 = 0. If we u4 = 0. Similarly to Since (4.4) and replacing ai =ex1, aj =ex2 and ak = at =ev2 in (2.2), we obtain ((ex1 ·ex2) ·ev2) ·ev2 = 2(ex1 ·ev2) · (ev2 ·ex2) = 2ey1 ·ey2. Replacing (4.7) and its equivalent for ey1 ·ey2 in (4.8), we obtain 2ev2 + ηx12 2(ev2 + ηy12 take ai =ey1, aj =ey2 and ak = at =ev1 in (2.2) we obtain ηx12 Let ai =ex1, aj =ey2 and ak = at =ev1 in (2.2), thus, we obtain ηxy12 shall take ak = at =ev2 or ak = at =ev3 we obtain ηxy12 Setting ai =ev1, aj = ey2 and ak = at =ev4 (respectively, ai = ex2, aj = ey1 and ak = at =ev3) in (2.2), we obtain ηxy21 If we take ai =ex1, aj =ex2, ak =ev2 and at =ev4 in (2.2), then using (4.4) we have ((ex1 ·ex2) ·ev2) ·ev4 = 2ex2 ·ey1. Therefore, ηxy21 u1 . Thus we getex1 ·ex2 =ev1 + αu1, ey1 ·ey2 =ev2 + αu2, ex1 ·ey2 =ev3 + αu3 and ex2 ·ey1 =ev4 + αu4 for some α ∈ F. Let ai = ex1, aj = ey1, and ak = at = ev1 in (2.2). Using the products in S0 and (4.4), we obtain ((ex1 ·ey1) ·ev1) ·ev1 = 0. Thus ((ex1 ·ey1) ·ev1) ·ev1 = 0 and Thus ex1 ·ey1 = ee − 3ef + ηxy11 . Similarly one can show that ex2 ·ey2 = ee − 3ef + ηxy22 Taking ai =ex1, aj =ey1, ak =ev1 and at =ev2, in (2.2), we obtain ((ex1 ·ey1) ·ev1) ·ev2 +ex1 ·ey1 = 2ee · (ex1 ·ey1) +ex2 ·ey2. u2 = 0. Analogously, one can verify that ηxy11 From the above equality, it is easy to see that ηxy11 = 0 (respectively ηxy12 therefore ηxy11 u1 = ηxy11 u3 = ηxy11 u4 = ηx12 and ηxy11 a + ηxy11 a + ηxy22 u4 = 0. = ηxy22 = ηxy12 = ηxy22 (4.9) . b . b a a a b = 0) a b a b b a 1 2 (ηxy11 a + ηxy11 Thus we have thatex1 ·ey1 =ex2 ·ey2. Let ai =ex1, aj =ex2 and ak = at = wty1 in (2.2), hence 0 =((ex1 ·ex2) ·ey1) ·ey1 − ((ex1 ·ey1) ·ey1) ·ex2 + ((ex2 ·ey1) ·ey1) ·ex1 =((ev1 + αu1) ·ey1) ·ey1 − ((ee − 3ef + ηxy11 b) ·ey1) ·ex2 + ((ev4 + αu4) ·ey1) ·ey1 )n1) ·ex2 =(ey1 ·ev1 + αey1 · u1) ·ey1 − (−ey1 + )ex2 · n1 = (α + =ex2 ·ey1 + αm2 ·ey1 −ex2 ·ey1 + If we take ai =ex1, aj =ey1, ak =ex2 and at =ey2 in (2.2), we obtain ηxy11 Lemma 4.3. There exists β ∈ F such that exi = xi + β mi, and eyi = yi + β ni. for = −3α, thereforeex1 ·ey1 =ex2 ·ey2 =ee − 3ef + αa − 3αb. thus, 2α = −(ηxy11 and ηxy11 + ηxy11 + ηxy11 + ηxy11 = α (cid:3) + ηpq11 i = 1, 2. (ηxy11 (ηxy11 ))u4, 1 2 1 2 ). a a a a a b b b b b WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 11 λti t = λti m1 m1 + λti y, where Λi x and Λi m2 m2 + λti Proof. Assume that there exist Λi eyi = yi + Λi n2 ∈ F. It is easy to see thatexi = xi + λxi Using the Lemma 4.1, we have that ex1 ·ev2 = −ey2, ex1 ·ev4 = −ex2 and ey2 ·ev4 =ey1. y ∈ N1 such that exi = xi + Λi mi mi andeyi = yi + λyi Thus one easily verifies that λx1 n1 = λy2 λy1 n1 . Therefore, (cid:3) x and n1 and n2 n2 and λti m2 and λy2 n2 , λx1 m1 = λx2 n2 = λx1 m1 = λx2 m2 . m1 = λy2 n1 n1 + λti m1, λti m2 , λti ni ni. n2 = λy1 Let us prove the following theorem Theorem 4.4. Let A be a finite dimensional Jordan superalgebra with solvable radical N such that N 2 = 0 and A/N ∼= K10. Then there exists a subsuperalgebra S ⊆ A such that S ∼= K10 and A = S ⊕ N . Proof. Recall that A. S. Shtern [13] proved that any irreducible Jordan superbimo- dule over K10 is isomorphic to Reg (K10). ByTheorem 3.3, we only need to consider this case. By Lemma 4.2, we can assume that there exists α ∈ N such that (4.10) It is easy to verifies the following equalities (ii) ey1 ·ey2 =ey2 + αu2 (i) ex1 ·ex2 =eu1 + αu1, (iv) ex2 ·ey1 =ey4 + αu4 (iii) ex1 ·ey2 =ey3 + αu3, (vi) ex2 ·ey2 =ee − 3ef + αa − 3αb (v) ex1 ·ey1 =ee − 3ef + αa − 3αb By Lemma (4.3), there is a β ∈ F such thatexi = xi + β mi and eyi = yi + β ni. ex1 ·ey1 = x1 · y1 + 2β(a − 3b), ex2 ·ey2 = x2 · y2 + 2β(a − 3b), ey1 ·ey2 = y1 · y2 + 2βu2, ex2 ·ey1 = x2 · y2 + 2βu3. ex1 ·ex2 = x1 · x2 + 2βu1, ex1 ·ey2 = x1 · y2 + 2βu3, Using (4.10) and (4.11), we get exi ·eyi = ee − 3ef , ex1 ·ex2 = ev1, ex1 ·ey2 = ev3, ex2 ·ey1 = ev4 and ey1 ·ey2 = ev2 if and only if, 2β = α. This equality has always a solution. Therefore the WPT holds in the case under consideration. (4.11) (cid:3) 5. Jordan superalgebra of superform. In this section we use the classification of irreducible J-bimodules obtained by E. Zelmanov and C. Martinez in [17], where J = J(V, f ) = (F·1⊕V0) ∔V1 be a Jordan superalgebra of nondegenerate super-symmetric superform f on a superspace V. We may assume that dim V1 > 1. Let v1, . . . , vn be an f -orthonormal basis of V0, i.e. f (vi, vi) = 1, f (vi, vj) = 0 for i 6= j, i, j = 1 . . . , n. Let w1, . . . , w2m be a basis of V1 such that f (w2p−1, w2p) = 1, 1 ≤ p ≤ m, and all the other products of basis elements are zero. We know that all products vi1 2m form a basis of C, where i1, . . ., in ∈ {0, 1} and k1, . . . , k2m are nonnegative integers and C denotes the Clifford superalgebra of V. Let Cr be the subspace in C spanned by the products of basis 1 · · · wk2m 1 · · · vin n wk1 12 G ´OMEZ-GONZ ´ALEZ F.A. elements of length at most r, and let J = (F·1+V0)∔V1 be the Jordan superalgebra of superform f . Let a be an even vector, V ′ = V ⊕ F · a. We extend the superform f to V ′ so that f (a, a) = 1, f (a, V) = 0. Denote by C′ r the subspace in C′ defined in the same way as Cr in C. 1 · · · vin n wk1 1 · · · wk2m In this section, for every element vi1 2m of the basis of C, we put into correspondence a pair (I, K), where I = (i1, . . . , in) is a n-tuple and K = (k1, . . . , k2m) is a 2m-tuple where is, kt satisfies the above conditions. We write ηI,K = vi1 2m = VI WK . Note that for any pair of elements ηI,K, ηI ′,K ′ ∈ C, the following relation holds ηI,K = ηI ′,K ′ if and only if I = I ′, and K = K ′. Thus every element of the basis of C has a unique representation in terms of (I, K). We denote V(0) = 1, V(1) = v1v2 · · · vn. 1 · · · wk2m 1 · · · vin n wk1 Let I, K be the following sets I = {I = (i1, . . . , in), ij = 0 or 1, j = 1, . . . , n }, K = {K = (k1, . . . , k2m), kj ∈ Z+ ∪ {0}, j = 1, . . . , 2m }, For I ∈ I, K ∈ K, we denote I = i1 + · · · + in, K = k1 + · · · + k2m and ηI,K = I + K. Some relations in C(+). From symmetric product in superalgebra C+ we have (5.1) (5.2) VI WK ◦ vj =(cid:16) − 1 2(cid:17)i1+···+ij−1 V(i1,...,ij +1,...,in)WK (1 + (−1)ηI,K −ij ), VI WK ◦ wp = 1 2 VI W(k1,...,kp+1,...,k2m)(1 + (−1)ηI,K )− kp+1VI W(k1,...,kp+1−1,...,k2m) for j = 1, . . . , n and p = 1, 3, . . . , 2m − 1. We note that a similar relation to (5.2) with some change of signs holds for even p. In this section, A0 = (S0 ⊕ N0) and (S1/N1) ∼= J1. Assume that S0 = F · 1 + F · ev1 + · · · + F ·evn, J0 ∼= S0 and A1/N1 = F · ¯w1 + F · ¯w2 + · · · + F · ¯w2m. We consider two cases for N . 5.1. N is isomorphic to Cr/Cr−2. Without loss of generality, we can take N0 = vect F h ηI,K, ηI,K = r, r − 1 and K is even i, N1 = vect F h ηI,K, ηI,K = r, r − 1 and K is odd i. WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 13 Using the notation introduced above, due to the equations (5.1) and (5.2), we have the following products: (5.3) (5.4) ±V(i1,...,ij−1,1,ij+1,...,in)WK ηI,K ·evj =( ±V(i1,...,ij−1,0,ij+1,...,in)WK ηI,K · ewp =( ±kp±1VI W(k1,...,kp±1−1,...,k2m) VI W(k1,...,kp+1,...,k2m) if if Firts, we prove three lemmas. ηI,K = r, ij = 1, ηI,K = r − 1, ij = 0. if if ηI,K = r, ηI,K = r − 1. Lemma 5.1. evj · ews = 0. Proof. Setting ai = ews, aj = al =evi and ak =evj in (2.2), we have (5.5) we may assume that there exist some scalars ξk (I,K) such that ((ews ·evi) ·evj ) ·evi = 0. (I,K)VI WK + XI, K ξk ηI,K =r−1 K odd evk · ews = XI, K ηI,K =r K odd ξk (I,K)VI WK. Let ξk (I,k)ηI,K ∈ N1 be a nonzero element and j 6= k, j, k ∈ {1, 2, . . . , n}. Using (5.3) and (5.5), we obtain the following relations: (a) If ηI,K = r − 1 and ij = 1, then (ξk (b) If ηI,K = r − 1, ij = ik = 0, then (ξk (c) If ηI,K = r, ij = 0, then (ξk (d) If ηI,K = r, ij = ik = 1, then (ξk From (a) - (d) and (5.5), we have (I,K)ηI,K ·evj) ·evk = 0. (I,K)ηI,K ·evj) ·evk = 0. (I,K)ηI,K ·evj) ·evk = 0. (I,K)ηI,K ·evj) ·evk = 0. (I,K)V(i1,...,ij−1,0,ij+1,...,in)WK + XI, K XI, K XI, K ηI,K =r K odd K odd ηI,K =r−1 ηI,K =r K odd evk · ews = XI, K K odd ηI,K =r−1 ξk ξk (I,K)V(i1,...,ij−1,1,ij+1,...,in)WK+ ξk (I,K)V(i1,...,ij−1,0,ij+1,...,ik−1,0,ik+1,...,in)WK + ξk (I,K)V(i1,...,ij−1,0,ij+1,...,ik−1,0,ik+1,...,in)WK = 0. 14 Thus, G ´OMEZ-GONZ ´ALEZ F.A. evk · ews = XI, K XI, K K odd (I,K)vi1 ξk ηI,K =r−1 ηI,K =r K odd n WK + XI, K n WK + XI, K ηI,K =r K odd ηI,K =r−1 K odd (I,K)vi1 ξk 1 · · · vj · · · vin (I,K)vi1 ξk 1 · · · vj · · · vk · · · vin n WK+ 1 · · · vij−1 j−1 vij+1 j+1 · · · vin (I,K)vi1 ξk 1 · · · vj · · · vk · · · vin n WK = 0. ηI,K =r−1 If we apply (5.5) to the obtained above equation for all j 6= k, we get (5.6) (I,K)v1 · · · vik ξk k · · · vnWK+ XK evk·ews = XI, K Substituting ai by ews and aj = ak = al byevk respectively in (2.2), we obtain (0,K)WK+ XK K=r−n K odd K=r K odd (5.7) K odd ξk ξk (1,K)V(1)WK. Applying (5.7) to (5.6), to get ((ews ·evk) ·evk) ·evk = zs ·evk. (5.8) (I,K)v1 · · · vik ξk k · · · vnWK. evk · ews = XI, K K odd I+K=r−1 Lemma 5.2. Substituting ai , aj, ak, and al by ews,evk,evj andevj respectively in (2.2), we have ((ews ·evk) ·evj) ·evj + ((ews ·evj) ·evj ) ·evk = ews ·evk Applying the obtained equality to (5.8), we haveevk · ews = 0. ewp · ewq = αp,q Proof. Since ewp ·ewq ∈ E0, we can assume that there exist some scalars αp,q (0,K)WK + XK αp,q (1,K)V(1)WK , where αp,q 0 + XK (0,K), αp,q 0 ∈ {0, 1} such that K=r−n K=r−1 αp,q n odd (I,K) (cid:3) ewp · ewq = αp,q 0 + XI,K ηI,K =r−1 K even αp,q (I,K)VI WK + XI,K ηI,K =r−1 K even αp,q (I,K)VI WK, (5.9) where αp,q 0 is 0 or 1. If we take ai = ewp, aj = ewq and ak = al =evi in (2.2), and use Lemma 5.1, then we obtain ((ewp ·ewq) ·evi) ·evi = ewp ·ewq. Combining (5.9) in the stated before equality, WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 15 we have XI,K ηI,K =r−1 K even (5.10) αp,q (I,K)(ηI,K ·eui) ·evi + XI,K XI,K ηI,K =r−1 K even ηI,K =r−1 K even αp,q (I,K)(ηI,K ·eui) ·evi = (I,K)ηI,K + XI,K αp,q ηI,K =r−1 K even αp,q (I,K)ηI,K. Let η = ηI,K be a nonzero element in N0. Using equality (5.3), one can easily prove the following statments (i) If ηI,K = r − 1, and ij = 0, then (η ·evj) ·evj = η. (ii) If ηI,K = r − 1, and ij = 1, then η ·evj = 0 therefore, (η ·evj ) ·evj = 0. (iii) If ηI,K = r, and ij = 0, then (η ·evj) = 0, thus (η ·evj ) ·evj = 0. (iv) If ηI,K = r, and ij = 1, then (η ·evj) ·evj = η. Using statments (i) - (iv), we note that if η = r − 1 and ij = 1 for some j then the left part of (5.10) is equal to zero and, consequently, the right part is zero. Thus, in the right part of (5.10) the only terms of length r − 1 are of type wk1 term of length r on the right hand side of (5.10) must contain every vj, but this is only possible if n is an odd integer. 2m . Now, if η = r and ij = 0 for some j then, evj · η = 0. Hence, every 1 · · · wk2m We have thus proved ewp · ewq = αp,q 0 + XK K=r−1 αp,q (0,K)WK XK K=r−n αp,q (1,K)V(1)WK. n odd (cid:3) Lemma 5.3. ewp = wp + XK K=r ξp (0,K)WK + XK K=r−n−1 n odd ξp (1,K)V(1)WK. Proof. Let p ∈ {1, . . . , 2m} be a fixed integer. We assume that there exist some scalars ξp (I,K) such that ξp (5.11) ηI,K =r−1 (I,K)ηI,K + XI,K Using Lemma 5.1, we have that ewp ·evj = 0, and therefore, ewp = wp + XI,K 0 = XI,K (I,K)ηI,K ·evj + XI,K ηI,K =r−1 (5.12) ηI,K =r ηI,K =r K odd K odd K odd K odd ξp ξp (I,K)ηI,K ·evj. ξp (I,K)ηI,K. 16 G ´OMEZ-GONZ ´ALEZ F.A. Let η = ξI,KηI,K be a nonzero element in (5.12). We shall use atatments (i) - (iv) from Lemma 5.2. verify that if I 6= (1) then ξ(I,K) = 0. Therefore, we see that the only elements in (5.11) of lenght r − 1 are of type V(1)WK . We note that If η = r − 1 and ij = 1, then η ·evj = 0. Using (5.12) one can easily Let η = r and ij = 1, then (η ·evj ) ·evj = η and therefore, if VI 6= 1, then ξ(I,K) = 0. Thus, the only elements of lenght r that are not zero on the right part of (5.12) are precisely those where ij = 0. As this is valid for every j, we have that the only elements of lenght r that appear in (5.11) are of type wk1 2m , with K = r. Thus, we have proved that 1 · · · wk2m ewp = wp + XK K=r ξp (0,K)WK + XK K=r−n−1 ξp (1,K)V(1)WK. n odd (cid:3) 5.2. N be isomorphic to u Cr/u Cr−2, where r is an even integer and u is an even vector. Without loss of generality we can take N0 = vect F h uVI WK ηI,K = r, r − 1 and K even i, N1 = vect F h uVI WK, ηI,K = r, r − 1 and K odd i. As in above case, one can easily verify that 1 2 )i1+···+ij−1 uV(i1,...,ij +1,...,in)WK((−1)ηI,K −ij − 1). if if ηI,K = r − 1, ij = 0, ηI,K = r, ij = 1, if if ηI,K = r, ηI,K = r − 1. Moreover, we have uVI W(k1,...,kp+1,...,k2m) ±uV(i1,...,ij−1,1,...,in)WK uVI WK ◦evj = (− uVI WK ·evj =( ±uV(i1,...,ij−1,1,...,in)WK uVI WK · ewp =( ∓kp±1uVI W(k1,...,kp±1−1,...,k2m) ewp · ewq = δp+1,q + XK (0,K)uW(k1,...,k2m) + XK ewp = wp + XK (0,K)uW(k1,...,k2m) + XK K=r−1 αp,q K=r ξi K=r−n n even K=r−n−1 n even Now, we note that there exist analogues to Lemmas 5.1, 5.2 and 5.3, implying the following equalities αp,q (1,K)uV(1)W(k1,...,k2m), ξi (1,K)uV(1)W(k1,...,k2m). We shall prove the following theorem Theorem 5.4. Let A be a finite-dimensional Jordan superalgebra, N be the solvable radical of A such that N 2 = 0 and A/N is isomorphic to the Jordan superalgebra WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 17 of superform J. Then A ∼= A/N ⊕ N if and only if, N ∈ M(A/N ; J (k)) where J (k) = C2k+1/C2k−1 if dim V0 = 2k + 1 or J (k) = aC2k/aC2k−2 if dim V0 = 2k. Proof. By Theorem 3.3 it suffices to prove the theorem when N is irreducible. So, by Theorem 7.7 in [17] we only need to consider the two cases. Using Lemma 5.1, we have ewp ·evi = 0 for i = 1, . . . , n; p = 1, . . . , 2m. Let p be an odd integer. Due to Lemmas 5.2 and 5.3, we can assume that ewp · ewq = δp+1,q + XK ewp = wp + XK K=r K=r−1 αp,q (0,K)WK + XK (0,K)WK + XK n odd ξp K=r−n K=r−n−1 n odd αp,q (1,K)V(1)WK , ξp (1,K)V(1)WK . (5.13) (5.14) Thus (5.15) ewp · ewq =wp · wq + XK K=r ξp (0,K)WK · wq + XK K=r−n−1 ξp (1,K)V(1)Wk · wq+ XK K=r ξp (0,K)wp · WK + XK K=r−n−1 n odd n odd ξp (1,K)wp · V(1)WK. αp,q K=r−1 (5.16) (−kp+1ξp (0,K))W(k1,...,kp+1−1,...,k2m)+ Using (5.2), (5.13) and (5.15), we have that ewp · ewq = δp+1,q if and only if, 0 = XK XK XK (0,K)WK + XK (0,K)W(k1,...,kp−1−1,...,k2m) + XK (1,K)V(1)W(k1,...,kp+1−1,...,k2m) + XK (1,K)V(1)WK+ kp−1ξq K=r−n K=r−n K=r−n K=r−1 K=r−1 αp,q n odd ξp ξq (1,K)V(1)W(k1,...,kp+1,...,k2m). n odd n odd Combining the above equality with a linear independence property of the elements VI WK we have the following relations: XK K=r−1 (5.17) αp,q (0,K)WK = XK XK K=r−1 K=r−1 kp+1ξp (0,K)W(k1...,kp+1−1,...,k2m)− (kp−1ξq (0,K))W(k1...,kp−1−1,...,k2m), 18 G ´OMEZ-GONZ ´ALEZ F.A. XK K=r−n n odd (5.18) αp,q (1,K)V(1)WK = − XK K=r−n ξp (1,K)V(1)W(k1...,kp+1,...,k2m)− n odd ξq (1,K)V(1)W(k1...,kp+1,...,k2m). XK K=r−n n odd Let αp,q (0,St)WSt be a nonzero element at the left part of (5.17), such that St = (s1, . . . , sp−1, sp, sp+1, . . . , sn) is a 2m-tupla, with St = r − 1. We shall find a 2m- tupla Sp+1 and Sp−1 on the right part of (5.17), such that St+1 = (s1, . . . , sp−1, sp, sp+1 + 1, . . . , sn) and St−1 = (s1, . . . , sp−1 + 1, sp, sp+1, . . . , sn). We observe that St−1 = St+1 = r. Applying similar arguments to above stated, and using (5.18), we have that for each Kt = (k1, . . . , kp, . . . , kn) we should take eKt = (k1, . . . , kp − 1, . . . , kn). Moreover, if Kt = r − n, then eKt = r − n − 1. It is easy to see that the equations (5.17) and (5.18) are respectively equivalent to (5.19) K=r−1(cid:16)αp,q XK (cid:16)αp,q XK K=r−n (0,S) − (sp+1 + 1)ξp (0,Sp+1) + (sp−1 + 1)ξq (1,Kp) + ξq 1, eKp + ξp 1, eKp(cid:17)V(1)WKt = 0. (0,Sp−1)(cid:17)WSt = 0, n odd Using the linear independance of WK , V(1)WK and (5.19), for each t ∈ {1, . . . , 2m}, we have (5.20) αp,q (0,St) − (sp+1 + 1)ξp (0,St+1) + (sp−1 + 1)ξq + ξp αp,q (1,Kt) + ξq (0,St−1) = 0, = 0. (1, eKt) (1, eKt) Hence, we have a solvable linear equation system if r 6= n. We note that an (cid:3) analogous procceding is valid if r is an even integer. Remark 5.5. By Lemmas 5.2 and 5.3, if n is an odd integer and r = n, then ewp = wp+ XK K=r We see that the system αp,q solution when αp,q (1,0) 6= 0. ξp (0,K)WK , and αp,q (0,K)WK +αp,q (1,0)V(1). ewp·ewq = δi+1,j + XK K=n−1 (0,St) − ξp (0,St+1) + ξq (0,St−1) = 0, and αp,q (1,0) = 0 has no 5.3. counter-examples to WPT for Jordan superalgebras of superform with radical Cr/Cr−2 and dim J0 = r. Now we will show that the restrictions imposed in the Theorem 5.4 are essential, and we have two cases to consider: WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 19 Case 1. Let n be an odd integer. Consider the superalgebra J = (F · 1 + F · v1 + · · · + F · vn + N0) ∔ (F · w1 + · · · + F · w2m + N1). where N0 = Spannh vi1 N1 = Spannh vi1 1 · · · vin 1 · · · vin n wk1 n wk1 1 · · · wk2m 2m , 1 · · · wk2m 2m , K is even, I + K = n − 1 or n i, K is odd , I + K = n − 1 or n i where i1, . . . , in are 0 or 1 and ki are nonnegative integers, K = k1 + · · · + k2m, I = i1 + · · · + in. All nonzero products of the basis elements of J are defined as follows v2 i = 1, w1 · w2 = 1 + v1 · · · vn = −w2 · w1, w2s−1 · w2s = −w2s · w2s−1 = 1 for s ∈ {2, 3, . . . , m}, vi1 1 · · · vin n wk1 1 · · · wk2m 2m · wp = 1 2 vi1 n wk1 1 · · · vin 1 · · · wkp−1 1 · · · wkp+1 p · · · wk2m 2m (1 + (−1)I+K)− kp+1vi1 1 · · · vin n wk1 p+1 · · · wk2m 2m if p = 2s − 1, s ∈ {1, . . . , m}, vi1 1 · · · vin n wk1 1 · · · wk2m 2m · wp = 1 · · · wkp+1 p · · · wk2m 2m (1 + (−1)I+K)+ 1 2 n wk1 1 · · · vin n wk1 vi1 1 · · · vin 1 · · · wkp−1−1 p−1 kp−1vi1 · · · wk2m 2m if p = 2s, s ∈ {1, . . . , m}, 1 · · · wk2m 2m · vj = vi1 n wk1 1 · · · vin 1 2 (− )i1+···+ij−1 vi1 1 · · ·vij +1 j · · · vin n wk1 1 · · · wk2m 2m (1 + (−1)I+K−ij ). We note that J/N = (F·1 + F·v1 +· · ·+ F·vn) ∔ (F·w1 +· · ·+ F·w2m) is a Jordan superalgebra isomorphic to Jordan superalgebra of superform, N is isomorphic to Cn/Cn−2. If we assume that the WPT is valid for J, then, for i = 1, . . . , 2m there exists ξKwk1 1 · · · wk2m ewi ∈ J1 such that ewi ≡ wi(modN1), and ew2i−1 · ew2i = 1. By Lemma 5.3 there exist βK, ξK , α, λ ∈ F such that ew1 = w1 + XK=n and ew2 = w2 + XK=n 2m + XK=n ew1 · ew2 = w1 · w2 + XK=n We observe that ew1 · ew2 = 1 if and only if v1 · · · vn + XK=n 1 · · · wk2m 2m . Hence, ξkw1 · wk1 1 · · · wk2m Using the fact that v1 · · · vn and wk1 contradiction. βKwk1 ωKwt1 2m are linearly independent, we have a βKwk1 1 · · · wk2m 2m 1 · · · wk2m 2m · w2. 1 · · · wt2m 2m = 0. Case 2 Let n be an even integer. Consider the superalgebra J = (F · 1 + F · v1 + · · · + F · vn + N0) ∔ (F · w1 + · · · + F · w2m + N1). 20 G ´OMEZ-GONZ ´ALEZ F.A. 1 · · · vin 1 · · · vin N0 is spanned by h uvi1 1 · · · wk2m 2m , n wk1 K is even i and N1 is spanned by h uvi1 K is odd i, where i1, . . . , in are 0 or 1 and ki are nonnegative integers, K = k1 + · · · + k2m, I = i1 + · · · + in and K + I = n or K + I = n − 1. 1 · · · wk2m 2m , n wk1 All nonzero products of the basis elements of J are defined as follows v2 i = 1, w1 · w2 = 1 + uv1 · · · vn = −w2 · w1, w2s−1 · w2s = −w2i · w2i−1 = 1 for s ∈ {2, 3, . . . , m}, uvi1 1 1 · · · vin 2 1 · · · wkp+1−1 n wk1 p+1 uvi1 1 1 · · · vin 2 1 · · · wkp−1−1 n wk1 p−1 uvi1 1 · · · vin n wk1 1 · · · wk2m 2m · wp = n wk1 1 · · · wkp+1 p · · · wk2m 2m (1 + (−1)I+K)+ kp+1uvi1 1 · · · vin · · · wk2m 2m , if p = 2s − 1, s ∈ {1, . . . , m}, uvi1 1 · · · vin n wk1 1 · · · wk2m 2m · wp = n wk1 1 · · · wkp+1 p · · · wk2m 2m (1 + (−1)I+K)− kp−1uvi1 1 · · · vin · · · wk2m 2m , if p = 2s, s ∈ {1, . . . , m}, uvi1 1 · · · vin 2m · vj = n wk1 1 · · · wk2m 1 2 (− )i1+···+ij−1+1uvi1 1 · · · vij +1 j · · · vin n wk1 1 · · · wk2m 2m (1 + (−1)I+K−ij ). It is easy to verify that J/N is a Jordan superalgebra of superform and N ∼= u Cn/u Cn−2. If we assume that the WPT is valid for J, then, for i = 1, . . . , 2m there exists ξKuwk1 ewi ∈ J1 such that ewi ≡ wi(modN1), and ew2i−1 · ew2i = 1, i ≥ 2. By an analogous to Lemma 5.3, we have that ew1 = w1 + XK=n−1 and ew2 = w2 + XK=n−1 It is clear that ew1 · ew2 = w1 · w2 + XK=n−1 fore ew1 ·ew2 = 1 if and only if uv1 · · · vn + XK=n−1 ωKuw1 · wk1 ωKuw1 · wt1 1 · · · wk2m 1 · · · wt2m 1 · · · wk2m 2m for some βK, ξK, α, λ ∈ F. again, uv1 · · · vn and uwk1 a contradiction. βKuwk1 1 · · · wk2m 2m 2m , ωK ∈ F. There- 1 · · · wk2m 2m = 0. Once 2m are linearly independent, consequently, we have 6. Superalgebra Dt and Kaplansky K3 In this section, we consider the Jordan superalgebra Dt = (F · e1 + F · e2) ∔ (F · x + F · y), and Kaplansky, K3 = (F · e) ∔ (F · x + F · y). We stress that Dt is a simple Jordan superalgebra if t 6= 0. If t = 0, then D0 contain K3. Unital irreducible superbimodules over Dt and K3 were classified by C. Martinez and E. Zelmanov in [16] and by M. Trushina in [15]. In this section, we shall use the examples, notations and ideas introduced by M. Trushina. WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 21 Let sl2 be a Lie algebra with the basis e, f, h and the multiplication given by [f, h] = 2f, [e, h] = 2e, [e, f ] = h, where [a, b] = ab − ba. We shall say that a module L with the basis l0, l1, . . . , ln is an irreducible sl2- module with standard basis l0, l1, . . . , ln if l0 · e = 0, ln · f = 0, li · h = (n − 2i)li, li · e = (−in + i(i − 1))li−1 li · f = li+1 for i > 0, for i < n. By Ra we denote the operator of right multiplication by a, we also denote it by the 1+t Y 2 capital letter A. One can easily check that the operators span the simple lie algebra sl2. In terms of operators above, it is easy to see that a superbimodule L with basis l0, l1, . . . , ln is an irreducible sl2-module with the standard basis l0, l1, . . . , ln if 1+t X ◦ Y, 2 1+t X 2, 2 2 li X ◦ Y = (6.1) l0 X 2 = 0, (n − 2i)li, 1 + t 2 li X 2 = ln Y 2 = 0, li Y 2 = li+1 for i < n. 1 + t 2 1 + t 2 (−in + i(i − 1)li−1) for any i > 0, In terms of right multiplication operators, equality (2.2) may be written as fol- lows: (6.2) Rai Raj Rak +(−1)ij+ik+jk Rak Raj Rai + (−1)jkR(aiak)aj = Rai Raj ak + (−1)ij+ik+jkRak Raj ai + (−1)ijRaj Raiak . Substituting ai = ak = x and aj = e1 (respectively ai = ak = y and aj = e1 ) in (6.2), we obtain [X 2, E] = 0 (respectively [Y 2, E] = 0), where E denote Re1 . 6.1. Jordan superalgebra Dt. In this section, we shall prove the following theo- rem. Theorem 6.1. Let A be a finite-dimensional Jordan superalgebra with a solvable radical N such that N 2 = 0 and A/N ∼= Dt, t 6= −1. Then A ∼= A/N ⊕ N Iff N ∈ M(A/N ; J (k)) where J (k) is a regular superbimodule. Proof. Using 3.3 and the Theorem 1.1 in [15], we need to consider three main cases. Here, S0 = F ·ee1 + F ·ee2 ∼= (Dt)0 and A1/N1 = F · ¯x + F · ¯y ∼= (Dt)1, A0 = S0 ⊕ N0. Case 1 Let n be a positive integer and suppose t ∈ R, t 6= 0, 1, − n+2 n . n+1, L2 n+1, N1 = Ln+2⊕Ln. Here, L1 n+1, Ln+2, We assume that where N0 = L1 Ln are the same as in Example 1 in [15]. n+1⊕L2 It is easy to see that E N1 ≡ 1 that there exist scalars βx 3,i, βx 4,i, βy 2 , therefore eei ·ex = 1 1,j and ξx,y 4,i, ξx,y 3,i, βy 2ex and eei ·ey = 1 2ey. Assume 2,j for i = −1, 0, 1, . . . , n and 22 G ´OMEZ-GONZ ´ALEZ F.A. j = 0, 1, . . . , n, such that ex =x + βx ey =y + βy (6.3) (6.4) 3,−1m + βx 3,nmY 2(n+1) + 3,−1m + βy 3,nmY 2(n+1) + x · y =ee1 + tee2 + 3,k−1 + αβx 4,k and γy nXk=0 γx 3,4,kmY 2k − γy 3,4,kmY 2k − nXk=1 nXk=1 nXk=1 nXk=1 (ξx,y 1,k − ξx,y 2,k )mY 2k+1E + 4,kmY 2k−1EY, βx βy 4,kmY 2k−1EY, ξx,y 2,k mY 2k+1. nXk=0 where γx 3,4,k = βx 3,4,k = βy 3,k−1 + αβy 4,k. Now, we haveex ·ey =ee1 + tee2, if and only if 3,nmY 2n+2 + 3,−1m + βx γx 3,4,kmY 2k − nXk=1 nXk=1 βx βy 4,kmY 2k−1EY(cid:17) · y− 4,kmY 2k−1EY(cid:17) · x+ 3,−1m + βy 3,nmY 2n+2 + γy 3,4,kmY 2k − (ξx,y 1,k − ξx,y 2,k )mY 2k+1E + ξx,y 2,k mY 2k+1 = 0 =(cid:16)βx (cid:16)βy nXk=0 nXk=1 nXk=1 nXk=0 nXk=1 nXk=1 (6.5) βx 3,−1mY + βx 3,nmY 2n+3 + γx 3,4,kmY 2k+1 + βx 4,kmY 2k+1E+ βy 3,n (1 + t)(n + 1) 2 mY 2n+1 + (1 + t)(n − (k − 1)) βy 4,kmY 2k−1E+ 2 nXk=1 2 nXk=1(cid:16) 1 + t n+1Xk=1 (ξx,y kγy 4 3,4,k + (−1)k (1 + t)n + 2 n+1Xk=1 2,k−1)mY 2k−1E + 1,k−1 − ξx,y βy 4,k(cid:17)mY 2k−1+ ξx,y 2,k−1mY 2k−1. Since α = (1+t)n+2 2(1+t)(n+1) , we have that (1 + t)βy 3,0 2 − nα(1 + t)βy 4,1 2 + ξx,y 3,nmY 2n+3+ 2,0(cid:21) mY + βx (cid:21) mY 2n+1+ 3,n 1,n − ξx,y 2,n + βx 3,n−1 + αβx 4,n + ξx,y 2,n + 4,1 + ξx,y 1,0 − ξx,y (1 + t)(n + 1)βy 2 2,0(cid:21) mY E +(cid:2)ξx,y (1 + t) 2 4,n(cid:3) mY 2n+1E+ 2,k−1(cid:21) mY 2k−1+ 3,k−2 + αβx 4,k−1 + [kβy 3,k−1 + α(k + (−1)k(n + 1)βy 4,k] + ξx,y 4,k−1 + βy 4,k (1 + t)(n − (k − 1)) 2 + ξx,y 1,k−1 − ξx,y 2,k−1(cid:21) mY 2k−1E = 0. (6.6) 3,−1 + (cid:20) βx (cid:20) βx (cid:20) (1 + t)nβy nXk=2(cid:20) βx nXk=2(cid:20) βx 2 WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 23 Note that fixing the ξ's in (6.4), we get βx 3,n = 0, 4,n = ξx,y βx 2,n − ξx,y 1,n, βy 4,1 = 2(ξx,y 2,0 − ξx,y 1,0 ) (t + 1)n , (6.7) βx 3,0 = βy 3,n = βy 4,k = βy 3,k−1 = 3,−1 − 2ξx,y 2,0 , nα(1 + t)βy 4,1 − 2βx (1 + t) 3,n−1 + αβx (1 + t)(n + 1) −2(βx 4,n + ξx,y 2,n) , 2 (1 + t)(n − (k − 1))h−βx k" −2(βx 3,k−2 + αβx (1 + t) 1 4,k−1 + ξx,y 2,k−1) 4,k−1 − ξx,y 1,k−1 + ξx,y k = 2, . . . , n 2,k−1i , − α(k + (−1)k(n + 1)β4,k# such that the equality (6.6) holds. Case 2. Let n be a positive integer, 1 t = − n n+2 . Consider the following cases: in [15].) N ∼= M(n + 1, n + 2), N ∼= M(n + 1, n), N ∼= fM(n1) and n1 6= n. (See example 2 (A) Assume that N ∼= M(n + 1, n + 2) where N0 is the irreducible sl2-module 2ey. Assume that there exist 0 = βy k lky 0,x, βz βx k lky, 0,xl0x + ξx,y k lk, 0,xl0x + 0,x, βu 0 ,. . ., βu n, βz , . . . , ξu,z nXk=0 2ex and eei ·ey = 1 0 , . . ., and βz with the standard basis l0, . . . , ln and N1 is spanned by l0x, l0y, l1y, . . . , lny. Since E M1 ≡ 1 ξu,z n , βu 0 n scalars such that x · y =ee1 + tee2 + ex = x + βx 2 , then eei ·ex = 1 nXk=0 nXk=0 0,x (cid:18) 1 − t 2 (cid:19) l0 + βy 2 (cid:19) lk+1 − n−1Xk=0 0,x (cid:18) 1 − t 2 (cid:19) + βy n + 2 (cid:19) + βx n (cid:18) (n + 1)t k−1 (cid:18) 1 + t 2 (cid:19) − βy ey = y + βy We observe thatex ·ey =ee1 + tee2 if and only if n + 2 (cid:19) ln n (cid:18) (n + 1)t k (cid:18) 1 + t 2(n − k)(cid:19) (−(k + 1)n + (k + 1)k)lk, 0 (cid:18) 1 + t 2 (cid:19) , n−1 (cid:18) 1 + t 2 (cid:19) , k (cid:18) 1 + t 2(n − k)(cid:19) (−(k + 1)n + (k + 1)k)lk, which gives rise to the system of equations k (cid:18) 1 + t nXk=0 n−1Xk=0 ξx,y k lk + βx k + βx 0 = ξx,y n + βy 0 = ξx,y 0 + βx 0 = ξx,y + βx (6.8) βy for k = 1, 2, . . . , n − 1. We note that the system (6.8) has always a solution. 24 G ´OMEZ-GONZ ´ALEZ F.A. (B) N ∼= M(n + 1, n) where N0 is the irreducible sl2-module with the standard basis l0, . . . , ln and N1 is spanned by l1x, . . . , lnx. ξx,y k lk. Let us find βx 2ey and there exist ξx,y k and βy 0 k=1 βx 2ex, eei ·ey = 1 As in the case (A), eei ·ex = 1 nXk=0 such that x · y =ee1 + tee2 + Pn k lkx, and ey = y +Pn Now, we note thatex ·ey =ee1 + tee2 if and only if k (cid:18) 1 + t nXk=0 2 (cid:19) k lk − nXk=1 nXk=1 ξx,y k lk − k=1 βy k lkx. 0 = βx βy k (cid:18) 1 + t , . . . , ξx,y n ∈ F k such that ex = x + 2 (cid:19) (−kn + k(k − 1))lk−1, which gives rise to the system of equations (6.9) 0 = ξx,y 0 − βy 0 = ξx,y k − βx 0 (cid:18) 1 − t k k(cid:18) 1 + t 2 (cid:19) n = ξx,y 2 (cid:19) − βy n − nβx k+1 (cid:18) 1 + t 2 (cid:19) , n (cid:18) 1 + t 2 (cid:19) (−(k + 1)n + (k + 1)k), for k = 1, 2, . . . , n − 1. System of equations (6.9) has always a solution. replacement of n by n1. (C) Finally, if N ∼= fM(n1), n1 6= n. This case is similar to Case (1), with the Case 3. Let t = 1. In this case, N is isomorphic to fM(n) or to a 1-dimensional vector space with a generator m such that mx = mx = 0, me = 1 3 in [15].) 2 m. (See example It only remains to consider the case when N is isomorphic to a 1-dimensional take t = 1 in the equation (6.6). Now we shall consider two subcases: holds if and only if η = 0. If we take ai = x, aj = y, ak = e1 and al = e2 in the 4 ηm and vector space. The case when N ∼= fM(n) is similar to Case (1). In particular we (A) If m is an even vector, then N0 = F·m. Assume that x·y =ee1 +tee2 +ηm, for some η ∈ F Since N1 = 0 we haveex = x, ey = y. Note that the equalityex·ey =ee1+tee2 equalitie (2.2) we obtained, 0 = ((ex ·ey)e1)e2 = ((e1 + te2 + ηm)e1)e2 = 1 x · y =ee1 + tee2. Letex = x + βum andey = y + βzm, henceex ·ey =ee1 + tee2 is always In this case N0 = 0 and N1 = F · m. Therefore, therefore, η = 0, thus the WPT is valid. (B) If m is odd vector. solvable. From Case (1) - (3), we conclude that it is possible to give some conditions for ηu and ηz ∈ N1, such that an analogue to WPT is valid under the Theorem (cid:3) conditions. WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 25 6.2. Jordan superalgebra K3. We shall proof the following theorem Theorem 6.2. Let A be a finite-dimensional Jordan superalgebra with solvable radical N , N 2 = 0 and such that A/N ∼= K3. Then there exists a subsuperalgebra S ⊆ A such that A/N ∼= S and A = S ⊕ N . Proof. Since Theorem 3.3 and [15], we have to consider two cases to know, N ∼= (K3)1 ∼= A1/N1 = F · ¯x + F · ¯y, and N = F · f ∔ (F · u + F · z), where f ↔ e, u ↔ x, can obtain an analogue of equality (6.5) substituting t = 0. This gives rise to the system of equations equivalent to 6.7. Reg K3 and N ∼= fM(n). But the second cases is analogous to case (1), for Dt, one Therefore, we consider N ∼= Reg K3. Assume that (K3)0 ∼= S0 = F ·ee1 and z ↔ y. Let ex and ey be some preimages of ¯x and ¯y respectively and suppose that xy =ee1 + ηf for some η ∈ F. Let α, β, γ and δ scalars such thatex = x + αu + βz and wy = y + γu + δz. We note that ex ·ey =ee1 if and only if α + δ = η. and the equality is always solvable. (cid:3) Remark 6.3. In the case of the Jordan superalgebra K3 ⊕ F · 1 we have that the irreducible superbimodules are the same as the ones for the Jordan superalgebra K3. In general, for any algebra A there exist an isomorphism of category of su- perbimodules over A, (Bimod A) into category of unital superbimodules over A#, (Bimod A#). Thus, the proof of the above theorem is also true if we substitute K3 by K3 ⊕ F · 1. 6.3. Counter-examples to WPT for Jordan superalgebras of type Dt, t 6= −1. Now we will show that restrictions imposed in Theorem 6.1 are essential. Let B = A ⊕ N be a superalgebra, where A0 = F · e1 + F · e2 + F · a1 + F · a2, A1 = F · x + F · y + F · v + F · w, N0 = F · a1 + F · a2 and N1 = F · v + F · w. All nonzero products of the basis elements of B are defined as follows: (6.10) e2 i = ei, eiaj = δijaj, eix = aix = 1 2 v, aiy = 1 2 w, eiv = 1 2 1 2 x, eiy = y, v, eiw = w, 1 2 1 2 (6.11) xw = vy = a1 + ta2, xy = e1 + te2 + a1 + (−2 − t)a2 for i = 0, 1. The products in (6.10) and (6.11) commute and anticommute respec- tively and t 6= −1. One easily verifies that B is a Jordan superalgebra, and B/N is a Jordan super- algebra isomorphic to Dt, t 6= −1, N ∼= Reg Dt and N 2 = 0. Consider the product xy = e1 + te2 + αa1 + βa2. Replacing ai = ak = x, aj = y and al = e1 in (2.2) we obtain ((xy) · x) · e1 − 1 2 (xy) · x = 0, thus we have 1 + t + α + β = 0, later on α + β = −1 − t and therefore, B is a Jordan superalgebra. 26 G ´OMEZ-GONZ ´ALEZ F.A. obtain λxw = 0 and therefore λ = 0. Now 2ex, eiey = 1 2ey. If we assume that the WPT is valid for B, then there are ex,ey such that ex ≡ x, ey ≡ y (mod N1) andexey = e1 + te2, eiex = 1 We note thatex = x + σv andey = y + ωw. Ifex = x + σv + λw, usingex2 = 0, we exey = xy + σyv + ωxw = e1 + te2 + a1 + (−2 − t)a2 − σ(a1 + ta2) + ω(a1 + ta2) Therefore, 1 − σ + ω = 0 and (−2 − t) − σt + ωt = 0, later on ω − σ = −1 and 0 = (−2 − t) + t(ω − σ) = −2 − 2t, thus t = −1 and this is a contradiction. 7. Main theorem Using the Theorems 4.4, 5.4, 6.1 and 6.2, we have the following theorem: Theorem 7.1. Let A be a finite dimensional Jordan superalgebra with solvable radical N such that N 2 = 0 and A/N ∼= J where J is a simple Jordan superalgebra. We set M(J; N1, . . . , Nt) = { V /V is a J-superbimodule such that homomorphic images of V do not contain subsuperbimodules isomorphic to Ni for i = 1, 2, . . . , t}, and M(J; N1, . . . , Nt) as an analogue of the class M(J; N1, . . . , Nt) for irreducible J-superbimodules Ni. If one of the following conditions holds: i) J ∼= K10; ii) J ∼= K3; iii) J is a superalgebra of a superform with even part of dimension n such that N ∈ M(J; Cn/Cn−2 (n is odd), u · Cn/u · Cn−2 (n is even)); iv) J ∼= Dt, t 6= −1, N ∈ M(Dt; Reg Dt); then there is a subsuperalgebra S ⊆ A such that S ∼= J and A = S ⊕ N , the restrictions of items iii) and iv) are essential. References 1. Th. Molien, On systems of higher complex numbers (Ueber Systeme hoherer complexer Zahlen), Math. Ann. XLI (1893) 83 -- 156. 2. Th. Molien, Correction to the article "On systems of higher complex numbers" (Berichtigung zu dem Aufsatze "Ueber Systeme hoherer complexer Zahlen"), Math. Ann. XLII (1893) 308 -- 312. 3. S. Epsteen, J.H. Maclagan-Wedderburn, On the structure of hypercomplex number systems, Trans. Am. Math. Soc. 6 (1905) 172 -- 178. 4. R.D. Schafer, The Wedderburn principal theorem for alternative algebras, Bull. Am. Math. Soc. 55 (1949) 604 -- 614. 5. A.A. Albert, The Wedderburn principal theorem for Jordan algebras, Ann. Math. 48 (1) (1947), 1 -- 7. 6. A.J. Penico, The Wedderburn principal theorem for Jordan algebras, Trans. Am. Math. Soc. 70 (1951) 404 -- 420. WEDDERBURN PRINCIPAL THEOREM FOR JORDAN SUPERALGEBRAS I. 27 7. V.G. Askinuze, A theorem on the splittability of J-algebras (Russian), Ukrain. Math. Z. 3 (1951) 381-398. 8. E.J. Taft, Invariant Wedderburn factors, Illinois J. Math. 1 (1957) 565 -- 573. 9. N.A. Pisarenko, The Wedderburn decomposition in finite dimensional alternative superalge- bras, Algebra Logic 32 (4) (1993) 231 -- 238; translation from Algebra Logika 32 (4) (1993) 428 -- 440. 10. V.G. Kac, Classification of simple Z-graded Lie superalgebras and simple Jordan superalge- bras, Commun. Algebra 5 (13) (1977) 1375-1400. 11. I.L. Kantor, Jordan and Lie superalgebras determined by a Poisson algebra, Algebra and Analysis 55 -- 80 (Tomsk, Russia, 1989); Amer. Math. Soc. Transl. Ser. 2 151 (1992). 12. E. Zelmanov, Semisimple finite dimensional Jordan superalgebras, Fong, Yuen (ed.) et al., Lie algebras, rings and related topics. Papers of the 2nd Tainan -- Moscow international algebra workshop 97, Taiwan, January 11 -- 17, 1997. Hong Kong: Springer (ISBN 962 -- 430 -- 110 -- 7/pbk). 227 -- 243 (2000). 13. A.S. Shtern, Representations of an excepcional Jordan superalgebra, Funct. Anal. Appl. 21(1 -- 3) (1987) 253 -- 254; translation from Funkts. Anal. Prilozh. 21 (3) (1987) 93 -- 94. 14. C.T.C. Wall, Graded Brauer groups; J.Reine Angew Math, 213 (1964) 187-199. 15. M. Trushina, Irreducible representations of a certain Jordan superalgebra, J. Algebra Appl. 4 (1) (2005) 1 -- 14. 16. C. Mart´ınez, E. Zelmanov, Unital Bimodules over the simple Jordan superalgebra Dt, Trans. Am. Math. Soc. 358 (8) (2006) 3637 -- 3649. 17. C. Mart´ınez, E. Zelmanov, Representation theory of Jordan superalgebras. I, Trans. Am. Math. Soc. 362 (2) (2010) 815 -- 846. 18. C. Mart´ınez, I. Shestakov, E. Zelmanov, Jordan bimodules over the superalgebras P(n) and Q(n), Trans. Am. Math. Soc. 362 (4) (2010) 2037 -- 2051. 19. F.A. G´omez Gonz´alez, Jordan superalgebras of type Mnm(F)(+) and the Wedderburn prin- cipal theorem (WPT), Commun. Algebra 44 (7) (2016) 2867 -- 2886. 20. F.A. G´omez Gonz´alez, R. Vel´asquez, Wedderburn principal theorem for orthosymplectic Jor- dan superalgebras Jospn2m(F), To appear in Algebra and Discrete Mathematics Instituto de Matematicas, Universidad de Antioquia, Colombia,
1708.08082	1	1708	2017-08-27T12:53:31	Semisimple Leibniz algebras and their derivations and automorphisms	[ "math.RA", "math.RT" ]	The present paper is devoted to the description of finite-dimensional semisimple Leibniz algebras over complex numbers, their derivations and automorphisms.	math.RA	math	SEMISIMPLE LEIBNIZ ALGEBRAS AND THEIR DERIVATIONS AND AUTOMORPHISMS SHAVKAT AYUPOV, KARIMBERGEN KUDAYBERGENOV, BAKHROM OMIROV, AND KAIMING ZHAO Abstract. The present paper is devoted to the description of finite-dimensional semisimple Leibniz algebras over complex numbers, their derivations and automor- phisms. AMS Subject Classifications (2010): 17A32, 17A60, 17B10, 17B20. Key words: Lie algebra, Leibniz algebra, simple algebra, semisimple algebra, irre- ducible module, derivation, automorphism. 1. Introduction In recent years the non-associative analogues of classical constructions become of interest in connection with their applications in many branches of mathematics and physics. Leibniz algebras present a "non-commutative" analogue of Lie algebras and they were introduced by Loday [12] as algebras satisfying the (right) Leibniz identity: [x, [y, z]] = [[x, y], z] − [[x, z], y]. Leibniz algebras preserve an important property of Lie algebras: the operator of right multiplication is a derivation. During the last decades the theory of Leibniz algebras has been actively investigated. Some (co)gomology and deformation properties; results on various types of decom- positions; structure of solvable and nilpotent Leibniz algebras; classifications of some classes of graded nilpotent Leibniz algebras were obtained in numerous papers devoted to Leibniz algebras, see, for example, [1,2,4,6,7,13] and reference therein. In fact, many results on Lie algebras have been extended to the Leibniz algebra case. For instance, the classical results on Cartan subalgebras [14], Levi's decomposition [5], properties of solvable algebras with given nilradical [9] and others from the theory of Lie algebras are also true for Leibniz algebras. From the classical theory of finite-dimensional Lie algebras it is known that an ar- bitrary semisimple Lie algebra is decomposed into a direct sum of simple ideals, which are completely classified [11]. In the paper [8] an example of semisimple Leibniz al- gebra, which can not be decomposed into a direct sum of simple ideals, is presented (see Example 3.6 also). This shows that the structure of semisimple Leibniz algebras is much more complicated than structure of semisimple Lie algebras. Thus, the natural problem arises - to describe semisimple Leibniz algebras. In fact, the structure depends on relations between semisimple Lie algebras and their modules. Due to Barnes' re- sult [5] an arbitrary semisimple Leibniz algebra L is represented as L = S +I, where S is a semisimple Lie algebra and I is the ideal generated by squares of elements of the algebra L. This means that the problem is focused to investigation of the relation between the ideal I and the semisimple Lie algebra S. 1 2 AYUPOV, KUDAYBERGENOV, OMIROV, AND ZHAO One of the main results of the present paper is the description on finite-dimensional semisimple Leibniz algebras (Theorem 3.5). Then next steps, which are crucial for any variety of algebras, are investigations of derivations and automorphisms of these algebras. Note that derivations and automorphisms of semisimple Lie algebras are completely described by their actions on corresponding simple ideals [11]. The de- scriptions of derivations and automorphisms of simple Leibniz algebras were studied in [3]. Here we present the description on derivations of semisimple Leibniz algebras. Since any semisimple Leibniz algebra can be represented as a direct sum of indecom- posable semisimple Leibniz algebras, the Lie algebra of derivations of such algebras is also represented as a direct sum of the Lie algebras of all derivations on indecom- posable semisimple Leibniz algebras. The automorphism group of semisimple Leibniz algebras is quite different from that of simple Leibniz algebras, and more subtle than their derivations. The paper is organized as follows. In Section 2 we give some preliminaries from the Leibniz algebra theory. Section 3 is devoted to the description of the structure of finite-dimensional com- plex semisimple Leibniz algebras (Theorem 3.5). We also determined a structure of irreducible modules over a Lie algebra which is a direct sum of two Lie algebras (see Theorem 3.1). In Section 4 we completely determine derivations of finite-dimensional complex semisimple Leibniz algebras (Theorem 4.5). Finally, In Section 5 we determine the group of automorphisms of finite-dimensional complex semisimple Leibniz algebras (Theorem 5.7). Throughout the paper, vector spaces and algebras are finite-dimensional over the field of complex numbers C. We will use ⊕ to denote direct sum of ideals in an algebra, while use + to denote direct sum of subalgebras or subspaces. Moreover, in the table of multiplication of an algebra the omitted products in terms of a given basis are assumed to be zero. 2. Preliminaries In this section we give necessary definitions and results on semisimple Lie algebras and their irreducible modules towards the description of semisimple Leibniz algebras. Definition 2.1. An algebra (L, [·, ·]) over a field F is called a Leibniz algebra if it satisfies the property [x, [y, z]] = [[x, y], z] − [[x, z], y] for all x, y ∈ L, which is called Leibniz identity. The Leibniz identity is a generalization of the Jacobi identity since under the con- dition of anti-symmetricity of the product "[· , ·]" this identity changes to the Jacobi identity. In fact, the definition above is the notion of the right Leibniz algebra, where "right" indicates that any right multiplication operator is a derivation of the algebra. In the present paper the term "Leibniz algebra" will always mean the "right Leib- niz algebra". The left Leibniz algebra is characterized by the property that any left multiplication operator is a derivation. Below we present an adapted version to our further using of the well-known Schur's Lemma ( [18, P. 57, Corollary 3]). Lemma 2.2. Let G be a complex Lie algebra, let V = V1 ⊕· · ·⊕Vm and W = W1 ⊕· · ·⊕ Wn be a completely reducible G-modules, where V1, . . . , Vn, W1, · · · , Wn are irreducible SEMISIMPLE LEIBNIZ ALGEBRAS 3 modules. Then any G-module homomorphism ϕ : V → W can be represented as m n ϕ = λijϕij, Xi=1 Xj=1 where the operator ϕij : Vi → Wj are fixed module homomorphisms and λij are complex numbers. Furthermore ϕij 6= 0 if and only if ϕij is an isomorphism if and only if Vi and Wj are isomorphic G-modules. Definition 2.3. A Lie algebra G is called to be reductive, if R(G) = Z(G), where R(G) and Z(G) are the solvable radical and the center of G, respectively. Lemma 2.4. [G, G] ⊕ Z(G), and [G, G] is either semisimple or 0. [10, Proposition 19.1] (a) Let G be a reductive Lie algebra. Then G = b) Let V be a finite-dimensional space and let G ⊂ gl(V) be a nonzero Lie algebra acting irreducibly on V. Then G is reductive with dim Z(G) ≤ 1. If in addition G ⊂ sl(V) then G is semisimple. Theorem 2.5. [10, Corollary 21.2]. Let G be a semisimple Lie algebra. Then the map λ 7→ Vλ induces a one-one correspondence between the set of all dominant integral functions Λ+ and the isomorphism classes of finite-dimensional irreducible G-modules. We need the following result from [17, Page 143, Theorem 6]. Theorem 2.6. Let G = G1 ⊕G2 be a semisimple Lie algebra. Let a Cartan subalgebra H of G be correspondingly decomposed as H = H1 ⊕ H2, where Hi is a Cartan subalgebra of Gi (i = 1, 2). Suppose that the fundamental root system Π of G with respect to H is decomposed into Π = Π1 ∪ Π2, where Π1 = {α1, . . . , αn} and Π2 {β1, . . . , βm} is the fundamental root system of Gi with respect to Hi (1 = 1, 2). (i) Suppose i is an irreducible representation of Gi with highest weight ωi and representation space Vi (i = 1, 2). If (x)(v1 ⊗ v2) = 1(x1)v1 ⊗ v2 + v1 ⊗ 2(x2)v2, x = x1 + x2, xi ∈ Gi, i = 1, 2, then is an irreducible representation of G with highest weight ω1 + ω2, where (ω1 + ω2)(h) = ω1(h1) + ω2(h2), h = h1 + h2, hi ∈ Hi, i = 1, 2. (ii) Conversely, every irreducible representation of G is obtained as in (i). For a Leibniz algebra L, a subspace generated by its squares I = span {[x, x] : x ∈ L} due to Leibniz identity becomes an ideal, and the quotient GL = L/I is a Lie algebra called liezation of L. Moreover, [L, I] = 0. In general, [I, L] 6= 0. Since we are interested in Leibniz algebras which are not Lie algebras, we will always assume that I 6= 0. Definition 2.7. A Leibniz algebra L is called simple if its liezation is a simple Lie algebra and the ideal I is a simple ideal. Equivalently, L is simple iff I is the only non-trivial ideal of L. Definition 2.8. Let G be a Lie algebra and V a (right) G-module. Endow the vector space L = G ⊕ V with the brackets as follows: [(g1, v1), (g2, v2)] := ([g1, g2], v1.g2), where v.g (sometimes denoted as [v, g]) is an action of an element g of G on v ∈ V. Then L is a Leibniz algebra, denoted as G ⋉ V . 4 AYUPOV, KUDAYBERGENOV, OMIROV, AND ZHAO Definition 2.9. A Leibniz algebra L is called semisimple if its liezation G is a semisim- ple Lie algebra. The following theorem recently proved by D. Barnes [5] presents an analogue of Levi–Malcev's theorem for Leibniz algebras. Theorem 2.10. Let L be a finite dimensional Leibniz algebra over a field of character- istic zero and let R be its solvable radical. Then there exists a semisimple Lie subalgebra S of L such that L = S +R. Application of the analogue of Levi's Theorem for Leibniz algebras to the case of semisimple Leibniz algebras implies the following Proposition 2.11. Let L be a finite-dimensional semisimple Leibniz algebra. Then L = S ⊕ I, where S is a semisimple Lie subalgebra of L. Moreover [I, S] = I. It should be noted that I is a nontrivial module over the Lie algebra S. The structure of simple Leibniz algebras is clear. By Proposition 2.8 a simple Leibniz algebra is a sum of a simple Lie algebra S and a simple module (the highest weight module) over S. Our problem concerns semisimple Leibniz algebras. Semisimple Leibniz algebras in general, are not direct sum of simple Leibniz algebras. See the next example. Example 2.12. Let L be a 10-dimensional Leibniz algebra. Let {e1, h1, f1, e2, h2, f2, x1, x2, x3, x4} be a basis of L such that I = {x1, x2, x3, x4}, and multiplication table of L has the following form: [sli 2, sli 2] : [ei, hi] = 2ei, [hi, ei] = −2ei [fi, hi] = −2fi, [hi, fi] = 2fi, [ei, fi] = hi, [fi, ei] = −hi, i = 1, 2, [I, sl1 2] : [I, sl2 2] : [x1, f1] = x2, [x3, f1] = x4, [x1, h1] = x1, [x3, h1] = x3, [x2, e1] = −x1, [x4, e1] = −x3, [x2, h1] = −x2, [x4, h1] = −x4, [x1, f2] = x3, [x2, f2] = x4, [x1, h2] = x1, [x2, h2] = x2, [x3, e2] = −x1, [x4, e2] = −x2, [x3, h2] = −x3, [x4, h2] = −x4, (omitted products of the basis vectors are equal to zero). 2] = I. Moreover, I splits over sl1 From this table of multiplications we conclude that L is semisimple and [I, sl1 2] = 2 (i.e. I = span{x1, x2} ⊕ span{x3, x4}) and over 2 (i.e. I = span{x1, x3} ⊕ span{x2, x4}). Actually, I is a simple ideal (an irreducible 2 ⊕sl2 2) ⋉I cannot be a direct sum of two nonzero [I, sl2 sl2 (sl1 ideals. See Example 3.6 for a more general case. 2)-module). Therefore, L = (sl1 2 ⊕sl2 3. Structure of semisimple Leibniz algebras First we give the generalized version of Theorem 2.6, which has an independent interest. Theorem 3.1. Let G = G1 ⊕ G2 be a direct sum of Lie algebras and let V be a finite- dimensional irreducible G-module with dim V > 1. Then there are finite-dimensional irreducible Gi-modules Vi so that V ∼= V1 ⊗ V2. Proof. Consider Ann(V) = {x ∈ G : V.x = 0} which is an ideal of G. Then V is a faithful irreducible module over G = G/Ann(V), in particular, for an element x ∈ G from v.x = 0 for all v ∈ V it follows that x = 0. SEMISIMPLE LEIBNIZ ALGEBRAS 5 Let ρ be the representation of G on V, i.e., ρ(¯g)(v) = v.¯g for any v ∈ V and ¯g ∈ G. Then we have the injective linear map ρ : G → gl(V). Thus G is also finite-dimensional. Applying part b) of Lemma 2.4 to G we know that ¯G is reductive, Z( ¯G) ≤ 1, and ¯G = [ ¯G, ¯G]⊕Z( ¯G). Denote the natural projections by ρ1 : ¯G → [ ¯G, ¯G] and ρ2 : ¯G → Z( ¯G), and define π = ρ1 ◦ ρ. Assume Z(G) = Cz 6= 0. Since [[v, x], z] = [[v, z], x] for all x ∈ G and v ∈ V, it follows that the mapping v ∈ V ֒→ [v, z] ∈ V is a G-module homomorphism of V. Taking into account that V is an irreducible G- module, by Lemma 2.2 there exists λ ∈ C such that [v, z] = λv for all v ∈ V. We know that V is a simple module over π(G) = [ ¯G, ¯G]. If π(G1) ∩ π(G2) 6= {0}, it has to be semisimple. We see that π(G1) ∩ π(G2) = [π(G1) ∩ π(G2), π(G1) ∩ π(G2)}] ⊂ [π(G1), π(G2)}] = 0, a contradiction. Thus π(G1) ∩ π(G2) = 0, and π(G) = π(G1) ⊕ π(G2). Since V is also an irreducible module over π(G1) ⊕ π(G2), then both π(G1) and π(G2) are semisimple. By Theorem 2.5, we deduce that there are finite-dimensional irreducible π(Gi)-modules Vi so that V ∼= V1 ⊗ V2. We can easily extend π(Gi)-module Vi into a Gi-module. This completes the proof. (cid:3) We say that a semisimple Leibniz algebra L = S +I is decomposable, if L = (cid:0)S1 +I1(cid:1) ⊕(cid:0)S2 +I2(cid:1) , where S1 +I1 and S2 +I2 are non-trivial semisimple Leibniz al- gebras. Otherwise, we say that L is indecomposable. Lemma 3.2. Any semisimple Leibniz algebra has the form: n where Si +Ii is an indecomposable semisimple Leibniz algebra for all i ∈ {1, . . . , n}. L = Mi=1 (cid:0)Si +Ii(cid:1) , Proof. Let L = S +I be a semisimple Leibniz algebra and let S = Si be a decompo- sition of simple Lie ideals Si. We know that [I, S] = I. The proof is by induction on n. Let n = 1, that is, S is a simple Lie algebra. Then L = S +I is an indecomposable semisimple Leibniz algebra. Suppose that the assertion of the theorem is true for all numbers least than n and n n Li=1 S = Si. Li=1 Consider a partition of the set {1, 2, . . . , n} = A ∪ B into union of disjoint subsets. Set IA =(cid:20)I,Li∈A Si(cid:21) and IB ="I, Lj∈B Sj# . Case 1. Let IA ∩ IB 6= {0} for any non trivial partition A ∪ B of the set {1, 2, . . . , n}. In this case L is indecomposable. 6 AYUPOV, KUDAYBERGENOV, OMIROV, AND ZHAO Case 2. Let IA ∩ IB = {0} for some partition A ∪ B of the set {1, 2, . . . , n}. Then L is decomposable and L = Mi∈A Si +IA! ⊕ Mj∈B Sj +IB! . Si +IA and Lj∈B By the hypothesis of the induction algebrasLi∈A as a direct sum of indecomposable algebras. The proof is complete. Sj +IB can be represented (cid:3) Further we need the following auxiliary result. Lemma 3.3. Let S be a semisimple Lie algebra and let I be an irreducible S-module. Then [I, S] = 0 if and only if dim I = 1. Proof. Let ρ be the representation of S on I, i.e., ρ(g)(v) = [v, g] for any v ∈ I and g ∈ S. Then we have the Lie algebra homomorphism ρ : S → gl(I). So ρ(S) is a subalgebra of gl(I). The result follows from Lemma 2.4. (cid:3) Remark 3.4. From Lemma 3.3 we conclude that for an indecomposable semisimple Leibniz algebra L = S +I the inequality dim I ≥ 2 holds true if I 6= 0. From Lemma 3.2, we need only to determine the structure of indecomposable semisimple Leibniz algebras. Suppose L is an indecomposable semisimple Leibniz alge- bra with a given Levi decomposition L = S +I. We may assume that S = ⊕m i=1Si and I = ⊕n i=1Ii where each Si is a simple Lie algebra and each Ii is an irreducible S-module. We say that Si and Sj are adjacent if there exists Ik such that [Ik, Si] = [Ik, Sj] = Ik. We say that Si and Sj are connected if there exist Sk1 = Si, Sk2, · · · , Skr = Sj such that Skl and Skl+1 are adjacent. Now we can prove our main result on indecomposable semisimple Leibniz algebras. Theorem 3.5. Let L = S +I be an indecomposable semisimple Leibniz algebra with I 6= {0}. Then (a). S = ⊕m (b). I = ⊕n (c). L = S ⋉ I; (d). for any 1 ≤ i ≤ m and 1 ≤ j ≤ n there is an irreducible Si-module Iij such that i=1Si where each Si is a simple Lie algebra; i=1Ii where each Ii is an irreducible S-module with [Ii, S] = Ii; Ii = ⊗m j=1Iij; (e). any Si and Sj are connected. Proof. Parts (a) and (b) are clear, (c) follows from the definition, (d) follows from Theorem 2.6. (e) The proof is by induction on n. Let n = 1. Then I is an irreducible S-module. By Theorem 2.6 for each j ∈ j=1Jj. Since S +I {1, . . . , m} there is an irreducible Sj-module Jj such that I = ⊗m is indecomposable, Lemma 3.3 implies that [Jj, Sj] = Jj for all j. Thus [I, Sj] = I for all j. This means that Si and Sj are adjacent. Suppose that the assertion (e) is true for all numbers least than n > 1. Consider a i=1 Ii. Assume that S +J is indecomposable. By Leibniz algebra S +J , where J = ⊕n−1 the hypothesis of the induction, Si and Sj are already connected. Now assume that S +J is decomposable. tion of S +J into a direct sum of p Let us consider the decomposi- indecomposable semisimple Leibniz algebras: Lt=1(cid:0)⊕i∈AtSi + ⊕s∈Bt Is(cid:1) , where A1, . . . , Ap and B1, . . . , Bp are partitions of {1, . . . , m} SEMISIMPLE LEIBNIZ ALGEBRAS 7 and {1, . . . n − 1}, respectively. Since L = S +I is indecomposable, for every s ∈ {1, . . . , p} there exists an index is ∈ As such that [In, Sis] = In. Thus Si1, . . . , Sip are mutually adjacent. On the other hand, since ⊕i∈AtSi + ⊕s∈Bt Is is indecomposable, by the hypothesis of the induction, for each i, j ∈ As (1 ≤ s ≤ p) algebras Si and Sj are connected. Thus Si and Sj are connected for every i, j ∈ {1, . . . , m}. The proof is complete. (cid:3) In our notations we present an example that generalizes the one in [8, Theorem 4.2] as a semisimple Leibniz algebra which can not be decomposed into the direct sum of simple ideals. Example 3.6. Let m, n ∈ N. Let L = (cid:0)sl1 with a basis 2L sl2 2(cid:1) +I be a semisimple Leibniz algebra {e1, h1, f1, e2, h2, f2, xi ⊗ yj : i = 0, 1, · · · , m; j = 0, 1, · · · , n} and the product table: [ei, hi] = −[hi, ei] = 2ei, [hi, fi] = −[fi, hi] = 2fi, [ei, fi] = −[fi, ei] = hi, [xi ⊗ yj, h1] = (m − 2i)xi ⊗ yj, [xi ⊗ yj, h2] = (n − 2j)xi ⊗ yj, [xi ⊗ yj, e1] = ixi+1 ⊗ yj, [xi ⊗ yj, e2] = jxi ⊗ yj+1, [xi ⊗ yj, f1] = (m − i)xi−1 ⊗ yj, [xi ⊗ yj, f2] = (n − j)xi ⊗ yj, where 0 ≤ i ≤ n and 0 ≤ j ≤ m. If V1 is the (m + 1)-dimensional irreducible sl1 and V2 is the (n+1)-dimensional irreducible sl2 2-module 4. Derivations on semisimple Leibniz algebras 2-module, then L =(cid:0)sl1 2L sl2 2(cid:1)⋉(V1⊗V2). In this section we describe all derivations of semisimple Leibniz algebras. We will use Der(L) to denote the Lie algebra of derivations of a Leibniz algebra L. Lemma 4.1. Let L = L1 ⊕ L2 be a Leibniz algebra such that [Li, Li] = Li. Then Der(L) = Der(L1) ⊕ Der(L2). Proof. This is similar to that of the Lie algebra case. We omit the details. (cid:3) From this lemma, in order to determine Der(L) for semisimple Leibniz algebras L we need only to consider indecomposable semisimple Leibniz algebras L. From now on in this section we assume that L = S +I is an indecomposable semisimple Leibniz algebra. We assume that r (4.1) S = niSi Mi=1 is a decomposition into the sum of simple Lie ideals where Si 6≃ Sj if i 6= j; and that (4.2) I = r+s Mi=1 miIi 8 AYUPOV, KUDAYBERGENOV, OMIROV, AND ZHAO is a decomposition into the sum of simple S-modules where Ii 6≃ Ij if i 6= j. We may further assume that Si ≃ Ii as S-modules and mi ≥ 0 for i ∈ {1, 2, · · · , r}. For any x ∈ S we have the inner derivation Rx : L → L, y 7→ [y, x], ∀ y ∈ L. Let RS = {Rx : x ∈ S} be the inner derivations of L, and Der(L)S,I = {d ∈ Der(L) : d(S) ⊂ I, d(I) = 0}, Der(L)I,I = {d ∈ Der(L) : d(I) ⊂ I, d(S) = 0}. Lemma 4.2. Let L = S +I be a semisimple Leibniz algebra. Then Der(L) = RS +Der(L)S,I +Der(L)I,I. Proof. The proof is similar to that in the case of simple Leibniz algebras (see [15, Theorem 3.2]). (cid:3) Lemma 4.3. (a). For any d ∈ Der(L)S,I, the restriction dS : S → I is an S-module homomorphism. (b). For any d ∈ Der(L)I,I, the restriction dI : I → I is an S-module homomor- phism. Proof. (a) Let d ∈ Der(L)S,I. For each x, y ∈ S we have d([x, y]) = [d(x), y] + [x, d(y)] = [d(x), y]. This means that dS is a S-module homomorphism. The proof of the part (b) is similar to (a). The proof is complete. (cid:3) The following result directly follows from Lemma 2.2. Lemma 4.4. Let S be a semisimple Lie algebra and V is a simple S-module. Let m, n ∈ Z+. Then (a). EndS(mV, nV) ≃ homC(Cm, Cn); (b). AutS(mV) ≃ GLm(C). The above four lemmata imply the following result. Theorem 4.5. Let L = S +I be a semisimple Leibniz algebra with decompositions given as in (4.1) and (4.2). Then (a). Each d ∈ Der(L)S,I is uniquely determined by a map in ⊕r (b). each d ∈ Der(L)I,I is uniquely determined by a map in ⊕r+s (c). i=1 homC(Cni, Cmi); i=1 homC(Cmi, Cmi); dim Der(L) = dim S + nimi + r Xi=1 m2 i . r+s Xi=1 5. Automorphisms of semisimple Leibniz algebras In this section we describe the group Aut(L) of automorphisms of semisimple Leibniz algebras L. We first have the following result. Lemma 5.1. Let L = ⊕n nonisomorphic indecomposable semisimple Leibniz algebras. Then i=1miLi be a semisimple Leibniz algebra where Li's are pairwise n (Smi ⋉ Aut(Li)mi) where Sm is the symmetric group of permutations. Aut(L) ∼= Yi=1 SEMISIMPLE LEIBNIZ ALGEBRAS 9 Proof. It suffices to consider an algebra of the form mL = L1 ⊕ . . . ⊕ Lm, where Li = L for all i. For 1 ≤ i, j ≤ m denote by πij an identical isomorphism from Li onto Lj. Then is a subgroup of Aut(mL), which isomorphic to the group Sm and we may identify its to Sm. (cid:8)π = ⊕m i=1πiσ(i) : σ ∈ Sm(cid:9) Let ϕ ∈ Aut(mL). Since Li is an indecomposable Leibniz algebra, ϕ(Li) is also an indecomposable Leibniz algebra. Thus there exists an index σϕ(i) such that ϕ(Li) = Lσϕ(i). It is clear that σϕ : i → σϕ(i) is a permutation of the set {1, . . . , m} and i=1πiσϕ(i) φi = πσϕ(i)i ◦ ϕLi It is easy to see that Aut(L)m is a normal subgroup of Aut(mL). So ϕ ∈ Sm ⋉ Aut(L)m. The proof is complete. (cid:3) is an automorphism of Li. Then ϕ = πϕ ◦ φ, where πϕ = ⊕m i=1 φi ∈ Aut(L)m. Thus Aut(mL) = SmAut(L)m. and φ = Qm Now we need to determine Aut(L) for indecomposable semisimple Leibniz algebras L. From now on in this section, we assume that L = S +I is an indecomposable semisimple Leibniz algebra with decompositions as in (4.1) and (4.2). We define the projections ρ1 : L → S, x + y 7→ x, ∀x ∈ S, y ∈ I; ρ2 : L → I, x + y 7→ y, ∀x ∈ S, y ∈ I. Let ϕ be an automorphism of L, ϕI = ρ2 ◦ ϕI, ϕ1 = ρ1 ◦ ϕS and ϕ2 = ρ2 ◦ ϕS. Clearly, ϕ is determined by ϕI, ϕ1 and ϕ2. It is clear that ϕ(I) = I. For any S-module V and any σ ∈Aut(S), we define the new S-module V σ = V with the action v · x = [v, x]′ = vσ(x), ∀ v ∈ V, x ∈ S. We know from [10] that V ≃ V σ if σ is an inner automorphism of S. If σ is not an inner automorphism of S, we generally do not have V ≃ V σ. Lemma 5.2. (a). ϕ1 ∈ Aut(S). (b). ϕI is an S-module isomorphism from I onto I ϕ1. (c). ϕ2 is an S-module hommorphism from S to I ϕ1. Proof. Let x, y ∈ S, then ϕ1([x, y]) + ϕ2([x, y]) = ϕ([x, y]) = [ϕ(x), ϕ(y)] = = [ϕ1(x) + ϕ2(x), ϕ1(y) + ϕ2(y)] = = [ϕ1(x), ϕ1(y)] + [ϕ2(x), ϕ1(y)]. This implies ϕ1([x, y]) = [ϕ1(x), ϕ1(y)], ϕ2([x, y]) = [ϕ2(x), ϕ1(y)]. Let x ∈ S, i ∈ I, then ϕI([i, x]) = ϕ([i, x]) = [ϕ(i), ϕ(x)] = [ϕI(i), ϕ1(x)]. The proof is complete. (cid:3) Lemma 5.3. (a). For σ ∈ Aut(S) and an S-module isomorphism σI : I → I σ there is ϕ ∈ Aut(L) such that ϕ1 = σ and ϕI = σI. (b). For σ ∈ Aut(S) there is ϕ ∈ Aut(L) such that ϕ1 = σ if and only if I ≃ I σ as S-modules. 10 AYUPOV, KUDAYBERGENOV, OMIROV, AND ZHAO Proof. (a) For σ ∈ Aut(S) and an S-module isomorphism σI : I → I σ set ϕ(x) = σ(x) + σI(i), x + i ∈ S + I. Then ϕ([x + i, y + j]) = ϕ([x, y] + [i, y]) = σ([x, y]) + σI([i, y]) = = [σ(x), σ(y)] + [σI(i), σ(y)] = = [σ(x) + σI(i), σ(y) + σI(j)] = [ϕ(x + i), ϕ(y + j)]. Thus ϕ is an automorphism such that ϕ1 = σ and ϕI = σI. (b) Let σ ∈ Aut(S). Suppose that there exists ϕ ∈ Aut(L) such that ϕ1 = σ. For every x ∈ S, i ∈ I we have ϕI([i, x]) = ϕ([i, x]) = [ϕ(i), ϕ(x)] = [ϕI(i), σ(x)]. This means that I ≃ I σ as S-modules. The converse assertion follows from the part (b). The proof is complete. (cid:3) The following example shows the existence of automorphism of S which can not be extended to the whole algebra L. Example 5.4. Let sl3 be the Lie algebra consisting of all traceless 3 × 3 complex matri- ces, and let V(Λ1) = C3 be the standard sl3-module (by left matrix multiplication). Its highest weight is the first fundamental weight Λ1. Consider the automorphism of sl3: σ : sl3 → sl3, x 7→ −xt, for all x ∈ sl3, where xt is the transpose of x. This automorphism σ cannot be extended to an automorphism of the whole simple Leibniz algebra sl3 ⋉ V (Λ1) since V(Λ1)σ ≃ V(Λ2) 6≃ V(Λ1), where Λ2 = −σ ◦ Λ1 (see [10, P. 116, Exersice 21.6]). We define the following subgroups of Aut(L): Aut(L)I ={ϕ ∈ Aut(L) : ϕS = idS}, Aut(L)S ={ϕ ∈ Aut(L) : ϕ(S) = S}, Aut(L)0 ={ϕ ∈ Aut(L) : ϕI = idI}. Lemma 5.5. Let L = S +I be a semisimple Leibniz algebra. Then we have the following isomorphisms: Aut(L)I ∼= AutS(I), Aut(L)0 ∼= homS(S, I). Proof. Let us prove the first isomorphism. For an arbitrary ϕ ∈ Aut(L)I and x ∈ S, i ∈ I we have ϕ([i, x]) = [ϕ(i), ϕ(x)] = [ϕ(i), x]. Thus ϕI ∈ AutS(I). Let now σ ∈ AutS(I). Set (5.1) ϕσ(x + i) = x + σ(i), x + i ∈ S + I. Then ϕσ ∈ Aut(L)I, and (5.1) implies that the mapping σ → ϕσ is an isomorphism from AutS(I) to Aut(L)I. Now we shall prove the second assertion. Let ϕ ∈ Aut(L)0. Since [i, x] = ϕ([i, x]) = [ϕ(i), ϕ(x)] = [i, ϕ1(x)], SEMISIMPLE LEIBNIZ ALGEBRAS 11 we obtain [i, ϕ1(x) − x] = 0 for all i ∈ I, x ∈ S. Thus [I, Sϕ1(x)−x] = 0, where Sϕ1(x)−x is the ideal of S generated by the element ϕ1(x) − x. Thus ϕ1(x) = x for all x ∈ S, i.e., ϕ1 = idS. So, we have (5.2) ϕ ∈ Aut(L)0 ⇒ ϕ1 = idS. The equalities ϕ([x, y]) = [x, y] + ϕ2([x, y]), [ϕ(x), ϕ(y)] = [x + ϕ2(x), y + ϕ2(y)] = [x, y] + [ϕ2(x), y], ∀x, y ∈ S imply that ϕ2([x, y]) = [ϕ2(x), y]. Therefore, ϕ2 ∈ homS(S, I). Let σ ∈ homS(S, I). Set (5.3) ϕσ(x + i) = x + σ(x) + i, x + i ∈ S + I. Then ϕσ ∈ Aut(L)0, and by (5.3) we see that the mapping σ → ϕσ is an isomorphism of the additive group homS(S, I) and the multiplicative group Aut(L)0. (cid:3) Denote Aut(S)0 = {ϕ ∈ Aut(S) : ∃ ψ ∈ Aut(L)S, ψS = ϕ}. Lemmata 5.2, 5.3, 5.5 and implication (5.2) gives us the following matrix representations of the above mentioned groups: 0 0 ϕ2 ϕI (cid:19) : ϕ1 ∈ Aut(S)0, ϕI ∈ IsoS(I, I ϕ1), ϕ2 ∈ homS(S, I ϕ1)(cid:27) , 0 ϕI (cid:19) : ϕI ∈ AutS(I)(cid:27) , 0 ϕI (cid:19) : ϕ1 ∈ Aut(S)0, ϕI ∈ IsoS(I, I ϕ1)(cid:27) , Aut(L) =(cid:26)(cid:18) ϕ1 Aut(L)I =(cid:26)(cid:18) idS Aut(L)S =(cid:26)(cid:18) ϕ1 Aut(L)0 =(cid:26)(cid:18) idS ϕ2 0 0 idI (cid:19) : ϕ2 ∈ homS(S, I)(cid:27) , where IsoS(I, I ϕ1) is the set of S-module isomorphisms. These representations show that Aut(L)I and Aut(L)0 are normal subgroups of Aut(L)S and Aut(L), respectively. Corollary 5.6. Let L = S +I be a semisimple Leibniz algebra. ∼= Aut(L)S/Aut(L)I. (a). Aut(S)0 (b). If all Si are not of type Al, Dl or E6, then Aut(S)0 = Aut(S). Now we can prove the main result of this section. Theorem 5.7. Let L = S +I be a semisimple Leibniz algebra. Then Aut(L) = Aut(L)S ⋉ Aut(L)0. Proof. Let ϕ ∈ Aut(L). Set ψ(x + i) = ϕ1(x) + ϕI(i), x + i ∈ S + I. Let us show that ψ is also an automorphism. Indeed, ψ([x + i, y + j]) = ψ([x, y] + [i, y]) = ϕ1([x, y]) + ϕI([i, y]) = [ϕ1(x), ϕ1(y)] + ϕ([i, y]) = [ϕ1(x), ϕ1(y)] + [ϕ(i), ϕ(y)] = [ϕ1(x), ϕ1(y)] + [ϕI(i), ϕ1(y) + ϕI(y)] = [ϕ1(x) + ϕI(i), ϕ1(y) + ϕI(j)] = [ψ(x + i), ψ(y + j)]. 12 AYUPOV, KUDAYBERGENOV, OMIROV, AND ZHAO It is clear that ψ ∈ Aut(L)S. Now consider the automorphism Then η = ψ−1 ◦ ϕ. η(x + i) = x + ϕ−1 I ϕ2(x) + i and therefore η = ψ−1 ◦ ϕ ∈ Aut(L)0. Since ϕ = ψ ◦ η, it follows that Aut(L) = Aut(L)S ⋉ Aut(L)0. The proof is complete. (cid:3) So far we have described the automorphism groups for all semisimple Leibniz algebras. K.Z. is partially supported by NSF of China (Grant 11271109) and NSERC. Acknowledgments References [1] Albeverio S., Ayupov. Sh.A., Omirov B.A., On nilpotent and simple Leibniz algebras, Comm. Algebra, 33, 2005, p. 159-172. [2] Albeverio S., Ayupov. Sh.A., Omirov B.A., Cartan subalgebras, weight spaces and criterion of solvability of finite dimensional Leibniz algebras, Rev. Mat. Complut., vol. 19(1), 2006, p. 183-195. [3] Ayupov Sh. A., Kudaybergenov K. K., Omirov A.B., Local and 2-local derivations and automor- phisms on simple Leibniz algebras, arXiv:1703.10506, 2017. [4] Balavoine D., D´eformations et rigidit´e g´eom´etrique des alg`ebres de Leibniz, Comm. Algebra, 24(3), 1996, p. 1017–1034. [5] Barnes D.W., On Levi's theorem for Leibniz algebras, B. Aust. Math. Soc., 86 (2) (2012), 184- 185. [6] Camacho L., G´omez J.R., Gonz´alez A.R., Omirov B.A., Naturally graded quasi-filiform Leibniz algebras, J. Symbol. Comp., 44 (2009), 527-539. [7] Camacho L., G´omez J.R., A. J. Gonz´alez, Omirov B.A., The classification of naturally graded p-filiform Leibniz algebras, Com. Algebra, 39 (1) (2011), 153-168. [8] Camacho L.M., G´omez-Vidal S., Omirov B.A., Karimjanov I.A., Leibniz algebras whose semisim- ple part is related to sl2, Bull. Malays. Math. Sci. Soc., 40 (2017), Issue 2, 599-615. [9] Casas J.M., Ladra M., Omirov B.A., Karimjanov I.A., Classification of solvable Leibniz algebras with null-filiform nilradical, Linear and Multilinear algebra, 61 (6) (2013), 758-774. [10] Humphreys J., Introduction to Lie algebras and representation theory, 1972 [11] Jacobson N., Lie algebras, Interscience Publishers, Wiley, New York, 1962. [12] Loday J.L., Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz, Enseign. Math. 39 (2) (1993), 269–293. [13] Loday J.-L., Pirashvili T., Leibniz representations of Leibniz algebras, J. Algebra, 181 (1996), 414-425. [14] Omirov B.A., Conjugacy of Cartan subalgebras of complex finite dimensional Leibniz algebras, Journal Algebra, vol. 302 (2006), p. 887-896. [15] Rakhimov I. S., Masutova K. K., Omirov B. A., On derivations of semisimple Leibniz algebras, B. Malays. Math. Sci. Soc., (2015), 1-12. [16] Rakhimov I.S., Omirov B.A., Turdibaev R.M., On description of Leibniz algebras corresponding to sl2, Algebr. Represent. Theory, 16 (2013), Issue 5, 1507-1519. [17] Wan Z., Lie Algebras, Pergamon Press Ltd, 1975. [18] Zhelobenko D.P., Compact Lie groups and their representations, AMS Providence, 1973. E-mail address: sh−[email protected] SEMISIMPLE LEIBNIZ ALGEBRAS 13 V.I.Romanovskiy Institute of Mathematics, Uzbekistan Academy of Sciences, 29, Dormon Yoli street, 100125 Tashkent, Uzbekistan E-mail address: [email protected] Department of Mathematics, Karakalpak State University, Ch. Abdirov 1, Nukus 230113, Uzbekistan E-mail address: [email protected] National University of Uzbekistan, 4, University street, 100174, Tashkent, Uzbek- istan E-mail address: [email protected] Department of Mathematics, Wilfrid Laurier University, Waterloo, Ontario, N2L 3C5, Canada College of Mathematics and Information Science, Hebei Normal (Teachers) Univer- sity, Shijiazhuang 050016, Hebei, PR China
1009.0196	1	1009	2010-09-01T14:54:38	Composition-Diamond Lemma for Non-associative Algebras over a Commutative Algebra	[ "math.RA" ]	We establish the Composition-Diamond lemma for non-associative algebras over a free commutative algebra. As an application, we prove that every countably generated non-associative algebra over an arbitrary commutative algebra $K$ can be embedded into a two-generated non-associative algebra over $K$.	math.RA	math	Composition-Diamond Lemma for Non-associative Algebras over a Commutative Algebra∗ Yuqun Chen, Jing Li and Mingjun Zeng School of Mathematical Sciences, South China Normal University Guangzhou 510631, P. R. China [email protected] yulin [email protected] [email protected] 0 1 0 2 p e S 1 ] . A R h t a m [ 1 v 6 9 1 0 . 9 0 0 1 : v i X r a Abstract: We establish the Composition-Diamond lemma for non-associative algebras over a free commutative algebra. As an application, we prove that every countably gener- ated non-associative algebra over an arbitrary commutative algebra K can be embedded into a two-generated non-associative algebra over K. Key words: Grobner-Shirshov basis; non-associative algebra; commutative algebra. AMS Mathematics Subject Classification(2000): 16S15, 13P10, 17Dxx, 13Axx 1 Introduction Grobner bases and Grobner-Shirshov bases theories were invented independently by A.I. Shirshov [23] for non-associative algebras and commutative (anti-commutative) non-associative algebras [21], for Lie algebras (explicitly) and associative algebras (implicitly) [22], for in- finite series algebras (both formal and convergent) by H. Hironaka [19] and for polynomial algebras by B. Buchberger (first publication in [13]). Grobner bases and Grobner-Shirshov bases theories have been proved to be very useful in different branches of mathematics, including commutative algebra and combinatorial algebra, see, for example, the books [1, 12, 14, 15, 17, 18], the papers [2, 3, 4, 5, 16], and the surveys [6, 9, 10, 11]. It is well known that every countably generated non-associative algebra over a field k can be embedded into a two-generated non-associative algebra over k. This result follows from Grobner-Shirshov bases theory for non-associative algebras by A.I. Shirshov [21]. Composition-Diamond lemmas for associative algebras over a polynomial algebra is established by A.A. Mikhalev and A.A. Zolotykh [20], for associative algebras over an associative algebra by L.A. Bokut, Yuqun Chen and Yongshan Chen [7], for Lie algebras over a polynomial algebra by L.A. Bokut, Yuqun Chen and Yongshan Chen [8]. In this paper, we establish the Composition-Diamond lemma for non-associative algebras over ∗Supported by the NNSF of China (No.10771077, 10911120389) and the NSF of Guangdong Province (No. 06025062). 1 a polynomial algebra. As an application, we prove that every countably generated non- associative algebra over an arbitrary commutative algebra K can be embedded into a two-generated non-associative algebra over K, in particular, this result holds if K is a free commutative algebra. 2 Composition-Diamond lemma for non-associative algebras over a commutative algebra Let k be a field, K a commutative associative k−algebra with unit, X a set and K(X) the free non-associative algebra over K generated by X. Let [Y ] denote the free abelian monoid generated by Y , X ∗ the free monoid generated by X and X ∗∗ the set of all non-associative words in X. Denote by N = [Y ]X ∗∗ = {u = uY uXuY ∈ [Y ], uX ∈ X ∗∗}. Let kN be a k- linear space spanned by N. For any u = uY uX, v = vY vX ∈ N, we define the multiplication of the words as follows uv = uY vY uXvX ∈ N. It is clear that kN is the free non-associative k[Y ]-algebra generated by X. Such an algebra is denoted by k[Y ](X), i.e., kN = k[Y ](X). Clearly, k[Y ](X) = k[Y ] ⊗ k(X). Now, we order the set N = [Y ]X ∗∗. Let > be a total ordering on X ∗∗. Then > is called monomial if (∀u, v, w ∈ X ∗∗) u > v ⇒ wu > wv and uw > vw. For example, the deg-lex ordering on X ∗∗ is monomial: uv > u1v1, if deg(uv) > deg(u1v1), otherwise u > u1 or u = u1, v > v1. Similarly, we define the monomial ordering on [Y ]. Suppose that both >X and >Y are monomial orderings on X ∗∗ and [Y ], respectively. For any u = uY uX, v = vY vX ∈ N, define u > v ⇔ uX >X vX or (uX = vX and uY >Y vY ). It is obvious that > is a monomial ordering on N in the sense of (∀u, v, w ∈ [Y ]X ∗∗) u > v ⇒ wu > wv, uw > vw and wY u > wY v. We will use this ordering in this paper. For any polynomial f ∈ k[Y ](X), f has a unique presentation of the form f = α ¯f ¯f + X αiui, where ¯f , ui ∈ [Y ]X ∗∗, ¯f > ui, α ¯f , αi ∈ k. ¯f is called the leading term of f . f is monic if the coefficient of ¯f is 1. 2 Let ⋆ 6∈ X. By a ⋆-word we mean any expression in [Y ](X ∪ {⋆})∗∗ with only one occurrence of ⋆. Let u be a ⋆-word and s ∈ k[Y ](X). Then we call us = u⋆7→s an s-word. It is clear that for s-word us, we can express us = uY (asb) for some a, b ∈ X ∗. Since > is monomial on [Y ]X ∗∗, we have following lemma. Lemma 2.1 Let s ∈ k[Y ](X) be a non-zero polynomial. Then for any s-word us = uY (asb), uY (asb) = uY (a¯sb). Now, we give the definition of compositions. Definition 2.2 Let f and g be monic polynomials of k[Y ](X), w = wY wX ∈ [Y ]X ∗∗ and a, b, c ∈ X ∗, where wY = L( ¯f Y , ¯gY ) , L and L( ¯f Y , ¯gY ) is the least common multiple of ¯f Y and ¯gY in k[Y ]. Then we have the following compositions. 1. X-inclusion If wX = ¯f X = (a(¯gX)b), then (f, g)w = L ¯f Y f − L ¯gY (a(g)b) is called the composition of X-inclusion. 2. Y -intersection only If ¯f Y + ¯gY > wY and wX = (a( ¯f X)b(¯gX)c), then (f, g)w = L ¯f Y (a(f )b(¯gX)c) − L ¯gY (a( ¯f X)b(g)c) is called the composition of Y -intersection only, where for u ∈ [Y ], u means the degree of u. w is called the ambiguity of the composition (f, g)w. Remark 1.In the case of Y -intersection only in Definition 2.2, ¯f X and ¯gX are disjoint. Remark 2. By Lemma 2.1, we have w > (f, g)w. Remark 3. In Definition 2.2, the compositions of f, g are the same as the ones in k(X), if Y = ∅. If this is the case, we have only composition of X-inclusion. Definition 2.3 Let S be a monic subset of k[Y ](X) and f, g ∈ S. A composition (f, g)w is said to be trivial modulo (S, w), denoted by (f, g)w ≡ 0 mod(S, w), if (f, g)w = X αiuisi, i where each si ∈ S, αi ∈ k, uisi si-word and w > ui ¯si. Generally, for any p, q ∈ k[Y ](X), p ≡ q mod(S, w) if and only if p−q ≡ 0 mod(S, w). S is called a Grobner-Shirshov basis in k[Y ](X) if all compositions of elements in S are trivial modulo S. 3 If a subset S of k[Y ](X) is not a Grobner-Shirshov basis then one can add to S all nontrivial compositions of polynomials of S and continue this process repeatedly so that we obtain a Grobner-Shirshov basis Sc that contains S. Such process is called the Shirshov algorithm. Lemma 2.4 Let S be a Grobner-Shirshov basis in k[Y ](X) and s1, s2 ∈ S. Let u1s1, u2s2 be s1, s2-words respectively. If w = u1s1 = u2s2, then u1s1 ≡ u2s2 mod(S, w). Proof: Clearly, wY = L( ¯s1 Y , ¯s2 There are three cases to consider. Case 1. X-inclusion. We may assume that ¯s1 X = (c( ¯s2 Y ) · t = L · t for some t ∈ [Y ]. X)d) for some c, d ∈ X ∗ and wX = (a( ¯s1 X )b) = (a(c( ¯s2 X )d)b) for some a, b ∈ X ∗. Thus, u1s1 − u2s2 = Y (a(c(s2)d)b) = t · (a( L · t ¯s1 Y (a(s1)b) − L · t ¯s2 L Y (c(s2)d))b) ¯s2 = t · (a(s1, s2)w1b) ≡ 0 L Y s1 − ¯s1 mod(S, w) where w1 = Ls1 X. Case 2. Y -intersection only. In this case, wX = (a( ¯s1 X)b( ¯s2 X)c), a, b, c ∈ X ∗ and then u1s1 − u2s2 = L · t ¯s1 Y (a(s1)b( ¯s2 X)c) − L · t ¯s2 Y (a( ¯s1 X)b(s2)c) = t · (s1, s2)w1 ≡ 0 mod(S, w) where w1 = LwX. Case 3. Y -disjoint and X-disjoint. In this case, L = ¯s1 Y and wX = (a( ¯s1 Y ¯s2 X)b( ¯s2 X)c), a, b, c ∈ X ∗. We have u1s1 − u2s2 = L · t ¯s1 X )c) − L · t ¯s2 X )b(s2)c) = t · ( X)c) − X)b(s2)c)) Y (a(s1)b( ¯s2 Y (a(s1)b( ¯s2 Y (a( ¯s1 L L Y (a( ¯s1 Y (a(s1)b( ¯s2 ¯s2 ¯s1 X )c) − ¯s1 Y (a( ¯s1 = t · ( ¯s2 = t · ((a(s1)b( ¯s2)c) − (a( ¯s1)b(s2)c)) = t · ((a(s1)b( ¯s2)c) − (a(s1)b(s2)c) + (a(s1)b(s2)c) − (a( ¯s1)b(s2)c)) = t · ((a(s1 − ¯s1)b(s2)c) − (a(s1)b(s2 − ¯s2)c)) ≡ 0 X)b(s2)c)) mod(S, w) since w = (a( ¯s1)b( ¯s2)c) > (a(s1 − ¯s1)b(s2)c) and w = (a( ¯s1)b( ¯s2)c) > (a(s1)b(s2 − ¯s2)c). This completes the proof. (cid:3) 4 Lemma 2.5 Let S ⊆ k[Y ](X) with each s ∈ S monic and Irr(S) = {w ∈ [Y ]X ∗∗w 6= u¯s, us is an s-word, s ∈ S}. Then for any f ∈ k[Y ](X), f = X uisi ≤ ¯f αiuisi + X vj≤ ¯f βjvj, where αi, βj ∈ k, uisi si-word, si ∈ S and vj ∈ Irr(S). Proof.Let f = P αiui ∈ k[Y ](X), where 0 6= αi ∈ k and u1 > u2 > · · · . If u1 ∈ Irr(S), then let f1 = f − α1u1. If u1 6∈ Irr(S), then there exists an s-word us such that ¯f = u¯s. Let f1 = f − α1us. In both cases, we have ¯f > ¯f1. Then the result follows from the induction on ¯f . (cid:3) i From the above lemmas, we reach the following theorem: Theorem 2.6 (Composition-Diamond lemma for k[Y ](X)) Let S ⊆ k[Y ](X) with each s ∈ S monic, > the ordering on [Y ]X ∗∗ defined as before and Id(S) the ideal of k[Y ](X) generated by S as k[Y ]-algebra. Then the following statements are equivalent: (i) S is a Grobner-Shirshov basis in k[Y ](X). (ii) If 0 6= f ∈ Id(S), then f = us for some s-word us, s ∈ S. (iii) Irr(S) = {w ∈ [Y ]X ∗∗w 6= u¯s, us is an s-word, s ∈ S} is a k-linear basis for the factor algebra k[Y ](XS) = k[Y ](X)/Id(S). Proof: (i) ⇒ (ii). Suppose 0 6= f ∈ Id(S). Then f = P αiuisi for some αi ∈ k, si- word uisi, si ∈ S. Let wi = uisi and w1 = w2 = · · · = wl > wl+1 ≥ · · · . We will prove the result by using induction on l and w1. If l = 1, then the result is clear. If l > 1, then w1 = u1s1 = u2s2. Now, by (i) and Lemma 2.4, u1s1 ≡ u2s2 mod(S, w1). Thus, α1u1s1 + α2u2s2 = (α1 + α2)u1s1 + α2(u2s2 − u1s1) ≡ (α1 + α2)u1s1 mod(S, w1). Therefore, if α1 + α2 6= 0 or l > 2, then the result follows from the induction on l. For the case α1 + α2 = 0 and l = 2, we use the induction on w1. Now the result follows. (ii) ⇒ (iii). By Lemma 2.5, Irr(S) generates the factor algebra. Moreover, if 0 6= h = P βjuj ∈ Id(S), uj ∈ Irr(S), u1 > u2 > · · · and β1 6= 0, then u1 = ¯h = u¯s, a contradiction. This shows that Irr(S) is a k-linear basis of the factor algebra. (iii) ⇒ (i). For any f, g ∈ S, since k[Y ]S ⊆ Id(S), we have h = (f, g)w ∈ Id(S). The result is trivial if (f, g)w = 0. Assume that (f, g)w 6= 0. Then, by Lemma 2.5, (iii) and by noting that w > (f, g)w = ¯h, we have (f, g)w ≡ 0 mod(S, w). This shows (i). (cid:3) Remark: Theorem 2.6 is the Composition-Diamond lemma for non-associative algebras when Y = ∅. 5 3 Applications Let A be an arbitrary K-algebra and A be presented by generators X and defining rela- tions S A = K(XS). Let K have a presentation by generators Y and defining relations R K = k[Y R] as a quotient algebra of the polynomial algebra k[Y ] over k. Then with a natural way, as k[Y ]-algebras, we have an isomorphism k[Y R](XS) → k[Y ](XSl, Rx, x ∈ X), X(fi + Id(R))ui + Id(S) 7→ X fiui + Id(S′), where fi ∈ k[Y ], ui ∈ X ∗∗, S′ = Sl∪{gxg ∈ R, x ∈ X}, Sl = {P fiui ∈ k[Y ](X) P(fi+ Id(R))ui ∈ S}. Then A has an expression A = k[Y R](XS) = k[Y ](XSl, gx, g ∈ R, x ∈ X). Theorem 3.1 Each countably generated non-associative algebra over an arbitrary com- mutative algebra K can be embedded into a two-generated non-associative algebra over K. Proof. Let the notation be as before. Let A be the non-associative algebra over K = k[Y R] generated by X = {xii = 1, 2, . . . }. We may assume that A = k[Y R](XS) is defined as above. Then A can be presented as A = k[Y ](XSl, gxi, g ∈ R, i = 1, 2, . . . ). By Shirshov algorithm, we can assume that, with the deg-lex ordering >Y on [Y ], R is a Grobner-Shirshov basis in the free commutative algebra k[Y ]. Let >X be the deg-lex ordering on X ∗∗, where x1 > x2 > . . . . We can also assume, by Shirshov algorithm, that with the ordering on [Y ]X ∗∗ defined as before, S′ = Sl ∪ {gxg ∈ R, x ∈ X} is a Grobner-Shirshov basis in k[Y ](X). Let B = k[Y ](X, a, bS1} where S1 consists of f1 = Sl, f2 = {gxg ∈ R, x ∈ X}, f3 = {a(bi) − xii = 1, 2, . . . }, f4 = {gag ∈ R}, f5 = {gbg ∈ R}. Clearly, B is a K-algebra generated by a, b. Thus, to prove the theorem, by using our Theorem 2.6, it suffices to show that with the ordering on [Y ](X ∪ {a, b})∗∗ as before, where a > b > xi, i = 1, 2, . . . , S1 is a Grobner-Shirshov basis in k[Y ](X, a, b). Denote by (i ∧ j)wij the composition of the type fi and type fj with respect to the ambiguity wij. Since S′ is a Grobner-Shirshov basis in k[Y ](X), we need only to check all compositions related to the following ambiguities wij: 1 ∧ 4, w14 = L( ¯f Y , ¯g)(z1( ¯f X)z2az3); 6 1 ∧ 5, w15 = L( ¯f Y , ¯g)(z1( ¯f X)z2bz3); 2 ∧ 4, w24 = L( ¯g′, ¯g)(z1xz2az3); 2 ∧ 5, w25 = L( ¯g′, ¯g)(z1xz2bz3); 3 ∧ 4, w34 = ¯ga(bi); 3 ∧ 5, w35 = ¯ga(bi); 4 ∧ 1, w41 = L(¯g, ¯f Y )(z1az2( ¯f X)z3); 4 ∧ 2, w42 = L(¯g, ¯g′)(z1az2xz3); 4 ∧ 4, w44 = L(g1, g2)a; 4 ∧ 5, w45 = L(¯g, ¯g′)(z1az2bz3); 5 ∧ 1, w51 = L(¯g, ¯f Y )(z1bz2( ¯f X)z3); 5 ∧ 2, w52 = L(¯g, ¯g′)(z1bz2xz3); 5 ∧ 4, w54 = L(¯g, ¯g′)(z1bz2az3); 5 ∧ 5, w55 = L(g1, g2)b; where g, g′, g1, g2 ∈ R, f ∈ Sl, z1, z2, z3 ∈ (X ∪{a, b})∗ and (z1v1z2v2z3) is some bracketing. Now, we prove that all the compositions are trivial. 1 ∧ 4, w14 = L( ¯f Y , ¯g)(z1( ¯f X)z2az3), where f ∈ Sl, g ∈ R. We can write ¯f X = (uxv), where u, v ∈ X ∗. Since S′ = {Sl, Rx, x ∈ X} is a Grobner- , where w = L( ¯f Y , ¯g) ¯f X, each Shirshov basis in k[Y ](X), we have (f, gx)w = P αiuisi αi ∈ k, si ∈ S′, ui ∈ [Y ]X ∗∗ and w > uisi . Then (1, 4)w14 = = L ¯f Y L ¯f Y (z1f z2az3) − (z1f z2az3) − L ¯g L ¯g (z1( ¯f X)z2gaz3) (z1(ugxv)z2az3) + L ¯g (z1(ugxv)z2az3) − (z1( ¯f X)z2gaz3) L ¯g = (z1( L ¯f Y f − L ¯g (ugxv))z2az3) + L ¯g g((z1(uxv)z2az3) − (z1( ¯f X)z2az3)) L ¯g = (z1(f, gx)wz2az3) + = X αi(z1uisi ≡ 0 mod(S1, w14). z2az3) g((z1( ¯f X)z2az3) − (z1( ¯f X)z2az3)) Similarly, (1, 5)w15 ≡ 0, (4, 1)w41 ≡ 0, (5, 1)w51 ≡ 0. 2 ∧ 4, w24 = L( ¯g′, ¯g)(z1xz2az3), where g, g′ ∈ R. If ¯g′ + ¯g > L, then since R is a Grobner-Shirshov basis in k[Y ], (g′, g)w = ( L ¯g′ g′ − ¯g g) = P αiuihi, where w = L( ¯g′, ¯g), each αi ∈ k, ui ∈ [Y ], hi ∈ R and w > uihi. Thus L (2, 4)w24 = L ¯g (z1xz2gaz3) L ¯g′ (z1g′xz2az3) − L ¯g′ g′ − L ¯g = ( g)(z1xz2az3) = X αiuihi(z1xz2az3) = X αiui(z1xz2hiaz3) ≡ 0 mod(S1, w24). 7 . Similarly, (2, 5)w25 ≡ 0, (4, 2)w42 ≡ 0, (4, 5)w45 ≡ 0, (5, 2)w52 ≡ 0 and (5, 4)w54 ≡ 0. 3 ∧ 4, w34 = ¯ga(bi), where g ∈ R. Let g = ¯g + r ∈ R. Then (3, 4)w34 = −¯gxi − ra(bi) ≡ −¯gxi − rxi ≡ gxi ≡ 0 mod(S1, w34). Similarly, (3, 5)w35 ≡ 0. 4 ∧ 4, w44 = L(g1, g2)a, where g1, g2 ∈ R. If ¯g1 + ¯g2 > L, then since R is a Grobner-Shirshov basis in k[Y ], (g1, g2)w = g1 − L g2) = P αiuihi, where w = L( ¯g1, ¯g2), each αi ∈ k, ui ∈ [Y ], hi ∈ R and ¯g2 ( L ¯g1 w > uihi. Thus (4, 4)w44 = (g1a) − L ¯g2 (g2a) L ¯g1 L ¯g1 g1 − g2)a = ( L ¯g2 = X αiuihia ≡ 0 mod(S1, w44). . If ¯g1 + ¯g2 = L, then (4, 4)w44 = L ¯g1 (g1a) − L ¯g2 (g2a) = ( ¯g2g1 − ¯g1g2)a ≡ ((g1 − ¯g1)g2 − (g2 − ¯g2)g1)a ≡ 0 mod(S1, w44). Similarly, (5, 5)w55 ≡ 0. Now we have proved that S1 is a Grobner-Shirshov basis in k[Y ](X, a, b). The proof is complete. (cid:3) A special case of Theorem 3.1 is the following corollary. Corollary 3.2 Every countably generated non-associative algebra over a free commuta- tive algebra can be embedded into a two-generated non-associative algebra over a free commutative algebra. Acknowledgement. The authors would like to express their deepest gratitude to Profes- sor L.A. Bokut for his kind guidance, useful discussions and enthusiastic encouragement. 8 References [1] William W. Adams and Philippe Loustaunau, An introduction to Grobner bases, Graduate Studies in Mathematics, Vol. 3, American Mathematical Society (AMS), 1994. [2] G.M. Bergman, The diamond lemma for ring theory, Adv. Math., 29(1978), 178-218. [3] L.A. Bokut, Insolvability of the word problem for Lie algebras, and subalgebras of finitely presented Lie algebras, Izvestija AN USSR (mathem.), 36(6)(1972), 1173- 1219. [4] L.A. Bokut, Imbeddings into simple associative algebras, Algebra i Logika, 15(1976), 117-142. [5] L.A. Bokut and Yuqun Chen, Grobner-Shirshov basis for free Lie algebras: after A.I. Shirshov, Southeast Asian Bull. Math., 31(2007), 1057-1076. [6] L.A. Bokut and Yuqun Chen, Grobner-Shirshov bases: Some new results, Proceedings of the Second International Congress in Algebra and Combinatorics, World Scientific, 2008, 35-56. [7] L.A. Bokut, Yuqun Chen and Yongshan Chen, Composition-Diamond lemma for ten- sor product of free algebras, Journal of Algebra, 323(2010), 2520-2537. [8] L.A. Bokut, Yuqun Chen and Yongshan Chen, Grobner-Shirshov bases for Lie algebras over a commutative algebra, arXiv:1005.7682 [9] L.A. Bokut, Y. Fong, W.-F. Ke and P.S. Kolesnikov, Grobner and Grobner-Shirshov bases in algebra and conformal algebras, Fundamental and Applied Mathematics, 6(3)(2000), 669-706. [10] L.A. Bokut and P.S. Kolesnikov, Grobner-Shirshov bases: from their incipiency to the present, Journal of Mathematical Sciences, 116(1)(2003), 2894-2916. [11] L.A. Bokut and P.S. Kolesnikov, Grobner-Shirshov bases, conformal algebras and pseudo-algebras, Journal of Mathematical Sciences, 131(5)(2005), 5962 -- 6003. [12] L.A. Bokut and G. Kukin, Algorithmic and Combinatorial algebra, Kluwer Academic Publ., Dordrecht, 1994. [13] B. Buchberger, An algorithmical criteria for the solvability of algebraic systems of equations [in German], Aequationes Math., 4, 374-383(1970). [14] B. Buchberger, G.E. Collins, R. Loos and R. Albrecht, Computer algebra, symbolic and algebraic computation, Computing Supplementum, Vol.4, New York: Springer- Verlag, 1982. [15] Bruno Buchberger and Franz Winkler, Grobner bases and applications, London Mathematical Society Lecture Note Series, Vol.251, Cambridge: Cambridge University Press, 1998. 9 [16] Yuqun Chen and Qiuhui Mo, Artin-Markov normal form for Braid group, Southeast Asian Bull. Math., 33(2009), 403-419. [17] David A. Cox, John Little and Donal O'Shea, Ideals, varieties and algorithms: An introduction to computational algebraic geometry and commutative algebra, Under- graduate Texts in Mathematics, New York: Spring-Verlag, 1992. [18] David Eisenbud, Commutative algebra with a view toward algebraic geometry, Grad- uate Texts in Math., Vol.150, Berlin and New York: Springer-Verlag, 1995. [19] H. Hironaka, Resolution of singulatities of an algebraic variety over a field if charac- teristic zero, I, II, Ann. Math., 79(1964), 109-203, 205-326. [20] A. A. Mikhalev and A. A. Zolotykh, Standard Grobner-Shirshov bases of free al- gebras over rings, I. Free associative algebras, International Journal of Algebra and Computation, 8(6)(1998), 689-726. [21] A.I. Shirshov, Some algorithmic problem for ε-algebras, Sibirsk. Mat. Z., 3(1962), 132-137. (in Russian) [22] A.I. Shirshov, Some algorithmic problem for Lie algebras, Sibirsk. Mat. Z., 3(2)(1962), 292-296 (in Russian); English translation in SIGSAM Bull., 33(2)(1999), 3-6. [23] Selected works of A.I. Shirshov, Eds Leonid A. Bokut, V. Latyshev, I. Shestakov, E. Zelmanov, Trs M. Bremner, M. Kochetov, Birkhauser, Basel, Boston, Berlin, 2009. 10
1903.03484	1	1903	2019-02-07T02:39:04	Cohomology and deformations of 3-dimensional Heisenberg Hom-Lie superalgebras	[ "math.RA" ]	In this paper, we study Hom-Lie superalgebras of Heisenberg type. For 3-dimensional Heisenberg Hom-Lie superalgebras, we describe their Hom-Lie super structures, compute the cohomology spaces and characterize their infinitesimal deformations.	math.RA	math	Cohomology and deformations of 3-dimensional Heisenberg Hom-Lie superalgebras Junxia Zhu, Liangyun Chen School of Mathematics and Statistics, Northeast Normal University, Changchun, 130024, CHINA Abstract In this paper, we study Hom-Lie superalgebras of Heisenberg type. For 3- dimensional Heisenberg Hom-Lie superalgebras, we describe their Hom-Lie super structures, compute the cohomology spaces and characterize their infinitesimal de- formations. Key words: Hom-Lie superalgebras, Lie superalgebras, Heisenberg, cohomology, de- formations 1 Introduction In recent years, Hom-Lie algebras and other Hom-algebras are widely studied, mo- tivated initially by instances appeared in Physics literature when looking for quantum deformations of some algebras of vector fields. Hom-Lie superalgebras, as a generaliza- tion of Hom-Lie algebras, are introduced in [3], [4]. Furthermore, the cohomology and deformation theories of Hom-algebras are studied in [1] [2], [6], [9] and so on, while the two theories of Hom-Lie superalgebras can be seen in [3], [5]. We will follow [7], [8] to define Heisenberg Hom-Lie superalgebras, which are a spe- cial case of 2-step nilpotent Hom-Lie superalgebras. The main idea of this paper is to characterize the infinitesimal deformations of Heisenberg Hom-Lie superalgebras using cohomology. Corresponding author(L. Chen): [email protected]. Supported by NNSF of China (No. 117701069), NSF of Jilin province(No. 20170101048JC) and the project of jilin province department of education (No. JJKH20180005K). 1 The paper proceeds as follows. In Section 2, we recall the definitions of Hom-Lie superalgebras. Section 3 is dedicated to introduce Heisenberg Hom-Lie superalgebras and classify three-dimensional Heisenberg Hom-Lie superalgebras. In Section 4, we review the cohomology theory and give the 2-nd cohomology spaces of Heisenberg Hom-Lie su- peralgebras of dimension three. In the last section, we characterize all the infinitesimal deformations of three-dimensional Heisenberg Hom-Lie superalgebras using cohomology. 2 Preliminaries Let V be a vector superspace over a field F; that is, a Z2-graded vector space with a di- rect sum decomposition V = V0 ⊕V1. The elements of Vj, j = 0, 1, are called homogeneous of parity j. The parity of homogenous element x is denoted by x. Moreover, the super- space End(V ) have a natural direct sum decomposition End(V ) = End(V )0 ⊕ End(V )1, where End(V )j = {f f (Vi) ⊆ Vi+j}, j = 0, 1. Elements of End(V )j are homogeneous of parity j. We review the definition of Hom-Lie superalgebra in [3]. Definition 2.1. A Hom-Lie superalgebra g = (V, [·, ·], α) is a triple consisting of a super- space V over a field F, an even bilinear map [·, ·] : V × V −→ V and an even superspace homomorphism α : V −→ V satisfying [x, y] = −(−1)xy[y, x](skew − supersymmetry), (cid:9)x,y,z (−1)xz[α(x), [y, z]] = 0(hom − Jacobi identity) (2.1) (2.2) for all homogenous elements x, y, z ∈ V , where (cid:9)x,y,z denotes the cyclic summation over x, y, z. We denote g0 = gV0, g1 = gV1 and then g = g0 ⊕ g1. It follows that g is a Hom-Lie algebra when g1 = 0. The classical Lie superalgebra can be obtained when α = id. A hom-Lie superalgebra is called multiplicative, if α([x, y]) = [α(x), α(y)], ∀x, y. It is obvious that the classical Lie superalgebras are a special case of multiplicative hom-Lie superalgebras. The center of Hom-Lie superalgebra g = (V, [·, ·], α) is defined by Z(g) = {x ∈ V : [x, y] = 0, ∀y ∈ V }. Two hom-Lie superalgebras (V, [·, ·]1, α) and (V, [·, ·]2, β) are said to be isomorphic if there exists an even bijective homomorphism φ : (V, [·, ·]1) → (V, [·, ·]2) satisfying φ([x, y]1) = [φ(x), φ(y)]2, ∀x, y ∈ V φ ◦ α = β ◦ φ. 2 In particular, (V, [·, ·], α) and (V, [·, ·], β) are isomorphic if and only if there exists an even automorphism φ such that β = φαφ−1. Let V be a vector superspace as before. A bilinear form B on V is called homogeneous of parity j if it satisfies B(x, y) = 0, ∀x, y ∈ V, x 6= y + j; skew-supersymmetric if B(x, y) = −(−1)xyB(y, x) for all homogenous elements x, y ∈ V ; non-degenerate if from B(x, y) = 0 for all x ∈ V , it follows that y = 0. In this paper, we only discuss multiplicative Hom-Lie superalgebras over the complex field C and the elements mentioned are homogenous. 3 Heisenberg Hom-Lie superalgebras Let g be a finite-dimensional Hom-Lie superalgebra with a 1-dimensional homogenous derived ideal such that [g, g] ⊂ Z(g). Let h ∈ Z(g) be the homogenous generator of [g, g]. Then a homogenous skew-supersymmetric bilinear form ¯B can be defined on g via [x, y] = ¯B(x, y)h, ∀x, y ∈ g. This induces a homogenous skew-supersymmetric bilinear form B on g/Z(g) via B(x + Z(g), y + Z(g)) = ¯B(x, y). Definition 3.1. A Hom-Lie superalgebra g is called a Heisenberg Hom-Lie superalgebra if the derived ideal [g, g] is generated by a homogenous element h ∈ Z(g) and B is non- degenerate. From now on, we will also denote a Hom-Lie superalgebra by (h, α), where h = (V, [·, ·]h) is a superalgebra and α is an even linear map. All brackets unmentioned in the following are zero. Let g = (V, [·, ·], α) be a 3-dimensional Heisenberg Hom-Lie superalgebra with a direct sum decomposition g = g0 ⊕ g1. Let h ∈ Z(g) be the homogenous generator of the derived ideal [g, g], We analyze the cases h ∈ g0 and h ∈ g1 separately. Case 1. If h ∈ g0, we have two subcases: (1.1) There are u1, u2 ∈ g0 such that {u1, u2, h} is a basis of g and [u1, u2] = h, which implies that g is a Hom-Lie algebra. (1.2) There are v1, v2 ∈ g1 such that {hv1, v2} is a basis of g and [v1, v2] = h. Then the Hom-Lie superalgebra will be denoted by (h1, α). Case 2. If h ∈ g1, there exist u ∈ g0, v ∈ g1 such that {uv, h} is a basis of g and [u, v] = h. In this case, we denote the Hom-Lie superalgebra by (h2, α). Theorem 3.2. Let g be a multiplicative Heisenberg Hom-Lie(non-Lie) superalgebra of dimension three. Then g must be isomorphic to one of the following: (1) (h1,  µ11µ22 0 0 0 µ11 0 0 0 µ22  ), µ11µ22 6= 0; 3 (2) (h1,  (3) (h1,  (4) (h2,  (5) (h2,  µ12µ21 0 0 0 0 µ21 0 µ12 0 0 0 0 0 µ11 µ12 0 0 0 µ0 0 0 µ11 0 0 µ0µ11  ), µ12µ21 6= 0; 0 0  );  );  ), (µ0 − 1)µ11 = 0, 0 0 µ11 where µ0, µij ∈ C, i, j = 1, 2. µ0 0 0 µ11 1 0 Proof. We analyze the cases h ∈ g1 and h ∈ g0 separately. Case 1. If h ∈ g1, there exists a basis {uv, h} of g such that [u, v] = h. Suppose that 0 µ0 0 0 µ11 µ12 0 µ21 µ22 α =   , µ0, µij ∈ C, i, j = 1, 2. We have that g is multiplicative if and only if α([ei, ej]) = [α(ei), α(ej)] for i, j = 1, 2, 3, which implies µ12 = 0 and µ22 = µ0µ11. Then we obtain that α =  (a)If µ21 = 0, we obtain a Heisenberg Hom-Lie superalgebra µ0 0 0 µ11 0 µ21 µ0µ11 0 0   . (b)If µ21 6= 0, let b0 0 0 φ =  0 b11 b21 0 0 b0b11 Then φαφ−1 =  =  b0 0 0 µ0 0 0 0 b11 b21 0 0 b0b11   0 µ11 (1 − µ0)µ11b11 0 0 µ0µ11  ). µ0 0 0 µ11 0 0 (h2,    , φ−1 =    µ0 0 0 µ11 0 µ21 µ0µ11 0 0   0 0     . −1b21 + µ21b0 µ0a11 4 −1 b0 0 0 −b0 0 −1 b11 −1b11 −2b21 0 0 b0 −1b11 −1   . −1 b0 0 0 −b0 0 −1 b11 −1b11 −2b21 0 0 b0 −1b11 −1   If µ0 6= 1 and µ11 6= 0, then b21 = −(1 − µ0)−1µ11 −1µ21b0b11 yields φαφ−1 =  µ0 0 0 µ11 0 0 0 0 µ0µ11   , which induces a Heisenberg Hom-Lie superalgebra like the one in (a). Otherwise, i.e.µ0 = 1 or µ11 = 0, then b0 = µ21 We can obtain a new Heisenberg Hom-Lie superalgebra −1 yields φαφ−1 =  µ0 0 0 µ11 1 0 0 0 µ11  . µ0 0 0 µ11 0 1 0 0 µ11 (h2,   ), (µ0 − 1)µ11 = 0. Case 2. If h ∈ g0, there exist v1, v2 ∈ g1 such that {hv1, v2} is a basis of g and [v1, v2] = h. In this case, we can get three Heisenberg Hom-Lie superalgebras: (h1,  and (h1,  µ11µ22 0 0 0 µ11 0 0 0 0 0 µ11 µ12 0 0 0  )(µ11µ22 6= 0), (h1,  µ12µ21 0 0 0 0 µ21 0 µ12 0  )(µ12µ21 6= 0) 0 0 µ22  ). 4 The adjoint cohomology of Heisenberg Hom-Lie superalgebras Let g = (V, [·, ·], α) be a Hom-Lie superalgebra. Let x1, · · · , xk be k homogeneous elements of V and (x1, · · · , xk) ∈ ∧kV . Then we denote by (x1, · · · , xk) = x1+· · ·+xk the parity of (x1, · · · , xk). The set C k α(g, g) of k-hom-cochains of g = (V, [·, ·], α) is the set of k-linear maps ϕ : ∧kV → V satisfying ϕ(x1, · · · , xi+1, xi · · · , xk) = −(−1)xixi+1ϕ(x1, · · · , xi, xi+1 · · · , xk), α(ϕ(x1, · · · , xk)) = ϕ(α(x1), · · · , α(xk)) (4.1) (4.2) for x1, · · · , xk ∈ V , 1 6 i 6 k − 1. In particular, C 0 α(g, g) = {x ∈ gα(x) = x}. Denote by ϕ the parity of ϕ and ϕ(x1, · · · , xk) = (x1, · · · , xk) + ϕ. We immediately get a direct sum decomposition C k α(g, g)0 ⊕ C k α(g, g) = C k α(g, g)1. 5 A k-coboundary operator δk(ϕ) : C k α(g, g) → C k+1 α (g, g) is defined by δk(ϕ)(x0, · · · , xk) = X06s<t6k (−1)t+xt(xs+1+···+xt−1) × ϕ(α(x0), · · · , α(xs−1), [xs, xt], α(xs+1), · · · , xt, · · · , α(xk)) + (−1)s+xs(ϕ+x0+···+xs−1)[αk−1(xs), ϕ(x0, · · · , xs, · · · , xk)], k Xs=1 where xi means that xi is omitted. The k-cocycles space, k-coboundaries space and k-th cohomology space are defined as: (1) Z k(g, g) = kerδk, Z k(g, g)j = Z k(g, g) ∩ C k (2) Bk(g, g) = Imδk−1, Bk(g, g)j = Bk(g, g) ∩ C k (3)H k(g, g) = Z k(g,g) α(g, g)j, j = 0, 1; α(g, g)j, j = 0, 1; Bk(g,g) = H k(g, g)0 ⊕ H k(g, g)1, where H k(g, g)j = Z k(g,g)j Bk(g,g)j , j = 0, 1. Theorem 4.1. The cohomology spaces of Heisenberg Hom-Lie superalgebras are: (1) For g = (h1,  µ11µ22 0 0 0 µ11 0 0 0 µ22 a22 + a33 a12δµ22,1 a13δµ11,1 0 0 a22 0  )(µ11µ22 6= 0),  + , 0 a33 0 0 0 0 δµ11,1a22 0 0 0 0 0 0 δµ22,1a34 0 0 − 1 2 µ22a22δµ11,1 − 1 2µ11a34δµ22,1 0 0  + ; H 1(g, g) =  H 2(g, g) =  H 1(g, g) =  H 2(g, g) =  (2) For g = (h1,  µ12µ21 0 0 0 0 µ21 2a22 a12δµ12µ21,1 a12δµ12µ21,1 0 µ12 0  )(µ12µ21 6= 0),  + , 0 a22 0 0 0 a22 0 0 0 0 −2µ21a23δµ12µ21,1 0 0 (3) For g = (h1,  0 0 0 0 µ11 µ12 0 0 0  ), 0 (a22 + a33)δµ12,0 H 1(g, g) =  0 0 a22δµ12,0 a13 0 0 a33δµ12,0 6  + ,  µ12a23δµ12µ21,1 −2µ12 a23δµ12µ21,1 0 0 0 0 2a23δµ12µ21,1 0 0 0 0  + ;  0 0 0 a22δµ11(µ11−1),0 a23δµ11,0 + µ12a22δµ11,1 a24δµ11,0 + ǫa22δµ11,1 0 δµ11,0a26 0 δµ12,0a14 a16 0 0 0 0 0 0 µ0 0 0 µ11 0 0 0 0 µ0µ11  ), 0 0 0 a22 a11 a21δµ0,µ11 a31δµ0(µ11−1),0 a32δ(µ0−1)µ11 ,0 a11 + a22  + , H 2(g, g) = 0 where ǫ = µ11   (4) For g = (h2,  −1µ12 2; H 1(g, g) =  H 2(g, g) = 0   (5) For g = (h2,   +,  + ;   0 0 a22δµ11,1 0 a32δµ11,0 a33δµ11,0 a34δµ0µ11(µ0µ11−1),0 0 0 0 0 1 2µ0a22δµ11,1 −µ0a34δµ0µ11,1 a26δ(µ0 2−1)µ11,0 a36δµ0(µ0−1)µ11,0 0 0 µ0 0 0 µ11 1 0 0 0 µ11  ), (µ0 − 1)µ11 = 0, H 1(g, g) =  H 2(g, g) = 0 0 0 0 a31δµ0,µ11 a32δ(µ0−1)µ11,0 a33 0 a33  + ,  a12δµ0,0δµ11,0 0 0 2a33δµ11,1 + a34δµ11,0 0 a32δµ11(µ11−1),0   a13δµ0,0δµ11,0 0 0 0 a33δµ11(µ11−1),0 a34δµ0,2δµ11,0 0 0 a16δµ0,0 µ0a36 0 a36  + . Proof. It is easy to obtain C k α(g, g) for k = 1, 2 by Eq.(4.1) and Eq.(4.2). µ11µ22 0 0 0 a22 a32δµ11,µ22 Taking g = (h1,  α(g, g) =  α(g, g) =  a11 0 0 C 1 C 2 0 a12δµ11,µ22 a13 a14δµ11,µ22 0 0 0 0 0 0 0 0 0 µ11 0 0 0 µ22 0 a23δµ11,µ22 a33   , µ11µ22 6= 0 for example,  , 0 0  . a25δµ11µ22,1 a35δµ11 2,1 a26δµ22 2,1 a36δµ11µ22,1 7 α(g, g)0. Then B2(g, g)0 = {δ1ϕ0ϕ0 ∈ C 1 a11 0 0 0 0 a22 a23 a32 a33 Moreover, we have ϕ0 ∈ Z 1(g, g)0 if and only if δ1(ϕ0) = 0. 0 2a32 a22 + a33 − a11 2a23 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Let ϕ0 =    ∈ C 1 α(g, g)0} =  0  In the same way, we suppose ϕ1 =   ∈ C 1(g, g)1 and immediately     and ψ0 =   ∈    ∈ C 2 Now suppose α =   α(g, g)0(ψ1 =  µ0 0 0 µ11 µ12 0 µ21 µ22 0 0 0 0 a22 a23 a24 0 a32 a33 a34 α(g, g)1) if and only if δ1(ψ0) = 0(δ1(ψ1) = 0). 0 a12 a13 a14 0 0 0 0 0 0 0 0 0 a21 a31 a12 a13 0 0 0 0 a15 a16 0 0 0 0 get Z 1(g, g)1 and B2(g, g)1. 0  + . 0 0 a25 a26 a35 b36 C 2 (ψ1 ∈ Z 2 α(g, g)1). We know that ψ0 ∈ Z 2 α(g, g)0 5 Infinitesimal deformations of Heisenberg Hom-Lie superalgebras Let g = (V, [·, ·]0, α) be a Hom-Lie superalgebra and ϕ : V × V −→ V be an even bilinear map commuting with α. A bilinear map [·, ·]t = [·, ·]0 + tϕ(·, ·) is called an infinitesimal deformation of g if ϕ satisfies [x, y]t = −[y, x]t, (cid:9)x,y,z (−1)xz[α(x), [y, z]t]t = 0 (5.1) (5.2) for x, y, z ∈ V . The previous E.q.(5.1) imply ϕ is skew-supersymmetric. We denote ϕ ◦ ψ(x, y, z) =(cid:9)x,y,z (−1)xzϕ(α(x), ψ(y, z)), and then Eq.(5.2) can be denoted by [·, ·]t ◦ [·, ·]t = 0. Lemma 5.1. Let g = (V, [·, ·]0, α) be a Hom-Lie superalgebra and [·, ·]t = [·, ·]0 + tϕ(·, ·) be an infinitesimal deformation of g = (V, [·, ·]0, α). Then ϕ ∈ Z 2(g, g)0. Proof. By Eq.(5.2), we have 0 = [·, ·]t ◦ [·, ·]t . =(cid:9)x,y,z (−1)xz([α(x), [y, z]t]0 + tϕ(α(x), [y, z]t)) (5.3) =(cid:9)x,y,z (−1)xz(cid:2)t([α(x), ϕ(y, z)]0 + ϕ(α(x), [y, z]0)) + t2ϕ(α(x), ϕ(y, z))(cid:3) . 8 Note that (cid:9)x,y,z (−1)xz([α(x), ϕ(y, z)]0 + ϕ(α(x), [y, z]0)) = (−1)xzδ2ϕ. Hence ϕ ∈ Z 2(g, g)0. By Eq.(5.3), we can see that [·, ·]t is an infinitesimal deformation if and only if ϕ ◦ ϕ = 0. Let gt = (V, [·, ·]t, α) and g′ [·, ·]0 + tϕ(·, ·) and [·, ·]′ id + tφ(φ ∈ C1 α(g, g)0), satisfying t = (V, [·, ·]′ t = [·, ·]0 + tψ(·, ·). t, α′) be two deformations of g, where [·, ·]t = If there exists a linear automorphism Φt = Φt([x, y]t) = [Φt(x), Φt(y)] ′ t, ∀x, y ∈ V we call the deformations gt and g t are equivalent if and only if ϕ1 − ψ1 ∈ B2(g, g)0. Therefore, the set of infinitesimal deformations of g can be parameterized by H 2(g, g)0. t are equivalent. It is obvious that gt and g ′ ′ A deformation gt of Hom-Lie superalgebras g is called trivial if it is equivalent to g. Corollary 5.2. All the infinitesimal deformations of the following Heisenberg Hom-Lie superalgebras are trivial: (1) (h1,  (2) (h1,  (3) (h2,  µ11µ22 0 0 µ12µ21 0 0 0 µ11 0 0 µ21 µ0 0 0 µ11 0 0 0 0 µ0µ11 0 0 µ22 0 µ12 0  ), µ11µ22 6= 0;  ), µ12µ21 6= 0;  ), µ0(µ0 2 − 1)µ11 6= 0. Proof. All the infinitesimal deformations of a Heisenberg Hom-Lie superalgebras are trivial if and only if H 2(g, g)0 = 0. In the following, we discuss the non-trivial infinitesimal deformations of Heisenberg Hom-Lie superalgebras. We will distinguish two separate cases: the ones that are also Lie superalgebras and those are not. We recall the classification of three-dimensional Lie superalgebras( [11]): Theorem 5.3. Let L = (V, [·, ·]) be a Lie superalgebras with a direct sum decomposition V = V0 ⊕ V1, where dimV0 = 1 and dimV1 = 2. There are e1 ∈ V0 and e2, e3 ∈ V1 such that {e1e2, e3} is a basis of V . Then L must be isomorphic to one of the following: 9 L1 : [e1, V1] = 0, [e2, e2] = e1, [e3, e3] = [e2, e3] = 0; L2 : [e1, V1] = 0, [e2, e2] = [e3, e3] = e1, [e2, e3] = 0; Lλ L4 : [e1, e2] = e2, [e1, e3] = e2 + e3, [V1, V1] = 0; L5 : [e1, e2] = 0, [e1, e3] = e2, [V1, V1] = 0. 3 : [e1, e2] = e2, [e1, e3] = λe3, [V1, V1] = 0; We construct a new Lie superalgebra L2 ′ = (V, [·, ·]0). Let {hv1, v2} be a basis of V satisfying [v1, v2]0 = h. Ther is an even bijective morphism φ : (V, [·, ·]0) → L2 φ(h) = e1, φ(v1) = e2 + ie3, φ(v2) = 1 2 e2 − 1 2 ie3 such that φ([x, y]0) = [φ(x), φ(y)], ∀x, y ∈ V. Then L2 replace L2 with it in Theorem 5.3. ′ is isomorphic to L2 and we shall Proposition 5.4. A non-trivial infinitesimal deformation of Heisenberg Hom-Lie super- algebra (h1, α), that is also a Lie superalgebra, is isomorphism to 0 0 0 0 µ11 µ12 0 0 0 ′ (L2 ,   ). Proof. Denote by (h1, α) = (V, [·, ·]0, α). There is a basis {hv1, v2} such that [v1, v2]0 = h and others are zero. the infinitesimal deformations of (g1, α) are trivial. If α =   µ11µ22 0 0 0 µ11 0 0 0 µ22 Consider α =   0 0 0 0 µ11 µ12 0 0 0  (µ11µ22 6= 0) or    . Let ϕ =   µ12µ21 0 0 0 0 µ21 0 µ12 0 0 0 0 a14δµ12,0 0 0 0 0 0 0 0 0 0  (µ12µ21 6= 0), all   be 0 a26δµ11,0 0 0 0 an even 2-cocycle and [·, ·]t = [·, ·]0 + tϕ(·, ·). Then ϕ ◦ ϕ = 0 and gt = (V, [·, ·]t, α′) is an infinitesimal deformation of (h1, α). Moreover, gt is a Lie superalgebra if and only if [·, ·]t ◦ [·, ·]t = 0, i.e. a26 = 0. All deformations are given in Table 1. µ11 µ12 [·, ·]t 6= 0 0 [v1, v2]t = h [v2, v2]t = a14h   a14 6= 0 Table 1 base change 1 0 0 1 1 0 0 0 2 a14 a14 6= 0 Hom-Lie superalgebra ′ 1   (L2 ,  µ12 ′ 0 0 0 0 µ11 µ12 0 0 ′ = − 1 2 a14µ11 0  ) Proposition 5.5. The non-trivial infinitesimal deformations of (h2, α), that are also Lie superalgebras, are isomorphism to: 10 (1) For α =  (a) λ, L3  (c)  −1, L3  (2) For α =  0,  µ0,  (a)  L3 (c)  L3 µ0 0 0 µ11 0 0 µ0 0 0 µ11 0 0 2 − 1)µ11 = 0), 0 0 µ0µ11 0 0 µ11   (µ0(µ0    (µ0−1)µ11 = 0,  0 1 2µ11 0 0 0 1 2ξ−1µ11 0 1 2ξµ11 1 2µ11    ;  µ0 0 0 µ11 1 0 0 0 0 0 0 0 0 0 η 0 0 µ11  ,   (b)  L4,  ,   0    , µ0 6= 0, 1.  µ0 0 0 0 0 −1 0 µ0 0 (b) L3 0,  −1 0 0 0 0 µ11 −µ11 0 −2µ11     , 1 0 0 µ11 0 0 0 0 µ11     , Proof. Denote by (h2, α) = (V, [·, ·]0, α). There is a basis {uv, h} such that [u, v]0 = h and others are zero. (1) Consider α =  Let ϕ =  µ0 0 0 µ11 0 0 0 0 µ0µ11 0 0 0 0 0 0 0 0 2−1)µ11,0 0 0 0 0 a36δµ0(µ0−1)µ11,0 a26δ(µ0 0 2 − 1)µ11 = 0.  , µ0(µ0   is an even 2-cocycle. Then ϕ ◦ ϕ = 0 and we obtain an infinitesimal deformation gt = ((V, [·, ·]0, α)′) is an infinitesimal deformation, where [·, ·]t = [·, ·]0 + tϕ(·, ·). It is easy to see that is also a Lie algebra. We analyze the cases (a)(µ0 − 1)µ11 = 0, (b)µ0 = −1 and (c)µ0 = 0 separately, which are given in Table 2. (2) Consider α =  Let ϕ =   µ0 0 0 µ11 0 1 0 0 µ11   , (µ0 − 1)µ11 = 0. 0 a12δµ0,0δµ11,0 a13δµ0,0δµ11,0 0 0 0 0 0 0 0 0 µ0a36 0 0 0 0 0 a36   be an even 2-cocycle. Then gt = ((V, [·, ·]0, α)′) is an infinitesimal deformation, where [·, ·]t = [·, ·]0 + tϕ(·, ·), if and only if ϕ ◦ ϕ = 0. We analyze the cases (a)µ0 = µ11 = 0, (b)µ0 = 1 and (c)µ0 6= 0, 1, µ11 = 0 separately. For case (a), ϕ◦ϕ = 0 implies a12 = a13 = 0 or a36 = 0. Furthermore, if a12 = a13 = 0, 11 (a) (µ0 − 1)µ11 = 0. [·, ·]t [u, v]t = h [u, h]t = a26v + a36h   a26a36 6= 0 [u, v]t = h [u, v]t = a36h a36 6= 0 [u, v]t = h [u, v]t = a26v a26 6= 0 (b) µ0 = −1. [·, ·]t [u, v]t = h [u, v]t = a36h a36 6= 0 (c) µ0 = 0. [·, ·]t [u, v]t = h [u, v]t = a26v a26 6= 0     Table 2 base change Hom-Lie superalgebra     −1 0 k1 0 τ a26 0 −τ a36 0 −1 −τ k2 −1 τ k1 −1 τ = (k1 − k2)−1 k1 6= k2, k1k2 = −a26 −1 k1 + k2 = −a36a26 0 0 −1 1 a36 a36 a36 0 0 −1 0     a36 6= 0 1 2 0 κ a26 0 0 −κa26 κ = (1 + a26)−1 0 κa26 κa26 1 2 3 2 a26 6= 0 base change 0 a36 0 −1 1 a36 0 −1 0 0 a36 a36 6= 0      L3 λ,  0 µ0 0 µ11 0 0 0 0 µ11     λ = k1k2 −1, λ 6= 0, −1 (µ0 − 1)µ11 = 0 µ0 0 0 µ11 0 0 0 0 µ11 (µ0 − 1)µ11 = 0 0 0 µ11 0 µ0 0 µ11 0 0  0, L3   −1, L3  (µ0 − 1)µ11 = 0         Hom-Lie superalgebra  L3 0,  0 −1 0 0 −µ11 µ11 0 2µ11 0     base change Hom-Lie superalgebra 0 1 2µ11 0 0 0 1 2 ξ−1µ11 ξ = −a26 − 1 2 0 1 2ξµ11 1 2 µ11     1 2 0 ρ a26 0 0 −ρa26 ρ = (1 + a26)−1 0 ρa26 ρa26 1 2 3 2    L3 0,  a26 6= 0 12 (a) µ0 = µ11 = 0. [·, ·]t [u, v]t = h [u, v]t = a36h a36 6= 0 (b) µ0 = 1. [·, ·]t [u, v]t = a36v + h [u, v]t = a36h a36 6= 0 (c) µ0 6= 0, 1, µ11 = 0. Table 3 base change a36 0 0 0 0 −1 0 a36 1 a36 a36 6= 0 −1     Hom-Lie superalgebra  L3 0,  0 0 0 0 0 0 0 0 µ22 µ22 = a36     base change 0 a36 0 0 a36 0 0 0 1 a36 6= 0     Hom-Lie superalgebra  L4,  1 0 0 µ11 0 0 0 0 µ11     [·, ·]t base change Hom-Lie superalgebra [u, v]t = µ0a36v + h [u, v]t = a36h a36 6= 0 a36 0 0   0 0 −1 (µ0 − 1)a36 µ0 0 a36 6= 0 0    L3 λ,  0 −1 0 µ0 0 µ0 0 0 0 λ 6= 0, 1 0     the deformation gt is also a Lie superalgebra. For cases (b) and (c), gt is an infinitesimal deformation for all ϕ. The deformation is also a Lie superalgebra if a12 = a13 = 0. The deformations of (a), (b) and (c) are given in Table 3. Proposition 5.4 and Proposition 5.5 give the infinitesimal deformations of Heisenberg Hom-Lie superalgebras that are also Lie superalgebras. Before discussing the rest defor- mations, we will recall some multiplicative Hom-Lie superalgebras and those can be find in the classification of multiplicative Hom-Lie superalgebras of [10]. Let V be a superspace with a direct sum decomposition V = V0 ⊕ V1, [·, ·] be an even bilinear map and σ be an even linear map on V . Let {e1e2, e3} be a basis of V . The following are three Hom-Lie superalgebras on V : L43,a 1,2 (a 6= 0), : [e1, e2] = 0, [e1, e3] = βe2, [e2, e2] = 0, [e2, e3] = 0, [e3, e3] = γe1, σ =   1,2 : [e1, e2] = 0, [e1, e3] = βe2, [e2, e2] = 0, [e2, e3] = νe1, [e3, e3] = γe1, σ =  (a 6= 0), L45,a 0 0 0 0 0 a 0 0 0 0 0 0 0 0 a 0 0 0     13 L46,a,b 1,2 : [e1, e2] = 0, [e1, e3] = βe2, [e2, e2] = 0, [e2, e3] = µe1, [e3, e3] = 0, σ =   0 0 0 0 0 a 0 0 b   (a2 + b2 6= 0). The following Proposition 5.6 and Proposition 5.7 will characterize the infinitesimal deformations of Heisenberg Hom-Lie superalgebras that are not Lie superalgebras. Proposition 5.6. An infinitesimal deformation of Heisenberg Hom-Lie superalgebra (h1, α), which is not a Lie superalgebra, is isomorphic to L46,a,0 (a 6= 0). 1,2 Proof. By the proof of Proposition 5.4, we know that (h1, α) have an infinitesimal deformation(not a Lie superalgebra) if and only if 0 0 0 0 0 µ12 0 0 0 α =    (µ12 6= 0), ϕ =  0 0 0 0 0 0 0 0 0 0 0 a26 0 0 0 0 0 0   (a26 6= 0). There is a basis {hv1, v2} of V such that [v1, v2]0 = h. Therefore gt = (V, [·, ·]t, α′) is an infinitesimal deformation, where [·, ·]t = [·, ·]0 + tϕ and [v1, v2]t = h, [h, v2]t = a26v1(a26 6= 0). Then the deformation gt is isomorphic to L46,µ12,0 : 1,2 [e1, e2] = 0, [e1, e3] = a26e2, [e2, e2] = 0, [e2, e3] = e1, [e3, e3] = 0, σ =  0 0 0 0 0 µ12 0 0 0  . Proposition 5.7. An infinitesimal deformation of (h2, α), which is not a Lie superalgebra, is isomorphic to L43,1 1,2 , L45,1 1,2 , or L46,1,0 1,2 . Proof. By the proof of Proposition 5.5, Heisenberg Hom-Lie superalgebra (h2, α) have an infinitesimal deformations if and only if α =  0 0 0 0 0 0 0 1 0   , ϕ =  0 a12 a13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0   . There is a basis {uv, h} of V such that [u, v]0 = h. Therefore gt = (V, [·, ·]t, α′) is an infinitesimal deformation, where [·, ·]t = [·, ·]0 + tϕ and [u, v]t = h, [v, v]t = a12u, [v, h]t = a13u. We analyze into three cases. (a) If a12 6= 0 and a13 6= 0, the infinitesimal deformation gt is isomorphic to L45,1 (b) If a12 = 0 and a13 6= 0, gt is isomorphic to: 1,2 : [e1, e2] = 0, [e1, e3] = e2, [e2, e2] = 0, [e2, e3] = a13e1, [e3, e3] = a12e3, σ =  0 0 0  . : [e1, e2] = 0, [e1, e3] = e2, [e2, e2] = 0, [e2, e3] = a13e1, [e3, e3] = 0, σ =  0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 1,2 L46,1,0  . 14 (c) If a12 6= 0 and a13 = 0, gt is isomorphic to: L43,1 1,2 : [e1, e2] = 0, [e1, e3] = e2, [e2, e2] = 0, [e2, e3] = 0, [e3, e3] = a12e1, σ =   0 0 0 0 0 1 0 0 0  . References [1] M. A. Alvarez, F. Cartes: Cohomology and deformations for the Heisenberg Hom-Lie algebras. Linear Multilinear Algebra. DOI:10.1080/03081087.2018.1487379. [2] F. Ammar, Z. Ejbehi, A. Makhlouf: Cohomology and deformations of Hom-algebras. J. Lie Theory 21 (2011), 813-836. [3] F. Ammar, A. Makhlouf: Cohomology of Hom-Lie superalgebras and q-deformed Witt superalgebra. Czech. Math. J. 63 (2013), 721-761. [4] F. Ammar, A. Makhlouf: Hom-Lie superalgebras and Hom-Lie admissible superal- gebras. J. Algebra 324 (2010), 1513-1528. [5] Y. Liu, L. Chen, Y. Ma: Hom-Nijienhuis operator and T-exentions of hom-Lie superalgebras. Linear Algebra Appl. 439 (2013), 2131-2144. [6] A. Makhlouf, S. D. Silvestrov: Notes on 1-parameter formal deformations of Hom- associative and Hom-Lie algebras. Forum Math. 22 (2010), 715-739. [7] R. Peniche, O.A. S´anchez-Valenzuela: On Heisenberg-like super group structures. Ann. Henri Poinca´re 10 (2010), 1395-1417. [8] M. Rodr´ıguez-Vallarte, G. Salgado, O. S´anchez-Valenzuela: Heisenberg Lie superal- gebra and their invariant superorthogonal and supersymplectic forms. J. Algebra 332 (2011), 71-86. [9] Y. Sheng: Representations of Hom-Lie algebras. Algebra. Represent. Theory 15 (2012), 1081-1098. [10] C. Wang, Q. Zhang, Z. Wei: A classification of low dimensional multiplicative Hom- Lie superalgebras. Open Math. 14 (2016), 613-628. [11] Z. Wang: A classification of low dimensional Lie superalgebras(Chinese). Master thesis, East China Normal University. 15
1701.01899	1	1701	2017-01-08T00:57:04	Buchberger-Zacharias Theory of Multivariate Ore Extensions	[ "math.RA" ]	We present Buchberger Theory and Algorithm of Gr\"obner bases for multivariate Ore extensions of rings presented as modules over a principal ideal domain. The algorithms are based on M\"oller Lifting Theorem.	math.RA	math	Buchberger-Zacharias Theory of Multivariate Ore Extensions Michela Ceria Dipartimento di Matematica Universit`a di Trento Teo Mora DIMA Universit`a di Genova [email protected] [email protected] ] . A R h t a m [ 1 v 9 9 8 1 0 . 1 0 7 1 : v i X r a January 10, 2017 Abstract We present Buchberger Theory and Algorithm of Grobner bases for multivari- ate Ore extensions of rings presented as modules over a principal ideal domain. The algorithms are based on Moller Lifting Theorem. In her 1978 Bachelor's thesis [53] Zacharias discussed how to extend Buchberger Theory [7, 8, 10] from the case of polynomial rings over a field to that of polynomials over a Noetherian ring with suitable effectiveness conditions. In the meantime a similar task was performed in a series of papers - Kandri-Rody– Kapur [20] merged the rewriting rules behind Euclidean Algorithm with Buchberger's rewriting, proposing a Buchberger Theory for polynomial rings over Euclidean do- mains; Pan [39] studied Buchberger Theory for polynomial rings over domains intro- ducing the notions of strong/weak Grobner bases - which culminated with [34]. Such unsorpassed paper, reformulating and improving Zacharias' intuition, gave efficient solutions to compute both weak and strong Grobner bases for polynomial rings over each Zacharias ring, with particolar attention to the PIR case. Its main contribution is the reformulation of Buchberger test/completion ("a basis F is Grobner if and only if each S-polynomial between two elements of F, reduces to 0") in the more flexible lifting theorem ("a basis F is Grobner if and only if each element in a minimal basis of the syzygies among the leading monomials lifts, via Buchberger reduction, to a syzygy among the elements of F"). The only further contribution to this ultimate paper is the survey [6] of Moller's results which reformulated them in terms of Szekeres Theory [50]. The suggestion of extending Buchberger Theory to non-commuative rings which satisfy Poincar´e-Birkhoff-Witt Theorem was put forward by Bergman [5], effectively applied by Apel–Lassner [3, 4] to Lie algebras and further extended to solvable poly- nomial rings [21, 22], skew polynomial rings [15, 16, 17] and to other algebras [1, 11, 25, 26] which satisfy Poincar´e-Birkhoff-Witt Theorem and thus, under the standard in- tepretation of Buchberger Theory in terms of filration/graduations [2, 29, 32, 49, 28], have the classical polynomial ring as associated graded rings. 1 In particular Weispfenning [52] adapted his results to a generalization of the Ore extension [38] proposed by Tamari [51] and its construction was generalized by his student Pesch [40, 41] thus introducing the notion of iterative Ore extension with com- muting variables R := R[Y1; α1, δ1][Y2; α2, δ2] · · · [Yn; αn, δn], R a domain; the concept has been called Ore algebra in [12] and is renamed here as multivariate Ore extension (for a different and promising approach to Ore algebras see [23]). Bergman's approach and most of all extensions are formulated for rings which are vector spaces over a field; in our knowledge the only instances in which the coeffi- cient ring R is presented as a D-module over a domain D (or at least as a Z-module) are Pritchard's [42, 43] extension of Moller Lifting Theory to non-commutative free algebras and Reinert's [44, 45] deep study of Buchberger Theory on Function Rings. Following the recent survey on Buchberger-Zacharias Theory for monoid rings R[S] over a unitary effective ring R and an effective monoid S [31], we propose here a Moller–Pritchard lifting theorem presentation of Buchberger-Zacharias Theory and re- lated Grobner basis computation algorithms for multivariate Ore extensions. The twist w.r.t. [31] is that there R[S] coincides with its associated graded ring; here R and its associated graded ring G(R) := R[Y1; α1][Y2; α2] · · · [Yn; αn] coincide as sets and as left R-modules between themselves and with the commutative polynomial ring R[Y1, · · · , Yn], but as rings have different multiplications. We begin by recalling Ore's original theory [38] of non-commutative polynomials R[Y], relaxing the original assumption that R is a field to the case in which R is a domain (Section 1.1) and Pesch's constructions of multivariate Ore extensions (Section 1.2) and graded multivariate Ore extensions (Section 1.3), focusing on the arithmetics of the main Example 14 R := R[Y1; α1][Y2; α2] · · · [Yn; αn], R := Z[x] αi(x) := cixei , ci ∈ Z \ {0}, ei ∈ N \ {0}. Next we introduce Buchberger Theory in multivariate Ore extensions recalling the notion of term-orderings (Section 2.1), definition and main properties of left, right, bi- lateral and restricted Grobner bases (Section 2.2) and Buchberger Algorithm for com- puting canonical forms in the case in which R is a field (Section 2.3). We adapt to our setting Szekeres Theory [50] (Section 3), Zacharias canonical rep- resentation with related algorithm (Section 4) and Moller Lifting Theorem (Section 5). The next Sections are the algorithmic core of the paper: we reformulate for multi- variate Ore extensions over a Zacharias ring – Moller's algorithm for computing the required Gebauer–Moller set (id est the minimal basis of the module of the syzygies among the leading monomials) for Buchberger test/completion of left weak bases (Section 6.1); – his reformulation of it which requires only l.c.m. computation in R for the case in which R is a PIR (Section 6.2); 2 – still in the case in which R is a PIR, Moller's completion of a left weak basis to a left strong one (Section 6.3); – Gebauer–Moller criteria [18] for producing a Gebauer–Moller set (Section 6.4); – Kandri-Rody–Weispfenning completion [21] of a left weak bases for producing a bilateral one (Section 7.1); – Weispfenning's [52] restricted completion (Section 7.2); – as a technical tool required by Weispfenning's restricted completion, how to pro- duce right Gebauer–Moller sets (Section 7.3). Finally we reverse to a theoretical survey summarizing the structural theorem for the case in which R is a Zacharias ring over a PID (Section 8), specializing to our setting Spear's Theorem [48, 28] (Section 9) and extending to it Lazard's Structural Theorem [24] (Section 10). In an appendix we discuss, as far as it is possible, how to extend this theory and algorithms to the case in which R is a PIR (Section A). 1 Recalls on Ore Theory 1.1 Ore Extensions Let R be a not necessarily commutative domain; Ore [38] investigated under which con- ditions the R-module R := R[Y] of all the formal polynomials is made a ring under the assumption that the multiplication of polynomials shall be associative and both-sided distributive and the limitation imposed by the postulate that the degree of a product shall be equal to the sum of the degree of the factors. It is clear that, due to the distributive property, it suffices to define the product of two monomials bYr · aY s or even more specifically, to define the product Y · r, r ∈ R; this necessarily requires the existence of maps α, δ : R → R such that Y · r = α(r)Y + δ(r) for each r ∈ R; Ore calls α(r) the conjugate and δ(r) the derivative of r. Under the required postulate clearly we have 1. for each r ∈ R, α(r) = 0 =⇒ r = 0, so that α is injective. It is moreover sufficient to consider, for each r, r′ ∈ R, the relations α(r + r′)Y + δ(r + r′) = Y · (r + r′) = Y · r + Y · r′ = (cid:0)α(r) + α(r′)(cid:1)Y + δ(r) + δ(r′), α(rr′)Y + δ(rr′) = Y · (rr′) = (Y · r) · r′ = α(r)α(r′)Y + α(r)δ(r′) + δ(r)r′, and, if R is a skew field, and r , 0, Y = (Y · r) · r−1 = α(r)α(r−1)Y + α(r)δ(r−1) + δ(r)r−1, to deduce that 3 2. α is a ring endomorphism; 3. the following conditions are equivalent: (a) for each d ∈ R \ {0} exists c ∈ R \ {0} : Y · c = dY + δ(c), α(c) = d; (b) α is a ring automorphism; 4. δ is an α-derivation of R id est an additive map satisfying δ(rr′) = α(r)δ(r′) + δ(r)r′ for each r, r′ ∈ R; 5. if R is a skew field, then each r ∈ R \ {0} satisfies α(r−1) = (α(r))−1 , δ(r−1) = − (α(r))−1 δ(r)r−1; 6. Im(α) ⊂ R is a subring, which is an isomorphism copy of R; 7. R1 := {r ∈ R : r = α(r)} ⊂ R is a ring, the invariant ring of R; 8. R0 := {r ∈ R : δ(r) = 0} ⊂ R is a ring, the constant ring of R; 9. {r ∈ R : Y · r = rY} = R0 ∩ R1. 10. If R is a skew field, such are also Im(α), R1 and R0. 11. Denoting Z := {z ∈ R : zr = rz for each r ∈ R} the center of R, we have {r ∈ R : f · r = r f for each f ∈ R} = R0 ∩ R1 ∩ Z. Moreover, if we consider two polynomials f (Y), g(Y) ∈ R \ {0}, f = aYm + f0, g = bYn +g0, a, b ∈ R\{0}, m, n ∈ N, f0, g0 ∈ R, deg( f0) < m, deg(g0) < n, we have f · g = aαm(b)Ym+n + h(Y), deg(h) < m + n; since α is an endomorphism, b , 0 =⇒ α(b) , 0 =⇒ αm(b) , 0 and since R is a domain it holds αm(b) , 0 , a =⇒ aαm(b) , 0 =⇒ f · g , 0. As a consequence 12. R is a domain. Definition 1. R with the ring structure described above is called an Ore extension and is denoted R[Y; α, δ]. Remark 2 (Ore). In an Ore extension R[Y; α, δ], denoting S = hα, δi the free semigroup over the alphabet {α, δ} and, for each d ∈ N and i ∈ N, 0 ≤ i ≤ d, Sd,i the set of the (cid:16)d i(cid:17) words in S of length d in which occur i instances of α and d − i instances of δ in an arbitrary order, we have Yd · r = d Xi=0 Xτ∈Sd,i τ(r)Yi 4 for each d ∈ N; for instance Y3 · r = α3(r)Y3 + δ3(r) + (cid:16)α2δ(r) + αδα(r) + δα2(r)(cid:17) Y2 + (cid:16)αδ2(r) + δαδ(r) + δ2α(r)(cid:17) Y. i=0 aiYn−i and g(Y) = Pm i=0 biYm−i in R we have i Xa=0 i−a ba Xb=0 Xτ∈Sm−a,i−a−b In particular, for f (Y) = Pn n+m Xi=0 g(Y) f (Y) = ciYn+m−i with c0 = b0αm(a0) and ci = τ(ab). Remark 3 (Ore). Under the assumption that (cf. 3.) α is an automorphism, each poly- i=1 Yi ¯ai for proper values ¯ai ∈ R. In fact we have ax = xα−1(a) − δ(α−1(a)) from which we can deduce inductively i=1 aiYi ∈ R can be uniquely represented asPn nomialPn proper expressions axn = xnα−n(a) + n Xi=1 (−1)ixn−iσin(a). ⊓⊔ 1.2 Multivariate Ore Extensions Definition 4. An iterative Ore extension is a ring (whose multiplication we denote ⋆) defined as R := R[Y1; α1, δ1][Y2; α2, δ2] · · · [Yn; αn, δn] where, for each i > 1, αi is an endomorphism and δi an αi-derivation of the iterative Ore extension Ri−1 := R[Y1; α1, δ1] · · · [Yi−1; αi−1, δi−1]. A multivariate Ore extension (or: Ore algebra [12]; or: iterative Ore extension with commuting variables [40, 41]) is an iterative Ore extension which satisfies – α jδi = δiα j, for each i, j, i , j, – αiα j = α jαi, δiδ j = δ jδi for j > i, – α j(Yi) = Yi, δ j(Yi) = 0 for j > i. Lemma 5 (Pesch). In an iterative Ore extension, for each i < j it holds Y j ⋆ Yi = YiY j ⇐⇒ α j(Yi) = Yi, δ j(Yi) = 0. Proof. For each i < j, we have Y j ⋆ Yi = α j(Yi)Y j + δ j(Yi). ⊓⊔ Lemma 6 (Pesch). An iterative Ore extension is a multivariate Ore extension iff Y j ⋆ Yi = YiY j for each i < j. ⊓⊔ 5 Proof. In fact, using Lemma 5 for each r ∈ R, we have Y j ⋆ Yi ⋆ r = Y j ⋆ (αi(r)Yi + δi(r)) = α j (αi(r)Yi + δi(r)) Y j + δ j (αi(r)Yi + δi(r)) = α jαi(r)YiY j + α jδi(r)Y j + δ j (αi(r)Yi) + δ jδi(r) = α jαi(r)YiY j + α jδi(r)Y j + δ jαi(r)Yi + δ jδi(r) and (by symmetry) YiY j ⋆ r = Yi ⋆ (α j(r)Y j + δi(r)) = αiα j(r)YiY j + δiα j(r)Y j + αiδ j(r)Yi + δiδ j(r). ⊓⊔ Thus the R-module structure of a multivariate Ore extension can be identified with that of the polynomial ring R[Y1, . . . , Yn] over its natural R-basis T := {Ya1 : (a1, . . . , an) ∈ Nn}, R (cid:27) R[T ] = SpanR{T}. We can therefore denote αYi := αi, δYi := δi for each i and, iteratively, 1 · · · Yan n ατYi := αταi, δτYi := δτδi, for each τ ∈ T . τ = Yd1 Remark that a multivariate Ore extension is not an algebra; in fact, if we define, for 1 · · · Ydn n and t = Ye1 1 · · · Yen n such that τ t t τ! := e1 d1! · · · en dn!, we have t τ!δ t τ t ⋆ r = Xτ∈Tτt ατ(r)τ, for each t ∈ T and r ∈ R. We can define, for each t ∈ T , a map θt : R → R, t τ!δ t τ ατ(r)τ, θt(r) = Xτ∈Tτt,τ,t Such maps αt and θt satisfy properties analogous of those of Ore's conjugate and so that t ⋆ r = αt(r)t + θt(r) for each t ∈ T and each r ∈ R. derivative: Lemma 7. With the present notation, for each t ∈ T , we have 1. for each r ∈ R, αt(r) = 0 =⇒ r = 0, 2. αt is a ring endomorphism; 6 3. the following conditions are equivalent: (a) for each d ∈ R \ {0} exists c ∈ R \ {0} : Y ⋆ c = dY + θt(c), αt(c) = d; (b) αt is a ring automorphism; 4. θt is an αt-derivation of R; 5. if R is a skew field, then each r ∈ R \ {0} satisfies αt(r−1) = (αt(r))−1 , θt(r−1) = − (αt(r))−1 θt(r)r−1; 6. Im(αt) ⊂ R is a subring, which is an isomorphism copy of R. We further have 7. if each αi is an automorphism, also each αt, t ∈ T , is such. ⊓⊔ 1.3 Associated graded Ore Extension Definition 8. A multivariate Ore extension R[Y1; α1, δ1][Y2; α2, δ2] · · · [Yn; αn, δn] where each δi is zero, will be called a graded Ore extension (or: Ore extension with zero derivations [40, 41]) and will be denoted R := R[Y1; α1][Y2; α2] · · · [Yn; αn]. Lemma 9. In a multivariate graded Ore extension, – since it is an Ore algebra, the αs commute, ⊓⊔ – and t ⋆ r = αt(r)th=: M(t ⋆ r)i for each t ∈ T and r ∈ R. Remark 10. Note that, since multivariate Ore extensions coincide, as left R-modules, with the classical polynomial rings R[Y1, . . . , Yn] and so have the same R-basis, namely T , they can share with the polynomial rings their standard T -valuation [50, 29, 2, 32] [30, §24.4,24.6]. This justifies the definition below. Definition 11. Given an Ore extension R := R[Y1; α1, δ1][Y2; α2, δ2] · · · [Yn; αn, δn] the corresponding graded Ore extension G(R) := R[Y1; α1][Y2; α2] · · · [Yn; αn] is called its associated graded Ore extension. ⊓⊔ Example 12. 1. The first non obvious example of Ore extension was proposed in 1948 by D.Ta- mari [51] in connection with the notion of "order of irregularity" introduced by Ore in [37]; it consists of the graded Ore extension. R := R[Y; α], R = Q[x] where α : R → R : x 7→ x2. 7 2. Such construction was generalized by Weispfenning [52] which introduced the rings R := R[Y; α], R = Q[x] where α : R → R : x 7→ xe, ei ∈ N 3. and extended by Pesch [40] to his iterated Ore extensions with power substitution R := R[Y1; α1][Y2; α2] · · · [Yn; αn], R = Q[x] where αi : R → R : x 7→ xei , ei ∈ N. 4. An Ore extension where α is invertible is discussed in [27]: R := R[S ; α], R = Q[D1, D2, D3] where α : R → R : f (D1, D2, D3) 7→ f (D2 + 2D1, D3, −D1) whose inverse is α−1 : R → R : f (D1, D2, D3) 7→ f (−D3, D1 + 2D3, D2). ⊓⊔ Note that, while as R-modules R and G(R) coincide both with the polynomial ring P = R[Y1, . . . , Yn], the three rings have, in general, different arithmetics; we will denote ⋆ the multiplication of R and ∗ those of G(R). Example 13. The ring of Example 12, 1. R := R[Y; α], R = Q[x] where α : R → R : x 7→ x2 is an Ore extension which is graded. The map δ : Q[x] → Q[x] : xi 7→ 2i−1 Xh=i xh is an α-derivation since = xh δ(xi · x j) = xh xh + 2i+2 j−1 Xh=i+ j 2i+2 j−1 Xh=2i+ j 2 j−1 Xh= j = x2i 2i+ j−1 Xh=i+ j 2i−1 Xh=i = α(xi)δ(x j) + δ(xi)x j; xh + x j xh thus S := R[Y; α, δ] is an Ore extension of which R is the associated graded Ore exten- sion. ⊓⊔ 8 Example 14. Since in Buchberger-Zacharias Theory, from an algorithmic point of view, the interest points are associated graded rings and thus the role of derivates is irrelevant, we illustrate the results for the Ore extensions with the zero-derivatives R := R[Y1; α1][Y2; α2] · · · [Yn; αn], R = Z[x], with αi(x) := cixei , ci ∈ Z \ {0}, ei ∈ N \ {0}. If we denote γ the map γ : N × N \ {0} → N, (a, e) 7→ ei = a−1 1 − ea Xi=0 1 − e i ∗ xb = cbγ(a,ei) i xea i bYa i . where the last equality holds for e , 1, we have Ya Note that γ(b, e) + ebγ(a, e) = b−1 Xi=0 Since α j(αi(x)) = ciα j(xei) = cicei R is a graded Ore extension if and only if ei + a−1 Xi=0 eb+i = a+b−1 Xi=0 ei = γ(a + b, e). (1-a) j xeie j and αi(α j(x)) = c jαi(xe j) = c jce j i xeie j, then cicei j xeie j = α j(αi(x)) = αi(α j(x)) = c jαi(xe j) = c jce j i xeie j id est cei−1 j = ce j−1 i . (1-b) We thus have(cid:16)n the indices as 2(cid:17) relations among the n coefficients ci. In particular we need to partition {1, . . . , n} = E ⊔ O ⊔ S , E = {i : 2 ei}, O = {i : 2 ∤ ei > 1}, S = {i : ei = 1}. If I := O ⊔ E = ∅ then each such equations are the trivial equality 1 = 1 and thus all ci are free. The situation is completely different when I := O ⊔ E , ∅; in fact, – for i ∈ S necessarily ci = ±1; – if a prime p divides at least a c j, j ∈ I, then it divides each ci, i ∈ I. As regards the sign of ci we can say that – if E , ∅ then – ci is positive for each i ∈ S ∪ O, – the sign of ci, i ∈ E, is undetermined but all the ci, i ∈ E, have the same sign. – if E = ∅ then the sign of ci, i ∈ S ∪ O is undetermined. For instance 9 – for e1 = e4 = 1, e2 = 5, e3 = 3 we have S = {1, 4}, O = {2, 3}, E = ∅ and 1 = c0 c4 1 = c0 3, c0 whence c1 = ±1, c4 = ±1, c2 = ±c2 3; 1 = c0 2, c2 4, c2 2 = c4 3, c0 2 = c4 4, c0 3 = c2 4; – for e1 = e4 = 1, e2 = 2, e3 = 3 we have S = {1, 4}, O = {3}, E = {2}, and c1 = c0 2, c2 1 = c0 1 = c0 4, c2 2 = c3, c0 2 = c4, c0 3 = c2 4; whence c1 = c4 = 1, c3 = c2 3, c0 2 > 0; – for e1 = 1, e2 = 2, e3 = 3, S = {1} E = {2}, O = {3}. Suppose c2 = 6, so both 2 = c3 we get c1 = 1 and the primes 2 and 3 divide c2. From c1 = c0 c3 = 36. We notice that 2 c3 and 3 c3, but neither 2 nor 3 divide c1; – for e1 = e4 = 1, e2 = 4, e3 = 8 we have S = {1, 4}, E = {2, 3}, O = ∅ and 1 = c0 2, c2 3, c2 2 = c3 whence c1 = c4 = 1, c2 = χ3, c3 = χ7, c2c3 > 0. c3 1 = c0 1 = c0 1 = c0 2, c7 3, c0 4, c7 3, c0 2 = c3 4, c0 3 = c7 4. As regards the values ci, 1 ≤ i ≤ n, setting ρ := we have Xj=1 n (e j − 1), χ := ρvt n (e j − 1) = Xj∈I c j = χe j−1 for each j ∈ {1, ...., n}. Yj=1 c j, (1-c) In fact, since if a prime p divides at least a c j, j ∈ I, then it divides each ci, i ∈ I, we h where p1, · · · , ph are the prime factors can express each ci, i ∈ I, as ci = pai,1 · · · pai,h of the squarefree associate √χ = p1 · · · ph of χ. 1 We have cie j−1 = c jei−1 =⇒ pai(e j−1) = pa j(ei−1) =⇒ ai(e j − 1) = a j(ei − 1) whence ai = a j ⇐⇒ ei = e j and ai > a j ⇐⇒ e j < ei. Thus the cis with minimal ei minimalize also all ai,l, 1 ≤ l ≤ h. We moreover have a j,l = ThereforeQn ai,i(e j−1) (ei−1) , 1 ≤ l ≤ h. l=1 pa j,l j=1 c j = Q j∈I c j = Qh Yj=1 c j = χ := ρvt n l = Qh Yl=1 ai,l ei−1 l p h = a j,l e j−1 l p h Yl=1 ai,l Pn l j=1(e j−1) ei−1 l=1 p ai,lρ ei−1 l whence l=1 p = Qh and (1-c). The formula (1-c) allows to reformulate (1-b) as c jei−1 = cie j−1 = χ(ei−1)(e j−1). (1-d) 10 Note that we have (eai i −1)(e χ a j j −1) = χ(ei−1)γ(ai,ei)(e j−1)γ(a j,e j) = ci(e j−1)γ(ai ,ei)γ(a j,e j) = ci(e = c j(ei−1)γ(ai,ei)γ(a j ,e j) = c j(eai a j j −1)γ(ai,ei) i −1)γ(a j ,e j) and ciγ(ai,ei)c jeai i γ(a j,e j) = cie a j j γ(ai,ei)c jγ(a j,e j) = χ eai i e a j j −1 . (1-e) To avoid cumbersome and useless case-to-case studies, let us simply assume ci > 0 for each i; under this restricted assumption, a series of inductive arguments allow to deduce Yi ∗ xα = cα i xαei Yi j i Ya j Yai i (cid:16)Ya j j ∗ xb0 = cb0γ(a j,e j) j ∗ xb0(cid:17) = Yai Yai i ∗(cid:18)cb0γ(a j,e j) j a j xb0e j (cid:19) = cb0γ(a j,e j) j Ya j = cb0γ(a j,e j) j j a j (cid:18)Yai i ∗ xb0e c b0e i a j j γ(ai,ei) j j (cid:19) Ya j xb0e = cb0γ(a j,e j) j b0e i c a j j γ(ai,ei) xb0e 11 a j j eai j i ! Ya j i Yai i Ya j j . a j j eai i Yai Yi ∗ xα = (Yi ∗ xα−1)x = cα−1 i j ∗ xb0 = cb0γ(a j,e j) Ya j x(α−1)ei (Yi ∗ x) = cα−1 i j Ya j j xb0e a j j x(α−1)ei cixeiYi = cα i xαei Yi. j ∗ xb0 = Y j(cid:16)Ya j−1 Ya j j ∗ xb0(cid:17) = Y j ∗(cid:18)cb0γ(a j−1,e j) j j a j−1 j a j−1 (cid:19) j Ya j−1 (cid:19) Ya j−1 (cid:19)e j! Ya j a j−1 j j j xb0e (cid:18)Y j ∗ xb0e x(cid:18)b0e c j Ya j xb0e j . b0e j a j−1 j a j = cb0γ(a j−1,e j) = cb0γ(a j−1,e j) j j = cb0γ(a j,e j) j cb0γ(a j ,e j) j b0e i c a j j γ(ai,ei) b0(e = χ a j j eai i −1) Substituing c j = χe j−1 and ci = χei−1 we get cb0γ(a j,e j) j b0e i c a j j γ(ai,ei) = χ(e j−1)b0γ(a j ,e j)χ (ei−1)b0e a j j γ(ai,ei) a j j a j (e j−1)b0 = χ = χ−b0(cid:16)1−e = χ a j 1−e j +(ei−1)b0e 1−e j a j j (1−eai i ) j (cid:17)−b0e a j a j a j j eai i −b0e j −b0+b0e j a j j eai i −1) = χ b0(e b0e . a j j γ(ai,ei) xb0e b0e i c a j j eai i Yai i Ya j j = χ b0(e a j j eai i −1)xb0e a j j eai i Yai i Ya j j (2-a) (2-b) (2-c) ai 1−e i 1−ei (2-d) (axa0Ya1 1 · · · Yan n ) ∗ (bxb0Yb1 1 · · · Ybn n ) = abχb0((Qn i=1 eai i )−1)xa0+b0Qn i=1 eai i Ya1+b1 1 · · · Yan +bn (2-e) n n i=2 eai (axa0Ya1 = axa0Ya1 = axa0Ya1 = abχb0((Qn = abχb0((Qn = abχb0((Qn = abχb0((Qn n ) ∗ (bxb0Yb1 1 · · · Ybn 1 · · · Yan n ) n ∗ bxb0)Yb1 1 (Ya2 1 · · · Ybn 2 · · · Yan i=2 eai i=2 eai i )−1)xb0Qn 1 (cid:16)bχb0((Qn 1 Ya2 +b2 i (cid:17) Yb1 1 ∗ xb0Qn i (cid:17) Yb1 i=2 eai xb0(Qn 1−e1 xb0Qn · · · Yan +bn i )−1)xa0 (cid:16)Ya1 i )−1)xa0 (cid:18)cb0Qn i )−1)xa0 χ i )−1)xa0+b0Qn Note that associativity is verified by (e1−1)b0Qn i=1 eai i Ya1+b1 i γ(a1,e1) i=2 eai i=2 eai i=2 eai i=2 eai i=1 eai a1 1−e 1 1 1 2 n i . 1 Ya2 +b2 2 i )ea1 i=2 eai n n · · · Yan+bn · · · Yan+bn 1 (cid:19) Ya1+b1 i ! Ya1+b1 1 1 i=1 eai Ya2 +b2 2 · · · Yan+bn n · · · Yan +bn n h(cid:16)axa0Ya1 = habχb0((Qn = abdχb0((Qn 1 · · · Yan i=1 eai n (cid:17) ∗(cid:16)bxb0Yb1 i )−1)xa0+b0Qn i=1 eai i )−1)+d0(cid:16)(cid:16)Qn 1 · · · Ybn i=1 eai i Ya1 +b1 (cid:17)−1(cid:17)xa0 +b0Qn n (cid:17)i ∗(cid:16)dxd0Yd1 · · · Yan+bn n i=1 eai 1 · · · Ydn n (cid:17) i ∗(cid:16)dxd0Yd1 1 · · · Ydn n (cid:17) i +d0Qn Ya1 +b1+d1 1 i=1 eai+bi i=1 eai+bi 1 i i and 1 · · · Yan 1 · · · Yan i=1 ebi (cid:16)axa0Ya1 = (cid:16)axa0Ya1 = abdχd0(cid:16)(cid:16)Qn b0((Qn = abdχ n (cid:17) ∗h(cid:16)bxb0Yb1 n (cid:17) ∗(cid:20)bdχd0(cid:16)(cid:16)Qn i (cid:17)−1(cid:17)+(cid:16)b0+d0(cid:16)Qn i )−1)+d0(cid:16)(cid:16)Qn i=1 eai 1 · · · Ybn i=1 ebi n (cid:17) ∗(cid:16)dxd0 Yd1 i (cid:17)−1(cid:17)xb0+d0Qn i (cid:17)(cid:17)((Qn (cid:17)−1(cid:17)xa0 +b0Qn i=1 eai i=1 eai i=1 eai+bi i=1 ebi i i )−1)xa0+(cid:16)b0+d0Qn i=1 eai+bi i i +d0Qn 1 · · · Ydn n (cid:17)i i=1 ebi i Yb1+d1 · · · Ybn+dn i=1 eai i=1 ebi i (cid:17)Qn (cid:21) 1 n i Ya1 +b1+d1 1 · · · Yan+bn+dn n n Yai+bi+di i Yi=1 · · · Yan+bn+dn n 2 Buchberger Theory ⊓⊔ 2.1 Term ordering For each m ∈ N, the free R-module Rm – the canonical basis of which will be denoted {e1, . . . , em} – is an (R, R)-bimodule with basis the set of the terms f = s Xi=1 c( f , ti)ti : c( f , ti) ∈ R \ {0}, ti ∈ T (m), t1 > · · · > ts. 12 T (m) := {tei : t ∈ T , 1 ≤ i ≤ m}. If we impose on T (m) a total ordering <, then each f ∈ Rm has a unique represen- tation as an ordered linear combination of terms t ∈ T (m) with coefficients in R: The support of f is the set supp( f ) := {t c( f , t) , 0}. cofficient and M( f ) := c( f , t1)t1 its maximal monomial. W.r.t. < we denote T( f ) := t1 the maximal term of f , lc( f ) := c( f , t1) its leading If we denote, following [44, 45], M(Rm) := {ctei c ∈ R \ {0}, t ∈ T , 1 ≤ i ≤ m}, for each f ∈ Rm \ {0} the unique finite representation above can be reformulated f = Xτ∈supp( f ) mτ, mτ = c( f , τ)τ as a sum of elements of the monomial set M(Rm). While a multivariate Ore extension does not satisfiy commutativity between terms and coefficients, it however satisfies t ⋆ r = rt for each r ∈ R \ {0}, t ∈ T , M(t ⋆ r) = αt(r)t, for each r ∈ R \ {0}, t ∈ T (m); (3) moreover, while R is not a monoid ring under the multiplication ⋆, so that in particular we cannot claim τ ⋆ ω ∈ T for τ, ω ∈ T , however τ ⋆ ω satisfies T(τ ⋆ ω) = τ ◦ ω (4) where we have denoted ◦ the (commutative) multiplication of T ; similarly, for n ∈ M(Rm) and ml, mr ∈ M(R) = {ct : c ∈ R\{0}, t ∈ T} we have M(ml⋆m⋆mr) = ml∗m∗mr. In conclusion w.r.t. each term ordering ≺ on T and each ≺-compatible term order- ing < on T (m), id est any well-ordering on T (m) which satisfies ω1 ≺ ω2 =⇒ ω1t < ω2t, tω1 < tω2 for each t ∈ T (m), ω1, ω2 ∈ T , it holds, for each l, r ∈ R and f ∈ R(m), T(l ⋆ f ⋆ r) = T(l) ◦ T( f ) ◦ T(r) and M(l ⋆ f ⋆ r) = M(M(l) ⋆ M( f ) ⋆ M(r)) = M(l) ∗ M( f ) ∗ M(r). As a consequence we trivially have (5) (6) Corollary 15. If ≺ is a term ordering on T and < is a ≺-compatible term ordering on T (m), then, for each l, r ∈ R and f ∈ R(m), 1. M(l ⋆ f ) = M(M(l) ⋆ M( f )) = M(l) ∗ M( f ); 2. M( f ⋆ r) = M(M( f ) ⋆ M(r)) = M( f ) ∗ M(r); 3. M(l ⋆ f ⋆ r) = M(M(l) ⋆ M( f ) ⋆ M(r)) = M(l) ∗ M( f ) ∗ M(r). 4. T(l ⋆ f ) = T(l) ◦ T( f ); 5. T( f ⋆ r) = T( f ) ◦ T(r); 6. T(l ⋆ f ⋆ r) = T(l) ◦ T( f ) ◦ T(r). 13 2.2 Grobner Bases In function of a term ordering < on T (m) which is compatible with a term ordering on T which, with a slight abuse of notation, we still denote <, we denote, for any set F ⊂ Rm, IL(F), IR(F), I2(F) the left (resp. right, bilateral) module generated by F, and – T{F} := {T( f ) : f ∈ F} ⊂ T (m); – M{F} := {M( f ) : f ∈ F} ⊂ M(Rm). – TL(F) := IL(T{F}) = {T(λ ⋆ f ) : λ ∈ T , f ∈ F} = {λ ◦ T( f ) : λ ∈ T , f ∈ F} ⊂ T (m); – ML(F) := {M(aλ ⋆ f ) : a ∈ R \ {0}, λ ∈ T , f ∈ F} = {m ∗ M( f ) : m ∈ M(R), f ∈ F} ⊂ M(Rm); – TR(F) := IR(T{F}) = {T( f ⋆ ρ) : ρ ∈ T , f ∈ F} = {T( f ) ◦ ρ : ρ ∈ T , f ∈ F} ⊂ T (m); – MR(F) := {M( f ⋆ bρ) : b ∈ R \ {0}, ρ ∈ T , f ∈ F} = {M( f ) ∗ n : n ∈ M(R), f ∈ F} ⊂ M(Rm); – T2(F) := I2(T{F}) = {T(λ ⋆ f ⋆ ρ) : λ, ρ ∈ T , f ∈ F} = {λ ◦ T( f ) ◦ ρ : λ, ρ ∈ T , f ∈ F} ⊂ T (m); – M2(F) := {M(aλ ⋆ f ⋆ bρ) : a, b ∈ R \ {0}, λ, ρ ∈ T , f ∈ F} = {m ∗ M( f ) ∗ n : m, n ∈ M(R), f ∈ F} ⊂ M(Rm). Following an intutition by Weispfenning [52] we further denote – IW(F) the restricted module generated by F, IW(F) := SpanR(a f ⋆ ρ : a ∈ R \ {0}, ρ ∈ T , f ∈ F), – TW(F) := TR(F), – MW(F) := {M(a f ⋆ ρ) : a ∈ R\{0}, ρ ∈ T , f ∈ F} = {aM( f )∗ ρ : a ∈ R\{0}, ρ ∈ T , f ∈ F} ⊂ M(Rm). If R is a skew field, for each set F ⊂ Rm we have ML(F) = M{IL(M{F})} = IL(M{F}) ∩ M(Rm), MR(F) = M{IR(M{F})} = IR(M{F}) ∩ M(Rm), = I2(M{F}) ∩ M(Rm), M2(F) = M{I2(M{F})} MW(F) = M{IW(M{F})} = IW(M{F}) ∩ M(Rm). (7) Notation 16. From now on, in order to avoid cumbersome notation and boring repeti- tions, we will drop the subscripts when it will be clear of which kind of module (left, right, bilateral, restricted) we are discussing. As a consequence, the four statements of (7) will be summarized as M(F) = M{I(M{F})} = I(M{F}) ∩ M(Rm). 14 Similarly, we formulate a (left,right,bilateral,restricted) definition simply for the bilateral case leaving to the reader the task to remove the irrelevant factors. For instance condition (ii) below is stated for the bilateral case; it would be refor- mulated: left case for each f ∈ I(F) there are g ∈ F, a ∈ R\{0}, λ ∈ T such that M( f ) = aλ∗M(g) = M(aλ ⋆ g), right case for each f ∈ I(F) there are g ∈ F, b ∈ R\{0}, ρ ∈ T such that M( f ) = M(g)∗bρ = M(g ⋆ bρ), restricted case for each f ∈ I(F) there are g ∈ F, a ∈ R\{0}, ρ ∈ T such that M( f ) = aM(g)∗ρ = ⊓⊔ Thus, if R is a skew field, the following conditions are equivalent and can be natu- M(ag ⋆ ρ), rally chosen as definition of Grobner bases: 1. M(I(F)) = M{I(F)} = M{I(M{F})} = I(M{F}) ∩ M(Rm), 2. for each f ∈ I(F) there is g ∈ F such that M(g) M( f ). But in general between these statements there is just the implication (2) =⇒ (1). Thus [39], there are two alternative natural definitions for the concept of Grobner bases: – a stronger one which satisfies the following equivalent conditions: (i). for each f ∈ I(F) there is g ∈ F such that M(g) M( f ), (ii). for each f ∈ I(F) there are g ∈ F, a, b ∈ R \ {0}, λ, ρ ∈ T such that (iii). M(I(F)) = M{I(F)} = M(F); M( f ) = aλ ∗ M(g) ∗ bρ = M(aλ ⋆ g ⋆ bρ), – and a weaker one which satisfies the following equivalent conditions: denoting τi := T(gi), one has and, equivalently, (iv). for each f ∈ I(F) there are gi ∈ F, ai, bi ∈ R \ {0}, λi, ρi ∈ T for which, – T( f ) = λi ◦ T(gi) ◦ ρi for each i, and lc( f ) = Pi aiαλi(lc(gi))αλiτi(bi) – M( f ) = Pi aiλi ∗ M(gi) ∗ biρi = Pi M(aiλi ⋆ gi ⋆ biρi); (v). M(I(F)) = M{I(F)} = M{I(M{F})} = I(M{F}) ∩ M(Rm); if moreover R is a skew field M(F) = M{I(M{F})} so that conditions (i-v) above are all equivalent and are also equivalent to (vi). T( f ) = λ ◦ T(g) ◦ ρ for some g ∈ F, λ, ρ ∈ T . 15 Example 17. Let us now specialize the ring of Example 14 to the case n = 3, e1 = 2, e2 = 3, e3 = 4, χ = 5, c1 = 5, c2 = 52, c3 = 53 and remark that 1 Ya2 2 Ya3 3 ∗ bxb0Yb1 axa0Ya1 As a consequence, for each (b0, b1, b2, b3), ( j0, j1, j2, j3) ∈ N4 3 = ab5b0(2a1 3a2 4a3−1)xa0+b02a1 3a2 4a3 Ya1+b1 1 Yb2 2 Yb3 1 Ya2 +b2 2 Ya3 +b3 3 . jx j0Y j1 1 Y j2 2 Y j3 3 ∈ IL(bxb0Yb1 1 Yb2 2 Yb3 3 ) if and only if a1 := j1 − b1 ≥ 0, a2 := j2 − b2 ≥ 0, a3 := j3 − b3 ≥ 0, a0 := j0 − b02a13a24a3 ≥ 0 (8) and b5b0(2a1 3a2 4a3−1) j. Z[y] ⊂ R Note that if we set y := 5x then for each (b1, b2, b3), ( j1, j2, j3) ∈ N3 and b(y), j(y) ∈ 1 Yb2 3 ∈ IL(b(y)Yb1 if and only if, not only (8) but also b(y2a1 3a2 4a3 ) j(y). Definition 18. Let I ⊂ Rm be a (left, right, bilateral, restricted) module and G ⊂ I. 2 Yb3 3 ) j(y)Y j1 1 Y j2 2 Y j3 – G will be called – a (left, right, bilateral, restricted) weak Grobner basis (Grobner basis for short) of I if M{I} = M(I) = M{I(M{G})} = I(M{G}) ∩ M(Rm), id est if G satisfies conditions (iv-v) w.r.t. the module I = I(G); in particular M{G} generates the (left, right, bilateral, restricted) module M(I) ⊂ Rm; – a (left, right, bilateral, restricted) strong Grobner basis of I if for each f ∈ I there is g ∈ G such that M(g) M( f ), id est if G satisfies conditions (i-iii) w.r.t. the module I = I(G). – We say that f ∈ Rm \ {0} has – a left Grobner representation in terms of G if it can be written as f = – a left (weak) Grobner representation in terms of G if it can be written as i=1 aiλi ⋆ gi, with ai ∈ R \ {0}, λi ∈ T , gi ∈ G and T(λi ⋆ gi) ≤ – a left (strong) Grobner representation in terms of G if it can be written as Pu i=1 li ⋆ gi, with li ∈ R, gi ∈ G and T(li) ◦ T(gi) ≤ T( f ) for each i; f = Pµ f = Pµ i=1 aiλi ⋆ gi, with ai ∈ R \ {0}, λi ∈ T , gi ∈ G and T( f ) for each i; T( f ) = λ1 ◦ T(g1) > λi ◦ T(gi) for each i; 16 – a right Grobner representation in terms of G if it can be written as f = – a right (weak) Grobner representation in terms of G if it can be written as i=1 gi ⋆ biρi, with bi ∈ R \ {0}, ρi ∈ T , gi ∈ G and T(gi ⋆ ρi) ≤ T( f ) – a right (strong) Grobner representation in terms of G if it can be written as Pu i=1 gi ⋆ ri, with ri ∈ R, gi ∈ G and T(gi) ◦ T(ri) ≤ T( f ) for each i; f = Pµ f = Pµ i=1 gi ⋆ biρi, with bi ∈ R \ {0}, ρi ∈ T , gi ∈ G and for each i; T( f ) = T(g1) ◦ ρ1 > T(gi) ◦ ρi for each i; – a bilateral (weak) Grobner representation in terms of G if it can be written i=1 aiλi ⋆ gi ⋆ biρi, with ai, bi ∈ R \ {0}, λi, ρi ∈ T , gi ∈ G and – a bilateral (strong) Grobner representation in terms of G if it can be written i=1 aiλi ⋆ gi ⋆ biρi, with ai, bi ∈ R \ {0}, λi, ρi ∈ T , gi ∈ G and – a restricted (weak) Grobner representation in terms of G if it can be written i=1 aigi ⋆ ρi, with ai ∈ R\{0}, ρi ∈ T , gi ∈ G and T(gi ⋆ ρi) ≤ T( f ) – a restricted (strong) Grobner representation in terms of G if it can be i=1 aigi ⋆ ρi, with ai ∈ R \ {0}, ρi ∈ T , gi ∈ G and as f = Pµ T(λi ⋆ gi ⋆ ρi) ≤ T( f ) for each i; as f = Pµ T( f ) = λ1 ◦ T(g1) ◦ ρ1 > λi ◦ T(gi) ◦ ρi for each i. as f = Pµ written as f = Pµ T( f ) = T(g1) ◦ ρ1 > T(gi) ◦ ρi for each i. for each i; – For f ∈ Rm \ {0}, F ⊂ Rm, an element h := NF( f , F) ∈ Rm is called a – (left, right, bilateral, restricted) (weak) normal form of f w.r.t. F, if f − h ∈ I(F) has a weak Grobner representation in terms of F, and h , 0 =⇒ M(h) < M{I(M{F})}; – (left, right, bilateral, restricted) strong normal form of f w.r.t. F, if f − h ∈ I(F) has a strong Grobner representation in terms of F, and h , 0 =⇒ M( f ) < M(F). ⊓⊔ Proposition 19. (cf. [44, 45]) For any set F ⊂ Rm \ {0}, among the following condi- tions: 1. f ∈ I(F) ⇐⇒ it has a (left, right, bilateral, restricted) strong Grobner repre- sentation f = Pµ i=1 aiλi ⋆ gi ⋆ biρi in terms of F which further satisfies T( f ) = T(λ1 ⋆ g1 ⋆ ρ1) > · · · > T(λi ⋆ gi ⋆ ρi) > · · · ; 2. f ∈ I(F) ⇐⇒ it has a (left, right, bilateral, restricted) strong Grobner repre- sentation in terms of F; 3. F is a (left, right, bilateral, restricted) strong Grobner basis of I(F); 17 4. f ∈ I(F) ⇐⇒ it has a (left, right, bilateral, restricted) weak Grobner represen- tation in terms of F; 5. F is a (left, right, bilateral, restricted) Grobner basis of I(F); f ∈ I(F) ⇐⇒ it has a (left, right) Grobner representation in terms of F; 6. 7. for each f ∈ Rm \ {0} and any (left, right, bilateral, restricted) strong normal form h of f w.r.t. F we have f ∈ I(F) ⇐⇒ h = 0; 8. for each f ∈ Rm \{0} and any (left, right, bilateral, restricted) weak normal form h of f w.r.t. F we have f ∈ I(F) ⇐⇒ h = 0; there are the implications (1) ⇔ (2) ⇒ (4) ⇔ (6) (7) ⇐ (3) ⇒ (5) ⇒ (8) m v t m If R is a skew field we have also the implication (4) =⇒ (2) and as a consequence also (5) =⇒ (3). Proof. The implications (1) =⇒ (2) =⇒ (4) ⇐⇒ (6), (3) =⇒ (5), (2) =⇒ (3) and (4) =⇒ (5) are trivial. Ad (3) =⇒ (1): for each f ∈ I2(F) by assumption there are elements g ∈ F, m = aλ, n = bρ ∈ M(R) such that M( f ) = M(m ⋆ g ⋆ n). Thus T( f ) = T(m ⋆ g ⋆ n) = λ ◦ T(g) ◦ ρ and, denoting f1 := f − m ⋆ g ⋆ n, we have T( f1) < T( f ) so the claim follows by induction, since < is a well ordering. Ad (5) =⇒ (4): similarly, for each f ∈ I2(F) by assumption there are elements gi ∈ F, T(gi) := τieli, mi = aiλi, ni = biρi ∈ M(R) such that – T( f ) = T(λi ⋆ gi ⋆ ρi) = λi ◦ τi ◦ ρieli for each i, – lc( f ) = Pi aiαλi(lc(gi))αλiτi(bi). induction, since T( f1) < T( f ) and < is a well ordering. It is then sufficient to denote f1 := f −Pi mi ⋆ gi ⋆ ni in order to deduce the claim by Ad (4) =⇒ (2): let f ∈ I2(F) \ {0}; (4) implies the existence of g ∈ F, λ, ρ ∈ T , such that T( f ) = λ ◦ T(g) ◦ ρ. Then setting f1 := f − lc( f )(cid:16)αλ (lc(g))(cid:17)−1 λ ⋆ g ⋆ ρ we deduce the claim by induction, since T( f1) < T( f ) and < is a well ordering. Ad (3) =⇒ (7) and (5) =⇒ (8): either – h = 0 and f = f − h ∈ I(F) or – h , 0, M(h) < M(I(F)), h < I(F) and f < I(F). Ad (7) =⇒ (2) and (8) =⇒ (4): for each f ∈ I(F), its normal form is h = 0 and ⊓⊔ f = f − h has a strong (resp.: weak) Grobner representation in terms of F. 18 Proposition 20. (Compare [30, Proposition 22.2.10]) If F is a (weak, strong) Grobner basis of I := I(F), then the following holds: 1. Let g ∈ Rm be a (weak, strong) normal form of f w.r.t. F. If g , 0, then T(g) = min{T(h) : h − f ∈ I(F)}. 2. Let f , f ′ ∈ Rm \ I be such that f − f ′ ∈ I. Let g be a (weak, strong) normal form of f w.r.t. F and g′ be a (weak, strong) normal form of f ′ w.r.t. F. Then – T(g) = T(g′) =: τ and – lc(g) − lc(g′) ∈ Iτ := {lc( f ) : f ∈ I, T( f ) = τ} ∪ {0} ⊂ R. Proof. 1. Let h ∈ Rm be such that h − f ∈ I; then h − g ∈ I and M(h − g) ∈ M{I}. If T(g) > T(h) then M(h − g) = M(g) < M{I}, giving a contradiction. 2. The assumption implies that f − g′ ∈ I so that, by the previous result, T(g) ≤ T(g′). Symmetrically, f ′ − g ∈ I and T(g′) ≤ T(g). Therefore T(g) = T(g′) = τ; morevoer, either – T(g − g′) < τ and M(g) = M(g′) so that lc(g) = lc(g′) or – T(g− g′) = τ and M(g− g′) = M(g)− M(g′) = (cid:16)lc(g)− lc(g′)(cid:17)τ; thus, since g − g′ ∈ I, lc(g) − lc(g′) ∈ Iτ. ⊓⊔ 2.3 Canonical forms (skew field case) If R := K is a skew field, for any set F ⊂ Rm we denote N(F) the (left, right, bilateral, restricted) order module N(F) := T (m) \ T(F) and K[N(F)] the (left, right, bilateral, restricted) K-module K[N(F)] := SpanK(N(F)). Definition 21. For any (left, right, bilateral, restricted) module I ⊂ Rm, the order module N(I) := T (m) \ T{I} is called the escalier of I. We easily obtain the notion, the properties and the computational algorithm (Fig- ure 1) of (left, right, bilateral, restricted) canonical forms: Lemma 22. (cf. [30, Lemma 22.2.12]) Let I ⊂ Rm be a (left, right, bilateral, restricted) module. If R = K is a skew field and denoting A the (left, right,bilateral, restricted) module A := Rm/I it holds 1. Rm (cid:27) I ⊕ K[N(I)]; 2. A (cid:27) K[N(I)]; 19 Figure 1: Canonical Form Algorithms (g,Pµ i=1 ciλi ⋆ gi) := LeftCanonicalForm( f , G) where G is the left Grobner basis of the left module I ⊂ Rm, f ∈ Rm, g ∈ K[N(I)], ci ∈ K \ {0}, λi ∈ T , gi ∈ G, f − g = Pµ of G, T( f − g) = λ1 ◦ T(g1) > λ2 ◦ T(g2) > · · · > λµ ◦ T(gµ). i=1 ciλi ⋆ gi is a left strong Grobner representation in terms h := f , i := 0, g := 0, While h , 0 do j=1 c jλ j ⋆ g j + h, %% f = g +Pi %% T( f − g) ≥ T(h); %% i > 0 =⇒ T( f −g) = λ1◦T(g1) > λ2◦T(g2) > · · · > λi◦T(gi) > T(h); If T(h) ∈ TL(G) do Else Let λ ∈ T , γ ∈ G : λ ◦ T(γ) = T(h) i := i + 1, ci := lc(h)αλ (lc(γ))−1 , λi := λ, gi := γ, h := h − ciλigi. %% T(h) ∈ N(I) g := g + M(h), h := h − M(h) (g,Pµ i=1 gi ⋆ diρi) := RightCanonicalForm( f , G) where G is the right Grobner basis of the right module I ⊂ Rm, f ∈ Rm, g ∈ K[N(I)], di ∈ K \ {0}, ρi ∈ T , gi ∈ G, f − g = Pµ of G, T( f − g) = T(g1) ◦ ρ1 > T(g2) ◦ ρ2 > · · · > T(gµ) ◦ ρµ. i=1 gi ⋆ diρi is a right strong Grobner representation in terms h := f , i := 0, g := 0, While h , 0 do If T(h) ∈ TR(G) do Let ρ ∈ T , γ ∈ G : T(γ) ◦ ρ = T(h) i := i + 1, di := α−1 h := h − gi ⋆ diρi. g := g + M(h), h := h − M(h) Else T(γ)(cid:16)lc(h) lc(γ)−1(cid:17) , ρi := ρ, gi := γ, 20 Figure 1 (cont.): Canonical Form Algorithms (g,Pµ i=1 ciλi ⋆ gi ⋆ ρi) := BilateralCanonicalForm( f , G) where G is the bilateral Grobner basis of the bilateral module I ⊂ Rm, f ∈ Rm, g ∈ K[N(I)], ci ∈ K \ {0}, λi, ρi ∈ T , gi ∈ G, f − g = Pµ in terms of G, T( f − g) = λ1 ◦ T(g1) ◦ ρ1 > λ2 ◦ T(g2) ◦ ρ2 > · · · > λµ ◦ T(gµ) ◦ ρµ. i=1 ciλi ⋆ gi ⋆ ρi is a bilateral strong Grobner representation h := f , i := 0, g := 0, While h , 0 do If T(h) ∈ T2(G) do Let λ, ρ ∈ T , γ ∈ G : λ ◦ T(γ) ◦ ρ = T(h) i := i + 1, ci := lc(h)αλ (lc(γ))−1 , λi := λ, ρi := ρ, gi := γ, h := h − ciλi ⋆ gi ⋆ ρi. g := g + M(h), h := h − M(h) Else (g,Pµ i=1 cigi ⋆ ρi) := RestrictedCanonicalForm( f , G) where G is the restricted Grobner basis of the restricted module I ⊂ Rm, f ∈ Rm, g ∈ K[N(I)], ci ∈ K \ {0}, ρi ∈ T , gi ∈ G, f − g = Pµ terms of G, T( f − g) = T(g1) ◦ ρ1 > T(g2) ◦ ρ2 > · · · > T(gµ) ◦ ρµ. i=1 cigi ⋆ ρi is a restricted strong Grobner representation in h := f , i := 0, g := 0, While h , 0 do If T(h) ∈ TW(G) do Let ρ ∈ T , γ ∈ G : T(γ) ◦ ρ = T(h) i := i + 1, ci := lc(h) lc(γ)−1, ρi := ρ, gi := γ, h := h − cigi ⋆ ρi. g := g + M(h), h := h − M(h) Else 21 3. for each f ∈ Rm, there is a unique g := Can( f , I) = Xt∈N(I) γ( f , t, <)t ∈ K[N(I)] such that f − g ∈ I. Moreover: (a) Can( f1, I) = Can( f2, I) ⇐⇒ f1 − f2 ∈ I; (b) Can( f , I) = 0 ⇐⇒ f ∈ I. 4. For each f ∈ Rm, f − Can( f , I) has a (left, right, bilateral, restricted) strong Grobner representation in terms of any Grobner basis. Definition 23. (cf. [30, Definition 22.2.13]) For each f ∈ Rm the unique element g := Can( f , I) ∈ K[N(I)] such that f − g ∈ I will be called the (left, right, bilateral, restricted) canonical form of f w.r.t. I. ⊓⊔ Corollary 24. (cf. [30, Corollary 22.3.14]) If R = K is a skew field, there is a unique set G ⊂ I such that – T{G} is an irredundant basis of T(I); – for each g ∈ G, lc(g) = 1; – for each g ∈ G, g = T(g) − Can(T(g), I). G is called the (left, right, bilateral, restricted) reduced Grobner basis of I. ⊓⊔ Note that the algorithm described for right canonical forms is assuming that each αi is an automorphism; alternatively we can assume that R is given as a right R-module in which case the theory can be developped symmetrically. 3 Szekeres Theory Let I ⊂ Rm be a (left-bilateral) module; if we denote for each τ ∈ T (m), Iτ the additive group Iτ := {lc( f ) : f ∈ I, T( f ) = τ} ∪ {0} ⊂ R, I := {Iτ : τ ∈ T (m)} and, for each ideal a ⊂ R , Ta and La the sets Ta := {τ ∈ T (m) : Iτ ⊇ a} ⊂ T (m) and La := {τ ∈ T (m) : Iτ = a} ⊂ T (m), we have 1. for each τ ∈ T (m), Iτ ⊂ R is a left ideal; 22 Tb; Lb, La = Ta \ Sb%a 2. for each ideals a, b ⊂ R, a ⊂ b =⇒ Ta ⊃ Tb; 3. Ta = Fb⊇a 4. for terms τ, ω ∈ T (m), τ ω =⇒ Iτ ⊂ Iω; 5. for each ideal a ⊂ R, Ta ⊂ T (m) is a right semigroup module. If R is a skew field, the situation is quite trivial: for any ideal I we have I = {(0), R}, TR = LR = T(I), T(0) = T (m), L(0) = T (m) \ T(I). ⊓⊔ Szekeres notation is related with a pre-Buchberger construction of "canonical" ide- als for the case of polynomial rings R[Y1, . . . , Yn] over a PID R. In connection recall that [13, 14] a not necessarily commutative ring R is called a (left, right, bilateral) B´ezout ring if every finitely generated (left, right, bilateral) ideal is principal and is called a B´ezout domain if it is both a B´ezout ring and is a domain, and remark that, if R is a noetherian (left, bilateral) B´ezout ring, then for each τ ∈ T (m), there is a value cτ ∈ R satisfying Iτ = I(cτ). Definition 25. With the present notation, we call Szekeres ideal each ideal Iτ ⊂ R and Szekeres level each set La ⊂ T (m), Szekeres semigroup each semigroup Ta ⊂ T (m). Finally, if R is a noetherian left B´ezout ring we call Szekeres generator each value cτ ∈ R satisfying Iτ = IL(cτ). Note that if R is a noetherian B´ezout ring, we have, ω τ =⇒ cτ L αλ(cω) for each λ, ρ ∈ T s.t. τ = λ ◦ ω ◦ ρ. Proposition 26 (Szekeres). [50] Let R be a noetherian left B´ezout ring and I ⊂ Rm be a (left,bilateral) module. Denote T := nτ ∈ T (m) s.t. cτ < I(αλ(cω), ω ∈ T (m), λ, ρ ∈ T , τ = λ ◦ ω ◦ ρ)o ⊂ T (m) and fix, for each τ ∈ T, any element fτ ∈ I such that1 M( fτ) = cττ. Then the basis S w := { fτ s.t. τ ∈ T} is a left/bilateral weak Grobner basis of I. Proof. For each f ∈ I, denoting τ := T( f ) we have lc( f ) ∈ IL(cτ) and lc( f ) = dcτ for suitable d ∈ R \ {0}. Thus if τ ∈ T we have M( f ) = dM( fτ); if, instead, τ < T there are suitable di, ∈ R \ {0}, ωi ∈ T ⊂ T (m), λi, ρi ∈ T for which λi ◦ ωi ◦ ρi = τ and cτ = Pi diαλi(cωi) so that M( f ) = dcττ = d diαλi(cωi)λi ◦ ωi ◦ ρi Xi = Xi (ddiλi) ·(cid:0)cωi ωi(cid:1) · ρi = Xi (ddiλi) ∗ M( fωi ) ∗ ρi. 1Of course for the extreme case Iτ = (0) so that cτ = 0, we have fτ := 0. ⊓⊔ 23 Remark 27. Remark that in the case in which each endomorphism ατ, τ ∈ T (m), is an automorphism, we can consider also right modules I to which we can associate Iτ = {lc( f ) : f ∈ I, T( f ) = τ} ∪ {0} which are right ideals themselves; in fact if we represent f ∈ Rm as (see Remark 3) f = Pn i=1 Yi ¯ai and we denote τI the right ideal τI := {c ∈ R : τc ∈ M{I}} ∪ {0} ⊂ R then Iτ is the right ideal ατ(τI). However, in this setting, Szekeres Theory can be built more easily by considering the ideals τI obtained through the right representation of Remark 3 and adapting to them the results reported above. Remark that if an endomorphism ατ is not invertible, in general Iτ is not an ideal but just an additive group. Finally note that for restricted modules, one apply verbatim, the classical Szekeres theory and subistitute in the results above each instance of αλ(cω), τ = λ ◦ ω ◦ ρ with cω, τ = ω ◦ ρ ⊓⊔ Example 28. In the Ore extension R := R[Y; α], R = Z2[x] where α : R → R : x 7→ x2 we can consider, as a left module, the two-sided ideal I2(x) = IL{xYi : i ∈ N}; we thus have Iτ = I(x) ⊂ R for each τ ∈ {Yi, i ≥ 0}, so that, setting a := I(x) ⊂ R, it holds I = {a}, Ta = La = {Yi : i ∈ N}, and S w = {xYi : i ∈ N} is both a weak and a strong Grobner basis of IL(x). For the right ideal IR(xY) the sets Iτ are not ideals; we have, e.g. IYi = {xφ(xei)φ(x) ∈ k[x]}. 4 Zacharias canonical representation Following Zacharias approach to Buchberger Theory [53], if each module I ⊂ Rm has a groebnerian property, necessarily the same property must be satisfied at least by the modules I ⊂ Rm ⊂ Rm and thus such property in Rm can be used to device a procedure granting the same property in Rm. The most elementary application of Zacharias approach is the generalization of the property of canonical forms from the case in which R = K is a skew field to the general case: all we need is an effective notion of canonical forms for modules in R: Definition 29 (Zacharias). [53] A ring R is said to have canonical representatives if there is an algorithm which, given an element c ∈ Rm and a (left,bilateral, right) module J ⊂ Rm, computes a unique element Rep(c, J) such that – c − Rep(c, J) ∈ J, 24 – Rep(c, J) = 0 ⇐⇒ c ∈ J. The set Zach(Rm/J) := Rep(J) := {Rep(c, J) : c ∈ Rm} (cid:27) Rm/J is called the canonical Zacharias representation of the module Rm/J. Remark that, for each c, d ∈ Rm and each module J ⊂ Rm, we have c − d ∈ J ⇐⇒ Rep(c, J) = Rep(d, J). Using Szekeres notation for a (left, right, bilateral) module I ⊂ Rm we obtain – the partition T (m) = L(I) ⊔ R(I) ⊔ N(I) of T (m) where ⊓⊔ – N(I) := L(0) = {ω ∈ T (m) : Iω = (0)}, – L(I) := LR = {ω ∈ T (m) : Iω = R}, – R(I) := nω ∈ T (m) : Iω < {(0), R}o ; – the canonical Zacharias representation Zach(Rm/I) := Rep(I) := nRep(c, I) : c ∈ Rmo = Ma∈I Mτ∈La Rep(a)τ Rep(Iτ)τ Zach(R/Iτ)τ (cid:27) Rm/I = Mτ∈T (m) = Mτ∈T (m) of the module Rm/I. If R has canonical representatives and there is an algorithm (cf. Definition 42 (c), (e)) which, given an element c ∈ Rm and a (left, right, bilateral) module J ⊂ Rm com- putes the unique canonical representative Rep(c, J), an easy adaptation of Figure 1 allows to extend, from the field coefficients case to the Zacharias ring [53, 31] coeffi- cients case, the notion of canonical forms, the algorithm (Figure 2) for computing them and their characterizing properties: Lemma 30. If R has canonical representatives, also R has canonical representatives. With the present notation and denoting A the (left, right, bilateral) module A := Rm/I it holds: 1. Rm (cid:27) I ⊕ Rep(I); 2. A (cid:27) Rep(I); 3. for each f ∈ Rm, there is a unique (left, right, bilateral) canonical form of f γ( f , τ, I, <) ∈ Rep(Iτ), γ( f , τ, I, <)τ ∈ Rep(I), g := Can( f , I) = Xa∈I Xτ∈La such that 25 – f − g ∈ I, – γ( f , τ, I, <) = Rep(γ( f , τ, I, <), Iτ) ∈ Rep(Iτ), for each τ ∈ T (m). Moreover: (a) Can( f1, I) = Can( f2, I) ⇐⇒ f1 − f2 ∈ I; (b) Can( f , I) = 0 ⇐⇒ f ∈ I; 4. for each f ∈ Rm, f −Can( f , I) has a (left, right, bilateral) (weak, strong) Grobner ⊓⊔ representation in terms of any (weak, strong) Grobner basis. 5 Moller's Lifting Theorem 5.1 Left case The validity of Eqs. (5) and (6) allows to intoduce the groebnerian terminology and, as in the standard theory of commutative polynomial rings over a field [30, § 21.1-2] or a Zacharias ring [53], the ability of imposing a T (m)-valuation on modules over R and its associated graded Ore extension S := G(R) (see Remark 10). The only twist w.r.t. the classical theory is that there the ring was coinciding with its associated graded ring; here they coincide as sets and as left R-modules, but as rings have two different multiplications. Consequently, denoting by ⋆ the one of R and by ∗ the one of S, given a finite basis F := {g1, . . . , gu} ⊂ Rm, gi = M(gi) − pi =: ciτieli − pi, with respect to the module M := IL(F) ⊂ Rm we need to consider the morphisms where the symbols {e1, . . . , eu} denote the common canonical basis of Su and Ru which, as R-modules, coincide. We can then consider – the T (m)-valuation v : Ru → T (m) defined, for each σ := Pu < {T ≺(hi) ◦ T<(gi)} = max < {T(hi ⋆ gi)} = max v(σ) := max < {T i=1 hiei ∈ Ru \ {0}, by ≺(hi) ◦ τieli} =: δǫ under which we further have Su = G(Ru); – the corresponding leading form LL(σ) := Ph∈H M(hh)eh ∈ Su – which is T (m)- homogeneous of T (m)-degree v(σ) = δǫ – where H := n j : T<(h j ⋆ g j) = T ≺(h j) ◦ τ jel j = δǫ = v(σ)o . 26 sL : Su → Sm SL : Ru → M ⊂ Rm u : sL hiei Xi=1 : SL hiei Xi=1 u u Xi=1 Xi=1 u := hi ∗ M(gi), := hi ⋆ gi, Figure 2: Canonical Form Algorithms (g,Pµ i=1 aiλi ⋆ gi) := LeftCanonicalForm( f , F) where R := R[T ], R a ring with canonical representatives, f ∈ Rm, F is the left Grobner basis of the left module I ⊂ Rm, g := Can( f , I) ∈ Rep(I), ai ∈ R \ {0}, λi ∈ T , gi ∈ F, f − g = Pµ i=1 aiλi ⋆ gi is a left weak Grobner representation in terms of F, h := f , µ := 0, g := 0 While h , 0 do Let cτ := M(h), γ := Rep(c, Iτ) h := h − γτ, g := g + γτ, If c , γ, let gi ∈ F, λi ∈ T , ai ∈ R \ {0} : h := h −Pν c − γ = Pν i=µ+1 aiλi ⋆ gi, µ := ν, i=µ+1 aiαλi (lc(gi)), T(g) = λi ◦ T(gi), µ < i ≤ ν, (g,Pµ i=1 gi ⋆ biρi) := RightCanonicalForm( f , F) where R := R[T ], R a ring with canonical representatives, f ∈ Rm, F is the right Grobner basis of the right module I ⊂ Rm, g := Can( f , I) ∈ Rep(I), bi ∈ R \ {0}, ρi ∈ T , gi ∈ F, f − g = Pµ i=1 gi ⋆ biρi is a right weak Grobner representation in terms of F, h := f , µ := 0, g := 0 While h , 0 do Let cτ := M(h), γ := Rep(c, Iτ) h := h − γτ, g := g + γτ, If c , γ, let gi ∈ F, ρi ∈ T , bi ∈ R \ {0} : h := h −Pν c − γ = Pν T(gi)(bi)ρi, µ := ν, i=µ+1 gi ⋆ α−1 i=µ+1 lc(gi)bi, T(g) = T(gi) ◦ ρi, µ < i ≤ ν, 27 Figure 2 (cont.): Canonical Form Algorithms (g,Pµ i=1 aiλi ⋆ gi ⋆ ρi) := BilateralCanonicalForm( f , F) where R := R[T ], R a ring with canonical representatives, f ∈ Rm, F is the bilateral Grobner basis of the bilateral module I ⊂ Rm, g := Can( f , I) ∈ Rep(I), ai ∈ R \ {0}, λi, ρi ∈ T , gi ∈ F, f − g = Pµ of F, i=1 aiλi ⋆ gi ⋆ ρi is a bilateral Grobner representation in terms h := f , µ := 0, g := 0 While h , 0 do Let cτ := M(h), γ := Rep(c, Iτ) h := h − γτ, g := g + γτ, If c , γ, let gi ∈ F, λi, ρi ∈ T , ai ∈ R \ {0} : c − γ = Pν h := h −Pν i=µ+1 aiαλi(lc(gi)), T(g) = λi ◦ T(gi) ◦ ρi, µ < i ≤ ν, i=µ+1 aiλi ⋆ gi ⋆ ρi, µ := ν, Definition 31. With the notation above and denoting for each set S ⊂ Ru, LL{S} := {LL(g) : g ∈ S} ⊂ Su, – for a left R-module N ⊂ Ru, a set B ⊂ N is called a left standard basis if IL(LL{B}) = IL(LL{N}); – for each h ∈ N a representation h = Xi hi ⋆ gi : hi ∈ R, gi ∈ B, is called a left standard representation in R in terms of B iff v(h) ≥ v(hi ⋆ gi), for each i; – if u ∈ ker(sL) is T (m)-homogeneous and U ∈ ker(SL) is such that u = LL(U), we say that u lifts to U, or U is a lifting of u, or simply u has a lifting; – a left Gebauer–Moller set for F is any T (m)-homogeneous basis of ker(sL); – for each T (m)-homogeneous element σ ∈ Ru, we say that SL(σ) has a left quasi- Grobner representation in terms of F if it can be written as SL(σ) = Pu i=1 li ⋆ gi with v(σ) > T(li ⋆ gi) = T(li) ◦ T(gi) for each i. 28 Remark 32. Note that each Grobner representation of SL(σ) in terms of F gives also a quasi-Grobner representation since T(li) ◦ T(gi) ≤ T(SL(σ)) < v(σ); on the other side, a quasi-Grobner representation grants only T(li) ◦ T(gi) < v(σ) but not necessarily T(li) ◦ T(gi) ≤ T(SL(σ)), since in principle we could have T(SL(σ)) < T(li) ◦ T(gi) < v(σ) so that we don't necessarily obtain a Grobner representation of the S-polynomial SL(σ). This relaxation was introduced by Gebauer-Moller in their reformulation of Buch- berger Theory for polynomial rings over a field [18]; in that setting, it allowed to better remove useless S-pairs and thus granted a more efficient reformulation of the algo- rithm; in the more general setting we are considering now, viz polynomials over rings, it becomes essential also for a smooth reformulation of the theory. ⊓⊔ Observe that if σ := Pu j=1 h je j ∈ ker(SL) then denoting δǫ := v(σ) and H := n j, 1 ≤ j ≤ u : T(h j) ◦ T(g j) = δǫo , its leading form LL(σ) := Pu v(σ) := δǫ ∈ T (m), satisfies j=1 d jλ je j ∈ Su is T (m)-homogeneous of T (m)-degree – 0 , d j ⇐⇒ j ∈ H and M(h j) = d jλ j, – Pu j=1 d jλ j ∗ M(g j) = P j∈H(d jλ j) ∗ (c jτ jel j) = (cid:16)P j∈H (cid:16)d jαλ j (c j)(cid:17) ·(cid:16)λ jτ j(cid:17)(cid:17) ǫ = 0, – P j∈H d jαλ j(lc(g j)) = 0 and λ j ◦ T(g j) = δǫ for each j ∈ H , so that in particular – ǫ = el j for each j ∈ H, and belongs to ker(sL). Theorem 33 (Moller; Left Lifting Theorem). [34] With the present notation and denoting GM(F) any left Gebauer–Moller set for F, the following conditions are equivalent: 1. F is a left Grobner basis of M; f ∈ M ⇐⇒ f has a left Grobner representation in terms of F; 2. 3. for each σ ∈ GM(F), the S-polynomial SL(σ) has a quasi-Grobner representa- tion SL(σ) = u Xi=1 ligi; 4. each σ ∈ GM(F) has a lifting lift(σ); 5. each T (m)-homogeneous element u ∈ ker(sL) has a lifting lift(u). Proof. 29 (1) =⇒ (2) Let f ∈ M; by assumption M( f ) = u Xi=1 aiλi ∗ M(gi) where (a1λ1, . . . , auλu) is T (m)-homogeneous of T (m)-degree T( f ). Therefore g := f −Pu g := Pu Thus, we can assume by induction the existence of a Grobner representation i=1(aiλi + li) ⋆ gi is the required Grobner i=1 aiλi ⋆ gi ∈ M and T(g) < T( f ). i=1 li ⋆ gi be a quasi-Grobner representation in terms of F; i=1 liei is the required lifing of σ. i=1 aiλiei, ai , 0 =⇒ λi◦ τieli = v(u), be a T (m)-homogeneous representation of f . i=1 li ⋆ gi of g; whence f := Pu (2) =⇒ (3) SL(σ) ∈ M and T(SL(σ)) < v(σ). (3) =⇒ (4) Let SL(σ) = Pu (4) =⇒ (5) Let u := Pu then T(li ⋆ gi) < v(σ) so that lift(σ) := σ −Pu element in ker(sL) of T (m)-degree v(u). Then there are cσ ∈ R, λσ ∈ T for which cσλσ ∗ σ, u = Xσ∈GM(F) For each σ ∈ GM(F) denote λσ ◦ v(σ) = v(u). ¯σ := σ − lift(σ) = LL(lift(σ)) − lift(σ) := u Xi=1 liσei and remark that T(liσ)◦ τieli ≤ v( ¯σ) < v(σ), SL(lift(σ)) = 0 and SL( ¯σ) = SL(σ). It is sufficient to define lift(u) := Xσ∈GM(F) cσλσ ⋆ lift(σ), and ¯u := Xσ∈GM(F) cσλσ ⋆ ¯σ i=1 li ⋆ gi. i=1 liei ∈ Ru satisfies (5) =⇒ (1) Let g ∈ M, so that there are li ∈ R, such that σ1 := Pu to obtain lift(u) = u − ¯u, LL(lift(u)) = u, SL(¯u) = SL(u), SL(lift(u)) = 0. g = SL(σ1) = Pu Denoting H := {i : T(li) ◦ τieli = v(σ1)}, then either – T(g) = v(σ1), so that M(g) = Pi∈H M(li) ∗ M(gi) ∈ M{IL(M{F})} and we – T(g) < v(σ1), 0 = Pi∈H M(li) ∗ M(gi) = sL(LL(σ1)) and the T (m)-homoge- neous element LL(σ1) ∈ ker(sL) has a lifting U := LL(σ1) −Pu i=1 l′iei with M(li) ⋆ gi and T(l′i) ◦ τieli < v(σ1), are through, or u Xi=1 l′i ⋆ gi = Xi∈H 30 so that g = SL(σ2) and v(σ2) < v(σ1) for σ2 := Xi<H liei +Xi∈H (li − M(li)) ei + u Xi=1 l′iei ∈ Ru and the claim follows by the well-orderedness of <. Theorem 34 (Janet-Schreyer). [19, 46, 47] With the same notation the equivalent conditions (1-5) imply that 6. {lift(σ) : σ ∈ GM(F)} is a left standard basis of ker(SL). Proof. (4) =⇒ (6) Let σ1 := Pu i=1 liei ∈ ker(SL) ⊂ Ru. Denoting H := {i : T(li) ◦ τieli = v(σ1)}, we have LL(σ1) = Xi∈H M(li)ei ∈ ker(sL) and there is a T (m)-homogeneous representation LL(σ1) = Xσ∈GM(F) cσλσ ∗ σ, λσ ◦ v(σ) = v(σ1), cσ ∈ R, λσ ∈ T . Then σ2 cσλσ ⋆ (σ − ¯σ) cσλσ ⋆ lift(σ) := σ1 − Xσ∈GM(F) = σ1 − Xσ∈GM(F) = (σ1 − LL(σ1)) + Xσ∈GM(F)  = Xi∈H (li − M(li)) + Xσ∈GM(F) cσλσ ⋆ ¯σ ⊓⊔ ei (9) λσ ⋆ liσ ei +Xi<H  li + Xσ∈GM(F) λσ ⋆ liσ satisfies both σ2 ∈ ker(SL) and v(σ2) < v(σ1); thus the claim follows by induc- tion. ⊓⊔ Example 35. Let us consider the ring of Example 17 and three elements f1, f2, f3 ∈ R with 2 Y2 3 , M( f2) = (5x − 1)Y2 M( f1) = (5x − 1)Y1Y2 Under the natural T -pseudovaluation on R3, an element 1 Y γ2 2 Y γ3 3 , βY β1 1 Y α2 2 Y α3 1 Y2Y2 1 Y β2 2 Y β3 3 , γY γ1 σ := (cid:16)αY α1 3 (cid:17) ∈ S3 3 , M( f3) = (5x − 1)Y2 1 Y2 2 Y3. 31 is homogeneous of T -degree Ya+2 1 Yb+2 2 Yc+2 3 iff α1 − 1 = β1 = γ1 =: a, α2 = β2 − 1 = γ2 =: b, α3 = β3 = γ3 − 1 =: c. Let us now specialize ourselves to the case a = b = c = 0 and consider the Z- 3 ; (9) is a syzygy in ker(sL) module of the homogeneous syzygies of T -degree Y2 iff 2 Y2 1 Y2 0 = sL(σ) = αY1 ∗ M( f1) + βY2 ∗ M( f2) + γY3 ∗ M( f3) 2 Y2 1 Y2 = (cid:16)α(y2 − 1) + β(y3 − 1) + γ(y4 − 1)(cid:17) Y2 3 . A minimal Gebauer-Moller set consists of σ1 := (−(y2 + y + 1)Y1, (y + 1)Y2, 0) and σ2 := (−(y2 + 1)Y1, 0, Y3). In fact a generic syzygy (9) satisfies α(y + 1) + β(y2 + y + 1) + γ(y2 + 1)(y + 1) = 0 so that (y + 1) β and setting β = (y + 1)δ we have α = −δ(y2 + y + 1)− γ(y2 + 1) whence σ := (cid:16)(cid:16)−δ(y2 + y + 1) − γ(y2 + 1)(cid:17) Y1, (y + 1)δY2, γY3(cid:17) = δσ1 + γσ2. 2 Y2 1 Y2 Remark 36. We can consider also the homogeneous syzygy of T -degree Y2 3 σ3 := (0, −(y2 + 1)(y + 1)Y2, (y2 + y + 1)Y3) − (y2 + 1)σ1 + (y2 + y + 1)σ2. Moreover, since setting we have note that 1 = (y2 + y + 1) − y(y + 1) = (y3 + y2 + y + 1) − y(y2 + y + 1) ςA := (yY1, Y2, 0) ∈ S3, ςB := (0, −yY2, Y3) ∈ S3 sL(ςA) = sL(ςB) = (y − 1)Y2 1 Y2 2 Y2 3 ; ςA − ςB := (yY1, (y + 1)Y2, Y3) = σ1 + σ2 ∈ ker(sL). Example 35 (cont.). Setting now τ := Ya have 1 Yb 2 Yc ⊓⊔ 3 and z := y2a3b4c, for the syzygy (9) we 0 = sL(σ) = ατY1 ∗ M( f1) + βτY2 ∗ M( f2) + γτY3 ∗ M( f3) 1 Y2 = (cid:16)ατ ∗ (y2 − 1) + βτ ∗ (y3 − 1) + γτ ∗ (y4 − 1)(cid:17) Y2 1 Y2 = (cid:16)α(z2 − 1) + β(z3 − 1) + γ(z4 − 1)(cid:17) Y2 2 Y2 3 τ 2 Y2 3 . 32 whence and α = −δ(z2 + z + 1) − γ(z2 + 1), β = (y + 1)δ Thus, {σ1, σ2} is a minimal basis of ker(sL). σ := βτ ∗ σ1 + γτ ∗ σ2. 5.2 Bilateral case Considering R as a left R-module, the adaptation of Moller lifting theorem to the bilat- eral case requires a few elementary adaptations; given a finite set F := {g1, . . . , gu} ⊂ Rm, gi = M(gi) − pi =: ciτieιi − pi, and the bilateral module M := I2(F), denote R := {a ∈ R : ah = a ⋆ h = h ⋆ a, for each h ∈ R} the commutative subring R ⊂ R of R consisting of the elements belonging to the center of R and remark that the subring of R generated by 1R is a subring of R and that R is also a subring of the center of the associated graded Ore extension S of R. Considering both the R-bimodule R ⊗ R Rop and the S-bimodule S ⊗ R Sop, which, as sets, coincide, we impose on the bilateral R-module (cid:0)R ⊗ R Rop(cid:1)u, whose canonical basis is denoted {e1, . . . , eu} and whose generic element has the shape aiλieℓibiρi, ai, bi ∈ R \ {0}, λi, ρi ∈ T , 1 ≤ ℓi ≤ u, the T (m)-graded structure given by the valuation v : (cid:0)R ⊗ R Rop(cid:1)u → T as Xi < {T(λi ⋆ gℓi ⋆ ρi)} = max < {λi ∗ T(gℓi) ∗ ρi} = max < {λi ◦ τℓi ◦ ρieιi} =: δǫ v(σ) := max for each so that σ := Xi aiλieℓibiρi ∈ (cid:0)R ⊗ R Rop(cid:1)u \ {0} G(cid:0)(cid:0)R ⊗ R Rop(cid:1)u(cid:1) = (cid:0)G(cid:0)R ⊗ R Rop(cid:1)(cid:1)u = (cid:0)S ⊗ R Sop(cid:1)u and its corresponding T (m)-homogeneous leading form is L2(σ) := Xh∈H where H := { j : λ j ◦ τ j ◦ ρ jeιℓ j (cid:0)R ⊗ R Rop(cid:1)u, ahλheℓhbhρh ∈ (cid:0)S ⊗ R Sop(cid:1)u = v(σ) = δǫ}; we also denote, for each set S ⊂ L2{S} := {L2(g) : g ∈ S} ⊂ (cid:0)S ⊗ R Sop(cid:1)u . 33 We can therefore consider the morphisms s2 : (cid:0)S ⊗ R Sop(cid:1)u → Sm S2 : (cid:0)R ⊗ R Rop(cid:1)u → Rm Definition 36. With the notation above : aiλieℓibiρi s2 Xi : S2 aiλieℓibiρi Xi := Xi := Xi aiλi ∗ M(gℓi) ∗ biρi, aiλi ⋆ gℓi ⋆ biρi. – for a bilateral R-module N, a set F ⊂ N is called a bilateral standard basis if I2(L2{F}) = I2(L2{N}); – for each h ∈ N a representation h = Xi aiλi ⋆ gℓi ⋆ biρi : ai, bi ∈ R \ {0}, λi, ρi ∈ T , gℓi ∈ F, is called a standard representation in R in terms of F iff v(h) ≥ v(λi ⋆ gℓi ⋆ ρi) = λi ◦ v(gℓi) ◦ ρi, for each i; – if u ∈ ker(s2) is T (m)-homogeneous and U ∈ ker(S2) is such that u = L2(U), we say that u lifts to U, or U is a lifting of u, or simply u has a lifting; – a bilateral Gebauer–Moller set for F is any T (m)-homogeneous basis of ker(s2); – for each T (m)-homogeneous element σ ∈ (cid:0)R ⊗ R Rop(cid:1)u, we say that S2(σ) has a bilateral quasi-Grobner representation in terms of G if it can be written as S2(σ) = Xi aiλi ⋆ gℓi ⋆ biρi : ai, bi ∈ R \ {0}, λi, ρi ∈ T , gℓi ∈ F with λi ◦ T(gℓi) ◦ ρi < v(σ) for each i. ⊓⊔ Theorem 37 (Moller–Pritchard). With the present notation and denoting GM(F) any bilateral Gebauer–Moller set for F, the following conditions are equivalent: 1. F is a bilateral Grobner basis of M; f ∈ M ⇐⇒ f has a bilateral Grobner representation in terms of F; 2. 3. for each σ ∈ GM(F), the bilateral S-polynomial S2(σ) has a bilateral quasi- Grobner representation S2(σ) = Pµ 4. each σ ∈ GM(F) has a lifting lift(σ); 5. each T (m)-homogeneous element u ∈ ker(s2) has a lifting lift(u). l=1 alλl ⋆ gℓl ⋆ blρl, in terms of F; Proof. 34 (1) =⇒ (2) Let f ∈ M; by assumption M( f ) = µ Xi=1 aiλi ∗ M(gℓi) ∗ biρi i=1 aiλieℓibiρi ∈ (cid:0)S ⊗ R Sop(cid:1)u is T (m)-homogeneous of T (m)-degree T( f ). i=1 aiλi ⋆ gℓi ⋆ biρi ∈ M and T(g) < T( f ). Thus, the claim follows by induction since < is a well-ordering. wherePµ Therefore g := f −Pµ (2) =⇒ (3) S2(σ) ∈ M and T(S2(σ)) < v(σ). (3) =⇒ (4) Let S2(σ) = µ Xi=1 aiλi ⋆ gℓi ⋆ biρi, v(σ) > λi ◦ τℓi ◦ ρieιℓi be a bilateral quasi-Grobner representation in terms of F; then lift(σ) := σ − µ Xi=1 aiλieℓibiρi is the required lifting of σ. (4) =⇒ (5) Let u := Pi aiλieℓibiρi ∈ (cid:0)S ⊗ R Sop(cid:1)u homogeneous element in ker(s2) of T (m)-degree v(u). Then there are λσ, ρσ ∈ T , aσ, bσ ∈ R \ {0}, for which , λi ◦ τℓi ◦ ρieιℓi = v(u), be a T (m)- u = Xσ∈GM(F) For each σ ∈ GM(F) denote aσλσ ∗ σ ∗ bσρσ, λσ ◦ v(σ) ◦ ρσ = v(u). ¯σ := σ − lift(σ) = L2(lift(σ)) − lift(σ) := µσ Xi=1 aiσλiσeℓiσbiσρiσ ∈ (cid:0)R ⊗ R Rop(cid:1)u and remark that λiσ ◦ τℓiσ ◦ ρiσeιℓiσ ≤ v( ¯σ) < v(σ) and S2( ¯σ) = S2(σ). It is sufficient to define lift(u) := Xσ∈GM(F) aσλσ ⋆ lift(σ) ⋆ bσρσ, and ¯u := Xσ∈GM(F) aσλσ ⋆ ¯σ ⋆ bσρσ to obtain lift(u) = u − ¯u, L2(lift(u)) = u, S2(¯u) = S2(u), S2(lift(u)) = 0. 35 (5) =⇒ (1) Let g ∈ M, so that there are λi, ρi ∈ T , ai, bi ∈ R \ {0}, 1 ≤ ℓi ≤ u, such that σ1 := Pµ i=1 aiλieℓibiρi ∈ (cid:0)R ⊗ R Rop(cid:1)u satisfies µ g = S2(σ1) = Xi=1 aiλi ⋆ gℓi ⋆ biρi. Denoting H := {i : λi ◦ T(gℓi) ◦ ρi = λi ◦ τℓi ◦ ρieιℓi = v(σ1)}, then either – v(σ1) = T(g) so that, for each i ∈ H, M(aiλi ⋆ M(gℓi) ⋆ biρi) = aiλi ∗ M(gℓi) ∗ biρi and M(g) = Xi∈H aiλi ∗ M(gℓi) ∗ biρi ∈ M{I2(M{F})}, and we are through, or – T(g) < v(σ1), in which case 0 = Pi∈H aiλi ∗ M(gℓi) ∗ biρi = s2(L2(σ1)) and the T (m)-homogeneous element L2(σ1) ∈ ker(s2) has a lifting a jλ jeℓ jb jρ j ∈ (cid:0)R ⊗ R Rop(cid:1)u U := L2(σ1) − Xj=1 ν with ν Xj=1 a jλ j ⋆ gℓ j ⋆ b jρ j = Xi∈H aiλi ⋆ gℓi ⋆ biρi and λ j ◦ τl j ◦ ρ jeιℓ j < v(σ1) so that g = S2(σ2) and v(σ2) < v(σ1) holds for σ2 := Xi<H aiλieℓibiρi + ν Xj=1 a jλ jeℓ jb jρ j ∈ (cid:0)R ⊗ R Rop(cid:1)u and the claim follows by the well-orderedness of <. Theorem 38 (Janet-Schreyer). With the same notation the equivalent conditions (1-5) imply that 6. {lift(σ) : σ ∈ GM(F)} is a bilateral standard basis of ker(S2). i=1 aiλieℓibiρi ∈ ker(S2) ⊂ (cid:0)R ⊗ R Rop(cid:1)u Proof. (4) =⇒ (6) Let σ1 := Pµ Denoting H := {i : λi ◦ τℓi ◦ ρieιℓi = v(σ1)}, we have ⊓⊔ . L2(σ1) = Xi∈H aiλieℓibiρi ∈ ker(s2) and there is a T (m)-homogeneous representation L2(σ1) = Xσ∈GM(F) aσλσ ∗ σ ∗ bσρσ, λσ ◦ v(σ) ◦ ρ = v(σ1) 36 with λσ, ρσ ∈ T , aσ, bσ ∈ R \ {0}. Then σ2 aσλσ ⋆ lift(σ) ⋆ bσρσ aσλσ ⋆ (σ − ¯σ) ⋆ bσρσ aσλσ ⋆ ¯σ ⋆ bσρσ := σ1 − Xσ∈GM(F) = σ1 − Xσ∈GM(F) = σ1 − L2(σ1) + Xσ∈GM(F) = Xi<H + Xσ∈GM(F) aiλieℓibiρi µσ Xi=1(cid:16)(cid:0)aσαλσ(aiσ)(cid:1) · (λσ ◦ λiσ)(cid:17)eℓiσ(cid:16)(cid:16)biσαρσ(bσ)(cid:17) · (ρiσ ◦ ρσ)(cid:17) satisfies both σ2 ∈ ker(S2) and v(σ2) < v(σ1); thus the claim follows by induc- tion. ⊓⊔ 5.3 Restricted case In order to deal with restricted modules, we need simply to adapt and simplify the bilateral case. Thus, we consider both the left R-modules R ⊗ Rop and R ⊗ Sop, which, as sets, coincide, we impose on the bilateral R-module (R ⊗ Rop)u, whose canonical basis is denoted {e1, . . . , eu} and whose generic element has the shape Xi aieℓi ρi, ai ∈ R \ {0}, ρi ∈ T , 1 ≤ ℓi ≤ u, the T (m)-graded structure given by the valuation v : (R ⊗ Rop)u → T as v(σ) := max < {T(gℓi ⋆ ρi)} = max < {T(gℓi) ∗ ρi} = max < {τℓi ◦ ρieιi} =: δǫ for each so that σ := Xi aieℓi ρi ∈ (cid:0)R ⊗ R Rop(cid:1)u \ {0} G(cid:0)(cid:0)R ⊗ Rop(cid:1)u(cid:1) = (cid:0)G(cid:0)R ⊗ Rop(cid:1)(cid:1)u = (cid:0)R ⊗ Sop(cid:1)u and its corresponding T (m)-homogeneous leading form is aheℓh ρh ∈ (cid:0)R ⊗ Sop(cid:1)u LW(σ) := Xh∈H = v(σ) = δǫ}; we also denote, for each set S ⊂ (R ⊗ Rop)u, where H := { j : τ j ◦ ρ jeιℓ j LW{S} := {LW (g) : g ∈ S} ⊂ (cid:0)R ⊗ Sop(cid:1)u . 37 We can therefore consider the morphisms sW : (cid:0)R ⊗ Sop(cid:1)u → Sm SW : (cid:0)R ⊗ Rop(cid:1)u → Rm Definition 39. With the notation above : sW  aieℓi ρi Xi aieℓi ρi : SW  Xi := Xi := Xi aiM(gℓi) ∗ ρi, aigℓi ⋆ ρi. – for a restricted module N, a set F ⊂ N is called a restricted standard basis if IW(LW{F}) = IW(LW{N}); – for each h ∈ N a representation h = Xi aigℓi ⋆ ρi : ai ∈ R \ {0}, ρi ∈ T , gℓi ∈ F, is called a standard representation in R in terms of F iff v(h) ≥ v(gℓi ⋆ ρi) = v(gℓi) ◦ ρi, for each i; – if u ∈ ker(sW) is T (m)-homogeneous and U ∈ ker(SW) is such that u = LW(U), we say that u lifts to U, or U is a lifting of u, or simply u has a lifting; – a restricted Gebauer–Moller set for F is any T (m)-homogeneous basis of ker(sW); – for each T (m)-homogeneous element σ ∈ (cid:0)R ⊗ R Rop(cid:1)u, we say that SW (σ) has a restricted quasi-Grobner representation in terms of G if it can be written as SW(σ) = Xi aigℓi ⋆ ρi : ai ∈ R \ {0}, ρi ∈ T , gℓi ∈ F with T(gℓi) ◦ ρi < v(σ) for each i. ⊓⊔ Theorem 40 (Moller–Pritchard). With the present notation and denoting GM(F) any restricted Gebauer–Moller set for F, the following conditions are equivalent: 1. F is a restricted Grobner basis of IW(F); f ∈ IW(F) ⇐⇒ f has a restricted Grobner representation in terms of F; 2. 3. for each σ ∈ GM(F), the restricted S-polynomial SW(σ) has a restricted quasi- Grobner representation SW(σ) = Pµ 4. each σ ∈ GM(F) has a lifting lift(σ); 5. each T (m)-homogeneous element u ∈ ker(sW) has a lifting lift(u). l=1 algℓl ⋆ ρl, in terms of F; Theorem 41 (Janet-Schreyer). With the same notation the equivalent conditions (1-5) imply that 6. {lift(σ) : σ ∈ GM(F)} is a restricted standard basis of ker(SW). 38 6 Grobner basis Computation for Multivariate Ore Ex- tensions of Zacharias Domains We recall the definition of Zacharias ring [53], [30, §26.1], [31]. Definition 42. A ring R with identity is called a (left) Zacharias ring if it satisfies the following properties: (a). R is a noetherian ring; (b). there is an algorithm which, for each c ∈ Rm, C := {c1, . . . ct} ⊂ Rm \ {0}, allows to decide whether c ∈ IL(C) in which case it produces elements di ∈ R : c = Pt i=1 dici; generators for the left syzygy R-modulen(d1, · · · , dt) ∈ Rt : Pt (c). there is an algorithm which, given {c1, . . . ct} ⊂ Rm \ {0}, computes a finite set of i=1 dici = 0o. Note that [34] for a ring R with identity which satisfies (a) and (b), (c) is equivalent to (d). there is an algorithm which, given {c1, . . . cs} ⊂ Rm \ {0}, computes a finite basis of the ideal IL({ci : 1 ≤ i < s}) : IL(cs). If R has canonical representatives, we improve the computational assumptions of computes the unique canonical representative Rep(c, J). Zacharias rings, requiring also the following property: (e). there is an algorithm which, given an element c ∈ Rm and a left module J ⊂ Rm, ⊓⊔ If R is a left Zacharias domain, the three algorithms proposed by Moller [34] for computing Grobner bases in the polynomial ring over R can be easily adapted to multi- variate Ore extensions of Zacharias domains, provided that each αi, and therefore each ατ, is an automorphism. 6.1 First algorithm Still considering a finite basis F := {g1, . . . , gu} ⊂ Rm, gi = M(gi) − pi =: ciτieli − pi, of the module M := IL(F) and denoting – H(F) := {{i1, i2, . . . , ir} ⊆ {1, . . . , u} : li1 = · · · = lir}; – for each H := {i1, i2, . . . , ir} ∈ H(F), – εH := eli1 = · · · = eli1 – τH := lcm (τi : i ∈ H) , , 39 – for each I ⊂ H, – τH,I := τH , τI – αH,I : R → R the morphism ατH,I ; – T(H) := τH εH, and, if R is a PID, – cH := lcm(αH,i(ci) : i ∈ H), – µ(H) := cHτH and – M(H) = cHT(H) = cHτHεH = µ(H)εH; – T := {T(H) : H ∈ H(F)}; – for any m = δǫ ∈ T, if T(gi) m, otherwise; – v(m) = (v(m)1, . . . , v(m)u) ∈ Ru the vector such that αti(m)(lc(gi)) 0 if T(gi) m, otherwise; δ τi 1 – for each i, 1 ≤ i ≤ u, ti(m) :=  v(m)i :=  SyzL (v(m)1, . . . , v(m)u) :=  – C(m) ⊂ Ru a finite basis of the syzygy module Xi=1 – S (m) := {(d1t1(m), . . . , dutu(m)) : (d1, . . . , du) ∈ C(m)}; (d1, . . . , du) ∈ Ru : u ; div(m)i = 0 – S(F) := Sm∈T S (m); – S′(F) ⊂ S(F) any subset satisfying – for each σ ∈ S(F) \ S′(F) exist σ j ∈ S′(F), d j ∈ R, τ j ∈ T , such that σ = P j d jτ j ∗ σ j; – R(F) := {Pi mi ⋆ gi : (m1, . . . , mu) ∈ S′(F)} , we have that (cf. [34], [30, Theorem 26.1.4]) Lemma 43. S(F) is a Gebauer–Moller set for F. Proof. Let us consider a generic T (m)-homogeneus element σ := u Xi=1 aiλiei ∈ Ru \ {0}, 40 with ai ∈ R, λi ∈ T , v(σ) := τǫ, and ai , 0 =⇒ λiτi = τ, ǫ = eli, and assume that it is a left syzygy in ker(sL). Denoting I := {i ≤ u : ai , 0} and setting m := δǫ := lcm{T(gi) : i ∈ H} v(σ), there is υ ∈ T : υδ = τ. With the present notation we also have δ = ti(m)τi; thus υti(m)τi = υδ = λiτi and λi = υti(m). We also have so that aiαλi(lc(gi)) u 0 = Xi=1 aiαλi(lc(gi)) 0 = α−1 u υ  Xi=1 = u Xi=1 α−1 υ (ai)αti(m)(lc(gi)), so that (α−1 υ (a1), . . . , α−1 υ (au)) ∈ SyzL(v(m)1, . . . , v(m)u). Therefore, if we enumerate as (d11, . . . , d1u), . . . , (dv1, . . . , dvu) a basis of C(m) and we denote s j := Pu υ (a1), . . . , α−1 we have (α−1 i=1 d jiti(m)ei, 1 ≤ j ≤ v, the elements of S (m), j=1 b j(d j1, . . . , d ju) for suitable b j ∈ R and u u σ = υ (au)) = Pv Xi=1 Xi=1 Xi=1 = υ ∗ = = u = = v Xj=1 Xj=1 v aiλiei aiυti(m)ei υ ∗ α−1 υ (ai)ti(m)ei b jd jiti(m)ei u v u Xi=1 Xj=1 αυ(b j)υ ∗ Xi=1 αυ(b j)υ ∗ s j. d jiti(m)ei Corollary 44. The following holds: 1. S′(F) is a Gebauer–Moller set for F. 2. F is a left Grobner basis of the module it generates iff each h ∈ R(F) has a left ⊓⊔ Grobner representation in terms of F. ⊓⊔ 41 1 Ya2 Example 45. If we consider the ring of Example 17 as a left Z[x]-module endowed 2 Ya3 with the Γ-pseudovaluation, Γ = {Ya1 : (a1, a2, a1) ∈ N3}, we obtain a similar solution as the one described in Example 35 . Expressing each M( fi) as M( fi) = lc( fi)T( fi), according Zacharias approach we need to compute a syzygy bases in Z[x] among αY1(lc( f1)) = (y2 − 1), αY2(lc( f2)) = (y3 − 1) and αY3(lc( f3)) = (y4 − 1); the natural solutions (−(y2 + y + 1), (y + 1), 0), (−(y2 − 1), 0, 1) produce σ1 and σ2. Example 46. Let us now specialize the ring of Example 14 to the case 3 n = 3, e1 = e2 = e3 = 1, c1 = 20, c2 = 6, c3 = 15, and let us consider four elements f1, f2, f3, f4 ∈ R with 3 , M( f3) = xY2 3 , M( f2) = x2Y2 M( f1) = xY1Y3 1 Y2Y2 2 Y2 1 Y3 2 Y3, M( f4) = xY2 1 Y2 2 Y2 3 . We have and Denoting T = {Y1Y3 m Y1Y3 2 Y2 3 1 Y2Y2 Y2 3 2 Y2 1 Y2 Y2 3 1 Y3 Y2 2 Y3 2 Y2 1 Y3 Y2 3 2 Y2 3 , Y2 1 Y2Y2 3 , Y2 1 Y2 2 Y2 3 , Y2 1 Y3 2 Y3, Y2 1 Y3 2 Y2 3}, (t1(m), . . . , t4(m)) (1, 0, 0, 0) (0, 1, 0, 0) (0, Y2, 0, 1) (0, 0, 1, 0) (Y1, Y2 2 , Y3, Y2) v(m) (x, 0, 0, 0) (0, x2, 0, 0) (0, 62x2, 0, x) (0, 0, x, 0) (20x, 64x2, 15x, 6x), b(1, 3) b(2, 4) b(3, 4) := (−3Y1, 0, 4Y3, 0), (0, −Y2, 0, 62x), : := (0, 0, −2Y3, 0, 5Y2) we have S (Y2 1 Y2 we can take S (Y2 2 Y2 3 ) = {b(2, 4)} and, since Y2 ∗ b(2, 4) = (0, −Y2 1 Y2 2 , 0, 62Y2 ∗ x) = (0, −Y2 3 ) = {b(1, 3), Y2 ∗ b(2, 4), b(3, 4)}; thus S′ := {b(1, 3), b(2, 4), b(3, 4)} 3 Y2 2 , 0, 63xY2) is the required Gebauer–Moller set. ⊓⊔ 6.2 Second algorithm Moller proposes an (essentially) equivalent alternative computation: for any s, 1 ≤ s ≤ u, let us consider the syzygy module (h1, . . . , hs) : Ss :=  hi ⋆ M(gi) = 0 ⊂ Rs s Xi=1 42 and let us compute S(F) = Su by inductively extending Ss−1 to Ss, the inductive seed being S1 = ∅. A direct application of the property (d) of a Zacharias ring allows to compute a Gebauer-Moller set via Definition 47. A subset H ⊂ {1, . . . , s} ∩ H(F), s ≤ u, is said to be maximal for a term δǫ ∈ T (m) if H = {i, 1 ≤ i ≤ s : τi δ, eli = ǫ}, basic if s ∈ H and H is maximal for T(H). For a basic subset H ⊂ {1, . . . , s} ∩ H(F), denote H× := H \ {s}. For any ds ∈ IL({αH,i(ci) : i ∈ H×}) : IL(αH,s(cs)), a syzygy associated to H and ds is any T (m)-homogeneous syzygy τH τi di Xi∈H× ei + ds τH τs es ∈ Ss where di ∈ R are suitable elements for which dsαH,s(cs) = −Pi∈H× diαH,i(ci). Theorem 48 (Moller). [34] With the present notation, denoting ⊓⊔ – {A1, . . . , Aµ} a T (m)-homogeneous basis of Ss−1, – H the set of all basic subsets H ⊂ {1, . . . , s} ∩ H(F), – {d1H, . . . , drHH} a basis of the ideal IL({αH,i(ci) s.t. i ∈ H×}) : IL(αH,s(cs)) for so that Pi∈H diαλi(ci) = 0,Pi∈H α−1 υ (ds)αH,s(cs) ∈ IL(αH,i(ci) : i ∈ H×) and α−1 α−1 j=1 u jd jH and S −PrH Therefore α−1 υ (ds) = PrH 43 υ (ds) ∈ IL(αH,i(ci) : i ∈ H×) : IL(αH,s(cs)). ⊓⊔ j=1 αυ(u j)υ ∗ D jH ∈ Ss−1. each basic subset H ∈ H , each j, 1 ≤ j ≤ rH – D jH ∈ Rs a syzygy associated to H and d jH, for each basic subset H ∈ H and the set {A1, . . . , Aµ} ∪ {D jH : H ∈ H , 1 ≤ j ≤ rH} is a T (m)-homogeneous basis of Ss. Proof. Let S := (d1λ1, . . . , dsλs) ∈ Ss, ds , 0, be a T (m)-homogeneous element of T (m)-degree δǫ and let K := {i, 1 ≤ i ≤ s : di , 0}; since by T (m)-homogeneity, τi δ and eιi = ǫ for each i ∈ K, we have T(K) δǫ; we also have di = 0 for each i < K and λiτi = δ, eιi = ǫ for each i ∈ K. For the set H := {i, 1 ≤ i ≤ s : τi τK , eιi = εK} clearly we have τH τK and K ⊆ H so that τH τK δ; we also have εH = εK = ǫ. Moreover ds , 0 implies s ∈ K ⊆ H so that H is basic. Since (d1λ1, . . . , dsλs) ∈ Ss, setting υ := δ τi ∗ ciτiǫ =  Xi∈H diλi ∗ M(gi) = Xi∈H diαλi(ci) , we have 0 = s Xi=1 δ di δǫ τH υ (di)αH,i(ci) = 0, whence 1 Ya2 Example 49. If we consider the ring of Example 17 as a left Z[x]-module endowed 2 Ya3 with the Γ-pseudovaluation, Γ = {Ya1 : (a1, a2, a1) ∈ N3}, we obtain a similar solution as the one described in Example 35 . Expressing each M( fi) as M( fi) = lc( fi)T( fi), according Zacharias approach we need to compute a syzygy bases in Z[x] among αY1(lc( f1)) = (y2 − 1), αY2(lc( f2)) = (y3 − 1) and αY3(lc( f3)) = (y4 − 1); the natural solutions (−(y2 + y + 1), (y2 + 1), 0), (−(y2 − 1), 0, 1) produce σ1 and σ3. Example 50. In Examples 46, the basic elements are the following: 3 s = 1 s = 2 s = 3 s = 4 T(H) H 2 Y2 Y1Y3 {1} 3 Y1Y3 2 Y2 {1} 3 1 Y2Y2 Y2 {2} 3 2 Y2 Y2 1 Y3 {1, 2} 3 2 Y2 Y1Y3 {1} 3 1 Y2Y2 Y2 {2} 3 1 Y3 Y2 {3} 2 Y3 2 Y2 1 Y3 Y2 {1, 2, 3} 3 2 Y2 Y1Y3 {1} 3 1 Y2Y2 Y2 {2} 3 Y2 1 Y3 {3} 2 Y3 Y2 1 Y2 2 Y2 {2, 4} 3 2 Y2 1 Y3 {1, 2, 3, 4} Y2 3 1 Y3 fH f1 f1 f2 f{1,2} = 4xY2 f1 f2 f3 f{1,2,3} = xY2 f1 f2 f3 f{2,4} = f4 f{1,2,3,4} = f{1,2,3}. 1 Y3 2 Y2 3 , 2 Y2 3 , ⊓⊔ Corollary 51. Assuming that the Zacharias domain R is a principal ideal domain and denoting2, for each i, j, 1 ≤ i < j ≤ u, eιi = eι j, b(i, j) := − B(i, j) := − α{i, j}, j(c j) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) α{i, j}, j(c j) α{i, j},i(ci) α{i, j},i(ci) lcm(τi, τ j) τ j lcm(τi, τ j) τi e j ei, lcm(τi, τ j) τ j lcm(τi, τ j) τi ⋆ g j ⋆ gi we have that {b(i, j) : 1 ≤ i < j ≤ u, eιi = eι j} is a Gebauer–Moller set for F, so that F is a Grobner basis of M, iff each B(i, j), 1 ≤ i < j ≤ u, eιi = eι j , has a weak Grobner representation in terms of F. ⊓⊔ 2Remember that α{i, j}, j = ατ for τ = lcm(τi,τ j) τ j . 44 Proof. Since, for any basic subset H ⊂ {1, . . . , s} ∩ H(F) we have I({α{i,s},i(ci) : i ∈ H×}) : I(α{i,s},s(cs)) = M(I(α{i,s},i(ci)) : I(α{i,s},s(cs))) ! = I lcm(α{i,s},i(ci), α{i,s},s(cs)) α{i,s},s(cs) and b(i, s) is the syzygy associated to {i, s} and lcm(α{i,s},i(ci),α{i,s},s(cs)) ⊓⊔ Example 52. In Examples 46, we obtain the following redundant Gebauer–Moller set (see Examples 65) α{i,s},s(cs) . lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) (i, j) (1, 2) 64 · 5x2 (1, 3) 60x (2, 3) 64 · 5x2 (1, 4) 60x (2, 4) 62x2 (3, 4) 30x lcm(τi, τ j) Y2 1 Y3 1 Y3 Y2 Y2 1 Y3 1 Y3 Y2 1 Y2 Y2 Y2 1 Y3 2 Y2 3 2 Y2 3 2 Y2 3 2 Y2 3 2 Y2 3 2 Y2 3 b(i, j) (−2234xY1, 5Y2 2 , (−3Y1, 0, (0, −5Y2 2 , (−3Y1, 0, (0, −Y2, 0, (0, 0, 4Y3, 3324xY3, 0, 0, −2Y3, 0) 0) 0) 10Y2) 62x) 5Y2) ⊓⊔ Corollary 53. Assuming that the Zacharias domain R is a principal ideal domain and that each αi is an automorphism denoting, for each i, j, 1 ≤ i < j ≤ u, eιi = eι j, ! lcm(τi, τ j) − eiα−1 := e jα−1 b(i, j) τi B(i, j) := g j ⋆ α−1 ! lcm(τi, τ j) we have that {b(i, j) : 1 ≤ i < j ≤ u, eιi = eι j} is a Gebauer–Moller set for F, so that F is a right Grobner basis of M, iff each B(i, j), 1 ≤ i < j ≤ u, eli = el j , has a right weak Grobner representation in terms of F. ⊓⊔ τi lcm(ci, c j) ci τ j τi c j τ j lcm(ci, c j) τ j lcm(ci, c j) ! lcm(τi, τ j) ! lcm(τi, τ j) c j τ j τi lcm(ci, c j) − gi ⋆ α−1 ci 6.3 Third algorithm: from weak to strong Grobner basis As regards strong Grobner bases, we have Definition 54. A set C ⊂ Rm is called a completion of F, if, for each subset H ⊂ H(F) which is maximal for T(H), it contains an element fH ∈ I(F) which satisfies 1. T( fH) = T(H) = τHεH, 2. lc( fH) = cH = gcd(cid:0)αH,i(lc(gi)) : i ∈ H(cid:1), fH has a Grobner representation in terms of F. 3. 45 Algorithm 55 (Moller). A completion of F can be inductively computed by mimicking the construction of Theorem 48 as follows: the result being trivial if #F = 1, we can assume to have already obtained a completion C(F×) of F× = {g1, . . . , gs−1}, s ≤ u; for each maximal subset H ⊂ {1, . . . , s}, if s < H we can take as fH the corresponding element in C(F×). If instead s ∈ H, then H× is maximal in F× for T(H×) and τH× τH; thus there is a corresponding element fH× in C(F×); let us compute the values s, t, d ∈ R such that αH,H×(lc( fH×))s + αH,s(lc(gs))t = gcd(αH,H×(lc( fH×)), αH,s(lc(gs))) = d and define fH := s τH τH× ⋆ fH× + t τH τs 1. T( fH) = T(H) = τHǫH, ⋆ gs which satisfies M( fH) = dT(H) = dτHǫH so that 2. lc( fH) = gcd(αH,H×(lc( fH×)), αH,s(lc(gs))) = gcd(cid:0)αH,i(lc(gi)) : i ∈ H(cid:1) = d;. 3. it is sufficient to substitute fH× with its Grobner representation, to obtain the ⊓⊔ Proposition 56 (Moller). With the present notation and under the assumption that R is a principal ideal domain, the following conditions are equivalent: required Grobner representation of fH. 1. F is a Grobner basis of M; 2. a completion of F is a strong Grobner basis of M. Proof. (1) =⇒ (2) Let f ∈ M and let f = Pu i=1 hi ⋆ gi be a Grobner representation; denoting H := { j : T(h j ⋆ g j) = T( f ) =: τǫ} we have τH τ, ǫH = ǫ. Thus, setting υ j := τ τ j for each j and λ := τ τH , ω j := τH τ j we have lc( f ) = P j∈H lc(h j)αυ j(lc(g j)) = P j∈H lc(h j)αλαω j(lc(g j)) ∈ I(cid:16)αλαω j (lc(g j)) : j ∈ H(cid:17) = αλ(cid:16)I(cid:16)αω j(lc(g j)) : j ∈ H(cid:17)(cid:17) = αλ(I(cH)) so that αλ(lc( fH)) = αλ(cH) lc( f ) and lc( f ) = dαλ(lc( fH)) with d ∈ R. In conclusion we have M( f ) = dλ ∗ M( fH). cK τK fK be a strong Grobner representation (2) =⇒ (1) Let f ∈ M and let f = PK⊂H(F) of it in terms of a completion of F; it is sufficient to substitute each fK with a Grobner representation of it in terms of F to obtain the required representation. ⊓⊔ Example 57. In the ring of Examples 17 and 35, we finally have (see Remark 36) f{1,2} = f{1,3} = f{1,2,3} = sL(ςA) = sL(ςB) = (y − 1)Y2 1 Y2 2 Y2 3 . 46 ⊓⊔ Example 58. The strong Grø"bner basis (see Examples 50) is since { f1, f2, f3, f4, f{1,2,3,4}} gcd(α{1,2},{1}(lc( f1)), α{1,2},2(lc( f2))) = gcd(αY1(x)), αY2 2 (x)) = gcd(20x)), 36x) = 4x gcd(α{1,2,3},{1,2}(lc( f{1,2})), α{1,2,3},3(lc( f3))) = gcd(α1(4x)), αY3(x)) = gcd(4x)), 15x) = x. Similarly, f{2,4} = f4 follows trivially from gcd(α{2,4},{2}(lc( f2)), α{2,4},4(lc( f4))) = gcd(αY2(x2)), α1(x)) = gcd(36x2)), x) = x. ⊓⊔ 6.4 Useless S-pairs and Gebauer-Moller sets Let us still assume that the Zacharias domain R is a principal ideal domain and we will use freely notations as M(i), M(i, j), M(i, j, k), 1 ≤ i, j, k ≤ u, instead of M({i}), M({i, j}), M({i, j, k}); we can then easily apply to the present setting the reformulation and improvement by Gebauer–Moller [18] of Buchberger Criteria [9]. However we must be aware that in this context, there is no chance of reformulating Buchberger's First Criterion. Remark 59. In fact we should at least require that M(i) ∗ M( j) = M(i) ∗ M( j) = M(i, j) id est not only lcm(τi, τ j) = τi ◦ τ j = τ j ◦ τi which is trivially true but also lcm(ατ j(ci), ατi(c j)) = c jατ j(ci) = ciατi(c j). This essentially requires ci ατi(c j) whence ατ j = Id; this sug- gests that Buchberger's First Criterion hardly can be applied except for the case of the commutative ring P = R[Y1, . . . , Yn], R a PIR, where it is stated as ατ j(ci) and c j If F ⊂ P and I(F) is an ideal of P, it holds M(i)M( j) = M(i, j) ⇐⇒ lcm(τi, τ j) = τiτ j, lcm(ci, c j) = cic j =⇒ NF(B(i, j), F) = 0. to the classical polynomial ring case. Note that the proof which considers the trival sysygies gig j − g jgi = 0 holds only ⊓⊔ Definition 60. A useful S-pair set for F is any subset GM ⊂ S(u) = n{i, j}, 1 ≤ i < j ≤ u, eli = el jo such that {b(i, j) : {i, j} ∈ GM} is a Gebauer–Moller set for F. 47 Corollary 61. With the present notation, under the assumption that R is a principal ideal domain, F is a Grobner basis of the left module M iff, denoting GM a useful S-pair set for F, each S-polynomial B(i, j), {i, j} ∈ GM has a Grobner representation in terms of F. ⊓⊔ Proof. By definition {b(i, j) : {i, j} ∈ GM} is a Gebauer–Moller set for F so that, by Theorem 33, F is a Grobner basis of M iff each S-polynomial B(i, j), {i, j} ∈ GM has a Grobner representation in terms of F. ⊓⊔ If we moreover define, – for each i, j : 1 ≤ i, j ≤ u, eli = el j , – c(i, j) := lcm(α{i, j},i(ci), α{i, j}, j(c j)), – τ(i, j) = lcm(τi, τ j) – µ(i, j) = c(i, j)τ(i, j) – and for each i, j, k : 1 ≤ i, j, k ≤ u, eli = el j = elk , – c(i, j, k) := lcm(α{i, j,k},{i, j}(c(i, j)), α{i, j,k},{i,k}(c(i, k)), α{i, j,k};{ j,k}(c( j, k))) – τ(i, j, k) = lcm(τi, τ j, τk) – µ(i, j, k) = c(i, j, k)τ(i, j, k), and we impose on the set S(u) := n{i, j}, 1 ≤ i < j ≤ u, eli = el jo the ordering ≺ defined by {i1, j1} ≺ {i2, j2} ⇐⇒ τ(i1, j1) < τ(i2, j2) τ(i1, j1) = τ(i2, j2), j1 < j2 τ(i1, j1) = τ(i2, j2), j1 = j2, i1 < i2, or or (10)  we obtain Definition 62. An S-element b(i, j), 1 ≤ i < j ≤ u, eli = el j , and the related S-pair {i, j} are called redundant if either (a). exists k > j, elk = eli = el j such that µ(i, j, k) = µ(i, j); µ(i, k) , µ(i, j) , µ( j, k) (b). or exists k < j, elk = eli = el j : µ( j, k) µ(i, j) , µ(k, j). Lemma 63 (Moller). The following holds 1. for each i, j, k : 1 ≤ i, j, k ≤ u, eli = el j = elk , it holds µ(i, j, k) µ(i, k) b(i, k) − µ(i, j, k) µ(i, j) b(i, j) + µ(i, j, k) µ(k, j) b(k, j) = 0. ⊓⊔ 48 2. R := (cid:8)b(i, j), 1 ≤ i < j ≤ u, eli = el j and not redundant(cid:9) is a useful S-element set. 3. Let G := {g1, . . . , gs}, s ≤ u, and let GM∗ ⊂ {{i, j}, 1 ≤ i < j < s, eli = el j} be a useful S-pair set for G∗ = {g1, . . . , gs−1}. Let M := {µ( j, s) : 1 ≤ j < s, el j = els} and let M′ ⊂ M be the set of the elements µ := µ( j, s) ∈ M such that there exists µ( j′, s) ∈ M : µ( j′, s) µ( j, s) , µ( j′, s). For each µ := M( j, s) ∈ M \ M′ let iµ, 1 ≤ iµ < s, be such that µ = M(iµ, s). Then GM := GM∗ ∪ {{iµ, s} : µ ∈ M \ M′} is a useful S-pair set for G. Proof. 1. (cf. [30, Lemma 25.1.4]) One has µ(i, j, k) µ(i, j) ∗ b(i, j) + µ(i, j, k) µ(k, j) ∗ b(k, j) µ(i, j, k) µ(i, k) ∗ b(i, k) − µ(i, k) ∗ µ(i, k) µ(i, j, k) µ(i, j) ∗ µ(i, j) µ(k, j) ∗ µ(k, j) µ(i, j, k) µ(i, j, k) µ(k) µ( j) ek − e j − µ(i, k) µ(i) µ(i, j) µ(i) µ(k, j) µ(k) ei! ei! ek! = − + µ(k) = µ(i, j, k) − µ(i, j, k) + µ(i, j, k) µ( j) µ( j) ek − e j − e j − µ(i) e j − µ( j) µ(i, j, k) ei! ei! ek! µ(i, j, k) µ(i, j, k) µ(k) µ(i) = 0 2. (cf. [30, Lemma 25.1.8]) In order to prove the claim by induction, it is sufficient to show that, for each redundant {i, j}, 1 ≤ i < j ≤ u, eli = el j =: ǫ, there are – {i1, j1}, . . . , {iρ, jρ}, . . . , {ir, jr}, 1 ≤ iρ < jρ ≤ u, eliρ – elements t1, . . . , tr ∈ T , – and coefficients c1, . . . cr ∈ R \ {0} = el jρ = ǫ, such that – b(i, j) = Pρ cρtρ ∗ b(iρ, jρ); – τ(i, j) = tρ ◦ τ(iρ, jρ), for each ρ; 49 – {iρ, jρ} ≺ {i, j}. In order to show this, we only need to consider the representation b(i, j) = µ(i, j, k) µ(i, k) ∗ b(i, k) + µ(i, j, k) µ(k, j) ∗ b(k, j) and to prove that {i, k} ≺ {i, j} ≻ {k, j}; this happens (according to the two cases of the definition) because (a) τ(i, k) τ(i, j, k) = τ(i, j) , τ(i, k) implies {i, k} ≺ {i, j} and the same argu- ment proves { j, k} ≺ {i, j}; (b) the same argument as that above proves { j, k} ≺ {i, j}, while {i, k} ≺ {i, j} because τ(i, k) ≤ τ(i, j) and k < j. 3. (cf. [30, Lemma 25.1.9]) Let i < s, eli = els =: ǫ, µ := µ(i, s). Then: – if there exists µ′ ∈ M such that µ(iµ′ , s) = µ′ iµ′ < s, {i, s} is redundant; – if i = iµ then {im, s} ∈ GM; – if i , iµ then b(i, s) = µ(i,iµ) b(i, iµ) − b(iµ, s). µ(i,iµ,s) µ(i, s) , µ′, then since ⊓⊔ Corollary 64. With the present notation, under the assumption that R is a principal ideal domain, F is a Grobner basis of M iff each S-polynomial B(i, j), {i, j} ∈ R has a Grobner representation in terms of F. ⊓⊔ Example 65. In connection with Lemma 63 we have (i, j, k) 1 Y3 (1, 2, 3) 64 · 5x2 Y2 (1, 2, 4) 64 · 5x2 Y2 1 Y3 1 Y3 Y2 (1, 3, 4) 60x 1 Y3 (2, 3, 4) 64 · 5x2 Y2 c(i, j, k) µ(i, j, k) 2 Y2 3 2 Y2 3 2 Y2 3 2 Y2 3 b(i, j, k) b(1, 2) − 2233xb(1, 3) + b(2, 3) = 0, b(1, 2) − 2233xb(1, 4) + 5Y2 ∗ b(2, 4) = 0, b(1, 3) − b(1, 4) + 2b(3, 4) = 0. b(2, 3) − 5Y2 ∗ b(2, 4) + 63b(3, 4) = 0. . Note that we obviously [19, 33] have also b(1, 2, 3) − b(1, 2, 4) + 2233xb(1, 3, 4) − b(2, 3, 4). Thus the redundant elements are b(2, 3) via 1 or 4, b(1, 2) via 4 and b(1, 4) via 3. But, as it is well-known, it is more efficient (if else for storing considerations) the algorithm sketched in Lemma 63.3 which for s = 2 stores (1, 2), for s = 3 stores (1, 3), for s = 4 removes (1, 2) and stores (2, 4) and (3, 4). 50 Thus the Gebauer Moller set is still {b(1, 3), b(2, 4), b(3, 4)} while b(1, 4) = b(1, 3) + 2b(3, 4), b(2, 3) = 5Y2 ∗ b(2, 4) + 63b(3, 4), b(1, 2) = 2233xb(1, 4) − 5Y2 ∗ b(2, 4). ⊓⊔ 7 Weispfenning Completions for Bilateral Grobner ba- sis for Multivariate Ore Extensions of Zacharias Do- mains 7.1 Kandri-Rody–Weispfenning completion The most efficient technique for producing bilateral Grobner bases G := I2(F) in a noetherian Ore extension is Kandri-Rody–Weispfenning completion [21]. Iteratively: – Repeat – Compute a left-Grobner basis G of the ideal IL(F); – for each g ∈ G, 1 ≤ i ≤ n, compute the normal form NF(g ⋆ Yi, IL(F)) of – set H := {NF(g ⋆ Yi, IL(G)), g ∈ G, 1 ≤ i ≤ n}, F := G ∪ H g ⋆ Yi w.r.t. G; until H = ∅. The rationale of the algorithm is Lemma 66 (Kandri-Rody–Weispfenning). For G ⊂ R the following conditions are equivalent: 1. IL(G) = I2(G); 2. for each τ ∈ T and each g ∈ G, g ⋆ τ ∈ IL(G); 3. for each i, 1 ≤ i ≤ n, and each g ∈ G, g ⋆ Yi ∈ IL(G). Proof. (1) =⇒ (2) ⇐⇒ (3) is trivial. (2) =⇒ (1) B2(G) := {λ ⋆ g ⋆ ρ : λ, ρ ∈ T , g ∈ G} is an R-linear basis of I2(G) and satisfies B2(G) = {λ ⋆ (g ⋆ ρ) : λ, ρ ∈ T , g ∈ G} ⊆ {λ ⋆ h : λ ∈ T , h ∈ IL(G)} ⊆ IL(G). ⊓⊔ 51 7.2 Weispfenning: Restricted Representation and Completion We can wlog assume that R is effectively given as a quotient R = R/I of a free monoid ring R := Zhvi (over Z and the monoid hvi of all words over the alphabet v) modulo a bilateral ideal I. We must restrict ourselves to the case in which < is a sequential term-ordering, id est for each τ ∈ T , the set {ω ∈ T : ω < τ} is finite. Lemma 67. [52] Let F := {g1, . . . , gu} ⊂ Rm, gi = M(gi) − pi =: ciτieιi − pi; set Ω := max<{T(gi) : 1 ≤ i ≤ u}. Let M be the bilateral module M := I2(F) and IW(F) the restricted module IW(F) := SpanR(a f ⋆ ρ : a ∈ R \ {0}, ρ ∈ T , f ∈ F). If every f ⋆ αυ(v), f ∈ F, v ∈ v, υ ∈ T , υ < Ω, has a restricted representation in terms of F w.r.t. a sequential term-ordering <, then every f ⋆ r, f ∈ F, r ∈ R, has a restricted representation in terms of F w.r.t. <. i=1 vi, vi ∈ v and prove the claim by induction on Thus we have a restricted representation in terms of F d jgi j ⋆(cid:16)ρ j ⋆ vν(cid:17) = Xj d jgi j ⋆ αρ j(vν)ρ j and since ρ j < T( f ) ≤ Ω each element gi j ⋆ αρ j(vν) can be substituted with its restricted representation whose existence is granted by assumption. ⊓⊔ Lemma 68. [52] Under the same assumption, if, for each g j ∈ F, both Yi⋆g j, 1 ≤ i ≤ n and each g j ⋆ αυ(v), v ∈ v, υ ∈ T , υ < Ω, have a restricted representation in terms of F w.r.t. <, then IW(F) = M. Proof. It is sufficient to show that, for each f ∈ IW(F), both each Yi ⋆ f ∈ IW(F), 1 ≤ i ≤ n and each f ⋆ r ∈ IW(F), r ∈ R. By assumption f = P j d jgi j ⋆ ρ j, d j ∈ R \ {0}, ρ j ∈ T , 1 ≤ i j ≤ u, so that Yi ⋆ f = Xj αi(d j) ⋆ (Yi ⋆ gi j) ⋆ ρ j and f ⋆ r = Xj d j(gi j ⋆ αρ j(r)) ⋆ ρ j; by assumption each Yi ⋆gi j has a restricted representation in terms of F; for the Lemma above, also each gi j ⋆ αρ j(r) has a restricted representation in terms of F. ⊓⊔ 52 Proof. We can wlog assume r = Qν ν ∈ N. vi ⋆ vν = Xj ν−1 f ⋆ Yi=1 ν−1 Yi=1 whence we obtain = Xj f ⋆ f ⋆ ν Yi=1 vi =  vi d jgi j ⋆ ρ j, τi j ◦ ρ j ≤ T( f ), Corollary 69. [52] Let F := {g1, . . . , gu} ⊂ Rm, gi = M(gi) − pi =: ciτieιi − pi. Let M be the bilateral module M := I2(F) and IW(F) the restricted module IW(F) := SpanR(a f ⋆ ρ : a ∈ R \ {0}, ρ ∈ T , f ∈ F). F is the bilateral Grobner basis of M iff 1. denoting GM(F) any restricted Gebauer–Moller set for F, each σ ∈ GM(F) has a restricted quasi-Grobner representation in terms of F; 2. for each g j ∈ F, both Yi ⋆ g j, 1 ≤ i ≤ n and each g j ⋆ αυ(v), v ∈ v, υ ∈ T , υ < Ω, have a restricted representation in terms of F w.r.t. <. 7.3 Gebauer-Moller sets for Restricted Grobner bases It is clear from Corollary 69 that the computation of a Grobner bases can be ob- tained via Weispfenning's completion, provided that we are able to produce restricted Gebauer-Moller sets; to do so, we need only to properly reformulate the results of Section 7.2. We begin by remarking that for each monomial cτ ∈ M(R) the function g 7→ cg ⋆ τ distributes, thus we can define a multiplication ⋄ : R × R → R by setting ciτi ⋄ c jτ j := cic jτ jτi = c jciτiτ j =: c jτ j ⋄ ciτi which of course is commutative and thus, granting the trivial syzygy allows to recover Buchberger First Criterium. gi ⋄ g j = g j ⋄ gi As a consequence, we can define the notion of restricted Grobner representation: – we say that f ∈ Rm \ {0} has a restricted Grobner representation in terms of G if it can be written as f = Pu i=1 li ⋄ gi, with li ∈ R, gi ∈ G and T(li) ◦ T(gi) ≤ Let us denote, for each i, j, 1 ≤ i < j ≤ u, eli = el j , T( f ) for each i. lcm(ci, c j) lcm(τi, τ j) lcm(ci, c j) ei lcm(τi, τ j) τi − ci gi ⋆ lcm(τi, τ j) τi bW(i, j) := = BW(i, j) := = e j c j M(i, j) M( j) ⋄ e j − lcm(ci, c j) g j ⋆ c j M(i, j) M( j) ⋄ g j − − ci τ j M(i, j) M(i) ⋄ ei ∈ ker(sW), lcm(τi, τ j) lcm(ci, c j) τ j M(i, j) M(i) ⋄ gi. and let us explicitly assume that 53 – for each g j ∈ F, both Yi ⋆ g j, 1 ≤ i ≤ n and each g j ⋆ αυ(v), v ∈ v, υ ∈ T , υ < T(g j), have a restricted representation in terms of F w.r.t. <. Lemma 70 (Buchberger's First Criterion). If m = 1, id est F ⊂ R and IW(F) is an ideal of R, then M(i) ⋄ M( j) = M(i, j) ⇐⇒ lcm(τi, τ j) = τiτ j and lcm(ci, c j) = cic j =⇒ NFW(BW(i, j), F) = 0. Proof. We will prove that BW(i, j) has a restricted Grobner representation in terms of F; thus the result will follow by the equivalence Proposition 19, (4) ⇐⇒ (8). Remark that pi := gi − M(i) = Xl satisfy T(pi) < T(gi), T(p j) < T(g j). Then it holds: ciltil and p j := g j − M( j) = Xk c jkt jk 0 = gi ⋄ g j − g j ⋄ gi = M(i) ⋄ g j + pi ⋄ g j − M( j) ⋄ gi − p j ⋄ gi, and BW(i, j) := M(i, j) M( j) ⋄ g j − M(i, j) M(i) ⋄ gi = M(i) ⋄ g j − M( j) ⋄ gi = p j ⋄ gi − pi ⋄ g j. There are then two possibilities: either – M(p j) ⋄ M(gi) , M(pi) ⋄ M(g j) in which case and T(BW(i, j)) = max(T(p j) ◦ T(gi), T(pi) ◦ T(g j)) BW(i, j) = p j ⋄ gi − pi ⋄ g j = Xk c jkgi ⋆ t jk −Xl cilg j ⋆ til is a restricted Grobner representation; – or M(p j) ⋄ M(gi) = M(pi) ⋄ M(g j), T(BW(i, j)) < T(p j) ◦ T(gi) = T(pi) ◦ T(g j), in which case BW(i, j) = p j ⋄ gi − pi ⋄ g j would not be a Grobner representation. But the latter case is impossible: in fact, from T(pi) ◦ T(g j) = T(p j) ◦ T(gi) < T(g j) ◦ T(gi) we deduce lcm(T(gi), T(g j)) , T(g j) ◦ T(gi) and T(i, j) , T(i) ◦ T( j) contradicting the assumption M(i, j) = M(i) ⋄ M( j). ⊓⊔ Definition 71. Denote A useful S-pair set for F is any subset n{i, j} : M(i) ⋄ M( j) = M(i, j)o ∅ Cu :=  GM ⊂ S(u) = n{i, j}, 1 ≤ i < j ≤ u, eli = el jo such that {b(i, j) : {i, j} ∈ GM ∪ Cu} is a Gebauer–Moller set for F. if M is an ideal otherwise. 54 Corollary 72. With the present notation, under the assumption that R is a principal ideal domain, F is a Grobner basis of M iff, denoting GM a useful S-pair set for F, each S-polynomial BW(i, j), {i, j} ∈ GM has a Grobner representation in terms of F. ⊓⊔ Proof. By definition {bW(i, j) : {i, j} ∈ GM ∪ Cu} is a Gebauer–Moller set for F so that, by Theorem 33, F is a Grobner basis of M iff each S-polynomial BW(i, j), {i, j} ∈ GM∪ Cu has a Grobner representation in terms of F. The claim is a direct consequence of Buchberger's First Criterion which states that for each {i, j} ∈ Cu, BW(i, j) has a weak Grobner representation in terms of F. ⊓⊔ Definition 73. An S-element b(i, j), 1 ≤ i < j ≤ u, eli = el j , and the related S-pair {i, j} are called redundant if either (a). exists k > j, elk = eli = el j such that M(i, j, k) = M(i, j); M(i, k) , M(i, j) , M( j, k), (b). or exists k < j, elk = eli = el j : M( j, k) M(i, j) , M( j, k). Lemma 74 (Moller). The following holds 1. for each i, j, k : 1 ≤ i, j, k ≤ u, eli = el j = elk , it holds ⊓⊔ c(i, j, k) c(i, k) b(i, k)∗ τ(i, j, k) τ(i, k) − c(i, j, k) c(i, j) b(i, j)∗ τ(i, j, k) τ(i, j) + c(i, j, k) c(k, j) b(k, j)∗ τ(i, j, k) τ(k, j) = 0. 2. R := (cid:8)b(i, j), 1 ≤ i < j ≤ u, eli = el j and not redundant(cid:9) is a useful S-element set. 3. Let G := {g1, . . . , gs}, s ≤ u, and let GM∗ ⊂ {{i, j}, 1 ≤ i < j < s, eli = el j} be a useful S-pair set for G∗ = {g1, . . . , gs−1}. Let M := {M( j, s) : 1 ≤ j < s, el j = els} and let M′ ⊂ M be the set of the elements µ := M( j, s) ∈ M such that either – there exists M( j′, s) ∈ M : M( j′, s) M( j, s) , M( j′, s) or – (in case M is an ideal) there exists iµ, 1 ≤ iµ < s : M(iµ) ⋄ M(s) = M(iµ, s) = µ. For each µ := M( j, s) ∈ M \ M′ let iµ, 1 ≤ iµ < s, be such that µ = M(iµ, s). Then GM := GM∗ ∪ {{iµ, s} : µ ∈ M \ M′} is a useful S-pair set for G. 55 Proof. 1. (cf. [30, Lemma 25.1.4]) One has M(i, j, k) M(k, j) ⋄ b(k, j) = − + M(i, j, k) M(i, k) ⋄ b(i, k) − M(i, j, k) M(i, k) ⋄ M(i, k) M(i, j) ⋄ M(i, j) M(k, j) ⋄ M(k, j) M(i, j, k) M(i, j, k) M(k) ⋄ ek − M( j) ⋄ e j − M(i, j) M(i, k) M(k, j) M(i, j, k) M(i, j) ⋄ b(i, j) + M(i) ⋄ ei! M(i) ⋄ ei! M(k) ⋄ ek! ⋄ ei! ⋄ ei! ⋄ ek! M(i) M(i, j, k) M(i) M(i, j, k) M(k) M( j) ⋄ e j − M(i, j, k) M(k) = M(i, j, k) − M(i, j, k) + M(i, j, k) M( j) M( j) ⋄ ek − ⋄ e j − ⋄ e j − = 0 2. (cf. [30, Lemma 25.1.8]) In order to prove the claim by induction, it is sufficient to show that, for each redundant {i, j}, 1 ≤ i < j ≤ u, eli = el j =: ǫ, there are – {i1, j1}, . . . , {iρ, jρ}, . . . , {ir, jr}, 1 ≤ iρ < jρ ≤ u, eliρ – elements t1, . . . , tr ∈ T , – and coefficients c1, . . . cr ∈ R \ {0} = el jρ = ǫ, such that – b(i, j) = Pρ cρtρ ⋄ b(iρ, jρ); – τ(i, j) = tρ ◦ τ(iρ, jρ), for each ρ; – {iρ, jρ} ≺ {i, j}. In order to show this, we only need to consider the representation b(i, j) = M(i, j, k) M(i, k) ⋄ b(i, k) − M(i, j, k) M(k, j) ⋄ b(k, j) and to prove that {i, k} ≺ {i, j} ≻ {k, j}; this happens (according to the two cases of the definition) because (a) τ(i, k) τ(i, j, k) = τ(i, j) , τ(i, k) implies {i, k} ≺ {i, j} and the same argu- ment proves { j, k} ≺ {i, j}; (b) the same argument as that above proves { j, k} ≺ {i, j}, while {i, k} ≺ {i, j} because τ(i, k) ≤ τ(i, j) and k < j. 56 3. (cf. [30, Lemma 25.1.9]) Let i < s, eli = els =: ǫ, µ := M(i, s). Then: iµ′ < s, {i, s} is redundant; – if there exists µ′ ∈ M such that M(iµ′ , s) = µ′ M(i, s) , µ′, then since – if i = iµ and M(iµ) ⋄ M(s) = M(iµ, s), then (M is an ideal) bW(iµ, s) ∈ Cs so that BW(iµ, s) has a restricted Grobner representation in terms of G by Buchberger's First Criterion; – if i = iµ and M(iµ) ⋄ M(s) , M(im, s) then {im, s} ∈ GM; – if i , iµ then b(i, s) = M(i, iµ)b(i, iµ) − b(iµ, s). ⊓⊔ Corollary 75. With the present notation, under the assumption that R is a principal ideal domain, F is a restricted Grobner basis of M iff M(i,iµ,s) ⋄ 1. each S-polynomial BW(i, j), {i, j} ∈ R, has a restricted Grobner representation in terms of F; T(g j), have a restricted representation in terms of F w.r.t. <. 2. for each g j ∈ F, both Yi ⋆ g j, 1 ≤ i ≤ n and each g j ⋆ αυ(v), v ∈ v, υ ∈ T , υ < ⊓⊔ 8 Structural Theorem for Multivariate Ore Extensions of Zacharias PIDs Theorem 76 (Structural Theorem). Let R be a left Zacharias principal ideal domain, R := R[Y1, . . . , Yn] a multivariate Ore extension of R, < a term-ordering, M ⊂ Rm a left module generated by a basis F := {g1, . . . , gu} ⊂ M, M(gi) = ciτieli, C(F) a completion of F, R := {B(i, j), 1 ≤ i < j ≤ u, eli = el j and not redundant}. Then the following conditions are equivalent: (1). F is a left Grobner basis of M; (1s). C(F) is a left strong Grobner basis of M; (2). B(F) := {λg : λ ∈ T , g ∈ F} is a Gauss generating set [30, Definition 21.2.1]; (3). f ∈ M ⇐⇒ it has a left Grobner representation in terms of F; f ∈ M ⇐⇒ it has a left strong Grobner representation in terms of C(F); (4). (5). for each f ∈ Rm \ {0} and any normal form h of f w.r.t. F, we have f ∈ M ⇐⇒ h = 0; (5s). for each f ∈ Rm \ {0} and any strong normal form h of f w.r.t. C(F), we have f ∈ M ⇐⇒ h = 0; 57 (6). for each f ∈ Rm\{0}, f −Can( f , M) has a strong Grobner representation in terms of C(F); (7). each B(i, j) ∈ R has a weak Grobner representation in terms of F; (8). for each element σ of a Gebauer–Moller set for F, the S-polynomial SL(σ) has a left quasi-Grobner representation in terms of F. Proof. (1) ⇐⇒ (1s) is Proposition 56; (1) ⇐⇒ (2) is trivial; (1) ⇐⇒ (5) ⇐⇒ (3) is Proposition 19; (1s) ⇐⇒ (4) ⇐⇒ (5s) is Proposition 19; (1) =⇒ (6) is the content of Section 6.3; (6) =⇒ (4) because for each f ∈ M, Can( f , M) = 0; (1) ⇐⇒ (7) is Corollary 51; (1) ⇐⇒ (3) ⇐⇒ (8) is Theorem 33. 9 Spear's Theorem For Grobner bases in a ring A given as quotient ⊓⊔ Π : Q := ZhY1, . . . , Yni ։ A (cid:27) Q/I, I := ker(Π) of a free assocative algebra, a general approach is to directly apply Spear's Theorem [48] [30, Proposition 24.7.3] [28], which, while not a tool for computation, can be helpful in order to understand the structure of A. For the present setting, denoting – fi j := Y jYi − α j(Yi)Y j − δ j(Yi), 1 ≤ i < j ≤ n, – C := { fi j : 1 ≤ i < j ≤ n}; – I := I2(C) and for each m ∈ N, – {e1, . . . , em} the canonical basis of Rm, – C(m) := { fi jι := fi jeι : 1 ≤ i < j ≤ n, 1 ≤ ι ≤ m}, 58 we have the presentation R = Q/I, I := ker(Π), Π : Q := RhY1, . . . , Yni ։ R and, for each free R-module Rm, m ∈ N, the projection Π extends to the canonical projections, still denoted Π, Π : Qm ։ Rm, ker(Π) = Im = I2(C(m)). Thus denoting – F := {g1, . . . , gu} ⊂ Rm, gi = M(gi) − pi =: ciτieli − pi, – M ⊂ Rm the module M := I2(F), – M′ := Π−1(M) = M + Im ⊂ Qm, we can reformulate Spear's result as Lemma 77. [28, Lemma 12] Assume F ⊂ M′ is a Grobner basis of M′ and denote ¯F := {Can(g, Im) : g ∈ F, T(g) ∈ T (m)} ⊂ R[Y1, . . . , Yn]m where Can(g, Im) denotes the canonical form of g ∈ Qm w.r.t. C(m) so that in particular g = Π(g) for each g ∈ ¯F. Then ¯F ⊔ C(m) is a Grobner basis of M′. Theorem 78 (Spear). [28, Theorem 13] With the present notation, the following holds: 1. if F is a reduced Grobner basis of M′, then {g ∈ F : g = Π(g)} = {Π(g) : g ∈ F, T(g) ∈ T (m)} = F ∩ R[Y1, . . . , Yn]m is a reduced Grobner basis of M; 2. if F ⊂ R[Y1, . . . , Yn]m, so that in particular Π( f ) = f for each f ∈ F, is the Grobner basis of M, then F ⊔ C(m) is a Grobner basis of M′. 3. Assume each m′ ∈ M′ has a Grobner representation in terms of F ⊂ M′. Set ¯F := {Can(g, Im) : g ∈ F, g < Im} ⊂ R[Y1, . . . , Yn]m where Can(g, Im) ∈ R[Y1, . . . , Yn]m denotes the canonical form of g ∈ Qm w.r.t. C(m) so that in particular g = Π(g) for each g ∈ ¯F. Then each m ∈ M has a Grobner representation in terms of ¯F. 4. if F ⊂ R[Y1, . . . , Yn]m, so that in particular Π( f ) = f for each f ∈ F, is such that each m ∈ M has a Grobner representation in terms of F, then each m′ ∈ M′ has a Grobner representation in terms of F ⊔ C(m). 59 Corollary 79. [28, Corollary 14] With the present notation and considering – the bilateral R-module(cid:0)R ⊗ R Rop(cid:1)u with canonical basis {e1, . . . , eu} – the bilateral Q-module(cid:0)Q ⊗ R Qop(cid:1)F+mG with canonical basis {e1, . . . , eu} ⊔ {ei jι : 1 ≤ i < j ≤ n, 1 ≤ ι ≤ m}, – the projections S2 : (cid:0)R ⊗ R Rop(cid:1)F → Rm : S2(ei) = gi, 1 ≤ i ≤ u, and – S2 : (cid:0)Q ⊗ R Qop(cid:1)F+mC → Qm : S2(ei) = gi, 1 ≤ i ≤ u, S2(ei jι) = fi jι, 1 ≤ i < j ≤ n, 1 ≤ ι ≤ m, – the map ¯Π : (cid:0)Q ⊗ R Qop(cid:1)F+mC → (cid:0)R ⊗ R Rop(cid:1)F (where each λ, ρ ∈ RhY1, ..., Yni, a, b ∈ R \ {0}) ¯Π Xi aiλieibiρi +Xi jι ai jιλi jιei jιbi jιρi jι aiΠ(λi)eibiΠ(ρi), = Xi if Σ ⊂ (cid:0)Q ⊗ R Qop(cid:1)F+mC is a bilateral standard basis of ker( S2), then ¯Π(Σ) is a bilateral standard basis of ker(S2). 10 Lazard Structural Theorem for Ore Extensions over a Principal Ideal Domain Let D be a commutative principal ideal domain, R := D[Y; α, δ] be an Ore extension and I ⊂ R be a bilateral ideal. ordered so that Let F := { f0, f1, . . . , fk} be a reduced minimal strong bilateral Grobner basis of I deg( f0) ≤ deg( f1) ≤ · · · ≤ deg( fk) and let us denote, for each i, ci := lc( fi), ri ∈ D \ {0} and pi ∈ R the content3 and the primitive of fi so that fi = ri pi; denoting P := p0 the primitive part of f0 and Gk+1 := rk ∈ D \ {0} the content of fk we have Theorem 80. With the present notation, for each i, 0 ≤ i < k, there is Hi+1 ∈ R, d(i) := deg(Hi) and Gi ∈ D \ {0} such that – f0 = G1 · · · Gk+1P, – f j = G j+1 · · · Gk+1H jP, 1 ≤ j ≤ k, 3Defined here as the greatest common divisor of the coefficients of fi in the principal ideal domain D . 60 and 1. 0 < d(1) < d(2) < · · · < d(k); 2. Gi ∈ D, 1 ≤ i ≤ k + 1, is such that ci−1 = Gici 3. P = p0 (the primitive part of f0 ∈ R); 4. Hi ∈ R is a monic polynomial of degree d(i), for each i; 5. Hi+1 ∈ (G1 ··· Gi, G2 ··· GiH1, . . . , G j+1 ··· GiH j, . . . , Hi−1Gi, Hi) for all i. 6. ri = Gi+1 · · · Gk. Proof. Let P and Gk+1 be, resp., the greatest common right divisor of {p0, . . . , pk} in R and the greatest common divisor of {r0, . . . , rk} in D; since a set {g0, . . . , gk} is a minimal strong Grobner basis if and only if the same is true for {rg0g, . . . , rgkg}, r ∈ D, g ∈ R, we can left divide by Gk+1 and right divide by P and assume wlog that P = Gk+1 = 1 and that both the greatest common right divisor of {p0, . . . , pk} and the greatest common divisor of {r0, . . . , rk} are 1. Setting d(i) := deg( fi) and ν(i) := d(i + 1) − d(i) for each i, by assumption we have d(i) ≤ d(i + 1). If d(i) = d(i + 1), let us define h := bi fi + bi+1 fi+1 ∈ I where c, bi, bi+1 ∈ D are such that bici + bi+1ci+1 = c, c being the greatest common divisor of ci and ci+1, so that cYd(i+1) = M(h) ∈ M(I); this implies the existence of j such that M( f j) M(h) M( fi+1) contradicting minimality; thus d(i) < d(i + 1) and this, in turn, implies (1) since d(i) = d(i) − deg(P). Both fiY ν(i) and fi+1 are in the ideal and have degree d(i + 1). Therefore, for c, bi, bi+1 ∈ D such that bici + bi+1ci+1 = c, c being the greatest common divisor of ci and ci+1, h := bi fiYd(i+1)−d(i) + bi+1 fi+1 ∈ I, so that cYd(i+1) = M(h) ∈ M(I) and M( f j) M(h) for some j. If ci+1 , c, necessarily deg( f j) < deg( fi+1) whence j < i + 1 and M( f j) M(h) M( fi+1) getting a contradiction. As a conclusion ci = Gi+1ci+1, for some Gi+1 ∈ D and (2). Since Gi+1 fi+1 − fiY ν(i) is a polynomial of degree less than d(i + 1) which reduces to zero by the Grobner basis, it follows that Gi+1 fi+1 ∈ I( f0, . . . , fi) for each i, 0 ≤ i < k; thus, inductively we obtain p0 R f j for each j ≤ i =⇒ p0 R f j for each j ≤ i + 1. Also ci L f j for each j ≤ i =⇒ Gi+1ci+1 = ci L Gi+1 fi+1 =⇒ ci+1 L f j for each j ≤ i + 1. 61 Therefore, the assumptions that the greatest common right divisor of {p0, . . . , pk} and the greatest common divisor of {r0, . . . , rk} are 1 imply that p0 = ck = 1 proving f0 and this is sufficient to deduce, by the (3); thus in particular f0 = c0 so that c0 inductive argument, that each ci left-divides fi and therefore coincides with ri. Inductively we obtain ri lc(P) = ci = Gi+1ci+1 = Gi+1ri+1 lc(P) = Gi+1 · · · Gk lc(P) thus proving (6); defining Hi the polynomial s.t. ciHiP = fi for all i we have lc(Hi) = 1 (proving (4)), d(i) + deg(P) = deg( fi) and Gi+1 fi+1 ∈ ( f0, . . . , fi) which proves (5) dividing out Gi+1 · · · Gk. ⊓⊔ A The PIR case While an understandable timor restrained us to violate Ore's tabu requiring degree preservation of product, it is well-known that Zacharias–Moller results are naturally stated for polynomials over PIRs and the restriction to PIDs is unnatural; we therefore sketch here the few modifications to the theory which are required in order to adapt it to Ore extensions R over a PIR R. The first delicate adaptation is required by formula (4); the natural solution is due to Gateva [15, 16, 17] which considered valuation over the semigroup with zero T ∪{o} instead of T setting ◦ : N × N → N ∪ {o} : u ◦ v =  Her theory however apply only to domains. T(u ⋆ v) u ⋆ v , 0 u ⋆ v = 0. o Thus in order to extend Corollary 15 we need to reformulate it as Corollary 15. If ≺ is a term ordering on T and < is a ≺-compatible term ordering on T (m), then, for each l, r ∈ R and f ∈ R(m), 1. M(l ⋆ f ) = M(M(l) ⋆ M( f )) provided lc(l) lc( f ) , 0; 2. M( f ⋆ r) = M(M( f ) ⋆ M(r)) provided lc( f ) lc(r) , 0; 3. M(l ⋆ f ⋆ r) = M(M(l) ⋆ M( f ) ⋆ M(r)) provided lc(l) lc( f ) lc(r) , 0. 4. T(l ⋆ f ) ≤ T(l) ◦ T( f ) equality holding provided that lc(l) lc( f ) , 0; 5. T( f ⋆ r) ≤ T( f ) ◦ T(r) equality holding provided that lc( f ) lc(r) , 0; 6. T(l⋆ f ⋆r) ≤ T(l)◦T( f )◦T(r) equality holding provided that lc(l) lc( f ) lc(r) , 0. If, moreover, R is a domain, then 7. T(l ⋆ f ) = T(l) ◦ T( f ); 8. T( f ⋆ r) = T( f ) ◦ T(r); 62 9. T(l ⋆ f ⋆ r) = T(l) ◦ T( f ) ◦ T(r). As regards Grobner basis computation we remark that the first and the third algo- rithms (Section 6.1 and 6.3 ) apply verbatim also in the PIR case; in the algorithm in fact we have {i} ∈ H(F) for each i, 1 ≤ i ≤ u and thus each m := T(gi) ∈ H is treated by the algorithm which (if the basis is minimal) produces also the annihilitator syzygy a j 0 where we denote, for each i ≤ u, ai ∈ R the annihilator of I(ci). (d1, . . . , du)) ∈ SyzL(v(m)1, . . . , v(m)u), d j :=  In the second algorithm (Section 6.2) the inductive seed becomes if j = i otherwise S1 = {b ∈ R : b lc(g1) = 0} = I(a1) ⊂ R, Therefore and, for each s, 1 < s ≤ u, {s} is basic for T(gs) provided the basis is minimal. Corollary 49. Assuming that the Zacharias ring R is principal and denoting, for each i, j, 1 ≤ i < j ≤ u, eli = el j, b(i, j) := − B(i, j) := α{i, j}, j(c j) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) lcm(cid:16)α{i, j},i(ci), α{i, j}, j(c j)(cid:17) α{i, j}, j(c j) α{i, j},i(ci) α{i, j},i(ci) lcm(τi, τ j) τ j lcm(τi, τ j) τi e j ei, lcm(τi, τ j) τ j lcm(τi, τ j) τi ⋆ g j ⋆ gi − := aiei := ai ⋆ gi, a(i) A(i) we have that {b(i, j) : 1 ≤ i < j ≤ u, eli = el j} ∪ {a(i), i ≤ u} is a Gebauer–Moller set for F, so that F is a Grobner basis of M, iff each B(i, j), 1 ≤ i < j ≤ u, eli = el j , and each A(i), i ≤ u, has a weak Grobner representation in terms of F. ⊓⊔ Corollary 51. Assuming that the Zacharias ring R is principal and that each αi is an automorphism denoting, for each i, j, 1 ≤ i < j ≤ u, eli = el j, b(i, j) B(i, j) a(i) A(i) c j := e jα−1 := g j ⋆ α−1 τ j lcm(ci, c j) τ j lcm(ci, c j) := eiα−1 τi (ai), := gi ⋆ α−1 τi (ai), c j τ j ! lcm(τi, τ j) ! lcm(τi, τ j) τ j − eiα−1 ci τi lcm(ci, c j) − gi ⋆ α−1 τi lcm(ci, c j) ci ! lcm(τi, τ j) τi ! lcm(τi, τ j) τi 63 we have that {b(i, j) : 1 ≤ i < j ≤ u, eli = el j} ∪ {a(i), i ≤ u} is a Gebauer–Moller set for F, so that F is a right Grobner basis of M, iff each B(i, j), 1 ≤ i < j ≤ u, eli = el j , and each A(i), i ≤ u, has a right weak Grobner representation in terms of F. ⊓⊔ Corollary 58. With the present notation, under the assumption that R is a principal ideal ring, F is a Grobner basis of I iff, denoting GM a useful S-pair set for F, each S- polynomial B(i, j), {i, j} ∈ GM, and each A(i), 1 ≤ i ≤ u, has a Grobner representation in terms of F. ⊓⊔ Corollary 61. With the present notation, under the assumption that R is a principal ideal domain, F is a Grobner basis of M iff each S-polynomial B(i, j), {i, j} ∈ R, and each A(i), 1 ≤ i ≤ u, has a Grobner representation in terms of F. ⊓⊔ Corollary 62. With the present notation, under the assumption that R is a principal ideal ring, F is a Grobner basis of M iff 1. 2. each S-polynomial BW(i, j), {i, j} ∈ R, and each A(i), 1 ≤ i ≤ u, has a restricted Grobner representation in terms of F; T(g j), have a restricted representation in terms of F w.r.t. <. 3. for each g j ∈ F, both Yi ⋆ g j, 1 ≤ i ≤ n and each g j ⋆ αυ(v), v ∈ v, υ ∈ T , υ < ⊓⊔ Finally we remark that a Lazard Structural Theorem for Ore Extensions over a Principal Ideal Domain can be easily obtained by adapting the result given by Norton– Salagean [35, 36], [30, § 33.3] for polynomial rings. References [1] Apel J., Grobnerbasen in Nichetkommutativen Algebren und ihre Anwendung, Dissertation, Leipzig (1988) [2] Apel J., Computational ideal theory in finitely generated extension rings, Theor. Comp. Sci. 224 (2000), 1–33 [3] Apel J., Lassner, W., An Algorithm for calculations in enveloping fields of Lie al- gebras, In: Proc. Int. Conf. on Comp. Algebra and its Appl. n Theoretical Physics JINR D11-85-792, Dubna (1985) 231–241 [4] Apel J., Lassner, W., Computation and Simplification in Lie fields, L. N. Comp. Sci. 378 (1987), 468–478, Springer [5] Bergman G.H., The Diamond Lemma for Ring Theory, Adv. Math. 29 (1978), 178–218 64 [6] E. Byerne, Grobner bases over commutative rings and Applications to coding theory in M. Sala et al. (Ed.) Grobner bases, Coding, Cryptography, Springer Risc XVI, (2009). 239–262 [7] Buchberger B., Ein Algorithmus zum Auffinden der Basiselemente des Restk- lassenringes nach einem nulldimensionalen Polynomideal, Ph. D. Thesis, Inns- bruck (1965) [8] Buchberger B., Ein algorithmisches Kriterium fur die Losbarkeit eines algebrais- chen Gleischunssystem, Aeq. Math. 4 (1970), 374–383 [9] Buchberger B., A Critorion for Detecting Unnecessary Reduction in the Con- struction of Grobner bases, L. N. Comp. Sci 72 (1979), 3–21, Springer [10] Buchberger B., Grobner Bases: An Algorithmic Method in Polynomial Ideal The- ory, in Bose N.K. (Ed.) Multidimensional Systems Theory (1985), 184–232, Rei- der [11] J. Bueso, J. Gomez-Torrecillas, and A. Verschoren. Methods in Non-Commutative Algebra (2003). Kluwer [12] Chyzak F., Salvy B. Non-commutative Elimination in Ore Algebras Proves mul- tivariate Identities J. Symb. Comp. 26 (1998), 187–227 [13] Cohn P.M., Noncommutative unique factorization domains Trans. A.M.S. 109 (1963), 313–331 [14] Cohn P.M., Ring with a week Algorithm Trans. A.M.S. 109 (1963), 332–356 [15] Gateva–Ivanova T., Groebner bases in skew polynomial rings, J. Algebra 138 (1991) 13–35 [16] Gateva–Ivanova T., Noetherian Properties of Skew Polynomial Rings with Bino- mial Relations, Trans. A.M.S. 345 (1994), 203–219, [17] Gateva–Ivanova T., Skew polynomial rings with binomial relations, J. Algebra 185 (1996) 710–753 [18] Gebauer R., Moller H.M., On an Installation of Buchbgerger's Algorithm, J. Symb. Comp. 6, (1988), 275–286 [19] Janet M. , Sur les syst`emes d'´equations aux d´eriv´ees partielles J. Math. Pure et Appl., 3 (1920), 65–151 [20] Kandri-Rody A., Kapur, D. Computing the Grobner basis of an ideal in polyno- mail rings over a Euclidean ring J. Symb. Comp. 6 (1990), 37–56 [21] Kandri-Rody, A., Weispfenning, W., Non-commutativer Grobner Bases in Alge- bras of Solvable Type, J. Symb. Comp. 9 (1990), 1–26 [22] Kredel, H. Solvable Polynomial rings Dissertation, Passau (1992) 65 [23] LaScala R., Levandovskyy V. Skew Polynomial Rings, Grobner bases and the Letterplace embedding of the free Associativ algebra, J. Symb. Comp. 48 (2013), 110–131 [24] Lazard D., Solving zero-dimensional algebraic systems J. Symb. Comp. 15 (1992), 117–132 [25] Levandovskyy V. G., Non-commutative Computer Algebra for Polynomial Al- gebras: Grobner Bases, Applications and ImplementationDissertation, Kaiser- slautern (2005) http://kluedo.ub.uni-kl.de/volltexte/2005/1883/ [26] Levandovskyy V. G., PBW Bases, Non-Degeneracy Conditions and Applications In: Buchweitz, R.-O., Lenzing, H. (Eds.), Representation of Algebras and Re- lated Topics (Proceedings of the ICRA X Conference),45. AMS. Fields Institute Communications,pp.229-246. [27] Mansfield E.L.,Szanto A. Elimination Theory for Differential Difference Poly- nomnials Proc. ISSAC 2002 (2002), 283–290, ACM [28] F. Mora, De Nugis Groebnerialium 4: Zacharias, Spears, Moller Proc. ISSAC'15 (2015), 191–198, ACM [29] T. Mora, Seven variations on standard bases, (1988) ftp://ftp.disi.unige.it/person/MoraF/PUBLICATIONS/7Varietions.tar.gz [30] T. Mora Solving Polynomial Equation Systems II: Macaulay's Paradigm and Grobner Technology, Cambridge University Press (2005) [31] T. Mora Zacharias Representeation of Effective Associative Rings, J. Symb. Comp. (submitted) [32] Mosteig E., Sweedler M. Valuations and filtrations, J. Symb. Comp. 34 (2002), 399–435 [33] Moller H.M., New constructive methods in classical ideal theory, J. Algebra 100 (1986) 138–178 [34] Moller H.M., On the construction of Grobner bases using syzygies, J. Symb. Comp. 6 (1988), 345–359 [35] Norton G.H., Salagean A., Strong Grobner bases for polynomials over a principal ideal ring, Bull. Austral. Math. Soc. 64 (2001), 505–528 [36] Norton G.H., Salagean A., Grobner bases and products of coefficient rings, Bull. Austral. Math. Soc. 65 (2002), 147–154 [37] Ore O., Linear equations in non-commutative fields , Ann. Math. 32 (1931), 463– 477 66 [38] Ore O., Theory of non-commutative polynomials , Ann. Math. 34 (1933), 480– 508. [39] Pan L., On the D-bases of polynomial ideals over principal ideal domains, J. Symb. Comp. 7 (1988), 55–69 [40] Pesch M., Grobner Bases in Skew Polynomial Rings Dissertation,Passau (1997) [41] Pesch M., Two-sided Grobner bases in Iterated Ore Extensions, Progress in Com- puter Science and Applied Logic 15 (1991), 225–243, Birkhauser [42] Pritchard F. L., A syzygies approach to non-commutative Grobner bases, Preprint (1994) [43] Pritchard F. L., The ideal membership problem in non-commutative polynomial rings, J. Symb. Comp. 22 (1996), 27–48 [44] Reinert B., A systematic Study of Grobner Basis Methods, Habilitation, Kaiser- slautern (2003) [45] Reinert B., Grobner Bases in Function Ring – A Guide for Introducing Reduction Relations to Algebraic Structures, J. Symb. Comp. 41 (2006), 1264–94 [46] Schreyer F.O., Die Berechnung von Syzygien mit dem verallgemeinerten Weier- strass'schen Divisionsatz, Diplomarbait, Hamburg (1980) [47] Schreyer F.O., A standard basis approach to syzygies of canonical curves, J. Reine angew. Math. 421 (1991), 83–123 [48] Spear D.A., A constructive approach to commutative ring theory, in Proc. of the 1977 MACSYMA Users' Conference, NASA CP-2012 (1977), 369–376 [49] Sweedler M. Ideal bases and valuation rings, Manuscript (1986) available at http://math.usask.ca/fvk/Valth.html [50] Szekeres L., A canonical basis for the ideals of a polynomial domain, Am. Math. Monthly 59 (1952), 379–386 [51] Tamari D. On a certain Classification of rings and semigroups Bull. A.M.S. 54 (1948), 153–158 [52] Weispfenning, V. Finite Grobner bases in non-noetherian Skew Polynomial Rings Proc. ISSAC'92 (1992), 320–332, A.C.M. [53] Zacharias G., Generalized Grobner bases in commutative polynomial rings, Bachelor's thesis, M.I.T. (1978) 67
1709.00647	1	1709	2017-09-03T00:47:58	Transferring Davey`s Theorem on Annihilators in Bounded Distributive Lattices to Modular Congruence Lattices and Rings	[ "math.RA" ]	Congruence lattices of semiprime algebras from semi--degenerate congruence--modular varieties fulfill the equivalences from B. A. Davey`s well--known characterization theorem for $m$--Stone bounded distributive lattices, moreover, changing the cardinalities in those equivalent conditions does not change their validity. I prove this by transferring Davey`s Theorem from bounded distributive lattices to such congruence lattices through a certain lattice morphism and using the fact that the codomain of that morphism is a frame. Furthermore, these equivalent conditions are preserved by finite direct products of such algebras, and similar equivalences are fulfilled by the elements of semiprime commutative unitary rings and, dualized, by the elements of complete residuated lattices.	math.RA	math	Transferring Davey's Theorem on Annihilators in Bounded Distributive Lattices to Modular Congruence Lattices and Rings Claudia MURES¸AN∗ University of Bucharest Faculty of Mathematics and Computer Science Academiei 14, RO 010014, Bucharest, Romania E -- mails: [email protected], [email protected] October 20, 2018 Abstract Congruence lattices of semiprime algebras from semi -- degenerate congruence -- modular varieties fulfill the equivalences from B. A. Davey's well -- known characterization theorem for m -- Stone bounded distributive lat- tices; moreover, changing the cardinalities in those equivalent conditions does not change their validity. I prove this by transferring Davey's Theorem from bounded distributive lattices to such congruence lattices through a certain lattice morphism and using the fact that the codomain of that morphism is a frame. Fur- thermore, these equivalent conditions are preserved by finite direct products of such algebras, and similar equivalences are fulfilled by the elements of semiprime commutative unitary rings and, dualized, by the ele- ments of complete residuated lattices. 2010 Mathematics Subject Classification: primary: 08A30; secondary: 08B10, 03G10. Keywords: congruence, (semi -- degenerate, congruence -- modular, congruence -- distributive) variety, commuta- tor, annihilator, (Stone, semiprime) algebra. 1 Introduction In [15, 16], I have transferred [5, Theorem 1] from bounded distributive lattices to residuated lattices, by using the reticulation of a residuated lattice. In [9], G. Georgescu and I have constructed the reticulation for algebras from congruence -- modular varieties. I have noticed that the kind of transfer from [15, 16] can be made in this general context, but referring to the lattices of congruences of the algebras from such varieties instead of their elements and enforcing some restrictions: the varieties must be semi -- degenerate and the algebras in question must be semiprime. It turns out that the transfer of these properties doesn't even necessitate the reticulation, but only part of its construction from [9], and it produces further equivalences, because changing the cardinality in those conditions forms other properties which are equivalent to those conditions. The present work contains two main results, the first of which is Theorem 2.29 below, stating the equivalences fulfilled by the congruence lattices of such algebras, which I am proving along with some related results, such as the fact that, if the congruence lattice of such an algebra satisfies the equivalent conditions from Theorem 2.29, then the reticulation of that algebra satisfies those conditions, as well, and the fact that those conditions are preserved by finite direct products of such algebras. While the structure of the proof of Theorem 2.29 resembles the one I have made for residuated lattices in [15, 16], the auxiliary results for that proof do not hold in the present context, which, in turn, produces new auxiliary results, so all the following results are new and original, excepting only the ones cited from other works or mentioned as being either well known or immediate from well -- known properties. The natural question that arises is whether other algebras fulfill analogues of Theorem 2.29 or [5, Theorem 1] for elements instead of congruences, as is the case for residuated lattices. I prove that semiprime commutative unitary rings do, in the second main result of the present paper: Theorem 3.32. It also turns out that complete residuated lattices fulfill the dual of Theorem 2.29 expressed for elements, because, for such residuated lattices, ∗Dedicated to the memory of my dear grandmother, Floara -- Marioara Mure¸san changing the cardinalities in the equivalent conditions from the analogue of Davey's Theorem for residuated lattices from [15, 16] produces other properties equivalent to those conditions. Of course, both residuated lattices, which are congruence -- distributive, thus semiprime, and form a semi -- degenerate variety, and semiprime commutative unitary rings, which are semiprime algebras from the semi -- degenerate congruence -- modular variety of commutative unitary rings, fulfill those equivalences for congruences, too. Section 2 contains Theorem 2.29 and the related results on congruence lattices. Section 3 contains a brief presentation of the situation in residuated lattices, the analogue of Theorem 2.29 for the elements of semiprime commutative unitary rings instead of the congruences from congruence lattices of semiprime algebras from semi -- degenerate congruence -- modular varieties, and some related results. Section 4 contains a brief layout of some directions for future research. 2 Transferring Davey's Theorem to Congruence Lattices For any set S, S shall denote the cardinality of S and by S < ∞ we shall specify the fact that S is finite. And, throughout this paper, m shall be an infinite cardinal, arbitrary but fixed. For brevity, instead of treating separate cases, we shall often use the fact that, in any bounded poset, W∅ is the minimum and V∅ is the maximum of that poset. Throughout this paper, we shall designate any algebra by its underlying set, unless there is danger of confusion. u∈U x∈U ∪V u∈U v∈V Ann(u)∩ \ Ann(u), hence Ann(U )∩Ann(V ) = \ Let (L,∨,∧, 0, 1) be an arbitrary bounded lattice. (Id(L),∨,∩,{0}, L) shall be the complete lattice of the ideals of L, which is well known to be distributive exactly when L is distributive. Let a ∈ L and U, V ⊆ L. We shall denote by (U ] the ideal of L generated by U , by (a] = ({a}] the principal ideal of L generated by a, by Ann(U ) the annihilator of U and by Ann(a) the annihilator of a in L: Ann(U ) = {x ∈ L (∀ u ∈ U ) (x ∧ u = 0)} and Ann(a) = Ann({a}) = {x ∈ L x ∧ a = 0}. Whenever specifying L is necessary, we shall denote (U ]L, (a]L, AnnL(U ) and AnnL(a) instead of (U ], (a], Ann(U ) and Ann(a), respectively. Obviously, Ann(a) = Ann((a]), U ⊆ Ann(Ann(U )) and Ann(U ) = \ Ann(v) = \ Ann(x) = Ann(U ∪V ). Clearly, if U ⊆ V , then Ann(V ) ⊆ Ann(U ), hence Ann(Ann(U )) ⊆ Ann(Ann(V )). Obviously, for any x ∈ L, Ann((x]) = Ann(x). It is straightforward that, if L is distributive, then Ann(U ) is an ideal of L. Recall that L is called a compact lattice iff all its elements are compact. Notice that, in a compact lattice L, the join of any non -- empty family U ⊆ L equals the join of a finite non -- empty subfamily of U . We shall denote by B(L) the set of the complemented elements of the bounded lattice L. If L is distributive, then B(L) is the Boolean center of L, which is a Boolean sublattice of L. We shall call B(L) the Boolean center of L regardless of whether L is distributive. We shall call L a Stone lattice iff, for all a ∈ L, there exists an e ∈ B(L) such that Ann(a) = (e]. We shall call L a strongly Stone lattice iff, for all U ⊆ L, there exists an e ∈ B(L) such that Ann(U ) = (e]. Trivially, if L is strongly Stone, then L is Stone. Since Ann(U ) = \ Ann(u) for any U ⊆ L and \ converse holds if B(L) is closed w.r.t. arbitrary joins. Now let M be a bounded lattice and f : L → M be a surjective lattice morphism. Then it is straightforward that the map I 7→ f (I) is a surjective lattice morphism from Id(L) to Id(M ), which fulfills f ((a]) = (f (a)] for all a ∈ L. Moreover, this lattice morphism preserves arbitrary joins. Indeed, if (Ji)i∈I is a family of ideals of L, then f (_ Ji, then x ≤ x1 ∨ . . . ∨ xn for some f (Ji), ai] for any non -- empty family (ai)i∈I ⊆ L having a join, it follows that the f (Ji). Now, if x ∈ _ f (Ji), thus f (_ Ji) ⊇ f ([ n ∈ N, i1, . . . , in ∈ I and x1 ∈ Ji1 , . . . , xn ∈ Jin . Then f (x) ≤ f (x1)∨. . .∨f (xn) ∈ f (Ji1)∨. . .∨f (Jin ) ⊆ _ thus f (x) ∈ _ f (Ji). Therefore f (_ f (Ji), hence f (_ Ji) ⊆ _ Ji) = _ f (Ji). i∈I Ji) ⊇ _ i∈I (ai] = (_ Ji) = [ i∈I i∈I i∈I i∈I u∈U i∈I i∈I i∈I i∈I i∈I i∈I i∈I i∈I For the arbitrary bounded lattice L, let us denote by: Ann(L) = {AnnL(U ) U ⊆ L}, PAnn(L) = {AnnL(a) a ∈ L}, P2Ann(L) = {AnnL(Ann)L(a)) a ∈ L} and 2Ann(L) = {AnnL(AnnL(U )) U ⊆ L}. Regarding the conditions below on subsets of L, note that they are trivially fulfilled by ∅, because AnnL(∅) = L = (1]L and AnnL(AnnL(∅)) = AnnL(L) = {0} = (0]L and, clearly, 0, 1 ∈ B(L). Here, I am specifying L through indexes in the notations, for clarity. Let us consider the following conditions on L, where κ is an arbitrary nonzero cardinality; note that (1)m,L, (2)m,L, (3)m,L, (4)m,L, (5)m,L express the conditions from [5, Theorem 1] in a way that makes sense without L being assumed distributive: (3)<∞,L P2Ann(L) is a Boolean sublattice of Id(L) such that (1)κ,L (1)<∞,L (1)L (2)κ,L (2)<∞,L L is a Stone lattice and B(L) is a Boolean sublattice of L; (2)L (3)κ,L for each U ⊆ L with U ≤ κ, there exists an e ∈ B(L) such that AnnL(U ) = (e]L; for each finite U ⊆ L, there exists an e ∈ B(L) such that AnnL(U ) = (e]L; L is a strongly Stone lattice; L is a Stone lattice and B(L) is a κ -- complete Boolean sublattice of L; L is a Stone lattice and B(L) is a complete Boolean sublattice of L; P2Ann(L) is a κ -- complete Boolean sublattice of Id(L) such that a 7→ AnnL(AnnL(a)) is a lattice morphism from L to P2Ann(L); a 7→ AnnL(AnnL(a)) is a lattice morphism from L to P2Ann(L); P2Ann(L) is a complete Boolean sublattice of Id(L) such that a 7→ AnnL(AnnL(a)) is a lattice morphism from L to P2Ann(L); for all a, b ∈ L, AnnL(a ∧ b) = (AnnL(a) ∪ AnnL(b)]L, and, for each U ⊆ L with U ≤ κ, there exists an x ∈ L such that AnnL(AnnL(U )) = AnnL(x); for all a, b ∈ L, AnnL(a ∧ b) = (AnnL(a) ∪ AnnL(b)]L, and, for each finite U ⊆ L, there exists an x ∈ L such that AnnL(AnnL(U )) = AnnL(x); for all a, b ∈ L, AnnL(a ∧ b) = (AnnL(a) ∪ AnnL(b)]L, and, for each U ⊆ L, there exists an x ∈ L such that AnnL(AnnL(U )) = AnnL(x); for all a, b ∈ L, AnnL(a ∧ b) = (AnnL(a) ∪ AnnL(b)]L; for each U ⊆ L with U ≤ κ, (AnnL(U ) ∪ AnnL(AnnL(U ))]L = L; for each finite U ⊆ L, (AnnL(U ) ∪ AnnL(AnnL(U ))]L = L; for each U ⊆ L, (AnnL(U ) ∪ AnnL(AnnL(U ))]L = L. (3)L (4)κ,L (4)<∞,L (4)L (iv)L (5)κ,L (5)<∞,L (5)L Theorem 2.1. [5, Theorem 1] If L is a bounded distributive lattice, then the conditions (1)m,L, (2)m,L, (3)m,L, (4)m,L and (5)m,L are equivalent. Remark 2.2. [15, 16] Notice that, if we denote by L′ the dual of the bounded lattice L, then the duals of conditions (1)κ,L through (5)L above are simply conditions (1)κ,L′ through (5)L′, respectively, that, with respect to L, can be expressed through co -- anihilators and generated filters (see also [15, 16]). In the case when L is a bounded distributive lattice, so is L′, thus conditions (1)m,L′, (2)m,L′, (3)m,L′, (4)m,L′ and (5)m,L′ are equivalent, as well. And, of course, all the following properties on L in this paper hold for L′, too. Let us also note that, for any i ∈ 1, 5, we have the following: for any nonzero cardinalities κ, µ such that κ ≤ µ, (i)µ,L implies (i)κ,L, and thus (i)µ,L is equivalent to (i)ν,L being valid for all nonzero cardinalities ν ≤ µ; (i)L is equivalent to (i)κ,L being valid for all nonzero cardinalities κ, as well as to (i)κ,L being valid for all nonzero cardinalities κ greater than a cardinality µ; (i)<∞,L is equivalent to (i)κ,L being valid for all finite nonzero cardinalities κ, as well as to (i)κ,L being valid for all finite nonzero cardinalities κ greater than a finite cardinality µ. By the above and Theorem 2.1, we get that, if L is a bounded distributive lattice, then conditions (1)L, (2)L, (3)L, (4)L and (5)L are equivalent. Since, in any bounded lattice L, AnnL(U ) = \ AnnL(u) for each U ⊆ L and, for any family (xi)i∈I ⊆ L xi]L, it follows that, for any bounded lattice L, (2)L implies (1)L. The converse u∈U having a meet, \ (xi]L = (^ i∈I i∈I holds if B(L) is a complete Boolean sublattice of L, case in which, therefore, L is Stone iff L is strongly Stone, that is: for any finite nonzero cardinality κ, (1)L is equivalent to (1)κ,L and thus also to (1)<∞,L. Also, for any family (Ui)i∈I of subsets of L: AnnL([ AnnL(u) = \ AnnL(u) = \ Ui) = \ AnnL(Ui). \ i∈I u∈Si∈I Ui i∈I u∈Ui i∈I Now assume that L is distributive. Then, for all n ∈ N∗ and all u1, . . . , un ∈ L, un}) = AnnL( n _ ui). i=1 Indeed, for all k ∈ 1, n, uk ≤ n _ i=1 ui, hence AnnL( _ i=1 ui) ⊆ AnnL(uk), therefore n \ i=1 n AnnL(ui) = AnnL({u1, . . . , AnnL( n _ i=1 ui) ⊆ n \ i=1 AnnL(ui). If x ∈ n \ i=1 AnnL(ui), then x ∧ ( n _ ui) = i=1 n _ i=1 (x ∧ ui) = 0, thus x ∈ AnnL( n _ ui), so i=1 the converse inclusion holds, as well. Therefore FAnn(L) = PAnn(L), from which it is easy to see that, for any i ∈ 1, 5 and any finite nonzero cardinality κ, (i)κ,L is equivalent to (i)<∞,L. It is immediate that, for any finite nonzero cardinality κ, (1)κ,L, (2)κ,L, (3)κ,L, (4)κ,L and (5)κ,L are equivalent. If, moreover, L is a frame, then, for any ∅ 6= U ⊆ L, \ u∈U AnnL(u) = AnnL(U ) = AnnL( _ u). Indeed, for all u∈U u∈U u∈U u) = _ u, hence AnnL( _ u) ⊆ AnnL(a), therefore AnnL( _ a ∈ U , a ≤ _ then x ∧ ( _ in a frame L, Ann(L) = PAnn(L), from which it is easy to see that, for any i ∈ 1, 5 and any finite nonzero cardinality κ, (i)κ,L is equivalent to (i)L, and it also follows that 2Ann(L) = P2Ann(L) ⊆ Ann(L) = PAnn(L), so that the second part of condition (4)κ,L is fulfilled for any nonzero cardinality κ, and thus (4)κ,L, (4)<∞,L, (4)L and (iv)L are equivalent. (x ∧ u) = 0, thus x ∈ AnnL( _ u) ⊆ \ AnnL(u). If x ∈ \ u), so the converse inclusion holds, as well. Therefore, AnnL(u), u∈U u∈U u∈U u∈U u∈U u∈U Corollary 2.3. If L is a bounded distributive lattice, then: • conditions (1)L, (2)L, (3)L, (4)L and (5)L are equivalent; • for any nonzero cardinality κ, (1)κ,L, (2)κ,L, (3)κ,L, (4)κ,L and (5)κ,L are equivalent; • for any finite nonzero cardinality κ, (1)κ,L, (2)κ,L, (3)κ,L, (4)κ,L, (5)κ,L, (1)<∞,L, (2)<∞,L, (3)<∞,L, (4)<∞,L and (5)<∞,L are equivalent; • if, moreover, L is a frame, then, for any nonzero cardinality κ and any h, i, j ∈ 1, 5, (iv)L, (h)κ,L, (i)<∞,L and (j)L are equivalent; in particular, (1)L is equivalent to (2)<∞,L, so that: L is a Stone lattice iff L is a strongly Stone lattice. Throughout this paper, all algebras shall be non -- empty, C shall be a semi -- degenerate congruence -- modular equational class of algebras of the same type and A shall be an arbitrary algebra from C. (Con(A),∨,∩, ∆A,∇A) shall be the complete modular lattice of the congruences of A, with ∆A = {(a, a) a ∈ A} and ∇A = A2, so that the set of the proper congruences of A is Con(A)\{∇A}. [·,·]A : (Con(A))2 → Con(A) shall be the commutator of A. Recall that [·,·]A is commutative, smaller than the intersection, increasing in both arguments and distributive in both arguments with respect to arbitrary joins [6]. Following [6], if φ is a proper congruence of A, then we call φ a prime congruence iff, for all θ, ζ ∈ Con(A), [θ, ζ]A ⊆ φ implies θ ⊆ φ or ζ ⊆ φ. We shall denote by Spec(A) the set of the prime congruences of A. Since C is semi -- degenerate, it follows that C has no skew congruences [6, Theorem 8.5, p. 85] and, for any member M of C, ∇M is a compact congruence of M [12] and each proper congruence of M is included in a prime congruence [1, Theorem 5.3], thus, if M is non -- trivial, that is M > 1, then Spec(M ) is non -- empty. Recall that the compact congruences of an algebra are exactly its finitely generated congruences. Following [9], for all θ ∈ Con(A), we shall denote by ρA(θ) the radical of θ: ρA(θ) = T{φ ∈ Spec(A) θ ⊆ φ}, for any θ, ζ ∈ Con(A), θ ≡A ζ iff ρA(θ) = ρA(ζ). and by ≡A the binary relation on Con(A) defined by: We have proven, in [9], that ≡A is an equivalence on Con(A), and we have denoted by λA : Con(A) → Con(A)/≡A the canonical surjection. Moreover, we have proven that ≡A is a congruence of the lattice Con(A), thus (Con(A)/≡A,∨,∧, λA(∆A), λA(∇A)) is a bounded lattice and λA is a surjective lattice morphism, where λA(θ) ∨ λA(ζ) = λA(θ ∨ ζ) and λA(θ) ∧ λA(ζ) = λA(θ ∩ ζ) for all θ, ζ ∈ Con(A). We have also proven that λA(θ ∩ ζ) = λA([θ, ζ]A) for all θ, ζ ∈ Con(A), from which, by using the distributivity of the commutator with respect to the join, it immediately follows that the bounded lattice Con(A)/≡A is distributive. Moreover, since Con(A) is a complete lattice and [·,·]A is distributive w.r.t. arbitrary joins, it follows that Con(A)/≡A is a complete lattice in which the meet is distributive w.r.t. arbitrary joins, that is Con(A)/≡A is a frame (see also [9]). Hence, from Corollary 2.3, we obtain: Corollary 2.4. For any nonzero cardinality κ and any h, i, j ∈ 1, 5, (h)κ,Con(A)/≡A (j)Con(A)/≡A are equivalent. , (i)<∞,Con(A)/≡A and We have proven in [9] that: • for any θ ∈ Con(A), λA(θ) = λA(∇A) iff θ = ∇A; indeed, λA(θ) = λA(∇A) iff ρA(θ) = ρA(∇A) = T∅ = ∇A iff no prime congruence of A includes θ iff θ = ∇A; θ ∈ Con(A), thus A is semiprime (see, also, [9]). • B(Con(A)) is a Boolean sublattice of Con(A), in which the commutator coincides to the intersection. We call A a semiprime algebra iff ρA(∆A) = ∆A. If C is congruence -- distributive, then ρA(θ) = θ for all Throughout the rest of this paper, the algebra A shall be semiprime. Then, as we have proven in [9]: • (◦) • (1◦) • (2◦) • (3◦) B(Con(A)/≡A) = B(Con(A))/≡A ; λA B(Con(A)): B(Con(A)) → B(Con(A)/≡A) = B(Con(A))/≡A is a Boolean isomorphism; for all θ ∈ Con(A): λA(θ) ∈ B(Con(A)/≡A) iff θ ∈ B(Con(A)); for all θ ∈ Con(A), λA(θ) = λA(∆A) iff θ = ∆A; indeed, λA(θ) = λA(∆A) implies θ ⊆ ρA(θ) = ρA(∆A) = ∆A, thus θ = ∆A; the converse implication is trivial; • (4◦) for all θ, ζ ∈ Con(A), [θ, ζ]A = ∆A iff θ ∩ ζ = ∆A; indeed, by (3◦): [θ, ζ]A = ∆A iff λA([θ, ζ]A) = λA(∆A) iff λA(θ) ∧ λA(ζ) = λA(∆A) iff λA(θ ∩ ζ) = λA(∆A) iff θ ∩ ζ = ∆A. I shall make repeated use of the surjectivity of λA : Con(A) → Con(A)/≡A, without mentioning it; the same goes for the remarks from this paper and the results I am recalling from [9], excepting (1◦), (2◦), (3◦) and (4◦). By Corollary 2.3, in the particular case when Con(A) is distributive, conditions (1)Con(A), . . . , (5)Con(A) are equivalent, if κ is a nonzero cardinality, then conditions (1)κ,Con(A), . . . , (5)κ,Con(A) are equivalent, and, if κ is fi- nite, then conditions (1)κ,Con(A), . . . , (5)κ,Con(A), (1)<∞,Con(A), . . . , (5)<∞,Con(A) are equivalent. An example of a semi -- degenerate congruence -- modular variety which is not congruence -- distributive is the variety of commutative unitary rings [10]. Throughout the rest of this section, unless there is danger of confusion, all annihilators of elements or subsets of Con(A), respectively Con(A)/≡A, shall be in the lattice Con(A), respectively Con(A)/≡A, and the same shall go for generated ideals. Remark 2.5. By (4◦), for all θ ∈ Con(A) and all Ω ⊆ Con(A), Ann(θ) = {ζ ∈ Con(A) θ ∩ ζ = ∆A} = {ζ ∈ Con(A) [θ, ζ]A = ∆A} and Ann(Ω) = {ζ ∈ Con(A) (∀ ω ∈ Ω) (ω ∩ ζ = ∆A)} = {ζ ∈ Con(A) (∀ ω ∈ Ω) ([ω, ζ]A = ∆A)}. Remark 2.6. Any annihilator of Con(A) is an ideal of Con(A). Indeed, let ω ∈ Con(A). Then ∆A ∩ ω = ∆A, thus ∆A ∈ Ann(ω), so Ann(ω) 6= ∅. Now let θ, ζ ∈ Con(A). If ζ ∈ Ann(ω) and θ ⊆ ζ, then θ ∩ ω ⊆ ζ ∩ ω = ∆A, thus θ ∈ Ann(ω). If θ, ζ ∈ Ann(ω), then [ω, θ ∨ ζ]A = [ω, θ]A ∨ [ω, ζ]A = ∆A ∨ ∆A = ∆A, hence θ ∨ ζ ∈ Ann(ω). Therefore Ann(ω) ∈ Id(Con(A)). Hence, for any Ω ⊆ Con(A), Ann(Ω) = \ Ann(ω) ∈ Id(Con(A)). ω∈Ω i∈I i∈I Ωi) = (\ Remark 2.7. Of course, the direct image of λA : Con(A) → Con(A)/≡A preserves arbitrary unions of subsets of Con(A). It also preserves arbitrary intersections, because, if (Ωi)i∈I is a family of subsets of Con(A), then λA(\ λA(Ωi). Since λA : Con(A) → Con(A)/≡A is a surjective lattice morphism, it follows that the map I 7→ λA(I) is a surjective lattice morphism from Id(Con(A)) to Id(Con(A)/≡A) which fulfills λA((θ]) = (λA(θ)] for all θ ∈ Con(A) and preserves arbitrary joins. Lemma 2.8. (i) If L is a bounded distributive lattice, then AnnL(U ) = AnnL((U ]L) for all U ⊆ L. (Ωi/≡A) = \ Ωi)/≡A = \ i∈I i∈I (ii) For all Ω ⊆ Con(A), Ann(Ω) = Ann((Ω]). Proof. (i) U ⊆ (U ]L, thus AnnL((U ]L) ⊆ AnnL(U ). If x ∈ AnnL(U ) and u ∈ (U ]L, then u ≤ u1 ∨ . . . ∨ un for some n ∈ N and u1, . . . , un ∈ U , thus x ∧ u ≤ x ∧ (u1 ∨ . . . ∨ un) = (x ∧ u1) ∨ . . . ∨ (x ∧ un) = 0 ∨ . . . ∨ 0 = 0, thus x ∈ AnnL((U ]L) since u is arbitrary in (U ]L, so the converse inclusion holds, as well. (ii) Ω ⊆ (Ω], thus Ann((Ω]) ⊆ Ann(Ω). Now let θ ∈ Ann(Ω) and ω ∈ (Ω], so that ω ⊆ ω1 ∨ . . . ∨ ωn for some n ∈ N and ω1, . . . , ωn ∈ Ω, thus [θ, ω]A ⊆ [θ, ω1 ∨ . . . ∨ ωn]A = [θ, ω1]A ∨ . . . ∨ [θ, ωn]A = ∆A ∨ . . . ∨ ∆A = ∆A, thus θ ∈ Ann((Ω]) since ω is arbitrary in (Ω], so the converse inclusion holds, as well. Lemma 2.9. If L is a bounded distributive lattice, then: • for any family (Ik)k∈K ⊆ Id(L), \ k∈K AnnL(Ik) = AnnL( _ Ik); k∈K • for any family (ak)k∈K ⊆ L having a join, \ k∈K AnnL(ak) = AnnL( _ ak). k∈K Proof. By Lemma 2.8, (i), \ AnnL(Ik) = AnnL( [ Ik) = AnnL(( [ Ik]L) = AnnL( _ Ik), hence, if _ ak k∈K k∈K k∈K k∈K k∈K exists, then \ AnnL(ak) = \ AnnL((ak]L) = AnnL( _ (ak]L) = AnnL(( _ ak]L) = AnnL( _ ak). k∈K k∈K k∈K k∈K k∈K Lemma 2.10. • For any family (Ik)k∈K ⊆ Id(Con(A)), \ k∈K Ann(Ik) = Ann( _ Ik). k∈K • For any family (θk)k∈K ⊆ Con(A), \ k∈K Ann(θk) = Ann( _ θk). k∈K Proof. Same as the proof of Lemma 2.9, but using (ii) from Lemma 2.8 instead of (i). Lemma 2.11. • For all θ ∈ Con(A), Ann(λA(θ)) = λA(Ann(θ)). • For all Ω ∈ Con(A), Ann(λA(Ω)) = λA(Ann(Ω)). Proof. Ann(λA(θ)) = {λA(ζ) ζ ∈ Con(A), λA(θ) ∧ λA(ζ) = λA(∆A)} = {λA(ζ) ζ ∈ Con(A), λA(θ ∩ ζ) = λA(∆A)} = {λA(ζ) ζ ∈ Con(A), θ ∩ ζ = ∆A} = λA(Ann(θ)), by (3◦). Therefore λA(Ann(Ω)) = λA( \ Ann(λA(ω)) = \ Ann(x) = Ann(λA(Ω)). Ann(ω)) = \ λA(Ann(ω)) = \ ω∈Ω ω∈Ω ω∈Ω x∈λA(Ω) Lemma 2.12. For all θ ∈ Con(A) and α ∈ B(Con(A)), λA(θ) ≤ λA(α) iff θ ⊆ α. Proof. Assume that λA(θ) ≤ λA(α), so that λA(θ ∨ α) = λA(θ) ∨ λA(α) = λA(α) ∈ λA(B(Con(A))) = B(Con(A)/≡A) by (1◦), hence θ ∨ α ∈ B(Con(A)) by (2◦), and thus θ ∨ α = α, again by (1◦), so θ ⊆ α. The fact that λA is order -- preserving proves the converse implication. Lemma 2.13. For all Γ, Ω ⊆ Con(A), γ ∈ Con(A) and α ∈ B(Con(A)): (i) λA(γ) ∈ λA(Ann(Ω)) iff γ ∈ Ann(Ω); (ii) λA(Γ) ⊆ λA(Ann(Ω)) iff Γ ⊆ Ann(Ω); (iii) λA(Ann(Γ)) = λA(Ann(Ω)) iff Ann(Γ) = Ann(Ω); (iv) λA((α]) = λA(Ann(Ω)) iff (α] = Ann(Ω). Proof. (i) Assume that λA(γ) ∈ λA(Ann(Ω)) = Ann(λA(Ω)) according to Lemma 2.11, so that, for all ω ∈ Ω, λA(γ ∩ ω) = λA(γ) ∧ λA(ω) = λA(∆A), thus, by (3◦), γ ∩ ω = ∆A, hence γ ∈ Ann(Ω). The converse implication is trivial. (ii) By (i). (iii) By (ii), λA(Ann(Γ)) = λA(Ann(Ω)) iff λA(Ann(Γ)) ⊆ λA(Ann(Ω)) and λA(Ann(Ω)) ⊆ λA(Ann(Γ)) iff Ann(Γ) ⊆ Ann(Ω) and Ann(Ω) ⊆ Ann(Γ) iff Ann(Γ) = Ann(Ω). (iv) Assume that λA((α]) = λA(Ann(Ω)), which implies (α] ⊆ Ann(Ω) by (ii). Now let θ ∈ Ann(Ω), so that λA(θ) ∈ λA(Ann(Ω)) = λA((α]) = (λA(α)], hence θ ∈ (α] by Lemma 2.12, thus Ann(Ω) ⊆ (α], therefore (α] = Ann(Ω). The converse implication is trivial. Proposition 2.14. For any nonzero cardinality κ, the properties (1)κ,Con(A) and (1)κ,Con(A)/≡A are equivalent. Proof. For the converse implication, assume that (1)κ,Con(A)/≡A is satisfied and let ∅ 6= Ω ⊆ Con(A) having Ω ≤ κ. Then λA(Ω) ≤ Ω ≤ κ, hence, by (1◦), Lemma 2.11 and Lemma 2.13, (iv), there exists an α ∈ B(Con(A)) such that λA(Ann(Ω)) = Ann(λA(Ω)) = (λA(α)] = λA((α]), therefore Ann(Ω) = (α]. For the direct implication, assume that (1)κ,Con(A) is satisfied and let U ⊆ Con(A)/≡A with U ≤ κ. For each u ∈ U , there exists an ωu ∈ Con(A) such that λA(ωu) = u. If we denote by Ω = {ωu u ∈ U} ⊆ Con(A), then λA(Ω) = U and Ω = U ≤ κ, hence, by Lemma 2.11 and (1◦), Ann(U ) = Ann(λA(Ω)) = λA(Ann(Ω)) = λA((α]) = (λA(α)] for some α ∈ B(Con(A)), so that λA(α) ∈ B(Con(A)/≡A). Proposition 2.15. Con(A) is a Stone lattice iff Con(A)/≡A is a Stone lattice. Proof. If Con(A) is a Stone lattice, then, for all θ ∈ Con(A), there exists an α ∈ B(Con(A)) such that Ann(θ) = (α], so that λA(α) ∈ B(Con(A)/≡A) and Ann(λA(θ)) = λA(Ann(θ)) = λA((α]) = (λA(α)] by (1◦) and Lemma 2.11, hence Con(A)/≡A is a Stone lattice. If Con(A)/≡A is a Stone lattice, then, by (1◦), Lemma 2.11 and Lemma 2.13, (iv), for all θ ∈ Con(A), there exists an α ∈ B(Con(A)) such that λA(Ann(θ)) = Ann(λA(θ)) = (λA(α)] = λA((α]), hence Ann(θ) = (α], therefore Con(A) is a Stone lattice. Proposition 2.16. For any nonzero cardinality κ, the properties (2)κ,Con(A) and (2)κ,Con(A)/≡A are equivalent. Proof. By Proposition 2.15 and (1◦), which ensures us that the Boolean algebras B(Con(A)) and B(Con(A)/≡A) are isomorphic. Proposition 2.17. Con(A) is a strongly Stone lattice iff Con(A)/≡A is a strongly Stone lattice. Proof. Same as the proof of Proposition 2.15. Remark 2.18. Since Con(A)/≡A is a frame, from Corollary 2.3, we get that: the lattice Con(A)/≡A is Stone iff it is strongly Stone. Corollary 2.19. The lattice Con(A) is Stone iff Con(A) is strongly Stone iff Con(A)/≡A is Stone iff Con(A)/≡A is strongly Stone. Let us consider the following conditions: (pann)L PAnn(L) is a sublattice of Id(L) such that the map a 7→ AnnL(a) is a lattice anti -- morphism from L to PAnn(L); (p2ann)L P2Ann(L) is a sublattice of Id(L) such that the map a 7→ AnnL(AnnL(a)) is a lattice morphism from L to P2Ann(L). Remark 2.20. Concerning the following results, recall that Con(A)/≡A is a frame, and thus PAnn(Con(A)/≡A) = Ann(Con(A)/≡A ) and P2Ann(Con(A)/≡A ) = 2Ann(Con(A)/≡A ). Lemma 2.21. (i) The map P 7→ λA(P ) from PAnn(Con(A)) to PAnn(Con(A)/≡A) is an order isomorphism. (ii) For all θ, ζ ∈ Con(A): Ann(θ∩ ζ) = Ann(θ)∨ Ann(ζ) iff Ann(λA(θ)∧ λA(ζ)) = Ann(λA(θ))∨ Ann(λA(ζ)). (iii) (pann)Con(A) is equivalent to (pann)Con(A)/≡A . (iv) If the equivalent conditions from (iii) are fulfilled, then the map from (i) is a lattice isomorphism. Proof. (i) By Lemma 2.11, this restriction of the direct image of λA takes Ann(θ) to Ann(λA(θ)) for all θ ∈ Con(A), thus it is well defined and surjective. By Lemma 2.13, (iii), it is also injective. By Lemma 2.13, (ii), for all θ, ζ ∈ Con(A), Ann(θ) ⊆ Ann(ζ) iff λA(Ann(θ)) ⊆ λA(Ann(ζ)), therefore this bijection and its inverse preserve order. So this map is an order isomorphism, which is a restriction of the lattice morphism I 7→ λA(I) from Id(Con(A)) to Id(Con(A)/≡A ). (ii) and (iii) By (i), the map P 7→ λA(P ) from PAnn(Con(A)) to PAnn(Con(A)/≡A) preserves all joins and so does its inverse. Lemmas 2.10 and 2.9 show that PAnn(Con(A)) and PAnn(Con(A)/≡A) always are inferior subsemilattices of Id(Con(A)), respectively Id(Con(A)/≡A). From (i) it follows that PAnn(Con(A)) is closed with respect to the join from the lattice Id(Con(A)) iff PAnn(Con(A)/≡A ) is closed with respect to the join from the lattice Id(Con(A)/≡A), and, if they are closed with respect to the join, then, for all θ, ζ ∈ Con(A), Ann(θ) ∨ Ann(ζ) = Ann(θ ∩ ζ) iff λA(Ann(θ) ∨ Ann(ζ)) = λA(Ann(θ ∩ ζ)), which in turn is equivalent to Ann(λA(θ)) ∨ Ann(λA(ζ)) = Ann(λA(θ ∩ ζ)) by Lemma 2.11 and the fact that the direct image of λA preserves the joins of ideals. Therefore PAnn(Con(A)) is a sublattice of Id(Con(A)) iff PAnn(Con(A)/≡A) is a sublattice of Id(Con(A)/≡A), with the expressions for the joins above. (iv) By (i). Lemma 2.22. phism. (i) The map Q 7→ λA(Q) from P2Ann(Con(A)) to P2Ann(Con(A)/≡A) is an order isomor- (ii) (p2ann)Con(A) is equivalent to (p2ann)Con(A)/≡A . (iii) If the equivalent conditions from (ii) are fulfilled, then the map from (i) is a lattice isomorphism. Proof. (i) By Lemma 2.11, this restriction of the direct image of λA takes Ann(Ann(θ)) to Ann(Ann(λA(θ))) for all θ ∈ Con(A), thus it is well defined and surjective. By Lemma 2.13, (iii), for any θ, ζ ∈ Con(A), λA(Ann(Ann(θ))) = λA(Ann(Ann(ζ))) iff Ann(Ann(θ)) = Ann(Ann(ζ)), so this map is also injective. By Lemma 2.13, (ii), for all θ, ζ ∈ Con(A), Ann(Ann(θ)) ⊆ Ann(Ann(ζ)) iff λA(Ann(Ann(θ))) ⊆ λA(Ann(Ann(ζ))), hence this bijection and its inverse preserve order, so this map is an order isomorphism, which is a restriction of the lattice morphism I 7→ λA(I) from Id(Con(A)) to Id(Con(A)/≡A). (ii) and (iii) follow from (i) in the same way in which properties (ii), (iii) and (iv) from Lemma 2.21 follow from (i). Another way to prove these facts is to notice that the equivalences in Lemma 2.21, (ii), hold if we replace the congruences by sets of congruences, and also apply Lemmas 2.9 and 2.10. are equivalent. is fulfilled for any nonzero cardinality κ. Proposition 2.23. For any nonzero cardinality κ, the properties (3)κ,Con(A) and (3)κ,Con(A)/≡A Proof. By Lemma 2.22 and the fact that the map from Lemma 2.22, (i), composed with the map from (3)κ,Con(A) equals the map from (3)κ,Con(A)/≡A composed with the the canonical surjective lattice morphism from Con(A) to Con(A)/≡A. Proposition 2.24. Properties (iv)Con(A) and (iv)Con(A)/≡A are equivalent. Proof. By Lemma 2.21, (ii), and the surjectivity of the map λA : Con(A) → Con(A)/≡A. Remark 2.25. Since Con(A)/≡A is a frame, 2Ann(Con(A)/≡A) ⊆ Ann(Con(A)/≡A ) = PAnn(Con(A)/≡A), which means that the second part of condition (4)κ,Con(A)/≡A Lemma 2.26. Ann(Con(A)) = PAnn(Con(A)). Proof. By Remark 2.25 and Lemma 2.11, for any Ω ⊆ Con(A)), λA(Ann(Ω)) = Ann(λA(Ω)) = Ann(λA(θ)) = λA(Ann(θ)) for some θ ∈ Con(A)), so that Ann(Ω) = Ann(θ) by Lemma 2.13, (iii). Proposition 2.27. For any nonzero cardinality κ, the properties (iv)Con(A), (4)κ,Con(A), (4)<∞,Con(A) and (4)Con(A) are equivalent. Proof. By Lemma 2.26, 2Ann(Con(A)) ⊆ Ann(Con(A)) = PAnn(Con(A)), which means that the second condi- tion in (4)κ,Con(A) is fulfilled for any nonzero cardinality κ, so that conditions (iv)Con(A), (4)κ,Con(A), (4)<∞,Con(A) and (4)Con(A) are equivalent. Proposition 2.28. For any nonzero cardinality κ, the properties (5)κ,Con(A) and (5)κ,Con(A)/≡A are equivalent. Proof. Assume that (5)κ,Con(A) is fulfilled and let U ⊆ Con(A)/≡A with U ≤ κ. For each u ∈ U , there exists an ωu ∈ Con(A) such that λA(ωu) = u. Let Ω = {ωu u ∈ U} ⊆ Con(A). Then Ω = U ≤ κ and λA(Ω) = U , thus Ann(Ω) ∨ Ann(Ann(Ω)) = Con(A) and hence, by Lemma 2.11, Ann(U ) ∨ Ann(Ann(U )) = Ann(λA(Ω)) ∨ Ann(Ann(λA(Ω))) = λA(Ann(Ω)) ∨ λA(Ann(Ann(Ω))) = λA(Ann(Ω) ∨ Ann(Ann(Ω))) = λA(Con(A)) = Con(A)/≡A, so the direct implication holds. is fulfilled and let Ω ⊆ Con(A) with Ω ≤ κ. Then λA(Ω) ⊆ Con(A)/≡A and λA(Ω) ≤ Ω ≤ κ, hence Ann(λA(Ω))∨Ann(Ann(λA(Ω))) = Con(A)/≡A = λA(Con(A)), so that λA(∇A) ∈ Ann(λA(Ω)) ∨ Ann(Ann(λA(Ω))), which means that λA(∇A) ≤ λA(θ) ∨ λA(ζ) for some θ, ζ ∈ Con(A) such that λA(θ) ∈ Ann(λA(Ω)) = λA(Ann(Ω)) and λA(ζ) ∈ Ann(Ann(λA(Ω))) = λA(Ann(Ann(Ω))), so that θ ∈ Ann(Ω) and ζ ∈ Ann(Ann(Ω)), by Lemma 2.11 and Lemma 2.13, (i). Hence λA(∇A) = λA(θ) ∨ λA(ζ) = λA(θ ∨ ζ), thus ∇A = θ ∨ ζ ∈ Ann(Ω) ∨ Ann(Ann(Ω)), hence Ann(Ω) ∨ Ann(Ann(Ω)) = Con(A), therefore the converse implication holds, as well. Now assume that (5)κ,Con(A)/≡A Theorem 2.29. For any nonzero cardinality κ and any h, i, j ∈ 1, 5, conditions (iv)Con(A), (h)κ,Con(A), (i)<∞,Con(A) and (j)Con(A) are equivalent. Proof. By Corollary 2.4 and Propositions 2.14, 2.16, 2.23, 2.27 and 2.28. Open problem 2.30. Determine what kinds of bounded modular lattices are congruence lattices of semiprime algebras from semi -- degenerate congruence -- modular varieties. It will follow that the equivalences in the last statement in Corollary 2.3 hold for all those kinds of bounded modular lattices. Corollary 2.31. Let κ be a nonzero cardinality. Then: the equivalent conditions (iv)Con(A), (1)κ,Con(A), . . . , (5)κ,Con(A), (1)<∞,Con(A), . . . , (5)<∞,Con(A), (1)Con(A), . . . , (5)Con(A) are fulfilled iff the equivalent conditions (iv)Con(A)/≡A , (1)Con(A)/≡A , . . . , (5)Con(A)/≡A are fulfilled. (1)<∞,Con(A)/≡A , . . . , (5)<∞,Con(A)/≡A , (1)κ,Con(A)/≡A , . . . , (5)κ,Con(A)/≡A , Proof. By any of the Propositions 2.24, 2.14, 2.16, 2.23, 2.27 and 2.28, along with Corollary 2.4 and Theorem 2.29. In [9], we have constructed the reticulation of A, L(A), which, by definition, is a bounded distributive lattice whose prime spectrum of ideals (or filters, but our construction in [9] fulfills this property for ideals) is homeomorphic to the prime spectrum of congruences of A, with respect to the Stone topologies. It is well known that, if two bounded distributive lattices have homeomorphic prime spectra of ideals, then they are isomorphic lattices, therefore the reticulation of A is unique up to a lattice isomorphism (or dual isomorphism, if we also consider the variant of the reticulation with the property above for filters). This is our construction of the reticulation of A from [9]: L(A) = K(A)/≡A, where K(A) is the set of the compact elements of the lattice Con(A), thus L(A) = Con(A)/≡A if the lattice Con(A) is compact, in particular if Con(A) is a finite lattice, in particular if A is finite. Following [9], I am denoting the restriction λA K(A): K(A) → L(A) of the canonical surjective lattice morphism λA : Con(A) → Con(A)/≡A by λA, as well. L(A) is a bounded sublattice of Con(A)/≡A, thus a bounded distributive lattice, hence, from Corollaries 2.3 and 2.4 and the obvious fact that a complete sublattice of a frame is a frame, we obtain: Corollary 2.32. • Conditions (1)L(A), . . . , (5)L(A) are equivalent. For any nonzero cardinality κ, (1)κ,L(A), . . . , (5)κ,L(A) are equivalent. For any finite nonzero cardinality κ, (1)κ,L(A), . . . , (5)κ,L(A), (1)<∞,L(A), . . . , (5)<∞,L(A) are equivalent. • If L(A) is a frame, in particular if L(A) is a complete sublattice of Con(A)/≡A , in particular if L(A) = Con(A)/≡A, in particular if Con(A) is a compact lattice, then, for any nonzero cardinality κ and any h, i, j ∈ 1, 5, (iv)L(A), (h)κ,L(A), (i)<∞,L(A) and (j)L(A) are equivalent, so, in particular, (1)L(A) is equiv- alent to (2)<∞,L(A), which means that: L(A) is a Stone lattice iff L(A) is a strongly Stone lattice. Corollary 2.33. If L(A) = Con(A)/≡A, in particular if Con(A) is a compact lattice, in particular if Con(A) is finite, in particular if A is finite, then: Con(A) is Stone iff Con(A) is strongly Stone iff L(A) is Stone iff L(A) is strongly Stone. Remark 2.34. Let M be a bounded sublattice of L and U, V ⊆ L. Then it is straightforward that AnnL(U ) ∩ M ⊆ AnnM (U ∩ M ), (U ∩ M ]L ∩ M = (U ∩ M ]M and, if L is distributive, then (U ]L ∩ (V ]L = (U ∩ V ]L. Lemma 2.35. If L is a bounded distributive lattice, M is a bounded sublattice of L and U ⊆ M , such that AnnL(U ) ∨ AnnL(AnnL(U )) = L, then AnnM (U ) ∨ AnnM (AnnM (U )) = M . Proof. If U is as in the hypothesis, then AnnM (U ) ∨ AnnM (AnnM (U )) ⊇ (AnnL(U ) ∩ M ) ∨ (AnnM (AnnL(U ) ∩ M ) ⊇ (AnnL(U ) ∩ M ) ∨ (AnnL(AnnL(U )) ∩ M ) ⊇ (AnnL(U ) ∩ M ) ∪ (AnnL(AnnL(U )) ∩ M ) = (AnnL(U ) ∪ AnnL(AnnL(U ))∩M , thus AnnM (U )∨AnnM (AnnM (U )) ⊇ (AnnL(U )∪AnnL(AnnL(U )]M∩(M ]M = (AnnL(U )∪ AnnL(AnnL(U ))]L ∩ M ∩ M = (AnnL(U ) ∨ AnnL(AnnL(U ))) ∩ M = L ∩ M = M . Lemma 2.36. If L is a bounded distributive lattice and M is a bounded sublattice of L, then, for any nonzero cardinality κ, (5)κ,L implies (5)κ,M . Proof. Assume that (5)κ,L is fulfilled, and let U ⊆ M ⊆ L with U ≤ κ, so that AnnL(U )∨AnnL(AnnL(U )) = L, hence AnnM (U ) ∨ AnnM (AnnM (U )) = M by Lemma 2.35. Proposition 2.37. If L is a bounded distributive lattice, M is a bounded sublattice of L, κ is a nonzero cardinality and the equivalent conditions (1)κ,L, . . . , (5)κ,L are fulfilled, then the equivalent conditions (1)κ,M , . . . , (5)κ,M are fulfilled. Proof. By Corollary 2.3 and Lemma 2.36. Corollary 2.38. For any nonzero cardinality κ, if the equivalent conditions (iv)Con(A),(1)κ,Con(A),. . ., (5)κ,Con(A), (1)<∞,Con(A), . . . , (5)<∞,Con(A), (1)Con(A), . . . , (5)Con(A) are fulfilled, then conditions (iv)L(A), (1)κ,L(A), . . . , (5)κ,L(A), (1)<∞,L(A), . . . , (5)<∞,L(A), (1)L(A), . . . , (5)L(A) (not necessarily equivalent) are fulfilled. Remark 2.39. Obviously, if L and M are isomorphic bounded lattices, then: (iv)L is equivalent to (iv)M , for any i ∈ 1, 5, (i)L is equivalent to (i)M , (i)<∞,L is equivalent to (i)<∞,M and, for any nonzero cardinality κ, (i)κ,L is equivalent to (i)κ,M . Remark 2.40. If L and M are bounded distributive lattices, then it is immediate that: B(L×M ) = B(L)×B(M ) and, for all a ∈ L and all b ∈ M , AnnL×M ((a, b)) = {(x, y) x ∈ AnnL(a), y ∈ AnnM (b)} = AnnL(a)× AnnM (b); since, for any x ∈ L and y ∈ M , (x, y) ≤ (a, b) means that x ≤ a and y ≤ b, it follows that ((a, b)]L×M = {(x, y) x ∈ (a]L, y ∈ (b]M} = (a]L × (b]M . And, for any set I and any families (ai)i∈I ⊆ L and (bi)i∈I ⊆ M , there exists _ bi in M , and, if these joins exist, then ai in L and _ i∈I (ai, bi) in L × M iff there exist _ ai, _ i∈I i∈I _ (ai, bi) = (_ bi); the same goes for arbitrary meets. From this, it is easy to obtain that, for any i∈I nonzero cardinality κ, (2)κ,L×M is fulfilled iff both (2)κ,L and (2)κ,M are fulfilled, and hence, by Corollary 2.3: i∈I i∈I Proposition 2.41. For any nonzero cardinality κ: condition (iv)L×M , respectively the equivalent conditions (1)κ,L×M , . . . , (5)κ,L×M , respectively (1)<∞,L×M , . . . , (5)<∞,L×M , respectively (1)L×M , . . . , (5)L×M , are fulfilled iff condition (iv)L, respectively the equivalent conditions (1)κ,L, . . . , (5)κ,L, respectively (1)<∞,L, . . . , (5)<∞,L, respectively (1)L, . . . , (5)L, as well as condition (iv)M , respectively the equivalent conditions (1)κ,M , . . . , (5)κ,M , respectively (1)<∞,M , . . . , (5)<∞,M , respectively (1)M , . . . , (5)M , are fulfilled. Throughout the rest of this section, B shall be a semiprime algebra from C. Then: Corollary 2.43. Remark 2.42. Con(A×B) is isomorphic to Con(A)×Con(B) and, as we have proven in [9], A×B is semiprime, Con(A × B)/≡A×B is isomorphic to Con(A)/≡A × Con(B)/≡B and L(A × B) is isomorphic to L(A) × L(B). (iv)Con(A×B)/≡A×B , (1)κ,Con(A×B)/≡A×B , (5)<∞,Con(A×B)/≡A×B , are fulfilled iff the equivalent conditions (iv)Con(A)/≡A , (1)Con(A×B)/≡A×B (1)κ,Con(A)/≡A , . . . , (5)κ,Con(A)/≡A , (1)<∞,Con(A)/≡A , . . . , (5)<∞,Con(A)/≡A , (1)Con(A)/≡A , . . . , (5)Con(A)/≡A , as well as the equivalent conditions (iv)Con(B)/≡B , (1)κ,Con(B)/≡B , . . . , (5)κ,Con(B)/≡B , (1)<∞,Con(B)/≡B , . . . , (5)<∞,Con(B)/≡B , (1)Con(B)/≡B , . . . , (5)Con(B)/≡B , are fulfilled. • For any nonzero cardinality κ: , (1)<∞,Con(A×B)/≡A×B , . . . , (5)Con(A×B)/≡A×B , (5)κ,Con(A×B)/≡A×B the equivalent conditions , . . . , . . . • For any nonzero cardinality κ: condition (iv)L(A×B), respectively the equivalent conditions (1)κ,L(A×B), . . . , (5)κ,L(A×B), respectively (1)<∞,L(A×B), . . . , (5)<∞,L(A×B), respectively (1)L(A×B), . . . , (5)L(A×B), are ful- filled iff condition (iv)L(A), respectively the equivalent conditions (1)κ,L(A), . . . , (5)κ,L(A), respectively (1)<∞,L(A), . . . , (5)<∞,L(A), respectively (1)L(A), . . . , (5)L(A), as well as condition (iv)L(B), respectively the equivalent conditions (1)κ,L(B), . . . , (5)κ,L(B), respectively (1)<∞,L(B), . . . , (5)<∞,L(B), respectively (1)L(B), . . . , (5)L(B), are fulfilled. Proof. By Remark 2.42, Proposition 2.41 and Corollary 2.3. Corollary 2.44. For any nonzero cardinality κ: the equivalent conditions (iv)Con(A×B), (1)κ,Con(A×B), . . . , (5)κ,Con(A×B), (1)<∞,Con(A×B), . . . , (5)<∞,Con(A×B), (1)Con(A×B), . . . , (5)Con(A×B) are fulfilled iff the equivalent condition (iv)Con(A), (1)κ,Con(A), . . . , (5)κ,Con(A), (1)<∞,Con(A), . . . , (5)<∞,Con(A), (1)Con(A), . . . , (5)Con(A), as well as the equivalent conditions (iv)Con(B), (1)κ,Con(B), . . . , (5)κ,Con(B), (1)<∞,Con(B), . . . , (5)<∞,Con(B), (1)Con(B), . . . , (5)Con(B), are fulfilled. Proof. By Corollaries 2.43 and 2.31. 3 Transferring Davey's Theorem to Commutative Unitary Rings Let (T,∨,∧,⊙,→, 0, 1) be a residuated lattice, which means that (T,∨,∧, 0, 1) is a bounded lattice, that we shall denote by S, (T,⊙, 1) is a commutative monoid and → is a binary operation on T which fulfills the law of residuation: for all a, b, c ∈ T , a ≤ b → c iff a ⊙ b ≤ c. Let us denote by S ′ the dual of the bounded lattice S. See more about residuated lattices in [7], [13], [17]. Residuated lattices form a semi -- degenerate congruence -- distributive variety, hence they are semiprime and thus their congruence lattices fulfill Theorem 2.1 and even Theorem 2.29. But they also fulfill a theorem of this form for elements, which can be expressed in the following way, since we notice that the bounded lattice of the filters of T is a bounded sublattice of that of the filters of S, for each e ∈ B(S), the filter of T generated by e coincides to the filter of S generated by e and the co -- annihilators in T coincide to the co -- annihilators in S: Theorem 3.1. [15, Theorem 5.2.6],[16, Theorem 3.13] Conditions (1)m,S ′, (2)m,S ′ , (3)m,S ′, (4)m,S ′ and (5)m,S ′ are equivalent. In [15, 16], I have proven Theorem 3.1 by transferring the dual of Theorem 2.1 from bounded distributive lattices to residuated lattices through the reticulation functor for residuated lattices. With the notation for the reticulation from Section 2, the construction from [9] identifies L(T ), up to a lattice isomorphism, as the bounded lattice PFilt(T ) of the principal filters of T , whose dual is a frame if T is complete. Since, for any nonzero cardinality κ and any i ∈ 1, 5, (i)κ,S ′ is equivalent to (i)κ,D, where D is the dual of PFilt(T ), according to [15, 16] and the above, it follows that: Theorem 3.2. If T is a complete residuated lattice and S ′ is the dual of the underlying bounded lattice of T , then, for any nonzero cardinality κ and any h, i, j ∈ 1, 5, conditions (iv)S ′ , (h)κ,S ′ , (i)<∞,S ′ and (j)S ′ are equivalent; in particular, T is co -- Stone iff T is strongly co -- Stone. Remark 3.3. If the commutator of A is associative, as it is, for example, in any commutative unitary ring, then Con(A) is a residuated lattice, in which [·,·]A is the multiplication, according to [9] (see also [4]), thus, from the above and the fact that Con(A) is a complete lattice, it follows that both the lattice Con(A) and its dual fulfill the equivalences in Theorem 2.29. Note that the commutator is not always associative [8]. Commutative unitary rings form a semi -- degenerate congruence -- modular equational class, thus their congru- ence lattices fulfill Theorem 2.29. Let us see that, similarly to what happens in (complete) residuated lattices, they also fulfill an analogue of Theorem 2.29 for elements instead of congruences. Commutative unitary rings form a semi -- degenerate congruence -- modular variety, thus their congruence lat- tices fulfill Theorem 2.29. Let us see that, like residuated lattices, commutative unitary rings fulfill Davey's Theorem for elements, too. Let (R, +,·, 0, 1) be a commutative unitary ring, (Id(R),∨ = +,∩,{0}, R) be the bounded modular lattice of the ideals of R, SpecId(R) the set of the prime ideals of R and ιγR : Id(R) → Con(R) the canonical lattice isomorphism: for all I ∈ Id(R), ιγR(I) = {(x, y) ∈ I 2 x− y ∈ I}. Note that, since ιγR is an order isomorphism, it preserves arbitrary intersections. Recall that SpecId(R) = {P ∈ Id(R) \ {R} (∀ I, J ∈ Id(R)) (I · J ⊆ P ⇒ I ⊆ P or J ⊆ P )} and [ιγR(I), ιγR(J)]R = ιγR(I · J) for all I, J ∈ Id(R), from which it is easy to deduce that ιγR(SpecId(R)) = Spec(R). For every U ⊆ R, hUiR shall be the ideal of R generated by U , so, for each x ∈ R, h{x}iR = xR. Let PId(R) and FGId(R) be the set of the principal ideals of R and that of the finitely It is straightforward that, for all x, a, b ∈ R, ιγR(xR) = CgR(x, 0) and generated ideals of R, respectively. CgR(a, b) = CgR(a − b, 0), hence ιγR(PId(R)) = PCon(R) and thus ιγR(FGId(R)) = K(R). Recall that, if we denote by E(R) the set of the idempotents of R, then (E(R),∨,∧ = ·,¬ , 0, 1) is a Boolean algebra, where, for every e, f ∈ E(R), ¬ e = 1 − e and e ∨ f = ¬ (¬ e ∧ ¬ f ) = 1 − (1 − e) · (1 − f ). Let L(R) = (K(R)/≡R,∨,∧, 0 = λR(∆R) = ∆R/≡R, 1 = λR(∇R) = ∇R/≡R) be the reticulation of R, as constructed in [9] (see the notations in Section 2). Let R∗ and µR : R → R∗ be the reticulation of R if we denote, for each and the reticulation function, respectively, as constructed in [2, 3] (see also [14, 18]): I ∈ Id(R), by √I = T{P ∈ SpecId(R) I ⊆ P} the radical of I, and by ∼R= {(I, J) ∈ (Id(R))2 √I = √J}, then ∼R is a congruence of the lattice Id(R), so Id(R)/∼R is a bounded lattice and the canonical surjection νR : Id(R) → Id(R)/∼R is a lattice morphism that fulfills: νR(I · J) = νR(I ∩ J) for all I, J ∈ Id(R), therefore the bounded lattice R∗ = (Id(R)/∼R,∨,∧,⊥= νR({0})) = {0}/∼R,⊤ = νR(R) = R/∼R) is distributive, since multiplication is distributive w.r.t. the join in Id(R). As shown in [2], R∗ has the prime spectrum of ideals homeomorphic to the prime spectrum of congruences of R, w.r.t. the Stone topologies, hence R∗ is a reticulation of R. The reticulation function µR : R → R∗ is defined, in [2], by: µR(x) = νR(xR) = xR/∼R for all x ∈ R. For any U ⊆ R, we denote the annihilator of U by AnnR(U ); so AnnR(U ) = {x ∈ R (∀ u ∈ U ) (u · x = 0)} ∈ Id(R). For any a ∈ R, we also denote AnnR(a) = AnnR({a}) = {x ∈ R u · a = 0}. Let us denote by Ann(R) = {AnnR(U ) U ⊆ R} and by 2Ann(R) = {AnnR(AnnR(U )) U ⊆ R}. Remark 3.4. It is well known and straightforward that Ann(R) ⊆ Id(R). Lemma 3.5. • For any U ⊆ R, AnnR(U ) = AnnR(hUiR). • For any V ⊆ Id(R), \ I ∈V AnnR(I) = AnnR( _ I). I ∈V Proof. Clearly, for any U ⊆ V ⊆ R, we have AnnR(V ) ⊆ AnnR(U ), hence AnnR(hUiR) ⊆ AnnR(U ). But the converse inclusion holds, as well, since, given any a ∈ hUiR and any x ∈ AnnR(U ), we have a = a1·u1+. . .+an·un for some n ∈ N∗, a1, . . . , an ∈ R and u1, . . . , un ∈ U , so that x · u1 = . . . = x · un = 0, therefore x · u = 0, so x ∈ AnnR(hUiR). Thus AnnR(U ) = AnnR(hUiR). Hence \ AnnR(a) = AnnR( [ AnnR(a) = \ AnnR(I) = \ a∈SI∈V I \ I ∈V a∈I I ∈V I) = AnnR(h [ IiR) = AnnR( _ I). I ∈V I ∈V I ∈V Regarding the results from [2] I am using, note that, since R is commutative, it follows that R is quasicom- mutative, thus, by [2, Theorem 3], R fulfills condition (∗) from [2]. It also follows that: Lemma 3.6. [2, Lemma, p. 1861] For all I ∈ Id(R), there exists a K ∈ FGId(R) such that K ⊆ I and √K = √I. Proposition 3.7. (i) R∗ = Id(R)/∼R = FGId(R)/∼R; (ii) ϕR : R∗ → Con(R)/≡R, for all I ∈ Id(R), ϕR(I/∼R) = ιγR(I)/≡R, is a lattice isomorphism; (iii) Con(R)/≡R = K(R)/≡R = L(R). Proof. (i) Lemma 3.6 says that, for all I ∈ Id(R), there exists a K ∈ FGId(R) such that K ⊆ I and K/∼R = I/∼R, therefore R∗ = Id(R)/∼R = FGId(R)/∼R. (ii) The map ϕR is defined by: ϕR(νR(I)) = λR(ιγR(I))) for all I ∈ Id(R), hence it makes the following diagram commutative, where the second equality in the bottom row will follow shortly: ιγR ϕR ✲ ✲ Con(R) ⊇ K(R) λR ❄ λR ❄ FGId(R) ⊆ Id(R) νR ❄ νR FGId(R) ❄ FGId(R)/∼R = Id(R)/∼R = R∗ Con(R)/≡R = K(R)/≡R = L(R) For all I, J ∈ Id(R), we have: I/∼R = J/∼R iff √I = √J iff ιγR(√I) = ιγR(√J) iff ιγR(T{P ∈ SpecId(R) I ⊆ P}) = ιγR(T{Q ∈ SpecId(R) J ⊆ Q}) iff T{ιγR(P ) P ∈ SpecId(R), I ⊆ P} = T{ιγR(Q) Q ∈ SpecId(R), J ⊆ Q} iff T{ιγR(P ) P ∈ SpecId(R), ιγR(I) ⊆ ιγR(P )} = T{ιγR(Q) Q ∈ SpecId(R), ιγR(J) ⊆ ιγR(Q)} iff T{φ φ ∈ Spec(R), ιγR(I) ⊆ φ} = T{ψ ψ ∈ Spec(R), ιγR(J) ⊆ ψ} iff ρR(ιγR(I)) = ρR(ιγR(J)) iff ιγR(I)/≡R = ιγR(J)/≡R iff ϕR(I/∼R) = ϕR(J/∼R ), hence ϕR is well defined and injective. Since ϕR ◦ νR = λR ◦ ιγR, which is surjective, it follows that ϕR is surjective. Clearly, ϕR preserves the join and the meet, so it is a lattice isomorphism. (iii) By (i) and (ii), Id(R)/∼R = FGId(R)/∼R, that is νR(Id(R)) = νR(FGId(R)), so ϕR(νR(Id(R))) = ϕR(νR(FGId(R))), which means that λR(ιγR(Id(R))) = λR(ιγR(FGId(R))), thus λR(Con(R)) = λR(K(R)), that is Con(R)/≡R = K(R)/≡R = L(R). Remark 3.8. Since B(Con(R)) is a Boolean sublattice of Con(R) and Con(R) is isomorphic to Id(R), it follows that B(Id(R)) is a Boolean sublattice of Id(R). Corollary 3.9. R∗ and L(R) are frames and ϕR : FGId(R)/∼R = Id(R)/∼R = R∗ → Con(R)/≡R = K(R)/≡R = L(R) is a lattice (and thus a frame) isomorphism, and, if we take, in the diagram above in the class of bounded lattices, the restrictions of the bounded lattice morphisms to the Boolean centers, then we get the following commutative diagram in the class of Boolean algebras, in which ιγR B(Id(R)): B(Id(R)) → B(Con(R)) and ϕR B(R∗): B(R∗) → B(L(R)) are Boolean isomorphisms: ιγR B(Id(R)) νR B(Id(R)) ❄ ϕR B(R∗) B(Id(R)) B(R∗) ✲ B(Con(R)) λR B(Con(R)) ❄ ✲ B(L(R)) Proof. By Remark 3.8, Proposition 3.7 and the fact that Con(R)/≡R is a frame, along with (1◦). homeomorphic prime spectra of ideals w.r.t. the Stone topologies. So the bounded distributive lattices L(R) and R∗ are isomorphic, which was to be expected, since they have R is called a Baer ring iff, for any a ∈ R, there exists an e ∈ E(R) such that AnnR(a) = eR. By analogy to the case of bounded lattices, we shall call R a strongly Baer ring iff, for any U ⊆ R, there exists an e ∈ E(R) such that AnnR(U ) = eR. Following [2, 3], we call R a semiprime ring iff p{0} = {0}, that is \ P = {0}, which is equivalent to ∆R = ιγR({0}) = ιγR( \ means that R is a semiprime algebra. P ∈SpecId(R) φ = ρR(∆R), which ιγR(P ) = \ P ) = \ P ∈SpecId(R) P ∈SpecId(R) φ∈Spec(R) Remark 3.10. [2, 3] Let I ∈ Id(R). Then: I/∼R = ⊤ iff νR(I) = νR(R) iff √I = √R = R iff no prime ideal of R includes I iff I = R. If R is semiprime, then we also have the following equivalences: I/∼R =⊥ iff νR(I) = νR({0}) iff I ⊆ √I = p{0} = {0} iff I = {0}. Remark 3.11. By Corollary 3.9 and Remarks 2.6 and 2.7, Ann(Id(R)) ⊆ Id(Id(R)), and the canonical sur- jective lattice morphism νR : Id(R) → R∗ = Id(R)/∼R preserves arbitrary intersections and joins and fulfills: (I]Id(R)/∼R = (I/∼R]R∗ for any I ∈ Id(R). Corollary 3.12. For any nonzero cardinality κ and any h, i, j ∈ 1, 5, (iv)R∗ , (h)κ,R∗ , (i)<∞,R∗ and (j)R∗ are equivalent. Proof. By Corollaries 2.31 and 3.9. Let us consider the following conditions on R, where κ is an arbitrary nonzero cardinality: for each U ⊆ R with U ≤ κ, there exists an e ∈ E(R) such that AnnR(U ) = eR; for each finite U ⊆ R, there exists an e ∈ E(R) such that AnnR(U ) = eR; R is a strongly Baer ring; R is a Baer ring and E(R) is a κ -- complete Boolean algebra; (1)κ,R (1)<∞,R (1)R (2)κ,R (2)<∞,R R is a Baer ring and E(R) is a Boolean algebra; (2)R (3)κ,R R is a Baer ring and E(R) is a complete Boolean algebra; 2Ann(R) is a κ -- complete Boolean sublattice of Id(R) such that I 7→ AnnR(AnnR(I)) is a lattice morphism from Id(R) to 2Ann(R); 2Ann(R) is a Boolean sublattice of Id(R) such that I 7→ AnnR(AnnR(I)) is a lattice morphism from Id(R) to 2Ann(R); 2Ann(R) is a complete Boolean sublattice of Id(R) such that I 7→ AnnR(AnnR(I)) is a lattice morphism from Id(R) to 2Ann(R);; for all I, J ∈ Id(R), AnnR(I ∩ J) = AnnR(I) ∨ AnnR(J), and, for each U ⊆ R with U ≤ κ, there exists a finite subset S ⊆ R such that AnnR(AnnR(U )) = AnnR(S); for all I, J ∈ Id(R), AnnR(I ∩ J) = AnnR(I) ∨ AnnR(J), and, for each finite U ⊆ R, there exists a finite subset S ⊆ R such that AnnR(AnnR(U )) = AnnR(S); for all I, J ∈ Id(R), AnnR(I ∩ J) = AnnR(I) ∨ AnnR(J), and, for each U ⊆ R, there exists a finite subset S ⊆ R such that AnnR(AnnR(U )) = AnnR(S); for all I, J ∈ Id(R), AnnR(I ∩ J) = AnnR(I) ∨ AnnR(J); (3)<∞,R (3)R (4)κ,R (4)<∞,R (4)R (iv)R (5)κ,R (5)<∞,R (5)R for each U ⊆ R with U ≤ κ, AnnR(U ) ∨ AnnR(AnnR(U )) = R; for each finite U ⊆ R, AnnR(U ) ∨ AnnR(AnnR(U )) = R; for each U ⊆ R, AnnR(U ) ∨ AnnR(AnnR(U )) = R. Remark 3.13. Clearly, the properties from the second paragraph of Remark 2.2 hold for R instead of L, too. For the motivation behind the change of the second part of conditions (4)κ,L, (4)<∞,L and (4)L into what we see above in conditions (4)κ,R, (4)<∞,R and (4)R, compare Lemma 2.26 above to Lemma 3.20 below. Throughout the rest of this paper, R shall be semiprime. Lemma 3.14. (i) [2, Lemma, p. 1863] µR E(R): E(R) → B(R∗) is a Boolean isomorphism. (ii) [18, Theorem 2.6] R is a Baer ring iff R∗ is a Stone lattice. Proposition 3.15. For any nonzero cardinality κ, conditions (2)κ,R and (2)κ,R∗ are equivalent. Proof. By Lemma 3.14, (i) and (ii). Lemma 3.16. νR B(Id(R)): B(Id(R)) → B(Id(R))/∼R = B(Id(R)/∼R) = B(R∗) is a Boolean isomorphism. Proof. By (1◦) and Corollary 3.9. Lemma 3.17. For any V, W ⊆ Id(R), I ∈ Id(R) and D ∈ B(Id(R)): (i) AnnR∗ (I/∼R) = AnnId(R)(I)/∼R and AnnR∗ (W/∼R) = AnnId(R)(W )/∼R; (ii) AnnR∗ (V /∼R ) = AnnR∗ (W/∼R) iff AnnId(R)(V )/∼R = AnnId(R)(W )/∼R iff AnnId(R)(V ) = AnnId(R)(W ); (iii) (D/∼R]R∗ = AnnR∗ (W/∼R ) iff (D]Id(R)/∼R = AnnId(R)(W )/∼R iff (D]Id(R) = AnnId(R)(W ). Proof. (i) By Corollary 3.9 and Lemma 2.11. (ii) By (i), Corollary 3.9 and Lemma 2.13, (iii). (iii) By (i), Corollary 3.9, Lemma 2.13, (iii), and Remark 3.11. Lemma 3.18. For any U ⊆ R and any I ∈ Id(R): • AnnId(R)(I) = (AnnR(I)]Id(R) and AnnId(R)(hUiR) = (AnnR(U )]Id(R); • AnnR∗ (µR(I)) = AnnR∗ (I/∼R) = (AnnR(I)]Id(R)/∼R = (AnnR(I)/∼R]R∗ ; • AnnR∗ (µR(U )) = AnnR∗ (µR(hUiR)) = (AnnR(U )]Id(R)/∼R = (AnnR(U )/∼R]R∗ ; • AnnId(R)(AnnId(R)(I)) = (AnnR(AnnR(I))]Id(R) and AnnR∗ (AnnR∗ (µR(U ))) = (AnnR(AnnR(U ))/∼R ]R∗ ; • AnnR∗ (I/∼R) = AnnId(R)(I)/∼R and AnnR∗ (AnnR∗ (I/∼R)) = AnnId(R)(AnnId(R)(I))/∼R . x∈U Proof. Let J ∈ Id(R). Then: J ∈ (AnnR(I)]Id(R) iff J ⊆ AnnR(I) iff x ∈ AnnR(I) for all x ∈ J iff x · y = 0 for all x ∈ J and all y ∈ I iff J · I = {0} iff J ∈ (AnnR(I)]Id(R), hence AnnId(R)(I) = (AnnR(I)]Id(R), thus AnnId(R)(hUiR) = (AnnR(hUiR)]Id(R) = (AnnR(U )]Id(R) by Lemma 3.5. By Lemma 3.17, (i), and Remark 3.11, AnnR∗ (I/∼R) = (AnnR(I)]Id(R)/∼R = (AnnR(I)/∼R ]R∗ . By Lemma 3.5 and Remark 3.11, AnnR∗ (µR(U )) = AnnR∗ ({µR(x)) x ∈ U}) = AnnR∗ ({νR(xR)) x ∈ U}) = \ AnnR∗ (νR(xR)) = AnnR∗ ( _ xR)) = AnnR∗ (νR(hUiR)) = AnnR∗ (hUiR/∼R). Therefore AnnR∗ (µR(I)) = AnnR∗ (I/∼R) and thus AnnR∗ (µR(U )) = AnnR∗ (hUiR/∼R) = AnnR∗ (µR(hUiR)), hence, by the above, Lemma 3.17, (i), Remark 3.11 and Lemma 3.5, AnnR∗ (µR(U )) = AnnR∗ (hUiR/∼R) = (AnnId(R)(hUiR)/∼R = (AnnR(U )]Id(R)/∼R = (AnnR(U )/∼R]R∗ . Hence AnnId(R)(AnnId(R)(I)) = AnnId(R)((AnnR(I)]Id(R)) = AnnId(R)(AnnR(I)) = (AnnR(AnnR(I))]Id(R) and AnnR∗ (AnnR∗ (µR(U ))) = AnnR∗ ((AnnR(U )/∼R]R∗ ) = AnnR∗ (AnnR(U )/∼R) = (AnnR(AnnR(U ))/∼R]R∗ . By the above, AnnR∗ (I/∼R) = AnnId(R)(I)/∼R, and thus AnnR∗ (AnnR∗ (I/∼R)) = AnnR∗ (AnnId(R)(I)/∼R) = AnnId(R)(AnnId(R)(I))/∼R . νR(xR)) = AnnR∗ (νR( _ x∈U x∈U Lemma 3.19. For any U, V ⊆ R and any W ⊆ R∗: (i) AnnR∗ (µR(U )) = AnnR∗ (µR(V )) iff AnnR∗ (hUiR/∼R) = AnnR∗ (hV iR/∼R) iff AnnR(U )/∼R = AnnR(V )/ ∼R iff AnnR(U ) = AnnR(V ); (ii) there exists a finite subset S ⊆ hUiR such that AnnR(U ) = AnnR(S); (iii) there exists a finite subset S ⊆ R such that AnnR∗ (W ) = AnnR∗ (µR(S)). Proof. (i) Let U ⊆ R. By Lemma 3.18 and Lemma 3.17, (ii), AnnR∗ (µR(U )) = AnnR∗ (µR(V )) iff AnnR∗ (hUiR/ ∼R) = AnnR∗ (hV iR/∼R) iff AnnId(R)(hUiR)/∼R = AnnId(R)(hV iR)/∼R, which is equivalent both to (AnnR(U )/ ∼R]R∗ = (AnnR(V )/∼R]R∗ and to AnnId(R)(hUiR) = AnnId(R)(hV iR), which in turn are equivalent to AnnR(U )/ ∼R = AnnR(V )/∼R and to (AnnR(U )]Id(R) = (AnnR(V )]Id(R), respectively, the latter of which is equivalent to AnnR(U ) = AnnR(V ). (ii) By Proposition 3.7, (i), for an appropriate finite subset S ⊆ hUiR, hUiR/∼R = hSiR/∼R, thus AnnR∗ (hUiR/ ∼R) = AnnR∗ (hSiR/∼R), hence AnnR(U ) = AnnR(S) by (i). (iii) Let W ⊆ R∗, so that W = {Ik/∼R k ∈ K} for some (Ik)k∈K ⊆ Id(R). As pointed out in Corollary 3.9, R∗ is a frame, thus AnnR∗ (W ) = AnnR∗ ( _ Ik)/∼R) = AnnR∗ (hSiR/∼R) = Ik, by Proposition 3.7, (i), and Lemma Ik/∼R) = AnnR∗ (( _ AnnR∗ (µR(hSiR)) = AnnR∗ (µR(S)) for some finite subset S ⊆ _ 3.18. k∈K k∈K k∈K Lemma 3.20. Ann(R) = {AnnR(S) S ⊆ R,S < ∞}. Proof. By Lemma 3.19, (ii). Proposition 3.21. (i) (1)1,R implies (1)<∞,R; (ii) (1)<∞,R implies (1)R∗ ; (iii) (1)R∗ implies (1)R; n i=1 n i=1 \ \ AnnR(ui) = (iv) for any nonzero cardinality κ: (1)κ,R, (1)<∞,R, (1)R and (1)R∗ are equivalent. Proof. (i) If n ∈ N∗, u1, . . . , un ∈ R and, for each i ∈ 1, n, AnnR(ui) = eiR for some ei ∈ E(R), then eiR = e1R·. . .·enR = (e1·. . .·en)R, and e1·. . .·en = e1∧. . .∧en ∈ E(R). AnnR({u1, . . . , un}) = (ii) Let W ⊆ R∗. By Lemma 3.19, (iii), there exists a finite subset S ⊆ R such that AnnR∗ (W ) = AnnR∗ (µR(S)). If (1)<∞,R is fulfilled, then AnnR(S) = eR for some e ∈ E(R). By Lemma 3.18 and Lemma 3.14, (i), it follows that AnnR∗ (W ) = AnnR∗ (µR(S)) = (AnnR(S)/∼R]R∗ = (eR/∼R]R∗ = (µR(e)]R∗ , and µR(e) ∈ B(R∗). (iii) Let U ⊆ R, so that µR(U ) ⊆ R∗. If (1)R∗ is fulfilled, then AnnR∗ (µR(U )) = (f ]R∗ for some f ∈ B(R∗). By Lemma 3.14, (i), f = µR(e) = eR/∼R = νR(eR) for some e ∈ E(R), so that eR ∈ B(Id(R)) by Lemma 3.16. By Lemma 3.18 and Lemma 3.17, (iii), AnnR∗ (hUiR/∼R) = AnnR∗ (µR(hUiR)) = (eR/∼R]R∗ , hence AnnId(R)(hUiR) = (eR]Id(R), that is (AnnR(U )]Id(R) = (eR]Id(R), so that AnnR(U ) = eR. (iv) By (i), (ii) and (iii). Remark 3.22. By Lemma 3.5, Ann(R) = {AnnR(I) I ∈ Id(R)} and 2Ann(R) = {AnnR(AnnR(I)) I ∈ Id(R)}. Let us consider the following conditions on R: (ann)R Ann(R) is a sublattice of Id(R) such that the map I 7→ AnnR(I) is (2ann)R a lattice anti -- morphism from Id(R) to Ann(R); 2Ann(R) is a sublattice of Id(R) such that the map I 7→ AnnR(AnnR(I)) is a lattice morphism from Id(R) to 2Ann(R). Lemma 3.23. isomorphism. (i) The map AnnR(I) 7→ AnnId(R)(I) (I ∈ Id(R)) from Ann(R) to PAnn(Id(R)) is an order (ii) The map AnnR(AnnR(I)) 7→ AnnId(R)(AnnId(R)(I)) (I ∈ Id(R)) from 2Ann(R) to P2Ann(Id(R)) is an order isomorphism. Proof. Clearly, these maps are surjective and order -- preserving. By Lemma 3.5, they are completely defined. By Lemma 3.18, the map from (i) is well defined and injective. By Lemma 3.18, Lemma 3.17, (i), and the injectivity of the map from (i), it follows that the map from (ii) is well defined and injective. Hence these maps are bijective. Clearly, their inverses are order -- preserving, as well. Lemma 3.24. to PAnn(R∗) is an order isomorphism. (i) The map AnnId(R)(I) 7→ AnnId(R)(I)/∼R = AnnR∗ (I/∼R) (I ∈ Id(R)) from PAnn(Id(R)) (ii) For all I, J ∈ Id(R): AnnId(R)(I∩J) = AnnId(R)(I)∨AnnId(R)(J) iff AnnR∗ (I/∼R∧J/∼R) = AnnR∗ (I/∼R) ∨AnnR∗ (J/∼R). (iii) (pann)Id(R) is equivalent to (pann)R∗ , and, if they are fulfilled, then the map from (i) is a lattice isomor- phism. Proof. By Lemma 2.21 and Corollary 3.9, with the equality in (i) holding by Lemma 3.18. Lemma 3.25. isomorphism. (i) The map AnnR(I) 7→ AnnR∗ (I/∼R) (I ∈ Id(R)) from Ann(R) to PAnn(R∗) is an order (ii) For all I, J ∈ Id(R): AnnR(I ∩ J) = AnnR(I) ∨ AnnR(J) iff AnnR∗ (I/∼R ∧ J/∼R ) = AnnR∗ (I/∼R) ∨ AnnR∗ (J/∼R ). (iii) (ann)R is equivalent to (pann)R∗ and, if they are fulfilled, then the map from (i) is a lattice isomorphism. Proof. By Remark 3.22, Lemma 3.23, (i), Lemma 3.24 and the fact that the map from (i) is the composition of the map from Lemma 3.24, (i), with the map from Lemma 3.23, (i). Lemma 3.26. (i) The map AnnId(R)(AnnId(R)(I)) 7→ AnnId(R)(AnnId(R)(I))/∼R = AnnR∗ (AnnR∗ (I/∼R)) (I ∈ Id(R)) from P2Ann(Id(R)) to P2Ann(R∗) is an order isomorphism. (ii) (p2ann)Id(R) is equivalent to (p2ann)R∗ and, if these conditions are fulfilled, then the map from (i) is a lattice isomorphism. Proof. By Lemma 2.22 and Corollary 3.9, with the equality in (i) holding by Lemma 3.18. Lemma 3.27. P2Ann(R∗) is an order isomorphism. (i) The map AnnR(AnnR(I)) 7→ AnnR∗ (AnnR∗ (I/∼R)) (I ∈ Id(R)) from 2Ann(R) to (ii) (2ann)R is equivalent to (p2ann)R∗ and, if these conditions are fulfilled, then the map from (i) is a lattice isomorphism. Proof. By Remark 3.22, Lemma 3.23, (ii), Lemma 3.26 and the fact that the map from (i) is the composition of the map from Lemma 3.26, (i), with the map from Lemma 3.23, (ii). Proposition 3.28. For any nonzero cardinality κ, the properties (3)κ,R and (3)κ,R∗ are equivalent. Proof. By Lemma 3.27 and the fact that the map from Lemma 3.27, (i), composed with the map from condition (3)κ,R equals the map from condition (3)κ,R∗ composed with the canonical surjective lattice morphism from Id(R) to Id(R)/∼R = R∗. Proposition 3.29. For any nonzero cardinality κ, conditions (4)κ,R, (4)<∞,R, (4)R, (iv)R and (iv)R∗ are equivalent. Proof. By Lemma 3.25, (ii), (iv)R is equivalent to (iv)R∗ . By Lemma 3.20, {AnnR(S) S ⊆ R,S < ∞} = Ann(R) ⊇ 2Ann(R), which means that, for any nonzero cardinality κ, the second property in (4)κ,R is fulfilled, so that conditions (4)κ,R, (4)<∞,R, (4)R and (iv)R are equivalent. Lemma 3.30. (i) B(Id(R)) = {eR e ∈ E(R)}. (ii) If U ⊆ R, then U ∩ AnnR(U ) ⊆ {0}. If I ∈ Id(R), then I ∩ AnnR(I) = {0}. (iii) If U ⊆ R such that AnnR(U ) ∨ AnnR(AnnR(U )) = R, then AnnR(U ) = eR for some e ∈ E(R). Proof. (i) This statement may be known, but, for the sake of completeness, I am deriving it from the statements in this paper. Let πR = (νR −1 B(Id(R))) ◦ µR E(R): E(R) → B(Id(R)) be the composition of the inverse of the Boolean isomorphism from Corollary 3.16 to the Boolean isomorphism from Lemma 3.14, (i). For any e ∈ E(R), since µR(e) = eR/∼R = νR(eR), it follows that πR(e) = eR. Hence B(Id(R)) = πR(E(R)) = {eR e ∈ E(R)}. (ii) If U ⊆ R and x ∈ U ∩ AnnR(U ), then x · x = 0, so that x = 0 since a semiprime commutative unitary ring has no nonzero nilpotents [11, p.125,126]. Now, if I ∈ Id(R), then 0 ∈ I ∩ AnnR(I). (iii) By (ii), it follows that AnnR(U ) ∈ B(Id(R)), having AnnR(AnnR(U )) as a complement, so that AnnR(U ) = eR for some e ∈ E(R) by (i). Proposition 3.31. For any nonzero cardinality κ: (i) (5)κ,R implies (1)κ,R; (ii) (5)<∞,R implies (5)R∗ ; (iii) (5)R∗ implies (5)R; (iv) (5)κ,R, (5)<∞,R, (5)R and (5)R∗ are equivalent. Proof. (i) By Lemma 3.30, (iii). (ii) Let W ⊆ R∗, so that, by Lemma 3.19, (iii), AnnR∗ (W ) = AnnR∗ (µR(S)) for some finite subset S ⊆ R. If (5)<∞,R is fulfilled, then AnnR(S) ∨ AnnR(AnnR(S)) = R. By Lemma 3.18, it follows that AnnR∗ (W ) ∨ AnnR∗ (AnnR∗ (W )) = AnnR∗ (µR(S))∨AnnR∗ (AnnR∗ (µR(S))) = (AnnR(S)/∼R]R∗∨(AnnR(AnnR(S))/∼R]R∗ = ((AnnR(S) ∨ AnnR(AnnR(S)))/∼R ]R∗ = (R/∼R]R∗ = (⊤]R∗ = R∗. (iii) Let U ⊆ R, so that µR(S) ⊆ R∗, thus, if (5)R∗ is fulfilled, then, by Lemma 3.18, R∗ = AnnR∗ (µR(U )) ∨ AnnR∗ (AnnR∗ (µR(U ))) = (AnnR(U )/∼R]R∗∨(AnnR(AnnR(U ))/∼R]R∗ = ((AnnR(U )∨AnnR(AnnR(U )))/∼R]R∗ , so that (AnnR(U ) ∨ AnnR(AnnR(U )))/∼R = ⊤, therefore AnnR(U ) ∨ AnnR(AnnR(U )) = R by Remark 3.10. (iv) By (i), (ii), (iii), Proposition 3.21 and Corollary 3.12. Theorem 3.32. For any nonzero cardinality κ and any h, i, j ∈ 1, 5: (i) (iv)R, (h)κ,R, (i)<∞,R and (j)R are equivalent; (ii) (iv)Id(R), (h)κ,Id(R), (i)<∞,Id(R) and (j)Id(R) are equivalent; (iii) the conditions from (i) are equivalent to those from (ii) and to the equivalent conditions (iv)R∗ , (h)κ,R∗ , (i)<∞,R∗ and (j)R∗ . In particular, R is a Baer ring iff R is a strongly Baer ring iff Id(R) is a Stone lattice iff Id(R) is a strongly Stone lattice iff R∗ is a Stone lattice iff R∗ is a strongly Stone lattice. Proof. By Corollaries 3.9 and 3.12 and Propositions 3.21, 3.15, 3.28, 3.29 and 3.31. Throughout the rest of this paper, S shall be a semiprime commutative unitary ring. Remark 3.33. By Corollary 3.9 and Remark 2.42, R × S is semiprime, Id(R × S) ∼= Id(R) × Id(S) and (R × S)∗ ∼= R∗ × S ∗. Corollary 3.34. For any nonzero cardinality κ and any h, i, j ∈ 1, 5: the equivalent conditions (iv)R×S, (h)κ,R×S, (i)<∞,R×S and (j)R×S are fulfilled iff the equivalent conditions (iv)R, (h)κ,R, (i)<∞,R and (j)R, as well as the equivalent conditions (iv)S, (h)κ,S, (i)<∞,S and (j)S, are fulfilled. Concerning the semiprimality condition, recall that it does not need to be enforced for residuated lattices or bounded distributive lattices because they are congruence -- distributive and thus semiprime. 4 Conclusions It may be possible to extend Theorem 2.29 to other kinds of congruence lattices, possibly of algebras from non -- modular commutator varieties. If the congruence lattices of the algebras from such varieties cover entire varieties of bounded lattices, then all lattices from those varieties fulfill [5, Theorem 1]. The study of such extensions of Davey's Theorem remains a theme for future research. Another important research theme is finding more classes of algebras in which, given an appropriate setting (regarding definitions for annihilators and a Boolean center), Davey's Theorem holds not only for congruences, but also for elements, as in the case of bounded distributive lattices, residuated lattices and commutative unitary rings. References [1] P. Agliano, Prime Spectra in Modular Varieties, Algebra Universalis 30 (1993), 581 -- 597. [2] L. P. Belluce, Spectral Spaces and Non -- commutative Rings, Communications in Algebra 19, Issue 7 (1991), 1855 -- 1865. [3] L. P. Belluce, Spectral Closure for Non -- commutative Rings, Communications in Algebra 25, Issue 5 (1997), 1513 -- 1536. [4] J. Czelakowski, Additivity of the Commutator and Residuation, Reports on Mathematical Logic 43 (2008), 109 -- 132. [5] B. A. Davey, m -- Stone Lattices, Canadian Journal of Mathematics 24, No. 6 (1972), 1027 -- 1032. [6] R. Freese, R. McKenzie, Commutator Theory for Congruence -- modular Varieties, London Mathematical Society Lecture Note Series 125, Cambridge University Press, 1987. [7] N. Galatos, P. Jipsen, T. Kowalski, H. Ono, Residuated Lattices: An Algebraic Glimpse at Substructural Log- ics, Studies in Logic and The Foundations of Mathematics 151, Elsevier, Amsterdam/ Boston /Heidelberg /London /New York /Oxford /Paris /San Diego/ San Francisco /Singapore /Sydney /Tokyo, 2007. [8] G. Georgescu, I. Voiculescu, Some Abstract Maximal Ideal -- like Spaces, Algebra Universalis 26 (1989), 90 -- 102. [9] G. Georgescu, C. Mure¸san, The Reticulation of a Congruence -- modular Algebra, arXiv: 1706.04270 [math.RA]. [10] J. Kaplansky, Commutative Rings, First Edition: University of Chicago Press, 1974; Second Edition: Polyg- onal Publishing House, 2006. [11] J. E. Kist, Two Characterizations of Commutative Baer Rings, Pacific Journal of Mathematics 50, No. 1, 1974. [12] J. Koll´ar, Congruences and One -- element Subalgebras, Algebra Universalis 9, Issue 1 (December 1979), 266 -- 267. [13] A. Iorgulescu, Algebras of Logic as BCK Algebras, Editura ASE, Bucharest, 2008. [14] A. Joyal, Le Th´eor`eme de Chevalley − Tarski et Remarques sur l'alg`ebre Constructive, Cahiers Topol. G´eom. Diff´er., 16 (1975), 256 -- 258. [15] C. Mure¸san, Algebras of Many -- valued Logic. Contributions to the Theory of Residuated Lattices, Ph. D. Thesis, 2009. [16] C. Mure¸san, Co -- Stone Residuated Lattices, Annals of the University of Craiova, Mathematics and Computer Science Series 40 (2013), 52 -- 75. [17] D. Piciu, Algebras of Fuzzy Logic, Editura Universitaria Craiova, Craiova, 2007. [18] H. Simmons, Reticulated Rings, Journal of Algebra 66, Issue 1 (September 1980), 169 -- 192.
1610.01848	1	1610	2016-10-06T12:56:22	Automorphisms for Some "symmetric" Multiparameter Quantized Weyl Algebras and Their Localizations	[ "math.RA", "math.QA" ]	In this paper, we study the algebra automorphisms and isomorphisms for a family of "symmetric" multiparameter quantized Weyl algebras $\A$ and some related algebras in the generic case. First, we compute the Nakayama automorphism for $\A$ and give a necessary and sufficient condition for $\A$ to be Calabi-Yau. We also prove that $\A$ is cancellative. Then we determine the automorphism group for $\A$ and its polynomial extension $\E$. As an application, we solve the isomorphism problem for $\{\A\}$ and $\{\E\}$. Similar results will be established for the Maltisiniotis multiparameter quantized Weyl algebra $\B$ and its polynomial extension $\F$. In addition, we prove a quantum analogue of the Dixmier conjecture for a simple localization $(\A)_{\mathcal{Z}}$ of $\A$. Moreover, we will completely determine the algebra automorphism group for $(\A)_{\mathcal{Z}}$.	math.RA	math	AUTOMORPHISMS FOR SOME "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS AND THEIR LOCALIZATIONS XIN TANG To Professor Yingbo Zhang on the occasion of her 70th birthday n (K) and give a necessary and sufficient condition for Aq,Λ Abstract. In this paper, we study the algebra automorphisms and isomorphisms for a family of "symmetric" multiparameter quantized Weyl algebras Aq,Λ n (K) and some related algebras in the generic case. First, we compute the Nakayama automorphism for Aq,Λ n (K) to be Calabi-Yau. We also prove that Aq,Λ n (K) is cancellative. Then we determine the automorphism group for Aq,Λ n (K) and its polyno- mial extension Aq,Λ n (K[t]). As an application, we solve the isomor- phism problem for {Aq,Λ n (K[t])}. Similar results will be established for the Maltisiniotis multiparameter quantized Weyl algebra Aq,Γ n (K[t]). In addition, we prove a quantum analogue of the Dixmier conjec- ture for a simple localization (Aq,Λ n (K). Moreover, we will completely determine the algebra automorphism group for (Aq,Λ n (K) and its polynomial extension Aq,Γ n (K)} and {Aq,Λ n (K))Z of Aq,Λ n (K))Z . Introduction Ever since it was introduced by Maltisiniotis in [26], the multiparam- eter quantized Weyl algebra Aq,Γ n (K) has been extensively studied in the literature [2, 14, 17, 12, 20, 32]. When n = 1, Aq,Γ n (K) is the rank-one quantized Weyl algebra Aq 1(K) is a K−algebra gen- erated by x, y subject to the relation xy − qyx = 1, whose prime ideals of Aq 1(K) was completely determined in [2]. There has been further research on the Maltisiniotis multiparameter quantized Weyl algebras Aq,Γ n (K) of 1(K) were classified in [17]. The automorphism group of Aq 1(K). Recall that Aq Date: September 24, 2018. 2010 Mathematics Subject Classification. Primary 16W20, 16W25, 16T20, 17B37, 16E40, 16S36; Secondary 16E65. Key words and phrases. Multiparameter Quantized Weyl Algebras, Algebra Au- tomorphisms, Isomorphism Problem, Quantum Dixmier Conjecture. This research is partially supported by a research mini-grant funded by the HBCU STEM Master's Program at Fayetteville State University. 1 2 X. TANG higher ranks. For instance, Rigal determined the prime spectrum and the automorphism group for Aq,Γ n (K) in the generic case [32]. Later on, Rigal's result has been improved by Goodearl and Hartwig in [18], using a result due to Jordan [20] that Aq,Γ n (K) has a simple localiza- tion when none of qi is a root of unity. As an application, they have further solved the isomorphism problem for the family of Maltsiniotis multiparameter quantized Weyl algebras {Aq,Γ n (K)}. Another family of multiparameter quantized Weyl algebras Aq,Λ n (K) with symmetric relations has been investigated in the literature [1]. We will refer to this family of algebras as the multiparameter quantized Weyl algebras of "symmetric type" or "symmetric" multiparameter quantized Weyl algebras. When n = 1, the multiparameter quantized Weyl algebra Aq,Λ n (K) is also isomorphic to the rank-one quantized Weyl algebra Aq n (K) are closely related in many aspects. n (K) and Aq,Γ n (K) have isomorphic simple localizations if none of the major parameters q1, · · · , qn is a root of unity. The prime ideals of Aq,Λ n (K) were classified in [1] in the generic case. When λij = 1, the algebra Aq,Λ 1 (K), whose automorphism group has recently been settled in the works [8, 9] for qi 6= 1, using the methods of discriminants and Mod p. n (K) is isomorphic to the tensor product Aq1 In particular, it was proved in [20, 1] that Aq,Λ 1 (K)K ⊗ · · · ⊗K Aqn 1(K). Note that Aq,Λ n (K) and Aq,Γ Since Aq,Λ n (K) can be presented as an iterated skew polynomial al- gebra, Aq,Λ n (K) is a twisted (or skew) Calabi-Yau algebra by the result in [25]. Thus, it is of interest to further determine when Aq,Λ n (K) is in- deed Calabi-Yau [15]. We will compute the Nakayama automorphism for Aq,Λ n (K) using the methods as developed in [25, 19] and establish a necessary and sufficient condition for Aq,Λ n (K) to be a Calabi-Yau alge- bra. We will also prove that Aq,Λ n (K) is universally cancellative when none of qi is a root of unity in the sense of [6]. Similar results will be es- tablished for the Maltisiniotis multiparameter quantized Weyl algebra Aq,Γ n (K) also fits into the class of generalized Wey al- gebras [4]. The isomorphism problem for some rank-one generalized Weyl algebras has been studied in [5, 31]. In general, it would be inter- esting to determine the automorphism group for Aq,Λ n (K) and obtain an isomorphism classification for the family {Aq,Λ n (K)}. One of the main objectives of this paper is to completely determine the algebra automorphism group for Aq,Λ n (K) and solve the isomorphism problem in the case where none of q1, · · · , qn is a root of unity. We will use the fact that Aq,Λ n (K))Z with respect to the submonoid Z of Aq,Λ n (K) generated by some normal elements. We n (K) has a simple localization (Aq,Λ n (K). The algebra Aq,Λ "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 3 will also study the automorphisms and solve the isomorphism problem for the polynomial extensions {Aq,Λ n (K[t])}. Since we don't put any conditions on the parameters λij and γij, we are not able to employ the method used in [9], due to the lack of information on the discriminants in the root of unity case, despite the recent progress in [10, 13, 24, 28]. We will follow the approach used [32] and [18] in the non-root of unity case. n (K[t])} and {Aq,Γ n (K) and Aq,Γ When q = 1 and n = 1, both Aq,Λ n (K) are indeed iso- morphic to the classical first Weyl algebra A1(K), which is generated by x, y subject to the relation xy − yx = 1. The automorphism group of A1(K) was first determined by Dixmier in [11], where Dixmier also asked the question whether each K−algebra endomorphism of the n−th Weyl algebra An(K) is an algebra automorphism when the base field K is of zero characteristic. Later on, Dixmier's question has been referred to as the Dixmier conjecture. The Dixmier conjecture has been proved to be stably equivalent to the Jacobian conjecture [21, 7, 34]. There have been several works studying a quantum analogue of the Dixmier conjecture for some quantum algebras [3, 30, 22, 23, 33]. In particular, it has recently been proved in [23] that each K−algebra endomorphism of a simple localization of Aq 1(K) is indeed an algebra automorphism when q is not a root of unity. It would be of great in- terest to establish such a quantum analogue for the simple localization (Aq,Λ n (K))Z in the generic case. Another main objective of this paper is to address such a problem. We will prove that each K−algebra endo- morphism of (Aq,Λ n (K))Z is indeed an algebra automorphism under the condition that the parameters qi are independent. We will be able to determine the K−algebra automorphism group for (Aq,Λ n (K))Z in this case. 1(K)K ⊗ · · · ⊗K Aq n (K). As an application, we prove that Aq,Λ The paper is organized as follows. In Section 1, we recall the defi- nition of Aq,Λ n (K) and establish some of its basic properties. Then we determine the height-one prime ideals, the normal elements and the center of Aq,Λ n (K) is univer- sally cancellative. We also compute the Nakayama automorphism of Aq,Λ n (K) as well. In Sec- tion 2, we completely determine the automorphism group for Aq,Λ n (K) and solve the isomorphism problem. We also study the automorphism group and the isomorphism problem for Aq,Λ n (K[t]). In Section 3, we prove a quantum analogue of the Dixmier conjecture for (Aq,Λ n (K). Similar results will be established for Aq,Γ n (K))Z and describe its automorphism group. n (K[t]) and Aq,Γ 4 X. TANG 1. Multiparameter Quantized Weyl Algebras of "Symmetric Type" In this section, we first recall the definitions of the Maltisiniotis mul- tiparameter quantized Weyl algebra Aq,Γ n (K) and the "symmetric" mul- tiparameter quantized Weyl algebra Aq,Λ n (K). Then we will establish some basic properties for Aq,Λ n (K). In particular, we will determine all the height-one prime ideals of Aq,Λ n (K), and describe the normal ele- ments and the center for Aq,Λ n (K) in the case where none of qi is a root of unity. As an application, we show that Aq,Λ n (K) is universally can- cellative. We will also verify that Aq,Λ n (K) is a twisted Calabi-Yau al- gebra, and compute its Nakayama automorphism, and determine when Aq,Λ n (K) is a Calabi-Yau algebra. Similar results will be established for the Maltsiniotis multiparameter quantized Weyl algebra Aq,Γ n (K) as well. Let K denote a field and n be any positive integer and set q = (q1, · · · , qn) with qi ∈ K∗ for i = 1, · · · , n. Let Γ = (γij) be a multi- plicatively skew-symmetric n × n matrix with γij ∈ K∗. Recall that the Maltisiniotis multiparameter quantized Weyl algebra Aq,Γ n (K) is de- fined to be a K−algebra generated by x1, y1, · · · , xn, yn subject to the following relations: yiyj = γijyjyi, xixj = qiγijxjxi, xiyj = γjiyjxi, xiyj = qiγjiyjxi, i−1 ∀i, j ∈ {1, 2, · · · , n}; ∀1 ≤ i < j ≤ n; ∀1 ≤ i < j ≤ n; ∀1 ≤ j < i ≤ n; xiyi − qiyixi = 1 + (qk − 1)ykxk, i = 1, · · · , n. Xk=1 Now we recall the definition of the so-called alternative (or "sym- metric") multiparameter quantized Weyl algebra Aq,Λ n (K) from [1]. Definition 1.1. Let q = (q1, · · · , qn) with qi ∈ K∗ and Λ = (λij)1≤i,j≤n be a square matrix in Mn(K∗) with λij = λ−1 ji and λii = 1. The multiparameter quantized Weyl algebra Aq,Λ n (K) of "symmetric type" is defined to be a K−algebra generated by x1, y1, · · · , xn, yn subject to the relations: xixj = λijxjxi, xiyj = λjiyjxi, yjyi = λjiyiyj, xjyi = λijyixj, xiyi − qiyixi = 1, ∀1 ≤ i < j ≤ n; ∀1 ≤ i < j ≤ n; ∀1 ≤ i ≤ n. ✷ "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 5 As we can see, the algebra Aq,Λ n (K) has more symmetric relations compared with the Maltisiniotis multiparameter quantized Weyl alge- bra Aq,Γ n (K). First of all, we have the following proposition. n (K). Next we recall a few well-known results for Aq,Λ Proposition 1.1. The algebra Aq,Λ n (K) is a Noetherian domain with a Gelfand-Kirillov dimension of 2n. Indeed, it has a K−basis given as follows: B = {ya1 1 xb1 1 · · · yan n xbn n a1, b1, · · · , an, bn ∈ Z≥0}. n (K) Proof: From the definition of Aq,Λ n (K), it is easy to see that Aq,Λ can be presented as an iterated skew polynomial algebra as follows: Aq,Λ As a result, Aq,Λ and Aq,Λ K−basis as stated. n (K) = K[y1][x1; τ2, δ2][y2; τ3, τ4, δ4] · · · [yn; τ2n−1, δ2n−1][xn; τ2n, δ2n]. n (K) is a (both left and right) Noetherian domain n (K) has a Gelfand-Kirillov dimension of 2n. It also has the ✷ Let D be any ring and σ = (σ1, · · · , σn) be a set of commuting au- tomorphisms of D and a = (a1, · · · , an) be a set of (non-zero) elements in the center of D such that σi(aj) = aj for i 6= j. The study of the generalized Weyl algebra A = D(σ, a) of degree n over the base ring D was initiated by Bavula. For the detailed definition of generalized Weyl algebras and their properties, we refer the readers to [4] and the references therein. Next, we show that Aq,Λ n (K) can be realized as a generalized Weyl algebra. For i = 1, · · · , n, let us set zi = xiyi − yixi. It is easy to verify that zi = 1 + (qi − 1)yixi = q−1 i (1 + (qi − 1)xiyi) and zjyi = yizj if i 6= j, zjxi = xizj if i 6= j, zjxj = q−1 zjyj = qjyjzj; j xjzj; zizj = zjzi. In particular, each zi is a normal element of Aq,Λ ziAq,Λ n (K)zi for i = 1, · · · , n. n (K) = Aq,Λ n (K) in the sense that Proposition 1.2. The algebra Aq,Λ n (K) can be presented as a general- ized Weyl algebra of degree n. As a result, Aq,Λ n (K) has a K−basis B consisting of elements which are monomials in xi, yi, zi given as follows: B = {za1 1 · · · zan n xb1 1 · · · xbn n , za1 1 · · · zan n yc1 1 · · · ycn n ai, bi, ci ∈ Z≥0}. Proof: We can set the base ring as D = K[z1, · · · , zn], which is the polynomial K−algebra in z1, · · · , zn. In addition, for i = 1, · · · , n, 6 X. TANG Let Z denote the submonoid of Aq,Λ qi−1 and define σi by σi(zj) = qδij we can choose ai = zi−1 i zj for j = 1, · · · , n. Now it is straightforward to check that Aq,Λ n (K) = D(σ, a) is a generalized Weyl algebra of degree n over the base ring D. By a basic property of the generalized Weyl algebras [4], we can see that Aq,Λ n (K) has the desired K−basis. ✷ n (K) generated by the normal elements z1, · · · , zn. Obviously, Z is an Ore set. Thus we can localize Aq,Λ n (K))Z denote the localization of Aq,Λ i = xiyi − yixi in Aq,Γ n (K) generated by z′ n. The following result has been established in [1]. Theorem 1.1. If none of qi is a root of unity, then the localization (AQ,Λ n (K) with respect to Z. Let (Aq,Λ n (K) with respect to Z. Similarly, one can set z′ n (K) for i = 1, · · · , n and denote by Z ′ the submonoid of Aq,Γ 1, · · · , z′ (Aq,Λ (Aq,Γ Theorem 3.2 in [20], the localization (Aq,Γ when none of qi is a root of unity. Thus, the result follows. n,K )Z is a simple algebra. Proof: Note that it has been proved in [1] that the localization n (K))Z of Aq,Λ n (K) is isomorphic to the corresponding localization n (K))Z ′ of the Maltsiniotis quantized Weyl algebra Aq,Γ n (K). By n (K))Z ′ is a simple algebra ✷ The prime ideals of the "symmetric" multiparameter quantized Weyl algebra Aq,Λ n (K) were completely determined by Akhavizadgan and Jor- dan in [1] under certain conditions on the parameters qi and λij in the case where K is algebraically closed. However, these conditions are only needed for describing the complete prime spectrum of Aq,Λ n (K). As we can see, one can still obtain a complete classification of the height-one prime ideals for the algebra Aq,Λ n (K) under the weaker condition that none of qi is a root of unity and K is any field. Theorem 1.2. If none of qi is a root of unity, then the height-one prime ideals of Aq,Λ n (K) are exactly given as follows: {(z1), (z2), · · · , (zn)}. Proof: First of all, by Lemma 3.2 in [1], each two-sided ideal (zi) of Aq,Λ n (K) generated by the normal element zi is a completely prime ideal. Note that the proof of Lemma 3.2 in [1] does not need the extra assumption that K is an algebraically closed field. In addition, by Corollary 4.1.12 in [27], the height of the prime ideal (zi) is at most one for i = 1, · · · , n. Since Aq,Λ n (K) is a domain, the zero ideal (0) is a prime ideal of Aq,Λ n (K). Thus, the height of (zi) is indeed 1 for i = 1, · · · , n. Since none of the parameters qi is a root of unity, the localization n (K))Z is a simple algebra. Thus each non-zero prime ideal P (Aq,Λ "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 7 n ∈ P . Thus we have the inclusion: (za1 of Aq,Λ n (K) has to meet with the Ore set Z. Suppose that we have za1 1 · · · zan n ) ⊆ P . Since each zi is a normal element, we have that zj ∈ P for some j. As a result, we have the inclusion: (zj) ⊆ P for some j. If P is of height one, then we have that P = (zj) for some j. Therefore, we have completed the proof. ✷ 1 · · · zan Corollary 1.1. If none of qi is a root of unity. Then the algebra Aq,Λ n (K) has a trivial center K. And the set of all normal elements of Aq,Λ n (K) is given as follows: N = {azl1 1 · · · zln n l1, · · · , ln ≥ 0, a ∈ K}. Proof: First of all, it is easy to check that each element azi1 n (K). Conversely, let w 6= 0 be a normal element of Aq,Λ 1 · · · zin n in the set N as given above is indeed a normal element for the alge- bra Aq,Λ n (K). Then the ideal (w) of Aq,Λ n (K) generated by the normal element w is a non-zero two-sided ideal of Aq,Λ n (K). Thus the localization (w)Z of the ideal (w) with respect to the Ore set Z is a non-zero two-sided ideal of the localization (Aq,Λ n (K))Z is a simple algebra when none of qi is a root of unity, we have that (w)Z = (Aq,Λ n (K))Z . Thus we have that 1 ∈ (w)Z, which implies that w′wzl1 n (K). So w is invertible in the localization (Aq,Λ n (K))Z . Note that the only invertible elements of (Aq,Λ n (K))Z are non-zero scalar multiples of the products of integer powers of the elements zi. Since w ∈ Aq,Λ n (K), we further have that w = azl1 n (K))Z. Since the localization (Aq,Λ n for some a ∈ K∗ and l1, · · · , ln ∈ Z≥0. n = 1 for some w′ ∈ Aq,Λ 1 · · · zln 1 · · · zln Let c be in the center of Aq,Λ n (K) is also a normal element of Aq,Λ n (K). Since any central element of n for n (K). Thus c = azl1 Aq,Λ 1 · · · zln some a ∈ K∗ and l1, · · · , ln ∈ Z≥0. Since c is in the center, we have for i = 1, · · · , n. Note that cyi = yic cyi = (azl1 1 · · · zln n )yi = qli i yi(azl1 1 · · · zln n ) = qli i yic n (K) is the base field K. for i = 1, · · · , n. Thus qli i = 1 for i = 1, · · · , n. Since qi is not a root of unity, we have li = 0 for i = 1, · · · , n. Thus c = a ∈ K∗. So the center of Aq,Λ ✷ The study of Zariski cancellation problem for noncommutative alge- bras has recently been initiated in the work of Bell and Zhang. We first recall a definition on the cancellation property from [6]. Definition 1.2. Let A be a K−algebra. 8 X. TANG (1) We call A cancellative if A[t] ∼= B[t] for some K−algebra B im- plies that A ∼= B. (2) We call A strongly cancellative if, for any d ≥ 1, the isomor- phism A[t1, · · · , td] ∼= B[t1, · · · , td] for some K−algebra B im- plies that A ∼= B. (3) We call A universally cancellative if, for any finitely generated commutative K−algebra and domain R such that R/I = K for some ideal I ⊂ R and any K−algebra B, A⊗R ∼= B ⊗R implies that A ∼= B. Note that if an algebra A is universally cancellative, then it is strongly cancellative; and if an algebra A is strongly cancellative, then it is can- cellative. Theorem 1.3. If none of qi is a root of unity, then the algebra Aq,Λ is universally cancellative. As a result, Aq,Λ and cancellative. n (K) n (K) is strongly cancellative Proof: If none of qi is a root of unity, then the algebra Aq,Λ has a trivial center K. Thus Aq,Λ Proposition 1.3 in [6]. As a result, the algebra Aq,Λ cancellative and cancellative n (K) n (K) is universally cancellative by n (K) is strongly ✷ Similarly, we have the following result concerning the cancellation property for the Maltisiniotis multiparameter quantized Weyl algebra Aq,Γ n (K). Theorem 1.4. If none of qi is a root of unity, the Maltisiniotis mul- tiparameter quantized Weyl algebra Aq,Γ n (K) is universally cancellative; and thus it is strongly cancellative and cancellative. Proof: It can be verified in a similar fashion that the center of the Maltisiniotis multiparameter quantum Weyl algebra Aq,Γ n (K) is also K when none of the parameters qi is a root of unity. Using Proposition 1.3 in [6], we know that the algebra Aq,Γ n (K) is universally cancellative. As a result, Aq,Γ ✷ Recall from [25, 29] that a K−algebra A is called µ−twisted (or skew) Calabi-Yau of dimension d, where µ is a K−algebra automorphism of A and d ∈ Z≥0, provided that n (K) is strongly cancellative and cancellative. (1) A is homologically smooth in the sense that as a module over Ae : = A⊗KAop, A has a finitely generated projective resolution of finite length. (2) Exti Ae(A, Ae) ∼= 0 or Aµ if i = d. "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 9 When the above two conditions hold, the automorphism µ is called the Nakayama automorphism of A. In general, the Nakayama automor- phism µ of an algebra A is unique up to an inner automorphism of A. An algebra A is called Calabi-Yau (or CY, for short) [15] if A is twisted Calabi-Yau and the Nakayama automorphism µA of A is inner . It has been proved that any Ore extension (or skew polynomial alge- bra) E = A[x; σ, δ] is a twisted Calabi-Yau algebra of dimension d + 1 if A is a twisted Calabi-Yau algebra of dimension d in [25], where a formula for computing the Nakayama automorphism µ of E is also es- tablished. It follows immediately that any iterated Ore extension (or iterated skew polynomial algebra) is a twisted Calabi-Yau algebra. The Nakayama automorphisms of iterated Ore extensions (or iterated skew polynomial algebras) have been further studied in [19], where a formula of computing the Nakayama automorphism for any reversible, diago- nalized iterated Ore extension is developed. In particular, it is proved in [19] that the Nakayama automorphism of any symmetric CGL ex- tension is an automorphism defined by the conjugation of a normal ele- ment. Since Aq,Λ n (K) is indeed a twisted Calabi-Yau algebra. We next compute the Nakayama automorphism for Aq,Λ n (K) and provide a necessary-sufficient condition for Aq,Λ In addition, we will iden- tify a normal element that defines the Nakayama automorphism µ of Aq,Λ n (K) is an iterated skew polynomial algebra, Aq,Λ n (K) to be a Calabi-Yau algebra. n (K). Theorem 1.5. The "symmetric" multiparameter quantized Weyl alge- bra Aq,Λ n (K) is a twisted Calabi-Yau algebra. The Nakayama automor- phism µ of Aq,Λ n (K) is given as follows: µ(xi) = qixi, µ(yi) = q−1 i yi for i = 1, · · · , n. As a result, the algebra Aq,Λ n (K) is a Calabi-Yau algebra if and only if qi = 1 for i = 1, · · · , n. Moreover, the Nakayama automorphism µ of Aq,Λ n (K) defined by the conjugation of z = z1z2 · · · zn. That is, n (K) is the automorphism of Aq,Λ µ(xi) = z−1xiz, µ(yi) = z−1yiz for i = 1, · · · , n. Proof: We prove this statement using induction on the index n. For simplicity, we will sometimes denote the algebra Aq,Λ n (K) by A instead. When n = 1, we have that A = Aq1 1 (K), the rank-one quantized Weyl algebra, which is a K−algebra generated by x1 and y1 subject to the only relation: x1y1−q1y1x1 = 1. In this case, we can present the algebra A as an iterated skew polynomial algebra in two different ways. First 10 X. TANG of all, we have that A = K[y1][x1; σ1, δ1] where σ1(y1) = q1y1 and δ1(y1) = 1. Applying Theorem 3.3 in [25], we have that A is a twisted Calabi-Yau algebra and its Nakayama automorphism µ is given as follows: µ(y1) = q−1 1 y1, µ(x1) = ax1 + b for some invertible element a ∈ K∗ and b in K[y1]. In addition, one can also present the algebra A as an iterated skew polynomial algebra as follows: where σ′ Yau algebra with its Nakayama automorphism µ defined by: 1 . Thus, A is a twisted Calabi- 1 x1 and δ′ 1(x1) = q−1 A = K[x1][y1; σ′ 1(x1) = −q−1 1, δ′ 1] µ(x1) = q1x1, µ(y1) = cy1 + d for some invertible element c ∈ K∗ and d in K[x1]. Since the Nakayama automorphism of A is unique up to an inner automorphism of A and the set of invertible elements in A is the set K∗, we have that b = d = 0 and a = q1 and c = q−1 1 . Suppose the statement is true for the subalgebra B of Aq,Λ n (K) gener- ated by x1, · · · , xn−1, y1, · · · , yn−1. That is, the algebra B is a twisted Calabi-Yau algebra with its Nakayama automorphism µ defined as fol- lows: for i = 1, · · · , n − 1. µ(xi) = qixi, µ(yi) = q−1 i yi Next we will present Aq,Λ n (K) as an iterated skew polynomial algebra over the base algebra B in two ways. We will denote the Nakayama automorphism of Aq,Λ n (K) = B[yn; σ1, δ1][xn; σ2, δ2] with n (K) by µ. First of all, we can have that Aq,Λ σ1(xi) = λinxi, for i = 1, · · · , n − 1 and σ1(yi) = λniyi, δ1(xi) = δ1(yi) = 0 σ2(yn) = qnyn, δ2(yn) = 1 and σ2(xi) = λnixi, σ2(yi) = λinyi, δ2(xi) = δ2(yi) = 0 for i = 1, · · · , n − 1. Since δ1 = 0, by Theorem 3.2 and Remark 3.4 in [25], the Nakayama automorphism µ of Aq,Λ n (K) is given as follow: in qixi = qixi, µ(yi) = λ−1 in λ−1 µ(xi) = λ−1 i yi = q−1 ni λ−1 ni q−1 i yi for i = 1, · · · , n − 1 and µ(yn) = aq−1 n yn, µ(xn) = cxn + d "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 11 for some a, c ∈ K∗ and d ∈ B[yn; σ1, δ1]. Second of all, we can have that Aq,Λ n (K) = B[xn; σ′ σ′ 1(xi) = λnixi, for i = 1, · · · , n − 1 and 1][yn; σ′ σ1(yi) = λinyi, δ′ 1(xi) = δ′ 1(yi) = 0 2] with 1, δ′ 2, δ′ 2(xn) = q−1 σ′ n xn, 2(xn) = −q−1 δ′ n and σ2(xi) = λinxi, σ2(yi) = λniyi, δ2(xi) = δ2(yi) = 0 for i = 1, · · · , n − 1. Therefore, the Nakayama automorphism of Aq,Λ n (K) is given as follows: µ(xi) = λ−1 in λ−1 ni qixi = qixi, µ(yi) = λ−1 ni λ−1 in q−1 i yi = q−1 i yi for i = 1, · · · , n − 1 and µ(xn) = a′qnxn, µ(yn) = c′yn + d′ for some a′, c′ ∈ K∗ and d′ ∈ B[xn; σ′ B[yn; σ1, δ1] of Aq,Λ polynomial algebra from the base field K as follows: 1]. Note that the subalgebra n (K) can also be constructed as an iterated skew 1, δ′ B[yn; σ1, δ1] = K[yn][x1; σ′′′ 2 , δ′′′ 2 ][y1; σ′′′ 3 , δ′′′ 3 ] · · · [yn−1; σ′′′ 2n−1, δ′′′ 2n−1]. Using Theorem 3.2 in [25] repeatedly, we have µ(yn) = q−1 n yn + e for some e ∈ B[yn; σ1, δ1]. Similarly, we can prove that µ(xn) = qnxn + f for some f ∈ B[xn; σ′ 1]. Comparing the various expressions of the Nakayama automorphism µ for the algebra Aq,Λ n (K), we have that d = d′ = e = f = 0 and a′qn = c = qn and c′ = aq−1 n . Thus the Nakayama automorphism µ of Aq,Λ n = q−1 n (K) is given as follows: 1, δ′ µ(xi) = qixi, µ(yi) = q−1 i yi for i = 1, · · · , n. Therefore, Aq,Λ for i = 1, · · · , n. n (K) is Calabi-Yau if and only if qi = 1 gation of z = z1 · · · zn. So we have completed the proof. It is straightforward to check that µ is indeed defined by the conju- ✷ We can determine the Nakayama automorphism µ for the multi- n (K) in a similar parameter Maltsiniotis quantized Weyl algebra Aq,Γ fashion. In particular, we have the following result. Theorem 1.6. The Maltsiniotis multiparameter quantized algebra Aq,Γ is a twisted Calabi-Yau algebra with a Nakayama automorphism µ de- fined as follows: n (K) µ(xi) = qn+1−i i xi, µ(yi) = q−n−1+i i yi for i = 1, · · · , n. As a result, the algebra Aq,Γ algebra if and only if each qi is an (n + 1 − i)-th root of unity. n (K) is a Calabi-Yau In 12 X. TANG particular, the Nakayama automorphism µ of Aq,Γ phism of Aq,Γ z = (x1y1 − y1x1) · · · (xnyn − ynxn) as follows: n (K) is the automor- n (K) defined by the conjugation of the normal element µ(xi) = z−1xiz, µ(yi) = z−1yiz for i = 1, · · · , n. Proof: The proof is similar to the one for Theorem 1.5, and we ✷ will not repeat the details here. Remark 1.1. The Nakayama automorphism µ of Aq,Λ n (K)) is completely determined by the parameters q1, · · · , qn, and it is inde- pendent of the rest parameters λij (or γij). n (K) (or Aq,Γ Remark 1.2. One can also present the algebra Aq,Λ n (K)) as a reversible, diagonalized iterated Ore extension and then apply the formula as developed in [19] to compute the Nakayama automorphism of Aq,Λ n (K)). Instead, we have closely followed the original idea as established in [25] for the purpose of providing more details. n (K) (or Aq,Γ n (K) (or Aq,Γ 2. Automorphisms and Isomorphisms for "Symmetric" Quantized Weyl Algebras ✷ n (K[t]) and classify the algebras Aq,Λ In this section, we determine the automorphism groups for Aq,Λ n (K) and Aq,Λ n (K[t]) up to algebra isomorphisms in the case where none of q1, · · · , qn is a root of unity. Note that we will not put any condition on the rest parameters λij. We have some similar results on the automorphisms and isomor- phisms for Aq,Γ n (K) and Aq,Λ n (K[t]). First of all, we establish a useful lemma, which is similar to the ones used in [32] and [16]. We will further set z0 = 1. Lemma 2.1. For any 1 ≤ i ≤ n, let a, b ∈ Aq,Λ n (K)\K such that ab = a0z0 + n Xj=1 αjzj with αj ∈ K and αi 6= 0. Then there exist λ, µ ∈ K∗ such that or a = λyi, b = µxi a = λxi, b = µyi. "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 13 Proof: For any 1 ≤ i ≤ n, we can regard the algebra Aq,Λ n (K) as a N−filtered algebra by assigning the degrees to its generators as follows: deg(xi) = 1 = deg(yi), deg(xj) = deg(yj) = 0 for all j 6= i. Note that the above assignment of degrees is compatible with the defining relations of Aq,Λ n (K), and the degree of any element in Aq,Λ n (K) is defined to be its degree when considered as an element in the corresponding graded algebra. Since qk 6= 1 for k = 1, · · · , n, we have j=0 αjzj with αi 6= 0, we have that deg(ab) = 2. Due to the degree consideration, we have the following three possibilities: that deg(zj) = 0 for j 6= i and deg(zi) = 2. Since ab = Pn or or deg(a) = 2, deg(b) = 0; deg(a) = 0, deg(b) = 2; deg(a) = deg(b) = 1. i + gxiyi + hy2 In the first two cases, we have that a = f x2 i or b = f x2 i for some elements f, g ∈ Aq,Λ n (K) whose terms do not involve any positive powers of xi and yi at all. As a result, we should have that either b or a is a scalar. This cannot happen. Thus only the third case is possible. That is, we have i + gxiyi + hy2 a = f xi + gyi, b = f ′xi + g′yi for some f, g, f ′, g′ ∈ Aq,Λ powers of xi and yi. It is easy to see that we have n (K) whose terms do not involve any positive f f ′ = 0, gg′ = 0. Suppose that f = 0, then g 6= 0, which implies that g′ = 0. Since b 6= 0, we have that f ′ 6= 0. So we have a = gyi, b = f ′xi. Note that ab is a K−linear combination of z1, · · · , zn, where zj = xjyj − yjxj for j = 0, · · · , n. Thus we have that both g and f ′ are in K∗. Moreover, we have a = λyi, b = µxi. If f ′ = 0, then we have that g′ 6= 0. Thus we have that g = 0. So we have a = f xi, As a result, we have the following: b = g′yi. a = λxi, b = µyi 14 X. TANG for some λ, µ ∈ K∗. So we have completed the proof. ✷ Lemma 2.2. Let ϕ be a K−algebra automorphism for Aq,Λ n (K). Then for each i = 1, · · · , n, we have that ϕ(zi) = (zj) for some j. In partic- ular, we have that or ϕ(xi) = αxj, ϕ(yi) = βyj ϕ(xi) = αyj, ϕ(yi) = βxj for some j and α, β ∈ K∗. Proof: Since ϕ is a K−algebra automorphism of Aq,Λ mutes all the height-one prime ideals of Aq,Λ n (K), it per- n (K). As a result, we have ϕ((zi)) = (zj) for some j. Thus ϕ(zi) = f zj for some f ∈ Aq,Λ have that φ−1((zj)) = (zi). Thus ϕ−1(zj) = gzi for some g ∈ Aq,Λ As a result, we have the following n (K). Conversely, we n (K). zi = ϕ−1(f )gzi which implies that ϕ−1(f )g = 1. Since the only invertible elements of n (K) are in K∗, we have that φ−1(f ) ∈ K∗ and g ∈ K∗. Thus we Aq,Λ have that ϕ(zi) = λzj for some λ ∈ K∗. Since xiyi = qizi−1 qi−1 , we have the following ϕ(xi)ϕ(yi) = qiλzj − 1 qi − 1 = qiλ qi − 1 zj − 1 qi − 1 . By Lemma 2.1, we have either σ(xi) = αxj and ϕ(yi) = βyj, or ϕ(xi) = αyj and ϕ(yi) = βxj for some α, β ∈ K∗ and j ∈ {1, · · · , n}. ✷ Theorem 2.1. Let ϕ ∈ AutK(Aq,Λ tion σ of the set {1, · · · , n} and αi ∈ K∗ for i = 1, · · · , n such that n (K)). Then there exist a permuta- ϕ(xi) = αixσ(i), ϕ(yi) = α−1 i yσ(i), qi = qσ(i), or ϕ(xi) = αiyσ(i), ϕ(yi) = −q−1 i α−1 i xσ(i), qi = q−1 σ(i). Proof: Since ϕ is a K−algebra automorphism of Aq,Λ n (K), we have that ϕ(xi) = αxi and ϕ(yi) = βyi or ϕ(xi) = αyj and ϕ(yi) = βxj for some j ∈ {1, · · · , n} and α, β ∈ K∗. Suppose that ϕ(xi) = αxj, ϕ(yi) = βyj. Then we have αβ(xjyj − qiyjxj) = 1, αβ(xjyj − qjyjxj) = αβ. "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 15 As a result, we have that αβ(qj −qi)yjxj = (1−αβ), which implies that αβ = 1, and qi = qj. So we have that β = α−1 and qi = qj. Suppose that ϕ(xi) = αyj and ϕ(yi) = βxj. Then we have the following αβ(yjxj − qixjyj) = 1, αβ(xjyj − qjyjxj) = αβ. As a result, we have that αβ(1 − qiqj)yjxj = (1 + qiαβ), which implies that qi = q−1 In either case, we can define a permutation σ of {1, · · · , n} by σ(i) = j. So the result has been proved. ✷ and β = −q−1 i α−1. j The following two corollaries follow directly from Theorem 2.1 and we state them here without proofs. Corollary 2.1. If qi 6= q±1 j for i 6= j, then AutK(Aq,Λ n (K)) ∼= (K∗)n. ✷ Corollary 2.2. If qi = qj for all i, j, then AutK(Aq,Λ where G is a finite subgroup of AutK(Aq,Λ morphisms Aq,Λ are compatible with the defining relations of Aq,Λ n (K)) ∼= G ⋉ (K∗)n n (K)) consisting of the auto- n (K) defined by the permutations of {1, · · · , n}, which n (K). ✷ Next we are going to give a more explicit description for the auto- morphism group of Aq,Λ n (K). Theorem 2.2. Let ϕ : Aq,Λ phism of Aq,Λ and a permutation σ : {1, 2, · · · , n} −→ {1, 2, · · · , n} such that n (K) be a K−algebra automor- n (K). Then there exist a partition {1, 2, · · · , n} = P1 ∪ P2 n (K) −→ Aq,Λ (1) for any i, j ∈ P1, we have λij = λσ(i)σ(j), qi = qσ(i), qj = qσ(j). (2) for i ∈ P1, j ∈ P2, we have λij = λ−1 σ(i)σ(j), qi = qσ(i), qj = q−1 σ(j). (3) for i, j ∈ P2, we have λij = λσ(i)σ(j), qi = q−1 σ(i), qj = q−1 σ(j). In particular, the automorphism ϕ is defined as follows: ϕ(xi) = αixσ(i), ϕ(xj) = αjyσ(j), ϕ(yj) = −q−1 ϕ(yi) = α−1 i α−1 i yσ(i), j xσ(j), ∀i ∈ P1; ∀j ∈ P2. Conversely, for any given partition P1 ∪ P2 of {1, · · · , n} and any permutation σ of {1, 2, · · · , n} satisfying the above conditions, we can 16 X. TANG define a K−algebra automorphism ϕ of Aq,Λ ϕ(yi) = α−1 j α−1 ϕ(xj) = αjyσ(j), n (K) as follows: ∀i ∈ P1; i yσ(i), j xσ(j), ∀j ∈ P2. ϕ(xi) = αixσ(i), ϕ(yj) = −q−1 Proof: Let ϕ be a K−algebra automorphism of Aq,Λ n (K) and σ be the corresponding permutation of {1, · · · , n} associated to ϕ. Let us set P1 = {i ∈ {1, · · · , n}ϕ(xi) = αixσ(i), ϕ(yi) = α−1 i yσ(i)} and P2 = {i ∈ {1, · · · , n}ϕ(xi) = αiyσ(i), ϕ(yi) = −q−1 i xσ(i)}. Then P1 ∪ P2 = {1, · · · , n} is a partition of the set {1, 2, · · · , n}. i α−1 Let i, j ∈ P1, then we have the following: qi = qσ(i), qj = qσ(j) and and ϕ(xi) = αixσ(i), ϕ(yi) = α−1 i yσ(i) ϕ(xj) = αjxσ(j), ϕ(yj) = α−1 j yσ(j). From the definition of Aq,Λ tions among the generators xi, xj, yi, and yj: n (K), we have the following commuting rela- xixj = λijxjxi, xiyj = λjiyjxi, yiyj = λijyjyi, yixj = λjixjyi. Applying the automorphism ϕ to both sides of the above equations, we will have the following: xσ(i)xσ(j) = λijxσ(j)xσ(i), xσ(i)yσ(j) = λjiyσ(j)xσ(i), yσ(i)yσ(j) = λijyσ(j)yσ(i), yσ(i)xσ(j) = λjixσ(j)yσ(i). By the definition of Aq,Λ n (K), we also have the following: xσ(i)xσ(j) = λσ(i)σ(j)xσ(j)xσ(i), xσ(i)yσ(j) = λσ(j)σ(i)yσ(j)xσ(i), yσ(i)yσ(j) = λσ(i)σ(j)yσ(j)yσ(i), yσ(i)xσ(j) = λσ(j)σ(i)xσ(j)yσ(i). Therefore, we have λij = λσ(i)σ(j) and qi = qσ(i) and qj = qσ(j) for any i, j ∈ P1. Let i ∈ P1 and j ∈ P2, then we have the following: and qi = qσ(i), qj = q−1 σ(j) ϕ(xi) = αixσ(i), ϕ(yi) = α−1 i yσ(i) "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 17 and ϕ(xj) = αjyσ(j), ϕ(yj) = −q−1 i α−1 j xσ(j). Applying the automorphism ϕ to the following equations: xixj = λijxjxi, xiyj = λjiyjxi, yiyj = λijyjyi, yixj = λjixjyi, we have the following equations: xσ(i)yσ(j) = λijyσ(j)xσ(i), xσ(i)xσ(j) = λjixσ(j)xσ(i), yσ(i)xσ(j) = λijxσ(j)yσ(i), yσ(i)yσ(j) = λjiyσ(j)yσ(i). From the definition of Aq,Λ xσ(i)yσ(j) = λσ(j)σ(i)yσ(j)xσ(i), xσ(i)xσ(j) = λσ(i)σ(j)xσ(j)xσ(i), n (K), we also have the following: yσ(i)xσ(j) = λσ(j)σ(i)xσ(j)yσ(i), yσ(i)yσ(j) = λσ(i)σ(j)yσ(j)yσ(i). As a result, we have that λij = λσ(j)σ(i) = λ−1 Let i, j ∈ P2, then we have the following: qj = q−1 σ(j) qi = q−1 σ(i), σ(i)σ(j) for i ∈ P1 and j ∈ P2. and and ϕ(xi) = αiyσ(i), ϕ(yi) = −q−1 i α−1 i xσ(i) ϕ(xj) = αjyσ(j), ϕ(yj) = −q−1 j α−1 j xσ(j). Apply the automorphism ϕ to both sides of the following equations: xixj = λijxjxi, xiyj = λjiyjxi, yiyj = λijyjyi, yixj = λjixjyi, and we have the following: yσ(i)yσ(j) = λijyσ(j)yσ(i), yσ(i)xσ(j) = λjixσ(j)xσ(i), xσ(i)xσ(j) = λijxσ(j)xσ(i), xσ(i)yσ(j) = λjiyσ(j)xσ(i). Additionally, we have the following: yσ(i)yσ(j) = λσ(i)σ(j)yσ(j)yσ(i), yσ(i)xσ(j) = λσ(j)σ(i)xσ(j)yσ(i), xσ(i)xσ(j) = λσ(i)σ(j)xσ(j)xσ(i), xσ(i)yσ(j) = λσ(j)σ(i)yσ(j)xσ(i). After comparing these equations, we have that λij = λσ(i)σ(j) for i, j ∈ P2. Conversely, it is straightforward to verify that a partition P1 ∪ P2 of {1, · · · , n} and a permutation σ of {1, · · · , n} satisfying the conditions can define an algebra automorphism ϕ of Aq,Λ n (K). We will not state the details here. ✷ 18 X. TANG Next, we solve the isomorphism problem for the family of "sym- metric" multiparameter quantized Weyl algebras. Given two data sets (n, q, Λ) and (m, q′, Λ′), we can define two "symmetric" multiparameter quantized Weyl algebras Aq,Λ m (K). We have the following result. n (K) and Aq′,Λ′ m, y′ n (K) and Aq′,Λ′ 1, · · · , x′ i is a root of unity, then Aq,Λ Theorem 2.3. Let Aq,Λ m (K) be two "symmetric" multi- parameter quantized Weyl algebras generated by x1, y1, · · · , xn, yn and x′ 1, y′ m respectively. Assume that none of the parameters qi and q′ m (K) as a K−algebra if and only if m = n and there exist a partition P1 ∪ P2 of {1, · · · , n} and a permutation σ of {1, 2, · · · , n} such that σ(j), λij = λ′ σ(i), q−1 (1) for i, j ∈ P1, we have qi = q′ (2) for i ∈ P1 and j ∈ P2, we have qi = q′ n (K) is isomorphic to Aq′,Λ′ σ(i), qj = q′ σ(j), λij = j = q′ σ(i)σ(j). λ′ σ(j)σ(i). (3) for i, j ∈ P2, we have q−1 i = q′ σ(i), q−1 j = q′ In particular, the corresponding isomorphism ϕ : Aq,Λ is defined as follows: σ(j), λij = λ′ σ(i)σ(j). n (K) −→ Aq′,Λ′ m (K) ϕ(xi) = αix′ σ(i), ϕ(yi) = α−1 i y′ σ(i) for i ∈ P1 and for i ∈ P2. ϕ(xi) = αiy′ σ(i), ϕ(yi) = −q−1 i α−1 i x′ σ(i) m (K) is 2m. If Aq,Λ Proof: First of all, the Gelfand-Kirillov dimension of Aq,Λ n (K) is 2n and the Gelfand-Kirillov dimension of Aq′,Λ′ n (K) is isomorphic to Aq′,Λ′ m (K) as a K−algebra, then they have the same Gelfand-Kirillov dimension. Thus we have m = n. The rest of the proof is to mimic the ones for Theorem 2.1 and Theorem 2.2 word in word. We will not repeat the details here. ✷ Let A be any K−algebra. A K−algebra automorphism h of the polynomial extension A[t] = A ⊗K K[t] is said to be triangular if there are a g ∈ AutK(A) and c ∈ K∗ and r in the center of A such that and h(w) = ct + r h(a) = g(a) ∈ A for any a ∈ A. For more details, we refer the reader to [8]. Denote by Aq,Λ n (K[t]) = Aq,Λ n (K)⊗KK[t] and Aq′,Λ′ m (K[t]) = Aq′,Λ′ m (K)⊗K K[t]. Then we have the following result. "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 19 n (K[t]) −→ Aq,Λ n (K[t]) be a K−algebra au- n (K[t]). Then there exist a partition P1 ∪ P2 of Theorem 2.4. Let ϕ : Aq,Λ tomorphism of Aq,Λ {1, · · · , n} and a permutation σ of {1, 2, · · · , n} such that σ(j), λij = λ′ σ(i), q−1 (1) for i, j ∈ P1, we have qi = q′ (2) for i ∈ P1 and j ∈ P2, we have qi = q′ σ(i), qj = q′ σ(j), λij = j = q′ σ(i)σ(j). λ′ σ(j)σ(i). (3) for i, j ∈ P2, we have q−1 i = q′ In particular, ϕ is defined as follows: σ(i), q−1 j = q′ σ(j), λij = λ′ σ(i)σ(j). ϕ(xi) = αixσ(i), ϕ(yi) = α−1 i yσ(i) ϕ(xi) = αiyσ(i), ϕ(yi) = −q−1 i α−1 i xσ(i) for i ∈ P1 and for i ∈ P2 and for some αi, a ∈ K∗ and b ∈ K. ϕ(t) = at + b Proof: Note that the localization of Aq,Λ n (K[t]) with respect to the Ore set K[t] − {0} is isomorphic to the algebra Aq,Λ n (K(t)), which is a "symmetric" multiparameter quantized Weyl algebra over the base field K(t). Since none of qi is a root of unity, Aq,Λ n (K(t)) is a simple K(t)−algebra. Thus the center of Aq,Λ n (K[t]) is K[t]. Since the center of Aq,Λ n (K[t]) is preserved by ϕ, we can restrict ϕ to a K−algebra au- tomorphism of K[t]. Thus we have ϕ(t) = at + b for some a ∈ K ∗ and b ∈ K. Let us denote the restriction of ϕ to K[t] by ϕ0. Note that the inverse of ϕ0 can be extended to a K−algebra ψ of Aq,Λ n (K[t]) as follows: ψ(t) = ϕ−1 0 (t), ψ(w) = w for any w ∈ Aq,Λ n (K). Now the composition ψ◦ϕ is actually a K[t]−algebra automorphism of Aq,Λ n (K[t]). Thus we can extend ψ◦ϕ to a K(t)−algebra automorphism of Aq,Λ n (K(t)) and we still denote the extension by ψ ◦ ϕ. By Theorem 2.2, there exist a partition P1 ∪ P2 of {1, · · · , n} and a permutation σ of {1, 2, · · · , n} such that ψ ◦ ϕ(xi) = αixσ(i), ψ ◦ ϕ(yi) = α−1 i yσ(i) with αi ∈ K(t)∗ for i ∈ P1 and ψ ◦ ϕ(xi) = αiyσ(i), ψ ◦ ϕ(yi) = −q−1 with αi ∈ K(t)∗ for i ∈ P2. Since ψ ◦ ϕ(xi) ∈ Aq,Λ αi ∈ K∗. Thus, we are done with the proof. i xσ(i) i α−1 n (K), we have that ✷ 20 X. TANG Theorem 2.5. Any K−algebra automorphism of Aq,Λ gular. In particular, we have the following: n (K[t]) is trian- AutK(Aq,Λ n (K)[t]) = (cid:18) AutK(Aq,Λ 0 n (K)) K K∗ (cid:19) . As a result, if Z ⊂ K, then we have LNDer(Aq,Λ LNDer(Aq,Λ n (K)) is the set of all locally nilpotent derivations of Aq,Λ n (K)) = {0}, where n (K). Proof: By Theorem 2.5, every K−algebra automorphism ϕ of Aq,Λ n (K[t]) can be restricted to Aq,Λ n (K) and ϕ(t) = at + b for some a ∈ K∗ and b ∈ K. Thus ϕ is triangular. As a result, we have the following: AutK(Aq,Λ n (K[t])) = (cid:18) AutK(Aq,Λ 0 n (K)) K K∗ (cid:19) . Let ∂ be a locally nilpotent derivation of Aq,Λ can define a K(t)−algebra automorphism ϕ of Aq,Λ n (K). Since Z ⊂ K, we n (K(t)) as follows: ϕ(t) = t, ϕ(x) = ∞ Xi=0 ti i! ∂i(x) for any x ∈ Aq,Λ for some j. Then we have n (K). If ∂ 6= 0, then ∂(xi) 6= 0 for some i or ∂(yj) 6= 0 ϕ(xi) = xi + t∂(xi) + other terms involving higher powers of t or ϕ(yj) = yj + t∂(yj) + other terms involving higher powers of t. This is a contradiction to the description of the K(t)−automorphisms of Aq,Λ n (K(t)). Therefore, we have ∂ = 0. So we have completed the proof. ✷ Theorem 2.6. The algebra Aq,Λ m (K[t]) if and only if m = n, and there exist a partition P1 ∪ P2 of {1, · · · , n} and a permutation σ of {1, 2, · · · , n} such that σ(i), qj = q′ n (K[t]) is isomorphic to Aq′,Λ′ (1) for i, j ∈ P1, we have qi = q′ (2) for i ∈ P1 and j ∈ P2, we have qi = q′ σ(j), λij = λ′ σ(i), q−1 σ(j), λij = j = q′ σ(i)σ(j). λ′ σ(j)σ(i). (3) for i, j ∈ P2, we have q−1 i = q′ σ(i), q−1 j = q′ σ(j), λij = λ′ σ(i)σ(j). In particular, the corresponding isomorphism ϕ : Aq,Λ is defined as follows: n (K[t]) −→ Aq′,Λ′ m (K[t]) ϕ(xi) = αix′ σ(i), ϕ(yi) = α−1 i y′ σ(i) "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 21 for i ∈ P1 and for i ∈ P2 and ϕ(xi) = αiy′ σ(i), ϕ(yi) = −q−1 i α−1 i x′ σ(i) for some a ∈ K∗ and b ∈ K. ϕ(t) = at + b Proof: Note that Aq,Λ n (K) is cancellative. Then Aq,Λ n (K) is isomorphic to Aq′,Λ′ n (K[t]) is isomor- m (K). phic to Aq′,Λ′ Now the result follows from Theorem 2.3. m (K[t]) if and only if Aq,Λ Let Aq,Γ n (K) and Aq′,Γ ✷ m (K) be two Maltsiniotis multiparameter quan- 1, · · · , x′ m, y′ tized Weyl algebras generated by x1, y1, · · · , xn, yn and x′ m respectively. Let us set Aq,Γ m (K[t]) = Aq′,Γ m (K) ⊗K K[t]. We have the following result for the automorphism group of the polynomial extension of the Maltsiniotis multiparameter quantized Weyl algebra. n (K) ⊗K K[t] and Aq ′,Γ′ n (K[t]) = Aq,Γ 1, y′ Theorem 2.7. Assume that none of qi, i = 1, 2, · · · , n is a root of unity. Then we have the following results. • If ϕ is a K−algebra automorphism of Aq,Γ n (K[t]), then there ex- ists a tuple (µ1, · · · , µn, a, b) ∈ (K∗)n+1 × K such that ϕ(t) = at + b and ϕ(xi) = µixi, ϕ(yi) = µ−1 i yi for each 1 ≤ i ≤ n. • For any a ∈ K∗ and b ∈ K, one can define a K−algebra auto- morphism ϕa,b of Aq,Γ n (K[t]) as follows: ϕa,b(t) = at + b, ϕa,b(xi) = xi, ϕa,b(yi) = yi for each 1 ≤ i ≤ n. • We have that AutK(Aq,Γ n (K[t])) ∼= (K∗)n ⋊ G where G is the subgroup of AutK(Aq,Γ n (K[t])) consisting of these ϕa,b. Proof: The proof is similar to the one for Theorem 2.4 and we ✷ will not repeat the details here. Moreover, we have the following theorem. Theorem 2.8. Any K−algebra automorphism of Aq,Γ gular. In particular, we have the following: n (K[t]) is trian- AutK(Aq,Γ n (K[t])) = (cid:18) AutK(Aq,Γ 0 n (K)) K K∗ (cid:19) . As a result, if Z ⊂ K, then we have LNDer(Aq,Γ LNDer(Aq,Γ n (K)) is the set of all locally nilpotent derivations of Aq,Γ n (K)) = {0}, where n (K). 22 X. TANG We skip the details. Proof: The proof is the same as the one used for Theorem 2.5. ✷ Next, we solve the isomorphism classification problem for the family of polynomial extensions {Aq,Λ n (K[t])} based on a result due to Good- earl and Hartwig [18]. For convenience, we will follow the notation used in [18]. Theorem 2.9. Assume none of q1, · · · , qn and q′ unity. Then Aq,Γ and only if n (K[t]) is isomorphic to Aq′,Γ′ 1, · · · , q′ m is a root of m (K[t]) as a K−algebra if • m = n; • There exists a sign vector ε ∈ {−1, 1}n such that we have q′ i = qεi i , ∀1 ≤ i ≤ n and γij γji q−1 i γji qiγij if (εi, εj) = (1, 1) if (εi, εj) = (−1, 1) if (εi, εj) = (1, −1) if (εi, εj) = (−1, −1) ∀1 ≤ i ≤ j ≤ n. γ′ ij =   If the above conditions are satisfied, then for any µ ∈ (K∗)n and ε ∈ {±1}n and a ∈ K∗ and b ∈ K, one can define a unique K−algebra isomorphism ϕµ,ε,a,b : Aq,Λ n (K[t]) −→ Aq′,Γ′ m (K[t]) by and and ϕµ,ε,a,b(t) = at + b i, µiy′ i, ϕµ,ε,a,b(xi) = (cid:26) µix′ ϕµ,ε,a,b(yi) = (cid:26) λiµ−1 i y′ i, −λiµ−1 i x′ i, εi = 1 εi = −1 εi = 1 εi = −1 for each i ∈ {1, · · · , n} with λ = λ(ε) ∈ (K∗)n recursively defined as follows: λ0 = 1, λi = q(εi−1)/2λi−1. Proof: Note that the algebra Aq,Γ n (K[t]) ∼= Aq′,Γ′ that Aq,Γ the result follows from Theorem 5.1 in [18]. m (K[t]) if and only if Aq,Γ n (K) is cancellative. Thus we have m (K). Now ✷ n (K) ∼= Aq′,Γ 3. The Quantum Dixmier Conjecture for (Aq,Λ n (K))Z In this section, we prove a quantum analogue of the Dixmier conjec- n (K) under the condi- n = 1 implies i1 = i2 = · · · = in = 0. In particular, n (K))Z is an ture for the simple localization (Aq,Λ tion that qi1 we will show that each K−algebra endomorphism of (Aq,Λ n (K))Z of Aq,Λ 2 · · · qin 1 qi2 "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 23 algebra automorphism. Furthermore, we will describe the automor- phism group for (Aq,Λ n (K))Z. Theorem 3.1. Suppose that qi1 in = 0. Then every K−algebra endomorphism ϕ of (Aq,Λ K−algebra automorphism. n = 1 implies i1 = i2 = · · · = n (K))Z is a 2 · · · qin 1 qi2 1 · · · zan Proof: Note that any invertible element of (Aq,Λ n (K))Z is of the n , where α ∈ K∗ and a1, · · · , an ∈ Z. Let ϕ be a n (K))Z , then ϕ sends invertible el- n (K))Z to its invertible elements. Therefore, for i = form αza1 K−algebra endomorphism of (Aq,Λ ements of (Aq,Λ 1, · · · , n, we have the following σ(zi) = λizai1 1 · · · zain n where λi ∈ K∗ and ai1, · · · , ain ∈ Z. Since yixi = zi−1 zixi = q−1 for i = 1, · · · , n. qi−1 and i xizi, ziyi = qiyizi for i = 1, · · · , n, we have that σ(zi) /∈ K∗ qi−1 , xiyi = qizi−1 Note that we can realize (Aq,Λ over the base algebra K[z±1 K[z±1 can write the image of xi under ϕ as follows: n ] defined by σi(zj) = qδij 1 , · · · , z±1 1 , · · · , z±1 n (K))Z as a generalized Weyl algebra n ] with the automorphisms σi of i zj. Thus for i = 1, · · · , n, we σ(xi) = Xk i fk ixki 1 · · · xki 1 n n +Xl i iyli 1 1 · · · yli n n gl i where k fk 1, · · · , ki = (ki 1 , · · · , z±1 i ∈ K[z±1 Since zixi = q−1 n ], and gl i ∈ K[z±1 i xizi, we have that n) ∈ (Z≥0)n, and l i 1, · · · , li = (li 1 , · · · , z±1 n ]. n) ∈ (Z≥0)n, and q−ai1ki 1 1 for fk i 6= 0 and q−ai2ki 2 2 · · · q−ainki n n = q−1 i for gl that aiiki and l i i 6= 0. Since qi1 i = 1 for fk 1 qi2 i 6= 0 and aiili n = 1 implies i1 = · · · = in = 0, we have ∈ (Z≥0)n i 6= 0 i = −1 for gl 6= 0. Since k i 6= 0 for some k i i ∈ (Z≥0)n, we cannot have both fk and gl spontaneously. Thus for i = 1, · · · , n, we have either i for some l · · · qainli n n = q−1 i 1 1 qai1li 2 · · · qin σ(xi) = Xk i fkixki 1 · · · xki n n 1 24 or X. TANG σ(xi) = Xl i gliyli 1 1 · · · yli n . n Since zjxi = xizj for i 6= j, we have ajiki i = 0 for j 6= i. Thus we have that aji = 0 for j 6= i and aii = ±1. As a result, we have i = 0 or ajili ϕ(zi) = λizi, or ϕ(zi) = λiz−1 i for i = 1, · · · , n. Suppose that we have σ(xi) = Xk i fk ixki 1 · · · xki n . n 1 Since ki i ≥ 0, we have that aii = 1. So we can conclude that ϕ(zi) = λizi. Since zixj = xjzi and ziyj = yjzj for i 6= j, we have that ϕ(xi) = fixi, and ϕ(yi) = giyi for some fi, gi ∈ K[z±1 n ]. Since we have xiyi = qizi−1 qi−1 , we have the following: 1 , · · · , z±1 fixigiyi = λiqizi − 1 qi − 1 . Thus we have fixigiyi = fig′ ixiyi = fig′ i qizi − 1 qi − 1 = λiqizi − 1 qi − 1 n ]. As a result, we have fig′ i ∈ K[z±1 1 , · · · , z±1 for some g′ Note that g′ i qbii gi = µ−1 i z−bi1 · · · z−bin Suppose that we have n 1 i = σi(gi). Therefore, we have that fi = µizbi1 1 for some µi ∈ K∗ and bi1, · · · , bin ∈ Z. i = 1 and λi = 1. n and · · · zbin σ(xi) = Xl i gl iyli 1 1 · · · yli n . n Then we have ajili for j 6= i. Since li ϕ(zi) = λiz−1 that ϕ(xi) = giyi and ϕ(yi) = fixi for some fi, gi ∈ K[z±1 Once again, since xiyi = qizi−1 i = 0 for j 6= i and aiili i = −1. Thus we have aji = 0 i ∈ Z≥0, we have aii = −1. As a result, we have for some λi ∈ K∗. Since zjxi = xizj for i 6= j, we have 1 , · · · , z±1 n ]. qi−1 , we have following: i giyifixi = λiqiz−1 i − 1 qi − 1 . Since yixi = zi−1 qi−1 , we have that giyifixi = gif ′ i yixi = gif ′ i zi − 1 qi − 1 = i − 1 λiqiz−1 qi − 1 "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 25 i ∈ K[z±1 . Since f ′ for some f ′ λi = q−1 and hi = −µ−1 bi1, · · · , bin ∈ Z. i q−bii−1 i i 1 , · · · , z±1 n ]. As a result, we have gif ′ i (fi), we have that fi = µizbi1 i = σ−1 z−bi1 1 i z = −1 and · · · zbii · · · zbin n i for some µi ∈ K∗ and · · · z−bii −1 · · · z−bin n 1 i Let ϕ be any K−algebra endomorphism of (Aq,Λ i = 1, · · · , n, we have either ϕ(zi) = λizi or ϕ(zi) = λiz−1 λi ∈ K∗. We will say that a K−algebra endomorphism ϕ of (Aq,Λ is of positive type if n (K))Z , then for each for some n (K))Z i ϕ(zi) = λizi for i = 1, · · · , n. Even though ϕ may not be of positive type, ϕ ◦ ϕ is a K−algebra endomorphism of (Aq,Λ n (K))Z of positive type. Next, we will show that each K−algebra endomorphism ϕ of (Aq,Λ n (K))Z of positive type is an automorphism, which implies that any K−algebra endomorphism ϕ of (Aq,Λ n (K))Z is an automorphism. Let ϕ be a K−algebra endomorphism of (Aq,Λ n (K))Z of positive type, then for each i = 1, · · · , n, we have the following: ϕ(xi) = µizbi1 n xi, ϕ(yi) = µ−1 1 for some µi ∈ K∗ and bi1, · · · , bin ∈ Z. i z−bi1 · · · zbin i qbii 1 Note that qi1 1 · · · qin n = 1 implies that i1 = · · · = in = 0 and · · · z−bin n yi, ϕ(zi) = zi xixj = λijxjxi, xiyj = λjiyjxi, yiyj = λijyjyi for i 6= j. After applying ϕ to both sides of the above commuting relations, we can conclude that ϕ(xi) = µizbii for i = 1, · · · , n. K−algebra automorphism of (Aq,Λ K−algebra endomorphism of (Aq,Λ yi, ϕ(zi) = zi i xi, ϕ(yi) = µ−1 It is obvious that ϕ has an inverse. Thus ϕ is a n (K))Z . So we have proved that every n (K))Z is an automorphism. ✷ i z−bii i qbii i Theorem 3.2. Maintain the condition that qi1 n = 1 implies i1 = · · · = in = 0 and let ϕ be a K−algebra automorphism of (Aq,Λ n (K))Z . Then there exists a partition P1 ∪ P2 of the set {1, · · · , n} with P1 = {l1, . . . , lk} and P2 = {lk+1, · · · , kn} such that 1 · · · qin (1) For any li ∈ P1 and lj ∈ P2 , we have that ϕ(xli) = αliz ai(k+1) lk+1 · · · zain ln zci li xli and and ϕ(yli) = βliz −ai(k+1) lk+1 · · · z−ain ln z−ci li yli ϕ(xlj ) = αlj zbj1 l1 · · · zbjk lk zcj lj ylj 26 X. TANG and ϕ(yj) = βlj z−bj1 l1 · · · z−bjk lk z−cj −1 lj xlj for some ai(k+1), · · · , ain, bj1, · · · , bjk ∈ Z, and ci, cj ∈ Z, and αli, βli, αlj , βlj ∈ K∗ such that αliβli = qci li , αlj βlj = −q−cj −1 lj , qbji li qaij lj = λ2 lilj . (2) For any given partition of P1 ∪ P2 of the set {1, · · · , n} sat- isfying the condition, let ϕ be a K−linear mapping defined as above. Then ϕ can be extended to a K−algebra automorphism of (Aq,Λ n (K))Z . Proof: Let ϕ be any K−algebra automorphism of (Aq,Λ n (K))Z . Then for i = 1, · · · , n, we have either Case 1: ϕ(xi) = µizbi1 · · · zbin 1 and ϕ(zi) = zi; or Case 2: ϕ(zi) = q−1 i z−1 z−bi1 ϕ(yi) = −µ−1 1 i q−bii−1 i i i · · · z−bii−1 n xi, and ϕ(yi) = µ−1 i qbii i z−bi1 1 · · · z−bin n yi, , and ϕ(xi) = µizbi1 1 · · · zbii i · · · zbin n yi, and · · · z−bin n xi. We can set P1 to be the set consisting of all i ∈ {1, · · · , n} such that the first case is true and P2 to be the set consisting of all i ∈ {1, · · · , n} such that the second case is true. It is obvious that P1 and P2 are disjoint and P1 ∪ P2 is a partition of {1, · · · , n}. The rest is verified by applying the automorphism ϕ to both sides of the following relations: xixj = λijxjxi, xiyj = λjiyjxi, yiyj = λijyjyi for any i 6= j. Conversely, it is straightforward to verify that the mapping ϕ defined by any given partition P1 ∪ P2 satisfying the condition can be extended to a K−algebra automorphism of (Aq,Λ ✷ As an immediate consequence of Theorem 3.2, we have the follow- n (K))Z in some special n (K))Z . ing corollary on the automorphism group of (Aq,Λ cases. Corollary 3.1. Let us further assume that qki j = λ2 kj = 0 for any i 6= j, then we have the following result. (1) Let ϕ be a K−algebra automorphism ϕ of (Aq,Λ i qkj ij implies ki = n (K))Z . For each i = 1, · · · , n, we have either ϕ(xi) = αizci i xi, ϕ(yi) = βiz−ci i yi for some ci ∈ Z and αi, βi ∈ K∗ such that αiβi = qci i , or ϕ(xi) = αizci i yi, ϕ(yi) = βiz−ci−1 i xi for some ci ∈ Z and αi, βi ∈ K∗ such that αiβi = −q−ci−1 i . "SYMMETRIC" MULTIPARAMETER QUANTIZED WEYL ALGEBRAS 27 (2) Let ϕ be a K−linear mapping defined as above. Then ϕ can be extended to a K−algebra automorphism of (Aq,Λ n (K))Z . ✷ Acknowledgements: The author would like to thank James Zhang for sharing an earlier version of the preprint [6]. Part of the results were announced during the special session on "New Developments in Non- commutative Algebra", the AMS Spring Western Sectional Meeting at University of Nevada, Las Vegas, Las Vegas, NV, April 18-19, 2015. The author would like to thank the organizers for the hospitality. References [1] M. Akhavizadegan and D. A. Jordan, Prime ideals of quantized Weyl algebras, Glasgow Math. J., 38 (1996), no. 3, 283 -- 297. [2] J. Alev and F. Dumas, Rigidit´e des plongements des quotients prim- q(sl(2)) dans lalg`ebre quantique de Weyl-Hayashi, itifs minimaux de U Nagoya Math. J., 143 (1996),119 -- 146. [3] E. Backelin, Endomorphisms of quantized Weyl algebras, Lett. Math. Phys., 97 (2011), no. 3, 317 -- 338. [4] V. V. Bavula, Generalized Weyl algebras and their representations, St. Petersburg Math. J., 4 (1993), no. 1, 71 -- 92. [5] V. V. Bavula and D. A. Jordan, Isomorphism problems and groups of automorphisms for generalized Weyl algebras, Trans. Amer. Math. Soc., 353 (2001), no. 2, 769 -- 794. [6] J. P. Bell and J. J. Zhang, Zariski cancellation problem for noncommu- tative algebras, preprint, arXiv:1601.04625. [7] A. Belov-Kanel and M. Kontsevich, The Jacobian conjecture is stably equivalent to the Dixmier conjecture, Mosc. Math. J., 7 (2007), no. 2, 209 -- 218. [8] S. Ceken, J. H. Palmieri, Y. H. Wang and J. J. Zhang, The discriminant controls automorphism groups of noncommutative algebras, Adv. Math., 269 (2015), 551 -- 584. [9] S. Ceken, J. H. Palmieri, Y. H. Wang and J. J. Zhang, The discriminant criterion and automorphism groups of quantized algebras, Adv. Math. 286 (2016), 754 -- 801 [10] K. Chan, A. Young and J. Zhang, Discriminant formulas and applica- tions, preprint, arXiv:1503.06327. [11] J. Dixmier, Sur les alg´ebres de Weyl, Bull. Soc. Math. France, 96 (1968), 209 -- 242. [12] H. Fujita, E. Kirkman and J. Kuzmanovich, Global and Krull dimen- sions of quantum Weyl algebras, J. Algebra, 216 (1999), 405 -- 416. [13] J. Gaddis, E. Kirkman and W. F. Moore, On the discriminant of twisted tensor products, preprint, arXiv:1606.03105. [14] A. Giaquinto and J. J. Zhang, Quantum Weyl algebras, J. Algebra, 176 (1995), 861 -- 881. [15] V. Ginzburg, Calabi-Yau algebras, preprint, arXiv:0612139v3. 28 X. TANG [16] J. Gomez-Torrecillas and L. EL Kaotit, The group of automorphisms of the coordinate ring of quantum symplectic space, Beitr. Algebra Geom., 43 (2002), 597 -- 601. [17] K. R. Goodearl, Prime ideals in skew polynomial rings and quantized Weyl algebras, J. Algebra 150 (1992), no. 2, 324 -- 377. [18] K. R. Goodearl and J. T. Hartwig, The isomorphism problem for multi- parameter quantized Weyl algebras, Sao Paulo J. Math. Sci., 9 (2015), 53 -- 61. [19] K. R. Goodearl and M. T. Yakimov, Unipotent and Nakayama auto- morphisms of quantum nilpotent algebras, preprint, arXiv:1311.0278. [20] D. A. Jordan, A simple localization of the quantized Weyl algebra, J. Algebra, 174 (1995), no. 1, 267 -- 281. [21] O. H. Keller, Ganze Cremona-Transformationen, Monatsh. Math. Phys., 47 (1939), no. 1, 299 -- 306. [22] A. P. Kitchin and S. Launois, Endomorphisms of quantum generalized Weyl algebras, Lett. Math. Phys., 104 (2014), 837 -- 848. [23] A. P. Kitchin and S. Launois, On the automorphisms of quantum Weyl algebras, preprint, arXiv:1511.01775. [24] J. Levitt and M. Yakimov, Quantized Weyl algebras at roots of unity, preprint, arXiv:1606.02121. [25] L.Y. Liu, S. Q. Wang and Q. S. Wu, Twisted Calabi-Yau property of Ore extensions, J. Noncommut. Geom., 8 (2014), no. 2, 587 -- 609. [26] G. Maltsiniotis, Groupes quantiques et structures diffrentielles, C. R. Acad. Sci. Paris S´er. I Math., 311 (1990), no. 12, 831 -- 834. [27] J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, Wiley -- Interscience, Chichester, 1987. [28] B. Nguyen, K. Trampel and M. Yakimov, Noncommutative discrimi- nants via Poisson primes, arXiv:1603.02585. [29] M. Reyes, D. Rogalski, and J. J. Zhang, Skew Calabi-Yau algebras and homological identities, Adv. Math., 264 (2014), 308 -- 354. [30] L. Richard, Sur les endomorphismes des tores quantiques, Comm. Al- gebra, 30 (2002), no. 11, 5283 -- 5306. [31] L. Richard and A. Solotar, Isomorphisms between quantum generalized Weyl algebras, J. Algebra Appl., 5 (2006), no. 3, 271 -- 285. [32] L. Rigal, Spectre de l'algb´ebre de Weyl quantique, Beitr. Algebra Geom., 37 (1996), 119 -- 148. [33] X. Tang, Algebra endomorphisms and derivations of some localized down-up algebras, J. Algebra Appl., 14, 1550034 (2015). [34] Y. Tsuchimoto, Endomorphisms of Weyl algebras and p -- curvature, Os- aka J. Math., 42(2) (2005), 435 -- 452. Department of Mathematics & Computer Science, Fayetteville State University, 1200 Murchison Road, Fayetteville, NC 28301 E-mail address: [email protected]
1101.1112	4	1101	2016-03-23T16:35:58	Some remarks on Mathieu subspaces over associative algebras	[ "math.RA" ]	In this paper, we generalize some of the results of [9], and add some new results. Furthermore, we take a closer look at strongly simple algebras, which are introduced in [9].	math.RA	math	Some remarks on Mathieu subspaces over associative algebras Michiel de Bondt June 8, 2018 Abstract In [9], the author takes a closer look at algebraic elements of radicals of Mathieu subspaces (of associative algebras) over a field, and suggests to look at integral elements with rings other than fields. But it seems more useful to look at so-called co-integral elements. We generalize his theory about algebraic radicals over fields to co-integral radicals over commuta- tive rings with unity. Furthermore, we show that over Artin rings, the concepts of integral- ity and co-integrality coincide. In addition, we define so-called uniform Mathieu subspaces, inspired by the fact that Mathieu subspaces with co- integral radicals are always of this type. Besides broadening the theory of [9] by means of the new concepts co-integrality and uniformity, we gen- eralize many of the results of [9] in other ways as well. Furthermore, we obtain several new results. In the last section, we disprove a conjecture by the author of [9] (in a version of [9] prior to finding the counterexample), by showing that so- called strongly simple algebras do not need to be fields over theirselves. Key words: Mathieu subspaces, radicals, co-integral elements, idempotents, uni- form Mathieu subspaces, strongly simple algebras, valuation domain. 2010 Mathematics Subject Classification: 16N40; 16D99; 16D70. 1 Introduction [9] is entitled 'Mathieu subspaces over associative algebras', in which the author W. Zhao introduces what the title expresses. The title of this paper has the words 'Some remarks on' in front, and can be seen as some remarks on the subject of [9], as well as some remarks on [9] itself. Let us first repeat the definition of Mathieu subspaces as formulated in [9]. Definition 1.1 (collecting [9, Def. 1.1] and [9, Def. 1.2]). Let M be an R- subspace (R-submodule) of an associative R-algebra A. Then we call M a ϑ-Mathieu subspace of A if the following property holds for all a, b, c ∈ A such that am ∈ M for all m ≥ 1: 1 (i) bam ∈ M when m ≫ 0, if ϑ = "left "; (ii) amc ∈ M when m ≫ 0, if ϑ = "right "; (iii) bam, amc ∈ M when m ≫ 0, if ϑ = "pre-two-sided "; (iv) bamc ∈ M when m ≫ 0, if ϑ = "two-sided ". If we replace every occurence of 'm ≫ 0' by 'm ≥ 1' in the above definition, then the cases "pre-two-sided " and "two-sided " coincide and we get the definition of a ϑ R-ideal of an associative R-algebra, or just the definition of a ϑ ideal of a ring since every ring is an associative Z-algebra. Thus the concept of Mathieu subspaces is a generalization of that of ideals. On the other hand, there is a single occurence of 'm ≥ 1' before the enumer- ation, which can be replaced by 'm ≫ 0'. Proposition 1.2 (summarizing [9, Prop. 2.1]). If we replace the single oc- curence of 'm ≥ 1' in definition 1.1 by 'm ≫ 0', then we actually keep the same definition. The definition of Mathieu subspace by Zhao was inspired by the following con- jecture of O. Mathieu in [5]. Mathieu Conjecture. Let G be a compact Lie group with Haar measure σ, and f a complex-valued G-finite function on G such that RG f mdσ = 0 for all m ≥ 1. Then for each G-finite function g on G, we have that RG gf mdσ = 0, for all m ≫ 0. This conjecture has been proved for the abelian case by Duistermaat and van der Kallen in [2], and can be reformulated in terms of Mathieu subspaces, as follows. Mathieu Conjecture. Let G be a compact Lie group with Haar measure σ, and A be the C-algebra of G-finite functions on G. Then the C-subspace of A consisting of functions whose integral over G with respect to σ is zero, is a Mathieu subspace of A. The Mathieu Conjecture resembles [6, Conjecture 7.1] by Zhao, which is given below, in both its structure and the fact that it implies the Jacobian conjecture. Vanishing Conjecture. Let P be a homogeneous polynomial in n variables over C. If ∆m(P m) = 0 for all positive m, then ∆m(P m+1) = 0, for all m large enough, where ∆ is the Laplace operator. Inspired by the Mathieu Conjecture, the authors found the following even more resembling equivalent formulation of the Vanishing Conjecture in [4, Th. 1.5]. Vanishing Conjecture. Let P be a homogeneous polynomial in n variables over C. If ∆m(P m) = 0 for all positive m, then ∆m(QP m) = 0, for all m large enough and all polynomials Q in n variables over C, where ∆ is the Laplace operator. 2 Later on, Zhao dropped the homogeneity condition on the polynomial P in [7], and replaced the Laplace operator by any differential operator with constant coefficients, which resulted in: Generalized Vanishing Conjecture. Let Λ be any differential operator with constant coefficients. If P is a polynomial over C such that Λm(P m) = 0 for all positive m, then for any polynomial Q over C we have that Λm(QP m) = 0, for all sufficiently large m. Other related conjectures are the Image Conjecture and the Dixmier Conjecture, the latter of which is actually equivalent to the Jacobian conjecture. See [9] and e.g. [8], [3], and the references in all these papers. Define the radical r(M ) = √M of M as the set {a ∈ A am ∈ M for all m ≫ 0}, where A is the associative algebra at hand. If the R-subspace M is an ideal of A, then r(M ) is the usual radical of M , which is an ideal itself when A is commutative. But r(M ) is not an ideal in general. r(M ) does not even need to be a vector space over R when R is a field. In proposition 1.2, we can rewrite the condition that am ∈ M for all m ≫ 0 by a ∈ r(M ). This was actually done in the original result [9, Prop. 2.1]. We will do the same in (ii) of 1.4 below. The radical of M plays a crucial role in the theory of Mathieu subspaces, and one can formulate the definition of Mathieu subspace entirely in terms of radicals, see [9, Lm. 2.3]. In order to avoid distinguishing cases of ϑ, we define a 'constraint' Cϑ(b, c) as follows. Definition 1.3. Let ϑ be any of the four types of Mathieu subspace and set Cϑ(b, c) := c = 1 b = 1 1 ∈ {b, c} 1 + 1 = 2 if ϑ = "left " if ϑ = "right " if ϑ = "pre-two-sided " if ϑ = "two-sided "   From definitions 1.1 and 1.3 and proposition 1.2, we obtain the following. Proposition 1.4. Let M be an R-subspace (R-submodule) of an associative R-algebra A. Then M is a ϑ-Mathieu subspace of A, if and only if any of the following property holds: (i) for all a such that am ∈ M for all m ≥ 1 and all b, c ∈ A such that Cϑ(b, c), we have bamc ∈ M when m ≫ 0; (ii) for all a ∈ r(M ) and all b, c ∈ A such that Cϑ(b, c), we have bamc ∈ M when m ≫ 0. In the proof of [9, Prop. 2.5], only the case ϑ = "left " is done, since the other cases are similar. This can be made precise using Cϑ(b, c). Let us prove for example the following reformulation of [9, Prop. 2.7] (without the condition that V is an R-space). 3 Proposition 1.5 (following [9, Prop. 2.7]). Let A and B be associative R- algebras and φ : A → B a surjective R-algebra homomorphism. Then V ⊆ B is a ϑ-Mathieu subspace of B = φ(A), if and only if φ−1(V ) is a ϑ-Mathieu subspace of A. Proof. The 'only if'-part follows from [9, Prop. 2.5], so assume that φ−1(V ) is a ϑ-Mathieu subspace of A. Since φ−1(V ) is an R-subspace of A and V = φ(φ−1(V )) by surjectivity of φ, we see that V is an R-subspace of B. Take a, b, c ∈ B such that Cϑ(b, c) and am ∈ V for all m ≥ 1. Then there exist a′, b′, c′ ∈ A such that Cϑ(b′, c′), φ(a′) = a, φ(b′) = b and φ(c′) = c. Therefore φ((a′)m) = am ∈ V for all m ≥ 1, i.e. (a′)m ∈ φ−1(V ) for all m ≥ 1. Since φ−1(V ) is a ϑ-Mathieu subspace of A, we have b′(a′)mc′ ∈ φ−1(V ) when m ≫ 0. Hence bamc ∈ φ(φ−1(V )) = V when m ≫ 0. 2 Co-integral elements in the radicals of arbi- trary subspaces We say that a is co-integral over R if aN R[a] = aN +1R[a] for some N ∈ N. If a is an invertible element of an associative algebra over R, the a is co-integral over R, if and only if a−1 is integral over R. This is because of the following, which explains the choice of the the term co-integral, too. The co-integrality condition aN R[a] = aN +1R[a] is equivalent to that a satisfies an algebraic relation over R whose trailing nonzero coefficient is equal to one, as opposed to the leading coefficient when a is integral over R. So if a is either integral or co-integral over R, say with corresponding algebraic relation p ∈ R[t], then by decomposing p = tN q(t−1) for some N ∈ N and a q ∈ R[t], we obtain the other for a−1, namely with corresponding algebraic relation q. If a is co-integral over R and a−1 /∈ A, then one can show that a is either a zero divisor in R[a] or zero itself: take V = R[a] in (ii) of proposition 2.8 later in this section. In this section, we extend results of the end of section 2 and of section 3 in [9], mainly by generalizing from fields to commutative rings, replacing 'integral' by 'co-integral'. Additionally, we change the order and setup on some points. We may omit proofs if those of the original results are already sufficient. Definition 2.1. Let A be an associative R-algebra. Call a co-integral over R if aN R[a] = aN +1R[a] for some N ∈ N. Define r ′(V ) := {a ∈ r(V ) a is co-integral over R} for R-subspaces V of A. Let (a)ϑ be the ϑ-ideal gener- ated by a when ϑ 6= "pre-two-sided ", and (a)"pre-two-sided" = Aa + aA be the sum of the left and right ideals generated by a. Define (S)ϑ in a similar manner for subsets S of A. Then (S)ϑ = Ps∈S(s)ϑ. ′(V ) is different because 'integral' is used instead of In [9], the definition of r ′(V ) when the base ring is a 'co-integral'. This is because the author only uses r field. In that case, the concept of co-integrality coincides with that of integrality, 4 so that the author can assume integrality and use co-integrality, which is the concept that really matters. So we need to distinguish co-integrality and integrality for commutative base rings in general. There are however rings other than fields for which both concepts coincide. Theorem 2.2. Assume A is an associative algebra over an Artin ring R. Take a ∈ A arbitrary. Then a is co-integral over R, if and only if a is integral over R. Proof. Let us first prove the 'if'-part. For that purpose, suppose that a is integral over R. From [1, Prop. 5.1], it follows that R[a] is a finitely generated R-module. From [1, Prop. 6.5], it follows that R[a] is an Artinian R-module. So we can apply the descending chain condition on Ra + Ra2 + Ra3 + ··· ⊇ Ra2 + Ra3 + ··· ⊇ Ra3 + ··· ⊇ ··· from which we deduce that a is co-integral over R. Hence the 'only if'-part remains to be proved. So let a be co-integral over R, i.e. aN R[a] = aN +1R[a] for some integer N ≥ 0. We distinguish two cases. • N = 0. Then 1 = aN ∈ aN +1R[a] = aR[a], so a is invertible. Consequently, a−1 is integral over R (see the second paragraph of this section). By the 'if'- part, a−1 is co-integral over R. So a is integral over R (see the second paragraph of this section again). • N ≥ 1. From lemma 2.3 below, it follows that aN ∈ a2N R[a], say that aN = a2N u(a), where u ∈ K[t]. Let e = aN u(a). Then eaN = aN e = a2N u(a) = aN , so eam = ame = am for every m ≥ N . Consequently, e2 = e. Now make the following definitions. R := Re A := aN R[a] a := ea Since eam = ame = am for every m ≥ N , we see that e ∈ R is unital in A ⊇ R. Hence A is an associative R-algebra by inclusion, with e being the multiplicative unit of both R and A. Furthermore, R is a homomorphic image of an Artin ring and hence Artinian itself. Since R[a] is commutative, it follows that am = amem = am for every m ≥ N , so {a, e} ⊆ A = aN R[a] = aN R[a] It follows that A = R[a] and that e ∈ a R[a], because N ≥ 1. The inclusion e ∈ a R[a] implies that a is co-integral over R, with the previous case N = 0 being in force. So a ∈ A is integral over R on account of the previous case, i.e. there is a polynomial p over R with leading coefficient e, such that p(a) = 0. 5 Assume without loss of generality that p has no terms of degree less than N . Then am = amem = am for all m ≥ N tells us that we can replace e by 1 and a by a without affecting p(a) = 0. So a is integral over R. Notice that in the proof of the 'only-if'-part of theorem 2.2, we only use the validity of the 'if'-part of theorem 2.2 for homomorphic images of R. So any ring R with the property that the 'if'-part of theorem 2.2 is satisfied for homomorphic images of R, satisfies theorem 2.2 as a consequence. Lemma 2.3. Let A be an associative R-algebra and a ∈ A. aN +1R[a], then aN ∈ amR[a] for every m ≥ 0. Proof. Take m minimum, such that aN /∈ am+1R[a]. If m ≤ N , then If aN R[a] = aN ∈ aN +1R[a] ⊆ aN +1−(N −m)R[a] = am+1R[a] Consequently, m > N . Furthermore, aN ∈ amR[a] = amR + am+1R[a] and amR ⊆ aN −maN R[a] = aN −maN +1R[a] = am+1R[a] This contradicts aN /∈ am+1R[a], so aN ∈ amR[a] for every m ≥ 0. Lemma 2.3 above can be seen as a replacement for [9, Lm. 3.3]. Co-integrality is in some sense the opposite of integrality, just as Artinian is in some sense the opposite of Noetherian. There is indeed a strong connection between both pairs of opposite concepts. Proposition 2.4. Let A be an associative R-algebra A and a ∈ A. Then we have the following. (i) If R is Noetherian, then a is integral over R, if and only if R[a] is Noethe- rian over R; (ii) If R is Artinian, then a is co-integral over R, if and only if R[a] is Artinian over R; Proof. We start with the 'if'-parts of (i) and (ii). The 'if'-part of (i) follows from the fact that the integrality of a over R is just the ascending chain condition on Ra ⊆ Ra + Ra2 ⊆ Ra + Ra2 + Ra3 ⊆ ··· The 'if'-part of (ii) follows from the fact that the co-integrality of a over R is just the descending chain condition on Ra + Ra2 + Ra3 + ··· ⊇ Ra2 + Ra3 + ··· ⊇ Ra3 + ··· ⊇ ··· The 'only if'-parts follow from [1, Prop. 5.1] and [1, Prop. 6.5], except that we need that a is integral over R over R instead of that a is co-integral over R in (ii). But that follows from theorem 2.2. 6 As opposed to the corresponding results in [9], lemma 2.5 below also describes the situation where a is not invertible. This allows us to give a proof of theorem 2.6 which is more direct than the proof of the corresponding results in [9]. Lemma 2.5 (following [9, Lm. 3.1] and [9, Lm. 3.2] more or less). Let A be an associative R-algebra and V an R-subspace of A. Assume a, b, c ∈ A such that bamc ∈ V when m ≫ 0 and aN R[a] = aN +1R[a] for some N ∈ N. Then bamc ∈ V for all m ≥ N . If additionally a is a unit in A, then bamc ∈ V for all m ∈ Z. Proof. Assume there is an m ≥ N (m ∈ Z) such that bamc /∈ V . Since bamc ∈ V for all m ≫ 0, there is a largest m ≥ N (m ∈ Z) such that bamc /∈ V . From aN R[a] = aN +1R[a], we deduce that aN c ∈ aN +1R[a]c, and multiplication with bam−N (which requires that a is a unit if m < N ) gives bamc ∈ bam+1R[a]c ⊆ V . Contradiction, so bamc ∈ V for all m ≥ N (m ∈ Z). Theorem 2.6 (combining [9, Th. 3.9] and [9, Th. 3.10] more or less). Let A be an associative R-algebra and M be a ϑ-Mathieu subspace of A over R. Suppose that a ∈ A and N ≥ 0. Then for (1) a ∈ r(M ) and aN R[a] = aN +1R[a]; (2) (aN )ϑ ⊆ M ; (3) a ∈ r(M ); we have (1) ⇒ (2) ⇒ (3). Proof. (2) ⇒ (3) follows immediately from the definitions of r(M ) and (aN )ϑ. To prove (1) ⇒ (2), take b, c ∈ A such that Cϑ(b, c), where Cϑ(b, c) is as in definition 1.3. Since a ∈ r(M ) and M is a ϑ-Mathieu subspace of A, we obtain by ii) of proposition 1.4 that bamc ∈ V for all m ≫ 0. As aN R[a] = aN +1R[a], we deduce from lemma 2.5 above that baN c ∈ M . Since b, c ∈ A such that Cϑ(b, c) were arbitrary, the desired result follows from definition 2.1. In particular, a ∈ r ′(M ) implies that (aN )ϑ ⊆ M for some N . Idempotents of arbitrary subspaces 2.1 Call a a semi-idempotent if a ∈ Ra2, a quasi-idempotent if a ∈ R∗a2, and an idempotent if a = a2. Here, R∗ denotes the set of units of R. The first of the three above definitions does not appear in [9]. The other two are taken from [9]. Lemma 2.7 (following [9, Lm. 2.9] more or less). Let a be a nonzero semi- idempotent of A and V be an R-subspace of A. Then a = ra2 for some r ∈ R. (i) If a is nilpotent, then a = 0. If a 6= 0 is not a zero divisor in R[a], then a = (r · 1)−1, which is a unit. If a 6= 0 is not a zero divisor in R[a] and a is an idempotent, then a = 1. 7 (ii) a ∈ r(V ) implies am ∈ V for all m ≥ 1. (iii) If a ∈ V is a (quasi-)idempotent, then a ∈ r(V ). If a ∈ r(V ) and V is a ϑ-Mathieu subspace of A, then (a)ϑ ⊆ V . Proof. Since a ∈ Ra2, we can write a = ra2 with r ∈ R. (i) If a is nilpotent, say that am = 0, then a = ra2 = r2a3 = ··· = rm−1am = 0 indeed. Furthermore, we obtain from a = ra2 that a(1 − ra) = 0. Thus if a 6= 0 is not a zero divisor in R[a], then ra = 1, whence a = (r · 1)−1. If additionally a is an idempotent, then we can take r = 1, so that a = 1−1 = 1. (ii) Assume a ∈ r(V ). Since a ∈ Ra2, we have aR[a] = a2R[a]. Hence by lemma 2.5 with b = c = 1 = N , am ∈ V for all m ≥ 1, as desired. (iii) If a ∈ V is a quasi-idempotent, then we can take r ∈ R∗, so that am = r1−ma ∈ V for all m ≥ 1. In particular, a ∈ V implies a ∈ r(V ). Next, assume that V is a Mathieu-subspace of A and a ∈ r(V ). Since a ∈ a2R[a], we have aR[a] = a2R[a]. Hence (a)ϑ ⊆ V on account of (1) ⇒ (2) or theorem 2.6 with N = 1. Proposition 2.8 (following [9, Prop. 3.4] more or less). Let V be an R-subspace ′(V ). Take N such that aN R[a] = aN +1R[a]. Then there exists of A and a ∈ r an idempotent e ∈ aN R[a] ⊆ V such that aN = aN e = eaN . Additionally, we have (i) a is nilpotent, if and only if aN = 0, if and only if e = 0, (ii) a is a unit in A, if and only if a 6= 0 is not a zero divisor in R[a], if and only if e = 1. Proof. From lemma 2.5 with b = c = 1, it follows that am ∈ V for all m ≥ N . Hence aN R[a] ⊆ V . If N = 0, then we take e = 1. If N ≥ 1, then we take e as in the case N ≥ 1 in the proof of theorem 2.2. In both cases, e ∈ aN R[a] and aN = aN e = eaN . (i) Since aN = aN e = eaN , we see that e = 0 implies aN = 0. Conversely e ∈ aN R[a] tells us that aN = 0 implies e = 0. Thus it remains to show that aN = 0 in case a is nilpotent, i.e. a ∈ r ′(0). This follows from aN R[a] ⊆ V , because we can take V = 0 when a ∈ r ′(0). (ii) If a 6= 0 is not a zero divisor in R[a], then we can cancel aN everywhere in aN = aN e = eaN , which gives e = 1. Since units are not zero divisors, it remains to show that a is a unit in case e = 1. Hence assume that e = 1. If N ≥ 1, then e ∈ tN R[t] tells us that a aN e = 1. If N = 0, then 1 = aN ∈ aN R[a] = aN +1R[a] = aR[a], which leads to a 1 as well, as desired. Theorem 2.9 (following [9, Th. 3.5] more or less). Let A be an associative R- algebra and V an R-subspace of A. Then the following statements are equivalent. 8 (1) Every non-unit of r ′(V ) is nilpotent. (2) Every zero divisor of r ′(V ) is nilpotent. (3) V contains no idempotents other than 0 and 1. Proof. Since (1) ⇒ (2) follows from the fact that zero divisors are non-units, the following remains to be proved. (2) ⇒ (3) Assume V contains an idempotent e /∈ {0, 1}. Then by (i) of lemma 2.7, e is a zero divisor because e 6= 1, but additionally e is not nilpotent because e 6= 0. This gives the desired result. (3) ⇒ (1) Assume a ∈ r ′(V ) is a non-unit, but not nilpotent. By proposition 2.8, R[a] contains an idempotent e, for which e 6= 0 and e 6= 1 on account of (i) and (ii) of proposition 2.8 respectively. This gives the desired result. If we take V = A = r(A) in the above theorem (just as in [9]), we obtain the following. Corollary 2.10 (following [9, Cor. 3.6] more or less). For every associative R-algebra A, the following statements are equivalent. (1) Every non-unit of r ′(A) is nilpotent. (2) Every zero divisor of r ′(A) is nilpotent. (3) A contains no idempotents other than 0 and 1. Lemma 2.11 (same as [9, Lm. 3.7]). Let A be an associative R-algebra. Then for the following three statements: (1) every non-unit of A is nilpotent; (2) A is a local R-algebra; (3) A contains no idempotents other than 0 and 1; we have (1) ⇒ (2) ⇒ (3). Proof (somewhat more direct than the original proof ). (1) ⇒ (2) It is a nice exercise for the reader to show that the nilpotent ele- ments of A form an ideal if (1) holds. Hence (1) implies (2). (2) ⇒ (3) Assume A is local and A has an idempotent e. Then e(1− e) = 0 = (1 − e)e. Since e and 1 − e cannot be contained in the same proper ideal of A and A is local, one of e and 1 − e must be a unit. The other must be zero, because units are not zero divisors and e(1 − e) = 0 = (1 − e)e. So e ∈ {0, 1}. 9 2.2 Quasi-stable algebras The following definition appears at the beginning of section 7 in [9], which has the same title as this subsection. Definition 2.12. Let A be an associative R-algebra. We say that A is ϑ-quasi- stable (or ϑ-stable), if every R-subspace V of A with 1 /∈ V is a ϑ-Mathieu subspace of A (or a ϑ-ideal of A respectively). If we combine lemma 2.11 with (3) ⇒ (1) of corollary 2.10 (just as in [9]), then we get the first assertion in the following. Corollary 2.13 (following [9, Cor. 3.8] and [9, Prop. 7.4]). For every associative ′(A) = A, the three statements in lemma 2.11 are equivalent. R-algebra A with r Furthermore, A is (two-sided) quasi-stable over R in case any of these three statements is fulfilled. Proof. Assume that the equivalent statements of lemma 2.11 are fulfilled. Let V be an R-subspace of A such that 1 6∈ V . It suffices to show that r(V ) ⊆ r((0)). ′(V ) such that a 6∈ r((0)). On account of the first So assume a ∈ r(V ) = r statement of lemma 2.11, a is invertible over R. Hence am ∈ V for all m ∈ Z by lemma 2.5. This contradicts 1 6∈ V , thus a ∈ r((0)). A generalization which applies to both integrality and co-integrality of a, is that some coefficient of the polynomial which has a as a root must be equal to 1, but not necessarily the leading or trailing nonzero coefficient. If a is invertible, then we can shift this polynomial to obtain a Laurant polynomial which has 1 as constant term. In other words, 1 ∈ aR[a]+a−1R[a−1]. Similarly, an invertible element a ∈ A is co-integral or integral over R, if and only if 1 ∈ aR[a] or 1 ∈ a−1R[a−1] respectively. Proposition 2.14 (generalizing [9, Prop. 7.4]). Let A be an associative R- algebra, such that every element of A is either invertible or nilpotent, and every invertible element a ∈ A satisfies 1 ∈ aR[a] + a−1R[a−1]. Then r ′(A) = A and A is integral and (two-sided) quasi-stable over R. Proof. We first show that each a ∈ A is both integral and co-integral over R. So take a ∈ A arbitrary. If a is nilpotent, say that am = 0, then clearly a is integral over R and amR[a] = am+1R[a]. Hence suppose that a is not nilpotent. Then a is invertible and 1 ∈ aR[a]+ a−1R[a−1] by assumption. Say that 1 = f (a) where f (z) is a Laurant polynomial without constant term. Let f (z) be the Laurant polynomial consisting of the terms rizi of f such that ri · 1 is not nilpotent. Then f has no constant term either. Since the nilpotent elements of the commutative algebra R[a, a−1] form an ideal of R[a, a−1], we have that 1 − f (a) = f (a) − f (a) is nilpotent. Hence f (a) = 1 − (1 − f (a)) is invertible in R[a, a−1] and Xi=0 (1 − f (a))i f (a) k = f (a) 1 = 1 − (1 − f (a)) 10 Notice that the leading term of f (z)Pk for some k ∈ N. i=0(1− f (z))i − 1 is −1 in case f (z) ∈ R[z−1]. Since every nonzero coefficient of f is invertible, the leading nonzero i=0(1 − f (z))i − 1 is invertible, regardles of whether f (z) ∈ i=0(1 − f (z))i − 1 is invertible, and ′(A) = A, and by corollary Similarly, the trailing nonzero of f (z)Pk coefficient of f (z)Pk R[z−1] or not. Hence a is integral over R. a is co-integral over R. Thus A is integral over R, r 2.13, A is two-sided quasi-stable over R. For more results about quasi-stable algebras, see section 7 of [9]. 2.3 Localization of the base ring We end this section with some results about integrality, co-integrality and lo- calization of the base ring. First, we formulate results about co-integrality and localization of the base ring. Proposition 2.15. Assume A is an associative R-algebra, and S ∋ 1 is a multiplicatively closed subset of R. (i) If saN ∈ aN +1R[a] for some s ∈ S, then s−1a is co-integral over R. (ii) If a ∈ A is co-integral over R, then for all s, s′ ∈ S, s−1a is co-integral over R and s−1s′a is co-integral over S−1R. (iii) If b ∈ S−1A is co-integral over R, then for all s, s′ ∈ S, s−1b is co-integral over R and s−1s′b is co-integral over S−1R. (iv) If b ∈ S−1A is co-integral over S−1R, then there exists an s ∈ S such that s−1b is co-integral over R. Furthermore, s−1s′b is co-integral over S−1R for all s, s′ ∈ S. Proof. (i) Multiplication of saN ∈ aN +1R[a] by s−N −1 gives (s−1a)N = s−N aN ∈ s−N −1aN +1R[a] ⊆ (s−1a)N +1R[s−1a]. (ii) Multiplication of aN ∈ aN +1R[a] by s−N gives (s−1a)N = s−N aN ∈ s−N −1aN +1sR[a] ⊆ (s−1a)N +1R[s−1a]. (s−1a)N ∈ (s−1a)N +1R[s−1a] in turn can be multiplied by (s′)N , to obtain the second claim. (iii) Replace a by b in the proof of (ii). (iv) Say that bN = bN +1p(b) for some univariate polynomial p over S−1R. Let s be the product of the denominators of the coefficients of p. Then (s−1b)N = s−N bN = s−N −1bN +1s p(b) ∈ (s−1b)N +1R[s−1b]. The second claim follows is a similar manner as the second claim in (iii). Although co-integrality seems a more useful concept than integrality in this context, (v) of the next theorem is about integrality and localization of the base ring. 11 Theorem 2.16. Assume A is an associative R-algebra, and S ∋ 1 is a multi- plicatively closed subset of R. Write φ : A → S−1A for the localization map. (i) If M is a ϑ-Mathieu subspace over S−1R of S−1A, then M is also a ϑ-Mathieu subspace over R over S−1A. (ii) If M is a ϑ-Mathieu subspace over R of S−1A, then φ−1(M ) is a ϑ- Mathieu subspace over R of A. (iii) If V ⊆ S−1A, then M := φ−1(V ) is a ϑ-Mathieu subspace over R of A, if and only if φ(M ) is a ϑ-Mathieu subspace over R of φ(A). (iv) Assume that V ⊆ S−1A and M := φ−1(V ) is a ϑ-Mathieu subspace over R of A, such that for each a ∈ S−1A such that am ∈ S−1M for all m ≥ 1, there exists an s ∈ S such that (sa)m ∈ φ(M ) for all m ≥ 1. Then S−1M is a ϑ-Mathieu subspace over S−1R of S−1A. (v) For a specific a as in (iv), i.e. am ∈ S−1M for all m ≥ 1, an s as in (iv) exists in case s′a is integral over R (not necessary co-integral) for some s′ ∈ S. Proof. (i) This follows from the trivial fact that an S−1R-subspace is also an R- subspace. (ii) This follows from [9, Prop. 2.5]. (iii) Since φ−1(φ(φ−1(V ))) = φ−1(V ), we have φ−1(φ(M )) = M . Hence taking V = φ(M ) in proposition 1.5 gives the desired result. (iv) Take any element a ∈ S−1A such that am ∈ S−1M for all m ≥ 1. By assumption, there exists an s ∈ S such that (sa)m ∈ φ(M ) for all m ≥ 1. By (iii), we see that φ(M ) is a ϑ-Mathieu subspace over R of φ(A). Thus for all b′, c′ ∈ φ(A) such that Cϑ(b′, c′), we have b′(sa)mc′ ∈ φ(M ) for all m ≫ 0, where Cϑ(b′, c′) is as in definition 1.3. Consequently, for all b′, c′ ∈ φ(A) such that Cϑ(b′, c′), we have b′amc′ ∈ S−1M for all m ≫ 0. For all b, c ∈ S−1M , there exists an s′ ∈ S such that s′b, s′c ∈ φ(A). Using this fact as far as b 6= 1 6= c, we deduce that for all b, c ∈ S−1M such that Cϑ(b, c), we have bamc ∈ S−1M for all m ≫ 0 as well. Hence it follows from (i) of proposition 1.4 that S−1M is a ϑ-Mathieu subspace over S−1R of S−1A. (v) Since am ∈ S−1M , there exist sm ∈ S such that smam ∈ φ(M ) for all m ≥ 1. Consequently, (s1s2 ··· sds′a)m ∈ φ(M ) for all d ∈ N, all m with 1 ≤ m ≤ d and all s′ ∈ S. By assumption, there exists an s′ ∈ S, and a monic f ∈ R[t], say of degree d, such that f (s′a) = 0. Therefore (s′a)m ∈ R · (s′a) + R · (s′a)2 + ··· + R · (s′a)d follows inductively for all m > d. By multiplication by 12 (s1s2 . . . sd)m on both sides, we see that (s1s2 ··· sds′a)m ∈ φ(M ) for all m > d as well. Thus (s1s2 ··· sds′a)m ∈ φ(M ) for all m ≥ 1, i.e. s = s1s2 ··· sds′ suffices. 3 Uniform Mathieu subspaces In this section, we generalize results of section 4 of [9], which is entitled 'Mathieu subspaces with algebraic radicals'. Hence you might expect a section about Mathieu subspaces with co-integral radicals, but it appears that such Mathieu subspaces are so-called uniform Mathieu subspaces, see theorem 3.9 below. Definition 3.1. Let M be an R-subspace (R-submodule) of an associative R- algebra A. Then we call M a uniform ϑ-Mathieu subspace of A if for all a ∈ A such that am ∈ M for all m ≥ 1, there exists an N ∈ N such that (aN )ϑ ⊆ M . Proposition 3.2. If M is a uniform ϑ-Mathieu subspace of an associative R- algebra A, then M is also a ϑ-Mathieu subspace of A. Proof. Assume M is a uniform ϑ-Mathieu subspace of an associative R-algebra A and am ∈ M for all m ≥ 1. Then there exists an N ∈ N such that (aN )ϑ ⊆ M , and we have the following when m ≥ N : (i) bam ∈ M for all b ∈ A, if ϑ = "left "; (ii) amc ∈ M for all c ∈ A, if ϑ = "right "; (iii) bam, amc ∈ M for all b, c ∈ A, if ϑ = "pre-two-sided "; (iv) bamc ∈ M for all b, c ∈ A, if ϑ = "two-sided ". Hence M is a ϑ-Mathieu subspace of A on account of definition 1.1. Notice that the difference between Mathieu subspace and uniform Mathieu sub- spaces is that for uniform Mathieu subspaces, the number N is the above propo- sition does not depend on the elements b and/or c of A, as opposed to regular Mathieu subspaces. The following propositions are analogs for uniform Mathieu subspaces of [9, Prop. 2.1] and proposition 1.4 respectively. Proposition 3.3 (following [9, Prop. 2.1] more or less). Let M be an R-subspace (R-submodule) of an associative R-algebra A. Then M is a uniform ϑ-Mathieu subspace of A, if and only if for all a ∈ r(M ), (aN )ϑ ⊆ M for some N ∈ N. Proof. In order to prove the 'only-if'-part, assume that M is a uniform ϑ- Mathieu subspace of A and a ∈ r(M ). Then there exists a k ∈ N such that (ak)m ∈ M for all m ≥ 1. Hence (akN )ϑ = ((ak)N )ϑ ⊆ M for some N ∈ N. This gives the 'only-if'-part. Since am ∈ M for all m ≥ 1 implies a ∈ r(M ), the 'if'-part follows as well. 13 Proposition 3.4. Let M be an R-subspace (R-submodule) of an associative R-algebra A. Then M is a uniform ϑ-Mathieu subspace of A, if and only if any of the following properties holds, where Cϑ(b, c) is as in definition 1.3: (i) for all a such that am ∈ M for all m ≥ 1, we have the following when m ≫ 0: bamc ∈ M for all b, c ∈ A such that Cϑ(b, c); (ii) for all a ∈ r(M ), we have the following when m ≫ 0: bamc ∈ M for all b, c ∈ A such that Cϑ(b, c). Proof. Comparing the proof of proposition 3.2 with definition 1.1, we see that an alternative definition of uniform Mathieu subspace can be obtained by inter- changing the quantification with m with that of b and/or c in the definition of Mathieu subspace as given in definition 1.1. Since this proposition and a possi- ble proof differs accordingly from proposition 1.4 and its proof respectively, the desired result follows. Remark 3.5. The proof of proposition 3.4 tells us how the proofs of several results about Mathieu subspaces can be turned into similar proofs for uniform Mathieu subspaces. Results with an analog for uniform Mathieu subspaces that can be proved in this manner are [9, Prop. 2.5 -- Lm. 2.8], proposition 1.5 and theorem 2.16. 3.1 Definitions of Gϑ(A) and Eϑ(A) ′(V ) = In [9], G(A) is defined as the set of all K-subspaces V of A such that r r(V ), where K = R is a field. Before we give another definition of G(A), we formulate a proposition. Recall that for subsets S of A, (S)ϑ is the ϑ-ideal generated by S when ϑ 6= "pre-two-sided ", and (S)"pre-two-sided" = AS + SA is the sum of the left and right ideals generated by S. Proposition 3.6. Let A be an associative R-algebra and V an R-subspace of A. Then r ′(V ) ⊆ r(cid:0)(e ∈ V e2 = e)ϑ(cid:1) ′(V ). From proposition 2.8, it follows that there exist an Proof. Take a ∈ r N ∈ N and an idempotent e ∈ V such that aN = aN e = eaN ∈ (e)ϑ. Hence a ∈ r(cid:0)(e ∈ V e2 = e)ϑ(cid:1). By proposition 3.6, condition (3.1) in the definition below is weaker than the condition r Definition 3.7. Let A be an associative R-algebra and ϑ 6= "pre-two-sided". Then we define Gϑ(A) as the set of all R-subspaces V of A, such that ′(V ) = r(V ) in [9]. Since the pre-two-sided case is a combination of both one-sided cases, we simply define r(V ) ⊆ r(cid:0)(e ∈ V e2 = e)ϑ(cid:1) (3.1) G"pre-two-sided"(A) := G"left"(A) ∩ G"right"(A) Let Eϑ(A) be the subset of ϑ-Mathieu subspaces of Gϑ(A). 14 We will show in corollary 3.23, which follows later, that in the commutative case, Eϑ(A) is just the set of all R-subspaces V of A for which we have equality in (3.1). The following theorem gives another definition of Gϑ(A) for the commutative case, namely r(V ) ⊆ r(cid:0)(r ′(V ))ϑ(cid:1) instead of (3.1), because r(cid:0)(e ∈ V e2 = e)ϑ(cid:1) is an ideal when A is commutative. Theorem 3.8. Let A be an associative R-algebra and V an R-subspace of A. Suppose that (e ∈ V e2 = e)ϑ is a ϑ-Mathieu subspace (which is obviously the case when ϑ 6= "pre-two-sided"). Suppose additionally that either r(cid:0)(e ∈ V e2 = e)ϑ(cid:1) or r ′(V ) is a ϑ-ideal of A. Then r(cid:0)(e ∈ V e2 = e)ϑ(cid:1) = r(cid:0)(r ′(V ))ϑ(cid:1) Proof. Let E = (e ∈ V e2 = e)ϑ. As e1R[e] = e2R[e] for every idempotent e, it follows that E ⊆ (r ′(V ))ϑ, so From proposition 3.6 and (ii) of [9, Lm. 2.2], we deduce that r(E) ⊆ r(cid:0)(r ′(V ))ϑ(cid:1) r ′(V ) ⊆ r(E) and respectively. If r(E) is a ϑ-ideal of A, then r(cid:0)r(E)(cid:1) = r(E) r(E) ⊆ r(cid:0)(r ′(V ))ϑ(cid:1) ⊆ r(cid:0)(r(E))ϑ(cid:1) = r(cid:0)r(E)(cid:1) = r(E) If r ′(V ) is a ϑ-ideal of A, then ′(V ))ϑ(cid:1) = r(cid:0)r r(E) ⊆ r(cid:0)(r ′(V ))ϑ(cid:1) in both cases. ′(V )(cid:1) ⊆ r(cid:0)r(E)(cid:1) = r(E) So r(E) = r(cid:0)(r The next theorem gives another definition of Eϑ. Theorem 3.9. Let A be an associative R-algebra. Then Eϑ(A) is the subset of uniform ϑ-Mathieu subspaces of Gϑ(A). Proof. The pre-two-sided case follows from both two-sided cases (take the largest of both N 's), so assume that ϑ 6= "pre-two-sided". Take any V ∈ Eϑ(A) and let a ∈ r(V ). By definition of Gϑ(A), we have aN ∈ (e ∈ V e2 = e)ϑ for some N ∈ N. Since (iii) of either lemma 2.7 or [9, Lm. 2.9] tells us that (e)ϑ ⊆ V for each idempotent e ∈ V , we see that (aN )ϑ ⊆ V . So V is a uniform ϑ-Mathieu subspace of A. ′(A). Proposition 3.10. Let A be an associative R-algebra such that A = r Then each R-subspace of A is contained in Gϑ(A) and each ϑ-Mathieu subspace of A is uniform. 15 ′(V ), and on account of Proof. Let V be an R-subspace of A. Then r(V ) = r proposition 3.6, we have V ∈ Gϑ(A) by definition of Gϑ. By definition of Eϑ, it follows from theorem 3.9 that each ϑ-Mathieu subspace of A is uniform. The rest of this section consists of generalizations of results of section 4 of [9]. Just as before, we may omit proofs if those of the original results are already sufficient. We start with a generalization of [9, Lm. 4.1]. Lemma 3.11 (generalizing [9, Lm. 4.1]). Let A be an associative algebra over an Artin ring R, and suppose that V is an R-subspace of A. Then for (1) A is integral (finite) over R; (2) V is integral (finite) over R; (3) every element of r(V ) is integral over R; (4) r(V ) = r ′(V ); (5) V ∈ Gϑ(A); we have (1) ⇒ (2) ⇒ (3) ⇒ (4) ⇒ (5). Proof. By [1, Th. 8.5] and [1, Th. 6.5], finite R-modules are Noetherian. By [1, Prop. 6.2], Noetherian R-modules are finite. Hence we can replace (finite) by (Noetherian) in (1) and (2). (1) ⇒ (2) Again by [1, Prop. 6.2] we obtain the Noetherian case of (1) ⇒ (2). The integral case of (1) ⇒ (2) is trivial. (2) ⇒ (3) Assume that (2) holds and take any a ∈ r(V ). We must show that a is integral over R, which is the same as that am is integral over R for some m ≥ 1. We can take m such that amR[am] is a subspace In the Noetherian case, of V . Now the integral case follows directly. R[am] = R · 1 + amR[am] is finite because of [1, Prop. 6.2], and am is integral over R by [1, Prop. 5.1]. (3) ⇒ (4) This follows directly from theorem 2.2. (4) ⇒ (5) This follows from proposition 3.6 and the definition of Gϑ(A). 3.2 Characterization of M ∈ Eϑ(A) in terms of idempo- tents Theorem 3.12 (following [9, Th. 4.2]). Let V ∈ Gϑ(A). Then V ∈ Eϑ(A), if and only if (e ∈ V e2 = e)ϑ ⊆ V . Proof. Just as in the proof of theorem 3.9, the pre-two-sided case follows by combining both one-sided cases. So assume again that ϑ 6= "pre-two-sided". The 'only-if'-part follows from (iii) of either lemma 2.7 or [9, Lm. 2.9]. 16 In order to prove the 'if'-part, assume that (e ∈ V e2 = e)ϑ ⊆ V , and take any a ∈ r(V ). By definition of Gϑ(A), we have aN ∈ (e ∈ V e2 = e)ϑ for some N ∈ N. Hence (aN )ϑ ⊆ (e ∈ V e2 = e)ϑ ⊆ V by assumption. So V is a uniform ϑ-Mathieu subspace of A. Corollary 3.13 (similar to [9, Cor. 4.3]). Let V ∈ Gϑ(A) such that V does not contain any nonzero idempotent. Then V is a (uniform) ϑ-Mathieu subspace of A. Just as in [9], let Iϑ,V denote the largest ϑ-ideal of A which is contained in V in case ϑ 6= "pre-two-sided ", and I"pre-two-sided",V := I"left",V + I"right",V Consequently, if r Proposition 3.14 (similar to [9, Prop. 4.5]). Let V ∈ Gϑ(A) such that Iϑ,V = (0)ϑ. Then V is a (uniform) ϑ-Mathieu subspace of A, if and only if V does not contain any nonzero idempotent. ′(A) = A and A has no proper ϑ-ideals other than (0)ϑ, then any proper R-subspace M of A is a (uniform) ϑ-Mathieu subspace of A, if and only if M does not contain any nonzero idempotent of A. Proof. By proposition 3.10, we can just follow the proof of [9, Prop. 4.5]. Corollary 3.15 (following [9, Cor. 4.6]). Let V be an R-subspace of an as- sociative R-algebra A and IV = I"two-sided",V . Assume that V ∈ G(A) or V /IV ∈ G(A/IV ), where G = G"two-sided". Then V is a (uniform) Mathieu subspace of A, if and only if V /IV does not contain any nonzero idempotent of the quotient R-algebra A/IV . Proof. By remark 3.5, we can just follow the proof of [9, Cor. 4.6], provided that we can prove that V ∈ G(A) implies V /IV ∈ G(A/IV ). So let V ∈ G(A). Since IV ⊆ V , we have am ∈ V ⇐⇒ (a + IV )m = am + IV ∈ V /IV Hence r(V )/IV = r(V /IV ). The forward implication still holds when we replace V by any E ⊆ A, so r(E)/IV ⊆ r(E/IV ) for any E ⊆ A. Now take E = (e ∈ V e2 = e)ϑ. Then E/IV ⊆ (e ∈ V /IV e2 = e)ϑ. Since r(V ) ⊆ r(E) by definition of G, we can conclude that r(V /IV ) = r(V )/IV ⊆ r(E)/IV ⊆ r(E/IV ) ⊆ r(cid:0)(e ∈ V /IV e2 = e)ϑ(cid:1) so that V /IV ⊆ G(A/IV ) by definition of G. Proposition 3.16 (following [9, Prop. 4.7] more or less). Assume that R is local and integrally closed in A. Then every V ∈ Gϑ(A) such that 1 /∈ V is a (uniform) ϑ-Mathieu subspace of A. 17 Proof. Since R is integrally closed in A and all idempotents of A are integral over R, we see that all idempotents of A must lie inside R · 1 ⊆ A. But on account of (i) of [1, Prop. 1.6], R · 1 is a local ring. Hence we deduce from lemma 2.11 that A has no idempotents other than 0 and 1. So if 1 /∈ V , then V does not contain any nonzero idempotent. Hence the desired result follows from corollary 3.13. 3.3 Posets of idempotents of R-subspaces of A If K is a field, and A ∋ a is a K-algebra, then it is clear that a is a quasi- idempotent, if and only if Ka contains a nonzero idempotent e. So (2) of [9, Prop. 4.8] is equivalent to that (2′) Ka does not have a nonzero idempotent and (e)ϑ ⊆ Ka for every idem- potent e of Ka. If (2) (or (2′)) does not hold, then Ka contains a nonzero idempotent e ∈ Ka, which is unique, and (e)ϑ = (a)ϑ. So (1) of [9, Prop. 4.8] can be replaced by (1′) Ka does have a nonzero idempotent and (e)ϑ ⊆ Ka for every idempotent e of Ka. Since Ka has at most one nonzero idempotent, the idempotens of Ka commute with one another. Hence the following proposition, which additionally shows that a is central in A in case Ka is a (pre-)two-sided Mathieu subspace and a is a quasi-idempotent, is indeed a generalization of [9, Prop. 4.8]. Proposition 3.17 (generalizing [9, Prop. 4.8]). Let A be an associative R- algebra. Suppose that V is an Artinian R-subspace of A, whose idempotents commute with one another, and let E := (e ∈ V e2 = e)ϑ. Then V is a (uniform) ϑ-Mathieu subspace of A, if and only if E ⊆ V . Furthermore, we have the following if V is indeed a ϑ-Mathieu subspace. (i) E is a ¯ϑ-unital associative algebra over R (with inherited ring operations and a multiplicative ¯ϑ-identity that differs from that of A), where ¯ϑ is ϑ with "left" and "right" interchanged. (ii) If V is a (pre-)two-sided Mathieu subspace of A, then the idempotents of V are central in A, and hence E does not depend on the choice of ϑ. In particular, V is a two-sided Mathieu subspace and E is a unital Abelian ring (with inherited ring operations and a multiplicative identity that differs from that of A) in that case. Proof. Assume that a ∈ r(V ), say that am ∈ V for all m ≥ N . Since the co-integrality of a is just the decending chain condition on RaN + RaN +1 + RaN +2 + ··· ⊇ RaN +1 + RaN +2 + ··· ⊇ RaN +2 + ··· ⊇ ··· ′(V ), and by (4) ⇒ (5) of lemma 3.11, we we see that a ∈ r have V ∈ Gϑ(A). Therefore, it follows from theorem 3.12 that V is a (uniform) ′(V ). Hence r(V ) = r 18 ϑ-Mathieu subspace of A, if and only if E ⊆ V . So it remains to prove (i) and (ii). (i) We only need to prove the case ϑ = "left", because the case ϑ = "right" is similar and the other two cases follow from (ii). So assume that ϑ = "left". Notice that for each idempotent e of A, ¯e := 1−e is another idempotent of A, and we have e¯e = 0. On account of Zorn's lemma and the descending chain condition of R-subspaces of V , we can choose an idempotent e ∈ V such that (¯e)ϑ ∩ V is minimal. Now take an arbitrary idempotent e′ ∈ V . Then e′′ := e + e′ − ee′ is contained in V , and e′′ is an idempotent because ¯e′′ = ¯e¯e′ = ¯e′¯e Furthermore, (¯e′)ϑ ∩ V ⊇ (¯e′′)ϑ ∩ V ⊆ (¯e)ϑ ∩ V , and the minimality of (¯e)ϑ ∩ V tells us that (¯e′)ϑ ∩ V ⊇ (¯e′′)ϑ ∩ V = (¯e)ϑ ∩ V . In particular, (¯e′)ϑ ⊇ (¯e′′)ϑ ∩ V ⊇ (¯e)ϑ ∩ (e′)ϑ Consequently, ¯ee′ = a¯e′ for some a ∈ A, and multiplication by e′ gives ¯ee′ = 0. Thus ee′ = (1 − ¯e)e′ = e′ − ¯ee′ = e′. Since e′ was arbitrary, we have e′ = e′e = ee′ for all idempotents e′ ∈ V . Using that E is a left ideal generated by elements with respect to which e is a right identity, we obtain that E is right-unital. (ii) Assume that V is a pre-two-sided Mathieu subspace of A. Take any idem- potent e′ ∈ V and take a ∈ A arbitrary. Then e′ + ¯e′ae′ is an idempotent as well, and by taking ϑ = "left", we see that e′ + ¯e′ae′ ∈ V , too. Since both e′ and e′ + ¯e′ae′ are idempotents of V , we have e′ + ¯e′ae′ = (e′ + ¯e′ae′)e′ = e′(e′ + ¯e′ae′) = e′ Hence ¯e′ae′ = 0. Adding e′ae′ gives ae′ = e′ae′ and e′ae′ = e′a follows in a similar manner. So every idempotent e′ ∈ V is central in A, and therefore E does not depend on ϑ. In particular, E is a two-sided ideal of A, so V has to be a (uniform) two-sided Mathieu subspace of A. Furthermore, the right identity e of E that we get by taking ϑ = "left" in (i), is a two-sided multiplicative identity of E. It is well-known that the idempotents of unital rings are central once they commute relatively (or with all nilpotent elements). The proof of that is based on the idempotence of e′ + ¯e′ae′ (the nilpotence of ¯e′ae′), so the idea to use that idempotent in the above proof was obvious. The assumption that the idempotents of V in proposition 3.17 commute relatively ensures that they form a lattice with respect to e ∧ e′ := ee′, e ∨ e′ := e + e′ − ee′, and (e ≤ e′) := (e = ee′ = e′e) (the idempotent property is just the 19 reflexivity of ≤). From the above proof, we can deduce that in some cases, that lattice has a top element. To obtain the existence of a top element, it is however not needed to make any assumptions on a certain lattice structure on the poset of idempotents whose ordering is given by (e ≤ e′) := (e = ee′ = e′e). More generally, one can even show that the lattice of idempotents of V must be complete, by proving the following. Proposition 3.18. Let A be an associative R-algebra and L be a set of idem- potents of A, which is a poset with respect to (e ≤ e′) := (e = ee′ = e′e). Suppose that every commutative R-subspace, generated by the multiplicative clo- sure of a subset of L, is either Noetherian or Artinian over R. Then we have the following. (i) Every chain of L has both a minimum and a maximum element. (ii) If L admits a lattice structure, then for every S ⊆ L there exist a finite S ′ ⊆ S such that V S = V S ′ and W S = W S ′. In particular, every lattice structure over L is complete. Proof. We only prove the Noetherian case here. Using ideas in the proof of 3.17, the reader may treat the Artinian case himself. Again, we write ¯e := 1 − e for idempotents e. (i) Let S be a chain of L. Since e ≤ e′ implies ee′ = e′e, we see that R[S] is commutative. Consequently, V := SR[S] is Noetherian over R and hence over R[S] as well by assumption. So we can take e∨, e∧ ∈ S such that e∨V and ¯e∧V are maximal. Suppose that there exists an e ∈ S such that e ≮ e∨. Then e ≥ e∨, thus ee∨ = e∨. Hence eV ⊇ e∨V , so eV = e∨V by definition of e∨. Multiplication of eV = e∨V by ¯e∨ and ¯e respectively gives ¯e∨eV = 0 = ¯ee∨V . By taking the elements e and e∨ of V respectively, we see that ¯e∨e = 0 = ¯ee∨. Adding e∨e subsequently gives e = e∨. Hence e ≤ e∨ for all e ∈ S, i.e. e∨ is maximum. Suppose next that there exists an e ∈ S such that e ≯ e∧. Then e ≤ e∧, thus e = ee∧. Multiplication by ¯e∧ gives e¯e∧ = 0, and adding ¯e¯e∧ subsequently gives ¯e∧ = ¯e¯e∧. Consequently, ¯e∧V ⊆ ¯eV , so ¯e∧V = ¯eV by definition of e∧. Multiplication of ¯e∧V = ¯eV by e and e∧ respectively gives ¯e∧eV = 0 = ¯ee∧V . Now a similar argument as in the previous paragraph tells us that e = e∧. Hence e ≥ e∧ for all e ∈ S, i.e. e∧ is minimum. (ii) Assume that L admits a lattice structure. Take any subset S of L. On account of (i) and Zorn's lemma, we can choose e∧, e∨ ∈ S which are minimal and maximal respectively. If S is closed under dyadic ∧ or ∨, then e∧ and e∨ have to be minimum and maximum respectively in S (see the proof of proposition 3.17). In general, we can take the closure ¯S of S under dyadic ∧ or ∨, and obtain that e∧ and e∨ are a finite meet and join of elements of S respectively (just like any element of ¯S). Furthermore, we can take for S ′ the union of both underlying finite sets. 20 3.4 Radicals of uniform ϑ-Mathieu subspaces in terms of radicals of Iϑ,M Lemma 3.19 (following [9, Lm. 4.9]). Let A be an associative R-algebra and M a ϑ-Mathieu subspace of A. Then r If M is even a uniform ϑ-Mathieu subspace of A, then r(M ) = r(Iϑ,M ). ′(M ) = r ′(Iϑ,M ). ′(Iϑ,M ) for ϑ-Mathieu subspaces M of A follows Proof. The claim r from the last claim of theorem 2.6. The last assertion of lemma 3.19 follows in a similar manner from the definition of uniform ϑ-Mathieu subspace. ′(M ) = r Theorem 3.20 (following [9, Th. 4.10] more or less). Let M be an R-subspace of A. Then the following statements are equivalent. (1) M is a uniform ϑ-Mathieu subspace of A, (2) r(M ) = r(Iϑ,M ). (3) For every R-subspace V of A such that Iϑ,M ⊆ V ⊆ M , V is a uniform ϑ-Mathieu subspace of A and r(V ) = r(Iϑ,M ). Proof. Since (1) ⇒ (2) follows from lemma 3.19 and (3) ⇒ (1) is trivial, we assume (2) to show (2) ⇒ (3). Let V be an R-subspace V of A such that Iϑ,M ⊆ V ⊆ M . Then r(V ) = r(Iϑ,M ) because of (2). Take a ∈ r(V ) arbitrary. Then a ∈ r(Iϑ,M ) and there exists an N ∈ N such that aN ∈ Iϑ,M and (aN )ϑ ⊆ Iϑ,M ⊆ V . Since a was arbitrary, we conclude that V is a uniform ϑ-Mathieu subspace of A, and (3) follows. Corollary 3.21 (following [9, Cor. 4.11]). Let A be a ϑ-simple associative R- algebra (i.e. A has no proper ϑ-ideals other than zero) and M a proper uniform ϑ-Mathieu subspace of A. Then r(M ) = r((0)ϑ) and all R-subspaces V ⊆ M are uniform (two-sided) Mathieu subspaces of A. Proof. Since A is ϑ-simple, we have Iϑ,M = (0)ϑ. By theorem 3.20, we get r(M ) = r((0)ϑ). Since (0)ϑ ⊆ V for all R-subspaces V of A, we additionally deduce from theorem 3.20 that all R-subspaces V ⊆ M are uniform two-sided Mathieu subspaces of A. If A is Abelian, then the definition of Gϑ(A) does not depend on ϑ. For that rea- son, we simply write G(A) in that case. The assumption that A is commutative in [9, Th. 4.12] can be weakened (or be replaced by that A is reduced to ensure that idempotents are central), but not without eliminating the 'commutative fact' that radicals of ideals are ideals themselves. Theorem 3.22 (generalizing [9, Th. 4.12]). Let A be an Abelian associative R-algebra and V ∈ G(A). If ϑ 6= "pre-two-sided", then V is a (uniform) (ϑ-) Mathieu subspace of A, if and only if r(V ) is the radical of some ϑ-ideal of A, which is the case when r(V ) is a ϑ-ideal itself. 21 Proof. Assume that ϑ 6= "pre-two-sided". The 'only if'-part follows directly from lemma 3.19. To prove the 'if'-part and the last claim along with it, suppose that r(V ) ∈ {J, r(J)} for some ϑ-ideal J of A. Since A is abelian, we obtain that (e)ϑ does not depend on ϑ. Hence by theorem 3.12, it suffices to show that (e)ϑ ⊆ V for every idempotent e ∈ V . So take any idempotent e ∈ V . Since the idempotents of any subset of A coincides with those of its radical, we have e ∈ J in any case (both when r(V ) = J and when r(V ) = r(J)). Using that (e)ϑ and J are ϑ-ideals, we deduce that (e)m ϑ ⊆ (e)ϑ ⊆ J for every m ≥ 1. Hence (e)ϑ ⊆ r(V ) in any case (both when r(V ) = J and when r(V ) = r(J)). So for arbitrary b ∈ A, we have ebm = (eb)m ∈ V when m ≫ 0. Consequently, for the R subspace Ve of A consisting of elements a ∈ A such that ea ∈ V , we have r(Ve) = A. Therefore, we have Ve = A on account of [9, Lm. 2.4], i.e. (e)ϑ ⊆ V . Corollary 3.23 (following [9, Cor. 4.13] more or less). Let A be an Abelian associative R-algebra and V ∈ G(A). If ϑ 6= "pre-two-sided" and r(V ) is a ϑ- ideal, then r(V ) is both a two-sided ideal itself and the radical of some two-sided ideal. 3.5 Unions and intersections of Mathieu subspaces M ∈ Eϑ(A) Proposition 3.24 (combining [9, Prop. 4.16] and [9, Prop. 4.18]). Let Mi (i ∈ I) be a family of proper ϑ-Mathieu subspaces of an associative R-algebra A. (i) If Ti∈I Mi ∈ Gϑ(A), then Ti∈I Mi is a proper (uniform) ϑ-Mathieu sub- (ii) If Si∈I Mi ∈ Gϑ(A), then Si∈I Mi is a proper (uniform) ϑ-Mathieu sub- space of A. space of A. More generally, suppose that J is a set of subsets of I, and define MJ := [j∈J \i∈j Mi (iii) If MJ ∈ Gϑ(A), then MJ is a proper (uniform) ϑ-Mathieu subspace of A. Using proposition 3.10 instead of [9, Lm. 4.1], we obtain the following from the paragraph that precedes [9, Prop. 4.20]. Proposition 3.25 (similar to [9, Prop. 4.20]). Let A be an associative R-algebra ′(A). If V is an R-subspace of A, then the following statements such that A = r hold. (i) There exists at least one ϑ-Mathieu subspace which is maximal among all the ϑ-Mathieu subspaces of A contained in V . 22 (ii) There exists a unique ϑ-Mathieu subspace which is minimum among all the ϑ-Mathieu subspaces of A containing V , namely the intersection of all ϑ-Mathieu subspaces of A that contain V . (iii) Let F be the collection of proper ϑ-Mathieu subspaces M of A such that M ⊇ V . If F 6= ∅, then F has at least one maximal element and a unique minimum element. Theorem 3.26 (following [9, Th. 4.21]). Let A be an associative R-algebra such ′(A) = A. Then every proper ϑ-Mathieu subspace of A is contained in a that r maximal proper ϑ-Mathieu subspace of A. In particular, A has at least one maximal proper ϑ-Mathieu subspace. If this ϑ-Mathieu subspace cannot be taken nonzero, then R · 1 = A and A is a field. Proof. Except for the last claim, we can just follow the proof of [9, Th. 4.21]. So assume that A does not have a nonzero proper ϑ-Mathieu subspace. By [9, Th. 6.2], A is a field which is isomorphic to the fraction field of R · 1, which in turn is an integral domain. Hence it suffices to show that R · 1 is a field itself. So let a ∈ R · 1. By ′(A) = A, a is co-integral over R. Now the desired result follows from (ii) of proposition 2.8, because R · 1 is a domain. Corollary 3.27 (similar to [9, Cor. 4.22]). Let V be an R-subspace of an asso- ′(A), and assume that the ϑ-ideal generated ciative R-algebra A such that A = r by V is not the whole algebra A. Then there exists a maximal nonzero ϑ-Mathieu subspace M of A such that V ⊆ M . r 4 Strongly simple algebras Our starting point is the following theorem, which is the main theorem of section 6 of [9]. Theorem 4.1 (following [9, Th. 6.2]). Assume R is a nontrivial commutative ring and A a nontrivial associative R-algebra such that A has no proper nonzero ϑ-Mathieu subspaces. Then R · 1 is an integral domain and A is isomorphic to the field of fractions of R · 1. The following definition appears at the beginning of section 6 in [9], which has the same title as this section. The above theorem says that variants of the below definition with ϑ 6= "two-sided " are unnecessary, since A is a field regardless of what ϑ is. Definition 4.2. Let R be a commutative ring and A an associative R-algebra. We say that A is strongly simple if A has no proper nonzero Mathieu subspaces over R. Since R-algebras are (R · 1)-algebras and strongly simple algebras over R can only be fraction fields of R · 1, we restrict to integral domains R with fraction field A = K from now on. 23 In an earlier version of [9], the author conjectured that the only strongly simple algebras would be fields over theirselves. But this is not true. The main result of this section is the following. Theorem 4.3. Let R be an integral domain with fraction field K. Then the following statements hold. (i) Each prime ideal p ⊂ R of height one is of the form p = M ∩ R for some Mathieu subspace M of K over R. (ii) If r ∈ R is anti-Archimedean, i.e. T∞ Mathieu subspace of K over R that contains r. n=1 rnR 6= (0), then there is no proper Corollary 4.4. Let R be an integral domain with fraction field K. Then for the following statements: (1) R is anti-Archimedian, i.e. every nonzero r ∈ R is anti-Archimedian; (2) K is strongly simple over R; (3) R does not have a prime ideal of height one; we have (1) ⇒ (2) ⇒ (3). Proof. (1) ⇒ (2) Since every nonzero Mathieu subspace of K over R contains a nonzero element of R (the numerator), the desired result follows from (ii) of theorem 4.3. (2) ⇒ (3) This follows from (i) of theorem 4.3. Proof of theorem 4.3. (i) Fix a prime ideal of height one of R and let S be the set of all elements of R that are not contained in the prime ideal at hand. By replacing R by the localization S−1R, we may assume that R is a local ring of dimension 1. Then it is known that there exists a valuation ring D (with the same fraction field K) that dominates R, i.e. D contains R and R ∩ mD = mR, where mR denotes the maximal ideal of the local ring R. Assume q is a nonzero prime ideal of D. Since K is the fraction field of R, the contraction q ∩ R of q, which is a prime ideal, is nonzero. But R has dimension one, so q ∩ R = mR. Now define M as the intersection of all nonzero prime ideals q of D. Then M ∩ R = mR. It is known that the ideals of D are totally ordered by inclusion, from which we can deduce that M is a prime ideal of D. So M is a prime ideal of height one. From lemma 4.5 below, it follows that M is a Mathieu subspace of K over D ⊇ R. 24 (ii) Assume r ∈ R is not Archimedian. Then there exists a nonzero a ∈ R such that rn a for all n ∈ N. Let M ∋ r be a Mathieu subspace of K over R. Then rn/a ∈ M for some n ∈ N. Since rn a, we have rnb = a for some b ∈ R. It follows that 1 = b rn/a ∈ M . Thus M = K by [9, Cor. 2.10]. Lemma 4.5. Assume D is a valuation ring with maximal ideal mD and fraction field K. Then any prime ideal of height one of D is a Mathieu subspace of K over D. Proof. Let p be a prime ideal of height one of D. Assume that p is not a Mathieu subspace of K over D, say that a ∈ p and b ∈ K such that an+1b /∈ p for infinitely many n ∈ N. Write b = t/d with t, d ∈ D. Since ta ∈ p, we have an/d /∈ D for infinitely many n ∈ N. Thus v(d) > v(an) = n v(a) for infinitely many n ∈ N. Hence the ideal q of D consisting of elements c such that v(c) > nv(a) for infinitely many n ∈ N is nonzero. In fact, it is a prime ideal, since v(c1c2) > 2nv(a) (for infinitely many n ∈ N) implies v(c1) > nv(a) or v(c2) > nv(a) (for infinitely many n ∈ N). Since q ( p, we have a contradiction with the height one assumption on p, so p is a Mathieu subspace of K over D. Example 4.6. Let k be a field. The valuation domain D = k[[x1, x2, x3, . . .]](cid:20) x2 x1 , x2 x2 1 , x2 x3 1 , . . .(cid:21)(cid:20) x3 x2 , x3 x2 2 , x3 x3 2 , . . .(cid:21) ··· with fraction field K = k((x1, x2, x3, . . .)) = k[[x1, x2, x3, . . .]][x−1 3 , . . .] 1 , x−1 2 , x−1 has value group Z[t], ordered by f (t) > g(t) ⇔ lc(f (t)−g(t)) > 0. Here, lc(f (t)) denotes the leading coefficient of f with respect to t. The value function v is defined by v(a) = min{degx1 s + t degx2 s + t2 degx3 s + ··· s is a term of a} 1 xα2 2 ··· xαn where min is with respect to the above ordering on Z[t]. So if a has a term s = xα1 n , and for every other term s′ of a, s′/s has positive degree in the variable of largest index which appears in s′/s, i.e. in which s′/s has nonzero degree, then v(a) = αntn−1 + αn−1tn−2 + ··· + α1. If g(t) ∈ Z[t], then there exists a h(t) ∈ Z[t] with positive leading coefficient such that deg h > deg g. Hence ng(t) < h(t) for all n ∈ N. On account of (1) ⇒ (5) of proposition 4.7 below, K is strongly simple over D. The following proposition classifies the valuation rings that are strongly simple. Proposition 4.7. Assume D is a valuation ring with maximal ideal mD and fraction field K. Then the following statements are equivalent. (1) D is not strongly simple, 25 (2) D has a prime ideal of height 1, (3) D has a nonzero element that is contained in every nonzero prime ideal of D, (4) D has a nonzero Archimedian element, i.e. an element a such that ∞ \n=1 Dan = (0) (5) The value group G of D has an element g such that for all h ∈ G, there exists an n ∈ N such that ng > h. Proof. (1) ⇒ (5) Assume D is not strongly simple. Let M be a proper nonzero Mathieu subspace of K over D. If M * mD, then M has an element a such that v(a) ≤ 0 and thus 1/a ∈ D, whence 1 = 1/a · a ∈ M . This contradicts [9, Cor. 2.10], so M ⊆ mD. Let a ∈ M be nonzero and b ∈ K. Take n ∈ N such that an/b ∈ M . Since M ⊆ mD, we have n v(a) − v(b) = v(an/b) > 0. Thus g = v(a) suffices. (2) ⇒ (1) This follows from lemma 4.5. 1 cn (3) ⇒ (2) Assume a is nonzero and contained in every nonzero prime ideal of D. Let p be the radical of (a). Then p is contained in every nonzero prime ideal of D, thus it suffices to show that p is prime. So assume c1c2 ∈ p. Then a cn 2 for some n, thus v(a) < nv(c1) + nv(c2). Hence v(a) < 2nv(c1) or v(a) < 2nv(c2). In the first case, a c2n 1 and thus c1 ∈ p. In the second case, c2 ∈ p. Thus p is prime, and its height is equal to one. (4) ⇒ (3) Assume a is a nonzero Archimedian element of D. Assume q is a nonzero prime ideal and b ∈ q. Since a is Archimedian, we have an ∤ b for some n ∈ N. Since D is a valuation ring, b an follows. Consequently a ∈ q, as desired. (5) ⇒ (4) Assume (5) and that g is as in (5). Take a such that v(a) = g. Then for all b ∈ D, there exists an n ∈ N such that v(an) = ng > v(b). Hence an ∤ b. This gives the desired result. References [1] M.F. Atiyah and I.G. Macdonald, Introduction to commutative algebra, Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont., 1969. [2] J.J. Duistermaat and W. van der Kallen, Constant terms in powers of a Laurent polynomial, Indag. Math. (NS) 9 (1998), no. 2, 221 -- 231. 26 [3] A. van den Essen, The Amazing Image Conjecture, arXiv:1006.5801v1, June 2010. [4] A. van den Essen and W. Zhao, Two results on homogeneous Hessian nilpo- tent polynomials, J. Pure Appl. Algebra 212 (2008), no. 10, 2190 -- 2193. [5] O. Mathieu, Some conjectures about invariant theory and their applications, Alg`ebre non commutative, groupes quantiques et invariants (Reims, 1995), S´emin. Congr. 2, Soc. Math. France, 263 -- 279 (1997). [6] W. Zhao, Hessian nilpotent polynomials and the Jacobian Conjecture, Trans. Amer. Math. Soc., 359 (2007), no. 1, 294 -- 274. [7] -- , A Vanishing Conjecture on differential operators with constant oeffi- cients, Acta Math. Vietnam., Vol. 32 (2007), no. 3, 259 -- 286. See also arXiv:0704.1691v2. [8] -- , Images of commuting differential operators of order one with constant leading coefficients, J. Algebra 324 (2010), no. 2, 231 -- 247. [9] -- , Mathieu subspaces of associative algebras, J. Algebra 350 (2012), no. 1, 245 -- 272. 27
1101.5107	1	1101	2011-01-26T17:03:53	The largest left quotient ring of a ring	[ "math.RA" ]	The left quotient ring (i.e. the left classical ring of fractions) $Q_{cl}(R)$ of a ring $R$ does not always exist and still, in general, there is no good understanding of the reason why this happens. In this paper, it is proved existence of the largest left quotient ring $Q_l(R)$, i.e. $Q_l(R) = S_0(R)^{-1}R$ where $S_0(R)$ is the largest left regular denominator set of $R$. It is proved that $Q_l(Q_l(R))=Q_l(R)$; the ring $Q_l(R)$ is semi-simple iff $Q_{cl}(R)$ exists and is semi-simple; moreover, if the ring $Q_l(R)$ is left artinian then $Q_{cl}(R)$ exists and $Q_l(R) = Q_{cl}(R)$. The group of units $Q_l(R)^$ of $Q_l(R)$ is equal to the set $\{s^{-1} t\, \| \, s,t\in S_0(R)\}$ and $S_0(R) = R\cap Q_l(R)^$. If there exists a finitely generated flat left $R$-module which is not projective then $Q_l(R)$ is not a semi-simple ring. We extend slightly Ore's method of localization to localizable left Ore sets, give a criterion of when a left Ore set is localizable, and prove that all left and right Ore sets of an arbitrary ring are localizable (not just denominator sets as in Ore's method of localization). Applications are given for certain classes of rings (semi-prime Goldie rings, Noetherian commutative rings, the algebra of polynomial integro-differential operators).	math.RA	math	The largest left quotient ring of a ring V. V. Bavula Abstract The left quotient ring (i.e. the left classical ring of fractions) Qcl(R) of a ring R does not always exist and still, in general, there is no good understanding of the reason why this happens. In this paper, it is proved existence of the largest left quotient ring Ql(R), i.e. Ql(R) = S0(R)−1R where S0(R) is the largest left regular denominator set of R. It is proved that Ql(Ql(R)) = Ql(R); the ring Ql(R) is semi-simple iff Qcl(R) exists and is semi-simple; moreover, if the ring Ql(R) is left artinian then Qcl(R) exists and Ql(R) = Qcl(R). The group of units Ql(R)∗ of Ql(R) is equal to the set {s−1t s, t ∈ S0(R)} and S0(R) = R ∩ Ql(R)∗. If there exists a finitely generated flat left R-module which is not projective then Ql(R) is not a semi-simple ring. We extend slightly Ore's method of localization to localizable left Ore sets, give a criterion of when a left Ore set is localizable, and prove that all left and right Ore sets of an arbitrary ring are localizable (not just denominator sets as in Ore's method of localization). Applications are given for certain classes of rings (semi-prime Goldie rings, Noetherian commutative rings, the algebra of polynomial integro-differential operators). Key Words: the largest left quotient ring of a ring, the largest left (regular) denominator set of a ring, the classical left quotient ring of a ring, denominator set, the maximal left quotient rings of a ring. Mathematics subject classification 2000: 16U20, 16P40, 16S32, 13N10. Contents 1. Introduction. 2. The largest left quotient ring of a ring. 3. The maximal left quotient rings of a ring. 4. The largest quotient ring of a ring. 5. Examples. 1 Introduction The aim of the paper is to introduce the following new concepts and to prove their existence for an arbitrary ring: the largest left quotient ring of a ring, the largest left (regular) denominator set of a ring, the maximal left quotient rings of a ring, the largest (two-sided) quotient ring of a ring, the maximal (two-sided) quotient rings of a ring, a (left) localization maximal ring. Throughout, module means a left module; R is a ring with 1; CR is the set of (left and right) regular elements of the ring R (i.e. CR is the set of non-zero-divisors of R); Qcl(R) is the left quotient ring of R (if exists); Denl(R, 0) is the set of regular left denominator sets S in R (S ⊆ CR). The largest left regular denominator set and the largest left quotient ring of a ring. • (Theorem 2.1) There exists the largest (w.r.t. inclusion) left regular denominators set S0(R) in R and so Ql(R) := S0(R)−1R is the largest left quotient ring of R. • (Corollary 2.5) The set Denl(R, 0) of left regular denominator sets of R is a complete lattice and an abelian monoid. 1 The next theorem describes the group of units Ql(R)∗ of the ring Ql(R) and its connection with S0(R). • (Theorem 2.8) 1. S0(Ql(R)) = Ql(R)∗ and S0(Ql(R)) ∩ R = S0(R). 2. Ql(R)∗ = hS0(R), S0(R)−1i, i.e. the group of units of the ring Ql(R) is generated by the sets S0(R) and S0(R)−1 := {s−1 s ∈ S0(R)}. 3. Ql(R)∗ = {s−1t s, t ∈ S0(R)}. 4. Ql(Ql(R)) = Ql(R). The next theorem gives an answer to the question of when the ring Ql(R) is semi-simple (Theorem 2.8 is a key step in proving Theorem 2.9, statements 2-5 is Goldie's Theorem). • (Theorem 2.9) The following statements are equivalent. 1. Ql(R) is a semi-simple ring. 2. Qcl(R) exists and is a semi-simple ring. 3. R is a left order in a semi-simple ring. 4. R has finite left rank, satisfies the ascending chain condition on left annihilators, and is a semi-prime ring. 5. A left ideal of the ring R is essential iff it contains a regular element. If one of these conditions hold then S0(R) = CR and Ql(R) = Qcl(R). The next result is a sufficient condition for the ring Ql(R) not being a semi-simple ring. • (Corollary 2.12) If there exists a finitely generated flat R-module which is not projective then the ring Ql(R) is not semi-simple. The next corollary gives a sufficient condition for existing of Qcl(R). • (Corollary 2.10) If Ql(R) is a left artinian ring then S0(R) = CR and Ql(R) = Qcl(R). Let Sr 0(R) and Qr(R) := RSr 0(R)−1 be the largest regular right denominator set in R and the largest right quotient ring of R, respectively. In general, the following natural questions have negative answers as the algebra I1 := Khx, d a field K of characteristic zero demonstrates [2]: dx ,R i of polynomial integro-differential operators over 0(R)? 0(R) or S0(R) ⊇ Sr Question 1. Is S0(R) = Sr Question 2. Is S0(R) ⊆ Sr Question 3. Are the rings Ql(R) and Qr(R) isomorphic? Though, for the algebra I1 the next question has positive answer [2]. Question 4. Are the rings Ql(R) and Qr(R) anti-isomorphic? Remark. The algebra I1 is neither left nor right Noetherian, it contains infinite direct sums of 0(R)? nonzero left (and right) ideals, neither left nor right quotient ring of I1 exists [2]. Notation: • Orel(R) := {S S is a left Ore set in R}; • Denl(R) := {S S is a left denominator set in R}; • Locl(R) := {S−1R S ∈ Denl(R)}; • Assl(R) := {ass(S) S ∈ Denl(R)} where ass(S) := {r ∈ R sr = 0 for some s = s(r) ∈ S}; 2 • Denl(R, a) := {S ∈ Denl(R) ass(S) = a} where a ∈ Assl(R); • Sa = Sa(R) = Sl,a(R) is the largest element of the poset (Denl(R, a), ⊆) and Qa(R) := a R is the largest left quotient ring associated to a, Sa exists (Theorem 2.1.(2)); Ql,a(R) := S−1 • In particular, S0 = S0(R) = Sl,0(R) is the largest element of the poset (Denl(R, 0), ⊆) and Ql(R) := S−1 0 R is the largest left quotient ring of R; • Locl(R, a) := {S−1R S ∈ Denl(R, a)}. For each denominator set S ∈ Denl(R, a) where a := ass(S), there are natural ring homomor- phisms R π→ R/a → S−1R. In Section 3, connections between the denominator sets Denl(R, a), Denl(R/a, 0) and Denl(S−1R, 0) are established (Lemma 3.2, Lemma 3.3 and Proposition 3.4; these results are too technical to explain in the Introduction). They are used to prove the follow- ing results. • (Lemma 3.3.(2)) Let a ∈ Assl(R) and π : R → R/a, a 7→ a+a. Then π−1(S0(R/a)) = Sa(R), π(Sa(R)) = S0(R/a) and Qa(R) = Ql(R/a). • (Lemma 3.7.(2)) The set max.Denl(R) of maximal elements of the poset (Denl(R), ⊆) of left denominator sets of an arbitrary ring R is a non-empty set. This means that the set max.Locl(R) of maximal left quotient rings of an arbitrary ring R is a non-empty set, and max.Locl(R) = {S−1R S ∈ max.Denl(R)}. Example. max.Denl(I1) = {I1\F } (Proposition 5.8.(3)) where F is the only proper ideal of the algebra of polynomial integro-differential operators I1 and max.Locl(I1) = {Frac(A1)} (Proposition 5.8.(2)) where Frac(A1) is the Weyl skew field, i.e. the classical (left and right) ring of fractions of the first Weyl algebra A1 = Khx, x dx i. A new class of rings is introduced, the class of left localization maximal rings, see Section 3 (intuitively, one can invert nothing more in these rings). For an arbitrary ring R, a criterion is given in terms of this class of rings of when a left localization S−1R of R is a maximal left localization of R. • (Theorem 3.11) Let a ring A be a left localization of a ring R, i.e. A ∈ Locl(R, a) for some a ∈ Assl(R). Then A ∈ max.Locl(R) iff Ql(A) = A and Assl(A) = {0}, i.e. A is a left localization maximal ring. Example. Let V be an infinite dimensional vector space with countable basis over a field K, let C := {ϕ ∈ EndK(V ) dimK(im(ϕ)) < ∞} be the ideal of compact operators/linear maps of the algebra EndK(V ). Then the factor algebra EndK(V )/C is a left localization maximal ring (Theorem 5.2.(5)). Moreover, the set F of Fredholm linear maps in V is the maximal left (resp. right; left and right) denominator set of the ring EndK(V ) and F −1EndK(V ) ≃ EndK(V )F −1 ≃ EndK(V )/C is the maximal left (resp. right; left and right) localization ring of EndK(V ) (Theorem 5.2). A slight extension/generalization of Ore's method of localization. Ore's method of left localization says that we can localize a ring R precisely at the left denominator sets Denl(R) of the ring R. Notice that each left denominator set is a left Ore set but not the other way round, in general. We introduce the concept of a localizable left Ore set and give a criterion of when a left Ore set is localizable (Theorem 3.15). Each left denominator set is a localizable left Ore set but not vice versa, in general. We extend Ore's method of localization to localizable left Ore sets and prove an analogue of Ore's Theorem (by using Ore's Theorem) for localizable left Ore sets. 3 • (Corollary 3.16) Let S be a localizable left Ore set in a ring R. Then there exists an ordered pair (Q, f ) where Q is a ring and f : R → Q is a ring homomorphism such that (i) for all s ∈ S, f (s) is a unit in Q; and if (Q′, f ′) is another pair satisfying the condition (i) then there is a unique ring ho- momorphism h : Q → Q′ such that f ′ = hf . The ring Q is unique up to isomorphism. The ring Q is isomorphic to the left localization of the ring R/p(S) at the left denomina- tor set π(S) ∈ Denl(R/p(S), 0) where the ideal p(S) of the ring R is defined in (11) and π : R → R/p(S), a 7→ a + p(S). In Section 4, we prove two-sided analogues of some of the results of Sections 2 and 3. In most cases the proofs are easy corollaries of the corresponding one-sided (i.e. left and right) results but there are some surprises. In particular, • (Theorem 4.15) Every (left and right) Ore set is localizable. Therefore we can localize at all the (left and right) Ore sets not just at the (left and right) denominator sets as in Ore's method of localization. • (Corollary 4.16) Let S be an Ore set in a ring R. Then there exists an ordered pair (Q, f ) where Q is a ring and f : R → Q is a ring homomorphism such that (i) for all s ∈ S, f (s) is a unit in Q; and if (Q′, f ′) is another pair satisfying the condition (i) then there is a unique ring ho- momorphism h : Q → Q′ such that f ′ = hf . The ring Q is unique up to isomor- phism. The ring Q is isomorphic to the localization of the ring R/p(S) at the denomi- nator set π(S) ∈ Den(R/p(S), 0) where the ideal p(S) of the ring R is defined in (19) and π : R → R/p(S), a 7→ a + p(S). In Section 5, the largest (left; right; left and right) and maximal (left; right; left and right) quotient rings are found for following rings: the endomorphism algebra EndK(V ) of an infinite dimensional vector space V with countable basis, semi-prime Goldie rings, the algebra I1 of poly- nomial integro-differential operators and Noetherian commutative rings. 2 The largest left quotient ring of a ring In this section, existence of the largest left quotient ring of a ring is proved (Theorem 2.1). Proofs of Theorems 2.8 and 2.9 are given. The largest left quotient ring of a ring. Let R be a ring. A multiplicatively closed subset S of R (i.e. a multiplicative sub-semigroup of (R, ·) such that 1 ∈ S and 0 6∈ S) is said to be a left Ore set if it satisfies the left Ore condition: for each r ∈ R and s ∈ S, SrT Rs 6= ∅. Let S be a (non-empty) multiplicatively closed subset of R, and let ass(S) := {r ∈ R sr = 0 for some s ∈ S} (if, in addition, S is a left Ore set then ass(S) is an ideal of the ring R). Then a left quotient ring of R with respect to S (a left localization of R at S) is a ring Q together with a homomorphism ϕ : R → Q such that (i) for all s ∈ S, ϕ(s) is a unit of Q, (ii) for all q ∈ Q, q = ϕ(s)−1ϕ(r) for some r ∈ R, s ∈ S, and (iii) ker(ϕ) = ass(S). If exists the ring Q is unique up to isomorphism, usually it is denoted by S−1R. Recall that S−1R exists iff S is a left Ore set and the set S = {s + ass(S) ∈ R/ass(S) s ∈ S} consists of regular elements ([5], 2.1.12). If the last two conditions are satisfied then S is called a left denominator set. Similarly, a right Ore set, the right Ore condition, the right denominator set and the right quotient ring RS−1 are defined. If both S−1R and RS−1 exist then they are isomorphic ([5], 2.1.4.(ii)). The left quotient ring of R with respect to the set CR of all regular elements is 4 called the left quotient ring of R, if exists, it is denoted by Fracl(R) or Qcl(R). Similarly, the right quotient ring, Fracr(R) = Qr cl(R), is defined. If both left and right quotient rings of R exist then they are isomorphic and we write simply Frac(R) or Q(R) in this case. Theorem 2.1 1. For each a ∈ Assl(R), the set Denl(R, a) is an ordered abelian semigroup (S1S2 = S2S1, and S1 ⊆ S2 implies S1S3 ⊆ S2S3) where the product S1S2 = hS1, S2i is the multiplicative subsemigroup of (R, ·) generated by S1 and S2. 2. Sa := Sa(R) := SS∈Denl(R,a) S is the largest element (w.r.t. ⊆) in Denl(R, a). The set Sa is called the largest left denominator set associated to a. 3. Let Si ∈ Denl(R, a), i ∈ I, where I is an arbitrary non-empty set. Then hSi i ∈ Ii := [∅6=J⊆I,J<∞Yj∈J Sj ∈ Denl(R, a) (1) the left denominator set generated by the left denominators sets Si, it is the least upper bound of the set {Si}i∈I in Denl(R, a), i.e. hSi i ∈ Ii =Wi∈I Si. Remark. Clearly, S1S2 = S1W S2, the join of S1 and S2. Proof. 1. It remains to show that S1S2 ∈ Denl(R, a), that is, S1S2 is a left Ore set in R with ass(S1S2) = a and rs = 0 where r ∈ R and s ∈ S implies tr = 0 for some t ∈ S. To prove that the left Ore condition holds for the multiplicatively closed set S = S1S2 we have to show that, for each s ∈ S and a ∈ R, there exist elements t ∈ S and b ∈ R such that ta = bs. We use induction on the length l(s) = min{n s = s1 · · · sn where all si ∈ S1 ∪ S2} of the element s. If l(s) = 1, i.e. s ∈ S1 ∪ S2, then the statement is obvious since S1 and S2 are left Ore sets. Suppose that n := l(s) > 1, and the statement holds for n′ < n. Fix a presentation s = s1 · · · sn−1sn where all si ∈ S1 ∪ S2. Then t1a = b1sn for some elements t1 ∈ S1 ∪ S2 and b1 ∈ R, and, by induction, t2b1 = bs1 · · · sn−1 for some elements t2 ∈ S and b ∈ R. Clearly, t = t2t1 ∈ S and ta = t2(t1a) = (t2b1)sn = bs, as required. Let us show that ass(S) = a. Clearly, ass(S1S2) ⊇ a. Let r ∈ ass(S1S2). Then sr = 0 for some element s = s1 · · · sm ∈ S where all si ∈ S1 ∪ S2. Without loss of generality we may assume that sodd ∈ S1 and seven ∈ S2. Then s2 · · · smr ∈ a, and so there exists s′ 2 ∈ S2 such that (s′ 2s2 ∈ S2. By induction on m, we conclude that r ∈ a. Therefore, ass(S) = a. 2 s3 · · · smr = 0 where s′′ 2s2)s3 · · · smr = s′′ 2 = s′ Finally, suppose that rs = 0 where r ∈ R and s ∈ S. We have to show that tr = 0 for some t ∈ S. Fix a presentation s = s1 . . . sn where si ∈ S1 ∪ S2. Then 0 = rs = rs1 · · · sn implies that t1rs1 · · · sn−1 = 0 for some t1 ∈ S1 ∪ S2 since S1, S2 ∈ Denl(R, a). Similarly, t2t1rs1 · · · sn−2 = 0 for some t2 ∈ S1 ∪ S2. Repeating the same argument several times we see that tntn−1 · · · t1r = 0 for some ti ∈ S1 ∪ S2. Notice that t := tntn−1 · · · t1 ∈ S and tr = 0, as required. 2. Let s1, s2 ∈ Sa. Then s1 ∈ S1 and s2 ∈ S2 for some S1, S2 ∈ Denl(R, a), and so s1, s2 ∈ S1S2 ∈ Denl(R, a), by statement 1. Therefore, Sa ∈ Denl(R, a), and so Sa is the largest element in Denl(R, a). 3. Statement 3 follows from statement 1. (cid:3) Definition. For each ideal a ∈ Assl(R), the ring Qa(R) := S−1 quotient ring associated to a. When a = 0, the ring Ql(R) := Qa(R) := S−1 left quotient ring of R and S0 = S0(R) is called the largest left regular denominator set of R. a R is called the largest left 0 R is called the largest The next obvious corollary shows that Ql(R) is a generalization of the classical left quotient ring Qcl(R). Corollary 2.2 1. If the classical left quotient ring Qcl(R) := C−1 R R exists then the set of regular elements CR of the ring R is the largest left regular denominator set and Ql(R) = Qcl(R). 2. Let R1, . . . , Rn be rings. Then Ql(Qn i=1 Ri) ≃Qn i=1 Ql(Ri). 5 Proof. It is obvious. (cid:3) Question 2.3 Can a ring monomorphism f : A → B be lifted (necessarily uniquely) to a ring monomorphism f ′ : Ql(A) → Ql(B)? In general, the answer is no. Proposition 2.4 (Corollary 9.9, [2]) Let K be a field of characteristic zero and A1 = Khx, d dx i be the ring of polynomial differential operators (the first Weyl algebra). Then the inclusion A1 → I1 cannot be lifted neither to a ring homomorphism Frac(A1) = Ql(A1) → Ql(I1) nor to Frac(A1) = Qr(A1) → Qr(I1). Let (P, ≤) be a partially ordered set, a poset, for short (a ≤ a for all a ∈ P ; a ≤ b and b ≤ a implies a = b; a ≤ b and b ≤ c implies a ≤ c). For a subset S of P , an element x ∈ P is called an upper bound (resp. a lower bound) if s ≤ x (resp. s ≥ x) for all s ∈ S. The least upper bound for S is the least element in the the set of all upper bounds for S. Similarly, the greatest lower bound for S is defined. A lattice is a poset such that each pair x, y of elements of the set S has both the the meet of x and y). It follows then by induction that every non-empty finite set of elements has the join and the meet. A lattice L is complete if every subset S of L has the least upper bound least upper bound xW y (called the join for x and y) and the greatest lower bound xV y (called (the join of S) denoted sup(S) or Ws∈S s and the greatest lower bound (the meet of S) denoted inf(S) or Vs∈S s. In a complete lattice there exists the greatest element sup(L) denoted by 1, and the smallest element inf(L) denoted by 0. By definition, sup(∅) = 0. Let (L, ≤) be a lattice and a, b ∈ L with a ≤ b. The set [a, b] := {x ∈ L a ≤ x ≤ b} is the interval between a and b. The interval [a, b] is a sublattice of L with inf([a, b]) = a and sup([a, b]) = b. Corollary 2.5 The abelian monoid Denl(R, 0) is a complete lattice such that S1S2 = S1W S1 and Vi∈I Si =WS∈Denl(R,0),S⊆∩i∈I Si Si where all Si ∈ Denl(R, 0). Proof. The largest and least elements of the poset Denl(R, 0) are S0(R) and {1} respec- tively. Clearly, {1} is the identity element of the semigroup Denl(R, 0). By Theorem 2.1.(3), each nonempty subset of Denl(R, 0) has the least upper bound, hence Denl(R, 0) is a complete lattice by Proposition 1.2, Sect. 3, [6] and Vi∈I Si =WS∈Denl(R,0),S⊆∩i∈I Si Si. (cid:3) Clearly, Vi∈I Si is the largest element of the set {S S ∈ Denl(R, 0), S ⊆Ti∈I Si}. Corollary 2.6 1. Let R be a ring. Each ring automorphism σ ∈ Aut(R) of the ring R has the unique extension σ ∈ Aut(Ql(R)) to an automorphism of the ring Ql(R) given by the rule σ(s−1r) = σ(s)−1σ(r) where s ∈ S0(R) and r ∈ R. 2. The group Aut(R) is a subgroup of the group Aut(Ql(R)). Moreover, Aut(R) = {τ ∈ Aut(Ql(R)) τ (R) = R}. Proof. 1. By the uniqueness of the set S0(R), σ(S0(R)) = S0(R) for all elements σ ∈ Aut(R). Now, statement 1 follows from the universal property of the localization at S0(R). 2. Statement 2 follows from statement 1. (cid:3) For a ring R, let R∗ be its group of units and Inn(R) := {ωu u ∈ R∗} be the group of inner automorphisms of the ring R where ωu(r) = uru−1 is the inner automorphism determined by the element u. The next proposition is used in the proof of Theorem 2.8. Proposition 2.7 Let R be a ring, S ∈ Denl(R, 0), and T ∈ Denl(S−1R, 0); and so R ⊆ S−1R ⊆ T −1(S−1R) are natural inclusions of rings. Then 1. T −1(S−1R) = T −1 1 (S−1R) for some T1 ∈ Denl(S−1R, 0) such that S, S−1 ⊆ T1; in particu- lar, sT1s−1 = T1 for all s ∈ S. 6 2. If, in addition, S ⊆ T then T ∩ R ∈ Denl(R, 0) and S ⊆ T ∩ R. Denl(S−1R, 0). Proof. 1. For each s ∈ S, T −1(S−1R) = ωs(T −1(S−1R)) = ωs(T )−1(S−1R) and ωs(T ) ∈ It suffices to take T1 := hS, S−1, ωs(T ) s ∈ Si in Denl(S−1R, 0) since clearly S, S−1 ∈ Denl(S−1R, 0) and, for each non-empty finite subset J of S, (Qs∈J ωs(T ))−1(S−1R) = T −1(S−1R). In more detail, T −1 1 (S−1R) = [∅6=J⊆I,J<∞ (Ys∈J ωs(T ))−1(S−1R) = [∅6=J⊆I,J<∞ T −1(S−1R) = T −1(S−1R). 2. The set T ′ := T ∩ R is a multiplicatively closed subset of CR that contains the set S. It remains to show that the left Ore condition holds for T ′ in R: for each elements t′ ∈ T ′ and r ∈ R, T ′r ∩ Rt′ 6= ∅. Since T ∈ Denl(S−1R, 0), T r ∩ S−1Rt′ 6= ∅. Take an element, say u, from the intersection. The element u can be written in two different ways as follows s−1 1 t1 · r = s−1 2 r2t′ for some s−1 R ∩ T = T ′ (since S ⊆ T ), and s1s−1 1 t1 ∈ T and s−1 1 t1 ∈ 2 r2 ∈ S−1R where s1, s2 ∈ S and t1, r2 ∈ R. Clearly, t1 = s1 · s−1 3 r3 for some s3 ∈ S and r3 ∈ R. Then the element 1 t1r = s3s1s−1 2 = s−1 s3t1 · r = s3s1 · s−1 2 r2t′ = r3r2t′ belongs to the intersection T r ∩ Rt′ since s3t1 ∈ T (since S ⊆ T ) and r3r2 ∈ R. (cid:3) The group of units Ql(R)∗ of Ql(R). For a ring R and its largest left quotient ring Ql(R), Theorem 2.8 is used in the proof of Theorem 2.9 and gives an answers to the following natural questions: • What is S0(Ql(R))? • What is S0(Ql(R)) ∩ R? • What is the group Ql(R)∗ of units of the ring Ql(R)? • Is the natural inclusion Ql(R) ⊆ Ql(Ql(R)) an equality? Theorem 2.8 1. S0(Ql(R)) = Ql(R)∗ and S0(Ql(R)) ∩ R = S0(R). 2. Ql(R)∗ = hS0(R), S0(R)−1i, i.e. the group of units of the ring Ql(R) is generated by the sets S0(R) and S0(R)−1 := {s−1 s ∈ S0(R)}. 3. Ql(R)∗ = {s−1t s, t ∈ S0(R)}. 4. Ql(Ql(R)) = Ql(R). Proof. 1 -- 3. It is obvious that G := hS0(R), S0(R)−1i ⊆ Ql(R)∗ ⊆ S0(Ql(R)). Applying Proposition 2.7.(2) in the situation where S = S0(R) and T = S0(Ql(R)) we see that S0(R) ⊆ T ′ := S0(Ql(R)) ∩ R ∈ Denl(R, 0), and so S0(R) = T ′, by the maximality of S0(R). Let q ∈ S0(Ql(R)). Then q = s−1t for some elements s ∈ S0(R) and t = sq ∈ S0(Ql(R)) ∩ R = S0(R). Therefore, S0(Ql(R)) ⊆ {s−1t s, t ∈ S0(R)} ⊆ G, and so G = Ql(R)∗ = S0(Ql(R)) = {s−1t s, t ∈ S0(R)}. This proves statements 1 -- 3. 4. Statement 4 follows from statement 1. (cid:3) Necessary and sufficient conditions for Ql(R) to be a semi-simple ring. A ring Q is called a ring of quotients if every element c ∈ CQ is invertible. A subring R of a ring of quotients Q is called a left order in Q if CR is a left Ore set and C−1 R R = Q. A ring R has finite left rank (i.e. finite left uniform dimension) if there are no infinite direct sums of nonzero left ideals in R. The next theorem gives an answer to the question of when Ql(R) is a semi-simple ring. 7 Theorem 2.9 The following properties of a ring R are equivalent. 1. Ql(R) is a semi-simple ring. 2. Qcl(R) exists and is a semi-simple ring. 3. R is a left order in a semi-simple ring. 4. R has finite left rank, satisfies the ascending chain condition on left annihilators and is a semi-prime ring. 5. A left ideal of R is essential iff it contains a regular element. If one of the equivalent conditions hold then S0(R) = CR and Ql(R) = Qcl(R). R R is a semi-simple ring. Proof. Goldie's Theorem states that 2 ⇔ 3 ⇔ 4 ⇔ 5. (3 ⇒ 1) If statement 3 holds then Ql(R) = C−1 (1 ⇒ 3) We have to show that CR is a left Ore set since then CR = S0(R) and Ql(R) = Qcl(R) is a semi-simple ring. Notice that S0(R) ⊆ CR. Let q ∈ CR. Then the Ql(R)-module monomorphism ·q : Ql(R) → Ql(R), x 7→ xq, is an isomorphism, and its inverse is necessarily of the form ·p for some element p ∈ Ql(R). Clearly, pq = 1 and qp = 1, i.e. q ∈ Ql(R)∗. By Theorem 2.8.(3), q = s−1t for some elements s, t ∈ S0(R). We have to show that, for each q ∈ CR and r ∈ R, there exists an element c ∈ CR such that crq−1 = crt−1s ∈ R. Since t ∈ S0(R), there exists an element s1 ∈ S0(R) such that s1rt−1 ∈ R. It suffices to take c = s1 since S0(R) ⊆ CR. (cid:3) The next corollary gives an interesting criterion of when the classical quotient ring Qcl(R) = C−1 R R exists. Corollary 2.10 If the ring Ql(R) is a left artinian ring then S0(R) = CR and Qcl(R) = Ql(R). Proof. Each left artinian ring is left Noetherian, hence the left Ql(R)-module Ql(R) has finite length. Now, we can repeat the proof of the implication 1 ⇒ 3 of Theorem 2.9, where, in fact, we only used the fact that the left Ql(R)-module Ql(R) has finite length. (cid:3) Proposition 2.11 (Proposition 11.6, [6]; [4]) Let A be a subring of a ring B. If M is a finitely generated flat A-module such that B ⊗A M is a projective B-module then M is a projective A- module. Corollary 2.12 If there exists a finitely generated flat R-module M which is not projective the the ring Ql(R) is not a semi-simple ring. Proof. If Ql(R) were a semi-simple ring then Ql(R)⊗R M would be a projective Ql(R)-module, and so M would be a projective R-module, by Proposition 2.11, a contradiction. (cid:3) 3 The maximal left quotient rings of a ring In this section, a new class of rings, the class of left localization maximal rings, is introduced. It is proved that, for an arbitrary ring R, the set of maximal elements of the poset (Denl(R), ⊆) is a non-empty set (Lemma 3.7.(2)), and therefore the set of maximal left quotient rings of the ring R is a non-empty set. A criterion is given (Theorem 3.11) for a left quotient ring of a ring to be a maximal left quotient ring of the ring. Many results on denominator sets are proved. In particular, for each denominator set S ∈ Denl(R, a), connections are established between the left denominator sets Denl(R, a), Denl(R/a, 0) and Denl(S−1R, 0). Proposition 3.1 1. For each ring A ∈ Locl(R, a) where a ∈ Assl(R), the set Denl(R, a, A) := {S ∈ Denl(R, a) S−1R = A} is an ordered submonoid of Denl(R, a), and 8 2. S(R, a, A) :=SS∈Denl(R,a,A) S is its largest element. In particular, S0(R) = S(R, 0, Ql(R)). 3. Let Si ∈ Denl(R, a, A), i ∈ I, where I is an arbitrary non-empty set. Then hSi i ∈ Ii ∈ Denl(R, a, A) (see (13)). Moreover, hSi i ∈ Ii is the least upper bound of the set {Si}i∈I in Denl(R, a, A) and in Denl(R, a). Proof. 1. In view of Theorem 2.1, it suffices to show that if S1, S2 ∈ Denl(R, a, A) then S1S2 ∈ Denl(R, a, A), i.e. (S1S2)−1R = A. By Theorem 2.1, S1S2 ∈ Denl(R, a). Notice that A = S−1 2 R. For each s = s1 · · · sn ∈ S1S2 where all si ∈ S1 ∪ S2, and, for each r ∈ R, 1 R = S−1 s−1r = s−1 n · · · s−1 1 r ∈ s−1 n · · · (s−1 2 A) ⊆ s−1 n · · · (s−1 3 A) ⊆ · · · ⊆ A. Therefore, (S1S2)−1R = A. 2. By Theorem 2.1, S(R, a, A) = hS S ∈ Denl(R, a, A)i ∈ Denl(R, a); and S(R, a, A)−1R = inj lim S−1R = inj lim A = A where the injective limit is over S ∈ Denl(R, a, A). 3. Repeat the proof of statement 2 replacing Denl(R, a, A) by I. (cid:3) Lemma 3.2 1. Let S ∈ Denl(R, a), b be an ideal of the ring R such b ⊆ a, and π : R → R/b, a 7→ a = a + b. Then π(S) ∈ Denl(R/b, a/b) and S−1R ≃ π(S)−1(R/b). 2. Let S1, S2 ∈ Denl(R) and S1 ⊆ S2. Then (a) a1 := ass(S1) ⊆ a2 := ass(S2); there is the R-ring homomorphism ϕ : S−1 1 R → S−1 2 R, s−1a 7→ s−1a; and ker(ϕ) = S−1 1 (a2/a1). (b) Let π1 : R → R/a1, a 7→ a = a + a1, and eS2 be the multiplicative submonoid of 1 (R/a1), ·) generated by π1(S2) and π1(S1)−1 = {s−1 s ∈ S1}. Then π1(S2), eS2 ∈ 1 R) ≃ π1(S2)−1(S−1 (S−1 Denl(S−1 1 R) ≃ S−1 1 R, S−1 2 (S−1 2 R. 1 (a2/a1)) and eS−1 Proof. 1. Since π is an epimorphism and b ⊆ a, π(S) is a left Ore set in R/a. Clearly, a/b ⊆ ass(π(S)). To prove that the inverse inclusion holds, let a ∈ ass(π(S)). Then 0 = sa for some element s ∈ π(S). Then b := sa ∈ b. Since b ⊆ a, we can find an element t ∈ S such that tb = 0. Then (ts)a = 0, and so a ∈ a. Therefore, ass(π(S)) = a/b. To prove that π(S) ∈ Denl(R/b), it remains us to show that as = 0, for some a ∈ R/b and s ∈ π(S), implies that ta = 0 for some t ∈ S. Clearly, b := as ∈ b ⊆ a, and so 0 = s1b = (s1a)s for some s1 ∈ S. It follows that s1a ∈ a since S ∈ Denl(R, a). We can find an element s2 ∈ S such that s2s1a = 0. Therefore, π(S) ∈ Denl(R/b, a/b). It suffices to take t = s2s1 ∈ S. By the universal property of the ring S−1R, there is the ring epimorphism S−1R → π(S)−1(R/b), s−1r 7→ π(s)−1π(r). By the universal property of the ring π(S)−1(R/b), there is the ring epimorphism π(S)−1(R/b) → S−1R, π(s)−1π(r) 7→ s−1r. Therefore, S−1R ≃ π(S)−1(R/b). 2a. The inclusion a1 ⊆ a2 is obvious. The image S1 of the Ore set S1 under the ring epi- morphism π : R → R/a2 is an Ore set in R/a2 since S1 ⊆ S2 and S2 ∈ Denl(R, a2). Therefore, each element of the subring of S−1 2 R generated by π(R) and S1 has the form π(s)−1π(r) for some s ∈ S1 and r ∈ R. By the universal property of the ring S−1 1 R, the map ϕ exists. An element s−1r ∈ S−1 2 R iff tr = 0 for some t ∈ S2 iff r ∈ a2. Therefore, ker(ϕ) = S−1 1 R, where s ∈ S1 and r ∈ R, belongs to the kernel of ϕ iff s−1r = 0 in S−1 1 (a2/a1). 2b. First, we prove that the set π1(S2) (resp. eS2) is a left Ore set in S−1 prove that for each element s2 ∈ π1(S2) (resp. s2 ∈ eS2) and s−1 r ∈ R, there exist elements t ∈ π1(S2) (resp. t ∈ eS2) and a ∈ S−1 1 R, i.e. we have to 1 R where s1 ∈ S1 and 1 r = as2. In the can be written as s−1 1 R, 1 (a2/a1), by statement 2(a). Therefore, there exists an element 3 r1 for some s3 ∈ S2 and r1 ∈ R. In S−1 1 R such that ts−1 1 r ∈ S−1 2 R, the element s−1 1 r − r1s2 ∈ ker(ϕ) = S−1 ring S−1 b := s3s−1 s4 ∈ S2 such that s4b = 0, and so s4s3 · s−1 and a = s4r. 1 rs−1 2 1 r = s4r · s2. It suffices to take t = s4s3 ∈ π1(S2) ⊆ eS2 9 It is obvious that S−1 ϕ(s) is a unit. Therefore, S−1 s ∈ eS2. Then 0 = ϕ(su) = ϕ(s)ϕ(u), and so u ∈ ker(ϕ) = S−1 1 (a2/a1) = ass(π1(S2)) = ass(eS2). 1 (a2/a1) ⊆ ass(π1(S2)) ⊆ ass(eS2). If u ∈ ass(eS2) then su = 0 for some 1 R and s ∈ π1(S2) (resp. s ∈ eS2). Then 0 = ϕ(v)ϕ(s), and so 1 (a2/a1) since ϕ(s) is a unit. This proves that π1(S2) ∈ Denl(S−1 1 (a2/a1) (by statement 2(a)) since If vs = 0 for some elements v ∈ S−1 ϕ(v) ∈ ker(ϕ) = S−1 1 R, S−1(a2/a1))). Now, it is obvious that π1(S2)−1(S−1 1 R, S−1(a2/a1)) 2 R ≃ 1 R) ≃ S−1 (resp. eS2 ∈ Denl(S−1 eS−1 2 (S−1 1 R). (cid:3) The set (Locl(R, a), →) is a poset where A1 → A2 if A1 = S−1 2 R for some denominator sets S1, S2 ∈ Denl(R, a) with S1 ⊆ S2, and A1 → A2 is the inclusion map in Lemma 3.2.(2a). If (S′ 1 R = (S1S′ 2) is another such a pair then, by Proposition 3.1.(1), A1 = S−1 1 R and A2 = S−1 1 R = S′−1 1 ⊆ S2S′ 2. 2 ∈ Denl(R, a) with S1S′ 1)−1R → A2 = S−1 2 R = (S2S′ 2 R = S′−1 1, S′ 2)−1R; S1S′ 1, S2S′ In the same way, the poset (Locl(R), →) is defined, i.e. A1 → A2 if there exists S1, S2 ∈ 2 R, A1 → A2 stands for the map ϕ : 1 R and A2 = S−1 Denl(R) such that S1 ⊆ S2, A1 = S−1 S−1 1 R → S−1 2 R (Lemma 3.2.(2a)). The map (·)−1R : Denl(R) → Locl(R), S 7→ S−1R, (2) is an epimorphism of the posets (Denl(R), ⊆) and (Locl(R), →). For each ideal a ∈ Assl(R), it induces the epimorphism of posets (Denl(R, a), ⊆) and (Locl(R, a), →), (·)−1R : Denl(R, a) → Locl(R, a), S 7→ S−1R. The sets Denl(R) and Locl(R) are the disjoint unions Denl(R) = Ga∈Assl(R) Denl(R, a), Locl(R) = Ga∈Assl(R) Locl(R, a). For each ideal a ∈ Assl(R), the set Denl(R, a) is the disjoint union Denl(R, a)) = GA∈Locl(R,a)) Denl(R, a, A). Let LDenl(R, a) := {S(R, a, A) A ∈ Locl(R, a)}, see Proposition 3.1.(2). The map (·)−1R : LDenl(R, a) → Locl(R, a), S 7→ S−1R, is an isomorphism of posets. For a left denominator set S ∈ Denl(R, a), there are natural ring homomorphisms R → R/a → S−1R. (3) (4) (5) (6) Lemma 3.3 and Proposition 3.4 establish connections between the left denominator sets Denl(R, a), Denl(R/a, 0) and Denl(S−1R, 0). Let S, T ∈ Denl(R). The denominator set T is called S-saturated if sr ∈ T , for some s ∈ S and r ∈ R, then r ∈ T , and if r′s′ ∈ T , for some s′ ∈ S and r′ ∈ R, then r′ ∈ T . Lemma 3.3 Let S ∈ Denl(R, a), π : R → R/a, a 7→ a + a, and σ : R → S−1R, r 7→ r/1. 1. Let T ∈ Denl(S−1R, 0) be such that π(S), π(S−1) ⊆ T . Then T ′ := σ−1(T ) ∈ Denl(R, a), T ′ is S-saturated, T = {s−1t′ s ∈ S, t′ ∈ T ′}, and S−1R ⊆ T ′−1R = T −1R. 2. π−1(S0(R/a)) = Sa(R), π(Sa(R)) = S0(R/a)) and Qa(R) = Sa(R)−1R = Ql(R/a). 10 Proof. 1. In the proof below we often identify the elements σ(s), s/1 and s in order to avoid cumbersome notation. By the very definition, T ′ is a multiplicatively closed set that contains S. To prove that T ′ is a left Ore set in R we have to show that, for any t′ ∈ T ′ and r ∈ R, there 1r = at′. Since T ∈ Denl(S−1R, 0) and σ(t′) ∈ T , exist elements t′ tσ(r) = s−1 2 r2 for some elements s2 ∈ S and r2 ∈ R. Clearly, r2 ∈ T ′ since π(S) ⊆ T and r2/1 = s2t ∈ T . Fix an element s3 ∈ S such that r3/1 := s3s1s−1 2 ∈ π(R). Then r3 ∈ T ′ since π(S)±1 ⊆ T . Multiplying the equality 1 r1σ(t′) for some elements t ∈ T , s1 ∈ S and r1 ∈ R. The element t is equal to s−1 1 ∈ T ′ and a ∈ R such that t′ s1s−1 2 r2σ(r) = r1σ(t′) by the element s3 on the left we obtain the equality r3r2σ(r) = s3r1σ(t′) ∈ π(R). Hence s4r3r2 ·r = s4s3r1 · t′ for some element s4 ∈ S. Now, take t′ 1 := s4r3r2 ∈ T ′ and a := s4s3r1 ∈ R. Since S ⊆ T ′, a ⊆ ass(T ′). If θu = 0 and vη = 0 for some elements u, v ∈ R and θ, η ∈ T ′ then 0 = σ(θ)σ(u) and 0 = σ(v)σ(η) and so 0 = σ(u) = σ(v) since the elements σ(θ) and σ(η) are units. Then, u, v ∈ a. This proves that T ′ ∈ Denl(R, a). The denominator set T ′ is S-saturated since sr = t′ ∈ T ′ (resp. rs = t′ ∈ T ′) for some s ∈ S and r ∈ R implies r/1 = s−1t′ ∈ T (resp. r/1 = t′s−1 ∈ T ), and so r ∈ T ′. Since π(S) ⊆ T , T = {s−1t′ s ∈ S, t′ ∈ T ′}. Then the inclusion and the equality, S−1R ⊆ T ′−1R = T −1R, are obvious. 2. Let S′ := π−1(S0(R/a)). Since the map π is surjective we have π(S′) = S0(R/a). We aim to show that S′ = Sa(R). Step 1: S′ ∈ Orel(R). We have to show that for any s ∈ S′ and r ∈ R there exist elements s1 ∈ S′ and r1 ∈ R such that s1r = r1s. Since π(S′) = S0(R/a) ∈ Orel(R/a), we have π(s′)π(r) = π(r′)π(s) for some elements s′ ∈ S′ and r′ ∈ R. Then π(s′r − r′s) = 0, and so s′r − r′s ∈ a, hence s′′(s′r − r′s) = 0 for some s′′ ∈ S. Note that π(S) ∈ Denl(R/a, 0) hence π(S) ⊆ S0(R/a), and so S ⊆ S′. Now, s1r = r1s where s1 = s′′s′ ∈ S′ and r1 = s′′r′ ∈ R. Therefore, S′ ∈ Orel(R). Step 2: ass(S′) = a. Since S ⊆ S′, we have the inclusion a ⊆ ass(S′). Let r ∈ ass(S′). Then s′r = 0 for some s′ ∈ S′, and so π(s′)π(r) = 0. It follows that π(r) = 0 since π(s′) ∈ S0(R/a), i.e. r ∈ ker(π) = a. This means that ass(S′) = a. Step 3: S′ ∈ Denl(R, a). In view of Steps 1 and 2, we have to show that the equality rs′ = 0 for some elements r ∈ R and s′ ∈ S′ implies that s′′r = 0 for some s′′ ∈ S′. Since π(r)π(s′) = 0 and π(s′) ∈ S0(R/a) = π(S′) we have the equality π(r) = 0 and so r ∈ a. Then s′′r = 0 for some element s′′ ∈ S ⊆ S′. Step 4: S′ = Sa(R). By Step 3, S′−1R = π(S′)−1R/a = S0(R/a)−1R/a = Ql(R/a). No- tice that Sa(R)−1R = π(Sa(R))−1R/a ⊆ Ql(R/a). Since π(Sa(R)) ∈ Denl(R/a, 0), we have π(Sa(R)) ⊆ S0(R/a), hence Sa(R) ⊆ π−1(S0(R/a)) = S′ ⊆ Sa(R), i.e. S′ = Sa(R). (cid:3) For S1, S2 ∈ Denl(R) such that S1 ⊆ S2, [S1, S2] := {T ∈ Denl(R) S1 ⊆ T ⊆ S2} is an interval in the posed Denl(R). If, in addition, S1, S2 ∈ Denl(R, a) then [S1, S2] ⊆ Denl(R, a) since S1 ⊆ S ⊆ S2 implies a = ass(S1) ⊆ ass(S) ⊆ ass(S2) = a, i.e. ass(S) = a. Proposition 3.4 Let S ∈ Denl(R, a); π : R → R/a, a → a = a + a; σ : R → S−1R, r → r/1; G := hπ(S), π(S)−1i ⊆ (S−1R)∗ (i.e. G is the subgroup of the group (S−1R)∗ of units of the ring S−1R generated by π(S)±1). 1. Let [σ−1(G), Sa(R)]S−com := {S1 ∈ [σ−1(G), Sa(R)] σ−1(Gπ(S1)) = S1} and [G, S0(S−1R)] := {T ∈ Denl(S−1R, 0) G ⊆ T ⊆ S0(S−1R)}. Then the map [σ−1(G), Sa(R)]S−com → [G, S0(S−1R)], S1 7→ eS1 := Gπ(S1), is an isomorphism of posets and abelian monoids with the inverse map T 7→ σ−1(T ) where Gπ(S1) is the multiplicative monoid generated by G and π(S1) in S−1R. In particular, Gπ(Sa(R)) = S0(S−1R), Sa(R) = σ−1(S0(S−1R)), Sa(R)−1R = Ql(R/a), 11 the monoid [σ−1(G), Sa(R)]S−com is a complete lattice (since [G, S0(S−1R)] is a complete lattice, as an interval of the complete lattice Denl(S−1R, 0), Corollary 2.5), and the map S1 7→ eS1 is a lattice isomorphism. 2. Consider the interval [G∩(R/a), S0(R/a)] in Denl(R/a, 0). Let [G∩(R/a), S0(R/a)]G−com := {T ∈ [G∩(R/a), S0(R/a)] GT ∩(R/a) = T }. Then [G∩(R/a), S0(R/a)]G−com ⊆ Denl(S−1R, 0) and the map [G ∩ (R/a), S0(R/a)]G−com → [G, S0(S−1R)], T 7→ GT, is an isomorphism of posets and abelian monoids with the inverse map T ′ 7→ T ′∩(R/a) where GT is the product in Denl(S−1R, 0). In particular, the monoid [G ∩ (R/a), S0(R/a)]G−com is a complete lattice. 3. The map [σ−1(G), Sa(R)]S−com → [G ∩ (R/a), S0(R/a)]G−com, S1 7→ Gπ(S1) ∩ (R/a), is an isomorphism of posets and abelian monoids with the inverse map S′ 7→ π−1(S′). Proof. 1. The equality ^σ−1(G) = G is obvious. Then, by Lemma 3.2.(2).(b), the map φ : S1 7→ eS1 is well-defined. By Lemma 3.3.(1), the map ψ : T 7→ σ−1(T ) is well-defined and ^σ−1(T ) = T (i.e. φψ = 1) since T = {s−1t s ∈ S, t ∈ σ−1(T )} (Lemma 3.3.(1)) and G ⊆ T . By the very definition of the set [S, Sa(R)]S−com, ψφ = 1, i.e. ψ = φ−1. For all elements S1, S2 ∈ [S, Sa(R)]S−com, ^S1^ S2 = ]S1S2 = Gπ(S1)π(S2) = Gπ(S1)Gπ(S2) = eS1eS2 = eS1^ eS2, and so the map φ is an isomorphism of posets and abelian monoids. Therefore, Gπ(Sa(R)) = S0(S−1R) and Sa(R) = σ−1(S0(S−1R)). Then, Sa(R)−1R = Ql(R/a), by Lemma 3.3.(1). The rest is obvious. 2. Let φ : T 7→ GT and ψ : T ′ 7→ T ′ ∩ R where R := R/a. Step 1: φ is well-defined. Since G ∈ Denl(S−1R, 0) and φ(T ) = GT (the product in Denl(S−1R, 0)), we have to show that T ∈ Denl(S−1R, 0). First, let us show that T ∈ Orel(S−1R), i.e. for each s−1r ∈ S−1R (where s ∈ S and r ∈ R) and t ∈ T we have to show that t1s−1r = at for some elements t1 ∈ T and a ∈ S−1R. Since T ∈ Denl(R, 0), t′r = at for some t′ ∈ T and a ∈ R. So, it suffices to take t1 = t′s (t1 ∈ T since π(S) ⊆ T ). Let us show that ass(T ) = 0 in S−1R. Suppose that t · s−1r = 0 for some t ∈ T , s ∈ S and r ∈ R, we have to show that s−1r = 0. There exist elements s1 ∈ S and t2 ∈ R such that s1t = t2s, hence t2 = s1ts−1 ∈ R ∩ GT = T . Then 0 = s1 · 0 = s1ts−1r = t2r, hence r = 0 (since T ∈ Denl(R, 0)), and so s−1r = 0, as required. Therefore, ass(T ) = 0 in S−1R. To finish the proof of Step 1, we have to prove that s−1rt = 0 for some s ∈ S, r ∈ R and t ∈ T implies s−1r = 0. This is obvious since s−1rt = 0 implies rt = 0 in R, hence r = 0 since T ∈ Denl(R, 0), and so s−1r = 0. Therefore, T ∈ Denl(S−1R, 0) for all T ∈ [G ∩ (R/a), S0(R/a)]G−com. In particular, [G ∩ (R/a), S0(R/a)]G−com ⊆ Denl(S−1R, 0). Step 2: ψ is well-defined. Let T ′ ∈ Denl(S−1R, 0). We have to show that T := R ∩ T ′ ∈ Denl(R, 0). It is obvious that assR(T ) = 0 since assR(T ) ⊆ assS−1R(T ′) = 0. If rt = 0 for some elements r ∈ R and t ∈ T then r = 0 since T ⊆ T ′, T ′ ∈ Denl(S−1R, 0) and R ⊆ S−1R. It remains to show that T ∈ Orel(R), that is, for any r ∈ R and t ∈ T , t1r = r1t for some elements r1 ∈ R and t1 ∈ T . Since T ⊆ T ′ and T ′ ∈ Denl(S−1R, 0), t′r = s−1 1t for some elements t′ ∈ T ′, s1 ∈ S and r′ 1 ∈ R. Fix an element s2 ∈ S such that t1 := s2s1t′ ∈ R. Then t1 ∈ R ∩ T ′ = T and t1r = s2s1t′r = s2s1s−1 1 · t = r1t where r1 := s2r′ ∈ R, as required. 1 r′ 1 r′ 1t = s2r′ By the very definition of the set [G ∩ R, S0(R)], ψφ = 1. To finish the proof of statement 2 it remains to show that φψ = 1, i.e. G(R ∩T ′) = T ′ for all T ′ ∈ [G, S0(S−1R)]. Since G, R ∩T ′ ⊆ T ′, the inclusion T ′′ := G(R ∩ T ′) ⊆ T ′ is obvious. The reverse inclusion follows from the fact that 12 any element t′ ∈ T ′ can be written as s−1t for some elements s ∈ S ⊆ G and t ∈ R, hence t = st′ ∈ R ∩ T ′ = T . 3. Statement 3 follows from statements 1 and 2. (cid:3) The elements of the set [S, Sa(R)]S−com are called S-complete and the elements of the set [G ∩ (R/a), S0(R/a)]G−com are called G-complete. Lemma 3.5 We keep the notation of Proposition 3.4.(1). If S1 ∈ [σ−1(G), Sa(R)]S−com then S1 is S-saturated. Proof. Notice that σ−1(G) ⊆ S1 ⊆ Sa(R) and S1 = σ−1(Gπ(S1)). If s1 := sr ∈ S1 (resp. s1 := rs ∈ S1) for some elements s ∈ S and r ∈ R then r/1 = s−1s1 ∈ Gπ(S1) (resp. r/1 = s1s−1 ∈ Gπ(S1)) hence r ∈ σ−1(Gπ(S1)) = S1, i.e. S1 is S-saturated. (cid:3) The maximal left quotient rings of a ring. Lemma 3.6 Let S1 ⊆ S2 ⊆ · · · ⊆ Si ⊆ · · · be an ascending chain in Denl(R), ai := ass(Si), S := Si≥1 Si. Then a1 ⊆ a2 ⊆ · · · ⊆ ai ⊆ · · · is the ascending chain in Assl(R), S ∈ Denl(R, a) where a := Si≥1 ai, S−1R = inj lim S−1 2 R → · · · → S−1 i R → · · · (Lemma 3.2.(2a)). i R where S−1 1 R → S−1 Proof. By Lemma 3.2.(2a), S ∈ Denl(R, a). For each number i = 1, 2, . . ., define the ring i+1R, s−1r 7→ s−1r. Then i R → S−1R which is a homomorphism φi : S−1 φi = φi+1νi for all i. Hence, there is the ring homomorphism φ : inj lim S−1 i R → S−1R, s−1r 7→ s−1r, and νi : S−1 surjection since S =Si≥1 Si and has kernel 0 since a =Si≥1 ai, i.e. φ is an isomorphism. (cid:3) i R → S−1 Consider the poset (Denl(R), ⊆). For each element S ∈ Denl(R), let [S, ·) := {S′ ∈ Denl(R) S ⊆ S′}. Lemma 3.7 1. For each element S ∈ Denl(R), there exists a maximal element in the poset ([S, ·), ⊆). 2. The set (max.Denl(R), ⊆) of maximal elements of the poset (Denl(R), ⊆) is a non-empty set. Proof. 1. Statement 1 follows from Lemma 3.6 and Zorn's Lemma. 2. Statement 2 follows from statement 1 and the fact that the set max.Denl(R) is the set of maximal elements of the poset [{1}, ·) = Denl(R). (cid:3) Definition. An element S of the set max.Denl(R) is called a maximal left denominator set of R and the ring S−1R is called a maximal left quotient ring of R. The intersection l.lrad(R) := \S∈max.Denl(R) ass(S) is called the left localization radical of the ring R. Proposition 3.8 Let a ∈ Assl(R), Q := Qa(R), Q∗ be the group of units of the ring Q and 1 . Let T ∈ Denl(Q, b) where b ∈ Assl(Q) and Q∗T be the multiplicative σ : R → Qa(R), r 7→ r sub-semigroup of (Q, ·) generated by Q∗ and T . Then 1. Q∗T ∈ Denl(Q, b). 2. If, in addition, Q∗ ⊆ T (eg, Q∗T from statement 1) then 13 (a) T ′ := σ−1(T ) ∈ Denl(R, b′) where b′ := σ−1(b) ⊇ a, Sa(R) ⊆ T ′, T = Q∗σ(T ′) (i.e. the monoid T is generated by Q∗ and σ(T ′)) and T ′−1R = T −1Q (i.e. the natural ring monomorphism T ′−1R → T −1Q, t−1r 7→ t−1r, is an isomorphism). (b) Sa(R) ⊆ Sb′(R) and Sb′ (R) = σ−1(Sb(Q)). (c) Qb′(R) = Ql(Q/b), i.e. the natural ring monomorphism Qb′(R) → Ql(Q/b), t−1r 7→ t−1r, is an isomorphism. Proof. 1. Clearly, T := Q∗T is a multiplicative monoid. We have to show that T ∈ Denl(Q, b). An element s of T is a product s = q1t1q2t2 · · · qntn for some elements qi ∈ Q and ti ∈ T . Step 1: T ∈ Orel(R). For any element s ∈ T and a ∈ Q, we have to find elements s1 ∈ T and a1 ∈ Q such that s1a = a1s. We use induction on n. When n = 1, i.e. s = q1t1, then s1a = a′t1 for some elements s1 ∈ T and a′ ∈ Q since T ∈ Denl(Q, b). It suffices to take a1 = a′q−1 since s1a = (a′q−1 1 ) · q1t1 = a1q1t1. Suppose that n > 1 and the statement is true for all n′ < n. Then sna = anqntn for some elements sn ∈ T and an ∈ Q. By induction, s′ nan = a1q1t1 · · · qn−1tn−1 for some elements s′ n ∈ T and a1 ∈ Q. It suffices to take s1 = s′ nsn since then 1 s1a = s′ nsna = s′ nanqntn = a1q1t1 · · · qn−1tn−1qntn = a1s. Step 2: ass(T ) = b. We have to show that sa = 0 for some a ∈ Q implies a ∈ b. We use induction on n. When n = 1, i.e. s = q1t1 then q1t1a = 0 implies t1a = 0 (since q1 is a unit), and so a ∈ b since T ∈ Denl(Q, b). Let n > 1. Suppose that the result is true for all n′ < n. Then 0 = sa = q1t1 · · · qn−1tn−1 · (qntna) implies (by induction) qntna ∈ b, hence tna ∈ b since qn is a unit. Then tn+1tna = 0 for some element tn+1 ∈ T since T ∈ Denl(Q, b), and finally a ∈ b since tn+1tn ∈ T . Step 3: T ∈ Denl(Q, b). We have to show that as = 0 for some a ∈ Q implies a ∈ b. We use induction on n. When n = 1, i.e. aq1t1 = 0, then aq1 ∈ b since T ∈ Denl(Q, b), and so a ∈ b since q1 is a unit. Let n > 1. Suppose that the result is true for all n′ < n. Then 0 = as = (aq1t1) · (q2t2 · · · qntn) implies (by induction) a1q1t1 ∈ b, hence tn+1aq1 ∈ b for some element tn+1 ∈ T (since T ∈ Denl(Q, b)), i.e. tn+1a ∈ b since q1 is a unit. Therefore, there exists an element tn+2 ∈ T such that tn+2tn+1a = 0. This means that a ∈ b since tn+2tn+1 ∈ T ∈ Denl(Q, b). 2a. By the very definition, T ′ is a monoid and a ⊆ b′. By Lemma 3.3.(2), Qa(R) = Q0(R/a), σ−1(S0(R/a)) = Sa(R) and σ(Sa(R)) = S0(R/a). Therefore, Sa(R) ⊆ T ′ since σ(Sa(R)) = S0(R/a) ⊆ Q∗ ⊆ T . Since Q∗ ⊆ T , we have the equality T = Q∗σ(T ′). Let us show that T ′ ∈ Denl(R, b′). 1 ∈ T ′ and r1 ∈ R such that t′ Step 1: T ′ ∈ Orel(R). We have to show that for any elements r ∈ R and t′ ∈ T ′ there exists elements t′ 1r = r1t′. Since T ∈ Denl(Q, b) and σ(t′) ∈ T , we have the equality tσ(r) = s−1r2σ(t′) in the ring Q for some elements t ∈ T , s ∈ S0(R/a) and r2 ∈ R. Since S0(R/a) ⊆ Q∗ and Q∗ ⊆ T (by the assumption), the product st ∈ T . Fix an element s1 ∈ S0(R/a) ⊆ Q∗ ⊆ T such that s1st = σ(t′ 2 ∈ T ′. Then s1 = σ(s′ 1 ∈ Sa(R) since σ(Sa(R)) = S0(R/a). By multiplying the equality tσ(r) = s−1r2σ(t′) by the element s1s on the left, we obtain the equality 1) for some element s′ 2) for some element t′ 2 ∈ R, necessarily t′ σ(t′ 2r − s′ 1r2t′) = 0, i.e. α := t′ s3t′ 2r − s′ 1r2t′ ∈ a. There exists an element s3 ∈ Sa(R) ⊆ T ′ such that s3α = 0, i.e. 2 · r = s3s′ Step 2: ass(T ′) = b′. Let r ∈ ass(T ′), i.e. t′r = 0 for some t′ ∈ T ′. Then σ(t′)σ(r) = 0 in Q, 1r2 · t′. Now, it suffices to take t′ 2 ∈ T ′ and r1 := s3s′ 1 := s3t′ 1r2 ∈ R. and so σ(r) ∈ b since σ(t′) ∈ T . Hence r ∈ b′. This means that ass(T ′) = b′. Step 3: T ′ ∈ Denl(R, b′). We have to show that if rt′ = 0 for some r ∈ R and t′ ∈ T ′ then r ∈ b′. Since σ(r)σ(t′) = 0 and σ(t′) ∈ T , we have the inclusion σ(r) ∈ b, hence r ∈ b′. Since σ(T ′) ⊆ T and ass(T ′) = b′, there is the (natural) ring monomorphism T ′−1R → T −1R, t′−1r 7→ t′−1r, 14 where t′ ∈ T ′ and r ∈ R. The monomorphism is, in fact, an isomorphism since T = Q∗σ(T ′), Sa(R) ⊆ T ′ and Q∗ = {σ(s)−1σ(t) where s, t ∈ Sa(R)} (by Lemma 3.3.(2) and Theorem 2.8(1- 3)). 2b. Let T and T ′ be as in statement 2a (it exists, by statement 1). By statement 2a, Sa(R) ⊆ T ′ and T ′ ∈ Denl(R, b′). Clearly, T ′ ⊆ Sb′(R) since Sb′(R) is the largest element of the poset (Denl(R, b′), ⊆). Therefore, Sa(R) ⊆ Sb′(R). Let T = Sb(Q). Then Q∗ ⊆ T by statement 1 and the maximality of Sb(Q). By statement 2a, T ′ := σ−1(Sb(Q)) ⊆ Sb′(R). Consider the ring epimorphism πa : R → R/a, a 7→ a + a. Notice that a ⊆ b′. Applying Lemma 3.2.(2) in the situation Sa(R) ⊆ Sb′(R), we see that πa(Sb′ (R)) ∈ Denl(Q, b). Therefore, Sb′(R) ⊆ T ′. This proves that Sb′ (R) = T ′. 2c. Applying statement 2a in the situation T = Sb(Q) ⊇ Q∗ and using the fact that Sb′(R) := σ−1(Sb(Q)) = σ−1(T ), we see that Sb′(R)−1R = Sb(Q)−1Q, i.e. Qb′(R) = Ql(Q/b) since Ql(Q/b) = Sb(Q)−1Q (by Lemma 3.3.(2)). (cid:3) Let max.Assl(R) be the set of maximal elements of the poset (Assl(R), ⊆). It is a subset of the set ass.max.Denl(R) := {ass(S) S ∈ max.Denl(R)} (7) which is a non-empty set, by Lemma 3.7.(2). Let max.Locl(R) be the set of maximal elements of the poset (Locl(R), →). By the very definition of Locl(R) and by Lemma 3.3.(2), max.Locl(R) = {S−1R S ∈ max.Denl(R)} = {Ql(R/a) a ∈ ass.max.Denl(R)}. (8) Definition. Each element of max.Locl(R) is called a maximal left localization ring (or a maximal left quotient ring) of the ring R. Theorem 3.9 Let S ∈ max.Denl(R), A = S−1R, A∗ be the group of units of the ring A; a := ass(S), πa : R → R/a, a 7→ a + a, and σa : R → A, r 7→ r 1 . Then 1. S = Sa(R), S = π−1 a (S0(R/a)), πa(S) = S0(R/a) and A = S0(R/a)−1R/a = Ql(R/a). 2. S0(A) = A∗ and S0(A) ∩ (R/a) = S0(R/a). 3. S = σ−1 a (A∗). 4. A∗ = hπa(S), πa(S)−1i, i.e. the group of units of the ring A is generated by the sets πa(S) and π−1 a (S) := {πa(s)−1 s ∈ S}. 5. A∗ = {πa(s)−1πa(t) s, t ∈ S}. 6. Ql(A) = A and Assl(A) = {0}. In particular, if T ∈ Denl(A, 0) then T ⊆ A∗. Proof. 1. Since S = Sa(R) (by Theorem 2.1.(2)), statement 1 follows from Lemma 3.3.(2). 2. S0(A) st.1= S0(Ql(R/a)) Ql(R/a)∗ st.1= A∗ and S0(A) ∩ (R/a) st.1= S0(Ql(R/a)) ∩ Thm 2.8.(1) = (R/a) Thm2.8.(1) = 3. S = Sa(R) σ−1 a (A∗). S0(R/a). Pr 3.4.(1) = σ−1 a (S0(Sa(R)−1R)) = σ−1 a (S0(Ql(R/a))) Thm 2.8.(1) = a (Ql(R/a)∗) st.1= σ−1 4. A∗ st.1= Ql(R/a)∗ Thm 2.8.(3) 5. A∗ st.1= Ql(R/a)∗ Thm 2.8.(3) = = hS0(R/a), S0(R/a)−1i st.1= hπa(S), πa(S)−1i. {p−1q p, q ∈ S0(R/a)} = {πa(s)−1πa(t) s, t ∈ S} since πa(S) = S0(R/a), by statement 1. 6. Ql(A) st.1= Ql(Ql(R/a)) Thm 2.8.(4) = Ql(R/a) st.1= A. Since S0(A) st.1= S0(Ql(R/a)) Thm 2.8.(1) = Ql(R/a)∗ = A∗, T ⊆ A∗. The fact that Assl(A) = {0} follows from Proposition 3.8.(2) and the maximality of A. (cid:3) The next theorem is a criterion of when a ring A ∈ Locl(R, a) is equal to Qa(R). 15 Theorem 3.10 Let A ∈ Locl(R, a), i.e. A = S−1R for some S ∈ Denl(R, a). Then A = Qa(R) iff Ql(A) = A. Proof. If A = Qa(R) then A = Ql(R/a), by Lemma 3.3.(2), and so Ql(A) = Ql(Ql(R/a)) = Ql(R/a) = A, by Theorem 2.8.(4). If A 6= Qa(R) then the monomorphism A → Qa(R), a 7→ a, is not surjective. Let S1 := S 1 R) ≃ π1(S2)−1A and S2 := Sa(R). Then S1 ⊆ S2 and, by Lemma 3.2.(2), Qa(R) ≃ π1(S2)−1(S−1 where π1 : R → R/a, a 7→ a + a. Therefore, A 6= Ql(A). (cid:3) Left localization maximal rings. We introduce a new class of rings, the left localization maximal rings, which turn out to be precisely the class of maximal left quotient rings of all rings. As a result, we have a characterization of the maximal left quotient rings of a ring (Theorem 3.11). Definition. A ring A is called a left localization maximal ring if A = Ql(A) and Assl(A) = {0}. A ring A is called a right localization maximal ring if A = Qr(A) and Assr(A) = {0}. A ring A which is a left and right localization maximal ring is called a left and right localization maximal ring (i.e. Ql(A) = A = Qr(A) and Assl(A) = Assr(A) = {0}). Example. Let A be a simple ring. Then Ql(A) is a left localization maximal ring and Qr(A) is a right localization maximal ring. Example. A division ring is a (left and right) localization maximal ring. More generally, a simple artinian algebra (i.e. the matrix algebra over a division ring) is a (left and right) localization maximal ring. The next theorem is a criterion of when a left quotient ring of a ring is a maximal left quotient ring of the ring. Theorem 3.11 Let a ring A be a left localization of a ring R, i.e. A ∈ Locl(R, a) for some a ∈ Assl(R). Then A ∈ max.Locl(R) iff Ql(A) = A and Assl(A) = {0}, i.e. A is a left localization maximal ring. Proof. (⇒) Theorem 3.9.(6). (⇐) Proposition 3.8. (cid:3) The next corollary is a criterion of when Sa1(R) ⊆ Sa2(R) where a1, a2 ∈ Assl(R). Corollary 3.12 Let a1, a2 ∈ Assl(R) and σi Sa1(R) ⊆ Sa2 (R) iff a1 ⊆ a2 and σ2(Sa1(R)) ⊆ Qa2(R)∗. : R → Qai(R), r 7→ r/1, for i = 1, 2. Then Proof. (⇒) By Lemma 3.2.(2a), a1 ⊆ a2 and, by Lemma 3.2.(2b), σ2(Sa1 (R)) ⊆ Qa2(R)∗. (⇐) Let Si := Sai(R) and Qi := Qai(R) for i = 1, 2. Let Q′ be the subring of Q2 generated by R/a2 and σ2(S1)±1. Since S1 ∈ Denl(R, a1), a1 ⊆ a2 and σ2(S1) ⊆ Q∗ 2, every element of Q′ has the form σ2(s1)−1σ2(r) for some elements s1 ∈ S1 and r ∈ R. By the universal property of Q1 = S−1 1 R, there exists a ring homomorphism Q1 → Q2 and so we have the commutative diagram of ring homomorphisms: R = R R/a1 Q1 / R/a2 / Q2 . Since Si = σ−1 we have the inclusion S1 ⊆ S2. (cid:3) (Q∗ i i ) for i = 1, 2 (Lemma 3.3.(2) and Theorem 2.8), using the commutative diagram 16 / / / / / / Theorem 3.13 Let S ∈ max.Denl(R), A = S−1R, a = ass(S) and σa : R → A, r 7→ r following statements are equivalent. 1 . Then the 1. A is a semi-simple ring. 2. Qcl(R/a) exists and is a semi-simple ring. If one of these conditions holds then A = Qcl(R) and S = σ−1 a (Qcl(R)∗). Proof. 1. Since A = Ql(R/a) and S = σ−1 a (A∗), by Theorem 3.9.(1,3), the results follow from Theorem 2.9. (cid:3) The ideal p(S) and the set Jl(R, S). Proposition 3.14 Let S1, S2 ∈ Orel(R) be such that a1 := ass(S1) ⊆ a2 := ass(S2). Then 1. S1S2 ∈ Orel(R, a) such that a2 ⊆ a := ass(S1S2) where S1S2 := hS1, S2i is the sub-semigroup of (R, ·) generated by S1 and S2. 2. If, in addition, S1, S2, S ∈ Denl(R) then for each i = 1, 2 there is the ring homomorphism S−1 i R → S−1R, s−1r 7→ s−1r, with kernel S−1 i a. Proof. 1. Step 1: 0 6∈ S := hS1, S2i. Suppose that 0 ∈ S, we seek a contradiction. Then s1s2 · · · sn = 0 for some elements si ∈ S1 ∪ S2 and n ≥ 1. Clearly, n ≥ 2. We may assume that n is the least possible. Then, by the minimality of n, either all seven ∈ S1 and sodd ∈ S2 or otherwise seven ∈ S2 and sodd ∈ S1. If s1 ∈ S1 then s2 ∈ S2 and s2s3 · · · sn ∈ a1 ⊆ a2, hence (s′ 2 ∈ S2. This contradicts to the minimality of n since 2s2 ∈ S2. So, s1 ∈ S2. If sn ∈ S1 then s′ s′ 1 ∈ S1. This is not possible by the previous case. Therefore s1, sn ∈ S2. Then (s′ ns1)s2 · · · sn−1 = 0 for some element s′ n ∈ S2. This contradicts to the minimality of n since s′ 2s2)s2 · · · sn = 0 for some element s′ 1s1s2 · · · sn = 0 for some element s′ ns1 ∈ S2. Therefore, 0 6∈ S. Step 2: S ∈ Orel(R). We have to show that for any elements s ∈ S and r ∈ R there exist elements s′ ∈ S and r′ ∈ R such that s′r = r′s. To prove this we use induction on n where s = s1s2 · · · sn, si ∈ S1 ∪ S2. The result is obvious when n = 1 since S1, S2 ∈ Orel(R). Suppose that n > 1 and the result is true for all n′ < n. Fix t1 ∈ S1 ∪ S2 such that t1r = r1sn for some element r1 ∈ R. By induction, t2r1 = r′s1 · · · sn−1 for some t2 ∈ S and r′ ∈ R. Then t2t1r = t2r1sn = r′s1 · · · sn. It suffices to take s′ = t2t1. By Step 2, a is an ideal of the ring R such that a 6= R, by Step 1. Clearly, a2 ⊆ a. 2. Statement 2 follows from statement 1 and Lemma 3.2.(2a). (cid:3) In order to give an answer to the question of when an Ore set S of a ring is a denominator set or, more generally, when the image of S is a denominator set in a factor ring, we introduce the ideal p(S). Let W be the family of all ordinal numbers. For each left Ore set S ∈ Orel(R), we attach the ideal of the ring R, p(S) := p(S)l := [α∈W pα (9) where the ideals pα of the ring R are defined recursively as follows: p1 := assl(S, R) + assr(S, R)R where assl(S, R) := {r ∈ R sr = 0 for some s = s(r) ∈ S} and assr(S, R) := {r ∈ R rs = 0 for some s = s(r) ∈ S}. Note that assl(S, R) is an ideal of the ring R since S ∈ Orel(R) and assr(S, R) is a right ideal of the ring R. pα+1 := π−1 α (assl(πα(S), R/pα) + assr(πα(S), R/pα) · R/pα) where πα : R → R/pα, a 7→ a + pα. If α is a limit ordinal then pα = [β<α pβ. 17 (10) (11) By the very definition, if α ≤ β then pα ≤ pβ. In a similar fashion, for a right Ore set S ∈ Orer(R), we can define the ideal p(S)r. For each left Ore set S ∈ Orel(R), let Jl(R, S) := {a a is an ideal of R such that πa(S) ∈ Denl(R/a, 0)} where πα : R → R/a, a 7→ a + a. Theorem 3.15 Let S ∈ Orel(R) and a = ass(S). Then 1. Jl(R, S) 6= ∅ iff p(S) 6= R. 2. If Jl(R, S) 6= ∅ then p(S) is the least (with respect to inclusion) element of Jl(R, S). 3. If, in addition, S ∈ Denl(R, a) then p(S) = a. 4. If the ring R satisfies the ascending chain condition on annihilator right ideals then S ∈ Denl(R, a) and p(S) = a. Proof. 1. (⇒) Suppose that Jl(R, S) 6= ∅. Fix a ∈ Jl(R, S). Clearly, assl(S, R), assr(S, R) ⊆ a, hence p1 ⊆ a. If pα ⊆ a then pα+1 ⊆ a since sa, bt ∈ pα ⊆ a for some s, t ∈ S and a, b ∈ R implies a, b ∈ a. If α is a limit ordinal and pβ ⊆ a for all β < α then pα = Sβ<α pβ ⊆ a. Therefore, p(S) ⊆ a, and so p(S) 6= R. (⇐) Suppose that p := p(S) 6= R. We claim that πp(S) ∈ Denl(R/p, 0) where πp : R → R/p, a 7→ a + a. If πp(s)πp(a) = 0 and πp(b)πp(t) = 0 in R/p for some elements s, t ∈ S and a, b ∈ R then sa, bt ∈ p =Sα∈W pα, i.e. sa, bt ∈ pα for some ordinal number α. Therefore, a, b ∈ pα+1 ⊆ p, i.e. πp(a) = πp(b) = 0. This proves that assl(πp(S), R/p) = 0 and assr(πp(S), R/p) = 0. To finish the proof of the fact that πp(S) ∈ Denl(R/p, 0) we have to show that πp(S) ∈ Orel(R/p), but his is obvious as epimorphisms respect left Ore sets provided their images are multiplicatively closed sets which is the case for πp(S) as p(S) 6= R. 2. In statement 1 we proved that p(S) ⊆ a for all ideals a ∈ Jl(R, S) and p(S) ∈ Jl(R, S), i.e. p(S) is the least element of the poset (Jl(R, S), ⊆). 3. Statement 3 is obvious. 4. By Lemma 1.1.2, [3], S ∈ Denl(R, a) and so p(S) = a, by statement 3. (cid:3) Localizable left Ore sets, a slight generalization of Ore's method of localization. Definition. A left Ore set S ∈ Orel(R) is called localizable if there exists a ring homomorphism ϕ : R → R′ where R′ is a ring such that, for all s ∈ S, ϕ(s) is a unit of the ring R′. The set of all localizable left Ore sets in R is denoted by orel(R). By Theorem 3.15, orel(R) = {S ∈ Orel(R) p(S) 6= R}. (12) Corollary 3.16 Let S be a localizable left Ore set in a ring R. Then there exists an ordered pair (Q, f ) where Q is a ring and f : R → Q is a ring homomorphism such that (i) for all s ∈ S, f (s) is a unit in Q; and if (Q′, f ′) is another pair satisfying the condition (i) then there is a unique ring homo- morphism h : Q → Q′ such that f ′ = hf . The ring Q is unique up to isomorphism. The ring Q is isomorphic to the left localization of the ring R/p(S) at the left denominator set π(S) ∈ Denl(R/p(S), 0) where the ideal p(S) of the ring R is defined in (11) and π : R → R/p(S), a 7→ a + p(S). Proof. Recall that p(S) = Sα∈W pα. By induction on α, all pα ⊆ ker(f ′). Therefore, p(S) ⊆ ker(f ′). Notice that Q′ ≃ πa(S)−1(R/a) for some a ∈ Jl(R, S). Let us identify this rings via the isomorphism above, then f ′(R) = R/a. By Theorem 3.15.(2), p(S) ⊆ a, and so there is a natural ring homomorphism ϕ : R/p(S) → πa(S)−1(R/a) such that, for all s ∈ S, ϕ(s) is a unit. By the universal property of the ring πp(S)(S)−1(R/p(S)) there is a unique homomorphism h : πp(S)(S)−1(R/p(S)) → Q′ such that f ′ = hf . This proves existence of the ring Q. It is unique up to isomorphism as it follows at once from the uniqueness of h. (cid:3) 18 The set LDenl(R) := {Sa(R) a ∈ Assl(R)} is the set of largest left denominator sets of the ring R. Let min.LDenl(R, a) and min.Locl(R, a) be the sets of minimal elements of the posets LDenl(R, a) and Locl(R, a) respectively. 4 The largest quotient ring of a ring In this section, the two-sided analogues of some results of Sections 2 and 3 are proved. The proofs are dropped in all cases where the two-sided analogues are direct corollaries of the one-sided results. In particular, it is proved existence of the largest quotient ring of a ring (Theorem 4.1) and that all Ore sets are localizable (Theorem 4.15). The results of this section are used in Section 5. Notation: • Ore(R) := Orel(R) ∩ Orer(R) = {S S is a left and right Ore set in R}; • Den(R) := Denl(R) ∩ Denr(R) = {S S is a left and right denominator set in R}. For each S ∈ Den(R), ass(S) := assl(S) = assr(S) is an ideal of the ring R where assl(S) := {r ∈ R sr = 0 for some s = s(r) ∈ S} and assr(S) := {r ∈ R rs = 0 for some s = s(r) ∈ S}; • Loc(R) := {S−1R = RS−1 S ∈ Den(R)}; • Ass(R) := {ass(S) S ∈ Den(R)}; • Den(R, a) := {S ∈ Den(R) ass(S) = a} where a ∈ Ass(R); • Sa = Sa(R) is the largest element of the poset (Den(R, a), ⊆) and Qa(R) := S−1 a R = RS−1 a is the largest (two-sided) quotient ring associated to a, Sa exists (Theorem 4.1.(2)); • In particular, S0 = S0(R) is the largest element of the poset (Den(R, 0), ⊆) and Q(R) := S−1 0 R = RS−1 0 is the largest (two-sided) quotient ring of R; • Loc(R, a) := {S−1R = RS−1 S ∈ Den(R, a)}. Remark. Subscripts 'l' and 'r' indicate that left and right versions of a definition/concept are considered respectively. The largest quotient ring of a ring. Theorem 4.1 1. For each a ∈ Ass(R), the set Den(R, a) is an ordered abelian semigroup (S1S2 = S2S1, and S1 ⊆ S2 implies S1S3 ⊆ S2S3) where the product S1S2 = hS1, S2i is the multiplicative subsemigroup of (R, ·) generated by S1 and S2. 2. Sa := Sa(R) := SS∈Den(R,a) S is the largest element (w.r.t. ⊆) in Den(R, a). The set Sa is called the largest denominator set associated to a. 3. Let Si ∈ Den(R, a), i ∈ I, where I is an arbitrary non-empty set. Then hSi i ∈ Ii := [∅6=J⊆I,J<∞Yj∈J Sj ∈ Den(R, a) (13) the denominator set generated by the denominators sets Si, it is the least upper bound of the set {Si}i∈I in Den(R, a), i.e. hSi i ∈ Ii =Wi∈I Si. In Section 5, we will see that for the algebra I1 of polynomial integro-differential operators the set S0(I1) (resp. the ring Q(I1)) is tiny comparing to the sets Sl,0(I1) and Sr,0(I1) (resp. to the rings Ql(I1) and Qr(I1)). 19 Corollary 4.2 The abelian monoid Den(R, 0) is a complete lattice such that S1S2 = S1W S1 and Vi∈I Si =WS∈Den(R,0),S⊆∩i∈I Si Si where all Si ∈ Den(R, 0). Clearly, Vi∈I Si is the largest element of the set {S S ∈ Den(R, 0), S ⊆Ti∈I Si}. Corollary 4.3 1. Let R be a ring. Each ring automorphism σ ∈ Aut(R) of the ring R has the unique extension σ ∈ Aut(Q(R)) to an automorphism of the ring Q(R) given by the rule σ(s−1r) = σ(s)−1σ(r) where s ∈ S0(R) and r ∈ R. 2. The group Aut(R) is a subgroup of the group Aut(Q(R)). Moreover, Aut(R) = {τ ∈ Aut(Q(R)) τ (R) = R}. Theorem 4.4 1. S0(Q(R)) = Q(R)∗ and S0(Q(R)) ∩ R = S0(R). 2. Q(R)∗ = hS0(R), S0(R)−1i, i.e. the group of units of the ring Q(R) is generated by the sets S0(R) and S0(R)−1 := {s−1 s ∈ S0(R)}. 3. Q(R)∗ = {s−1t s, t ∈ S0(R)} = {ts−1 s, t ∈ S0(R)}. 4. Q(Q(R)) = Q(R). Proof. 1 -- 3. It is obvious that G := hS0(R), S0(R)−1i ⊆ Q(R)∗ ⊆ S0(Q(R)). Applying Proposition 2.7.(2) and its right analogue in the situation where S = S0(R) and T = S0(Q(R)) we see that S0(R) ⊆ T ′ := S0(Q(R)) ∩ R ∈ Denl(R, 0) ∩ Denr(R, 0), and so T ′ ∈ Den(R, 0) and S0(R) = T ′, by the maximality of S0(R). Let q ∈ S0(Q(R)). Then q = s−1t = t1s−1 for some elements s, s1 ∈ S0(R), t, t1 ∈ R, t = sq ∈ S0(Q(R)) ∩ R = S0(R) 1 and t1 = qs1 ∈ S0(Q(R)) ∩ R = S0(R). Therefore, S0(Q(R)) ⊆ {s−1t s, t ∈ S0(R)} ⊆ G and S0(Q(R)) ⊆ {ts−1 s, t ∈ S0(R)} ⊆ G, and so G = Q(R)∗ = S0(Q(R)) = {s−1t s, t ∈ S0(R)} = {ts−1 s, t ∈ S0(R)}. This proves statements 1 -- 3. 4. Statement 4 follows from statement 1. (cid:3) Theorem 4.5 Let R be a ring and CR be the set of regular elements of the ring R (i.e. the set of non-zero-divisors). Then the following statements are equivalent. 1. Q(R) is a semi-simple ring. 2. The rings Ql,cl(R) and Qr,cl(R) exist and are semi-simple. 3. The rings Ql(R) and Qr(R) are semi-simple. If one of the equivalent conditions holds then S0(R) = CR and Q(R) ≃ Ql,cl(R) ≃ Qr,cl(R) ≃ Ql(R) ≃ Qr(R). Proof. (1 ⇒ 3) Since Q(R) ⊆ Ql(R), Q(R) ⊆ Qr(R) and the ring Q(R) is a semi-simple, by Lemma 3.2.(2), the inclusions are, in fact, equalities (since the regular elements of any semi-simple ring are the units). Therefore, the rings Ql(R) and Qr(R) are semi-simple. (2 ⇔ 3) Statements 2 and 3 are equivalent, by Theorem 2.9. Moreover, Ql,cl(R) ≃ Qr,cl(R) ≃ Ql(R) ≃ Qr(R). (2 ⇒ 1) Recall that Ql,cl(R) = C−1 R where CR is the set of regular ele- ments of the ring R. Since the rings Ql,cl(R) and Qr,cl(R) exist, they are R-isomorphic. Therefore, S0(R) = CR and Q(R) ≃ Ql,cl(R) ≃ Qr,cl(R). In particular, Q(R) is a semi-simple ring. (cid:3) R R and Qr,cl(R) = RC−1 Proposition 4.6 If the ring Ql(R) is a left artinian ring and the ring Qr(R) is a right artinian ring then S0(R) = CR and Q(R) ≃ Ql,cl(R) ≃ Qr,cl(R) ≃ Ql(R) ≃ Qr(R). 20 Proof. By Corollary 2.10, Sl,0(R) = CR and Ql,cl(R) ≃ Ql(R). By the right version of Corollary 2.10, Sr,0(R) = CR and Qr,cl(R) ≃ Qr(R). Then S0(R) = CR and Q(R) ≃ Ql,cl(R) ≃ Qr,cl(R) ≃ Ql(R) ≃ Qr(R). (cid:3) The maximal quotients rings of a ring. Lemma 4.7 Let S ∈ Den(R, a), π : R → R/a, a 7→ a + a, and σ : R → S−1R, r 7→ r/1. 1. Let T ∈ Den(S−1R, 0) be such that π(S), π(S−1) ⊆ T . Then T ′ := σ−1(T ) ∈ Den(R, a), T ′ is S-saturated, T = {s−1t′ s ∈ S, t′ ∈ T ′}, and S−1R ⊆ T ′−1R = T −1R. 2. π−1(S0(R/a)) = Sa(R), π(Sa(R)) = S0(R/a)) and Qa(R) = Q(R/a). Proposition 4.8 Let S ∈ Den(R, a); π : R → R/a, a → a = a + a; σ : R → S−1R, r → r/1; G := hπ(S), π(S)−1i ⊆ (S−1R)∗ (i.e. G is the subgroup of the group (S−1R)∗ of units of the ring S−1R generated by π(S)±1). Let [σ−1(G), Sa(R)] := {S ∈ Den(R, a) σ−1(G) ⊆ S ⊆ Sa(R)}. 1. Let [σ−1(G), Sa(R)]S−com := {S1 ∈ [σ−1(G), Sa(R)] σ−1(Gπ(S1)) = S1} and [G, S0(S−1R)] := {T ∈ Den(S−1R, 0) G ⊆ T ⊆ S0(S−1R)}. Then the map [σ−1(G), Sa(R)]S−com → [G, S0(S−1R)], S1 7→ eS1 := Gπ(S1), is an isomorphism of posets and abelian monoids with the inverse map T 7→ σ−1(T ) where Gπ(S1) is the multiplicative monoid generated by G and π(S1) in S−1R. In particular, Gπ(Sa(R)) = S0(S−1R), Sa(R) = σ−1(S0(S−1R)), Sa(R)−1R ≃ Sa(R)−1R ≃ Q(R/a), the monoid [σ−1(G), Sa(R)]S−com is a complete lattice (since [G, S0(S−1R)] is a complete lattice, as an interval of the complete lattice Den(S−1R, 0), Corollary 4.2), and the map S1 7→ eS1 is a lattice isomorphism. 2. Consider the interval [G∩(R/a), S0(R/a)] in Den(R/a, 0). Let [G∩(R/a), S0(R/a)]G−com := {T ∈ [G∩(R/a), S0(R/a)] GT ∩(R/a) = T }. Then [G∩(R/a), S0(R/a)]G−com ⊆ Den(S−1R, 0) and the map [G ∩ (R/a), S0(R/a)]G−com → [G, S0(S−1R)], T 7→ GT, is an isomorphism of posets and abelian monoids with the inverse map T ′ 7→ T ′∩(R/a) where GT is the product in Den(S−1R, 0). In particular, the monoid [G ∩ (R/a), S0(R/a)]G−com is a complete lattice. 3. The map [σ−1(G), Sa(R)]S−com → [G ∩ (R/a), S0(R/a)]G−com, S1 7→ Gπ(S1) ∩ (R/a), is an isomorphism of posets and abelian monoids with the inverse map S′ 7→ π−1(S′). Lemma 4.9 1. For each element S ∈ Den(R), there exists a maximal element in the poset ([S, ·), ⊆). 2. The set (max.Den(R), ⊆) of maximal elements of the poset (Den(R), ⊆) is a non-empty set. Definition. An element S of the set max.Den(R) is called a maximal denominator set of R and the ring S−1R = RS−1 is called a maximal quotient ring of R. Proposition 4.10 Let a ∈ Ass(R), Q := Qa(R), Q∗ be the group of units of the ring Q and 1 . Let T ∈ Den(Q, b) where b ∈ Ass(Q) and Q∗T be the multiplicative σ : R → Qa(R), r 7→ r sub-semigroup of (Q, ·) generated by Q∗ and T . Then 1. Q∗T ∈ Den(Q, b). 21 2. If, in addition, Q∗ ⊆ T (eg, Q∗T from statement 1) then (a) T ′ := σ−1(T ) ∈ Den(R, b′) where b′ := σ−1(b) ⊇ a, Sa(R) ⊆ T ′, T = Q∗σ(T ′) (i.e. the monoid T is generated by Q∗ and σ(T ′)) and T ′−1R = T −1Q (i.e. the natural ring monomorphism T ′−1R → T −1Q, t−1r 7→ t−1r, is an isomorphism). (b) Sa(R) ⊆ Sb′(R) and Sb′ (R) = σ−1(Sb(Q)). (c) Qb′(R) = Q(Qa(R)/b), i.e. the natural ring monomorphism Qb′(R) → Q(Qa(R)/b), t−1r 7→ t−1r, is an isomorphism. Let max.Ass(R) be the set of maximal elements of the poset (Ass(R), ⊆). It is a subset of the set ass.max.Den(R) := {ass(S) S ∈ max.Den(R)} (14) which is a non-empty set, by Lemma 4.9.(2). Let max.Loc(R) be the set of maximal elements of the poset (Loc(R), →). By the very definition of Loc(R) and by Lemma 4.7.(2), max.Loc(R) = {S−1R S ∈ max.Den(R)} = {Q(R/a) a ∈ ass.max.Den(R)}. (15) Definition. Each element of max.Loc(R) is called a maximal localization ring (or a maximal quotient ring) of the ring R. Theorem 4.11 Let S ∈ max.Den(R), A = S−1R, A∗ be the group of units of the ring A; a := ass(S), πa : R → R/a, a 7→ a + a, and σa : R → A, r 7→ r 1 . Then 1. S = Sa(R), S = π−1 a (S0(R/a)), πa(S) = S0(R/a) and A = S0(R/a)−1(R/a) = (R/a)S0(R/a)−1 = Q(R/a). 2. S0(A) = A∗ and S0(A) ∩ (R/a) = S0(R/a). 3. S = σ−1 a (A∗). 4. A∗ = hπa(S), πa(S)−1i, i.e. the group of units of the ring A is generated by the sets πa(S) and π−1 a (S) := {πa(s)−1 s ∈ S}. 5. A∗ = {πa(s)−1πa(t) s, t ∈ S}. 6. Q(A) = A and Ass(A) = {0}. In particular, if T ∈ Den(A, 0) then T ⊆ A∗. The next theorem is a criterion of when a ring A ∈ Loc(R, a) is equal to Qa(R). Theorem 4.12 Let A ∈ Loc(R, a), i.e. A = S−1R for some S ∈ Den(R, a). Then A = Qa(R) iff Q(A) = A. Localization maximal rings. We introduce a new class of rings, the localization maximal rings, which turn out to be precisely the class of maximal quotient rings of all rings. As a result, we have a characterization of the maximal quotient rings of a ring (Theorem 4.13). Definition. A ring A is called a localization maximal ring if A = Q(A) and Ass(A) = {0}. Clearly, a left and right localization maximal ring A (i.e. Ql(A) = A = Qr(A) and Ass(A) = Assr(A) = {0}) is a localization maximal but vice versa, in general, as the example of the algebra I1 shows (see Section 5 for details). The next theorem is a criterion of when a quotient ring of a ring is a maximal quotient ring of the ring. Theorem 4.13 Let a ring A be a localization of a ring R, i.e. A ∈ Loc(R, a) for some a ∈ Ass(R). Then A ∈ max.Loc(R) iff Q(A) = A and Ass(A) = {0}, i.e. A is a localization maximal ring. 22 The next corollary is a criterion of when Sa1(R) ⊆ Sa2(R) where a1, a2 ∈ Ass(R). Corollary 4.14 Let a1, a2 ∈ Ass(R) and σi : R → Qai(R), r 7→ r/1, for i = 1, 2. Then Sa1(R) ⊆ Sa2(R) iff a1 ⊆ a2 and σ2(Sa1 (R)) ⊆ Qa2(R)∗. Proof. (⇒) By Lemma 3.2.(2a), a1 ⊆ a2 and, by Lemma 3.2.(2b) and its right analogue, σ2(Sa1(R)) ⊆ Qa2(R)∗. (⇐) Let Si := Sai(R) and Qi := Qai(R) for i = 1, 2. Let Q′ be the subring of Q2 generated by R/a2 and σ2(S1)±1. Since S1 ∈ Den(R, a1), a1 ⊆ a2 and σ2(S1) ⊆ Q∗ 2, every element of Q′ has the form σ2(s)−1σ2(r) = σ2(r′)σ2(s′)−1 for some elements s, s′ ∈ S1 and r, r′ ∈ R. By the universal property of Q1 = S−1 1 , there exists a ring homomorphism Q1 → Q2 and so we have the commutative diagram of ring homomorphisms: 1 R = RS−1 R = R R/a1 Q1 / R/a2 / Q2 . Since Si = σ−1 i (Q∗ i ) for i = 1, 2 (Lemma 4.7.(2)), we have the inclusion S1 ⊆ S2. (cid:3) All Ore sets are localizable, a slight generalization of Ore's method of localization. Definition. An Ore set S ∈ Ore(R) is called localizable if there exists a ring homomorphism ϕ : R → R′ where R′ is a ring such that, for all s ∈ S, ϕ(s) is a unit of the ring R′. We will see that all Ore sets are localizable (Theorem 4.15 and Corollary 4.16). For each right Ore set S ∈ Orer(R) of the ring R, let p(S)r := Sα∈W p′ of the ideal p(S)l defined in (9) where α be the right version p′ α+1 := π′−1 α (R/pα · assl(π′ α(S), R/pα) + assr(π′ α(S), R/pα)), π′ α : R → R/p′ α, a 7→ p′ α. If α is a limit ordinal then p′ α = [β<α p′ α. For each Ore set S ∈ Ore(R), let J (R, S) := {a a is an ideal of R such that πa(S) ∈ Den(R/a, 0)} (16) (17) where πa : R → R/a, a 7→ a + a. p(S)r =Sα∈W p′ α coincides since, for all α ∈ W, It follows from (9) that the ideals p(S)l = Sα∈W pα and pα = p′ α. (18) Moreover, for all α ∈ W, pα+1 := π−1 α (assl(πα(S), R/pα) + assr(πα(S), R/pα)) and p(S) = [α∈W pα. (19) Theorem 4.15 Let S ∈ Ore(R). Then 1. J (R, S) 6= ∅. 2. p(S) is the least (with respect to inclusion) element of J (R, S). 3. If, in addition, S ∈ Den(R, a) then p(S) = a. 23 / / / / / / Proof. The theorem follows at once from Theorem 3.15 and its analogue for the right Ore sets in R provided we show that p(S) 6= R. Suppose that p(S) = R, we seek a contradiction. Then pα ∩ S 6= ∅ for some α. We can assume that α is the least possible. Then α is necessarily not a limit ordinal. Then 0 6∈ πα−1(S) since otherwise we would have pα−1 ∩ S 6= ∅, a contradiction. Therefore, πα−1(S) ∈ Ore(R/pα−1). Replacing R and S by R/pα−1 and πα−1(S) respectively we may assume that α = 1, i.e. s′ ∈ p1 = assl(S, R) + assr(S, R) for some s′ ∈ S. Then a + b = s′ for some elements a, b ∈ R such that sa = 0 and bt = 0 for some elements s, t ∈ S. Then S ∋ ss′t = s(a + b)t = 0, a contradiction. The proof of the theorem is complete. (cid:3) Corollary 4.16 Let S be an Ore set in a ring R. Then there exists an ordered pair (Q, f ) where Q is a ring and f : R → Q is a ring homomorphism such that (i) for all s ∈ S, f (s) is a unit in Q; and if (Q′, f ′) is another pair satisfying the condition (i) then there is a unique ring homomor- phism h : Q → Q′ such that f ′ = hf . The ring Q is unique up to isomorphism. The ring Q is isomorphic to the localization of the ring R/p(S) at the denominator set π(S) ∈ Den(R/p(S), 0) where the ideal p(S) of the ring R is defined in (19) and π : R → R/p(S), a 7→ a + p(S). 5 Examples In this section, the largest (left; right; left and right) quotient ring and the maximal (left; right; left and right) quotient rings are found for the following rings: the endomorphism algebra EndK(V ) of an infinite dimensional vector space with countable basis, semi-prime Goldie rings, the algebra I1 of polynomial integro-differential operators, and Noetherian commutative rings. The endomorphism algebra EndK(V ) of an infinite dimensional vector space V with countable basis. For a vector space V , let F = F (V ) := {ϕ ∈ EndK(V ) dimK(ker(ϕ)) < ∞, dimK(coker(ϕ)) < ∞} be the set of Fredholm linear maps/operators in V . Lemma 5.1 Let V be an infinite dimensional vector space with countable basis and R := EndK(V ). Then 1. Let ϕ ∈ R and dimK(im(ϕ)) = ∞. Then (a) αϕβ = 1 for some elements α and β of R (necessarily, α is a surjection and β is an injection). Moreover, α and β can be chosen to satisfy the following conditions: (b) If ϕ is a surjection then ϕβ = 1 for some (necessarily, injective) map β ∈ R that can (c) If ϕ is an injection then αϕ = 1 for some (necessarily, surjective) map α ∈ R which 2. The ideal of compact operators C := {ϕ ∈ R dimK(im(ϕ)) < ∞} is the only proper ideal of V = ker(α)L im(ϕ) and V = ker(ϕ)L im(β). be chosen to satisfy the equality V = ker(ϕ)L im(β). can be chosen to satisfy the equality V = ker(α)L im(ϕ). the ring R. 3. Let ϕ ∈ F (V ) and c ∈ R. Then (a) If ϕc = 0 then c ∈ C. (b) If cϕ = 0 then c ∈ C. 4. Let ϕ ∈ R. Then (a) ker(ϕ) 6= 0 iff ϕc = 0 for some element 0 6= c ∈ C. (b) im(ϕ) 6= V iff cϕ = 0 for some element 0 6= c ∈ C. 24 5. Let ϕ ∈ R. Then ϕ is a right regular element in R (i.e. ϕψ = 0 implies ψ = 0) iff the map ϕ : V → V is an injection. 6. Let ϕ ∈ R. Then ϕ is a left regular element in R (i.e. ψϕ = 0 implies ψ = 0) iff the map ϕ : V → V is a surjection. 7. Let ϕ ∈ R. Then ϕ is a regular element of R iff ϕ is a unit of R. So, CR = AutK(V ) and Qcl(R) = Qcl,l(R) = Qcl,r(R) = R. Proof. 1. This is obvious. 2. Statement 2 follows from statement 1. 3a. im(c) ⊆ ker(ϕ), and so c ∈ C. 3b. ker(c) ⊇ im(ϕ), and so dimK(im(ϕ)) ≤ dimK(coker(ϕ)) < ∞, i.e. c ∈ C. 4. This is obvious. 5. Statement 5 follows from statement 4a. 6. Statement 6 follows from statement 4b. 7. Statement 7 follows from statements 5 and 6. (cid:3) Theorem 5.2 Let V be an infinite dimensional vector space with countable basis and R := EndK(V ). Then 1. Assl(R) = Assr(R) = Ass(R) = {0, C}. 2. Sl,0(R) = Sr,0(R) = S0(R) = AutK(V ) and Ql(R) = Qr(R) = Q(R) = R. 3. Sl,C(R) = Sr,C(R) = SC(R) = F and Ql,C(R) = Qr,C(R) = QC(R) = R/C. 4. max.Assl(R) = max.Assr(R) = max.Ass(R) = {C}. 5. R/C is a localization maximal ring and a left (resp. right; left and right) localization maximal ring. Proof. 1. Statement 1 follows from statements 2 and 3. 2. Statement 2 follows from Lemma 5.1.(7). 4. Statement 4 follows from statement 3. 3. Let π : R → R/C, a 7→ a + C. The ordered pair (R/C, π) satisfies the following conditions (i) for all s ∈ F , π(s) is a unit; (ii) for all q ∈ R/C, q = π(s)−1π(r) = π(r1)π(s1)−1 for some s, s1 ∈ F and r, r1 ∈ R; (iii) ker(π) = C and C = assl(F ) = assr(F ), by Lemma 5.1.(3). By Ore's Theorem, R/C = F −1R = RF −1. To finish the proof of statement 3 it suffices to show that every regular element π(r) in R/C is invertible. Since π(r) regular in R/C, dimK(ker(r)) < ∞ (suppose that dimK(ker(r)) = ∞, we seek a contradiction. Fix a complement subspace U of ker(r) in V , i.e. V = ker(r)L U . Let p be the projection onto ker(r). Then π(r)π(p) = 0 but π(p) 6= 0 ∞, we seek a contradiction. Fix a complement subspace W of im(r) in V , i.e. V = im(r)L W . Let q be the projection onto W . Then π(q)π(r) = 0 but π(q) 6= 0 since dimK(W ) = dimK(coker(r)) = ∞, a contradiction). Therefore, r ∈ F , and so π(r) is a unit. since dimK(ker(r)) = ∞, a contradiction). Similarly, since π(r) is regular in R/C, dimK(coker(r)) < ∞ (suppose that dimK(coker(r)) = 5. Statement 5 follows from statements 3 and 4, see Theorem 3.11. (cid:3) Semi-prime Goldie rings. A ring R is called a left Goldie ring if R has finite left uniform dimension and R satisfies the a.c.c. on left annihilators. Similarly, a right Goldie ring is defined. A left and right Goldie ring is called a Goldie ring. The reader is referred to the books [3], [5] and [6] for more details. Lemma 5.3 Let R be a ring. 25 1. If S and T are left (resp. right; left and right) Ore sets in R then so is the semigroup ST generated by S and T in R provided 0 6∈ ST . 2. If S ∈ Den(R, a) and C ∈ Den(R, 0) then CS ∈ Den(R, a), i.e. the monoid CS generated by C and S in R is a denominator set with ass(CS) = a. nr = r′ n ∈ R such that t′ n ∈ R such that s′ ntn. Since S is a left Ore set, there are elements s′ Proof. 1. It suffices to prove statement 1, say, for left Ore sets S and T . We have to show that for elements s1t1s2t2 · · · sntn ∈ ST (where si ∈ S and ti ∈ T ) and r ∈ R there exist elements θ ∈ ST and r′ ∈ R such that θr = r′s1t1 · · · sntn. Since T is a left Ore set, there are elements t′ n ∈ T and r′ n ∈ S and r′′ nsntn. Repeating these two steps n − 1 more times we find elements s′ 1 ∈ R such that s′ 1t′ 1 · · · s′ 2. Let us show that CS is a multiplicative set. Suppose that c1s1 · · · cnsn = 0 for some elements ci ∈ C and si ∈ S, we seek a contradiction. Then s1c2s2 · · · cnsn = 0 since c1 is a regular element, and so c2s2 · · · cnsns′ n ∈ S since S ∈ Den(R, a). Repeating this argument several times we come to the situation where c′s′ = 0 for some elements c′ ∈ C and s′ ∈ S, i.e. s′ = 0 (since c′ is a regular element), a contradiction. 1, . . . , s′ 1 s1t1 · · · sntn. Now, set θ := s′ nr = r′′ 1, . . . , t′ n ∈ S and r′ := r′′ n−1 ∈ T and r′′ 1 ∈ R. n = 0 for some element s′ n−1 ∈ S; t′ 1t′ nt′ nr = r′′ 1 · · · s′ nr′ n = r′′ nsn. Therefore, s′ nt′ nt′ By statement 1, CS is an Ore set in R. To finish the proof of statement 2 it suffices to show that ass(CS) = a, i.e. c1s1 · · · cnsnr = 0 for some ci ∈ C, si ∈ S and r ∈ R, implies that r ∈ a. The element c1 is a regular element, hence s1 · · · cnsnr = 0, and so c2s2 · · · cnsnrs′ 1 = 0 for some s′ 1 ∈ S since S ∈ Den(R, a). Repeating the same two steps n − 1 more times gives rs′ n = 0 for some elements s′ 2 · · · s′ i ∈ S, i.e. r ∈ a. (cid:3) 1s′ Corollary 5.4 Let R be a prime Goldie ring and CR be the set of regular elements of R. Then Ass(R) = {0}, S0(R) = CR, Q0(R) = Qcl(R) is the only maximal localization of the ring R. Proof. Let a ∈ Ass(R) and S ∈ Den(R, a). By Lemma 5.3.(2), CRS ⊆ Sa(R). Since Qcl(R) is a simple artinian ring (i.e. the matrix ring over a division ring), we see that a = 0, and so CR = S0(R). (cid:3) R R ≃ RC−1 Let R be a semi-prime Goldie ring which is not a prime ring and CR be its set of regular elements. Let Min(R) = {p1, . . . , ps} be the set of minimal primes of the ring R. By Goldie's Theorem, Qcl(R) := C−1 i=1 Ri is the direct product of simple artinian rings Ri (i.e. Ri is a matrix ring over a division ring). The ring R can be identified with its image under i6=j=1 Rj) (Proposition 3.2.2, [5]). For each non-empty set I of the set {1, . . . , s}, let the ring homomorphism i be is the group of units of the ring RI . A the ring monomorphism σ : R → Qcl(R), r 7→ r/1. For each i = 1, . . . , s, pi = σ−1(Qs σI : R → RI := Qi∈I Ri be the composition of σ and the projection Qs the group of units of the ring Ri. Then R∗ subset A of a set B is called a proper subset of B if A 6= ∅, B. I = Qi∈I R∗ i=1 Ri → RI . Let R∗ R ≃ Qs i Theorem 5.5 Let R be a semi-prime Goldie ring which is not a prime ring and {p1, . . . , ps} be the set of its minimal prime ideals. Then 1. Ass(R) := {a(I) :=Ti∈I pi ∅ $ I ⊆ {1, . . . , s}}. 2. For each proper subset I of {1, . . . , s}, Sa(I)(R) = σ−1 I (R∗ I ×RCI ) and Qa(I)(R) = Qcl(R)/RCI ≃ RI where CI := {1, . . . , s}\I. 3. S0(R) = CR ⊆ Sa(R) for all a ∈ Ass(R). 4. Sa(I)(R) ⊆ Sa(J)(R) iff I ⊇ J where I and J are proper subsets of {1, . . . , s}. 5. If I and J are proper subsets of {1, . . . , s} such that I ⊇ J then, by statement 4 and the universal property of localization, there is the unique ring homomorphism Qa(I)(R) = Qi∈I RI → Qa(J)(R) =Qj∈J Rj which is necessarily the projection ontoQj∈J Rj inQi∈I Ri. 26 6. max.Ass(R) = {pi, i = 1, . . . , s}, {Qpi(R) = Ri i = 1, . . . , s} is the set of maximal localiza- tions of the ring R and ass(Qpi)(R) = pi for all i = 1, . . . , s. Proof. 1 -- 3. By Theorem 4.5, S0(R) = CR and Q(R) = Qcl(R). By Lemma 5.3.(2), CR ⊆ Sa(R) for all a ∈ Ass(R) (and statement 3 follows), and there is a natural homomorphism Qcl(R) → Qa(R), r 7→ r/1. By Lemma 3.2, the ring Qa(R) is a localization of the ring Qcl(R). Moreover, by Proposition 4.10.(2), the map Ass(R) → Ass(Qcl(R)), a 7→ S−1a, is a bijection with the inverse map and the map b 7→ R ∩ b, {Sa(R) a ∈ Ass(R)} → {Sb(Qcl(R)) b ∈ Ass(Qcl(R))}, SR∩b(R) 7→ Sb(Qcl(R)), is a bijection and where SR∩b(R) = σ−1 R∩b(Sb(Qcl(R))) (20) (21) σR∩b : R → QR∩b(R) = SR∩b(R)−1R ≃ Sb(Qcl(R))−1Qcl(R). (22) QRCI (Qcl(R)) (by Theorem 4.12, since every regular element of RI is a unit). Now, statement 1 Clearly, Ass(Qcl(R)) = Ass(Qs i=1 Ri) = {0}S{RI , ∅ 6= I & {1, . . . , s}} and RI = Qcl(R)/RCI = follows from (20) and the fact that pi = R ∩Qj6=i Rj. For every proper subset I of {1, . . . , s}, CI × RI . Then, by (21), SR∩RI (R) = CI ), and, by (22), Qa(CI)(R) = QR∩RI (R) ≃ SRI (Qcl(R))−1Qcl(R) ≃ R ∩ RI = a(CI), σR∩RI = σCI and SRI (Qcl(R)) = R∗ σ−1 R∩RI (SRI (Qcl(R))) = σ−1 RCI . CI (R∗ 4. Statement 4 follows from statement 2. 5. Statement 5 follows from statement 2. 6. Statement 6 follows from statements 2, 4 and 5. (cid:3) The algebra I1 of polynomial integro-differential operators. Let us collect some facts for the algebra I1 which are used in the proofs of Theorem 5.7 and Proposition 5.8. For the details the reader is referred to [1] or [2]. Throughout, K is a field of characteristic zero and K ∗ is its group of units; P1 := K[x] is a polynomial algebra in one variable x over K; ∂ := d dx ; EndK(P1) is the algebra of all K-linear maps from P1 to P1, and AutK(P1) is its group of units (i.e. the group of all the invertible linear maps from P1 to P1); the subalgebras A1 := Khx, ∂i and I1 := Khx, ∂,R i operators respectively where R : P1 → P1, p 7→ R p dx, is the integration, i.e. R : xn 7→ xn+1 of EndK(P1) are called the (first) Weyl algebra and the algebra of polynomial integro-differential n+1 for all n ∈ N. The algebra I1 is neither left nor right Noetherian and not a domain. Moreover, it contains infinite direct sums of nonzero left and right ideals, [1]. The algebra I1 is generated by the elements ∂, H := ∂x and R (since x = R H) that satisfy the defining relations (Proposition 2.2, [1]): ∂Z = 1, [H,Z ] =Z , [H, ∂] = −∂, H(1 −Z ∂) = (1 −Z ∂)H = 1 −Z ∂. The elements of the algebra I1, eij :=Z i ∂j −Z i+1 ∂j+1, i, j ∈ N, (23) satisfy the relations eijekl = δjkeil where δjk is the Kronecker delta function. Notice that eij = R i e00∂j. The algebra I1 =Li∈Z I1,i is a Z-graded algebra (I1,iI1,j ⊆ I1,i+j for all i, j ∈ Z) where D1R i =R i D1 D1 ∂iD1 = D1∂i I1,i =  27 if i > 0, if i = 0, if i < 0, (24) the algebra D1 := K[H]LLi∈N Keii is a commutative non-Noetherian subalgebra of I1, Heii = eiiH = (i + 1)eii for i ∈ N (notice that Li∈N Keii is the direct sum of non-zero ideals of D1); (R i D1)D1 ≃ D1, R i d 7→ d; D1 (D1∂i) ≃ D1, d∂i 7→ d, for all i ≥ 0 since ∂iR i = 1. Notice that the maps ·R i : D1 → D1R i, d 7→ dR i, and ∂i· : D1 → ∂iD1, d 7→ ∂id, have the same kernel Li−1 j=0 Kejj. Each element a of the algebra I1 is the unique finite sum a =Xi>0 a−i∂i + a0 +Xi>0 Z i ai + Xi,j∈N λij eij (25) where ak ∈ K[H] and λij ∈ K. This is the canonical form of the polynomial integro-differential The factor algebra I1/F is canonically isomorphic to the skew Laurent polynomial algebra B1 := operator [1]. The algebra I1 has the only proper ideal F = Li,j∈N Keij ≃ M∞(K) and F 2 = F . K[H][∂, ∂−1; τ ], τ (H) = H + 1, via ∂ 7→ ∂, R 7→ ∂−1, H 7→ H (where ∂±1α = τ ±1(α)∂±1 for all elements α ∈ K[H]). The algebra B1 is canonically isomorphic to the (left and right) localization A1,∂ of the Weyl algebra A1 at the powers of the element ∂ (notice that x = ∂−1H). Therefore, they have the common skew field of fractions, Frac(A1) = Frac(B1), the first Weyl skew field. The algebra I1 admits the involution ∗ over the field K: ∂∗ = R , R ∗ = ∂ and H ∗ = H, i.e. it is a K-algebra anti-isomorphism ((ab)∗ = b∗a∗) such that a∗∗ = a. Therefore, the algebra I1 is self-dual, i.e. it is isomorphic to its opposite algebra Iop 1 . As a result, the left and right properties of the algebra I1 are the same. Clearly, e∗ ij = eji for all i, j ∈ N, and so F ∗ = F . The next theorem describes the one-sided largest quotient rings of the algebra I1. Theorem 5.6 (Theorem 9.7, [2]) 1. Sr,0(I1) = I1T AutK(K[x]) and the largest regular right quotient ring Qr(I1) of I1 is the subalgebra of EndK(K[x]) generated by I1 and Sr,0(I1)−1 := {s−1 s ∈ Sr,0(I1)}. 2. Sl,0(I1) = Sr,0(I1)∗ and Sl,0(I1) 6= Sr,0(I1). 3. The rings Ql(I1) and Qr(I1) are not isomorphic. The next theorem describes the (two-sided) largest quotient ring of the algebra I1, it is tiny comparing with the one-sided largest quotient rings of the algebra I1. S0(I1)∗ = S0(I1) where ∗ is the involution of the algebra I1. Theorem 5.7 Let M := (K[H] + F )T AutK(K[x]). Then 1. S0(I1) = Sl,0(I1)T Sr,0(I1), S0(I1) is a proper subset of the sets Sl,0(I1) and Sr,0(I1), and 2. Sl,0(I1)T Sr,0(I1) = M and M is the set of regular elements of the algebra K[H] + F . 3. Let M0 := D1T AutK(K[x]). Then M0 ⊆ M, M = M0(1 + F )∗ = (1 + F )∗M0 and M0T(1 + F )∗ = (1 + F0)∗ where F0 :=Li∈N Keii. M0(1 + F0)` F0 = M0 ∪ {0} + F0, Qcl(D1) := M−1 0 M0(1 + F0)` F0 = 5. Q(I1) = S0(I1)−1I1 =Pi∈Z Qcl(D1)vi+F =Pi∈Z(M−1 0 M0∪{0})vi+F =Pi∈Z viQcl(D1)+ F =Pi∈Z vi(M−1 4. M0 is the set of regular elements of the commutative (non-Noetherian) algebra D1; D1 = 0 M0 ∪ {0}) + F where Qcl(D1) is the classical ring of fractions of the com- 0 M0 ∪ {0} + F0. 0 D1 = M−1 mutative ring D1 and M−1 6. Q(I1) $ Ql(I1) and Q(I1) $ Qr(I1). if i ≥ 1, if i = 0, if , i ≤ −1. R i 1 ∂i vi :=  28 Proof. 2. Recall that Sr,0(I1) = I1T AutK(K[x]) (Theorem 9.7.(2), [2]), Sl,0(I1) = Sr,0(I1)∗ (Theorem 9.7.(3), [2]), and if a ∈ I1\F and a ∈ I1T AutK(K[x]) then a ∈ Pi≤0 D1vi + F (Theorem 6.2.(1), [2]). Since (vi)∗ = v−i and d∗ = d for all elements d ∈ D1, Sl,0(I1)\ Sr,0(I1) = Sl,0(I1)\ Sl,0(I1)∗ ⊆ (K[H] + F )\ AutK(K[x]) ⊆ Sl,0(I1)\ Sl,0(I1)∗ = Sl,0(I1)\ Sr,0(I1) i.e. Sl,0(I1)T Sr,0(I1) = M. Let R be the set of regular elements of the algebra K[H] + F . Clearly, M ⊆ R. Let r ∈ R. Then r 6∈ F . By Proposition 6.1.(1), [2], indK[x](r) = 0. Then, by Theorem 6.2.(3), [2], rK[x] is a bijection iff rK[x] is an injection iff rK[x] is a surjection where rK[x] : K[x] → K[x], p 7→ rp. Therefore, to prove the equality M = R it suffices to show that rK[x] is an injection. Suppose that ker(rK[x]) 6= 0, we seek a contradiction. Since r 6∈ F , by Lemma 6.10, [2], there exists an idempotent f ∈ F such that ker(rK[x]) = im(fK[x]). In particular, rf = 0 but f 6= 0, a contradiction. Therefore, M = R. 1. Since Sl,0(I1) 6= Sr,0(I1) (Theorem 9.7.(3,4), [2]), the set S0(I1) is a proper subset of the sets Sl,0(I1) and Sr,0(I1). Clearly, S0(I1) ⊆ M = Sl,0(I1)\ Sr,0(I1) = Sl,0(I1)\ Sr,0(I1)∗ = (Sl,0(I1)\ Sl,0(I1)∗)∗ = Sl,0(I1)\ Sr,0(I1) = M. We have used the fact that Sl,0(I1) = Sr,0(I1)∗ (Theorem 9.7.(3), [2]). In view of statement 2 and the equality M∗ = M, to finish the proof of statement 1 it suffices to show that M is a left Ore set in I1, that is, for any elements a ∈ I1 and u = α + u1 ∈ M where α ∈ K[H] and u1 ∈ F , there exists an element b ∈ M such that bau−1 ∈ I1 (where the product bau−1 is taken in EndK(K[x]), i=−n αivi + f for some natural number recall that I1 ⊆ EndK(K[x])). The element a is a sum Pn that F M−1 ⊆ F and α · u−1 = α(α + u1)−1 ∈ (1 + F )∗ ⊆ M. Let b′ :=Qn For all elements β ∈ K[H], βvi = viτ i(β) where τ ∈ AutK(K[H]) and τ (H) = H + 1. Notice i=−n τ i(α). Then n where αi ∈ K[H]. b′au−1 = b′( αiviu−1 + f u−1) = nXi=−n nXi=−n αivi τ i(b′) α αu−1 + b′f u−1 ∈ I1, since τ i(b′) α ∈ K[H], αu−1 ∈ M and b′f u−1 ∈ b′ · F M−1 ⊆ b′ · F ⊆ F . Let I := {i ∈ N b ∗ xi = 0}. Since H ∗ xi = (i + 1)xi for all i ∈ N, we see that I = {i ∈ N i + 1 is a root of the polynomial 3. Since D1 = K[H] + F0 ⊆ K[H] + F , the inclusion M0 ⊆ M follows. It is obvious that b′ ∈ K[H]}. Then the element b := b′ +Pi∈I eii ∈ M and bau−1 ∈ I1. M0T(1 + F )∗ = (1 + F0)∗. To prove the equality M = M0(1 + F )∗ we have to show that each element u ∈ M is a product vw for some elements v ∈ M0 and w ∈ (1 + F )∗. Notice that u = α + f for some elements α ∈ K[H] and f ∈ F . Choose an element, say g ∈ F0, such that v := α + g ∈ M0 (see the proof of statement 1). Then u = α + g + f − g = v(1 + v−1(f − g)) = vw where w := 1 + v−1(f − g) ∈ (1 + F )∗ since M−1 0 F ⊆ F . Therefore, M = M0(1 + F )∗. Then M = (1 + F )∗M0 since, for all elements s ∈ M0, s(1 + F )∗s−1 ⊆ (1 + F )∗. 4. Let R be the set of regular elements of the ring D1. Then M0 ⊆ R. Let r ∈ R, in order to prove that the inverse inclusion holds, M0 ⊇ R, we have to show that r ∈ M0 or, equivalently, i.e. r ∗ Kxi ⊆ Kxi for all i ∈ N. If r 6∈ AutK(K[x]) then r ∗ Kxi = 0 for some i, and so reii = 0 since im(eii) = Kxi where eii ∈ D1, a contradiction. Therefore, r ∈ Autk(K[x]), and so M0 = R. that r ∈ AutK(K[x]). Notice that K[x] = Li∈N Kxi is the direct sum of r-invariant subspaces, Clearly, D1 ⊇ M0(1 + F0)` F0. To prove that the reverse inclusion holds we have to show that every element d ∈ D1\F0 belongs to the set M0(1 + F0). Notice that d = α + f for some elements 0 6= α ∈ K[H] and f ∈ F0. Choose an element, say g ∈ F0, such that u := α + g ∈ M0. 29 Then d = α + g + f − g = u(1 + u−1(f − g)) ∈ M0(1 + F0) since M−1 D1 = M0(1 + F0)` F0, and so D1 = M0 ∪ {0} + F0. Then M−1 0 D1 = M−1 M−1 0 F0 ⊆ F0. Therefore, 0 M(1 + F )` F0 = 0 M ∪ {0} + F0 since M−1F0 ⊆ F0. 5. By statements 3 and 4, for all i ∈ Z, M−1D1vi = M−1 ⊆ (M−1 0 (1 + F )∗D1vi = M−1 0 M0 ∪ {0} + F )vi ⊆ (M−1 0 (1 + F )∗(M0 ∪ {0} + F0)vi 0 M0 ∪ {0})vi + F. Therefore, S0(I1)−1I1 = M−1I1 =Xi∈Z M−1D1vi + M−1F =Xi∈Z 0 M0 ∪ {0} + F0)vi + F =Xi∈Z = Xi∈Z (M−1 (M−1 0 M0 ∪ {0})vi + F Qcl(D1)vi + F. Applying the involution ∗ to the inclusion M−1D1vi ⊆ (M−1 equalities M∗ = M, D∗ {0}) + F for all i ∈ Z. Then 1 = D1 and v∗ i = v−i we obtains the inclusion v−iD1M−1 ⊆ v−i(M−1 0 M0 ∪ {0})vi + F and using the 0 M0 ∪ I1S0(I1)−1 = I1M−1 =Xi∈Z viD1M−1 + F M−1 =Xi∈Z 0 M0 ∪ {0} + F ) + F =Xi∈Z = Xi∈Z vi(M−1 vi(M−1 0 M0 ∪ {0}) + F viQcl(D1) + F. 6. The inclusion Q(I1) ⊆ Ql(I1) (resp. Q(I1) ⊆ Qr(I1)) is a proper inclusion by statement 5 and Theorem 9.7, [2]. Another way to prove this is as follows. Suppose that Q(I1) = Ql(I1), we seek a contradiction. The algebra I1 has the involution ∗, and so it is isomorphic to its opposite algebra of I1. Therefore, Q(I1) = Qr(I1), but the rings Ql(I1) and Qr(I1) are not isomorphic, by Theorem 9.7.(4), [2], a contradiction. (cid:3) Proposition 5.8 1. Assl(I1) = Assr(I1) = Ass(I1) = {0, F } and max.Assl(I1) = max.Assr(I1) = max.Ass(I1) = {F }. 2. Sl,F (I1) = Sr,F (I1) = SF (I1) = I1\F and Ql,F (I1) = Qr,F (I1) = QF (I1) = Frac(B1) = Frac(A1). 3. max.Denl(I1) = max.Denr(I1) = max.Den(I1) = {I1\F }. Proof. 2. Let πF : I1 → I1/F = B1, a 7→ a + F . Then π−1 F (B1\{0}) = I1\F is a multiplicative set of the algebra I1 with assl(I1\F ) = assr(I1\F ) = F since for all i, j ∈ N, ∂i+1eij = 0, eijR i+1 = 0 and F = Li,j∈N Keij is the only proper ideal of the algebra I1 and the algebra I1/F = B1 is a domain. Since B1 is a Noetherian domain and Frac(A1) = Frac(B1) = (B1\{0})−1B1 = B1(B1\{0})−1, we see that, by the universal property of localization, Frac(I1) = (I1\F )−1I1 = I1(I1\F )−1 and so I1\F is a left and right denominators set of the algebra I1 with ass(I1\F ) = F . Now, statement 2 follows from Theorem 3.9.(1). 1. Statement 1 follows from statement 2. 3. Statement 3 follows from statements 1 and 2. (cid:3) Noetherian commutative rings. Example. If R is a commutative ring with a single minimal prime ideal p (eg, R is a commutative domain) then Ass(R) = {0}, S0(R) = R\p and Q(R) = Rp, the localization of the ring R at the prime ideal p. Lemma 5.9 Let R be a commutative ring, CR be its set of regular elements and S ∈ Ore(R). Then 30 1. p(S) = ass(S). 2. CRS ∈ Den(R, ass(S)). Proof. 1. The equality follows from the definition of the ideal p(S), see (18) and (19). 2. Obvious. (cid:3) primes with s ≥ 2. Then S0 = S0(R) = R\Ss Let R be a Noetherian commutative ring and Min(R) = {p1, . . . , ps} be its set of minimal i=1 pi is the set of regular elements of the ring R (i.e. the set of non-zero-divisors of R) and S−1 i=1 Ri is the direct product of local commutative artinian rings (Ri, mi) where mi is the maximal ideal of the ring Ri = Rpi, the localization of the ring R at pi. For each non-empty subset I of the set {1, . . . , s}, let the ring homomorphism 0 R, r 7→ r/1 i where σI : R → RI := Qi∈I Ri be the composition of the ring monomorphism σ : R → S−1 0 R → RI . The group of units R∗ 0 R ≃Qs I of the ring RI is equal to Qi∈I R∗ i = Ri\mi is the group of units of the ring Ri. For each number i = 1, . . . , s, p′ and the projection S−1 R∗ is an ideal of the ring S−1 i := mi ×Qj6=i Rj 0 R. Then p′′ i := σ−1(p′ i) is an ideal of the ring R. Proposition 5.10 Let R be a Noetherian commutative ring, {p1, . . . , ps} be the set of its minimal prime ideals, s ≥ 2 and {p′′ s } be as above. Then 1 , . . . , p′′ 1. Ass(R) := {0}S{a(I) :=Ti∈I p′′ i ∅ $ I $ {1, . . . , s}}. 2. For each proper subset I of {1, . . . , s}, Sa(I)(R) = σ−1 I (R∗ RI where CI := {1, . . . , s}\I. I ×RCI ) = R\Si∈I pi and Qa(I)(R) ≃ 3. S0(R) = R\Ss i=1 pi ⊆ Sa(R) for all a ∈ Ass(R). 4. Sa(I)(R) ⊆ Sa(I)(R) iff I ⊇ J where I and J are proper subsets of {1, . . . , s}. 5. If I and J are proper subsets of {1, . . . , s} such that I ⊇ J then, by statement 4 and the universal property of localization, there is the unique ring homomorphism Qa(I)(R) = Qi∈I RI → Qa(J)(R) =Qj∈J Rj which is necessarily the projection ontoQj∈J Rj inQi∈I Ri. i , i = 1, . . . , s}, {Qp′′ i (R) = Rpi i = 1, . . . , s} is the set of maximal 6. max.Ass(R) = {p′′ localizations of the ring R. Proof. 1 -- 3. Clearly, S0 := S0(R) = CR where CR = R\Ss i=1 pi is the set of regular elements of the ring R. By Lemma 5.9.(2), S0 ⊆ Sa(R) for all a ∈ Ass(R) and statement 3 follows. We identify the ring R with its image under the ring monomorphisms R → S−1 i=1 Ri, r 7→ r/1. By statement 3 and Proposition 3.8.(2), the map Ass(R) → Ass(S−1 0 R), a 7→ S−1a, is a bijection with the inverse map 0 R =Qs and the map b 7→ R ∩ b, {Sa(R) a ∈ Ass(R)} → {Sb(S−1 0 R) b ∈ Ass(S−1 0 R)}, SR∩b(R) 7→ Sb(S−1 0 R), (26) (27) (28) is a bijection and where SR∩b(R) = σ−1 R∩b(Sb(S−1 0 R)) σR∩b : R → QR∩b(R) = SR∩b(R)−1R ≃ Sb(S−1 0 R)−1S−1 0 R. 0 R) ⊇ R since RI = ass(σ−1 0 R) = {0}S{RI ∅ & I & {1, . . . , s}}. Let R be the RHS of the equality. Then Claim: Ass(S−1 Ass(S−1 CI )) where CI is the complement of the proper set I in {1, . . . , s}. To prove that the reverse inclusion holds, i.e. Ass(S−1 0 R) ⊆ R, we have to show that, for each multiplicatively closed subset S of S−1 0 R that does not entirely consists of non-zero- divisors, ass(S) = RI for some proper subset I of {1, . . . , s}. For each element r ∈ R, the subset CI (R∗ 31 0 R = Qs of {1, . . . , s}, supp(r) := {i πi(r) ∈ R∗ where πi : R → S−1 i and i ∈ {1, . . . , s}} is called the support of the element r j=1 Rj → Ri. For all elements s, t ∈ S, supp(st) = supp(s) ∩ supp(t). Let supp(S) := Ts∈S supp(s). Clearly, there exists an element, say s ∈ S, such that supp(s) = Supp(S) (eg, s is the product of all elements of S with all possible distinct supports). Replacing s with sn for some natural number we may additionally assume that πi(s) = 0 for all elements j 6∈ I i = 0 for all i = 1, . . . , s). Then clearly ass(S) = ass({si}i∈N) = RCsupp(S). (take n such that mn This finishes the proof of the Claim. Statement 1 follows from the Claim and (26). For every proper subset I of {1, . . . , s}, R ∩ RI = CI ×RI . Then, by (27), SR∩RI (R) = σ−1 R∩RI (SRI (S−1 0 R)) = a(CI), σI = σR∩RCI and SRI (S−1 σ−1 CI (R∗ CI × RI ) and, by (28), QR∩RI (R) ≃ SRI (S−1 0 R) = R∗ statement 2 it remains to show that SR∩RI (R) = R\Si∈I pi for each proper subset I of the set {1, . . . , s}. An element s ∈ R belongs to the set Sa(CI) = SR∩RI (R) = σ−1 σ(r) ∈ R∗ CI × RI ) iff CI (R∗ 0 R)−1(S−1 0 R) ≃ RCI . To finish the proof of CI × RI iff s 6∈Si∈CI pi iff s ∈ R\Si∈CI pi. 4. Statement 4 follows from statement 2. 5. Statement 5 follows from statement 2. 6. Statement 6 follows from statements 2, 4 and 5. (cid:3) References [1] V. V. Bavula, The algebra of integro-differential operators on a polynomial algebra, Journal of the London Math. Soc. (Arxiv:math.RA: 0912.0723). [2] V. V. Bavula, The algebra of integro-differential operators on an affine line and its modules, Arxiv.RA:1011.2997. [3] A. V. Jategaonkar, Localization in Noetherian Rings, London Mathematical Society, LNS 98, Cambridge Univ. Press, London, 1986. [4] S. Jondrup, On finitely generated flat modules, II, Math. Scand. 27 (1970) 105 -- 112. [5] J. C. McConnell and J. C. Robson, Noncommutative Noetherian rings, Wiley, Chichester, 1987. [6] B. Stenstrom, Rings of Quotients, Springer-Verlag, Berlin Heidelberg New York, 1975. Department of Pure Mathematics University of Sheffield Hicks Building Sheffield S3 7RH UK email: [email protected] 32
1111.6035	1	1111	2011-11-09T04:30:33	Basis of Representation of Universal Algebra	[ "math.RA" ]	We say that there is a representation of the universal algebra B in the universal algebra A if the set of endomorphisms of the universal algebra A has the structure of universal algebra B. Therefore, the role of representation of the universal algebra is similar to the role of symmetry in geometry and physics. Morphism of the representation is the mapping that conserves the structure of the representation. Exploring of morphisms of the representation leads to the concepts of generating set and basis of representation. The set of automorphisms of the representation of the universal algebra forms the group. Twin representations of this group in basis manifold of the representation are called active and passive representations. Passive representation in basis manifold is underlying of concept of geometric object and the theory of invariants of the representation of the universal algebra.	math.RA	math	Basis of Representation of Universal Algebra Aleks Kleyn Abstract. We say that there is a representation of the universal algebra B in the universal algebra A if the set of endomorphisms of the universal algebra A has the structure of universal algebra B. Therefore, the role of representation of the universal algebra is similar to the role of symmetry in geometry and physics. Morphism of the representation is the mapping that conserves the structure of the representation. Exploring of morphisms of the representation leads to the concepts of generating set and basis of representation. The set of automorphisms of the representation of the universal algebra forms the group. Twin representations of this group in basis manifold of the representation are called active and passive representations. Passive representation in basis man- ifold is underlying of concept of geometric object and the theory of invariants of the representation of the universal algebra. Contents 1. Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. Generating Set of Representation . . . . . . . . . . . . . . . . . . . . . 4. Basis of representation . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Basis Manifold of Representation . . . . . . . . . . . . . . . . . . . . . 6. Geometric Object of Representation of Universal Algebra . . . . . . . 7. Examples of Basis of Representation of Universal Algebra . . . . . . . 7.1. Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2. Representation of Group on the Set . . . . . . . . . . . . . . . . . . . 8. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. Special Symbols and Notations . . . . . . . . . . . . . . . . . . . . . . 1 2 2 11 15 17 20 20 20 20 21 22 1. Preface This paper is based on the chapter [3]-3. The role of representation of the universal algebra is similar to the role of sym- metry in geometry and physics. In both cases we study the structured set of trans- formations; knowledge of the structure of this set gives us additional information about the object under study. Aleks [email protected]. http://sites.google.com/site/AleksKleyn/. http://arxiv.org/a/kleyn_a_1 . http://AleksKleyn.blogspot.com/ . 1 2 Aleks Kleyn We consider the theory of representations of universal algebra as an extension of the theory of universal algebra. Any algebraic structure assumes a set of mappings preserving this structure. Mapping that preserves the structure of the representa- tion of universal algebra, is called a morphism of representations. Automorphism of representation is special case of morphism of representations. The study of automorphisms representation is directly related to the necessity to answer the question what is the structure of the generating set of representation. If the representation has minimal generating set, then such set is called basis of the represntation. Automorphism of representation maps basis into basis. The set of automorphisms of representation forms a group. Twin representations of this group in basis manifold of the representation are called active and passive representations. Passive representation in basis manifold is underlying of concept of geometric object and the theory of invariants of the representation of the universal algebra. 2. Conventions Convention 2.1. In [4], an arbitrary operation of algebra is denoted by letter ω, and Ω is the set of operations of some universal algebra. Correspondingly, the universal algebra with the set of operations Ω is denoted as Ω-algebra. Similar notations we see in [2] with small difference that an operation in the algebra is denoted by letter f and F is the set of operations. I preferred first case of notations because in this case it is easier to see where I use operation. (cid:3) Convention 2.2. Let A be Ω1-algebra. Let B be Ω2-algebra. Notation A ∗ / B means that there is representation of Ω1-algebra A in Ω2-algebra B. (cid:3) Without a doubt, the reader may have questions, comments, objections. I will appreciate any response. 3. Generating Set of Representation Definition 3.1. Let f : A → ∗M be representation of Ω1-algebra A in Ω2-algebra M . The set N ⊂ M is called stable set of representation f , if f (a)(m) ∈ N for each a ∈ A, m ∈ N . (cid:3) We also say that the set M is stable with respect to the representation f . Theorem 3.2. Let f : A → ∗M be representation of Ω1-algebra A in Ω2-algebra M . Let set N ⊂ M be subalgebra of Ω2-algebra M and stable set of representation f . Then there exists representation fN : A → ∗N such that fN (a) = f (a)N . Representation fN is called subrepresentation of representation f . / Basis of Representation of Universal Algebra 3 Proof. Let ω1 be n-ary operation of Ω1-algebra A. Then for each a1, ..., an ∈ A and each b ∈ N (fN (a1)...fN (an)ω1)(b) = (f (a1)...f (an)ω1)(b) = f (a1...anω1)(b) = fN (a1...anω1)(b) Let ω2 be n-ary operation of Ω2-algebra M . Then for each b1, ..., bn ∈ N and each a ∈ A fN (a)(b1)...fN (a)(bn)ω2 = f (a)(b1)...f (a)(bn)ω2 = f (a)(b1...bnω2) = fN (a)(b1...bnω2) We proved the statement of theorem. (cid:3) From the theorem 3.2, it follows that if fN is subrepresentation of representation f , then the mapping (id : A → A, idN : N → M ) is morphism of representations. Theorem 3.3. The set1 Bf of all subrepresentations of representation f generates a closure system on Ω2-algebra M and therefore is a complete lattice. Proof. Let (Kλ)λ∈Λ be the set off subalgebras of Ω2-algebra M that are stable with respect to representation f . We define the operation of intersection on the set Bf according to rule \ fKλ = f∩Kλ We defined the operation of intersection of subrepresentations properly. ∩Kλ is subalgebra of Ω2-algebra M . Let m ∈ ∩Kλ. For each λ ∈ Λ and for each a ∈ A, f (a)(m) ∈ Kλ. Therefore, f (a)(m) ∈ ∩Kλ. Therefore, ∩Kλ is the stable set of representation f . (cid:3) We denote the corresponding closure operator by J(f ). Thus J(f, X) is the intersection of all subalgebras of Ω2-algebra M containing X and stable with respect to representation f . Theorem 3.4. Let2 f : A → ∗M be representation of Ω1-algebra A in Ω2-algebra M . Let X ⊂ M . Define a subset Xk ⊂ M by induction on k. X0 = X x ∈ Xk => x ∈ Xk+1 x1 ∈ Xk, ..., xn ∈ Xk, ω ∈ Ω2(n) => x1...xnω ∈ Xk+1 x ∈ Xk, a ∈ A => f (a)(x) ∈ Xk+1 Then ∞ [ k=0 Xk = J(f, X) 1This definition is similar to definition of the lattice of subalgebras ([4], p. 79, 80) 2The statement of theorem is similar to the statement of theorem 5.1, [4], p. 79. 4 Aleks Kleyn Proof. If we put U = ∪Xk, then by definition of Xk, we have X0 ⊂ J(f, X), and if Xk ⊂ J(f, X), then Xk+1 ⊂ J(f, X). By induction it follows that Xk ⊂ J(f, X) for all k. Therefore, (3.1) U ⊂ J(f, X) If a ∈ U n, a = (a1, ..., an), where ai ∈ Xki, and if k = max{k1, ..., kn}, then a1...anω ∈ Xk+1 ⊂ U . Therefore, U is subalgebra of Ω2-algebra M . If m ∈ U , then there exists such k that m ∈ Xk. Therefore, f (a)(m) ∈ Xk+1 ⊂ U for any a ∈ A. Therefore, U is stable set of the representation f . Since U is subalgebra of Ω2-algebra M and is a stable set of the representation f , then subrepresentation fU is defined. Therefore, (3.2) J(f, X) ⊂ U From (3.1), (3.2), it follows that J(f, X) = U . (cid:3) Definition 3.5. J(f, X) is called subrepresentation generated by set X, and X is a generating set of subrepresentation J(f, X). In particular, a generat- ing set of representation f is a subset X ⊂ M such that J(f, X) = M . (cid:3) It is easy to see that the definition of generating set of representation does not depend on whether representation is effective or not. For this reason hereinafter we will assume that the representation is effective and we will use convention for effective T ⋆-representation in remark [3]-2.1.9. We also will use notation R ◦ m = R(m) for image of m ∈ M under the endomorphism of effective representation. According to the definition of product of mappings, for any endomorphisms R, S the following equation is true (3.3) (R ◦ S) ◦ m = R ◦ (S ◦ m) The equation (3.3) is associative law for ◦ and allows us to write expression without brackets. From theorem 3.4, it follows next definition. R ◦ S ◦ m Definition 3.6. Let X ⊂ M . For each x ∈ J(f, X) there exists Ω2-word defined according to following rules. (1) If m ∈ X, then m is Ω2-word. (2) If m1, ..., mn are Ω2-words and ω ∈ Ω2(n), then m1...mnω is Ω2-word. (3) If m is Ω2-word and a ∈ A, then am is Ω2-word. Ω2-word w(f, X, m) represents given element m ∈ J(f, X). We will identify an element m ∈ J(f, X) and corresponding it Ω2-word using equation Similarly, for an arbitrary set B ⊂ J(f, X) we consider the set of Ω2-words3 w(f, X, B) = {w(f, X, m) : m ∈ B} m = w(f, X, m) 3The expression w(f, X, m) is a special case of the expression w(f, X, B), namely w(f, X, {m}) = {w(f, X, m)} 5 (cid:3) Basis of Representation of Universal Algebra We also use notation w(f, X, B) = (w(f, X, m), m ∈ B) Denote w(f, X) the set of Ω2-words of representation J(f, X). Theorem 3.7. Endomorphism R of representation generates the mapping of Ω2-words f : A → ∗M w[f, X, R] : w(f, X) → w(f, X ′) X ⊂ M X ′ = R ◦ X such that (1) If m ∈ X, m′ = R ◦ m, then (2) If w[f, X, R](m) = m′ m1, ..., mn ∈ w(f, X) m′ 1 = w[f, X, R](m1) ... m′ n = w[f, X, R](mn) then for operation ω ∈ Ω2(n) holds w[f, X, R](m1...mnω) = m′ 1...m′ nω (3) If then m ∈ w(f, X) m′ = w[f, X, R](m) a ∈ A w[f, X, R](am) = am′ Proof. Statements (1), (2) of the theorem are true by definition of the endomorpism R. Because r = id, the statement (3) of the theorem follows from the equation [3]- (2.2.4). (cid:3) Remark 3.8. Let R be endomorphism of representation f . Let m ∈ J(f, X) m′ = R ◦ m X ′ = R ◦ X The theorem 3.7 states that m′ ∈ Jf (X ′) . The theorem 3.7 also states that Ω2-word representing m relative X and Ω2-word representing m′ relative X ′ are generated according to the same algorithm. This allows considering of the set of Ω2-words w(f, X ′, m′) as mapping W (f, X, m) : X ′ → w(f, X ′, m′) W (f, X, m)(X ′) = W (f, X, m) ◦ X ′ such that, if for certain endomorphism R then X ′ = R ◦ X m′ = R ◦ m W (f, X, m) ◦ X ′ = w(f, X ′, m′) = m′ 6 Aleks Kleyn The mapping W (f, X, m) is called coordinates of element m relative to set X. Similarly, we consider coordinates of a set B ⊂ J(f, X) relative to the set X W (f, X, B) = {W (f, X, m) : m ∈ B} = (W (f, X, m), m ∈ B) Denote W (f, X) the set of coordinates of representation J(f, X). (cid:3) Theorem 3.9. There is a structure of Ω2-algebra on the set of coordinates W (f, X). Proof. Let ω ∈ Ω2(n). Then for any m1, ..., mn ∈ J(f, X) , we assume (3.4) W (f, X, m1)...W (f, X, mn)ω = W (f, X, m1...mnω) According to the remark 3.8, (3.5) (W (f, X, m1)...W (f, X, mn)ω) ◦ X = W (f, X, m1...mnω) ◦ X = w(f, X, m1...mnω) follows from the equation (3.4). According to rule (2) of the definition 3.6, from the equation (3.5), it follows that (3.6) (W (f, X, m1)...W (f, X, mn)ω) ◦ X = w(f, X, m1)...w(f, X, mn)ω = (W (f, X, m1) ◦ X)...(W (f, X, mn) ◦ X)ω From the equation (3.6), it follows that the operation ω defined by the equation (3.4) on the set of coordinates is defined properly. (cid:3) Theorem 3.10. There exists the representation of Ω1-algebra A in Ω2-algebra W (f, X). Proof. Let a ∈ A. Then for any m ∈ J(f, X) we assume (3.7) aW (f, X, m) = W (f, X, am) According to the remark 3.8, (3.8) (aW (f, X, m)) ◦ X = W (f, X, am) ◦ X = w(f, X, am) follows from the equation (3.7). According to rule (3) of the definition 3.6, from the equation (3.8), it follows that (3.9) (aW (f, X, m)) ◦ X = aw(f, X, m) = a(W (f, X, m) ◦ X) From the equation (3.9), it follows that the representation (3.7) of Ω1-algebra A in Ω2-algebra W (f, X) is defined properly. (cid:3) Theorem 3.11. Let f : A → ∗M be representation of Ω1-algebra A in Ω2-algebra M . For given sets X ⊂ M , X ′ ⊂ M , let map R1 : X → X ′ Basis of Representation of Universal Algebra 7 agree with the structure of representation f , i. e. ω ∈ Ω2(n) x1, ..., xn, x1...xnω ∈ X, R1(x1...xnω) ∈ X ′ => R1(x1...xnω) = R1(x1)...R1(xn)ω x ∈ X, a ∈ A, R1(ax) ∈ X ′ => R1(ax) = aR1(x) Consider the mapping of Ω2-words w[f, X, R1] : w(f, X) → w(f, X ′) that satisfies conditions (1), (2), (3) of the theorem 3.7 and such that There exists unique endomorphism of Ω2-algebra M x ∈ X => w[f, X, R1](x) = R1(x) R : M → M defined by rule which is the morphism of representations J(f, X) and J(f, X ′). R ◦ m = w[f, X, R1](w(f, X, m)) Proof. We prove the theorem by induction over complexity of Ω2-word. If w(f, X, m) = m, then m ∈ X. According to condition (1) of theorem 3.7, R ◦ m = w[f, X, R1](w(f, X, m)) = w[f, X, R1](m) = R1(m) Therefore, mappings R and R1 coinside on the set X, and the mapping R agrees with structure of representation f . Let ω ∈ Ω2(n). Let the mapping R be defined for m1, ..., mn ∈ J(f, X). Let w1 = w(f, X, m1) ... wn = w(f, X, mn) If m = m1...mnω, then according to rule (2) of definition 3.6, According to condition (2) of theorem 3.7, w(f, X, m) = w1...wnω R ◦ m = w[f, X, R1](w(f, X, m)) = w[f, X, R1](w1...wnω) = w[f, X, R1](w1)...w[f, X, R1](wn)ω = (R ◦ m1)...(R ◦ mn)ω Therefore, the mapping R is endomorphism of Ω2-algebra M . Let the mapping R be defined for m1 ∈ J(f, X), w1 = w(f, X, m1). Let a ∈ A. If m = am1, then according to rule (3) of definition 3.6, According to condition (3) of theorem 3.7, w(f, X, am1) = aw1 R ◦ m = w[f, X, R1](w(f, X, m)) = w[f, X, R1](aw1) = aw[f, X, R1](w1) = aR ◦ m1 From equation [3]-(2.2.4), it follows that the mapping R is morphism of the repre- sentation f . 8 Aleks Kleyn The statement that the endomorphism R is unique and therefore this endomor- phism is defined properly follows from the following argument. Let m ∈ M have different Ω2-words relative the set X, for instance (3.10) m = x1...xnω = ax Because R is endomorphism of representation, then, from the equation (3.10), it follows that (3.11) R ◦ m = R ◦ (x1...xnω) = (R ◦ x1)...(R ◦ xn)ω = R ◦ (ax) = aR ◦ x From the equation (3.11), it follows that (3.12) R ◦ m = (R ◦ x1)...(R ◦ xn)ω = aR ◦ x From equations (3.10), (3.12), it follows that the equation (3.10) is preserved un- der the mapping. Therefore, the image of m does not depend on the choice of coordinates. (cid:3) Remark 3.12. The theorem 3.11 is the theorem of extension of mapping. The only statement we know about the set X is the statement that X is generating set of the representation f . However, between the elements of the set X there may be rela- tionships generated by either operations of Ω2-algebra M , or by transformation of representation f . Therefore, any mapping of set X, in general, cannot be extended to an endomorphism of representation f .4 However, if the mapping R1 is coordi- nated with the structure of representation on the set X, then we can construct an extension of this mapping and this extension is endomorphism of representation f . (cid:3) Definition 3.13. Let X be the generating set of the representation f . Let R be the endomorphism of the representation f . The set of coordinates W (f, X, R ◦ X) is called coordinates of endomorphism of representation. (cid:3) Definition 3.14. Let X be the generating set of the representation f . Let R be the endomorphism of the representation f . Let m ∈ M . We define superposition of coordinates of the representation f and the element m as coordinates defined according to rule (3.13) W (f, X, m) ◦ W (f, X, R ◦ X) = W (f, X, R ◦ m) Let Y ⊂ M . We define superposition of coordinates of the representation f and the set Y according to rule (3.14) W (f, X, Y ) ◦ W (f, X, R ◦ X) = (W (f, X, m) ◦ W (f, X, R ◦ X), m ∈ Y ) (cid:3) Theorem 3.15. Endomorphism R of representation f : A → ∗M generates the mapping of coordinates of representation (3.15) W [f, X, R] : W (f, X) → W (f, X) 4In the theorem 4.7, requirements to generating set are more stringent. Therefore, the theorem 4.7 says about extension of arbitrary mapping. A more detailed analysis is given in the remark 4.9. Basis of Representation of Universal Algebra 9 such that (3.16) W (f, X, m) → W [f, X, R] ⋆ W (f, X, m) = W (f, X, R ◦ m) W [f, X, R] ⋆ W (f, X, m) = W (f, X, m) ◦ W (f, X, R ◦ X) Proof. According to the remark 3.8, we consider equations (3.13), (3.15) relative to given generating set X. The word (3.17) W (f, X, m) ◦ X = w(f, X, m) corresponds to coordinates W (f, X, m); the word (3.18) W (f, X, R ◦ m) ◦ X = w(f, X, R ◦ m) corresponds to coordinates W (f, X, R ◦ m). Therefore, in order to prove the theo- rem, it is sufficient to show that the mapping W [f, X, R] corresponds to mapping w[f, X, R]. We prove this statement by induction over complexity of Ω2-word. If m ∈ X, m′ = R ◦ m, then, according to equations (3.17), (3.18), mappings W [f, X, R] and w[f, X, R] are coordinated. Let for m1, ..., mn ∈ X mappings W [f, X, R] and w[f, X, R] be coordinated. Let ω ∈ Ω2(n). According to the theorem 3.9 (3.19) W (f, X, m1...mnω) = W (f, X, m1)...W (f, X, mn)ω Because R is endomorphism of Ω2-algebra M , then from the equation (3.19), it follows that (3.20) W (f, X, R ◦ (m1...mnω)) = W (f, X, (R ◦ m1)...(R ◦ mn)ω) = W (f, X, R ◦ m1)...W (f, X, R ◦ mn)ω From equations (3.19), (3.20) and the statement of induction, it follows that the mappings W [f, X, R] and w[f, X, R] are coordinated for m = m1...mnω. Let for m1 ∈ M mappings W [f, X, R] and w[f, X, R] are coordinated. Let a ∈ A. According to the theorem 3.10 (3.21) W (f, X, am1) = aW (f, X, m1) Because R is endomorphism of representation f , then, from the equation (3.21), it follows that (3.22) W (f, X, R ◦ (am1)) = W (f, X, a R ◦ m1) = aW (f, X, R ◦ m1) From equations (3.21), (3.22) and the statement of induction, it follows that map- pings W [f, X, R] and w[f, X, R] are coordinated for m = am1. (cid:3) Corollary 3.16. Let X be the generating set of the representation f . Let R be the endomorphism of the representation f . The mapping W [f, X, R] is endomorphism of representation of Ω1-algebra A in Ω2-algebra W (f, X). (cid:3) Hereinafter we will identify mapping W [f, X, R] and the set of coordinates W (f, X, R ◦ X). Theorem 3.17. Let X be the generating set of the representation f . Let R be the endomorphism of the representation f . Let Y ⊂ M . Then (3.23) (3.24) W (f, X, Y ) ◦ W (f, X, R ◦ X) = W (f, X, R ◦ Y ) W [f, X, R] ⋆ W (f, X, Y ) = W (f, X, R ◦ Y ) 10 Aleks Kleyn Proof. The equation (3.23) follows from the equation R ◦ Y = (R ◦ m, m ∈ Y ) as well from equations (3.13), (3.14). The equation (3.24) is corollary of equations (3.23), (3.16). (cid:3) Theorem 3.18. Let X be the generating set of the representation f . Let R, S be the endomorphisms of the representation f . Then (3.25) (3.26) W (f, X, S ◦ X) ◦ W (f, X, R ◦ X) = W (f, X, R ◦ S ◦ X) W [f, X, R] ⋆ W [f, X, S] = W [f, X, R ◦ S] Proof. The equation (3.25) follows from the equation (3.23), if we assume Y = S◦X. The equation (3.26) follows from the equation (3.25) and chain of equations (W [f, X, R] ⋆ W [f, X, S]) ⋆ W (f, X, Y ) = W [f, X, R] ⋆ (W [f, X, S] ⋆ W (f, X, Y )) = (W (f, X, Y ) ◦ W (f, X, S ◦ X)) ◦ W (f, X, R ◦ X) = W (f, X, Y ) ◦ (W (f, X, S ◦ X) ◦ W (f, X, R ◦ X)) = W (f, X, Y ) ◦ W (f, X, R ◦ S ◦ X) = W (f, X, R ◦ S) ⋆ W (f, X, Y ) The concept of superposition of the coordinates is very simple and resembles a kind of Turing machine. If element m ∈ M has form either (cid:3) or m = m1...mnω m = am1 then we are looking for the coordinates of elements mi to substitute them in an appropriate expression. As soon as an element m ∈ M belongs to the generating set of Ω2-algebra M , we choose the coordinates of the corresponding element of the second factor. Therefore, we require that the second factor in the superposition has been the set of coordinates of the image of the generating set X. We can generalize the definition of the superposition of coordinates and assume that one of the factors is a set of Ω2-words. Accordingly, the definition of the superposition of coordinates has the form w(f, X, Y ) ◦ W (f, X, R ◦ X) = W (f, X, Y ) ◦ w(f, X, R ◦ X) = w(f, X, R ◦ Y ) The following forms of writing an image of the set Y under endomorphism R are equivalent. (3.27) R ◦ Y = W (f, X, Y ) ◦ (R ◦ X) = W (f, X, Y ) ◦ (W (f, X, R ◦ X) ◦ X) From equations (3.23), (3.27), it follows that (3.28) (W (f, X, Y ) ◦ W (f, X, R ◦ X)) ◦ X = W (f, X, Y ) ◦ (W (f, X, R ◦ X) ◦ X) The equation (3.28) is associative law for composition and allows us to write ex- pression W (f, X, Y ) ◦ W (f, X, R ◦ X) ◦ X without brackets. Basis of Representation of Universal Algebra 11 Consider equation (3.25), where we see change in the order of endomorphisms in a superposition of the coordinates. This equation also follows from the chain of equations, where we can immediately see when order of endomorphisms changes W (f, X, m) ◦ W (f, X, R ◦ S ◦ X) ◦ X (3.29) = (R ◦ S) ◦ m = R ◦ (S ◦ m) = R ◦ ((W (f, X, m) ◦ W (f, X, S ◦ X)) ◦ X) = W (f, X, m) ◦ W (f, X, S ◦ X) ◦ W (f, X, R ◦ X) ◦ X From the equation (3.29), it follows that coordinates of endomorphism act over coordinates of element of Ω2-algebra M from the right. Definition 3.19. Let X ⊂ M be generating set of representation Let the mapping f : A → ∗M H : M → M be endomorphism of the representation f . Let the set X ′ = H ◦ X be the image of the set X under the mapping H. Endomorphism H of representation f is called regular on generating set X, if the set X ′ is the generating set of represen- tation f . Otherwise, endomorphism H of representation f is called singular on generating set X, (cid:3) Definition 3.20. Endomorphism of representation f is called regular, if it is regular on every generating set. (cid:3) Theorem 3.21. Automorphism R of representation f : A → ∗M is regular endomorphism. Proof. Let X be generating set of representation f . Let X ′ = R ◦ X. According to theorem 3.7 endomorphism R forms the map of Ω2-words w[f, X, R]. Let m′ ∈ M . Since R is automorphism, then there exists m ∈ M , R ◦ m = m′. According to definition 3.6, w(f, X, m) is Ω2-word, representing m relative to generating set X. According to theorem 3.7, w(f, X ′, m′) is Ω2-word, representing of m′ relative to generating set X ′ w(f, X ′, m′) = w[f, X, R](w(f, X, m)) Therefore, X ′ is generating set of representation f . According to definition 3.20, automorphism R is regular. (cid:3) 4. Basis of representation Definition 4.1. If the set X ⊂ M is generating set of representation f , then any set Y , X ⊂ Y ⊂ M also is generating set of representation f . If there exists minimal set X generating the representation f , then the set X is called basis of representation f . (cid:3) Theorem 4.2. The generating set X of representation f is basis iff for any m ∈ X the set X \ {m} is not generating set of representation f . 12 Aleks Kleyn Proof. Let X be generating set of representation f . Assume that for some m ∈ X there exist Ω2-word (4.1) w = w(f, X \ {m}, m) Consider element m′ such that it has Ω2-word (4.2) w′ = w(f, X, m′) that depends on m. According to the definition 3.6, any occurrence of m into Ω2- word w′ can be substituted by the Ω2-word w. Therefore, the Ω2-word w′ does not depend on m, and the set X \ {m} is generating set of representation f . Therefore, X is not basis of representation f . (cid:3) Remark 4.3. The proof of the theorem 4.2 gives us effective method for constructing the basis of the representation f . Choosing an arbitrary generating set, step by step, we remove from set those elements which have coordinates relative to other elements of the set. If the generating set of the representation is infinite, then this construction may not have the last step. If the representation has finite generating set, then we need a finite number of steps to construct a basis of this representation. As noted by Paul Cohn in [4], p. 82, 83, the representation may have inequivalent bases. For instance, the cyclic group of order six has bases {a} and {a2, a3} which we cannot map one into another by endomorphism of the representation. (cid:3) Remark 4.4. We write a basis also in following form X = (x, x ∈ X) If basis is finite, then we also use notation X = (xi, i ∈ I) = (x1, ..., xn) (cid:3) Remark 4.5. We introduced Ω2-word of x ∈ M relative generating set X in the definition 3.6. From the theorem 4.2, it follows that if the generating set X is not a basis, then a choice of Ω2-word relative generating set X is ambiguous. However, even if the generating set X is a basis, then a representation of m ∈ M in form of Ω2-word is ambiguous. If m1, ..., mn are Ω2-words, ω ∈ Ω2(n) and a ∈ A, then5 (4.3) a(m1...mnω) = (am1)...(amn)ω It is possible that there exist equations related to specific character of representa- tion. For instance, if ω is operation of Ω1-algebra A and operation of Ω2-algebra M , then we require that Ω2-words a1...anωx and a1x...anxω describe the same element of Ω2-algebra M .6 In addition to the above equations in Ω2-algebra there may be relations of the form (4.4) w1(f, X, m) = w2(f, X, m) m /∈ X 5For instance, let {e1, e2} be the basis of vector space over field k. The equation (4.3) has the form of distributive law a(b1e1 + b2e2) = (ab1)e1 + (ab2)e2 6For vector space, this requirement has the form of distributive law (a + b)e1 = ae1 + be1 Basis of Representation of Universal Algebra 13 The feature of the equation (4.4) is that this equation cannot be reduced.7 On the set of Ω2-words w(f, X), above equations determine equivalence ρ(f ) generated by representation f . It is evident that for any m ∈ M the choice of appropriate Ω2-word is unique up to equivalence relations ρ(f ). However, if during the construction, we obtain the equality of two Ω2-word relative to given basis, then we can say without worrying about the equivalence ρ(f ) that these Ω2-words are equal. A similar remark concerns the mapping W (f, X, m) defined in the remark 3.8.8 (cid:3) Theorem 4.6. Automorphism of the representation f maps a basis of the repre- sentation f into basis. Proof. Let the mapping R be automorphism of the representation f . Let the set X be a basis of the representation f . Let X ′ = R ◦ X. Assume that the set X ′ is not basis. According to the theorem 4.2 there exists such m′ ∈ X ′ that X ′ \ {x′} is generating set of the representation f . According to the theorem [3]-2.3.5, the mapping R−1 is automorphism of the representation f . According to the theorem 3.21 and definition 3.20, the set X \ {m} is generating set of the representation f . The contradiction completes the proof of the theorem. (cid:3) Theorem 4.7. Let X be the basis of the representation f . Let R1 : X → X ′ be arbitrary mapping of the set X. Consider the mapping of Ω2-words w[f, X, R1] : w(f, X) → w(f, X ′) that satisfies conditions (1), (2), (3) of the theorem 3.7 and such that There exists unique endomorphism of representation f 9 x ∈ X => w[f, X, R1](x) = R1(x) defined by rule R : M → M R ◦ m = w[f, X, R1](w(f, X, m)) 7See for instance sections [3]-5.5.2, [3]-5.5.3. 8If vector space has finite basis, then we represent the basis as matrix e = (cid:16)e1 We present the mapping W (f, e, v) as matrix ... e2(cid:17) W (f, e, v) =   v1 ... vn   Then W (f, e, v)(e′) = W (f, e, v)(cid:16)e′ 1 ... n(cid:17) = e′   v1 ... vn   (cid:16)e′ 1 ... n(cid:17) e′ has form of matrix product. 9This statement is similar to the theorem [1]-4.1, p. 135. 14 Aleks Kleyn Proof. The statement of theorem is corollary theorems 3.7, 3.11. (cid:3) Corollary 4.8. Let X, X ′ be the bases of the representation f . Let R be the automorphism of the representation f such that X ′ = R ◦ X. Automorphism R is uniquely defined. (cid:3) Remark 4.9. The theorem 4.7, as well as the theorem 3.11, is the theorem of ex- tension of mapping. However in this theorem, X is not arbitrary generating set of the representation, but basis. According to remark 4.3, we cannot determine the coordinates of any element of basis through the remaining elements of the same basis. Therefore, we do not need to coordinate the mapping of the basis with representation. (cid:3) Theorem 4.10. The set of coordinates W (f, X, X) corresponds to identity trans- formation W [f, X, E] = W (f, X, X) Proof. The statement of the theorem follows from the equation m = W (f, X, m) ◦ X = W (f, X, m) ◦ W (f, X, X) ◦ X (cid:3) Theorem 4.11. Let W (f, X, R ◦ X) be the set of coordinates of automorphism R. There exists set of coordinates W (f, R ◦ X, X), corresponding to automorphism R−1. The set of coordinates W (f, R ◦ X, X) satisfy to equation (4.5) W (f, X, R ◦ X) ◦ W (f, R ◦ X, X) = W (f, X, X) W [f, X, R−1] = W [f, X, R]−1 = W (f, R ◦ X, X) Proof. Since R is automorphism of the representation f , then, according to the theorem 4.6, the set R ◦ X is a basis of the representation f . Therefore, there exists the set of coordinates W (f, R ◦ X, X). The equation (4.5) follows from the chain of equations W (f, X, R ◦ X) ◦ W (f, R ◦ X, X) = W (f, X, R ◦ X) ◦ W (f, X, R−1 ◦ X) = W (f, X, R−1 ◦ R ◦ X) = W (f, X, X) (cid:3) Remark 4.12. In Ω2-algebra M there is no universal algorithm for determining the set of coordinates W (f, R ◦ X, X) for given set W (f, X, R ◦ X).10 We assume that in the theorem 4.11 this algorithm is given implicitly. It is evident also that the set of Ω2-words (4.6) W (f, X, R ◦ X) ◦ W (f, R ◦ X, X) ◦ X in general, does not coincide with the set of Ω2-words (4.7) W (f, X, X) ◦ X The theorem 4.11 states that sets of Ω2-words (4.6) and (4.7) coincide up to equiv- alence generated by the representation f . (cid:3) 10In vector space, the matrix of numbers corresponds to linear transformation. Accordingly, the inverse matrix corresponds to inverse transformation. Basis of Representation of Universal Algebra 15 Theorem 4.13. The group of automorphisms G(f ) of effective representation f in Ω2-algebra M generates effective representation in Ω2-algebra M . Proof. From the corollary 4.8, it follows that if automorphism R maps a basis X into a basis X ′, then the set of coordinates W (f, X, X ′) uniquely determines an automorphism R. From the theorem 3.15, it follows that the set of coordinates W (f, X, X ′) determines the mapping of coordinates relative to the basis X un- der automorphism of the representation f . From the equation (3.27), it follows that automorphism R acts from the right on elements of Ω2-algebra M . From the equation (3.25), it follows that the representation of group covariant from right. According to the theorem 4.10, the set of coordinates W (f, X, X) corresponds to identity transformation. From the theorem 4.11, it follows that the set of coor- dinates W (f, R ◦ X, X) corresponds to transformation, inverse to transformation W (f, X, R ◦ X). (cid:3) 5. Basis Manifold of Representation The set B(f ) of bases of representation f is called basis manifold of repre- sentation f . According to theorem 4.6, automorphism R of the representation f generates transformation (5.1) R : Y → R ◦ Y W (f, X, R ◦ Y ) = W (f, X, Y ) ◦ W (f, X, R ◦ X) of the basis manifold of representation. This transformation is called active. Ac- cording to the theorem [3]-2.3.5, we defined active representation A(f ) : G(f ) → B∗(f ) of group G(f ) in basis manifold B(f ). According to the corollary 4.8, this repre- sentation is single transitive. Remark 5.1. According to remark 4.3, it is possible that there exist bases such that there is no active transformation between them. Then we consider the orbit of selected basis as basis manifold. Therefore, it is possible that the representation f has different basis manifolds. We will assume that we have chosen a basis manifold. (cid:3) Definition 5.2. Automorphism S of representation A(f ) is called passive trans- formation of the basis manifold of representation. We also use notation to denote the image of basis X under passive transformation S. (cid:3) S(X) = S ⋆ X Theorem 5.3. Passive transformation of basis has form (5.2) S : Y → S ⋆ Y W (f, X, S ⋆ Y ) = W (f, X, S ⋆ X) ◦ W (f, X, Y ) Proof. According to the equation (5.1), active transformation acts from right on coordinates of basis. The equation (5.2) follows from theorems [3]-2.5.12, [3]- 2.5.13, [3]-2.5.15, according to these theorems, passive transformation acts from left on coordinates of basis. (cid:3) 16 Aleks Kleyn Theorem 5.4. Let S be passive transformation of the basis manifold of the repre- sentation f . Let X be the basis of the representation f , X ′ = S ⋆X. For basis Y , let there exists an active transformation R such that Y = R ◦ X. Assume Y ′ = R ◦ X ′. Then S ⋆ Y = Y ′. Proof. According to equation (5.1), active transformation of coordinates of basis Y has form (5.3) R : X ′ = S ⋆ X → Y ′ W (f, X, Y ′) = W (f, X, X ′) ◦ W (f, X, Y ) Let Y ′′ = S ⋆ Y . From the equation (5.2), it follows that (5.4) S : Y = R ◦ X → Y ′′ W (f, X, Y ′′) = W (f, X, X ′) ◦ W (f, X, Y ) From match of expressions in equations (5.3), (5.4), it follows that Y ′ = Y ′′ = S⋆Y . Therefore, the diagram B(f ) S B(f ) R R / B(f ) S / B(f ) is commutative. (cid:3) Theorem 5.5. There exists single transitive passive representation P (f ) : G(f ) → ∗B(f ) Proof. Since A(f ) is single transitive representation of group G(f ), then according to the definition 5.2 and the theorem [3]-2.3.5, the set of passive transformations forms group. (cid:3) Let passive transformation S maps basis Y into basis (5.5) Y ′ = S ⋆ Y W (f, X, Y ′) = W (f, X, S ⋆ X) ◦ W (f, X, Y ) Let passive transformation T maps basis Y ′ into basis (5.6) Y ′′ = T ⋆ Y ′ W (f, X, Y ′′) = W (f, X, T ⋆ X) ◦ W (f, X, Y ′) Denote T ⋆ S the product of passive transformations T and S (5.7) Y ′′ = T ⋆ S ⋆ Y W (f, X, Y ′′) = W (f, X, T ⋆ S ⋆ X) ◦ W (f, X, Y ) From equations (5.5), (5.6), it follows that (5.8) Y ′′ = T ⋆ S ⋆ Y W (f, X, Y ′′) = W (f, X, T ⋆ X) ◦ W (f, X, S ⋆ X) ◦ W (f, X, Y ) / / Basis of Representation of Universal Algebra 17 From equations (5.7), (5.8) and theorem 5.5, it follows that (5.9) W (f, X, T ⋆ S ⋆ X) = W (f, X, T ⋆ X) ◦ W (f, X, S ⋆ X) 6. Geometric Object of Representation of Universal Algebra An active transformation changes bases and elements of Ω2-algebra uniformly and coordinates of element relative basis do not change. A passive transformation changes only the basis and it leads to change of coordinates of element relative to the basis. Let passive transformation S ∈ G(f ) maps basis Y ∈ B(f ) into basis Y ′ ∈ B(f ) (6.1) W (f, X, Y ′) = W (f, X, S ⋆ X) ◦ W (f, X, Y ) Let element m ∈ M has Ω2-word (6.2) m = W (f, Y, m) ◦ Y relative to basis Y and has Ω2-word (6.3) m = W (f, Y ′, m) ◦ Y ′ relative to basis Y ′. From (6.1) and (6.3), it follows that (6.4) W (f, Y, m) ◦ Y = W (f, Y ′, m) ◦ Y ′ = W (f, Y ′, m) ◦ W (f, Y, Y ′) ◦ Y = W (f, Y ′, m) ◦ W (f, Y, S ⋆ Y ) ◦ Y Comparing (6.2) and (6.4) we get (6.5) W (f, Y, m) = W (f, Y ′, m) ◦ W (f, Y, S ⋆ Y ) Because S is automorphism we get from (6.5) and the theorem 4.11 (6.6) W (f, Y ′, m) = W (f, Y, m) ◦ W (f, S ⋆ Y, Y ) Coordinate transformation (6.6) does not depend on element m or basis Y , but is defined only by coordinates of element m relative to basis Y . Theorem 6.1. Coordinate transformations (6.6) form contravariant effective T ⋆- representation of group G(f ) which is called coordinate representation in Ω2- algebra. Proof. According to corollary 3.16, the transformation (6.6) is the endomorphism of representation of Ω1-algebra A into Ω2-algebra W (f, X).11 Suppose we have two consecutive passive transformations S and T . Coordinate transformation (6.6) corresponds to passive transformation S. Coordinate trans- formation (6.7) W (f, Y ′′, m) = W (f, Y ′, m) ◦ W (f, T ⋆ Y, Y ) corresponds to passive transformation T . Product of coordinate transformations (6.6) and (6.7) has form (6.8) W (f, Y ′′, m) = W (f, Y, m) ◦ W (f, S ⋆ Y, Y ) ◦ W (f, T ⋆ Y, Y ) = W (f, Y, m) ◦ W (f, T ⋆ S ⋆ Y, Y ) 11This transformation does not generate an endomorphism of the representation f . Coordinates change because basis relative which we determinate coordinates changes. However, element of Ω2- algebra M , coordinates of which we are considering, does not change. 18 Aleks Kleyn and is coordinate transformation corresponding to passive transformation T ⋆ S. It proves that coordinate transformations form contravariant T ⋆-representation of group G(f ). Suppose coordinate transformation does not change coordinates of selected basis. Then unit of group G(f ) corresponds to it because representation is single transitive. Therefore, coordinate representation is effective. (cid:3) Let f be representation of Ω1-algebra A in Ω2-algebra M . Let g be representa- tion of Ω1-algebra A in Ω3-algebra N . Passive representation gp coordinated with passive representation fp, if there exists homomorphism h of group G(f ) into group G(g). Consider diagram12 ∗B(f ) H P (f ) G(f ) n n n / ∗B(g) 7n P (g) G(g) n n n n n n n n n f ′ h Since mappings P (f ), P (g) are isomorphisms of group, then mapping H is homo- morphism of groups. Therefore, mapping f ′ is representation of group G(f ) in basis manifold B(g). According to design, passive transformation H(S) of Ω3-algebra N corresponds to passive transformation S of Ω2-algebra M . (6.9) Y ′ N = H(S) ⋆ YN Then coordinate transformation in N gets form (6.10) W (g, Y ′ P , m) = W (g, YN , m) ◦ W (g, H(S) ⋆ YP , YP ) Definition 6.2. Orbit (W (g, YN , n) ◦ W (g, H(S) ⋆ YN , YN ), S ⋆ YM , S ∈ G(f )) is called geometric object in coordinate representation defined in Ω2-al- gebra M . For any basis Y ′ M = S ⋆ YM corresponding point (6.10) of orbit defines coordinates of geometric object in coordinate space of representation relative basis Y ′ (cid:3) M . Definition 6.3. Orbit (W (g, YN , m) ◦ W (g, H(S) ⋆ YN , YN ), H(S) ⋆ YN , S ⋆ YM , S ∈ G(f )) is called geometric object defined in Ω2-algebra M . For any basis Y ′ M = S⋆YM corresponding point (6.10) of orbit defines coordinates of a geometric object in Ω2-algebra M relative to basis Y ′ M and the corresponding vector p = W (g, Y ′ P , p) ◦ Y ′ P is called representative of geometric object in Ω2-algebra M . (cid:3) Since a geometric object is an orbit of representation, we see that according to the theorem [3]-2.4.13 the definition of the geometric object is a proper definition. We also say that p is a geometric object of type H 12We can relax definition of consistency of representations and assume that N is Ω2-algebra. Then it is sufficient to require that the mapping (h, H) is morphism of representations. However in this case, elements of affine representation cannot be geometric object of vector space. / O O / / 7 O O Basis of Representation of Universal Algebra 19 Definition 6.2 introduces a geometric object in coordinate space. We assume in definition 6.3 that we selected a basis of representation g. This allows using a representative of the geometric object instead of its coordinates. Theorem 6.4 (invariance principle). Representative of geometric object does not depend on selection of basis Y ′ M . Proof. To define representative of geometric object, we need to select basis YM , basis YP and coordinates of geometric object W (g, YP , p). Corresponding represen- tative of geometric object has form p = W (g, YP , p) ◦ YP Suppose we map basis YM to basis Y ′ Y ′ M by passive transformation M = S ◦ YM According building this forms passive transformation (6.9) and coordinate trans- formation (6.10). Corresponding representative of geometric object has form p′ = W (g, Y ′ P , p′) ◦ Y ′ P = W (g, YP , p) ◦ W (g, H(S) ⋆ YP , YP ) ◦ W (g, YP , H(S) ⋆ YP ) ◦ YP = W (g, YP , p) ◦ YP = p Therefore representative of geometric object is invariant relative selection of basis. (cid:3) Theorem 6.5. The set of geometric objects of type H is Ω3-algebra. Proof. Let pi = W (g, YP , pi) ◦ YP i = 1, ..., n For operation ω ∈ Ω3(n) we assume (6.11) p1...pnω = W (g, YP , p1)...W (g, YP , pn)ω ◦ YP Since for arbitrary endomorphism S of Ω2-algebra M , the mapping W (g, H(S) ⋆ YN , YN ) is endomorphism of Ω3-algebra N , then the definition (6.11) is correct. (cid:3) Theorem 6.6. There exists the representation of Ω1-algebra A in Ω3-algebra N of geometric objects of type H. Proof. Let For a ∈ A, we assume p = W (g, YP , p) ◦ YP (6.12) ap = aW (g, YP , p1) ◦ YP Since for arbitrary endomorphism S of Ω2-algebra M , the mapping W (g, H(S) ⋆ YN , YN ) is endomorphism of representation g, then the definition (6.12) is correct. (cid:3) 20 Aleks Kleyn 7. Examples of Basis of Representation of Universal Algebra 7.1. Vector Space. Consider the vector space V over the field F . Given the set of vectors e1, ..., en, according to algorithm of construction of coordinates over vector space, coordinates include such elements as e1 + e2 and ae1. Recursively using rules, contained in the definition 3.6, we conclude that the set of vectors e1, ..., en, generates the set of linear combinations a1e1 + ... + anen According to the theorem 4.2, the set of vectors e1, ..., en, is a basis if for any i, i = 1, ..., n, vector ei is not linear combination of other vectors. This requirement is equivalent to the requirement of linear independence of vectors. 7.2. Representation of Group on the Set. Consider the representation from the example [3]-2.6.2. We can consider the set M as union of orbits of the repre- sentation of the group G. We can select for basis of the representation the set of points such that one and only one point belongs to each orbit. If X is the basis of representation, A ∈ X, g ∈ G, then Ω2-word has form A + g. Since there is no op- erations on the set M , then there is no Ω2-word containing different elements of the basis. If representation of group G is single transitive, then basis of representation consists of one point. Any point of the set M can be such point. 8. References [1] Serge Lang, Algebra, Springer, 2002 [2] S. Burris, H.P. Sankappanavar, A Course in Universal Algebra, Springer- Verlag (March, 1982), eprint http://www.math.uwaterloo.ca/ snburris/htdocs/ualg.html (The Millennium Edition) [3] Aleks Kleyn, Representation Theory: Representation of Universal Algebra, Lambert Academic Publishing, 2011 [4] Paul M. Cohn, Universal Algebra, Springer, 1981 9. Index active representation of group G(f ) in basis manifold of representation 15 active transformation of basis manifold of representation 15 basis manifold of representation 15 basis of representation 11 coordinate representation in Ω2-algebra 17 coordinates of a geometric object in Ω2- algebra M 18 coordinates of element m of representation f relative to set X 6 coordinates of endomorphism of representation 8 coordinates of geometric object in coordinate space of representation 18 endomorphism of representation regular on generating set X 11 endomorphism of representation singular on generating set X 11 equivalence generated by representation f 13 generating set of representation 4 generating set of subrepresentation 4 geometric object defined in Ω2-algebra M 18 geometric object in coordinate representation defined in Ω2-algebra M 18 geometric object of type H 18 invariance principle in representation of universal algebra 19 Ω2-word of element of representation relative to generating set 4 passive representation of group G(f ) in basis manifold of representation 16 passive transformation of the basis manifold of representation 15 regular endomorphism of representation 11 representative of geometric object in Ω2- algebra 18 set of coordinates of representation 6 set of Ω2-words of representation 5 stable set of representation 2 subrepresentation generated by set X 4 subrepresentation of representation 2 superposition of coordinates of the representation f and the element m 8 21 10. Special Symbols and Notations A(f ) active representation of group G(f ) in basis manifold B(f ) 15 B(f ) basis manifold of representation f 15 Bf lattice of subrepresentations of representation f 3 J(f ) closure operator of representation f 3 P (f ) passive representation of group G(f ) in basis manifold B(f ) 16 R ◦ m image of m under endomorphism R of effective representation 4 S ⋆ X image of basis X under passive transformation S 15 W (f, X) set of coordinates of representation J(f, X) 6 W (f, X, m) coordinates of element m of representation f relative to set X 5 (W (g, YN , n) ◦ W (g, H(S) ⋆ YN , YN ), S ⋆ YM , S ∈ G(f )) geometric object in coordinate representation defined in Ω2-algebra M 18 (W (g, YN , m) ◦ W (g, H(S) ⋆ YN , YN ), H(S) ⋆ YN , S ⋆ YM , S ∈ G(f )) geometric object defined in Ω2-algebra M 18 W (f, X, B) set of coordinates of set B ⊂ J(f, X) 6 w(f, X, B) set of Ω2-words representing set B ⊂ J(f, X) 4 W (f, X, m) ◦ W (f, X, R ◦ X) superposition of coordinates of the representation f and the element m 8 w(f, X, m) Ω2-word representing element m ∈ J(f, X) 4 w(f, X) set of Ω2-words of representation J(f, X) 5 22 1 1 0 2 v o N 9 ] . A R h t a m [ 1 v 5 3 0 6 . 1 1 1 1 : v i X r a Базис представления универсальной алгебры Александр Клейн Аннотация. Если множество эндоморфизмов универсальной алгебры A имеет структуру универсальной алгебры B, то мы говорим, что опреде- лено представление универсальной алгебры B в универсальной алгебре A. Следовательно, роль представления универсальной алгебры аналогич- на роли симметрии в геометрии и физике. Морфизм представления - это отображение, сохраняющее структуру представления. Изучение морфиз- мов представлений ведёт к понятиям множества образующих и базиса представления. Множество автоморфизмов представления универсальной алгебры порождает группу. Парные представления этой группы в много- бразии базисов представления называются активным и пассивным пред- ставлениями. Пассивное представление является основой понятия геомет- рического объекта и теории инвариантов представления универсальной алгебры. Содержание 1. Предисловие . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Соглашения . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. Множество образующих представления . . . . . . . . . . . . . . . . 4. Базис представления . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Многообразие базисов представления . . . . . . . . . . . . . . . . . . 6. Геометрический объект представления универсальной алгебры . . 7. Примеры базиса представления универсальной алгебры . . . . . . . 7.1. Векторное пространство . . . . . . . . . . . . . . . . . . . . . . . . 7.2. Представление группы на множестве . . . . . . . . . . . . . . . . . 8. Список литературы . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. Предметный указатель . . . . . . . . . . . . . . . . . . . . . . . . . . 10. Специальные символы и обозначения . . . . . . . . . . . . . . . . . 1 2 2 12 16 18 21 21 21 21 22 23 1. Предисловие Эта статья написана на основе главы [3]-3. Роль представления универсальной алгебры аналогична роли симметрии в геометрии и физике В обоих случаях мы изучаем структурированные множе- ства преобразований, и знание структуры этого множества даёт нам дополни- тельные сведения об изучаемом объекте. [email protected]. http://sites.google.com/site/AleksKleyn/ . http://arxiv.org/a/kleyn_a_1. http://AleksKleyn.blogspot.com/. 1 2 Александр Клейн Мы рассматриваем теорию представлений универсальной алгебры как рас- ширение теории универсальной алгебры. Любая алгебраическая структура пред- полагает множество отображений, сохраняющих эту структуру. Отображение, сохраняющее структуру представления универсальной алгебры, называется морфизмом представлений. Частным случаем морфизма представлений является автоморфизм пред- ставления. Изучение автоморфизмов представления непосредственно связано с необходимостью ответить на вопрос какова структура множества образую- щих представления. Если представление имеет минимальное множество обра- зующих, то такое множество называется базисом представления. Автоморфизм представления отображает базис в базис. Множество автоморфизмов представления порождает группу. Парные пред- ставления этой группы в многобразии базисов представления называются ак- тивным и пассивным представлениями. Пассивное представление является ос- новой понятия геометрического объекта и теории инвариантов представления универсальной алгебры. 2. Соглашения Соглашение 2.1. В [4] произвольная операция алгебры обозначена буквой ω, и Ω - множество операций некоторой универсальной алгебры. Соответствен- но, универсальная алгебра с множеством операций Ω обозначается Ω-алгебра. Аналогичные обозначения мы видим в [2] с той небольшой разницей, что опе- рация в алгебре обозначена буквой f и F - множество операций. Я выбрал первый вариант обозначений, так как в этом случае легче видно, где я ис- пользую операцию. (cid:3) Соглашение 2.2. Пусть A - Ω1-алгебра. Пусть B - Ω2-алгебра. Запись A ∗ / B означает, что определено представление Ω1-алгебры A в Ω2-алгебре B. (cid:3) Без сомнения, у читателя могут быть вопросы, замечания, возражения. Я буду признателен любому отзыву. 3. Множество образующих представления Определение 3.1. Пусть f : A → ∗M представление Ω1-алгебры A в Ω2-алгебре M . Множество N ⊂ M называется стабильным множеством представления f , если f (a)(m) ∈ N для любых a ∈ A, m ∈ N . (cid:3) Мы также будем говорить, что множество M стабильно относительно пред- ставления f . Теорема 3.2. Пусть f : A → ∗M представление Ω1-алгебры A в Ω2-алгебре M . Пусть множество N ⊂ M явля- ется подалгеброй Ω2-алгебры M и стабильным множеством представления f . Тогда существует представление fN : A → ∗N / Базис представления универсальной алгебры 3 такое, что fN (a) = f (a)N . Представление fN называется подпредставле- нием представления f . Доказательство. Пусть ω1 - n-арная операция Ω1-алгебры A. Тогда для лю- бых a1, ..., an ∈ A и любого b ∈ N (fN (a1)...fN (an)ω1)(b) = (f (a1)...f (an)ω1)(b) = f (a1...anω1)(b) = fN (a1...anω1)(b) Пусть ω2 - n-арная операция Ω2-алгебры M . Тогда для любых b1, ..., bn ∈ N и любого a ∈ A fN (a)(b1)...fN (a)(bn)ω2 = f (a)(b1)...f (a)(bn)ω2 = f (a)(b1...bnω2) = fN (a)(b1...bnω2) Утверждение теоремы доказано. (cid:3) Из теоремы 3.2 следует, что если fN - подпредставление представления f , то отображение (id : A → A, idN : N → M ) является морфизмом представлений. Теорема 3.3. Множество1 Bf всех подпредставлений представления f по- рождает систему замыканий на Ω2-алгебре M и, следовательно, является полной структурой. Доказательство. Пусть (Kλ)λ∈Λ - семейство подалгебр Ω2-алгебры M , ста- бильных относительно представления f . Операцию пересечения на множестве Bf мы определим согласно правилу \ fKλ = f∩Kλ Операция пересечения подпредставлений определена корректно. ∩Kλ - подал- гебра Ω2-алгебры M . Пусть m ∈ ∩Kλ. Для любого λ ∈ Λ и для любого a ∈ A, f (a)(m) ∈ Kλ. Следовательно, f (a)(m) ∈ ∩Kλ. Следовательно, ∩Kλ - стабиль- ное множество представления f . (cid:3) Обозначим соответствующий оператор замыкания через J(f ). Таким обра- зом, J(f, X) является пересечением всех подалгебр Ω2-алгебры M , содержащих X и стабильных относительно представления f . Теорема 3.4. Пусть2 f : A → ∗M представление Ω1-алгебры A в Ω2-алгебре M . Пусть X ⊂ M . Определим под- множество Xk ⊂ M индукцией по k. X0 = X x ∈ Xk => x ∈ Xk+1 x1 ∈ Xk, ..., xn ∈ Xk, ω ∈ Ω2(n) => x1...xnω ∈ Xk+1 x ∈ Xk, a ∈ A => f (a)(x) ∈ Xk+1 1Это определение аналогично определению структуры подалгебр ([4], стр. 93, 94) 2Утверждение теоремы аналогично утверждению теоремы 5.1, [4], стр. 94. 4 Тогда Александр Клейн ∞ [ k=0 Xk = J(f, X) Доказательство. Если положим U = ∪Xk, то по определению Xk имеем X0 ⊂ J(f, X), и если Xk ⊂ J(f, X), то Xk+1 ⊂ J(f, X). По индукции следует, что Xk ⊂ J(f, X) для всех k. Следовательно, (3.1) U ⊂ J(f, X) Если a ∈ U n, a = (a1, ..., an), где ai ∈ Xki, и если k = max{k1, ..., kn}, то a1...anω ∈ Xk+1 ⊂ U . Следовательно, U является подалгеброй Ω2-алгебры M . Если m ∈ U , то m ∈ Xk для некоторого k. Следовательно, f (a)(m) ∈ Xk+1 ⊂ U для любого a ∈ A. Следовательно, U - стабильное множество представления f . Так как U - подалгеброй Ω2-алгебры M и стабильное множество представ- ления f , то определено подпредставление fU . Следовательно, (3.2) J(f, X) ⊂ U Из (3.1), (3.2), следует J(f, X) = U . (cid:3) Определение 3.5. J(f, X) называется подпредставлением, порождён- ным множеством X, а X - множеством образующих подпредставления J(f, X). В частности, множеством образующих представления f будет такое подмножество X ⊂ M , что J(f, X) = M . (cid:3) Нетрудно видеть, что определение множества образующих представления не зависит от того, эффективно представление или нет. Поэтому в дальнейшем мы будем предполагать, что представление эффективно и будем опираться на соглашение для эффективного T ⋆-представления в замечании [3]-2.1.9. Мы также будем пользоваться записью R ◦ m = R(m) для образа m ∈ M при эндоморфизме эффективного представления. Соглас- но определению произведения отображений, для любых эндоморфизмов R, S верно равенство (3.3) (R ◦ S) ◦ m = R ◦ (S ◦ m) Равенство (3.3) является законом ассоциативности для ◦ и позволяет записать выражение без использования скобок. Из теоремы 3.4 следует следующее определение. R ◦ S ◦ m Определение 3.6. Пусть X ⊂ M . Для любого x ∈ J(f, X) существует Ω2- слово, определённое согласно следующему правилам. (1) Если m ∈ X, то m - Ω2-слово. (2) Если m1, ..., mn - Ω2-слова и ω ∈ Ω2(n), то m1...mnω - Ω2-слово. (3) Если m - Ω2-слово и a ∈ A, то am - Ω2-слово. Базис представления универсальной алгебры 5 Ω2-слово w(f, X, m) представляет данный элемент m ∈ J(f, X). Мы будем отождествлять элемент m ∈ J(f, X) и соответствующее ему Ω2-слово, выражая это равенством m = w(f, X, m) Аналогично, для произвольного множества B ⊂ J(f, X) рассмотрим множе- ство Ω2-слов3 w(f, X, B) = {w(f, X, m) : m ∈ B} Мы будем также пользоваться записью Обозначим w(f, X) множество Ω2-слов представления J(f, X). (cid:3) w(f, X, B) = (w(f, X, m), m ∈ B) Теорема 3.7. Эндоморфизм R представления f : A → ∗M порождает отображение Ω2-слов w[f, X, R] : w(f, X) → w(f, X ′) X ⊂ M X ′ = R ◦ X такое, что (1) Если m ∈ X, m′ = R ◦ m, то (2) Если w[f, X, R](m) = m′ m1, ..., mn ∈ w(f, X) m′ 1 = w[f, X, R](m1) ... m′ n = w[f, X, R](mn) то для операции ω ∈ Ω2(n) справедливо w[f, X, R](m1...mnω) = m′ 1...m′ nω (3) Если то m ∈ w(f, X) m′ = w[f, X, R](m) a ∈ A w[f, X, R](am) = am′ Доказательство. Утверждения (1), (2) теоремы справедливы в силу опреде- ления эндоморфизма R. Утверждение (3) теоремы следует из равенства [3]- (2.2.4), так как r = id. (cid:3) 3Выражение w(f, X, m) является частным случаем выражения w(f, X, B), а именно w(f, X, {m}) = {w(f, X, m)} 6 Александр Клейн Замечание 3.8. Пусть R - эндоморфизм представления f . Пусть m ∈ J(f, X) m′ = R ◦ m X ′ = R ◦ X Теорема 3.7 утверждает, что m′ ∈ Jf (X ′) . Теорема 3.7 также утверждает, что Ω2-слово, представляющее m, относительно X и Ω2-слово, представляющее m′, относительно X ′ формируются согласно одному и тому же алгоритму. Это позволяет рассматривать множество Ω2-слов w(f, X ′, m′) как отображение W (f, X, m) : X ′ → w(f, X ′, m′) W (f, X, m)(X ′) = W (f, X, m) ◦ X ′ такое, что, если для некоторого эндоморфизма R то X ′ = R ◦ X m′ = R ◦ m W (f, X, m) ◦ X ′ = w(f, X ′, m′) = m′ Отображение W (f, X, m) называется координатами элемента m относи- тельно множества X. Аналогично, мы можем рассмотреть координаты мно- жества B ⊂ J(f, X) относительно множества X W (f, X, B) = {W (f, X, m) : m ∈ B} = (W (f, X, m), m ∈ B) Обозначим W (f, X) множество координат представления J(f, X). (cid:3) Теорема 3.9. На множестве координат W (f, X) определена структура Ω2- алгебры. Доказательство. Пусть ω ∈ Ω2(n). Тогда для любых m1, ..., mn ∈ J(f, X) положим (3.4) W (f, X, m1)...W (f, X, mn)ω = W (f, X, m1...mnω) Согласно замечанию 3.8, из равенства (3.4) следует (3.5) (W (f, X, m1)...W (f, X, mn)ω) ◦ X = W (f, X, m1...mnω) ◦ X = w(f, X, m1...mnω) Согласно правилу (2) определения 3.6, из равенства (3.5) следует (3.6) (W (f, X, m1)...W (f, X, mn)ω) ◦ X = w(f, X, m1)...w(f, X, mn)ω = (W (f, X, m1) ◦ X)...(W (f, X, mn) ◦ X)ω Из равенства (3.6) следует корректность определения (3.4) операции ω на мно- жестве координат. (cid:3) Теорема 3.10. Определено представление Ω1-алгебры A в Ω2-алгебре W (f, X). Доказательство. Пусть a ∈ A. Тогда для любого m ∈ J(f, X) положим (3.7) aW (f, X, m) = W (f, X, am) Согласно замечанию 3.8, из равенства (3.7) следует (3.8) (aW (f, X, m)) ◦ X = W (f, X, am) ◦ X = w(f, X, am) Базис представления универсальной алгебры 7 Согласно правилу (3) определения 3.6, из равенства (3.8) следует (3.9) (aW (f, X, m)) ◦ X = aw(f, X, m) = a(W (f, X, m) ◦ X) Из равенства (3.9) следует корректность определения (3.7) представления Ω1- алгебры A в Ω2-алгебре W (f, X). (cid:3) Теорема 3.11. Пусть f : A → ∗M представление Ω1-алгебры A в Ω2-алгебре M . Для заданных множеств X ⊂ M , X ′ ⊂ M , пусть отображение согласовано со структурой представления f , т. е. R1 : X → X ′ ω ∈ Ω2(n) x1, ..., xn, x1...xnω ∈ X, R1(x1...xnω) ∈ X ′ => R1(x1...xnω) = R1(x1)...R1(xn)ω x ∈ X, a ∈ A, R1(ax) ∈ X ′ => R1(ax) = aR1(x) Рассмотрим отображение Ω2-слов w[f, X, R1] : w(f, X) → w(f, X ′) удовлетворяющее условиям (1), (2), (3) теоремы 3.7, и такое, что Существует единственный эндоморфизм Ω2-алгебры M x ∈ X => w[f, X, R1](x) = R1(x) определённый правилом R : M → M который является морфизмом представлнений J(f, X) и J(f, X ′). R ◦ m = w[f, X, R1](w(f, X, m)) Доказательство. Мы будем доказывать теорему индукцией по сложности Ω2- слова. Если w(f, X, m) = m, то m ∈ X. Согласно условию (1) теоремы 3.7, R ◦ m = w[f, X, R1](w(f, X, m)) = w[f, X, R1](m) = R1(m) Следовательно, на множестве X отображения R и R1 совпадают, и отображе- ние R согласовано со структурой представления f . Пусть ω ∈ Ω2(n). Пусть отображение R определено для m1, ..., mn ∈ J(f, X). Пусть w1 = w(f, X, m1) ... wn = w(f, X, mn) Если m = m1...mnω, то согласно правилу (2) определения 3.6, w(f, X, m) = w1...wnω 8 Александр Клейн Согласно условию (2) теоремы 3.7, R ◦ m = w[f, X, R1](w(f, X, m)) = w[f, X, R1](w1...wnω) = w[f, X, R1](w1)...w[f, X, R1](wn)ω = (R ◦ m1)...(R ◦ mn)ω Следовательно, отображение R является эндоморфизмом Ω2-алгебры M . Пусть отображение R определено для m1 ∈ J(f, X), w1 = w(f, X, m1). Пусть a ∈ A. Если m = am1, то согласно правилу (3) определения 3.6, Согласно условию (3) теоремы 3.7, w(f, X, am1) = aw1 R ◦ m = w[f, X, R1](w(f, X, m)) = w[f, X, R1](aw1) = aw[f, X, R1](w1) = aR ◦ m1 Из равенства [3]-(2.2.4) следует, что отображение R является морфизмом пред- ставления f . Единственность эндоморфизма R, а следовательно, корректность его опре- деления, следует из следующего рассуждения. Допустим, m ∈ M имеет раз- личные Ω2-слова относительно множества X, например (3.10) m = x1...xnω = ax Так как R - эндомоморфизм представления, то из равенства (3.10) следует (3.11) R ◦ m = R ◦ (x1...xnω) = (R ◦ x1)...(R ◦ xn)ω = R ◦ (ax) = aR ◦ x Из равенства (3.11) следует (3.12) R ◦ m = (R ◦ x1)...(R ◦ xn)ω = aR ◦ x Из равенств (3.10), (3.12) следует, что равенство (3.10) сохраняется при отоб- ражении. Следовательно, образ m не зависит от выбора координат. (cid:3) Замечание 3.12. Теорема 3.11 - это теорема о продолжении отображения. Един- ственное, что нам известно о множестве X - это то, что X - множество об- разующих представления f . Однако, между элементами множества X могут существовать соотношения, порождённые либо операциями Ω2-алгебры M , ли- бо преобразованиями представления f . Поэтому произвольное отображение множества X, вообще говоря, не может быть продолжено до эндоморфизма представления f .4 Однако, если отображение R1 согласованно со структурой представления на множестве X, то мы можем построить продолжение этого отображения, которое является эндоморфизмом представления f . (cid:3) Определение 3.13. Пусть X - множество образующих представления f . Пусть R - эндоморфизм представления f . Множество координат W (f, X, R ◦ X) на- зывается координатами эндоморфизма представления. (cid:3) 4В теореме 4.7, требования к множеству образующих более жёсткие. Поэтому теорема 4.7 говорит о продолжении произвольного отображения. Более подробный анализ дан в за- мечании 4.9. Базис представления универсальной алгебры 9 Определение 3.14. Пусть X - множество образующих представления f . Пусть R - эндоморфизм представления f . Пусть m ∈ M . Мы определим суперпози- цию координат эндоморфизма представления f и элемента m как координа- ты, определённые согласно правилу (3.13) W (f, X, m) ◦ W (f, X, R ◦ X) = W (f, X, R ◦ m) Пусть Y ⊂ M . Мы определим суперпозицию координат эндоморфизма пред- ставления f и множества Y согласно правилу (3.14) W (f, X, Y ) ◦ W (f, X, R ◦ X) = (W (f, X, m) ◦ W (f, X, R ◦ X), m ∈ Y ) (cid:3) Теорема 3.15. Эндоморфизм R представления f : A → ∗M порождает отображение координат представления (3.15) такое, что (3.16) W [f, X, R] : W (f, X) → W (f, X) W (f, X, m) → W [f, X, R] ⋆ W (f, X, m) = W (f, X, R ◦ m) W [f, X, R] ⋆ W (f, X, m) = W (f, X, m) ◦ W (f, X, R ◦ X) Доказательство. Согласно замечанию 3.8, мы можем рассматривать равен- ства (3.13), (3.15) относительно заданного множества образующих X. При этом координатам W (f, X, m) соответствует слово (3.17) W (f, X, m) ◦ X = w(f, X, m) а координатам W (f, X, R ◦ m) соответствует слово (3.18) W (f, X, R ◦ m) ◦ X = w(f, X, R ◦ m) Поэтому для того, чтобы доказать теорему, нам достаточно показать, что отоб- ражению W [f, X, R] соответствует отображение w[f, X, R]. Мы будем доказы- вать это утверждение индукцией по сложности Ω2-слова. Если m ∈ X, m′ = R◦m, то, согласно равенствам (3.17), (3.18), отображения W [f, X, R] и w[f, X, R] согласованы. Пусть для m1, ..., mn ∈ X отображения W [f, X, R] и w[f, X, R] согласованы. Пусть ω ∈ Ω2(n). Согласно теореме 3.9 (3.19) W (f, X, m1...mnω) = W (f, X, m1)...W (f, X, mn)ω Так как R - эндоморфизм Ω2-алгебры M , то из равенства (3.19) следует (3.20) W (f, X, R ◦ (m1...mnω)) = W (f, X, (R ◦ m1)...(R ◦ mn)ω) = W (f, X, R ◦ m1)...W (f, X, R ◦ mn)ω Из равенств (3.19), (3.20) и предположения индукции следует, что отображения W [f, X, R] и w[f, X, R] согласованы для m = m1...mnω. Пусть для m1 ∈ M отображения W [f, X, R] и w[f, X, R] согласованы. Пусть a ∈ A. Согласно теореме 3.10 (3.21) W (f, X, am1) = aW (f, X, m1) 10 Александр Клейн Так как R - эндоморфизм представления f , то из равенства (3.21) следует (3.22) W (f, X, R ◦ (am1)) = W (f, X, a R ◦ m1) = aW (f, X, R ◦ m1) Из равенств (3.21), (3.22) и предположения индукции следует, что отображения W [f, X, R] и w[f, X, R] согласованы для m = am1. (cid:3) Следствие 3.16. Пусть X - множество образующих представления f . Пусть R - эндоморфизм представления f . Отображение W [f, X, R] является эндо- морфизмом представления Ω1-алгебры A в Ω2-алгебре W (f, X). (cid:3) В дальнейшем мы будем отождествлять отображение W [f, X, R] и множе- ство координат W (f, X, R ◦ X). Теорема 3.17. Пусть X - множество образующих представления f . Пусть R - эндоморфизм представления f . Пусть Y ⊂ M . Тогда (3.23) (3.24) W (f, X, Y ) ◦ W (f, X, R ◦ X) = W (f, X, R ◦ Y ) W [f, X, R] ⋆ W (f, X, Y ) = W (f, X, R ◦ Y ) Доказательство. Равенство (3.23) является следствием равенства R ◦ Y = (R ◦ m, m ∈ Y ) а также равенств (3.13), (3.14). Равенство (3.24) является следствием равенств (3.23), (3.16). (cid:3) Теорема 3.18. Пусть X - множество образующих представления f . Пусть R, S - эндоморфизмы представления f . Тогда (3.25) (3.26) W (f, X, S ◦ X) ◦ W (f, X, R ◦ X) = W (f, X, R ◦ S ◦ X) W [f, X, R] ⋆ W [f, X, S] = W [f, X, R ◦ S] Доказательство. Равенство (3.25) следует из равенства (3.23), если положить Y = S ◦ X. Равенство (3.26) следует из равенства (3.25) и цепочки равенств (W [f, X, R] ⋆ W [f, X, S]) ⋆ W (f, X, Y ) = W [f, X, R] ⋆ (W [f, X, S] ⋆ W (f, X, Y )) = (W (f, X, Y ) ◦ W (f, X, S ◦ X)) ◦ W (f, X, R ◦ X) = W (f, X, Y ) ◦ (W (f, X, S ◦ X) ◦ W (f, X, R ◦ X)) = W (f, X, Y ) ◦ W (f, X, R ◦ S ◦ X) = W (f, X, R ◦ S) ⋆ W (f, X, Y ) Концепция суперпозиции координат очень проста и напоминает своеобраз- ную машину Тюринга. Если элемент m ∈ M имеет вид (cid:3) или m = m1...mnω m = am1 то мы ищем координаты элементов mi, для того чтобы подставить их в соот- ветствующее выражение. Как только элемент m ∈ M принадлежит множеству Базис представления универсальной алгебры 11 образующих Ω2-алгебры M , мы выбираем координаты соответствующего эле- мента из второго множителя. Поэтому мы требуем, чтобы второй множитель в суперпозиции был множеством координат образа множества образующих X. Мы можем обобщить определение суперпозиции координат и предположить, что один из множителей является множеством Ω2-слов. Соответственно, опре- деление суперпозиции координат имеет вид w(f, X, Y ) ◦ W (f, X, R ◦ X) = W (f, X, Y ) ◦ w(f, X, R ◦ X) = w(f, X, R ◦ Y ) Следующие формы записи образа множества Y при эндоморфизме R экви- валентны. (3.27) R ◦ Y = W (f, X, Y ) ◦ (R ◦ X) = W (f, X, Y ) ◦ (W (f, X, R ◦ X) ◦ X) Из равенств (3.23), (3.27) следует, что (3.28) (W (f, X, Y ) ◦ W (f, X, R ◦ X)) ◦ X = W (f, X, Y ) ◦ (W (f, X, R ◦ X) ◦ X) Равенство (3.28) является законом ассоциативности для операции композиции и позволяет записать выражение W (f, X, Y ) ◦ W (f, X, R ◦ X) ◦ X без использования скобок. Рассмотрим равенство (3.25) где мы видим изменение порядка эндоморфиз- мов в суперпозиции координат. Это равенство также следует из цепочки ра- венств, где мы можем непосредственно видеть, когда изменяется порядок эн- доморфизмов W (f, X, m) ◦ W (f, X, R ◦ S ◦ X) ◦ X (3.29) = (R ◦ S) ◦ m = R ◦ (S ◦ m) = R ◦ ((W (f, X, m) ◦ W (f, X, S ◦ X)) ◦ X) = W (f, X, m) ◦ W (f, X, S ◦ X) ◦ W (f, X, R ◦ X) ◦ X Из равенства (3.29) следует, что координаты эндоморфизма действуют на ко- ординаты элементы Ω2-алгебры M справа. Определение 3.19. Пусть X ⊂ M - множество образующих представления Пусть отображение f : A → ∗M H : M → M является эндоморфизмом представления f . Пусть множество X ′ = H ◦ X яв- ляется образом множества X при отображении H. Эндоморфизм H представ- ления f называется невырожденным на множестве образующих X, если множество X ′ является множеством образующих представления f . В против- ном случае, эндоморфизм H представления f называется вырожденным на множестве образующих X, (cid:3) Определение 3.20. Эндоморфизм представления f называется невырож- денным, если он невырожден на любом множестве образующих. (cid:3) 12 Александр Клейн Теорема 3.21. Автоморфизм R представления является невырожденным эндоморфизмом. f : A → ∗M Доказательство. Пусть X - множество образующих представления f . Пусть X ′ = R ◦ X. Согласно теореме 3.7 эндоморфизм R порождает отображение Ω2-слов w[f, X, R]. Пусть m′ ∈ M . Так как R - автоморфизм, то существует m ∈ M , R ◦ m = m′. Согласно определению 3.6, w(f, X, m) - Ω2-слово, представляющее m относи- тельно множества образующих X. Согласно теореме 3.7, w(f, X ′, m′) - Ω2-сло- во, представляющее m′ относительно множества образующих X ′ w(f, X ′, m′) = w[f, X, R](w(f, X, m)) Следовательно, X ′ - множество образующих представления f . Согласно опре- делению 3.20, автоморфизм R - невырожден. (cid:3) 4. Базис представления Определение 4.1. Если множество X ⊂ M является множеством образу- ющих представления f , то любое множество Y , X ⊂ Y ⊂ M также является множеством образующих представления f . Если существует минимальное мно- жество X, порождающее представление f , то такое множество X называется базисом представления f . (cid:3) Теорема 4.2. Множество образующих X представления f является бази- сом тогда и только тогда, когда для любого m ∈ X множество X \ {m} не является множеством образующих представления f . Доказательство. Пусть X - множество образующих расслоения f . Допустим для некоторого m ∈ X существует Ω2-слово (4.1) w = w(f, X \ {m}, m) Рассмотрим элемент m′, для которого Ω2-слово w′ = w(f, X, m′) (4.2) зависит от m. Согласно определению 3.6, любое вхождение m в Ω2-слово w′ может быть заменено Ω2-словом w. Следовательно, Ω2-слово w′ не зависит от m, а множество X \ {m} является множеством образующих представления f . Следовательно, X не является базисом расслоения f . (cid:3) Замечание 4.3. Доказательство теоремы 4.2 даёт нам эффективный метод по- строения базиса представления f . Выбрав произвольное множество образую- щих, мы шаг за шагом исключаем те элементы множества, которые имеют координаты относительно остальных элементов множества. Если множество образующих представления бесконечно, то рассмотренная операция может не иметь последнего шага. Если представление имеет конечное множество обра- зующих, то за конечное число шагов мы можем построить базис этого пред- ставления. Как отметил Кон в [4], cтр. 96, 97, представление может иметь неэквива- лентные базисы. Например, циклическая группа шестого порядка имеет бази- сы {a} и {a2, a3}, которые нельзя отобразить один в другой эндоморфизмом представления. (cid:3) Базис представления универсальной алгебры 13 Замечание 4.4. Мы будем записывать базис также в виде X = (x, x ∈ X) Если базис - конечный, то мы будем также пользоваться записью X = (xi, i ∈ I) = (x1, ..., xn) (cid:3) Замечание 4.5. Мы ввели Ω2-слово элемента x ∈ M относительно множества образующих X в определении 3.6. Из теоремы 4.2 следует, что если множество образующих X не является базисом, то выбор Ω2-слова относительно множе- ства образующих X неоднозначен. Но даже если множество образующих X является базисом, то представление m ∈ M в виде Ω2-слова неоднозначно. Если m1, ..., mn - Ω2-слова, ω ∈ Ω2(n) и a ∈ A, то5 (4.3) a(m1...mnω) = (am1)...(amn)ω Возможны равенства, связанные со спецификой представления. Например, ес- ли ω является операцией Ω1-алгебры A и операцией Ω2-алгебры M , то мы можем потребовать, что Ω2-слова a1...anωx и a1x...anxω описывают один и тот же элемент Ω2-алгебры M .6 Помимо перечисленных равенств в Ω2-алгебре могут существовать соотно- шения вида (4.4) w1(f, X, m) = w2(f, X, m) m /∈ X Особенностью равенства (4.4) является то, что это равенство не может быть приведенно к более простому виду.7 На множестве Ω2-слов w(f, X), перечисленные выше равенства определя- ют отношение эквивалентности ρ(f ), порождённое представлением f . Очевидно, что для произвольного m ∈ M выбор соответствующего Ω2-слова однозначен с точностью до отношения эквивалентности ρ(f ). Тем не менее, ес- ли в процессе построений мы получим равенство двух Ω2-слов относительно заданного базиса, мы можем утверждать, что эти Ω2-слова равны, не заботясь об отношении эквивалентности ρ(f ). Аналогичное замечание касается отображения W (f, X, m), определённого в (cid:3) замечании 3.8.8 Теорема 4.6. Автоморфизм представления f отображает базис представ- ления f в базис. 5Например, пусть {e1, e2} - базис векторного пространства над полем k. Равенство (4.3) принимает форму закона дистрибутивности a(b1e1 + b2e2) = (ab1)e1 + (ab2)e2 6Для векторного пространства это требование принимает форму закона дистрибутивно- сти (a + b)e1 = ae1 + be1 7Смотри, например, подразделы [3]-5.5.2, [3]-5.5.3. 8Если базис векторного пространства - конечен, то мы можем представить базис в виде матрицы строки e = (cid:16)e1 ... e2(cid:17) 14 Александр Клейн Доказательство. Пусть отображение R - автоморфизм представления f . Пусть множество X - базис представления f . Пусть X ′ = R ◦ X. Допустим множество X ′ не является базисом. Согласно теореме 4.2 суще- ствует m′ ∈ X ′ такое, что X ′\{x′} является множеством образующих представ- ления f . Согласно теореме [3]-2.3.5 отображение R−1 является автоморфизмом представления f . Согласно теореме 3.21 и определению 3.20, множество X \{m} является множеством образующих представления f . Полученное противоречие доказывает теорему. (cid:3) Теорема 4.7. Пусть X - базис представления f . Пусть R1 : X → X ′ произвольное отображение множества X. Рассмотрим отображение Ω2-слов w[f, X, R1] : w(f, X) → w(f, X ′) удовлетворяющее условиям (1), (2), (3) теоремы 3.7, и такое, что Существует единственный эндоморфизм представлнения f 9 x ∈ X => w[f, X, R1](x) = R1(x) определённый правилом R : M → M R ◦ m = w[f, X, R1](w(f, X, m)) Доказательство. Утверждение теоремы является следствием теорем 3.7, 3.11. (cid:3) Следствие 4.8. Пусть X, X ′ - базисы представления f . Пусть R - авто- морфизм представления f такой, что X ′ = R ◦ X. Автоморфизм R определён однозначно. (cid:3) Замечание 4.9. Теорема 4.7, так же как и теорема 3.11, является теоремой о продолжении отображения. Однако здесь X - не произвольное множество об- разующих представления, а базис. Согласно замечанию 4.3, мы не можем опре- делить координаты любого элемента базиса через остальные элементы этого же базиса. Поэтому отпадает необходимость в согласованности отображения базиса с представлением. (cid:3) Мы можем представить отображение W (f, e, v) в виде матрицы столбца W (f, e, v) =   v1 ... vn   Тогда W (f, e, v)(e′) = W (f, e, v)(cid:16)e′ 1 ... n(cid:17) = e′ имеет вид произведения матриц. 9Это утверждение похоже на теорему [1]-1, с. 104.   v1 ... vn   (cid:16)e′ 1 ... n(cid:17) e′ Базис представления универсальной алгебры 15 Теорема 4.10. Набор координат W (f, X, X) соответствует тождествен- ному преобразованию W [f, X, E] = W (f, X, X) Доказательство. Утверждение теоремы следует из равенства m = W (f, X, m) ◦ X = W (f, X, m) ◦ W (f, X, X) ◦ X (cid:3) Теорема 4.11. Пусть W (f, X, R ◦ X) - множество координат автоморфиз- ма R. Определено множество координат W (f, R ◦ X, X), соответствующее автоморфизму R−1. Множество координаты W (f, R ◦ X, X) удовлетворяют равенству (4.5) W (f, X, R ◦ X) ◦ W (f, R ◦ X, X) = W (f, X, X) W [f, X, R−1] = W [f, X, R]−1 = W (f, R ◦ X, X) Доказательство. Поскольку R - автоморфизм представления f , то, согласно теореме 4.6, множество R ◦ X - базис представления f . Следовательно, опреде- лено множество координат W (f, R ◦ X, X). Равенство (4.5) следует из цепочки равенств W (f, X, R ◦ X) ◦ W (f, R ◦ X, X) = W (f, X, R ◦ X) ◦ W (f, X, R−1 ◦ X) = W (f, X, R−1 ◦ R ◦ X) = W (f, X, X) (cid:3) Замечание 4.12. В Ω2-алгебре M не существует универсального алгоритма определения множества координат W (f, R ◦ X, X) для заданного множества W (f, X, R ◦ X).10 Мы полагаем, что в теореме 4.11 этот алгоритм задан неявно. Очевидно также, что множество Ω2-слов (4.6) W (f, X, R ◦ X) ◦ W (f, R ◦ X, X) ◦ X вообще говоря, не совпадает с множеством Ω2-слов (4.7) W (f, X, X) ◦ X Теорема 4.11 утверждает, что множества Ω2-слов (4.6) и (4.7) совпадают с точ- ностью до отношения эквивалентности, порождённой представлением f . (cid:3) Теорема 4.13. Группа автоморфизмов G(f ) эффективного представления f в Ω2-алгебре M порождает эффективное представление в Ω2-алгебре M . Доказательство. Из следствия 4.8 следует, что если автоморфизм R отоб- ражает базис X в базис X ′, то множество координат W (f, X, X ′) однозначно определяет автоморфизм R. Из теоремы 3.15 следует, что множество координат W (f, X, X ′) определяет правило отображения координат относительно базиса X при автоморфизме представления f . Из равенства (3.27) следует, что авто- морфизм R действует справа на элементы Ω2-алгебры M . Из равенства (3.25) следует, что представление группы ковариантно справа. Согласно теореме 4.10 набор координат W (f, X, X) соответствует тождественному преобразованию. Из теоремы 4.11 следует, что набор координат W (f, R ◦ X, X) соответствует преобразованию, обратному преобразованию W (f, X, R ◦ X). (cid:3) 10В векторном пространстве линейному преобразованию соответствует матрица чисел. Соответственно, обратному преобразованию соответствует обратная матрица. 16 Александр Клейн 5. Многообразие базисов представления Множество B(f ) базисов представления f называется многообразием базисов представления f . Согласно теореме 4.6, автоморфизм R представления f порождает преобра- зование (5.1) R : Y → R ◦ Y W (f, X, R ◦ Y ) = W (f, X, Y ) ◦ W (f, X, R ◦ X) многообразия базисов представления. Это преобразование называется актив- ным. Согласно теореме [3]-2.3.5, определено активное представление A(f ) : G(f ) → B∗(f ) группы G(f ) в многообразии базисов B(f ). Согласно следствию 4.8, это пред- ставление однотранзитивно. Замечание 5.1. Согласно замечанию 4.3, могут существовать базисы представ- ления f , не связанные активным преобразованием. В этом случае мы в качестве многообразия базисов будем рассматривать орбиту выбранного базиса. Следо- вательно, представление f может иметь различные многообразия базисов. Мы будем предполагать, что мы выбрали многообразие базисов. (cid:3) Определение 5.2. Автоморфизм S представления A(f ) называется пассив- ным преобразованием многообразия базисов представления. Мы будем пользоваться записью S(X) = S ⋆ X для обозначения образа базиса X при пассивном преобразовании S. (cid:3) Теорема 5.3. Пассивное преобразование базиса имеет вид (5.2) S : Y → S ⋆ Y W (f, X, S ⋆ Y ) = W (f, X, S ⋆ X) ◦ W (f, X, Y ) Доказательство. Согласно равенству (5.1), активное преобразование действу- ет на координаты базиса справа. Равенство (5.2) следует из теорем [3]-2.5.12, [3]-2.5.13, [3]-2.5.15, согласно которым пассивное преобразование действует на координаты базиса слева. (cid:3) Теорема 5.4. Пусть S - пассивное преобразование многообразия базисов пред- ставления f . Пусть X - базис представления f , X ′ = S ⋆X. Пусть для базиса Y существует активное преобразование R такое, что Y = R ◦ X. Положим Y ′ = R ◦ X ′. Тогда S ⋆ Y = Y ′. Доказательство. Согласно равенству (5.1), активное преобразование коорди- нат базиса Y имеет вид (5.3) R : X ′ = S ⋆ X → Y ′ W (f, X, Y ′) = W (f, X, X ′) ◦ W (f, X, Y ) Базис представления универсальной алгебры 17 Пусть Y ′′ = S ⋆ Y . Из равенства (5.2) следует, что (5.4) S : Y = R ◦ X → Y ′′ W (f, X, Y ′′) = W (f, X, X ′) ◦ W (f, X, Y ) Из совпадения выражений в равенствах (5.3), (5.4) следует, что Y ′ = Y ′′ = S ⋆ Y . Следовательно, коммутативна диаграмма B(f ) S B(f ) R R / B(f ) S / B(f ) Теорема 5.5. Существует однотранзитивное пассивное представление P (f ) : G(f ) → ∗B(f ) Доказательство. Поскольку A(f ) - однотранзитивное представление группы G(f ), то согласно определению 5.2 и теореме [3]-2.3.5, множество пассивных преобразований порождает группу. (cid:3) Пусть пассивное преобразование S отображает базис Y в базис (cid:3) (5.5) Y ′ = S ⋆ Y W (f, X, Y ′) = W (f, X, S ⋆ X) ◦ W (f, X, Y ) Пусть пассивное преобразование T отображает базис Y ′ в базис (5.6) Y ′′ = T ⋆ Y ′ W (f, X, Y ′′) = W (f, X, T ⋆ X) ◦ W (f, X, Y ′) Обозначим T ⋆ S произведение пассивных преобразований T и S (5.7) Y ′′ = T ⋆ S ⋆ Y W (f, X, Y ′′) = W (f, X, T ⋆ S ⋆ X) ◦ W (f, X, Y ) Из равенств (5.5), (5.6) следует, что (5.8) Y ′′ = T ⋆ S ⋆ Y W (f, X, Y ′′) = W (f, X, T ⋆ X) ◦ W (f, X, S ⋆ X) ◦ W (f, X, Y ) Из равенств (5.7), (5.8) и теоремы 5.5 следует, что (5.9) W (f, X, T ⋆ S ⋆ X) = W (f, X, T ⋆ X) ◦ W (f, X, S ⋆ X) / / 18 Александр Клейн 6. Геометрический объект представления универсальной алгебры Активное преобразование изменяет базисы и элементы Ω2-алгебры согла- совано и координаты элемента относительно базиса не меняются. Пассивное преобразование меняет только базис, и это ведёт к изменению координат эле- мента относительно базиса. Допустим пассивное преобразование S ∈ G(f ) отображает базис Y ∈ B(f ) в базис Y ′ ∈ B(f ) (6.1) W (f, X, Y ′) = W (f, X, S ⋆ X) ◦ W (f, X, Y ) Допустим элемент m ∈ M имеет Ω2-слово (6.2) m = W (f, Y, m) ◦ Y относительно базиса Y и имеет Ω2-слово (6.3) m = W (f, Y ′, m) ◦ Y ′ относительно базиса Y ′. Из (6.1) и (6.3) следует, что (6.4) W (f, Y, m) ◦ Y = W (f, Y ′, m) ◦ Y ′ = W (f, Y ′, m) ◦ W (f, Y, Y ′) ◦ Y = W (f, Y ′, m) ◦ W (f, Y, S ⋆ Y ) ◦ Y Сравнивая (6.2) и (6.4) получаем, что (6.5) W (f, Y, m) = W (f, Y ′, m) ◦ W (f, Y, S ⋆ Y ) Так как S - автоморфизм представления, то из (6.5) и теоремы 4.11 следует (6.6) W (f, Y ′, m) = W (f, Y, m) ◦ W (f, S ⋆ Y, Y ) Преобразование координат (6.6) не зависит от элемента m или базиса Y , а определенно исключительно координатами элемента m относительно базиса Y . Теорема 6.1. Преобразования координат (6.6) порождают контравариант- ное эффективное T ⋆-представление группы G(f ), называемое координатным представлением в Ω2-алгебре. Доказательство. Согласно следствию 3.16, преобразование (6.6) является эн- доморфизмом представления Ω1-алгебры A в Ω2-алгебре W (f, X).11 Допустим мы имеем два последовательных пассивных преобразования S и T . Преобразование координат (6.6) соответствует пассивному преобразованию S. Преобразование координат (6.7) W (f, Y ′′, m) = W (f, Y ′, m) ◦ W (f, T ⋆ Y, Y ) соответствует пассивному преобразованию T . Произведение преобразований координат (6.6) и (6.7) имеет вид (6.8) W (f, Y ′′, m) = W (f, Y, m) ◦ W (f, S ⋆ Y, Y ) ◦ W (f, T ⋆ Y, Y ) = W (f, Y, m) ◦ W (f, T ⋆ S ⋆ Y, Y ) 11Это преобразование не порождает эндоморфизма представления f . Координаты меня- ются, поскольку меняется базис, относительно которого мы определяем координаты. Однако элемент Ω2-алгебры M , координаты которого мы рассматриваем, не меняется. Базис представления универсальной алгебры 19 и является координатным преобразованием, соответствующим пассивному пре- образованию T ⋆S. Это доказывает, что преобразования координат порождают контравариантное T ⋆-представление группы G(f ). Если координатное преобразование не изменяет координаты выбранного ба- зиса, то ему соответствует единица группы G(f ), так как пассивное представ- ление однотранзитивно. Следовательно, координатное представление эффек- тивно. (cid:3) Пусть f - представление Ω1-алгебры A в Ω2-алгебре M . Пусть g - представ- ление Ω1-алгебры A в Ω3-алгебре N . Пассивное представление gp согласовано с пассивным представлением fp, если существует гомоморфизм h группы G(f ) в группу G(g). Рассмотрим диаграмму12 ∗B(f ) H P (f ) n n n G(f ) h n n n n n n n n n f ′ / ∗B(g) 7n P (g) G(g) Так как отображения P (f ), P (g) являются изоморфизмами группы, то отоб- ражение H является гомоморфизмом групп. Следовательно, отображение f ′ является представлением группы G(f ) в многообразии базисов B(g). Согласно построению, пассивному преобразованию S Ω2-алгебры M соответствует пас- сивное преобразование H(S) Ω3-алгебры N . (6.9) Y ′ N = H(S) ⋆ YN Тогда координатное преобразование в N принимает вид (6.10) W (g, Y ′ P , m) = W (g, YN , m) ◦ W (g, H(S) ⋆ YP , YP ) Определение 6.2. Мы будем называть орбиту (W (g, YN , n) ◦ W (g, H(S) ⋆ YN , YN ), S ⋆ YM , S ∈ G(f )) геометрическим объектом в координатном представлении, опреде- лённом в Ω2-алгебре M . Для любого базиса Y ′ M = S ⋆ YM соответствующая точка (6.10) орбиты определяет координаты геометрического объекта в координатном пространстве представления относительно базиса Y ′ M . (cid:3) Определение 6.3. Мы будем называть орбиту (W (g, YN , m) ◦ W (g, H(S) ⋆ YN , YN ), H(S) ⋆ YN , S ⋆ YM , S ∈ G(f )) геометрическим объектом, определённым в Ω2-алгебре M . Для любого базиса Y ′ M = S ⋆ YM соответствующая точка (6.10) орбиты определяет ко- ординаты геометрического объекта в Ω2-алгбре M относительно базиса Y ′ M и соответствующий вектор p = W (g, Y ′ P , p) ◦ Y ′ P называется представителем геометрического объекта в Ω2-алгбре M . (cid:3) 12Мы можем ослабить определение согласованности представлений и предположить, что N - Ω2-алгебра. Тогда нам достаточно потребовать, чтобы отображение (h, H) было морфиз- мом представлений. Но тогда элементы аффинного представления не могут быть геометри- ческим объектом векторного пространства. / O O / / 7 O O 20 Александр Клейн Так как геометрический объект - это орбита представления, то согласно теореме [3]-2.4.13 определение геометрического объекта корректно. Мы будем также говорить, что p - это геометрический объект типа H Определение 6.2 строит геометрический объект в координатном простран- стве. Определение 6.3 предполагает, что мы выбрали базис представления g. Это позволяет использовать представитель геометрического объекта вместо его координат. Теорема 6.4 (принцип инвариантности). Представитель геометрическо- го объекта не зависит от выбора базиса Y ′ M . Доказательство. Чтобы определить представителя геометрического объекта, мы должны выбрать базис YM , базис YP и координаты геометрического объек- та W (g, YP , p). Соответствующий представитель геометрического объекта име- ет вид Базис Y ′ M связан с базисом YM пассивным преобразованием p = W (g, YP , p) ◦ YP Y ′ M = S ◦ YM Согласно построению это порождает пассивное преобразование (6.9) и коор- динатное преобразование (6.10). Соответствующий представитель геометриче- ского объекта имеет вид p′ = W (g, Y ′ P , p′) ◦ Y ′ P = W (g, YP , p) ◦ W (g, H(S) ⋆ YP , YP ) ◦ W (g, YP , H(S) ⋆ YP ) ◦ YP = W (g, YP , p) ◦ YP = p Следовательно, представитель геометрического объекта инвариантен относи- тельно выбора базиса. (cid:3) Теорема 6.5. Множество геометрических объектов типа H является Ω3- алгеброй. Доказательство. Пусть pi = W (g, YP , pi) ◦ YP i = 1, ..., n Для операции ω ∈ Ω3(n) мы положим (6.11) p1...pnω = W (g, YP , p1)...W (g, YP , pn)ω ◦ YP Так как для произвольного эндоморфизма S Ω2-алгебры M отображение W (g, H(S)⋆ YN , YN ) является эндоморфизмом Ω3-алгебры N , то определение (6.11) кор- ректно. (cid:3) Теорема 6.6. Определено представление Ω1-алгебры A в Ω3-алгебре N гео- метрических объектов типа H. Доказательство. Пусть Для a ∈ A мы положим p = W (g, YP , p) ◦ YP (6.12) ap = aW (g, YP , p1) ◦ YP Базис представления универсальной алгебры 21 Так как для произвольного эндоморфизма S Ω2-алгебры M отображение W (g, H(S)⋆ YN , YN ) является эндоморфизмом представления g, то определение (6.12) кор- ректно. (cid:3) 7. Примеры базиса представления универсальной алгебры 7.1. Векторное пространство. Рассмотрим векторное пространство V над полем F . Если дано множество векторов e1, ..., en, то, согласно алгоритму построения координат над векторным пространством, координаты включают такие элементы как e1 + e2 и ae1. Рекурсивно применяя правила, приведенные в определении 3.6, мы придём к выводу, что множество векторов e1, ..., en, порождает множество линейных комбинаций a1e1 + ... + anen Согласно теореме 4.2, множество векторов e1, ..., en, является базисом при условии, если для любого i, i = 1, ..., n, вектор ei не является линейной комби- нацией остальных векторов. Это требование равносильно требованию линейной независимости векторов. 7.2. Представление группы на множестве. Рассмотрим представление из примера [3]-2.6.2. Мы можем рассматривать множество M как объединение ор- бит представления группы G. В качестве базиса представления можно выбрать множество точек таким образом, что одна и только одна точка принадлежит каждой орбите представления. Если X - базис представления, A ∈ X, g ∈ G, то Ω2-слово имеет вид A + g. Поскольку на множестве M не определены опера- ции, то не существует Ω2-слово, содержащее различные элементы базиса. Если представление группы G однотранзитивно, то базис представления состоит из одной точки. Этой точкой может быть любая точка множества M . 8. Список литературы [1] Серж Ленг, Алгебра, М. Мир, 1968 [2] S. Burris, H.P. Sankappanavar, A Course in Universal Algebra, Springer- Verlag (March, 1982), eprint http://www.math.uwaterloo.ca/ snburris/htdocs/ualg.html (The Millennium Edition) [3] Aleks Kleyn, Representation Theory: Representation of Universal Algebra, Lambert Academic Publishing, 2011 [4] П. Кон, Универсальная алгебра, М., Мир, 1968 9. Предметный указатель Ω2-слово элемента представления относительно множества образующих 5 стабильное множество представления 2 суперпозиция координат эндоморфизма представления f и элемента m 9 активное представление группы G(f ) в эндоморфизм представления, многообразии базисов представления 16 вырожденный на множестве образующих X 11 активное преобразование многообразия эндоморфизм представления, невырожденный на множестве образующих X 11 базисов представления 16 базис представления 12 геометрический объект в координатном представлении, определённом в Ω2- алгебре M 19 геометрический объект типа H 20 геометрический объект, определённый в Ω2-алгбре M 19 координатное представление в Ω2- алгебре 18 координаты геометрического объекта в Ω2-алгбре M 19 координаты геометрического объекта в координатном пространстве представления 19 координаты элемента m представления f относительно множества X 6 координаты эндоморфизма представления 8 многообразие базисов представления 16 множество Ω2-слов представления 5 множество координат представления 6 множество образующих подпредставления 4 множество образующих представления 4 невырожденный эндоморфизм представления 11 отношение эквивалентности, порождённое представлением f 13 пассивное представление группы G(f ) в многообразии базисов представления 17 пассивное преобразование многообразия базисов представления 16 подпредставление представления 3 подпредставление, порождённое множеством X 4 представитель геометрического объекта в Ω2-алгбре 19 принцип инвариантности в представлении универсальной алгебры 20 22 10. Специальные символы и обозначения A(f ) активное представление группы G(f ) в многообразии базисов B(f ) 16 B(f ) многообразие базисов представления f 16 Bf структура всех подпредставлений представления f 3 J(f ) оператор замыкания представления f 3 P (f ) пассивное представление группы G(f ) в многообразии базисов B(f ) 17 R ◦ m образ m при эндоморфизме R эффективного представления 4 S ⋆ X образ базиса X при пассивном преобразовании S 16 W (f, X) множество координат представления J(f, X) 6 W (f, X, m) координаты элемента m представления f относительно множества X 6 (W (g, YN , n) ◦ W (g, H(S) ⋆ YN , YN ), S ⋆ YM , S ∈ G(f )) геометрический объект в координатном представлении, определённом в Ω2-алгебре M 19 (W (g, YN , m) ◦ W (g, H(S) ⋆ YN , YN ), H(S) ⋆ YN , S ⋆ YM , S ∈ G(f )) геометрический объект, определённый в Ω2-алгбре M 19 W (f, X, B) множество координат множества B ⊂ J(f, X) 6 w(f, X, B) множество Ω2-слов, представляющих множество B ⊂ J(f, X) 5 W (f, X, m) ◦ W (f, X, R ◦ X) суперпозиция координат эндоморфизма представления f и точки m 9 w(f, X, m) Ω2-слово, представляющее элемент m ∈ J(f, X) 5 w(f, X) множество Ω2-слов представления J(f, X) 5 23
1311.0341	1	1311	2013-11-02T04:54:43	A Symplectic Representation of $\mathrm{E}_7$	[ "math.RA" ]	We explicitly construct a particular real form of the Lie algebra $\mathfrak{e}_7$ in terms of symplectic matrices over the octonions, thus justifying the identifications $\mathfrak{e}_7\cong\mathfrak{sp}(6,\mathbb{O})$ and, at the group level, $\mathrm{E}_7\cong\mathrm{Sp}(6,\mathbb{O})$. Along the way, we provide a geometric description of the minimal representation of $\mathfrak{e}_7$ in terms of rank 3 objects called cubies.	math.RA	math	A Symplectic Representation of E7 Department of Mathematics, Oregon State University, Corvallis, OR 97331, USA Tevian Dray [email protected] Corinne A. Manogue Department of Physics, Oregon State University, Corvallis, OR 97331, USA [email protected] Robert A. Wilson School of Mathematical Sciences, Queen Mary, University of London, London E1 4NS, UK [email protected] October 26, 2013 Abstract We explicitly construct a particular real form of the Lie algebra e7 in terms of sym- plectic matrices over the octonions, thus justifying the identifications e7 ∼= sp(6, O) and, at the group level, E7 ∼= Sp(6, O). Along the way, we provide a geometric description of the minimal representation of e7 in terms of rank 3 objects called cubies. 1 Introduction The Freudenthal-Tits magic square [1, 2] of Lie algebras provides a parametrization in terms of division algebras of a family of Lie algebras that includes all of the exceptional Lie algebras except g2. The "half-split" version of the magic square, in which one of the division algebras is split, is given in Table 1. The interpretation of the Lie algebra real forms appearing in the first two rows of the magic square as su(3, K) and sl(3, K) has been discussed in [3, 4]; see also [5, 6]. Freudenthal [7] provided an algebraic description of the symplectic geometry of e7, and Barton & Sudbery [6] advanced this description to the Lie algebra level by interpreting the third row of the magic square as sp(6, K). We continue this process here, by providing a natural symplectic interpretation of the minimal representation of e7 = e7(−25). R C H O 3 1 0 2 v o N 2 ] . A R h t a m [ 1 v 1 4 3 0 . 1 1 3 1 : v i X r a R′ C′ H′ O′ su(3, R) sl(3, R) ∼= sp(6, R) f4(4) c3(3) su(3, C) sl(3, C) su(3, 3, C) e6(2) ∼= su(3, H) c3 a5(−7) ∼= sl(3, H) d6(−6) e7(−5) ∼= su(3, O) f4 e6(−26) ∼= sl(3, O) e7(−25) e8(−24) Table 1: The "half-split" 3 × 3 magic square of Lie algebras. 1 2 Freudenthal's Description of e7 Let X , Y ∈ H3(O) be elements of the Albert algebra, that is, 3 × 3 Hermitian matrices whose components are octonions. There are two natural products on the Albert algebra, namely the Jordan product and the Freudenthal product X ◦ Y = 1 2(cid:16)X Y + YX(cid:17) (1) (2) (4) X ∗ Y = X ◦ Y − 1 2(cid:16)(trX )Y + (trY)X(cid:17) + 1 2(cid:16)(trX )(trY) − tr(X ◦ Y)(cid:17)I which can be thought of as a generalization of the cross product on R3 (with the trace of the Jordan product playing the role of the dot product). The Lie algebra e6 = e6(−26) acts on the Albert algebra H3(O). The generators of e6 fall into one of three categories; there are 26 boosts, 14 derivations (elements of g2), and 38 remaining rotations (the remaining generators of f4). For both boosts and rotations, φ ∈ e6 can be treated as a 3 × 3, tracefree, octonionic matrix; boosts are Hermitian, and rotations are anti-Hermitian. Such matrices φ ∈ e6 act on the Albert algebra via X 7−→ φX + X φ† (3) where † denotes conjugate transpose (in O). Since the derivations can be obtained by suc- cessive rotations (or boosts) through nesting, it suffices to consider the boosts and rotations, that is, to consider matrix transformations. 1 The dual representation of e6 is formed by the duals φ′ of each φ ∈ e6, defined via tr(cid:0)φ(X ) ◦ Y(cid:1) = −tr(cid:0)X ◦ φ′(Y)(cid:1) for X , Y ∈ H3(O). It is easily checked that φ′ = φ on rotations, but that φ′ = −φ on boosts. Thus, φ′ = −φ† (5) for both boosts and rotations. We can regard e7 as the conformal algebra associated with e6, since e7 consists of the 78 elements of e6, together with 27 translations, 27 conformal translations, and a dilation. In fact, Freudenthal [7] represents elements of e7 as Θ = (φ, ρ, A, B) (6) where φ ∈ e6, ρ ∈ R is the dilation, and A, B ∈ H3(O) are elements of the Albert algebra, representing (null) translations. What does Θ act on? Freudenthal [7] explicitly constructs the minimal representation of e7, which consists of elements of the form 1Since all rotations can be obtained from pairs of boosts, it would be enough to consider boosts alone. P = (X , Y, p, q) (7) 2 where X , Y ∈ H3(O), and p, q ∈ R. But how are we to visualize these elements? Freudenthal does tell us that Θ acts on P via X 7−→ φ(X ) + 1 3 ρ X + 2B ∗ Y + A q Y 7−→ 2A ∗ X + φ′(Y) − p 7−→ tr(A ◦ Y) − ρ p q 7−→ tr(B ◦ X ) + ρ q 1 3 ρ Y + B p (8) (9) (10) (11) But again, how are we to visualize this action? We conclude this section by giving two further constructions due to Freudenthal [7]. There is a "super-Freudenthal" product ∗ taking elements P of the minimal representation of e7 to elements of e7, given by 2 where where P ∗ P = (φ, ρ, A, B) φ = hX , Yi ρ = − A = − 1 B = 1 4 1 tr(cid:0)X ◦ Y − pq I(cid:1) 2(cid:0)Y ∗ Y − p X(cid:1) 2(cid:0)X ∗ X − q Y(cid:1) (12) (13) (14) (15) (16) hX, Y iZ = Y ◦ (X ◦ Z) − X ◦ (Y ◦ Z) − (X ◦ Y ) ◦ Z + 1 3 tr(X ◦ Y )Z (17) Finally, e7 preserves the quartic invariant J = tr(cid:0)(X ∗ X ) ◦ (Y ∗ Y)(cid:1) − p det X − q det Y − which can be constructed using P ∗ P. 1 4(cid:0)tr(X ◦ Y) − pq(cid:1)2 (18) 3 The Symplectic Structure of so(k + 2, 2) An analogous problem has been analyzed for the 2 × 2 magic square, which is shown in Table 2; the interpretation of the first two rows was discussed in [8]; see also [5]. Dray, Huerta, and Kincaid showed first [9] (see also [10]) how to relate SO(4, 2) to SU(2, H′ ⊗ C), 2We use ∗ to denote this "super-Freudenthal" product because of its analogy to the Freudenthal product ∗, with which there should be no confusion. Neither of these products is the same as the Hodge dual map, also denoted ∗, used briefly in Sections 3 and 4. 3 R C H O R′ C′ H′ O′ so(2) ∼= su(2, R) so(2, 1) ∼= sl(2, R) so(3, 2) ∼= sp(4, R) so(3) ∼= su(2, C) so(3, 1) ∼= sl(2, C) so(4, 2) ∼= su(2, 2, C) so(5, 4) so(6, 4) so(5) ∼= su(2, H) so(5, 1) ∼= sl(2, H) so(9) ∼= su(2, O) so(9, 1) ∼= sl(2, O) so(6, 2) so(8, 4) so(10, 2) so(12, 4) Table 2: The "half-split" 2 × 2 magic square of Lie algebras. and later [11] extended their treatment to the full 2 × 2 magic square of Lie groups in Table 2. In the third row, their Clifford algebra description of SU(2, H′ ⊗ K) is equivalent to a symplectic description as Sp(4, K), with K = R, C, H, O. Explicitly, they represent so(k + 2, 2), where k = K = 1, 2, 4, 8, in terms of actions on 4 × 4 matrices of the form (19) P0 =(cid:18) p I X −eX q I(cid:19) where X is a 2 × 2 Hermitian matrix over K, representing so(k + 1, 1), p, q ∈ R, I denotes the 2 × 2 identity matrix, and tilde denotes trace-reversal, that is, eX = X − tr(X) I. The matrix P0 can be thought of as the upper right 4 × 4 block of an 8 × 8 Clifford algebra representation, and the action of so(k + 2, 2) on P0 is obtained as usual from (the restriction of) the quadratic elements of the Clifford algebra. The generators A ∈ so(k + 2, 2) can be chosen so that the action takes the form P0 7−→ AP0 ± P0A (20) where the case-dependent signs are related to the restriction from 8 × 8 matrices to 4 × 4 matrices. Following Sudbery [5], we define the elements A of the symplectic Lie algebra sp(4, K) by the condition where Solutions of (21) take the form 3 AΩ + ΩA† = 0 Ω =(cid:18) 0 −I 0(cid:19) I A = φ − 1 2 ρ I B A 2 ρ I! −φ† + 1 (21) (22) (23) where both A and B are Hermitian, tr(φ) = 0, and ρ ∈ R. But generators of so(k + 2, 2) take exactly the same form: φ represents an element of so(k +1, 1), A and B are (null) trans- lations, and ρ is the dilation. Direct computation shows that the generators A of so(k + 2, 2) 3Care must be taken with the isometry algebra of Im(K), corresponding to Im(tr(φ)) 6= 0. Such elements can however also be generated as commutators of elements of the form (23), so we do not consider them separately. 4 do indeed satisfy (21); the above construction therefore establishes the isomorphism so(k + 2, 2) ∼= sp(4, K) (24) as claimed. We can bring the representation (19) into a more explicitly symplectic form by treating X as a vector-valued 1-form, and computing its Hodge dual ∗X, defined by ∗X = Xǫ 1 ǫ =(cid:18) 0 −1 0(cid:19) where is the Levi-Civita tensor in two dimensions. Using the identity we see that P = P0 I ⊗ ǫ takes the form ǫXǫ = eXT P =(cid:18) p ǫ −(∗X)T ∗X q ǫ(cid:19) (25) (26) (27) (28) which is antisymmetric, and whose block structure is shown in Figure 1. The diagonal blocks, labeled 00 and 11, are antisymmetric, and correspond to p and q, respectively, whereas the off-diagonal blocks, labeled 01 and 10, contain equivalent information, corresponding to ∗X. Note that ∗X does not use up all of the degrees of freedom available in an off-diagonal block; the set of all antisymmetric 4 × 4 matrices is not an irreducible representation of sp(4, K). The action of sp(4, K) on P is given by P 7−→ AP + P AT (29) for A ∈ sp(4, K), that is, for A satisfying (21). action (29) is just the antisymmetric square 4 When working over K = R or C, the v ∧ w 7−→ Av ∧ w + v ∧ Aw (30) of the natural representation v 7−→ Av, with v ∈ K4. 4 Cubies Before generalizing the above construction to the 3 × 3 magic square, we first consider the analog of ∗X. Let X ∈ H3(O) be an element of the Albert algebra, which we can regard as 4Thus, (29) can be used if desired to determine the signs in (20). 5 00 01 10 11 Figure 1: The block structure of a 4 × 4 antisymmetric matrix in terms of 2 × 2 blocks. A binary labeling of the blocks is shown on the left; on the right, blocks with similar shading contain equivalent information. a vector-valued 1-form with components Xa is a vector-valued 2-form with components b, with a, b ∈ {1, 2, 3}. The Hodge dual ∗X of X (∗X )abc = Xa mǫmbc (31) where ǫabc denotes the Levi-Civita tensor in three dimensions, that is, the completely anti- symmetric tensor satisfying ǫ123 = 1 (32) and where repeated indices are summed over. We refer to ∗X as a cubie. We also introduce the dual of ǫabc, the completely antisymmetric tensor ǫabc satisfying and note the further identities ǫmnsǫmns = 6 b ǫamn ǫbmn = 2 δa ǫabm ǫcdm = δa ǫabc ǫdef = δa c c δb d δb − δa d − δa e δc d δc d δb f + δb e δb d δc f − δb e δa d δa f + δc e δc d δa f − δc e δb d δa f f e δb In particular, we have (∗X )amnǫbmn = 2Xa b Operations on the Albert algebra can be rewritten in terms of cubies. For instance, trX = (cid:0)∗(X Y)(cid:1)abc = (cid:0)∗(X ◦ Y)(cid:1)abc = tr(X ◦ Y) = = (trX )(trY) = 1 2 1 2 1 1 2 Xabc ǫabc Xamn Ypbc ǫmnp 1 4(cid:0)Xamn Ypbc + Yamn Xpbc(cid:1) ǫmnp 8(cid:0)Xamn Ypbc + Yamn Xpbc(cid:1) ǫmnp ǫbca 8(cid:0)Xamn Ypbc + Ypbc Xamn(cid:1) ǫmnp ǫbca Xabc Ydef ǫabc ǫdef 1 6 (33) (34) (35) (36) (37) (38) (39) (40) (41) (42) from which the components of ∗(X ∗ Y) can also be worked out. In the special case where the components of X and Y commute, contracting both sides of (36) with X ⊗ Y yields 1 2 Xc mYd n ǫamn ǫbcd = (X ∗ Y)a b (cid:0)∗(X ∗ Y)(cid:1)abc = 1 2 (Xb mYc n − Xc mYb n) ǫamn (43) (44) or equivalently providing two remarkably simple expressions for the Freudenthal product, albeit only in a very special case. We will return to this issue below. Lemma 1. The action of φ ∈ e6 on cubies is given by Xa mǫmbc 7−→ φa mXm nǫnbc + Xa nφ′ b mǫnmc + Xa nφ′ c mǫnbm (45) Proof. Consider the expression Qnbc = φ′ n mǫmbc + φ′ b mǫnmc + φ′ c mǫnbm (46) which is completely antisymmetric, and hence vanishes unless n, b, c are distinct. But then which vanishes, since tr(φ′) = −tr(φ) = 0. Thus, (3) becomes Qnbc = tr(φ′) ǫnbc Xa mǫmbc 7−→(cid:0)φa = φa nXn nXn m + Xa m ǫmbc + Xa nφ† n nφ′ b m(cid:1) ǫmbc mǫnmc + Xa nφ′ c mǫnbm as claimed, where we have used both (5) and (47). A similar result holds for the action of φ′. 5 The Symplectic Structure of e7 The representation (6) can be written in block form, which we also call Θ, namely 5 Θ = φ − 1 3 ρ I B A φ′ + 1 3 ρ I! (47) (48) (49) where I denotes the 3 × 3 identity matrix. By analogy with Section 3, we would like Θ to act on ∗X , which has 3 indices, and the correct symmetries to be an off-diagonal block 5The derivations g2 ⊂ e6 require nested matrix transformations of the form (49). 7 100 101 110 111 000 001 010 011 Figure 2: The block structure of a 6 × 6 × 6 antisymmetric tensor in terms of 3 × 3 × 3 "cubies". A binary labeling of the cubies is shown on the pulled-apart cube on the left; on the right, cubies with similar shading contain equivalent information. of a rank 3 antisymmetric tensor P, whose components make up a 6 × 6 × 6 cube, which we divide into 3 × 3 × 3 cubies, as shown in Figure 2; compare Figure 1. We identity the diagonal cubies, labeled 000 and 111, with p ∗I and q ∗I, respectively, the cubie labeled 011 with ∗X , the cubie labeled 100 with ∗Y, and then let antisymmetry do the rest. Explicitly, we have P abc = a ≤ 3, b ≤ 3, c ≤ 3 p ǫabc (∗Y)abc a ≥ 4, b ≤ 3, c ≤ 3 (∗X )abc a ≤ 3, b ≥ 4, c ≥ 4 a ≥ 4, b ≥ 4, c ≥ 4 q ǫabc (50) where we have introduced the convention that a = a−3, and where the remaining components are determined by antisymmetry. 6 In the complex case, we could begin with the natural action of Θ on 6-component complex vectors, and then take the antisymmetric cube, that is, we could consider the action u ∧ v ∧ w 7−→ Θu ∧ v ∧ w + u ∧ Θv ∧ w + u ∧ v ∧ Θw with u, v, w ∈ C6, or equivalently P abc 7−→ Θa mP mbc + Θb mP amc + Θc mP abm Lemma 2. The action of the dilation Θ = (0, ρ, 0, 0) ∈ e7 on P is given by (52). Proof. From (49), we have Θa b = ± 1 3 b ρ δa (51) (52) (53) 6Note that P is a cube, and has components P abc with a, b, c ∈ {1, 2, 3, 4, 5, 6}, whereas ǫabc, ∗Xabc, and ∗Yabc are the components of cubies, which are subblocks of P, with a, b, c ∈ {1, 2, 3}. 8 with the sign being negative for a = b ≤ 3 and positive for a = b ≥ 3. Thus, (52) becomes P abc 7−→ ± 1 3 ρ P abc ± 1 3 ρ P abc ± 1 3 ρ P abc (54) where the signs depend on which of a, b, c are "small" (≤ 3) or "large" (≥ 4). Examining (50), it is now easy to see that p 7→ −ρ p, q 7→ +ρ p, X 7→ + ρ 3 Y, exactly as required by (8) -- (11). 3 X , and Y 7→ − ρ Lemma 3. If the elements of A, B ∈ H3(O) commute with those of P, then the action of the translations Θ = (0, 0, A, 0) and Θ = (0, 0, 0, B) on P is given by (52). Proof. Set Θ = (0, 0, A, 0) and consider the action of Θ on p, X , Y, and q, needing to verify (8) -- (11) with φ = 0, ρ = 0, and B = 0. From (49), we have b =(Aa 0 Θa b a ≤ 3, b ≥ 4 otherwise (55) Since A has one "small" index and one "large" index, it acts as a lowering operator, e.g. mapping cubie 100 to 000, and thus maps q 7→ X 7→ Y 7→ p. In particular, this confirms the lack of a term involving A in (11). Considering terms involving q, we look at cubie 011, where the only nonzero term of (52) is (∗X )abc 7−→ Aa mq ǫmbc = q (∗A)abc which verifies (8) in this case. We next look at cubie 000, where (52) becomes p ǫabc 7−→ Aa m(∗Y)mbc + Ab m(∗Y)mca + Ac m(∗Y)mab which is clearly antisymmetric, so we can use (33) and (37) to obtain p 7−→ 1 2 Aa m(∗Y)mbc ǫabc = Aa mYm a = tr(AY) = tr(A ◦ Y) which is (10), where we have used commutativity only in the last equality. Finally, turning to cubie 100, (52) becomes (∗Y)abc 7−→ Ab m(∗X )cam + Ac m(∗X )bma = Ab mXc n ǫnam + Ac mXb n ǫnma or equivalently, using (37) and (43), 2 Ya b 7−→ 2 Ae mXf n ǫamn ǫbef = 4 (X ∗ Y)a b (56) (57) (58) (59) (60) which is (9). This entire argument can be repeated with only minor changes if Θ = (0, 0, 0, B). 9 Over R or C, we're done; Lemmas 1, 2, and 3 together suffice to show that the action (52) is the same as the Freudenthal action (8) -- (11). Unfortunately, the action (52) fails to satisfy the Jacobi identity over H or O. However, we can still use Lemmas 1, 2, and 3 to reproduce the Freudenthal action in those cases, as follows. Lemma 4. The action of Θ = (φ, 0, 0, 0) ∈ e7 on P is determined by P abc 7−→ Θa mP mbc + P amcΘb m + P abmΘc m when acting on elements of the form (50), which extends to all of e7 by antisymmetry. Proof. From (49), we have Θa φa φ′ a 0 b a ≤ 3, b ≤ 3 b a ≥ 4, b ≥ 4 otherwise b = (61) (62) (64) (65) Inserting (62) into (61) now yields precisely (45) when acting on X ; the argument for the action on Y is similar. Furthermore, using a argument similar to that used to prove Lemma 1 to begin with, (52) acts on p via p ǫabc 7−→ φa mp ǫmbc + φb mp ǫamc + φc mp ǫabm (63) which is completely antisymmetric in a, b, c, and therefore proportional to tr(φ) = 0. The argument for the action on q is similar, with φ replaced by φ′. Although (61) itself is only antisymmetric in its last two indices, that suffices to define an action on cubies 000, 011, 100, and 111; the action on the remaining 4 cubies is uniquely determined by requiring that antisymmetry be preserved. We now have all the pieces, and can state our main result. Theorem 1. The Lie algebra e7 acts symplectically on cubes, that is, e6 ⊂ e7 acts on cubes via (61), as do real translations and the dilation, and all other e7 transformations can then be constructed from these transformations using linear combinations and commutators. Proof. Lemmas 2 and 3 are unchanged by the use of (61) rather than (52), since the com- ponents of Θ commute with those of P in both cases, and Lemma 4 verifies that e6 acts via (61), as claimed. It only remains to show that the remaining generators of e7 can be obtained from these elements via commutators. Using (8) -- (11), it is straightforward to compute the commutator of two e7 transformations of the form (6). Letting φ = Q ∈ e6 be a boost, so that Q† = Q and tr(Q) = 0, and using the identity for any A, B, X ∈ H3(O), we obtain − (A ◦ B) ∗ X =(cid:0)B − tr(B)I(cid:1) ◦ (A ∗ X ) + A ∗ (B ◦ X ) (cid:2)(0, 0, A, 0), (Q, 0, 0, 0)(cid:3) = (0, 0, A ◦ Q, 0) We can therefore obtain the null translation (0, 0, Q, 0) for any tracefree Albert algebra element Q as the commutator of (0, 0, I, 0) and (Q, 0, 0, 0); a similar argument can be used to construct the null translation (0, 0, 0, Q). 10 Thus, all generators of e7 can be implemented either as a symplectic transformation on cubes via (61), or as the commutator of two such transformations. 6 Discussion We have showed that the algebraic description of the minimal representation of e7 introduced by Freudenthal naturally corresponds geometrically to a symplectic structure. Along the way, we have emphasized both the similarities and differences between e7 and so(10, 2). Both of these algebras are conformal ; their elements divide naturally into generalized rotations (e6 or so(9, 1), respectively), translations, and a dilation. Both act naturally on a representation built out of vectors (3×3 or 2×2 Hermitian octonionic matrices, respectively), together with two additional real degrees of freedom (p and q). In the 2 × 2 case, the representation (19) contains just one vector; in the 3 × 3 case (7), there are two. This at first puzzling difference is fully explained by expressing both representations as antisymmetric tensors, as in (28) and (50), respectively, and as shown geometrically in Figures 1 and 2. In the complex case, we have shown that the symplectic action (52) exactly reproduces the Freudenthal action (8) -- (11). The analogy goes even further. In 2n dimensions, there is a natural map taking two n-forms to a 2n × 2n matrix. When acting on P, this map takes the form P 7−→ P acdP ef b ǫacdef b (66) where ǫ now denotes the volume element in six dimensions, that is, the completely antisym- metric tensor with ǫ123456 = 1. It is not hard to verify that, in the complex case, (66) is (a multiple of) P ∗ P, as given by (12) -- (16). Similarly, the quartic invariant (18) can be expressed in the complex case as J ∼ P gabP cdeP f hiP jkl ǫabcdef ǫghijkl (67) up to an overall factor. Neither the form of the action (52), nor the expressions (66) and (67), hold over H or O. This failure should not be a surprise, as trilinear tensor products are not well defined over H, let alone O. Nonetheless, Theorem 1 does tell us how to extend (52) to the octonions. Although it is also possible to write down versions of (66) and (67) that hold over the octonions, by using case-dependent algorithms to determine the order of multiplication, it is not clear that such expressions have any advantage over the original expressions (12) -- (16) and (18) given by Freudenthal. Despite these drawbacks, it is clear from our construction that e7 should be regarded as a natural generalization of the traditional notion of a symplectic Lie algebra, and fully deserves the name sp(6, O). Acknowledgments We thank John Huerta for discussions, and for coining the term "cubies". This work was supported in part by the John Templeton Foundation. 11 References [1] Hans Freudenthal. Lie Groups in the Foundations of Geometry. Adv. Math., 1:145 -- 190, 1964. [2] Jacques Tits. Alg`ebres Alternatives, Alg`ebres de Jordan et Alg`ebres de Lie Exception- nelles. Indag. Math., 28:223 -- 237, 1966. [3] Tevian Dray and Corinne A. Manogue. Octonions and the Structure of E6. Comment. Math. Univ. Carolin., 51:193 -- 207, 2010. [4] Corinne A. Manogue and Tevian Dray. Octonions, E6, and Particle Physics. J. Phys.: Conference Series, 254:012005, 2010. [5] A. Sudbery. Division Algebras, (Pseudo)Orthogonal Groups and Spinors. J. Phys., A17:939 -- 955, 1984. [6] C. H. Barton and A. Sudbery. Magic Squares and Matrix Models of Lie Algebras. Adv. Math., 180:596 -- 647, 2003. [7] Hans Freudenthal. Beziehungen der E7 und E8 zur Oktavenebene, I. Proc. Kon. Ned. Akad. Wet., A57:218 -- 230, 1954. [8] Corinne A. Manogue and Jorg Schray. Finite Lorentz Transformations, Automorphisms, and Division Algebras. J. Math. Phys., 34:3746 -- 3767, 1993. [9] Joshua Kincaid and Tevian Dray. Division Algebra Representations of SO(4, 2). (2013; in preparation). [10] Joshua James Kincaid. Master's http://ir.library.oregonstate.edu/xmlui/handle/1957/30682. Oregon thesis, Division Algebra Representations State University, 2012. of SO(4, 2). at Available [11] Tevian Dray, John Huerta, and Joshua Kincaid. The 2 × 2 Lie Group Magic Square. (2013; in preparation). 12
1109.6894	1	1109	2011-09-30T16:49:19	Zero divisors in reduction algebras	[ "math.RA", "math-ph", "math-ph", "math.RT" ]	We establish the absence of zero divisors in the reduction algebra of a Lie algebra g with respect to its reductive Lie sub-algebra k. The class of reduction algebras include the Lie algebras (they arise when k is trivial) and the Gelfand--Kirillov conjecture extends naturally to the reduction algebras. We formulate the conjecture for the diagonal reduction algebras of sl type and verify it on a simplest example.	math.RA	math	Zero divisors in reduction algebras S. Khoroshkin◦⋄ and O. Ogievetsky⋆♯1 ◦Institute of Theoretical and Experimental Physics, 117218 Moscow, Russia ⋄ Higher School of Economics, Myasnitskaya 20, 101000, Moscow, Russia ⋆Centre de Physique Th´eorique2, Luminy, 13288 Marseille, France ♯ French-Russian Poncelet laboratory, UMI 2615 du CNRS Independent University of Moscow, 11 B. Vlasievski per., 119002 Moscow, Russia Abstract We establish the absence of zero divisors in the reduction algebra of a Lie algebra g with respect to its reductive Lie sub-algebra k. The class of reduction algebras include the Lie algebras (they arise when k is trivial) and the Gelfand -- Kirillov conjecture extends naturally to the reduction algebras. We formulate the conjecture for the diagonal reduction algebras of sl type and verify it on a simplest example. 1 Preliminaries Let k be a reductive Lie subalgebra of a Lie algebra g; that is, the adjoint action of k on g is completely reducible (in particular, k is reductive). Fix a triangular decomposition of the Lie algebra k, k = n− + h + n+ . (1.1) Denote by ∆+ and ∆− the sets of positive and negative roots in the root system ∆ = ∆+ ∪ ∆− of k. For each root α ∈ ∆ let hα = α∨ ∈ h be the corresponding coroot vector. Denote by U(h) the ring of fractions of the commutative algebra U(h) relative to the set of denominators { hα + l α ∈ ∆, l ∈ Z } . (1.2) The elements of this ring can also be regarded as rational functions on the vector space h∗ . The elements of U(h) ⊂ U(h) are then regarded as polynomial functions on h∗ . Let U(k) ⊂ ¯A = U(g) be the rings of fractions of the algebras U(k) and A = U(g) relative to the set of denominators (1.2). These rings are well defined, because both U(k) and U(g) satisfy the Ore condition relative to (1.2); we give a short proof in the second part of Appendix. Define Z(g, k) to be the double coset space of ¯A by its left ideal I+ := ¯An+, generated by elements of n+, and the right ideal I− := n− ¯A, generated by elements of n−, Z(g, k) := 1On leave of absence from P.N. Lebedev Physical Institute, Theoretical Department, Leninsky prospekt 53, 119991 Moscow, Russia 2Unit´e Mixte de Recherche (UMR 6207) du CNRS et des Universit´es Aix -- Marseille I, Aix -- Marseille II et du Sud Toulon -- Var; laboratoire affili´e `a la FRUMAM (FR 2291) 1 ¯A/(I+ + I− ). The space Z(g, k) is an associative algebra with respect to the multiplication map a ⋄ b := aP b . (1.3) Here P is the extremal projector [AST] of the Lie algebra k corresponding to the triangular decomposition (1.1). We call Z(g, k) the reduction algebra associated to the pair (g, k). See [Z, KO] for details. Let p be an adk-invariant complement of k in g. Choose a linear basis {pK} of p, K runs through a certain set I of indices. We assume that 1) each basis vector is a weight vector, that is [h, pK] = µK(h)pK; (1.4) 2) the set I of indices of basis vectors is equipped with a total order (cid:22), compatible with the natural order of their weights, that is, if µK − µL is a sum of simple roots of k with integer nonnegative coefficients, then necessarily K (cid:22) L. The reduction algebra has the following general properties, see [Z, KO, KO1]: (i) Z(g, k) is free as a left U(h)-module and as a right U(h)-module. As a generating (over U(h)) subspace one can take a projection of the space S(p) of symmetric tensors on p to Z(g, k), that is a subspace of Z(g, k), formed by linear combinations of images of the powers pν, where p ∈ p and ν ≥ 0. Assignments deg(fX ) = l for the image of any product of l elements from p, X = pK1pK2 · · · pKl, and deg(Y ) = 0 for any Y ∈ U(h) define the structure of a filtered algebra on Z(g, k). The subspace Z(g, k)(k) of elements of degree not greater than k is a free left U(h)-module and a free right U(h)-module, with a generating subspace formed by linear combinations of images of the powers pν, where p ∈ p and k ≥ ν ≥ 0. Moreover, the images of monomials ( ¯L is understood as the multi-index) p ¯L := pn1 L1pn2 L2 · · · pnm Lm , L1 ≺ L2 ≺ . . . ≺ Lm , k = n1 + · · · + nm , (1.5) in Z(g, k)(k) are linearly independent over U(h) and their projections to the quotient Z(g, k)(k)/Z(g, k)(k−1) form a basis of the left U(h)-module Z(g, k)(k)/Z(g, k)(k−1). (ii) The algebra Z(g, k) is the unital associative algebra, generated by U(h) and all fpL, with the weight relations (1.4) and the ordering relations fpI ⋄ fpJ = X K,L:K(cid:22)L BIJ KL fpK ⋄ fpL +X L CIJ LfpL + DIJ , I ≻ J , (1.6) where BIJ KL, CIJ L and DIJ are certain elements of U(h). 2 (iii) Assume that g is finite-dimensional. Then the monomials form a basis of the left U(h)-module Z(g, k). fpI1 ⋄ fpI2 ⋄ · · · ⋄ fpIa, I1 (cid:22) I2 (cid:22) . . . (cid:22) Ia , (1.7) Moreover, any expression in Z(g, k) can be written in the ordered form by a repeated application of (1.6) as instructions "replace the left hand side by the right hand side". The main goal of this note is to extend this list by the property (iv) The algebra Z(g, k) has no zero divisors. We discuss the possible generalizations of the Gelfand-Kirillov conjecture for reduction algebras; we also present a proof of the property (iii), different from the proof in [KO1]. 2 Absence of zero divisors Let R≺ be the set of ordering relations (1.6). The right hand sides of relations in R≺ contain quadratic, linear and degree zero terms; the algebra Z(g, k) is thus filtered by the degree in the generators fpL; in the previous section, the members of this filtrations were denoted by Z(g, k)(k). By the standard argument, it is sufficient to prove the absence of zero divisors in the associated graded algebra. Note that this graded algebra is isomorphic to the reduction algebra, related to the pair (g′, k), where g′ is the semidirect sum of k and its module p. In other words, g′ is isomorphic to k ⊕ p as a vector space, the Lie brackets between the elements of k and between elements of k and of p are the same as in g, the other brackets vanish. We thus denote the associated graded algebra by Z(g′, k); its generators we shall denote by the same symbols fpL. We have also the structure of a filtered space on U(h). The filtration is given by the degree ∂ of a rational function (the degree is defined to be the difference of the total degrees of the numerator and denominator as polynomials in several variables). The subspace U(h)(k) ⊂ U(h), k ∈ Z, consists of rational functions on h∗ of degree not greater than k. We have U(h)(k) ⊂ U(h)(k+1), ∩kU(h)(k) = 0, ∪kU(h)(k) = U(h). Moreover, hh′ ∈ U(h)(k+l), if h ∈ U(h)(k) h′ ∈ U(h)(l), that is, U(h) is a filtered ring, and the associated graded quotient gr U(h) is isomorphic to the ring U(h) of rational functions on h∗ with poles on hyperplanes hγ = 0, γ ∈ ∆. Recall that the algebra Z(g′, k) is a free left (and right) U(h)-module with a basis fp ¯L, see (1.5). For anyz = h1fp ¯K1 + · · · + hmfp ¯Km ∈ Z(g′, k) with hj ∈ U(h) we set ∂(z) ≤ k if ∂(hj) ≤ k for all j. This definition does not depend on a choice of a linear basis in S(p): 3 one can choose instead of fp ¯K the image of an arbitrary basis of symmetric algebra of p. Let Z′(g, k)(k), where k ∈ Z, be the subspace in Z(g′, k) formed by elements of degree not greater than k. We have Z′(g, k)(k) ⊂ Z′(g, k)(k+1), ∩kZ′(g, k)(k) = 0, ∪kZ′(g, k)(k) = Z(g′, k). Lemma 1. (i) Z(g′, k) is a filtered algebra with respect to the filtration {Z′(g, k)(k)}. The filtrations on Z(g′, k) and on U(h) are compatible, that is Z(g′, k) is the filtered module over the filtered ring U(h). (ii) The associated graded quotient algebra gr Z(g′, k) is isomorphic to the tensor product U(h) ⊗ S(p). Proof. The nontrivial part of both statements concerns the multiplication structure and immediately follows from the structure of the extremal projector P : it has the form P = 1 +X hixiyi , where hi ∈ U(h), ∂(hi) < 0 and xi ∈ U(n−), yi ∈ U(n+). Moving in a ⋄ b ≡ aP b mod (I+ + I− ) the elements xi to the left through a and yi to (cid:3) the right through b we find that a ⋄ b = ab + terms of lower degree ∂. Corollary 2. The algebra Z(g, k) has no zero divizors. Proof. The commutative algebra U(h) ⊗ S(p) clearly has no zero divisors. The absence of zero divisors in the filtered algebra Z(g′, k) follows, since its associated graded quotient algebra is U(h) ⊗ S(p). Then the absence of zero divisors in the filtered algebra for Z(g, k) follows, since its associated graded quotient algebra is Z(g′, k). (cid:3) 3 Quotient rings of reduction algebras Suppose that g is finite-dimensional. The arguments of the previous section imply the noetherian property of reduction algebras. Indeed, the commutative ring U(h) is noethe- rian as a localization of the noetherian polynomial ring U(h). The tensor product U(h) ⊗ S(p) is noetherian as well by the Hilbert basis theorem: "if R is a left noetherian ring, then the polynomial ring R[x] is also a left noetherian ring". The filtered ring is noethe- rian if the associated graded quotient is noetherian, see e.g., [GK]. This implies that both Z(g′, k) and Z(g, k) are noetherian rings. We summarize all statements as Proposition 3. Suppose g is finite-dimensional. Then the reduction algebra Z(g, k) is a (left) noetherian domain. 4 A (left) noetherian domain is an Ore domain, see [GK], Lemma 2. We can thus form fields of fractions of reduction algebras. It is natural to conjecture that there is a relation between the validity of the Gelfand -- Kirillov conjecture for the field of fractions of U(g) and the field of fractions of Z(g, k) (maybe up to finite extensions of the centers of the fields of fractions, as in [GK2]). In particular, we conjecture that the Gelfand -- Kirillov conjecture holds for the diagonal reduction algebra DR(cid:0)sl(n)(cid:1) (see [KO2] for definitions and notation). More precisely, we conjecture that the field of fractions of DR(cid:0)sl(n)(cid:1) is isomorphic to the field of fractions of the algebra A n(n−1) ,2(n−1), where Ak,l is the unital associative algebra, generated by 2k + l variables u1, ..., uk, v1, ..., vk, y1, ..., yl, with relations [ui, vj] = δi,j, [ui, uj] = [vi, vj] = 0, [yα, ui] = [yα, vi] = 0 and [yα, yβ] = 0. Note that the field of fractions of U(sl(n)) is isomorphic to the field of fractions of the algebra A n(n−1) 2 ,n−1. 2 Example. The diagonal reduction algebra DR(cid:0)sl(2)(cid:1) for the Lie algebra sl(2), see [KO2], has generators z+, z−, t and h with the defining relations z+t = tz+ h + 4 h + 2 , z+z− = z−z+ h(h + 3) (h + 1)(h + 2) − t2 1 h + h , tz− = z−t h + 2 h (3.1) and the h-weight relations [h, z+] = 2z+ , [h, t] = 0 , [h, z−] = 2z− . (3.2) The central elements are C (1) = (h + 2)t , C (2) = z−z+ h + 3 h + 2 + t2 1 4 + h(h + 4) . 4 The following formulas define a homomorphism from the algebra DR(cid:0)sl(2)(cid:1) to a certain localization of the algebra A1,2 (with the Weyl variables x and d ν and ζ): dx and commuting variables z− 7→ x−1 , t 7→ ν 2(E + 1) , z+ 7→ xf (E) , h 7→ 2E . (3.3) Here E is the Euler operator, E = x d dx , and f (E) := − 2(E + 1) 2E + 3 (cid:16)E(E + 2) + ν2 16(E + 1)2 + ζ(cid:17) . Now, the following formulas define a homomorphism from the algebra A1,2 to a certain localization of DR(cid:0)sl(2)(cid:1): x 7→ z−1 − , d dx 7→ 1 2 z−h , ν 7→ 2C (1) , ζ 7→ C (2) . (3.4) 5 The extensions to the fields of fractions of the maps (3.3) and (3.4) are inverse to each other and establish an isomorphism of the fields of fractions of the algebras DR(cid:0)sl(2)(cid:1) and A1,2. Moreover, through the homomorphism (3.3) the operators of the generators of the algebra DR(cid:0)sl(2)(cid:1) act naturally on the space VM with the basis {xj+M j ∈ Z}; here M provides thus a two-parameter family of representations of the algebra DR(cid:0)sl(2)(cid:1). The is either a parameter from C/Z or can be considered as a variable. This construction family is unique in the following sense. Proposition 4. Let V be the space with the basis {vj j ∈ Z}. Assume that the operators z+, z−, t and h act on V by the formulas z− : vj 7→ vj−1 , t : vj 7→ βjvj , z+ : vj 7→ γjvj+1 , h : vj 7→ αjvj with all coefficients αj, βj, γj non-vanishing. Then this module is isomorphic to VM with a non-integer M. Proof. The defining relations (3.1) -- (3.2) imply (taking into account that the coefficients are invertible): (αj + 2)βj = αjβj−1 , αj = αj−1 + 2 , γj−1 = αj − β2 j αj + αj(αj + 3) (αj + 1)(αj + 2) γj . The first two of these relations imply αj = 2j + M with some M and βj = ν αj + 2 with some ν . Let γj := αj + 3 αj + 2 γj . Then the recurrence for γj becomes γj−1 = γj + αj + 1 αj (cid:16)αj − (cid:17) . β2 j αj Substituting the expression for βj and using the identity 4(y + 1) y2(y + 2)2 = 1 y2 − 1 (y + 2)2 one easily solves the recurrence for γj and obtains the assertion. (cid:3) 6 Appendix 1. Proof of the statement (iii), section 1. The present proof uses a result about cubic monomials from [KO1] and then refers to the diamond, or composition, lemma, see [Bo, Be]. We shall prove the statement (iii) in several steps. Denote by I(fpI1 ⋄ fpI2) the right hand side of (1.6). We understand (1.6) as the set of I(fpI1 ⋄fpI2) ( stands for "replace") in the free algebra with the instructions fpI1 ⋄fpI2 weight generators fpI. Lemma 5. Any polynomial, cubic in the generators, acquires an ordered form after a repeated application of instructions (1.6). The proof is given in [KO1]. Proof of statement (iii), section 1. (a) Recall the filtration Z(g, k)(k) defined in (i), section 1. By the statement (ii), section it follows that Z(g, k)(⋄k) ⊂ Z(g, k)(k). The opposite inclusion holds as well because the filtration by the ⋄-degree. Let Z(g, k)(⋄k) be the subspace of elements of degree not greater 1, the algebra Z(g, k) is generated by fpI and, due to the form (1.6) of relations, has a than k with respect to the product ⋄. Since fpI1 ⋄ fpI2 ⋄ . . . ⋄ fpIk = fpI1PfpI2P . . . PfpIk, algebra Z(g, k) is generated by fpI. We conclude that the two filtrations coincide. Therefore, every pIgpJ pK, I (cid:22) J (cid:22) K, is in Z(g, k)(⋄3) and, by lemma 5, can be ordered. The cardinalities of the sets {pIgpJ pK I (cid:22) J (cid:22) K} and {fpI ⋄ fpJ ⋄ fpK I (cid:22) J (cid:22) K} are equal, so due to (1.5) the image of the set { fpI ⋄ fpJ ⋄ fpK I (cid:22) J (cid:22) K} is a basis of Z(g, k)(⋄3)/Z(g, k)(⋄2). To generalize this statement to higher degrees, we use the diamond lemma. (b) We shortly remind a slightly simplified version of the diamond lemma assertion, following [O]. Let A be the free associative algebra on letters {x1, . . . , xN }. Fix an order on the set {x1, . . . , xN }. Write an element f ∈ A in the reduced form, as a linear combination of different words. Denote by f the highest symbol of f , that is, the lexicographically highest word in the reduced form of the element f . Let B be a quotient algebra of A by a set S = {r1, . . . , rM } of relations. Every relation r we write in the form r = ⌊r⌋, all terms in ⌊r⌋ are smaller than r; we understand it as an instruction to replace r by the right hand side. Taking, if necessary, linear combinations of relations, we always assume that all r's are different. Let S = {r1, . . . , rM }. Given an expression, it might happen that there are different ways of applying instruc- tions to it. This is called ambiguity. An ambiguity happens iff there are (maybe after several applications of the instructions) two or more places in which different subwords of the form r, r ∈ S, enter the expression. Suppose that all possible sequences of applications of the instructions to every word terminate and lead to one and the same result; so the result depends on the initial word but not on the way of using the instructions. By construction, the resulting final expression 7 does not contain any subword from S. In this situation one says that all ambiguities are resolvable. Two types of ambiguities are called minimal; these are overlaps and inclusions. The overlap ambiguity is of the sort ri = ab and rj = bc for some words a, b, c and some i, j = 1, . . . , M. The inclusion ambiguity is of the sort ri = abc and rj = b for some words a, b, c and some i, j = 1, . . . , M. The diamond lemma states the equivalence of three assertions: (1) all ambiguities are resolvable; (2) all minimal ambiguities are resolvable; (3) B, as a vector space, possesses a basis consisting of images of normal words - words, which do not have subwords belonging to S. We return to our situation (the reduction algebra). The only minimal ambiguities for the instructions (1.6) are overlaps fpI1 ⋄ fpI2 ⋄ fpI3 with I1 ≻ I2 ≻ I3. We can start by either I12 or I23; each time we will arrive, by lemma 5, to an ordered expression. By (a) of the present proof, the ordered expressions are linearly independent. Therefore, the two expressions coincide and the ambiguities are resolvable. The set of normal words is given by (1.7), which finishes the proof of the PBW property for the reduction algebra. ⊓⊔ 2. Ore conditions. Let A be an associative algebra and S a multiplicative (that is, multiplicatively closed) subset in A. The Ore conditions, ensuring that one can localize with respect to the "set of denominators" S, are ∀ a ∈ A, s ∈ S ∃ a ∈ A, s ∈ S : as = sa , (A.1) or aS ∩ sA 6= ∅, and if sa = 0 for some s ∈ S, a ∈ A then ∃ s ∈ S : as = 0 . (A.2) Let S be multiplicativly generated by elements hµ + k, k ∈ Z; we assume that the algebra A is a sum of its h-weight components (h is the Cartan subalgebra of k), that is, each element in A can be written as a sum of h-homogeneous elements; recall that an element x is homogeneous of weight α ∈ h∗ if h(x) = α(h)x for any h ∈ h. For an arbitrary element h we denote by h the commutator with h, h(x) := [h, x]. We shall verify the Ore conditions in this situation. Condition (A.1). It is clearly sufficient to check this condition only for multiplicative generators of the set S. Let s = hµ + k and let a be an arbitrary element of A. Decompose a into a sum of h-homogeneous components, a = a1 + ... + aN . Then hµ(aj) = µjaj with 8 some numbers µj or aj(hµ + k + µj) = (hµ + k)aj for all j = 1, ..., N. Therefore, for s := (hµ + k + µ1)(hµ + k + µ2)...(hµ + k + µN ) the product as = (a1 + ... + aN )(hµ + k + µ1)(hµ + k + µ2)...(hµ + k + µN ) is divisible from the left by hµ + k, that is, representable in the form sa. (cid:3) Condition (A.2). Again, it is enough to check the condition for generators of S only: if the condition (A.2) holds for • s1 ∈ S and all {a ∈ A : s1a = 0} and for • s2 ∈ S and all {a ∈ A : s2a = 0} then s1s2a = 0(cid:0)= s1(s2a)(cid:1) implies 0 = (s2a)s1 = s2(as1) with some s1 ∈ S and therefore (as1)s2 = 0 with some s2 ∈ S. By multiplicativity, s1s2 belongs to S. Let s = hµ+k and let a be an element of A such that (hµ+k)a = 0. Let a = a1+...+aN be the h-weight decomposition of a as in the check of the condition (A.1). Then by the weight argument, (hµ + k)aj = 0 for all j = 1, ..., N. But (hµ + k)aj = aj(hµ + k + µj) by homogeneity of aj. Therefore, for s := (hµ + k + µ1)(hµ + k + µ2)...(hµ + k + µN ) we obtain as = (a1 + ... + aN )(hµ + k + µ1)...(hµ + k + µN ) = 0 . The check is completed. (cid:3) Acknowledgments S. K. was supported by the RFBR grant 11-01-00980, joint SFBRU-RFBR grant 11-02- 90453, and by Federal Agency for Science and Innovations of Russian Federation under contract 14.740.11.0347. References [AST] R. M. Asherova, Yu. F. Smirnov and V. N. Tolstoy, Description of a certain class of projection operators for complex semi-simple Lie algebras (Russian), Matem. Zametki 26 (1979) 15 -- 25. [Be] [Bo] G. Bergman, The diamond lemma for ring theory, Adv. Math. 29 (1978) 178- 218. L. Bokut′, Embeddings into simple associative algebras (Russian), Algebra i Logika 15 (1976) 117 -- 142. 9 [GK] [GK2] [KO] [KO1] I. M. Gelfand and A. A. Kirillov, Sur les corps li´es aux alg`ebres enveloppantes des alg`ebres de Lie, Inst. Hautes ´Etudes Sci. Publ. Math. 31 (1966) 5 -- 19. I. M. Gelfand and A. A. Kirillov, Structure of the field of fractions related to a simple splitting Lie algebra, Func. Anal. i Prilozhen. 3 1 (1966) 509 -- 523. S. Khoroshkin and O. Ogievetsky, Mickelsson algebras and Zhelobenko operators, J. Algebra 319 (2008) 2113 -- 2165; arXiv: math.QA/0606259 S. Khoroshkin and O. Ogievetsky, Diagonal reduction algebras of gl type; Funk- tsional. Anal. i Prilozhen. 44 no.3 (2010) 27 -- 49 (English transl.: Funct. Anal. Appl. 44 (2010) 182 -- 198); arXiv:0912.4055. [KO2] S. Khoroshkin and O. Ogievetsky, Structure Constants of Diagonal Reduction Algebras of gl Type, SIGMA 7 (2011), no. 064; arXiv:1101.2647. [O] [Z] O. Ogievetsky, Uses of quantum spaces, Contemp. Math. 294 (2002) 161 -- 232. D. Zhelobenko, Representations of reductive Lie algebras, Nauka, Moscow (1994). 10
1312.0178	2	1312	2018-01-01T14:33:51	Generalized Hopf-Ore extensions	[ "math.RA" ]	We derive necessary and sufficient conditions for an Ore extension of a Hopf algebra to have a Hopf algebra structure of a certain type. This construction generalizes the notion of Hopf-Ore extension, called a generalized Hopf-Ore extension. We describe the generalized Hopf-Ore extensions of the enveloping algebras of Lie algebras. For some Lie algebras g, the generalized Hopf-Ore extensions of U(g) are classified.	math.RA	math	GENERALIZED HOPF-ORE EXTENSIONS LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Abstract. We derive necessary and sufficient conditions for an Ore extension of a Hopf algebra to have a Hopf algebra structure of a certain type. This construction generalizes the notion of Hopf-Ore extension, called a generalized Hopf-Ore extension. We describe the generalized Hopf-Ore extensions of the enveloping algebras of Lie algebras. For some Lie algebras g, the generalized Hopf-Ore extensions of U(g) are classified. 1 ] . A R h t a m [ 2 v 8 7 1 0 . 2 1 3 1 : v i X r a Introduction For some special algebras, to investigate Hopf algebra structures over them is an effec- tive method for studying and classifying Hopf algebras. The algebraic structures of Ore extensions have been studied extensively in the past several years. In particular, one can consider Hopf algebra structures over them. For example, many Hopf algebras have been constructed and classified by means of Ore extensions [1, 2, 7, 8]. In [7], a class of Hopf algebra structures over Ore extension were defined and studied. Let A be a Hopf algebra and A[z; τ, δ] an Ore extension of A. Under certain conditions, A[z; τ, δ] becomes a Hopf al- gebra by setting ∆z = z ⊗ r1 + r2 ⊗ z for some group-like elements r1, r2 ∈ A, which is called Hopf-Ore extension of A [7, Definition 1.0]. Recently, general Hopf algebra structures over Ore extension were discussed in [3]. In this paper, we generalize the notion in [7] by setting ∆z = z ⊗ r1 + r2 ⊗ z + x ⊗ y for some r1, r2, x, y ∈ A such that A[z; τ, δ] is a Hopf algebra. We call A[z; τ, δ] with this type of Hopf algebra structure a generalized Hopf-Ore extension of A. In [9], two kinds of connected Hopf algebras A(λ1, λ2, α) and B(λ) were constructed and were used to classify connected Hopf algebras of GK-dimension three over an algebraically closed field of characteristic zero. These Hopf algebras can be regarded as the generalized Hopf-Ore extensions of the enveloping algebras of 2-dimensional Lie algebras. Moreover, the half quantum group U ≥0 q (sl(3)) (see [4]) can be constructed by the generalized Hopf-Ore extension. This paper is organized as follows. In Section 1, we give a sufficient and necessary con- dition for A[z; τ, δ] to have this type of Hopf algebra structures and study the properties of the generalized Hopf-Ore extension A[z; τ, δ]. For two such Hopf algebra structures over Ore extensions, we give a sufficient and necessary condition for them to be isomorphic. In Section 2, we describe the generalized Hopf-Ore extensions of the enveloping algebras U(g) of Lie algebras g. The generalized Hopf-Ore extensions of U(g) are classified when g is a 1-dimensional Lie algebra over an arbitrary field, a 2-dimensional Lie algebra over an arbitrary field, and an n-dimensional abelian Lie algebra over a field of characteristic zero, respectively. Throughout, we work over a field k. Let k× denote the set of all non-zero elements of k, which is a multiplicative group. We refer to [6] for basic notions concerning Hopf algebras. 2010 Mathematics Subject Classification. 16T05,16S36,16S30. Key words and phrases. Hopf algebra, Ore extension, enveloping algebra, half quantum group. This research is supported by NNSF of China (Grant No.11571298). 1 2 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN 1. Hopf algebra structures on Ore extensions Let A be a k-algebra. Let τ be an algebra endomorphism of A and δ a τ-derivation of A. The Ore extension A[z; τ, δ] of the algebra A is an algebra generated by the variable z and the algebra A with the relation If {ai i ∈ I} is a k-basis of A, {aiz j i ∈ I, j ∈ N} is a k-basis of A[z; τ, δ] (see [5, 2.1]). za = τ(a)z + δ(a), a ∈ A. (1) Furthermore, assume that A has a Hopf algebra structure. Then we can define a Hopf algebra structure on A[y; τ, δ], which generalizes the Hopf-Ore extension defined in [7]. Definition 1.1. Let A be a Hopf algebra and H = A[z; τ, δ] an Ore extension of A. If there is a Hopf algebra structure on H such that A is a Hopf subalgebra of H and ∆(z) = z ⊗ r1 + x ⊗ y + r2 ⊗ z for some r1, r2, x, y ∈ A, then H is called a generalized Hopf-Ore extension of A. In this case, we also say that H has a Hopf algebra structure determined by (r1, r2, x, y). Note that r1, r2 are nonzero. When x ⊗ y = 0, one recover the definition of usual Hopf-Ore extension in [7]. Let A be a Hopf algebra. Recall that an element g ∈ A is a group-like element if ∆g = g ⊗ g and ε(g) = 1. Let G(A) denote the group of group-like elements in A. For g, h ∈ G(A), an element a ∈ A is a (g, h)-primitive element if ∆(a) = a ⊗ g + h ⊗ a. Let Pg,h(A) denote the set of all (g, h)-primitive elements of A. When g = h = 1, a (1, 1)-primitive element is simply called a primitive element of A, and P1,1(A) is simply written as P(A). Proposition 1.2. Let A be a Hopf algebra, H = A[z; τ, δ] and r1, r2, x, y ∈ A. If H has a Hopf algebra structure determined by (r1, r2, x, y), then r1, r2 ∈ G(A), and one of the followings is satisfied: (a) x = 0 or y = 0; (b) x = αr2 and y = βr1 for some α, β ∈ k×; (c) x ∈ Pr3,r2(A) and y ∈ Pr1,r3(A) for some r3 ∈ G(A). Proof. Note that H is a free left A-module under left multiplication with the basis {zi i ≥ 0}. Consequently, H ⊗ H ⊗ H is a free left A ⊗ A ⊗ A-module with the basis {zi ⊗ z j ⊗ zs i, j, s ≥ 0}. By comparing (∆ ⊗ id)∆(z) = z ⊗ r1 ⊗ r1 + x ⊗ y ⊗ r1 + r2 ⊗ z ⊗ r1 + ∆x ⊗ y + ∆r2 ⊗ z with one gets (id ⊗ ∆)∆(z) = z ⊗ ∆r1 + x ⊗ ∆y + r2 ⊗ z ⊗ r1 + r2 ⊗ x ⊗ y + r2 ⊗ r2 ⊗ z, ∆(r1) = r1 ⊗ r1, ∆(r2) = r2 ⊗ r2, x ⊗ y ⊗ r1 + ∆(x) ⊗ y = x ⊗ ∆(y) + r2 ⊗ x ⊗ y. (2) (3) Since r1 , 0 and r2 , 0, it follows from (2) that r1 and r2 are both group-like elements. We have x = 0 or y = 0 or x, y , 0. When x, y , 0 and x = αr2 for some α ∈ k×, the equation (3) becomes αr2 ⊗ y ⊗ r1 + αr2 ⊗ r2 ⊗ y = αr2 ⊗ ∆(y) + r2 ⊗ αr2 ⊗ y. Applying ε ⊗ ε ⊗ id to both sides, one gets that y = ε(y)r1. Let β = ε(y). Then x = αr2 and y = βr1 for some α, β ∈ k× in this case. When x, y , 0 and x , αr2 for any α ∈ k×, {r2, x} are linearly independent. Hence we have ∆(x) = r2 ⊗ x + u and ∆(y) = y ⊗ r1 + w for some u, w ∈ A ⊗ A with u , 0. Then it follows from (3) that u ⊗ y = x ⊗ w, which implies u = x ⊗ u′ and w = w′ ⊗ y for some non-zero elements u′, w′ ∈ A. Hence x ⊗ u′ ⊗ y = x ⊗ w′ ⊗ y, and so u′ = w′. Thus, by GENERALIZED HOPF-ORE EXTENSIONS 3 ∆(x) = r2 ⊗ x + x ⊗ u′ and (∆ ⊗ id)∆(x) = (id ⊗ ∆)∆(x), we have that ∆u′ = u′ ⊗ u′. Hence u′ is a group-like element. This completes the proof by letting r3 = u′. (cid:3) If x = 0 or y = 0, then x ⊗ y = 0, and so all Hopf algebra structures on H determined by (r1, r2, x, y) are the same. Hence we may assume that x = y = 0 in this case. If x = αr2 and y = βr1 for some α, β ∈ k×, let z′ = z + αβr2, δ′(a) = δ(a) + αβ(r2a − τ(a)r2) for a ∈ A. Then δ′ is a τ-derivation of A, A[z′; τ, δ′] = A[z; τ, δ] as algebras, and ∆z′ = z′ ⊗ r1 + r2 ⊗ z′. Thus, A[z; τ, δ] has a Hopf algebra structure determined by (r1, r2, x, y) if and only if A[z′; τ, δ′] has a Hopf algebra structure determined by (r1, r2, 0, 0). In this case, we have A[z′; τ, δ′] = A[z; τ, δ] as Hopf algebras. Notation 1.3. Suppose that H = A[z; τ, δ] has a Hopf algebra structure determined by (r1, r2, x, y). By Proposition 1.2 and the discussion above, we assume in what follows that x is an (r3, r2)-primitive element, y is an (r1, r3)-primitive element for some group-like element r3 ∈ A. In particular, we assume that x = 0 if and only if y = 0. Corollary 1.4. Let A be a Hopf algebra and H = A[z; τ, δ], r1, r2, x, y ∈ A. If H has a Hopf algebra structure determined by (r1, r2, x, y), then ε(z) = 0, S (z) = r−1 2 xr−1 3 yr−1 1 − r−1 2 zr−1 1 . (4) (5) Proof. The first equation follows from Notation 1.3 and (ε ⊗ id)∆(z) = z. The second equation follows from Notation 1.3 and (S ⊗ id)∆(z) = ε(z) = 0. (cid:3) By Notation 1.3, we have ∆(z) = z ⊗ r1 + x ⊗ y + r2 ⊗ z, ∆(x) = x ⊗ r3 + r2 ⊗ x and ∆(y) = y ⊗ r1 + r3 ⊗ y. Replacing the generating element z by z′ = r−1 3 z, the quaternion (r1, r2, x, y) ⊗ z′, by (r′ ∆(x′) = x′ ⊗ 1 + r′ 3 y), we have ∆(z′) = z′ ⊗ r′ 2, x′, y′) = (r−1 + x′ ⊗ y′ + r′ 2 3 r1, r−1 3 x, r−1 2 ⊗ x′ and ∆(y′) = y′ ⊗ r′ 3 r2, r−1 + 1 ⊗ y′. 1, r′ 1 1 Notation 1.5. Preserving the above notation, we assume in what follows that if H = A[z; τ, δ] has a Hopf algebra structure determined by (r1, r2, x, y), then r1, r2 ∈ G(A), x ∈ P1,r2 (A), y ∈ Pr1,1(A) and the element z satisfies the relation Under this assumption, the equation (5) becomes ∆(z) = z ⊗ r1 + x ⊗ y + r2 ⊗ z. S (z) = r−1 2 (xy − z)r−1 1 . (6) (7) Let Ada(b) := P a1bS (a2) for any a, b ∈ A. The next theorem gives a sufficient and necessary condition for an Ore extension of a Hopf algebra to have a generalized Hopf- Ore extension structure, which generalizes [7, Theorem 1.3]. Theorem 1.6. Let A be a Hopf algebra, r1, r2 ∈ G(A), x ∈ P1,r2(A), y ∈ Pr1,1(A) and H = A[z; τ, δ]. Then H has a Hopf algebra structure determined by (r1, r2, x, y) if and only if the following conditions are satisfied: (a) there is a character χ : A → k such that τ(a) = P χ(a1)Adr1 (a2), ∀a ∈ A; (b) P χ(a1)Adr1 (a2) = P Adr2 (a1)χ(a2), ∀a ∈ A; (c) ∆(τ(a))(x ⊗ y) + ∆(δ(a)) = (x ⊗ y)∆(a) + P δ(a1) ⊗ r1a2 + P r2a1 ⊗ δ(a2), ∀a ∈ A. Proof. Similarly to [7, Theorem 1.3], the proof can be divided into three steps. First we show that the comultiplication ∆ of A can be extended to H = A[z; τ, δ] by (6) if and only if the conditions (a)-(c) are satisfied. Secondly, we prove that if the conditions (a)-(c) are satisfied, then H admits an extension of counit from A by (4). Lastly, we show that if the conditions (a)-(c) are satisfied, then H has antipode S extending the antipode of A by (7). 4 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Step 1. Assume that the comultiplication ∆ of A can be extended to H by (6). Then by (1), we have ∆(z)∆(a) = ∆(τ(a))∆(z) + ∆(δ(a)). By (6), we have ∆(z)∆(a) = (z ⊗ r1 + x ⊗ y + r2 ⊗ z)(X a1 ⊗ a2) = X τ(a1)z ⊗ r1a2 + X δ(a1) ⊗ r1a2 + X(x ⊗ y)(a1 ⊗ a2) + X r2a1 ⊗ τ(a2)z + X r2a1 ⊗ δ(a2) = X(τ(a1) ⊗ r1a2)(z ⊗ 1) + X(r2a1 ⊗ τ(a2))(1 ⊗ z) + X δ(a1) ⊗ r1a2 + X(x ⊗ y)(a1 ⊗ a2) + X r2a1 ⊗ δ(a2) and ∆(τ(a))∆(z) + ∆(δ(a)) = X(τ(a)1 ⊗ τ(a)2)(z ⊗ r1 + x ⊗ y + r2 ⊗ z) + ∆(δ(a)) = X(τ(a)1 ⊗ τ(a)2r1)(z ⊗ 1) + X(τ(a)1r2 ⊗ τ(a)2)(1 ⊗ z) + X(τ(a)1 ⊗ τ(a)2)(x ⊗ y) + ∆(δ(a)). It follows that (8) ∆(τ(a)) = τ(a1) ⊗ r1a2r−1 1 ∆(τ(a)) = r2a1r−1 2 ⊗ τ(a2) (9) ∆(τ(a))(x ⊗ y) + ∆(δ(a)) = X δ(a1) ⊗ r1a2 + X(x ⊗ y)(a1 ⊗ a2) + X r2a1 ⊗ δ(a2) for any a ∈ A. The last equation coincides with the equation in (c). Define χ : A → A by χ(a) := P τ(a1)r1S (a2)r−1 ∆(χ(a)) = X(τ(a1) ⊗ r1a2r−1 = X τ(a1)r1S (a4)r−1 = X τ(a1)r1S (a2)r−1 1 , where S is the antipode of A. Then 1 )(r1S (a4)r−1 1 ⊗ r1a2S (a3)r−1 1 ⊗ r1S (a3)r−1 1 ) 1 ⊗ 1 = χ(a) ⊗ 1. 1 Hence χ(a) ∈ k, and so χ can be regarded as a linear map from A to k. It is easy to check that χ(1) = 1 and χ(ab) = χ(a)χ(b) for any a, b ∈ A. Hence χ is a character of A. One can recover τ from χ as follows: P χ(a1)r1a2r−1 = τ(a). This proves (a). Then by (8) and (9), we have = P τ(a1)r1S (a2)r−1 1 r1a3r−1 1 1 X χ(a1)r1a2r−1 1 ⊗ r1a3r−1 1 = X r2a1r−1 2 ⊗ χ(a2)r1a3r−1 1 . = P r2a1r−1 Applying id ⊗ ε to both sides, one gets P χ(a1)r1a2r−1 Conversely, assume that conditions (a)-(c) are satisfied. Then (8) and (9) are clearly satis- fied too. Thus, by the computation above, one can see that the comultiplication ∆ of A can be extended to H = A[z; τ, δ] by (6). 2 χ(a2). This proves (b). 1 Step 2. Assume that the conditions (a)-(c) are satisfied. Note that ε(x) = ε(y) = 0. Apply- ing ε ⊗ ε to the equation in (c), one gets ε(δ(a)) = ε(δ(a)) + ε(δ(a)), and so ε(δ(a)) = 0 for all a ∈ A. Thus, if we put ε(z) = 0, then ε(z)ε(a) = ε(τ(a))ε(z) + ε(δ(a)) for all a ∈ A. Therefore, H admits an extension of counit from A by (4). Step 3. Assume that the conditions (a)-(c) are satisfied. In order to show that H has antipode S extending the antipode of A by (7), it is enough to show that if we put S (z) = r−1 2 (xy − z)r−1 By ε(a)1 = P a1S (a2), we have then S (a)S (z) = S (z)S (τ(a)) + S (δ(a)) for all a ∈ A. 1 0 = δ(ε(a)1) = X δ(a1S (a2)) = X δ(a1)S (a2) + X τ(a1)δ(S (a2)). (10) GENERALIZED HOPF-ORE EXTENSIONS 5 Applying m ◦ (id ⊗ S ) to the equation in (c), we have ε(a)xyr−1 1 = X τ(a)1 xyr−1 1 S (τ(a)2) + X δ(a1)S (a2)r−1 1 + X r2a1S (δ(a2)). Then by (a), (b), (9), (10) and the above equation, we have S (a)r−1 2 xyr−1 1 = X S (a1)r−1 = X S (a1)r−1 2 ε(a2)xyr−1 2 τ(a2)1 xyr−1 1 1 S (τ(a2)2) + X S (a1)r−1 2 δ(a2)S (a3)r−1 1 2 r2a2S (δ(a3)) + X S (a1)r−1 2 xyr−1 2 xyr−1 1 S (τ(a)) + X S (a1)r−1 1 S (τ(a)) − X r−1 =r−1 =r−1 2 δ(a2)S (a3)r−1 1 + S (δ(a)) 2 χ(a1)δ(S (a2))r−1 1 + S (δ(a)). Again by (a) and (b), we have = X r−1 2 (xy − z)r−1 Now putting S (z) = r−1 S (a)r−1 2 2 χ(a1)r2S (a3)r−1 2 χ(S (a2)) = r−1 2 χ(a1)τ(S (a2)). 1 . Then for any a ∈ A, we have S (a)S (z) = S (a)r−1 = S (a)r−1 2 xyr−1 1 − S (a)r−1 2 zr−1 1 2 χ(a1)δ(S (a2))r−1 1 + S (δ(a)) 1 2 (xy − z)r−1 1 S (τ(a)) − X r−1 2 χ(a1)τ(S (a2))zr−1 1 S (τ(a)) − r−1 1 S (τ(a)) − r−1 = S (z)S (τ(a)) + S (δ(a)). = r−1 2 xyr−1 − X r−1 2 xyr−1 = r−1 2 xyr−1 = r−1 1 2 zχ(a1)S (a2)r−1 2 zr−1 1 S (τ(a)) + S (δ(a)) 1 + S (δ(a)) This completes the proof. (cid:3) Corollary 1.7. Let A be a Hopf algebra and H = A[z; τ, δ]. If H has a Hopf algebra structure determined by some (r1, r2, x, y) of elements A, then (a) χ is (convolution) invertible in A∗ with χ−1 = χ ◦ S , where χ is the character determined by τ as in Theorem 1.6(a); (b) τ is an algebra automorphism with τ−1(a) = P χ−1(a1)r−1 all a ∈ A; (c) r1r2 = r2r1. 1 a2r1 = P r−1 2 a1r2χ−1(a2) for Proof. (a) is known. (b) follows from a straightforward verification. By Theorem 1.6(b), we have P χ(a1)r1a2r−1 2 χ(r1). Since χ(r1) , 0, we have r1r2 = r2r1. This shows (c). 2 χ(a2). Taking a = r1, then χ(r1)r1r1r−1 = P r2a1r−1 = r2r1r−1 (cid:3) 1 1 Corollary 1.8. Let A be a Hopf algebra and H = A[z; τ, δ]. Assume that H has a Hopf algebra structure determined by some (r1, r2, x, y) of elements of A. (a) If A is cocommutative, then r−1 (b) If A is commutative, then χ ∈ Z(A∗), the center of the dual algebra A∗ of A. 2 r1 ∈ Z(A), the center of A; 1 r2, r−1 Proof. (a) Assume that A is cocommutative. r−1 2 r1 is the inverse of r−1 P r2a2r−1 ar−1 1 r2 = P χ−1(a1)χ(a2)a3r−1 r−1 1 r2 ∈ Z(A). 2 χ(a1), which implies that P χ(a1)a2r−1 1 r2 = P χ−1(a1)χ(a2)r−1 1 r2. By Theorem 1.6(b), P χ(a1)r1a2r−1 1 r2 = P χ(a1)r−1 1 r2a3 = r−1 1 1 r2 ∈ Z(A) since = P r2a1r−1 2 χ(a2) = 1 r2a2 for any a ∈ A. Hence 1 r2a for any a ∈ A, and so It suffices to show r−1 6 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN (b) Assume A is commutative. Then the equation in Theorem 1.6(b) becomes P χ(a1)a2 = P a1χ(a2) for all a ∈ A. Hence for any f ∈ A∗, we have P χ(a1) f (a2) = P f (a1)χ(a2). Thus, χ ∈ Z(A∗). (cid:3) Corollary 1.9. If x ⊗y satisfies ∆(τ(a))(x ⊗y) = (x ⊗y)∆(a) for any a ∈ A, then the equation in Theorem 1.6(c) becomes ∆(δ(a)) = X δ(a1) ⊗ r1a2 + X r2a1 ⊗ δ(a2). (11) (cid:3) Proof. It is clear. Notation 1.10. Let A be a Hopf algebra and H = A[z; τ, δ] which has a Hopf algebra structure determined by some (r1, r2, x, y) of elements of A. Denote the generalized Hopf- Ore extension H = A[z; τ, δ] by H = A(χ, r1, r2, x, y, δ), where χ : A → k is a character such that τ(a) = P χ(a1)Adr1 (a2), r1 and r2 are group-like elements, x is a (1, r2)-primitive element, y is a (r1, 1)-primitive element, and the equations in Theorem 1.6(b) and (c) are satisfied for {χ, r1, r2, x, y, δ}. 2, x′, y′, δ′) of Hopf Two Hopf-Ore extensions H = A(χ, r1, r2, x, y, δ) and H′ = A′(χ′, r′ algebras A and A′ are said to be isomorphic if there is a Hopf algebra isomorphism Ψ : H → H′ such that Ψ(A) = A′. 1, r′ Let m′, 1, ∆′, ε′, S ′ denote the multiplication, the unit, the comultiplication, the counit and the antipode of H′, respectively. 2, x′, y′, δ′) Proposition 1.11. Two Hopf-Ore extensions A(χ, r1, r2, x, y, δ) and A′(χ′, r′ are isomorphic if there exists a scalar λ ∈ k×, a group-like element r ∈ G(A′), an element b ∈ A′ and a Hopf algebra isomorphism Φ : A → A′ such that 1, r′ i , i = 1, 2, + r′ (a) Φ(ri) = rr′ (b) ∆′(b) = b ⊗ r′ 1 (c) χ′Φ = χ, (d) δ′ = λr−1ΦδΦ−1 + δ′′, 2 ⊗ b + λr−1Φ(x) ⊗ r−1Φ(y) − x′ ⊗ y′, hence ε′(b) = 0, where δ′′ is an inner τ′-derivation of A′ defined by δ′′(a′) = τ′(a′)b − ba′ for all a′ ∈ A′, τ′ is determined by χ′ as in Theorem 1.6(a). The converse holds if A (or A′) has no zero- divisors. Proof. Let H = A(χ, r1, r2, x, y, δ) and H′ = A′(χ′, r′ 2, x′, y′, δ′), and let τ and τ′ be the algebra automorphisms of A and A′ induced by χ and χ′ as in Theorem 1.6(a), respectively. Then τ and τ′ are algebra automorphisms by Corollary 1.7(b). 1, r′ Assume that there exists some λ ∈ k×, a group-like element r ∈ G(A′), b ∈ A′ and a Hopf algebra isomorphism Φ : A → A′ such that the conditions (a)-(d) are satisfied. Let Ψ(a) = Φ(a) for all a ∈ A and Ψ(z) = λ−1r(z′ + b). Then a straightforward computation shows that Ψ can be uniquely extended to an algebra isomorphism from H to H′. Furthermore, one can check that ∆′(Ψ(z)) = (Ψ ⊗ Ψ)(∆(z)) and ε′(Ψ(z)) = ε(z). Hence Ψ is a bialgebra isomorphism. Consequently, Ψ is a Hopf algebra isomorphism since any bialgebra map between two Hopf algebras is a Hopf algebra map. Conversely, assume that A (or A′) has no zero-divisors and that there is a Hopf algebra isomorphism Ψ : H → H′ such that Ψ(A) = A′. Then Φ = ΨA is a Hopf algebra isomorphism from A to A′, and so neither of A′ and A has zero-divisors. Hence Ψ(z) = γ′z′ + η′ for some γ′, η′ ∈ A′ and γ′ , 0. Similarly, Ψ−1(z′) = γz + η for some γ, η ∈ A and γ , 0. Thus z′ = ΨΨ−1(z′) = Φ(γ)γ′z′ + Φ(γ)η′ + Φ(η). By comparing the coefficients of GENERALIZED HOPF-ORE EXTENSIONS 7 z′, we have Φ(γ)γ′ = 1. Similarly, one gets Φ−1(γ′)γ = 1 from z = Ψ−1Ψ(z). Applying Φ on it, we have γ′Φ(γ) = 1. Thus γ′ is invertible with the inverse Φ(γ). Since Ψ is a coalgebra map, ∆′(Ψ(z)) = (Ψ ⊗ Ψ)(∆(z)) and ε′(Ψ(z)) = ε(z). Hence we have ∆′(γ′)(z′ ⊗ r′ 2 ⊗ z′) + ∆′(η′) = (γ′z′ + η′) ⊗ Φ(r1) + Φ(x) ⊗ Φ(y) + Φ(r2) ⊗ (γ′z′ + η′) 1 and ε′(η′) = 0. By comparing the two sides of this equation, we have + x′ ⊗ y′ + r′ ∆′(γ′)(1 ⊗ r′ ∆′(γ′)(r′ 1) = γ′ ⊗ Φ(r1), 2 ⊗ 1) = Φ(r2) ⊗ γ′, ∆′(γ′)(x′ ⊗ y′) + ∆′(η′) = η′ ⊗ Φ(r1) + Φ(x) ⊗ Φ(y) + Φ(r2) ⊗ η′. (12) (13) (14) Applying ε′ ⊗ id to both sides of (12), we obtain Φ(r1) = ε′(γ′)−1γ′r′ 1 since γ′ is invertible. Hence ε′(γ′)−1γ′ is a group-like element. Similarly, applying id ⊗ ε′ to both sides of (13), we have Φ(r2) = ε′(γ′)−1γ′r′ 2. Let λ := ε′(γ′)−1 and r := λγ′. Then λ , 0, r is a group-like i , i = 1, 2. Let b = γ′−1η′. Then one gets the first equation in (b) element and Φ(ri) = rr′ by multiplying ∆′(γ′−1) on the two sides of (14) from the left. From ε′(η′) = 0, one gets ε′(b) = 0. Since Ψ is an algebra map, we have Ψ(z)Ψ(a) = Ψ(τ(a))Ψ(z) + Ψ(δ(a)). This means that γ′τ′(Φ(a))z′ + γ′δ′(Φ(a)) + η′Φ(a) = Φ(τ(a))γ′z′ + Φ(τ(a))η′ + Φ(δ(a)). By comparing its two sides, we have γ′τ′(Φ(a)) = Φ(τ(a))γ′, γ′δ′(Φ(a)) + η′Φ(a) = Φ(τ(a))η′ + Φ(δ(a)). (15) (16) Since Φ is a Hopf algebra isomorphism, it follows from Theorem 1.6(a) and (15) that X γ′χ′(Φ(a1))r′ 1 Φ(a2)(r′ 1 1)−1 = X χ(a1)Φ(r1)Φ(a2)Φ(r1)−1γ′ 1)−1r−1γ′ 1)−1. = X χ(a1)rr′ = X χ(a1)γ′r′ Φ(a2)(r′ Φ(a2)(r′ 1 Hence P χ′(Φ(a1))Φ(a2) = P χ(a1)Φ(a2). Applying ε′ on its both sides, one gets χ′(Φ(a)) = χ(a) for all a ∈ A, i.e., χ′Φ = χ. Since Φ is bijective, we may substitute a by Φ−1(a′) in (16), where a′ ∈ A′. Then we have γ′δ′(a′) + η′a′ = γ′τ′(a′)γ′−1η′ + ΦδΦ−1(a′), here we use the relation (15). Define δ′′ : A′ → A′ by δ′′(a′) = τ′(a′)b − ba′ for all a′ ∈ A′. Then δ′ = λr−1ΦδΦ−1 + δ′′ by γ′−1η′ = b and λ−1r = γ′. (cid:3) Corollary 1.12. Let A(χ, r1, r2, x, y, δ) be a generalized Hopf-Ore extension of a Hopf al- gebra A. Then as generalized Hopf-Ore extensions, we have (a) A(χ, r1, r2, x, y, δ) (cid:27) A(χ, r1, r2, αx, βy, αβδ) for all α, β ∈ k×. (b) A(χ, r1, r2, α(1−r2), y, δ) (cid:27) A(χ, r1, r2, 0, 0, δ+δ′), where α ∈ k and δ′(a) = α(τ(a)y−ya) for all a ∈ A. (c) A(χ, r1, r2, x, β(1−r1), δ) (cid:27) A(χ, r1, r2, 0, 0, δ+δ′′), where β ∈ k and δ′′(a) = β(τ(a)x−xa) for all a ∈ A. (d) Assume that x and y are linearly dependent and x < kG(A). Then r1 = r2 = 1. Furthermore, if char(k) , 2, then A(χ, 1, 1, x, αx, δ) (cid:27) A(χ, 1, 1, 0, 0, δ + δ′), where α ∈ k and δ′(a) = 1 (e) A(χ, r1, r2, x, y, δ) (cid:27) A′(χΦ−1, Φ(r1), Φ(r2), Φ(x), Φ(y), ΦδΦ−1), where Φ : A → A′ is a Hopf algebra isomorphism. 2 α(τ(a)x2 − x2a) for all a ∈ A. 8 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Proof. (a) Let A(χ, r1, r2, x, y, δ) = A[z; τ, δ] and A(χ, r1, r2, αx, βy, αβδ) = A[z′; τ, αβδ]. An isomorphism Ψ : A[z; τ, δ] → A[z′; τ, αβδ] is given by Ψ(a) = a for all a ∈ A and Ψ(z) = α−1β−1z′. (b) Let A(χ, r1, r2, α(1 − r2), y, δ) = A[z; τ, δ] and A(χ, r1, r2, 0, 0, δ + δ′) = A[z′; τ, δ + δ′]. An isomorphism Ψ : A[z; τ, δ] → A[z′; τ, δ + δ′] is given by Ψ(a) = a for all a ∈ A and Ψ(z) = z′ + αy. (c) Let A(χ, r1, r2, x, β(1 − r1), δ) = A[z; τ, δ] and A(χ, r1, r2, 0, 0, δ + δ′′) = A[z′; τ, δ + δ′′]. An isomorphism Ψ : A[z; τ, δ] → A[z′; τ, δ + δ′′] is given by Ψ(a) = a for all a ∈ A and Ψ(z) = z′ + βx. (d) By Notation 1.5, x is a (1, r2)-primitive element and y is an (r1, 1)-primitive element. Hence r1 = r2 = 1 since x and y are linearly dependent and x < kG(A). Assume that char(k) , 2. Let A(χ, 1, 1, x, αx, δ) = A[z; τ, δ] and A(χ, 1, 1, 0, 0, δ + δ′) = A[z′; τ, δ + δ′]. An isomorphism Ψ : A[z; τ, δ] → A[z′; τ, δ + δ′] is given by Ψ(a) = a for all a ∈ A and Ψ(z) = z′ + 1 2 αx2. (e) It follows from Proposition 1.11 by putting λ = 1, r = 1, b = 0 there. (cid:3) Example 1.13. Let A = kG be the group algebra of a group G over k. If H = A[z; τ, δ] has a Hopf algebra structure determined by (r1, r2, x, y), then x = α(1 − r2), y = β(1 − r1) for some α, β ∈ k×. By Corollary 1.12(b) and (c), H (cid:27) A(χ, r1, r2, 0, 0, δ′) as generalized Hopf-Ore extensions. Hence H is a usual Hopf-Ore extension. Thus, it follows from [7, Propositionp 2.2] that every generalized Hopf-Ore extension of A = kG is isomorphic to a usual Hopf-Ore extension A(χ, 1, r, 0, 0, δ), where χ is a group character, r is an element of the center of the group G, and δ is given by δ(g) = α(g)(1 − r)g, g ∈ G, for some 1-cocycle α ∈ Z1 χ(kG) means a linear map α : kG → k satisfying α(gh) = α(g) + χ(g)α(h) for all g, h ∈ G. χ(kG). Note that a 1-cocycle α ∈ Z1 Example 1.14. The half quantum group U ≥0 q (sl(3)) is the upper triangular Hopf subal- gebra of quantum group Uq(sl(3)). Let A be a Hopf algebra generated by K1, K2, K−1 1 , K−1 2 , E1, E2, subject to the relations KiK j = K jKi, KiK−1 KiE jK−1 = qa ji E j. i i = K−1 i Ki = 1, (17) (18) Then A is a Hopf algebra with the comultiplication ∆, counit ε and antipode S given by ∆(Ei) = Ki ⊗ Ei + Ei ⊗ 1, ∆(Ki) = Ki ⊗ Ki, ∆(K−1 ε(Ei) = 0, S (Ei) = −K−1 ε(Ki) = 1, S (Ki) = K−1 i i Ei, ) = K−1 i ⊗ K−1 i , , i i i ) = qK−1 where q ∈ k×, 1 ≤ i, j ≤ 2 and a11 = a22 = 2, a12 = a21 = −1. Define τ(Ki) = q−1Ki, τ(K−1 for i = 1, 2, τ(E1) = q−1E1 and τ(E2) = qE2. Then τ can be uniquely extended to an algebra automorphism of A. Let H = A[z; τ, 0] be the Ore extension of A. Taking r1 = 1, r2 = K1K2, x = (q − q−1)K2E1 and y = E2. Then (q − q−1)K2E1 is a (K2, K1K2)-primitive element and E2 is a (1, K2)-primitive element. Let ∆z = z ⊗ 1 + (q − q−1)K2E1 ⊗ E2 + K1K2 ⊗ z. By Proposition 1.2, H has a Hopf algebra structure determined by (1, K1K2, (q − q−1)K2E1, E2). Let I = hz − E1E2 + q−1E2E1i be the ideal of H generated by z − E1E2 + q−1E2E1. Then one can check that I is a Hopf ideal of H. Moreover, H/I is isomorphic to the half quantum group U ≥0 q (sl(3)) as a Hopf algebra. 2. Classification of generalized Hopf-Ore extensions of some Hopf algebras In this section, we investigate the generalized Hopf-Ore extensions for the enveloping al- gebras of Lie algebras. In particular, we classify the generalized Hopf-Ore extensions of GENERALIZED HOPF-ORE EXTENSIONS 9 U(g) when g is a 1-dimensional or 2-dimensional Lie algebra over an arbitrary field, or an n-dimensional abelian Lie algebra over a field with characteristic zero. By the discussion in the last section, each generalized Hopf-Ore extension A[z; τ, δ] of a Hopf algebra A can be written as A(χ, r1, r2, x, y, δ), where χ : A → k is a character, r1, r2 ∈ G(A), x ∈ P1,r2(A), y ∈ Pr1,1(A) and Theorem 1.6(b)-(c) are satisfied for (χ, r1, r2, x, y, δ). We also assume that x = 0 if and only if y = 0. 2.1. Generalized Hopf-Ore extensions of U(g). Let g be a Lie algebra and U(g) its en- veloping algebra. Then U(g) is a Hopf algebra as usual. It is well-known that U(g) has no zero-divisors. Let {aλ λ ∈ Λ} be a fixed ordered basis of g. Let us use the notations in [6, pp. 73]. Say functions n : Λ → N with finite support if n(λ) , 0 ⇔ λ ∈ {λ1, · · · , λm}, where λ1 < λ2 < · · · < λm. Then an denotes the basis monomial an(λ1) , and any b ∈ U(g) may be written as b = Pn αnan, where αn ∈ k are almost all zero. Define . Then for m ≤ n if m(λ) ≤ n(λ) for all λ ∈ Λ. Let n! = Πλ∈Λn(λ)! and  = Πλ∈Λ  · · · an(λm) an(λ2) λ2 m(λ) n(λ)  n m  all basis monomials an, λ1 λm ∆an = X 0≤m≤n n m   am ⊗ an-m. (19) Lemma 2.1. ([6, Proposition 5.5.3]) Let g be a Lie algebra over k and H = U(g). Then (a) H is connected with H0 = k1. (b) If chark = 0, then P(H) = g. (c) If chark = p > 0, then P(H) = g, the restricted Lie algebra spanned by all {apr g, r ≥ 0}. a ∈ When chark = p > 0, it follows from Lemma 2.1 that {apr This is a totally ordered set if we define apr particular, apr λ ≤ aps λ′ if and only if λ = λ′ and r = s. Lemma 2.2. Let I be an ordered set and B = {bi chark = 0 (resp., chark = p > 0). Let c ∈ U(g). = aps λ λ λ ∈ Λ, r ≥ 0} is a basis of g. λ′ when λ < λ′ or λ = λ′ and r ≤ s. In i ∈ I} a basis of g (resp., g) when (a) If chark = 0 (resp., chark = p > 2) and ∆(c) = c ⊗ 1 + 1 ⊗ c + Pi, j∈I αi jbi ⊗ b j for some αi j ∈ k, then αi j = α ji for any i, j ∈ I, i.e., ∆(c) = c ⊗ 1 + 1 ⊗ c + X αi j(bi ⊗ b j + b j ⊗ bi) + X αiibi ⊗ bi. i, j∈I,i< j i∈I In this case, c − Pi, j∈I,i< j αi jbib j − 1 (b) If chark = 2 and ∆(c) = c ⊗ 1 + 1 ⊗ c + Pi, j∈I αi jbi ⊗ b j for some αi j ∈ k, then αi j = α ji for any i , j in I and αii = 0 for any i ∈ I, i.e., ∆(c) = c ⊗ 1 + 1 ⊗ c + X αi j(bi ⊗ b j + b j ⊗ bi). i ∈ g (resp., ∈ g). 2 Pi∈I αiib2 In this case, c − Pi, j∈I,i< j αi jbib j ∈ g. i, j∈I,i< j Proof. (a) follows from the cocommutativity of U(g) immediately. For (b), by chark = 2, ∆(b2) = b2 ⊗ 1 + 1 ⊗ b2 for any b ∈ g. Hence bi ⊗ bi can't appear in the expression of ∆(c). (cid:3) The usual Hopf-Ore extension of U(g) has been classified in [7, Proposition 2.3] when chark = 0. We will discuss the generalized Hopf-Ore extension of U(g) for an arbitrary field k. 10 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Let H = U(g)[z; τ, δ] be an Ore extension of U(g). If H has a Hopf algebra structure determined by (r1, r2, x, y), then r1 = r2 = 1, x and y are primitives by Lemma 2.1. Since r1 = r2 = 1 and U(g) is cocommutative, any character χ of U(g) satisfies Theorem 1.6(b). Since U(g) is generated by g as an algebra, Theorem 1.6(a) is equivalent to that there exists a character χ of U(g) such that τ(a) = a + χ(a) for all a ∈ g. In this case, the equation in Theorem 1.6(c) becomes ∆(τ(a))(x ⊗ y) + ∆(δ(a)) = (x ⊗ y)∆(a) + X δ(a1) ⊗ a2 + X a1 ⊗ δ(a2), (20) where a ∈ U(g). We claim that if a, b ∈ U(g) satisfy (20) then so does ab. In fact, since ∆(τ(u)) = ∆(P χ(u1)u2) = P τ(u1) ⊗ u2 = P u1 ⊗ τ(u2) for any u ∈ U(g), we have ∆(τ(ab))(x ⊗ y) + ∆(δ(ab)) = ∆(τ(a))∆(τ(b))(x ⊗ y) + ∆(δ(a))∆(b) + ∆(τ(a))∆(δ(b)) = ∆(τ(a))(∆(τ(b))(x ⊗ y) + ∆(δ(b))) + ∆(δ(a))∆(b) = ∆(τ(a))((x ⊗ y)∆(b) + P δ(b1) ⊗ b2 + P b1 ⊗ δ(b2)) + ∆(δ(a))∆(b) = (∆(τ(a))(x ⊗ y) + ∆(δ(a)))∆(b) + ∆(τ(a))(P δ(b1) ⊗ b2 + P b1 ⊗ δ(b2)) = ((x ⊗ y)∆(a) + P δ(a1) ⊗ a2 + P a1 ⊗ δ(a2))∆(b) + P τ(a1)δ(b1) ⊗ a2b2 + P a1b1 ⊗ τ(a2)δ(b2) = (x ⊗ y)∆(a)∆(b) + P δ(a1)b1 ⊗ a2b2 + P a1b1 ⊗ δ(a2)b2 + P τ(a1)δ(b1) ⊗ a2b2 + P a1b1 ⊗ τ(a2)δ(b2) = (x ⊗ y)∆(ab) + P δ(a1b1) ⊗ a2b2 + P a1b1 ⊗ δ(a2b2) = (x ⊗ y)∆(ab) + P δ((ab)1 ⊗ (ab)2 + P(ab)1 ⊗ δ((ab)2). This shows the claim. Thus, Theorem 1.6(c) is equivalent to that (20) is satisfied for any a ∈ g since U(g) is generated by g as an algebra. On the other hand, when c ∈ P(U(g)), (20) becomes ∆(δ(c)) = δ(c) ⊗ 1 + 1 ⊗ δ(c) + [x, c] ⊗ y + x ⊗ [y, c] − χ(c)x ⊗ y, (21) where [u, v] = uv − vu for any u, v ∈ U(g). Summarizing the discussion above together with Theorem 1.6, we have the following proposition. Proposition 2.3. Let H be a generalized Hopf-Ore extension of U(g). Then H is iso- morphic to U(g)(χ, 1, 1, x, y, δ) for some x, y ∈ P(U(g)), a character χ of U(g) and a τ-derivation δ of U(g) such that (21) is satisfied for all c ∈ g, where τ is an algebra automorphism of U(g) determined by τ(c) = c + χ(c), ∀c ∈ g. In what follows, denote U(g)(χ, 1, 1, x, y, δ) by U(g)(χ, x, y, δ) simply. Let AutHopf(U(g)) denote the group of all Hopf algebra automorphisms of U(g). 2.2. The case of dim(g)=1. Throughout this subsection, let g = ka be a 1-dimensional Lie algebra. Then G(U(g)) = {1} by Lemma 2.1(a). For any α ∈ k, there is a character χα : U(g) → k determined by χα(a) = α. Moreover, any character of U(g) is equal to some χα with α ∈ k. Note that χ0 = ε, the counit of U(g). Each character χα induces an algebra automorphism τα of U(g) given by τα(a) = a + α as stated in Proposition 2.3. In particular, τ0 is exactly the identity map on U(g), and a τ0- derivation is a usual derivation. Obviously, a derivation δ of U(g) is uniquely determined by the value δ(a). Let δ0 be the derivation of U(g) determined by δ0(a) = a. For any α ∈ k×, one can define a Hopf algebra automorphism Φα of U(g) by Φα(a) = αa. The we have the following lemma. Lemma 2.4. The map Φ : k× → AutHopf(U(g)), α 7→ Φα, is a group isomorphism. GENERALIZED HOPF-ORE EXTENSIONS Proof. It follows from a straightforward verification. 11 (cid:3) Now let H be a generalized Hopf-Ore extension of U(g) Then by Proposition 2.3, H = U(g)(χα, x, y, δ), where α ∈ k, x, y ∈ P(U(g)) and δ is a τα-derivation of U(g) such that (21) is satisfied for all c ∈ g. In case chark = 0, x = βa and y = γa for some β, γ ∈ k by Lemma 2.1. Note that we always assume that x = 0 if and only if y = 0. By Corollary 1.12(d), H is isomorphic to a usual Hopf-Ore extension U(g)(χα, 0, 0, δ′). Then by (21), ∆(δ′(a)) = δ′(a) ⊗ 1 + 1 ⊗ δ′(a), and so δ′(a) ∈ g by Lemma 2.1. Proposition 2.5. Assume chark = 0. Then up to isomorphism, there are three generalized Hopf-Ore extensions of U(g): U(g)(ε, 0, 0, 0), U(g)(ε, 0, 0, δ0), U(g)(χ1, 0, 0, 0), where δ0 is the derivation of U(g) given before. Proof. We first show that U(g)(ε, 0, 0, 0), U(g)(ε, 0, 0, δ0) and U(g)(χ1, 0, 0, 0) are not iso- morphic to each other as generalized Hopf-Ore extensions of U(g). Since εΦ = ε , χ1 for any Hopf algebra automorphism Φ of U(g), it follows from Proposition 1.11 that both U(g)(ε, 0, 0, 0) and U(g)(ε, 0, 0, δ0) are not isomorphic to U(g)(χ1, 0, 0, 0). Suppose that U(g)(ε, 0, 0, 0) and U(g)(ε, 0, 0, δ0) were isomorphic generalized Hopf-Ore extensions of U(g). Since G(U(g)) = {1}, it follows from Proposition 1.11 that there exists an element b ∈ U(g) such that ∆(b) = b ⊗ 1 + 1 ⊗ b and δ0 = δ′′, where δ′′ is an inner derivation of U(g) defined by δ′′(a′) = a′b − ba′ for all b′ ∈ U(g). Since U(g) is commutative, δ′′ = 0. This shows δ0 = 0, a contradiction. Therefore, U(g)(ε, 0, 0, 0) and U(g)(ε, 0, 0, δ0) are not isomorphic generalized Hopf-Ore extensions of U(g). Now let H be a generalized Hopf-Ore extensions on U(g). By the discussion before, we may assume that H = U(g)(χα, 0, 0, δ), where α ∈ k and δ is a τα-derivation of U(g) with δ(a) ∈ g. Hence δ(a) = ηa for some η ∈ k. If α = 0 and η = 0, then χα = χ0 = ε and δ = 0, and hence H = U(g)(ε, 0, 0, 0). If α = 0 and η , 0, then δ = ηδ0, and so it follows from Corollary 1.12(a) that H is isomorphic to U(g)(ε, 0, 0, δ0) as a generalized Hopf-Ore extension of U(g). Finally, if α , 0, then using Proposition 1.11 with λ = 1, r = 1, b = −δ(a) and Φ = Φα, one can check that H is isomorphic to U(g)(χ1, 0, 0, 0) as a generalized Hopf-Ore extension of U(g). (cid:3) In case chark = p > 0, x, y ∈ g by Lemma 2.1. If x = y = 0, then H = U(g)(χα, 0, 0, δ), and δ(a) ∈ g as above. Now assume that x , 0 and y , 0. Then (21) becomes ∆(δ(a)) = δ(a) ⊗ 1 + 1 ⊗ δ(a) − αx ⊗ y. If chark = p = 2, then it follows from Lemma 2.2(b) that α = 0. In general, if α = 0, then τ0(a) = a and δ(a) ∈ g. Now assume that α , 0. Then chark = p > 2. By Lemma 2.2(a), we have that x = λy for some λ ∈ k×. By Corollary 1.12(d), H is isomorphic to a usual Hopf-Ore extension. Thus, we have the following proposition. Proposition 2.6. Assume that chark = p > 0. Then up to isomorphism, there are four classes of generalized Hopf-Ore extensions of U(g) as follows: (a) U(g)(ε, 0, 0, 0); (b) U(g)(ε, 0, 0, δ1), where δ1 is a derivation of U(g) with 0 , δ1(a) ∈ g; (c) U(g)(χ1, 0, 0, 0); (d) U(g)(ε, x, y, δ2), where 0 , x, y ∈ g with kx , ky, and δ2 is a derivation of U(g) with δ2(a) ∈ g. 12 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Moreover, U(g)(ε, 0, 0, 0), U(g)(ε, 0, 0, δ1), U(g)(χ1, 0, 0, 0) and U(g)(ε, x, y, δ2) are pair- wise non-isomorphic generalized Hopf-Ore extensions on U(g). Proof. Let H be a generalized Hopf-Ore extension of U(g). By the discussion above, we may assume H = U(g)(χα, 0, 0, δ) for some α ∈ k and τα-derivation δ with δ(a) ∈ g, or H = U(g)(ε, x, y, δ2) for some 0 , x, y ∈ g with kx , ky, and derivation δ2 with δ(a) ∈ g. Similarly to the proof of Proposition 2.5, one can show that as generalized Hopf-Ore ex- tensions of U(g), U(g)(ε, 0, 0, 0), U(g)(ε, 0, 0, δ1) and U(g)(χ1, 0, 0, 0) are pairwise non- isomorphic, U(g)(χ1, 0, 0, 0) and U(g)(ε, x, y, δ2) are not isomorphic, and U(g)(χα, 0, 0, δ) is isomorphic to one of U(g)(ε, 0, 0, 0), U(g)(ε, 0, 0, δ1) and U(g)(χ1, 0, 0, 0). Now we consider U(g)(ε, 0, 0, δ) and U(g)(ε, x, y, δ2), where δ = 0 or δ = δ1. If U(g)(ε, 0, 0, δ) and U(g)(ε, x, y, δ2) are isomorphic as generalized Hopf-Ore extensions of U(g), then by Proposition 1.11, there exists an element b ∈ U(g) such that ∆b = b ⊗ 1 + 1 ⊗ b − x ⊗ y. By Lemma 2.2, we have that p > 2 and kx = ky, which contradicts kx , ky. Hence U(g)(ε, x, y, δ2) is neither isomorphic to U(g)(ε, 0, 0, 0) nor isomorphic to U(g)(ε, 0, 0, δ1) as a generalized Hopf-Ore extension of U(g). This completes the proof. (cid:3) For two generalized Hopf-Ore extensions in Proposition 2.6(b) or (d), one can use Propo- sition 1.11 to determine when they are isomorphic. 2.3. The case of dim(g)=2. It is well known that up to isomorphism, there are two 2- dimensional Lie algebras over k: the abelian Lie algebra g1 = ka ⊕ kb and the non-abelian Lie algebra g2 = ka ⊕ kb with [a, b] = a. Obviously, U(g1) and U(g2) are not isomor- phic as (Hopf) algebras. Consequently, a generalized Hopf-Ore extension of U(g1) is not isomorphic to a generalized Hopf-Ore extension of U(g2) by Proposition 1.11. For any α, β ∈ k, one can define a character χα,β : U(g1) → k by χα,β(a) = α and χα,β(b) = β. Moreover, any character of U(g1) has the form χα,β for some α, β ∈ k. The algebra automorphism τα,β of U(g1) induced by χα,β is given by τα,β(a) = a + α and τα,β(b) = b + β (see Proposition 2.3). Let χ1 = χ0,1 and τ1 = τ0,1. Note that χ0,0 = ε and τ0,0 = id. Similarly, for any α ∈ k, one can define a character χα : U(g2) → k by χα(a) = 0 and χα(b) = α. Moreover, any character of U(g2) has the form χα for some α ∈ k. The algebra automorphism τα of U(g2) induced by χα is given by τα(a) = a and τα(b) = b + α. Note that χ0 = ε and τ0 = id. Let GL(gi) be the group of all linear automorphisms of gi, i = 1, 2. Under the basis {a, b} of gi, GL(gi) is isomorphic to GL2(k), the group of all invertible 2 × 2-matrices over k. For any A =  α11 α12 α21 α22  ∈ GL2(k), the corresponding linear automorphism φA ∈ GL(gi) is determined by φA(a, b) = (a, b)A. Since g1 is abelian, φA is a Lie algebra automorphism of g1 for any A ∈ GL2(k). It is easy to check that φA is a Lie algebra automorphism of g2 if and only if α21 = 0 and α22 = 1. Let G2 be the subgroup of GL2(k) consisting of all matrices of form  α β 0 1  with α ∈ k× and β ∈ k. If φA is a Lie algebra automorphism of gi, then φA can be uniquely extended to a Hopf algebra automorphism of U(gi), denoted by ΦA. Conversely, assume that chark = 0 and Φ is a Hopf algebra automorphism of U(gi). Then it follows from Lemma 2.1 that the restriction Φgi is a Lie algebra automorphism of gi. Thus, we have the following lemma. Lemma 2.7. Under the hypotheses above, we have GENERALIZED HOPF-ORE EXTENSIONS 13 (a) The map GL2(k) → AutHopf(U(g1)), A 7→ ΦA is a group monomorphism. (b) The map G2 → AutHopf(U(g2)), A 7→ ΦA is a group monomorphism. (c) If chark = 0, then the two maps in (1) and (2) are group isomorphisms. Lemma 2.8. In U(g2), we have (a) [an, b] = nan and [bn, a] = Pn−1 (b) if chark = p > 0 then [apn i=0 (−1)n−i(cid:16)n i(cid:17)abi, ∀n ≥ 1; and [bpn , b] =  a, n = 0 0, n ≥ 1 , a] = (−1)pn a, ∀n ≥ 0. Proof. (a) follows by induction on n, and (b) follows from (a). (cid:3) Now we discuss the generalized Hopf-Ore extensions on U(gi) in three cases: chark = 0, chark > 2 and chark = 2, respectively. Case 1: chark = 0. In this case, we only consider the generalized Hopf-Ore extensions of U(g2). The generalized Hopf-Ore extensions of U(g1) will be considered in Section 2.4. Let H be a generalized Hopf-Ore extension on U(g2). Then by Proposition 2.3, we may assume that H = U(g2)(χα, x, y, δ), where α ∈ k, x, y ∈ g2 and δ is a τα-derivation of U(g2) such that (21) is satisfied for all c ∈ g. If x = y = 0 or kx = ky , 0, then by Corollary 1.12(d), H (cid:27) U(g2)(χα, 0, 0, δ) with δ(g2) ⊆ g2, a usual Hopf-Ore extension of U(g2). In case that x, y ∈ g are linearly independent, we have (x, y) = (a, b)A for some A =  α11 α12 α21 α22  ∈ GL2(k). By (21), we have ∆(δ(b)) = δ(b) ⊗ 1 + 1 ⊗ δ(b) + [x, b] ⊗ y + x ⊗ [y, b] − χα(b)x ⊗ y = δ(b) ⊗ 1 + 1 ⊗ δ(b) + (2 − α)α11α12a ⊗ a − αα21α22b ⊗ b +(1 − α)α11α22a ⊗ b + (1 − α)α12α21b ⊗ a. 2 λα11α12a2 + λα12α21ab + 1 By Lemma 2.2(a), (1 − α)α11α22 = (1 − α)α12α21, and so α = 1. Let λ = det(A)−1 ∈ k× and b′ = 1 2 λα21α22b2 ∈ U(g2). Then a straightforward computation shows that ∆(b′) = b′ ⊗ 1 + 1 ⊗ b′ + λx ⊗ y − a ⊗ b. Let δ′′ : U(g2) → U(g2) be defined by δ′′(u) = τ1(u)b′ − b′u for all u ∈ U(g2), and let δ′ = λδ + δ′′. Then it follows from Proposition 1.11 that U(g2)(χ1, x, y, δ) is isomorphic, as a generalized Hopf-Ore extension of U(g2), to U(g2)(χ1, a, b, δ′). Again by (21), we have ∆(δ′(a)) = δ′(a) ⊗ 1 + 1 ⊗ δ′(a) − a ⊗ a, ∆(δ′(b)) = δ′(b) ⊗ 1 + 1 ⊗ δ′(b). Hence δ′(b) ∈ g2. By Lemma 2.2(a), δ′(a) + 1 2 a2 ∈ g2. Proposition 2.9. Assume that char(k) = 0. Then up to isomorphism, there are two classes of generalized Hopf-Ore extensions of U(g2) as follows: (a) U(g2)(χα, 0, 0, δ1), where α ∈ k and δ1 is a τα-derivation of U(g2) with δ1(g2) ⊆ g2; (b) U(g2)(χ1, a, b, δ2), where δ2 is a τ1-derivation of U(g2) with δ2(a) + 1 δ2(b) ∈ g2. 2 a2 ∈ g2 and Moreover, U(g2)(χα, 0, 0, δ1) and U(g2)(χ1, a, b, δ2) are not isomorphic generalized Hopf- Ore extensions of U(g2). Furthermore, U(g2)(χα, 0, 0, δ1) is isomorphic, as a Hopf algebra, to the enveloping algebras of some 3-dimensional Lie algebra. Proof. The first claim follows from the discussion above, the second one follows from Proposition 1.11, and the last one is obvious. (cid:3) 14 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Using Proposition 1.11, it is easy to determine when two generalized Hopf-Ore extensions in the same class of Proposition 2.9 are isomorphic. Case 2: chark = p > 2. Firstly, let H be a generalized Hopf-Ore extension of U(g1). By Proposition 2.3, we may assume that H = U(g1)(χα,β, x, y, δ), where α, β ∈ k, x, y ∈ g1 and δ is a τα,β-derivation of U(g1) such that (21) is satisfied for all c ∈ g. If x = y = 0 or kx = ky , 0, then by Corollary 1.12(d), H is isomorphic to U(g1)(χα,β, 0, 0, δ) with δ(g1) ⊆ g1. If (α, β) , (0, 0), then one can choose a matrix A ∈ GL2(k) such that (α, β) = (0, 1)A. In this case, χα,βΦ A−1 = χ1. By Corollary 1.12(e) and Lemma 2.7(a), U(g1)(χα,β, 0, 0, δ) (cid:27) U(g1)(χ1, 0, 0, ΦAδΦ A−1 ). Hence H is isomorphic to U(g1)(ε, 0, 0, δ) or U(g1)(χ1, 0, 0, δ) as a generalized Hopf-Ore extension of U(g1), where δ is a derivation or a τ1-derivation of U(g1) with δ(g1) ⊆ g1. If x, y ∈ g1 are linearly independent, then by (21) we have δ(a) = δ(a) ⊗ 1 + 1 ⊗ δ(a) − αx ⊗ y, δ(b) = δ(b) ⊗ 1 + 1 ⊗ δ(b) − βx ⊗ y. Thus, by Lemma 2.2(a), α = β = 0, and so H = U(g1)(ε, x, y, δ) with δ(g1) ⊆ g1. Summa- rizing the discussion above, we have the following proposition. Proposition 2.10. Assume that char(k) = p > 2. Then up to isomorphism, there are three classes of generalized Hopf-Ore extensions of U(g1) as follows: (a) U(g1)(ε, 0, 0, δ1), where δ1 is a derivation of U(g1) with δ1(g1) ⊆ g1; (b) U(g1)(χ1, 0, 0, δ2), where δ2 is a τ1-derivation of U(g1) with δ2(g1) ⊆ g1; (c) U(g1)(ε, x, y, δ3), where 0 , x, y ∈ g1 with kx , ky, and δ3 is a derivation of U(g1) with δ3(g1) ⊆ g1. Moreover, as generalized Hopf-Ore extensions, U(g1)(ε, 0, 0, δ1), U(g1)(χ1, 0, 0, δ2) and U(g1)(ε, x, y, δ3) are pairwise non-isomorphic. Proof. The first claim follows from the discussion above, and the second one follows from Proposition 1.11. (cid:3) Next, let H be a generalized Hopf-Ore extension of U(g2). Then similarly, we may assume that H = U(g2)(χα, x, y, δ), where α ∈ k, x, y ∈ g2 and δ is a τα-derivation of U(g2) such that (21) is satisfied for all c ∈ g. If x = y = 0 or kx = ky , 0, then H (cid:27) U(g2)(χα, 0, 0, δ) with δ(g2) ⊆ g2, a usual Hopf-Ore extension of U(g2). Now assume that x, y ∈ g2 are linearly independent. Then x = Pi≥0(αiapi ) for some almost all zero elements αi, α′ i . Then by (21) and Lemma 2.8, we have i bpi i and β′ = − Pi≥0 β′ i ∈ k. Let α′ = − Pi≥0 α′ ) and y = Pi≥0(βiapi i , βi, β′ i bpi + α′ + β′ ∆(δ(a)) = δ(a) ⊗ 1 + 1 ⊗ δ(a) + α′a ⊗ y + β′ x ⊗ a, ∆(δ(b)) = δ(b) ⊗ 1 + 1 ⊗ δ(b) + α0a ⊗ y + β0 x ⊗ a − αx ⊗ y. By Lemma 2.2(a), we have the following equations (): α′βi = β′αi, ∀i ≥ 1; (1 − α)α0βi = (1 − α)β0αi, ∀i ≥ 1; ααiβ j = αα jβi, ∀ j > i ≥ 1; (1 − α)α0β′ i αα′ = αα′ = (1 − α)β0α′ jβ′ i , ∀ j > i ≥ 0. i , ∀i ≥ 0; ααiβ′ i , ∀i ≥ 0; jβi, ∀i ≥ 1, j ≥ 0; = β′α′ = αα′ α′β′ i i β′ j j Suppose α , 1. If α0 , 0, then β0 = γα0 for some γ ∈ k. Thus, from the equations above one gets that βi = γαi, ∀i ≥ 1 and β′ i , ∀i ≥ 0. Hence y = γx, a contradiction. This i = γα′ GENERALIZED HOPF-ORE EXTENSIONS 15 shows that α0 = 0. Then we have β0αi = β0α′ = 0, ∀i ≥ 0, and so β0 = 0 since x , 0. i Furthermore, suppose α , 0. Then from the equations above, one can see that x and y are linearly dependent over k, a contradiction. Thus, we have proven that either α = 0 or α = 1, and that α0 = β0 = 0 when α = 0. In case α = 0, α0 = β0 = 0, and the equations () become α′βi = β′αi, ∀i ≥ 1; α′β′ i = β′α′ i , ∀i ≥ 0. Then a similar argument as above shows that α′ = β′ = 0. Hence δ(a), δ(b) ∈ g2. In case α = 1, the equations () become α′βi = β′αi, ∀i ≥ 1; α′β′ i αiβ j = α jβi, ∀ j > i ≥ 1; αiβ′ j = β′α′ = α′ i , ∀i ≥ 0; jβi, ∀i ≥ 1, j ≥ 0; α′ i β′ j = α′ jβ′ i , ∀ j > i ≥ 0. The above equations are equivalent to that x − α0a and y − β0a are linearly dependent. Moreover, either x − α0a , 0 or y − β0a , 0 since x and y are linearly independent. If x − α0a = 0, then y − β0a , 0, α′ = 0, and α0 , 0 by x , 0. In this case, we may assume x = a by Corollary 1.12(a), i.e., α0 = 1. Then by Lemma 2.2(a), δ(a) − 1 2 β′a2 ∈ g2 and δ(b) − 1 2 β0a2 ∈ g2. Similarly, if y − β0a = 0, then x − α0a , 0, β′ = 0, y = a, δ(a) − 1 2 α0a2 ∈ g2. Now suppose that x − α0a , 0 and y − β0a , 0. Then y − β0a = λ(x − α0a) for some λ ∈ k×. By Corollary 1.12(a), we may assume λ = 1. Then y − β0a = x − α0a and β′ = α′. Since x and y are linearly independent, β0 , α0. Hence y = (β0 − α0)a + x, and so (β0 − α0)−1y = a + (β0 − α0)−1 x. Again by Corollary 1.12(a), we may assume β0 − α0 = 1. Then y = a + x, and so 2 α′a2 ∈ g2 and δ(b) − 1 ∆(δ(a)) = δ(a) ⊗ 1 + 1 ⊗ δ(a) + α′a ⊗ a + α′(a ⊗ x + x ⊗ a), ∆(δ(b)) = δ(b) ⊗ 1 + 1 ⊗ δ(b) + α0a ⊗ a + α0(a ⊗ x + x ⊗ a) − x ⊗ x. Thus, by Lemma 2.2(a), we have δ(a)− 1 2 α′a2−α′ax ∈ g2 and δ(b)− 1 2 α0a2−α0ax+ 1 2 x2 ∈ g2. Summarizing the discussion above, we have the following proposition. Proposition 2.11. Assume that char(k) = p > 2. Then each generalized Hopf-Ore exten- sions of U(g2) is isomorphic to one of the followings: (a) U(g2)(χα, 0, 0, δ1), where δ1 is a τα-derivation of U(g2) with δ1(g2) ⊆ g2; (b) U(g2)(ε, x, y, δ2), where x = Pi≥1 αiapi , 0 and y = Pi≥1 βiapi 0 for some almost all zero elements αi, α′ i ∈ k with kx , ky and Pi≥0 α′ δ2 is a derivation of U(g2) with δ2(g2) ⊆ g2; (c) U(g2)(χ1, x, y, δ3), where 0 , x, y ∈ g2, and δ3 is a τ1-derivation of U(g2) such that one of the followings is satisfied: +Pi≥0 α′ i , βi, β′ +Pi≥0 β′ = Pi≥0 β′ i bpi , = 0, i bpi i i + β′ i bpi ) , β0a for some almost all zero elements βi, β′ 2 β′a2 ∈ g2, δ3(b) − 1 2 β0a2 ∈ g2, where β′ = − Pi≥0 β′ i ; i bpi ) , α0a for some almost all zero elements αi, α′ 2 α′a2 ∈ g2, δ3(b) − 1 2 α0a2 ∈ g2, where α′ = − Pi≥0 α′ i ; + α′ ) , α0a for some almost all zero elements αi, α′ i bpi i ∈ k, i ∈ k, δ3(a) − 1 (1) x = a, y = Pi≥0(βiapi (2) y = a, x = Pi≥0(αiapi (3) x = Pi≥0(αiapi y = a + x, δ3(a) − 1 α′ = − Pi≥0 α′ i . δ3(a) − 1 + α′ 2 α′a2 − α′ax ∈ g2 and δ3(b) − 1 2 α0a2 − α0ax + 1 i ∈ k, 2 x2 ∈ g2, where Moreover, as generalized Hopf-Ore extensions of U(g2), U(g2)(χα, 0, 0, δ1), U(g2)(ε, x, y, δ2) and U(g2)(χ1, x, y, δ3) are pairwise non-isomorphic. 16 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Proof. The first statement follows from the discussion above, and the second one follows from Proposition 1.11. (cid:3) Case 3: chark = 2. Firstly, let H be a generalized Hopf-Ore extension on U(g1). By Proposition 2.3, we may assume that H = U(g1)(χα,β, x, y, δ), where α, β ∈ k, x, y ∈ g1 and δ is a τα,β-derivation of U(g1) such that (21) is satisfied for all c ∈ g1. If x = y = 0, then an argument similar to Case 2 shows that H is isomorphic to U(g1)(ε, 0, 0, δ) or U(g1)(χ1, 0, 0, δ), where δ is a derivation or τ1-derivation of U(g1) with δ(g1) ⊆ g1. If x , 0 and y , 0, then by (21), we have ∆(δ(a)) = δ(a) ⊗ 1 + 1 ⊗ δ(a) − αx ⊗ y, ∆(δ(b)) = δ(b) ⊗ 1 + 1 ⊗ δ(b) − βx ⊗ y. By Lemma 2.2(b), α = β = 0. Hence δ(a), δ(b) ∈ g1. Thus, we have the following proposition. Proposition 2.12. Assume that char(k) = 2. Then each generalized Hopf-Ore extensions on U(g1) is isomorphic to one of the followings: (a) U(g1)(ε, 0, 0, δ1), where δ1 is a derivation of U(g1) with δ1(g1) ⊆ g1; (b) U(g1)(χ1, 0, 0, δ2), where δ2 is a τ1-derivation of U(g1) with δ2(g1) ⊆ g1; (c) U(g1)(ε, x, y, δ3), where 0 , x, y ∈ g1, and δ3 is a derivation of U(g1) with δ3(g1) ⊆ g1. Moreover, U(g1)(ε, 0, 0, δ1), U(g1)(χ1, 0, 0, δ2) and U(g1)(ε, x, y, δ3) are pairwise non-isomorphic generalized Hopf-Ore extensions on U(g1). Proof. The first statement follows from the discussion above, and the second one follows from Proposition 1.11. (cid:3) Next, let H be a generalized Hopf-Ore extension on U(g2). Then similarly, we may assume that H = U(g2)(χα, x, y, δ), where α ∈ k, x, y ∈ g2 and δ is a τα-derivation of U(g2) such that (21) is satisfied for all c ∈ g. If x = y = 0, then H (cid:27) U(g2)(χα, 0, 0, δ) with δ(g2) ⊆ g2, a usual Hopf-Ore extension on U(g2). Now assume x , 0 and y , 0. Then x = Pi≥0(αia2i i b2i ) for some almost all zero elements αi, α′ i . Then by (21) and Lemma 2.8, we have ) and y = Pi≥0(βia2i + β′ i and β′ = Pi≥0 β′ i ∈ k. Let α′ = Pi≥0 α′ i , βi, β′ + α′ i b2i ∆(δ(a)) = δ(a) ⊗ 1 + 1 ⊗ δ(a) + α′a ⊗ y + β′ x ⊗ a, ∆(δ(b)) = δ(b) ⊗ 1 + 1 ⊗ δ(b) + α0a ⊗ y + β0 x ⊗ a − αx ⊗ y. (22) (23) By Lemma 2.2(b), we have the following equations (): α′βi = β′αi, ∀i ≥ 0; (1 − α)α0βi = (1 − α)β0αi, ∀i ≥ 1; ααiβ j = αα jβi, ∀ j > i ≥ 1; (1 − α)α0β′ i αα′ i β0, ∀i ≥ 0; ααiβ′ i β′ jβi, ∀i ≥ 1, j ≥ 0; jβ′ = ααiβi = 0, ∀i ≥ 0; i , ∀ j > i ≥ 0. = αα′ = αα′ = (1 − α)α′ i , ∀i ≥ 0; = β′α′ α′β′ i αα′ j i β′ i j = (1 − α)α′ Suppose α , 0 and α , 1. If α0 , 0, then β0 = 0 by αα0β0 = 0. Then from (1 − α)α0βi = (1 − α)β0αi, ∀i ≥ 1 and (1 − α)α0β′ i β0, ∀i ≥ 0, one gets βi = 0, ∀i ≥ 1 and i β′ = 0, ∀i ≥ 0. Hence y = 0, a contradiction. Thus, α0 = 0. Similarly, β0 = 0. Since i x , 0, there exists an integer i0 ≥ 1 or i0 ≥ 0 such that αi0 , 0 = 0. Then from ααiβ j = αα jβi, ∀ j > i ≥ 1 and for some i0 ≥ 1, then βi0 ααiβ′ = 0, ∀i ≥ 0. Hence y = 0, j a contradiction. Similarly, if α′ , 0 for some i0 ≥ 0, then y = 0, a contradiction. Thus, we i0 have proven that either α = 0 or α = 1. jβi, ∀i ≥ 1, j ≥ 0, one gets βi = 0, ∀i0 , i ≥ 1 and β′ = 0 by ααi0 βi0 , 0 or α′ i0 , 0. If αi0 = αα′ i GENERALIZED HOPF-ORE EXTENSIONS 17 In case α = 0, the equations () become α′βi = β′αi, ∀i ≥ 1; α′β′ i α0βi = β0αi, ∀i ≥ 1; α0β′ i = β′α′ = α′ i , ∀i ≥ 0, i β0, ∀i ≥ 0. If kx = ky, then y = γx for some γ ∈ k×. In this case, we may assume that y = x by Corollary 1.12(a). Moreover, δ(a) − α′ax ∈ g2 and δ(b) − α0ax ∈ g2. Now assume that kx , ky. If α0 , 0, then β0 = γα0 for some γ ∈ k. Then from α0βi = β0αi, ∀i ≥ 1 and α0β′ i , ∀i ≥ 0. This implies y = γx. i By x, y , 0, γ , 0 and kx = ky, a contradiction. Hence α0 = 0. Similarly, we have β0 = 0. Then from α′βi = β′αi, ∀i ≥ 1 and α′β′ i , ∀i ≥ 0, a similar argument as above shows i that α′ = β′ = 0. In this case, δ(a) ∈ g2 and δ(b) ∈ g2. i β0, ∀i ≥ 0, one gets βi = γαi, ∀i ≥ 1 and β′ = β′α′ = γα′ = α′ i In case α = 1, the equations (*) become α′β′ α′βi = β′αi, ∀i ≥ 0; i αiβ j = α jβi, ∀ j > i ≥ 1; αiβ′ j α′ i β′ i β′ = αiβi = 0, ∀i ≥ 0; α′ i j = β′α′ = α′ = α′ i , ∀i ≥ 0; jβi, ∀i ≥ 1, j ≥ 0; jβ′ i , ∀ j > i ≥ 0. These equations implies that x − α0a and y − β0a are linearly dependent. If x − α0a = y − β0a = 0, then x = α0a and y = β0a. Since α0β0 = 0, α0 = 0 or β0 = 0, and hence x = 0 or y = 0, a contradiction. It follows that either x − α0a , 0 or y − β0a , 0. β′ i0 , 0 (or α′ i0 If x − α0a , 0, then y − β0a = γ(x − α0a) for some γ ∈ k, and there is an integer i0 ≥ 1 = 0 (or i0 ≥ 0) such that αi0 (or α′ ), hence γ = 0. Thus, y = β0a and i0 β0 , 0, which implies α0 = 0 by α0β0 = 0. In this case, we may assume y = a by Corollary 1.12(a), i.e., β0 = 1. Hence β′ = 0, and α′ = 0 by α′β0 = β′α0. Then by Lemma 2.2(b), δ(a), δ(b) ∈ g2. = 0). However, βi0 = γαi0 (or β′ i0 , 0). Hence βi0 = 0 (or β′ i0 = 0) by αi0 βi0 = γα′ i0 Similarly, if y − β0a , 0, then x − α0a = 0. We may assume x = a. Moreover, we have α′ = β0 = β′ = 0 and δ(a), δ(b) ∈ g2. In this case, H (cid:27) U(g2)(χ1, a, y, δ). We claim that U(g2)(χ1, a, y, δ) (cid:27) U(g2)(χ1, y, a, δ) as generalized Hopf-Ore extensions on U(g2). In fact, let b′ = ay. Then ∆(b′) = ∆(a)∆(y) = (a⊗1+1⊗a)(y⊗1+1⊗y) = ay⊗1+1⊗ay+a⊗y+y⊗a = b′ ⊗ 1 + 1 ⊗ b′ + a ⊗ y − y ⊗ a. Define a map δ′′ : U(g2) → U(g2) by δ′′(u) = τ1(u)b′ − b′u for any u ∈ U(g2). Then δ′′ is a τ1-derivation of U(g2). By Lemma 2.8(b), a straightforward computation shows that δ′′(a) = δ′′(b) = 0. Hence δ′′ = 0, and so δ + δ′′ = δ. Thus, it follows from Proposition 1.11 that U(g2)(χ1, a, y, δ) and U(g2)(χ1, y, a, δ) are isomorphic generalized Hopf-Ore extensions on U(g2). Summarizing the discussion above, we have the following proposition. Proposition 2.13. Assume char(k) = 2. Then each generalized Hopf-Ore extension on U(g2) is isomorphic to one in the followings: + α′ i b2i i ∈ k, α′ = Pi≥0 α′ ) , 0 for some almost all zero elements i , and δ2 is a derivation of U(g2) with δ2(a) − α′ax ∈ g2 and (a) U(g2)(χα, 0, 0, δ1), where α ∈ k and δ1 is a τα-derivation of U(g2) with δ1(g2) ⊆ g2. (b) U(g2)(ε, x, x, δ2), where x = Pi≥0(αia2i αi, α′ δ2(b) − α0ax ∈ g2. (c) U(g2)(ε, x, y, δ3), where x = Pi≥1 αia2i for some almost all zero elements αi, α′ and δ3 is a derivation of U(g2) with δ3(g2) ⊆ g2. (d) U(g2)(χ1, x, a, δ4), where x = Pi≥1 αia2i elements αi, α′ , 0, y = Pi≥1 βia2i i ∈ k with kx , ky and Pi≥0 α′ + Pi≥0 α′ = 0, and δ4 is a τ1-derivation of U(g2) with δ4(g2) ⊆ g2. i b2i + Pi≥0 β′ = Pi≥0 β′ , 0 for some almost all zero + Pi≥0 α′ , 0 = 0, i , βi, β′ i b2i i b2i i i i ∈ k with Pi≥0 α′ i Moreover, U(g2)(χα, 0, 0, δ1), U(g2)(ε, x, x, δ2), U(g2)(ε, x, y, δ3) and U(g2)(χ1, x, a, δ4) are pairwise non-isomorphic generalized Hopf-Ore extensions on U(g2). 18 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN Proof. The first statement follows from the discussion above, and the second one follows from Proposition 1.11. (cid:3) 2.4. The case of dim(g)=n with n ≥2 and char(k)=0. Throughout this subsection, as- sume that char(k) = 0 and g is an n-dimensional abelian Lie algebra. Let {a1, a2, · · · , an} be a fixed basis of g over k. For any α = (α1, α2, · · · , αn) ∈ kn, one can define a character χα : U(g) → k by χα(ai) = αi, 1 ≤ i ≤ n. Moreover, any character of U(g) is equal to some χα, α ∈ kn. The algebra automorphism τα of U(g) induced by χα is given by τα(ai) = ai + αi, 1 ≤ i ≤ n (see Proposition 2.3). Note that χ0 = ε and τ0 = id. Let 1 = (1, 0, · · · , 0) ∈ kn. Let Mn(k) be the algebra of all n × n-matrices over k, and GLn(k) the group of all invertible matrices in Mn(k). For any A = (αi j) ∈ Mn(k), one can define a linear endomorphism φA ∈ Endk(g) by φA(ai) = Pn j=1 α jia j, 1 ≤ i ≤ n. Moreover, the map φ : Mn(k) → Endk(g), A 7→ φA, is an algebra isomorphism. Let GL(g) be the group of all linear automorphisms of g. Then A ∈ GLn(k) if and only if φA ∈ GL(g). Since g is abelian, φA is a Lie algebra endomorphism (or automorphism) of g for any A ∈ Mn(k) (or GLn(k)). Therefore φA can be uniquely extended to a Hopf algebra endomorphism (or automorphism) ΦA of U(g). Conversely, since chark = 0, any Hopf algebra endomorphism (or automorphism) of U(g) is equal to some ΦA, A ∈ Mn(k) (or A ∈ GLn(k)). Thus, we have the following lemma. Lemma 2.14. Under the hypotheses above, we have (a) The map φ : Mn(k) → Endk(g), A 7→ φA is an algebra isomorphism. (b) The map Φ : GLn(k) → AutHopf(U(g)), A 7→ ΦA is a group isomorphism. Let α ∈ kn. Since U(g) is generated as an algebra by g, a τα-derivation of U(g) is deter- mined by its value on g. Suppose that δ is a τα-derivation of U(g). Since g is abelian, δ(aia j) = δ(a jai) for all 1 ≤ i, j ≤ n, which means that αiδ(a j) = α jδ(ai), 1 ≤ i, j ≤ n. If δ(g) ⊆ g, then the restriction δg can be regarded as a linear endomorphism of g, i.e., δg ∈ End(g). By Lemma 2.14(a), there is a matrix A = (αi j) ∈ Mn(k) such that δg = φA. In this case, αiδ(a j) = α jδ(ai) 1 ≤ i, j ≤ n, if and only if αiαl j = α jαli, 1 ≤ i, j, l ≤ n. (24) Conversely, if a matrix A = (αi j) ∈ Mn(k) satisfies (24), then φA can be uniquely extended to a τα-derivation of U(g), denoted by δα,A. In particular, if α = 0, δ0,A is a derivation of U(g) for any A ∈ Mn(k). Now let H be a generalized Hopf-Ore extension on U(g). By Proposition 2.3, we may assume that H = U(g)(χα, x, y, δ), where α = (α1, α2, · · · , αn) ∈ kn, x, y ∈ g and δ is a τα-derivation of U(g) such that (21) is satisfied for all c ∈ g. Firstly, assume that x = y = 0 or kx = ky , 0. Then by Corollary 1.12(d), H (cid:27) U(g)(χα, 0, 0, δ), a usual Hopf-Ore extension over U(g). By Lemma 2.1 and (21), one knows that δ(g) ⊆ g. If α , 0, then we may choose a matrix P ∈ GLn(k) such that α = 1P. In this case, χα = χ1ΦP, and it follows from Corollary 1.12(e) and Lemma 2.7 that H is isomorphic to U(g)(χ1, 0, 0, ΦPδΦ P−1(g) ⊆ g, and hence P−1 = δ1,A for some A = (αi j) ∈ Mn(k). Since δ1,A is a τ1-derivation, it follows from ΦPδΦ (24) that αl j = 0 for all 1 ≤ l ≤ n and 2 ≤ j ≤ n. Therefore, δ1,A(a1) = Pn j=1 α j1a j and δ1,A(ai) = 0 for all 2 ≤ i ≤ n. Let b = −δ1,A(a1). Then ∆(b) = b ⊗ 1 + 1 ⊗ b. Define δ′′ : U(g) → U(g) by δ′′(u) = τ1(u)b − bu, u ∈ U(g). Then δ′′(a1) = τ1(a1)b − ba1 = (a1 + 1)b − ba1 = b = −δ1,A(a1) and δ′′(ai) = τ1(ai)b − bai = aib − bai = 0 for all 2 ≤ i ≤ n. Hence δ′′ = −δ1,A, and so U(g)(χ1, 0, 0, δ1,A) (cid:27) U(g)(χ1, 0, 0, 0) by Propo- sition 1.11. Let A, B ∈ Mn(k). By Proposition 1.11 and Lemma 2.14(b), one can check P−1). Obviously, ΦPδΦ GENERALIZED HOPF-ORE EXTENSIONS 19 that U(g)(ε, 0, 0, δ0,A) (cid:27) U(g)(ε, 0, 0, δ0,B) as generalized Hopf-Ore extensions if and only if B = λPAP−1 for some λ ∈ k× and P ∈ GLn(k). Now assume that x, y ∈ g are linearly independent. Then there is an invertible matrix A in GLn(k) such that ΦA(x) = a1 and ΦA(y) = a2. Since χαΦ A−1 = χαA−1 , we have general- ized Hopf-Ore extension isomorphism U(g)(χα, x, y, δ) (cid:27) U(g)(χαA−1, a1, a2, ΦAδΦ A−1 ) by Corollary 1.12(e). Thus, we may assume that H = U(g)(χα, a1, a2, δ). Then by (21), we have ∆(δ(ai)) = δ(ai) ⊗ 1 + 1 ⊗ δ(ai) − αia1 ⊗ a2, 1 ≤ i ≤ n. By Lemma 2.2(a), αi = 0, 1 ≤ i ≤ n. Hence α = 0 and δ(g) ⊆ g. It follows that δ = δ0,A for some A ∈ Mn(k). Lemma 2.15. Let A = (αi j), B = (βi j) ∈ Mn(k). Then U(g)(ε, a1, a2, δ0,A) is isomorphic to U(g)(ε, a1, a2, δ0,B) as a generalized Hopf-Ore extension of U(g) if and only if there is a matrix P = (pi j) ∈ GLn(k) with pi1 = pi2 = 0 for all 2 < i ≤ n such that B = (p11 p22 − p12 p21)−1PAP−1. Proof. Suppose U(g)(ε, a1, a2, δ0,A) (cid:27) U(g)(ε, a1, a2, δ0,B) as generalized Hopf-Ore exten- sions of U(g). Then by Proposition 1.11 and Lemma 2.14, there is an element b ∈ U(g), a scale λ ∈ k× and a matrix P = (pi j) ∈ GLn(k) such that ∆(b) = b ⊗ 1 + 1 ⊗ b + λΦP(a1) ⊗ ΦP(a2) − a1 ⊗ a2 and δ0,B = λΦPδ0,AΦ P−1 . Hence B = λPAP−1 and ∆(b) = b ⊗ 1 + 1 ⊗ b + λ Pn = b ⊗ 1 + 1 ⊗ b + λ Pn i, j=1 pi1 p j2ai ⊗ a j − a1 ⊗ a2 i=1 pi1 pi2ai ⊗ ai +(λp11 p22 − 1)a1 ⊗ a2 + λp21 p12a2 ⊗ a1 + P3≤i≤n λ(p11 pi2a1 ⊗ ai + pi1 p12ai ⊗ a1) + P2≤i< j≤n λ(pi1 p j2ai ⊗ a j + p j1 pi2a j ⊗ ai). Then by Lemma 2.2(a), we have λ(p11 p22 − p12 p21) = 1, p11 pi2 = pi1 p12, 3 ≤ i ≤ n, and pi1 p j2 = p j1 pi2, 2 ≤ i < j ≤ n. It follows that pi1 = pi2 = 0 for all 3 ≤ i ≤ n and λ = (p11 p22 − p12 p21)−1. Conversely, suppose that there is a matrix P = (pi j) ∈ GLn(k) with pi1 = pi2 = 0 for all 2 < i ≤ n such that B = (p11 p22 − p12 p21)−1PAP−1. Then by det(P) , 0 and pi1 = pi2 = 0 for all 2 < i ≤ n, one gets p11 p22 − p12 p21 , 0. Let λ = (p11 p22 − p12 p21)−1 and b = 1 P−1 and + λp12 p21a1a2. Then λp11 p22 − 1 = λp12 p21, δ0,B = λΦPδ0,AΦ i i=1 pi1 pi2(ai ⊗ 1 + 1 ⊗ ai)2 i=1 pi1 pi2a2 2 λ P2 +λp12 p21(a1 ⊗ 1 + 1 ⊗ a1)(a2 ⊗ 1 + 1 ⊗ a2) 2 λ P2 i=1 pi1 pi2a2 +λp12 p21a1a2 ⊗ 1 + λp12 p211 ⊗ a1a2 + λp12 p21a1 ⊗ a2 + λp12 p21a2 ⊗ a1 i=1 pi1 pi2ai ⊗ ai i=1 pi1 pi2a2 i ⊗ 1 + 1 ⊗ 1 + λ P2 i 2 λ P2 2 λ P2 ∆(b) = 1 = 1 = b ⊗ 1 + 1 ⊗ b + λ P2 i=1 pi1 pi2ai ⊗ ai +(λp11 p22 − 1)a1 ⊗ a2 + λp21 p12a2 ⊗ a1 = b ⊗ 1 + 1 ⊗ b + λ P2 = b ⊗ 1 + 1 ⊗ b + λΦP(a1) ⊗ ΦP(a2) − a1 ⊗ a2. i, j=1 pi1 p j2ai ⊗ a j − a1 ⊗ a2 It follows from Proposition 1.11 that U(g)(ε, a1, a2, δ0,A) (cid:27) U(g)(ε, a1, a2, δ0,B) as general- ized Hopf-Ore extensions of U(g). (cid:3) Define a relation ∼ on Mn(k) as follows: for A, B ∈ Mn(k), A ∼ B if and only if there is a scalar λ ∈ k× and a matrix P ∈ GLn(k) such that B = λPAP−1. Then ∼ is an equivalence relation on Mn(k). Let Mn(k)/ ∼ be the corresponding set of equivalence classes. Similarly, define a relation ∼′ on Mn(k) as follows: for A, B ∈ Mn(k), A ∼′ B if and only if there 20 LAN YOU, ZHEN WANG, AND HUI-XIANG CHEN is a matrix P = (pi j) ∈ GLn(k) with pi1 = pi2 = 0 for all 3 ≤ i ≤ n such that B = (p11 p22 − p12 p21)−1PAP−1. Then ∼′ is also an equivalence relation on Mn(k). Let Mn(k)/ ∼′ be the corresponding set of equivalence classes. Proposition 2.16. Each generalized Hopf-Ore extension of U(g) is isomorphic to one of the followings: (a) U(g)(ε, 0, 0, δ0,A), where A ∈ Mn(k)/ ∼; (b) U(g)(χ1, 0, 0, 0); (c) U(g)(ε, a1, a2, δ0,B), where B ∈ Mn(k)/ ∼′. Moreover, as generalized Hopf-Ore extensions of U(g), U(g)(ε, 0, 0, δ0,A), U(g)(χ1, 0, 0, 0) and U(g)(ε, a1, a2, δ0,B) are pairwise non-isomorphic. Proof. The first claim follows from the discussion above and Lemma 2.15, the second one follows from Proposition 1.11. (cid:3) When n = 2, g is exactly the Lie algebra g1 given in Subsection 2.3. Thus, let n = 2 in Proposition 2.16, one gets the classifications of the generalized Hopf-Ore extensions of U(g1) with char(k) = 0. In this case, two matrices A, B ∈ M2(k) satisfy A ∼′ B if and only if there is a matrix P ∈ GL2(k) such that B = det(P)−1PAP−1. Remark 2.17. Assume that k is an algebraically closed field of characteristic zero. In [9], author constructed two kinds of connected Hopf algebras A(λ1, λ2, α) and B(λ) and proved that every connected Hopf algebra of GK-dimension three is isomorphic to one of the fol- lowing: the enveloping algebra U(g) for a 3-dimensional Lie algebra g, the Hopf algebras A(0, 0, 0), A(0, 0, 1), A(1, 1, 1), A(1, λ, 0) and B(λ), λ ∈ k. Clearly, one can check that these Hopf algebras are the generalized Hopf-Ore extensions of the enveloping algebras of some 2-dimensional Lie algebras. ACKNOWLEDGMENTS This work is supported by the National Natural Science Foundation of China (Grant No. 11571298). References [1] M. Beattie, S. Dascalescu, L. Grunenfelder, On the number of types of finite dimensional Hopf algebras, Invent. Math. 136(1) (1999)1-7. [2] M. Beattie, S. Dascalescu, L. Grunenfelder, Constructing pointed Hopf algebras by Ore extensions, J. of Algebra 225(2)(2000) 743-770. [3] K.A. Brown, S. O'Hagan, J.J. Zhang and G. Zhuang, Connected Hopf algebras and iterated Ore extensions, J. of Pure and Applied Algebra 219(6) (2015) 2405-2433. [4] C. Cibils, Half-quantum groups at roots of unity, path algebras, and representation type, Int. Math. Res. Notices (1997) 1997 (12): 541-553. [5] J.C. McConnell and J.C.Robson, Noncommutative Noetherian rings, Wiley-Interscience, New York. 1987. [6] S. Montgomery, Hopf Algebras and Their Actions on Rings, CBMS Reg. Conf. Ser. Math. 82, Amer. Math. Soc., Providence, RI, 1993. [7] A. N. Panov, Ore extensions of Hopf algebras, Mathematical Notes 74(3) (2003) 401-410. [8] Z. Wang, L. You, H. X. Chen, Representations of Hopf-Ore extensions of group algebras and pointed Hopf algebras of rank one, Algebras and Representation Theory 18(3) (2015) 801-830. [9] G. Zhuang, Properties of connected Hopf algebras of finite Gelfand-Kirillov dimension, J. London Math. Soc. 87 (2)(2013) 877 - 898. GENERALIZED HOPF-ORE EXTENSIONS 21 College of Mathematical Science, Yangzhou University, Yangzhou 225002, China; School of Mathematics and Physics, Yancheng Institute of Technology, Yancheng 224051, China E-mail address: [email protected] School of Mathematics and Physics, Yancheng Institute of Technology, Yancheng 224051, China E-mail address: [email protected] College of Mathematical Science, Yangzhou University, Yangzhou 225002, China E-mail address: [email protected]
1903.09932	2	1903	2019-10-03T15:36:31	On Leibniz algebras whose centralizers are ideals	[ "math.RA" ]	This paper concerns the study of Leibniz algebras, a natural generalization of Lie algebras, from the perspective of centralizers of elements. We study conditions on Leibniz algebras under which centralizers of all elements are ideals. We call a Leibniz algebra, a CL-algebra if centralizers of all elements are ideals. We discuss nilpotency of CL-algebras.	math.RA	math	ON LEIBNIZ ALGEBRAS WHOSE CENTRALIZERS ARE IDEALS PRATULANANDA DAS AND RIPAN SAHA Abstract. This paper concerns the study of Leibniz algebras, a natural gen- eralization of Lie algebras, from the perspective of centralizers of elements. We study conditions on Leibniz algebras under which centralizers of all ele- ments are ideals. We call a Leibniz algebra, a CL-algebra if centralizers of all elements are ideals. We discuss nilpotency of CL-algebras. 1. Introduction In [11], J.-L. Loday introduced some new types of a non-anticommutative ver- sion of Lie algebras along with their (co)homologies. Historically, such algebraic structures had been studied by A. Bloh who called them D-algebras [7]. In this new type of algebras, the bracket satisfies Leibniz identity instead of Jacobi iden- tity and such algebras are called Leibniz algebras. In group theory, centralizers of a group give many useful pieces of information about the structure of the group. Investigations of centralizers is an interesting research area in group theory [3, 6]. Centralizers of Lie algebras are studied in many articles [5, 14, 13]. The present article contributes to the development of the theory of non-associative algebras from the perspective of centralizers of Leibniz algebras. Leibniz algebras whose subalgebras are ideals are studied in [9]. In this article, we first study conditions on Leibniz algebras for which centralizers of all elements of Leibniz algebras are ideals. Leibniz algebras which satisfy the required conditions will be called CL-algebras. The main question addressed in this paper is whether a finite-dimensional nilpotent Leibniz algebra is a CL- algebra or not. We show in Theorem 4.3 that any nilpotent complex Leibniz algebra up to dimension four is a CL-algebra. We define a notion of CL-property in a Leibniz algebra. Elements which satisfy the CL-property are called CL- elements. In [12], Mukherjee and Saha have introduced a notion of a finite group action on a Leibniz algebra. We show that a CL-element is preserved under the group actions on Leibniz algebras. At the end of this article, we mention some questions for further study. In this section, we discuss some basics of Leibniz algebras. 2. Preliminaries Definition 2.1. Let K be a field. A Leibniz algebra is a vector space L over K, equipped with a bracket operation, which is K-bilinear and satisfies the Leibniz 2010 Mathematics Subject Classification. 17A32, 17A30, 17B30. Key words and phrases. Leibniz algebra; centralizer; CL-algebra; nilpotent Leibniz algebra; group action. 1 2 PRATULANANDA DAS AND RIPAN SAHA identity: [x, [y, z]] = [[x, y], z] − [[x, z], y] for x, y, z ∈ L. Note that any Lie algebra is automatically a Leibniz algebra as in the presence of skew symmetry the Leibniz identity is the same as the Jacobi identity. A Leibniz algebra (L, [ , ]) is called an abelian algebra if [x, y] = 0 for all x, y ∈ L. A morphism between two Leibniz algebras L1 and L2 is a K-linear map f : L1 → L2 which satisfies f ([x, y]) = [f (x), f (y)] for all x, y ∈ L1. Example 2.2. Consider a differential Lie algebra (L, d) with the Lie bracket [ , ]. Then L has a Leibniz algebra structure with the bracket operation [x, y]d := [x, dy]. The new bracket on L is called the derived bracket. Definition 2.3. A Leibniz algebra L is said to be simple if it contains only the following ideals: {0}, I, L. Here I denotes ideal generated by elements of the form [x, x] for all x ∈ L. Given a Leibniz algebra L, we define the following two sided ideals: L1 = L and Lk+1 = [Lk, L], L[1] = L and L[k+1] = [L[k], L[k]] for k ≥ 1. Definition 2.4. A Leibniz algebra L is called a nilpotent Leibniz algebra if there exist n ∈ N such that Ln = 0. Suppose n ∈ N is least and Ln = 0 then L is called an n-step nilpotent Leibniz algebra. Example 2.5. [4] Consider a three dimensional vector space L spanned by ] : L × L −→ L by [e1, e3] = e2 {e1, e2, e3} over C. Define a bilinear map [ , and [e3, e3] = e1, all other products of basis elements being 0. Then (L, [ , ]) is a Leibniz algebra over C of dimension 3. The Leibniz algebra L is nilpotent and is denoted by λ6 in the classification of three dimensional nilpotent Leibniz algebras. Definition 2.6. A Leibniz algebra L is called a solvable Leibniz algebra if there exist n ∈ N such that L[n] = 0. Note that any nilpotent Leibniz algebra is a solvable Leibniz algebra but the converse is not true in general. For example, the subalgebra of gl(n, R), n ≥ 2, consisting of upper triangular matrices, b(n, R), is solvable but not nilpotent [8, p. 10]. Definition 2.7. An ideal P of a nilpotent Leibniz algebra L is called nilpotent if P is nilpotent as a Leibniz algebra. 2.1. Classsification of nilpotent Leibniz algebras upto dimension 4. Here we recall some well-known classification results for nilpotent Leibniz algebras. Theorem 2.8. [10] Let L be a 2-dimensional nilpotent Leibniz algebra. Then L is either abelian or isomorphic to µ1 : [e1, e1] = e2. ON LEIBNIZ ALGEBRAS WHOSE CENTRALIZERS ARE IDEALS 3 Theorem 2.9. [1] Let L be a 3-dimensional nilpotent Leibniz algebra. Then L is isomorphic to one of the following pairwise non-isomorphic algebras: λ1 : abelian, λ2 : [e1, e1] = e3, λ3 : [e1, e2] = e3, [e2, e1] = −e3, λ4(α) : [e1, e1] = e3, [e2, e2] = αe3, [e1, e2] = e3, λ5 : [e2, e1] = e3, [e1, e2] = e3, λ6 : [e1, e1] = e2, [e2, e1] = e3. Theorem 2.10. [2] There exist up to isomorphism five one parametric fami- lies and twelve concrete representatives of nilpotent complex Leibniz algebras of dimension four, namely: ρ1 : [e1, e1] = e2, [e2, e1] = e3, [e3, e1] = e4; ρ2 : [e1, e1] = e3, [e1, e2] = e4, [e2, e1] = e3, [e3, e1] = e4; ρ3 : [e1, e1] = e3, [e2, e1] = e3, [e3, e1] = e4; ρ4(α) : [e1, e1] = e3, [e1, e2] = αe4, [e2, e1] = e3, [e2, e2] = e4, [e3, e1] = e4, α ∈ {0, 1}; ρ5 : [e1, e1] = e3, [e1, e2] = e4, [e3, e1] = e4; ρ6 : [e1, e1] = e3, [e2, e2] = e4, [e3, e1] = e4; ρ7 : [e1, e1] = e4, [e1, e2] = e3, [e2, e1] = −e3, [e2, e2] = −2e3 + e4; ρ8 : [e1, e2] = e3, [e2, e1] = e4, [e2, e2] = −e3; ρ9(α) : [e1, e1] = e3, [e1, e2] = e4, [e2, e1] = −αe3, [e2, e2] = −e4, α ∈ C; ρ10(α) : [e1, e1] = e4, [e1, e2] = αe4, [e2, e1] = −αe4, [e2, e2] = e4, [e3, e3] = e4, α ∈ C; ρ11 : [e1, e2] = e4, [e1, e3] = e4, [e2, e1] = −e4, [e2, e2] = e4, [e3, e1] = e4; ρ12 : [e1, e1] = e4, [e1, e2] = e4, [e2, e1] = −e4, [e3, e3] = e4; ρ13 : [e1, e2] = e3, [e2, e1] = e4; ρ14 : [e1, e2] = e3, [e2, e1] = −e3, [e2, e2] = e4; ρ15 : [e2, e1] = e4, [e2, e2] = e3; ρ16(α) : [e1, e2] = e4, [e2, e1] = (1 + α)/(1 − α)e4, [e2, e2] = e3, α ∈ C\{1}; ρ17 : [e1, e2] = e4, [e2, e1] = −e4, [e3, e3] = e4. 4 PRATULANANDA DAS AND RIPAN SAHA 3. CL-algebras In this section, we introduce conditions so that for a Leibniz algebra satisfying these conditions centralizers of each element are ideals. We introduce a subclass of collection of all Leibniz algebras, a member of this subclass is called a CL- algebra. We give some examples of CL-algebras and show that the centralizer of an element in a CL-algebra is an ideal (cf. Theorem 3.10). Definition 3.1. Let L be a Leibniz algebra over a field K. We define centralizer of x ∈ L as follows: CL(x) = {y ∈ L [x, y] = 0 = [y, x]}. The right centralizer of x ∈ L is defined as C r L(x) = {y ∈ L [x, y] = 0}. The left centralizer of x ∈ L is defined as C l L(x) = {y ∈ L [y, x] = 0}. So, the centralizer of x ∈ L is both left and right centralizer of x. Remark 3.2. In case of a Lie algebra both the left and the right centralizers are same. Observe that for any x ∈ L, the square element [x, x] ∈ C r L(x) but [x, x] may not belongs to C l L(x). For example, consider the algebra with multiplication [e3, e3] = e1, [e1, e3] = e2, and all other products are zero. We have CL(e3) = < e2 >, and [e3, e3] = e1 /∈ C l L(e3). Lemma 3.3. For any Leibniz algebra L and any x ∈ L, CL(x) is a Leibniz subalgebra. Proof. Let y1, y2 ∈ CL(x). From the bilinearity of bracket operation, y1 − y2 ∈ CL(x). Now, [x, [y1, y2]] = [[x, y1], y2] − [[x, y2], y1] = [0, y2] − [0, y1] = 0. Thus, [y1, y2] ∈ CL(x). Similarly, [y2, y1] ∈ CL(x). So, CL(x) is a Leibniz subalgebra. (cid:3) Definition 3.4. A Leibniz algebra L is called a left CL-algebra if it satisfies the following conditions: (1) (2) [[x, a], y] = 0, [[a, x], y] = 0, ∀ y ∈ CL(x), and a, x ∈ L. A Leibniz algebra L is called a right CL-algebra if it satisfies the following con- ditions: (3) (4) [[x, a], y] = 0, [y, [a, x]] = 0, ∀ y ∈ CL(x), and a, x ∈ L. Definition 3.5. A Leibniz algebra L is called a CL-algebra if it is both a left and a right CL-algebra. Remark 3.6. To verify a Leibniz algebra is a CL-algebra, we need to check the following conditions for all x ∈ L, ON LEIBNIZ ALGEBRAS WHOSE CENTRALIZERS ARE IDEALS 5 (1) [[x, L], CL(x)] = 0, (2) [[L, x], CL(x)] = 0, (3) [CL(x), [L, x]] = 0. Example 3.7. Any Abelian Leibniz algebra is a CL-algebra. The following ex- ample shows that the converse part may not be true. Example 3.8. Let L be a vector space of dimension 2 over a field F and {a, b} be a basis of L. Define the bracket [ , ] by the following rule: [a, a] = b, [b, a] = [b, b] = [a, b] = 0. It is enough to check the conditions of CL-algebras for the basis elements. For the basis elements, we have, CL(a) = < b >, [[a, L], CL(a)] = 0, [[L, a], CL(a)] = 0, [CL(a), [L, a]] = 0; CL(b) = L, [[b, L], CL(b)] = 0, [[L, b], CL(b)] = 0, [CL(b), [L, b]] = 0. Thus, L is a CL-algebra, which is non-abelian. Example 3.9. Any 3-step nilpotent Leibniz algebra L is a CL-algebra. As L is a 3-step nilpotent Leibniz algebra, we have L3 = 0. Thus, by Definition 3.4 L is a CL-algebra. Theorem 3.10. Suppose L is a Leibniz algebra and x ∈ L. Centralizers CL(x) in L are left (right) ideal of L if and only if L is a left (right) CL-algebra. Proof. Suppose L is a left CL-algebra. Let y ∈ CL(x) and a ∈ L. We show that [a, y] ∈ CL(x). [[x, [a, y] = [[x, a], y] − [[x, y], a] = −[0, a] = 0. Similary, using the second equation of left CL-algebra, one can easily show [[a, y], x] = 0. Thus, [a, y] ∈ CL(x). So, CL(x) is a left ideal of L. Conversely, suppose that CL(x) is a left ideal of L for all x ∈ L. It is easy to see from the Leibniz identity that L is a left CL-algebra. By a similar argument L is a right CL-algebra. (cid:3) Remark 3.11. Theorem 3.10 gives us a motivation for the conditions of CL- algebras. Note that by the conditions in Definition 3.4, a Leibniz algebra L is a left (respectively, right) CL-algebra implies CL(x) is a left (respectively, right) ideal for all x ∈ L. Corollary 3.12. For all x ∈ L, centralizers CL(x) in L are ideals of L if and only if L is a CL-algebra. Definition 3.13. An element a ∈ L is said to have the left CL-property if for all y ∈ CL(x) and x ∈ L, [[x, a], y] = 0, [[a, x], y] = 0. An element a ∈ L is said to have the right CL-property if for all y ∈ CL(x) and x ∈ L, we have [[x, a], y] = 0, [y, [a, x]] = 0. 6 PRATULANANDA DAS AND RIPAN SAHA Definition 3.14. An element a ∈ L is said to have the CL-property if it satisfies both left and right CL-property. Remark 3.15. We call an element having CL-property a CL-element. Note that the additive identity 0 is a CL-element. Lemma 3.16. Suppose y ∈ CL(x) and a ∈ L satisfies the CL-property then [a, y], [y, a] ∈ CL(x). Proof. Suppose y ∈ CL(x) and a ∈ L satisfies the CL-property. From the Leibniz identity, [[a, y], x] = [a, [y, x]] + [[a, x], y] = 0. Similarly, it is easy to check that [x, [a, y]] = [[y, a], x] = [x, [y, a]] = 0. Thus, (cid:3) [a, y], [y, a] ∈ CL(x). Theorem 3.17. The collection S of all elements of L satisfying the CL-property forms a Leibniz subalgebra. Thus, (S, [ , ]) is a CL-algebra. Proof. To show S is a Leibniz subalgebra, we need only to check that elements of S are closed under the bracket of L. Suppose a, b ∈ L have the CL-property. The set S is non-empty as 0 ∈ S. Suppose z = [a, b] and y ∈ CL(x). From the Leibniz identity we have, [[x, z], y]−[[x, y], z] = [x, [z, y]], [[x, z], y] = [x, [z, y]]. Since a, b satisfy CL-property, it follows from Lemma 3.16, [z, y] = [[a, b], y] = [a, [b, y]] + [[a, y], b] ∈ CL(x). Similarly, [y, z] ∈ CL(x). We also have, [x, [z, y]] = 0, [[z, x], y] = [z, [x, y]] − [[z, y], x] = 0, [y, [z, x]] = [[y, z], x] + [[y, x], z] = 0. Thus, z = [a, b] ∈ S. Similarly, one can show that [b, a] ∈ S. Therefore, S is a subalgebra of L. (cid:3) Proposition 3.18. Let L1 and L2 be two Leibniz algebras and f : L1 → L2 be an isomorphism between L1 and L2 then f (CL(x)) = CL(f (x)). Proof. Let z ∈ f (CL(x)). So z = f (y1) for some y1 ∈ CL(x). Now, [z, f (x)] = [f (y1), f (x)] = f ([y1, x]) = f (0) = 0. Thus, z ∈ CL(f (x)) and f (CL(x)) ⊆ CL(f (x)). Suppose z ∈ CL(f (x)). This implies [f (x), z] = 0. As f is an isomorphism, z = f (y2) for some y2 ∈ L1. Now, [f (x), z] = 0, [f (x), f (y2)] = 0, f ([x, y2]) = 0, [x, y2] = 0, Since f is an isomorphism. This implies y2 ∈ CL(x) and f (y2) ∈ f (CL(x)). So, CL(f (x)) ⊆ f (CL(x)). Therefore, f (CL(x)) = CL(f (x)). (cid:3) ON LEIBNIZ ALGEBRAS WHOSE CENTRALIZERS ARE IDEALS 7 Theorem 3.19. Let L1 and L2 be two Leibniz algebras and f : L1 → L2 be an isomorphism between L1 and L2. If a ∈ L1 has the CL-property then f (a) also has the CL-property in L2. Proof. Suppose a ∈ L1 has the CL-property. For y ∈ CL(x) and using Proposi- tion 3.18, we have, [[f (x), f (a)], f (y)] = [f ([x, a]), f (y)] = f ([[x, a], y]) = f (0) = 0, [[f (a), f (x)], f (y)] = [f ([a, x]), f (y)] = f ([[a, x], y]) = f (0) = 0, [f (y), [f (a), f (x)]] = [f (y), f ([a, x])] = f ([y, [a, x]]) = f (0) = 0. Thus, the set of all CL-elements are preserved under the isomorphism of Leibniz algebras. (cid:3) Remark 3.20. The Theorem 3.19 may be used to check whether a Leibniz mor- phism is an isomorphism or not. 4. Nilpotency of CL-algebras In this section, we study how nilpotent Leibniz algebras are related to CL- algebras. We attempt to answer the following questions: Question 4.1. Is every finite dimensional CL-algebra a nilpotent Leibniz alge- bra? Question 4.2. Is every finite dimensional nilpotent Leibniz algebra a CL-algebra? To answer Question 4.1, we consider the following example. Let K denotes the complex field and L is a 3 dimensional Leibniz algebra generated by the basis elements {e1, e2, e3} with bracket [e3, e3] = e1, [e3, e2] = e2, [e2, e3] = −e2 and all other product are 0. Now, L2 = < e1, e2 >, L3 = L4 = · · · = < e2 > . Thus, L is not nilpotent but L is a CL-algebra as CL(e1) = < e1, e2, e3 >, CL(e2) = < e1, e2 >, CL(e3) = < e1 >. Now, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0; [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0; [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0. Thus not all CL-algebras are nilpotent. To answer Question 4.2, we consider the following theorem. Theorem 4.3. Nilpotent complex Leibniz algebras up to dimension 4 are CL- algebras. Proof. Suppose L is a finite dimensional complex Leibniz algebra. We consider the following four cases: Case:1 Suppose the dimension of L is 1. So, L is abelian and any abelian Leibniz algebra is automatically a CL-algebra. 8 PRATULANANDA DAS AND RIPAN SAHA Case:2 Suppose the dimension of L is 2. By Theorem 2.8, up to isomorphisms there are only two nilpotent Leibniz algebras of dimension 2. µ1 : CL(e1) =< e2 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0; CL(e2) =< e1, e2 >= L, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0. Therefore, two dimensional nilpotent Leibniz algebras are CL-algebras. Case:3 Suppose the dimension of L is 3. By Theorem 2.9, upto isomorphisms there are only six nilpotent Leibniz algebras of dimension 3. λ2 : CL(e1) =< e2, e3 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e1, e2, e3 >= L, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0. λ3 : CL(e1) =< e1, e3 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0. λ4(α) : CL(e1) =< e3 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0. λ5 : CL(e1) =< e1, e3 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0. λ6 : CL(e1) =< e3 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0. Therefore, three dimensional nilpotent Leibniz algebras are CL-algebras. Case:4 Suppose L is four dimensional and L =< e1, e2, e3, e4 >. By the classifica- tion Theorem 2.10 of nilpotent Leibniz algebras of dimension four, we have the following seventeen non-isomorphic classes: ρ1 : CL(e1) =< e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ2 : CL(e1) =< e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ON LEIBNIZ ALGEBRAS WHOSE CENTRALIZERS ARE IDEALS 9 ρ3 : CL(e1) =< e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ4(α) : CL(e1) =< e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ5 : CL(e1) =< e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ6 : CL(e1) =< e2, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e1, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e2, e3, e4 >= L, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ7 : CL(e1) =< e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ8 : CL(e1) =< e1, e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ9(α) : CL(e1) =< e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ10(α) : CL(e1) =< e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ11 : CL(e1) =< e1, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ12 : CL(e1) =< e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. 10 PRATULANANDA DAS AND RIPAN SAHA ρ13 : CL(e1) =< e1, e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ14 : CL(e1) =< e1, , e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ15 : CL(e1) =< e1, e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ16(α) : CL(e1) =< e1, e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e3, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. ρ17 : CL(e1) =< e1, e3, e4 >, [[e1, L], CL(e1)] = 0, [[L, e1], CL(e1)] = 0, [CL(e1), [L, e1]] = 0, CL(e2) =< e2, e3, e4 >, [[e2, L], CL(e2)] = 0, [[L, e2], CL(e2)] = 0, [CL(e2), [L, e2]] = 0, CL(e3) =< e1, e2, e4 >, [[e3, L], CL(e3)] = 0, [[L, e3], CL(e3)] = 0, [CL(e3), [L, e3]] = 0, CL(e4) =< e1, e2, e3, e4 >, [[e4, L], CL(e4)] = 0, [[L, e4], CL(e4)] = 0, [CL(e4), [L, e4]] = 0. Therefore, four dimensional nilpotent complex Leibniz algebras are CL-algebras. Thus, we have proved that nilpotent complex Leibniz algebras of dimension less than equal to four are CL-algebras. (cid:3) 5. Group actions and CL-algebras In [12], authors have defined a notion of a finite group action on Leibniz alge- bra. In this section, we study the CL-property of Leibniz algebras under actions of finite groups. Definition 5.1. Let L be a Leibniz algebra and G be a finite group. The group G is said to act from the left if there exists a function φ : G × L → L, (g, x) 7→ φ(g, x) = gx satisfying the following conditions. (1) For each g ∈ G the map x 7→ gx, denoted by ψg is linear. (2) ex = x for all x ∈ L, where e ∈ G is the group identity. (3) g1(g2x) = (g1g2)x for all g1, g2 ∈ G and x ∈ L. (4) g[x, y] = [gx, gy] for all g ∈ G and x, y ∈ L. The following is an alternative formulation of the above definition. Proposition 5.2. A finite group G acts on a Leibniz algebra L if and only if there exists a group homomorphism ψ : G → IsoLeib(L, L), g 7→ ψ(g) = ψg, from the group G to the group of Leibniz algebra isomorphisms from L to L, where ψg(x) = gx is the left translation by g. ON LEIBNIZ ALGEBRAS WHOSE CENTRALIZERS ARE IDEALS 11 Remark 5.3. Let G be a finite group and K[G] be the associated group ring. If G acts on a Leibniz algebra L then L may be viewed as a K[G]-module. Definition 5.4. Suppose L is a Leibniz algebra equipped with an action of a finite group G. A map f : L → L is said to be an equivariant map if f (gx) = gf (x). Lemma 5.5. Suppose a finite group G acts on a Leibniz algebra L. Then there is a map CL(x) → CL(gx). Proof. For x ∈ L and y ∈ CL(x), we define a map p : CL(x) → CL(gx), y 7→ gy Note that the map is well defined as [gx, gy] = g[x, y] = g0 = 0 and [gy, gx] = g[y, x] = g0 = 0. (cid:3) Note that the map as defined above is just restriction of the group action map φ to CL(x). Theorem 5.6. A CL-element of L is preserved under the action of G on L. Proof. Let a ∈ L satisfies the CL-property, we need to show that for all g ∈ G, ga also satisfies the CL-property. As a satisfies the CL-property, we have for all x ∈ L and y ∈ CL(x) [[x, a], y] = 0, [[a, x], y] = 0, [y, [a, x]] = 0. From the Lemma 5.5, gy ∈ CL(gx). Thus, we have, [[x, ga], y] = [g[g−1x, a], y] = g[[g−1x, a], g−1y] = g0 = 0. Similarly, one can show that [[ga, x], y] = 0, [y, [ga, x]] = 0. So, ga also satisfies the CL-property. (cid:3) Corollary 5.7. An equivariant automorphism of Leibniz algebras preserves CL- elements. Proof. Suppose a Leibniz algebra L is equipped with an action of a finite group G. Let x ∈ L be a CL-element. By Theorem 5.6, gx is also a CL-element for all g ∈ G. Suppose f : L → L is an equivariant automorphism. From the Theorem 3.19 and 5.6, f (gx) = gf (x) is also a CL-element. (cid:3) Further questions We end this article with the following interesting questions: (1) In the Theorem 4.3, we proved nilpotent complex Leibniz algebras up to dimension four are CL-algebras. What about nilpotent Leibniz algebras of dimension greater than four, are they CL-algebras? In general, are all nilpotent Leibniz algebras CL-algebras? (2) What is the relationship of CL-algebras with solvable Leibniz algebras? 12 PRATULANANDA DAS AND RIPAN SAHA Acnowledgements The second author would like to thank Prof. Ayupov of Institute of Mathe- matics, Uzbekistan Academy of Sciences and Prof. Karimbergen of Karakalpak State University, Uzbekistan for valuable discussions on this problem in a CIMPA research school on "Non-associative algebra and applications" held at Tashkent, Uzbekistan. The second author is especially thankful to Dr. Abror Kh. Khu- doyberdiyev for reading a draft version of the article and his useful suggestions. The second author expresses his gratitude to CIMPA, France for their financial help to attend the research school. References [1] S. Albeverio, Sh. A. Ayupov, B. A. Omirov, On nilpotent and simple Leibniz alge- bras, Comm. in Algebra, vol. 33(1), 2005, pp. 159-172. [2] S. Albeverio, B.A. Omirov, I.S. Rakhimov, Classification of 4-dimensional nilpotent complex Leibniz algebras. Extracta mathematicae, vol. 21(3), (2006), pp. 197-210. [3] A. R. Ashrafi, On finite groups with a given number of centralizers. Algebra Colloq. 7: 2 (2000), 139-146. [4] Sh. A. Ayupov and B. A. Omirov, On some classes of nilpotent Leibniz algebras, Siberian Math. J. 42 (1) (2001) 18-29. [5] Y. Barnea, I. M. Isaacs, (2003), Lie algebras with few centralizer dimensions, J. Algebra, 259:284-299. [6] S. M. Belcastro, G. J. Sherman, (1994), Counting centralizers in finite groups. Math. Mag. 5:111-114. [7] A. Bloh, On a generalization of Lie algebra notion, Math. in USSR Doklady, 1965, 165(3):471-473. [8] J. E. Humphreys, Introduction to Lie algebras and representation theory, Graduate Texts in Mathematics, 9, New York : Springer-Verlag, 1972. [9] L. Kurdachenko, N. Semko, I. Subbotin, (2017), The Leibniz algebras whose subal- gebras are ideals, Open Mathematics, 15(1), pp. 92-100. [10] J.L. Loday, Cup product for Leibniz cohomology and dual Leibniz algebras, Math. Scand. 77, 189-196 (1995). [11] J. L. Loday , T. Pirashvili, Universal enveloping algebras of Leibniz algebras and (co)homology, Math. Ann., 296, 139-158 (1993). [12] G. Mukherjee, R. Saha, Cup-Product for Equivariant Leibniz Cohomology and Zin- biel Algebras, Algebra Colloquium, 26 : 2 (2019), 271-284. [13] Saffarnia, Somayeh; Moghaddam, Mohammad Reza R.; Rostamyari, Mohammad, Centralizers in Lie algebras, Indian J. Pure Appl. Math. 49 (2018), no. 1, 39-49. [14] A. J. Zapirain (2005), Centralizer sizes and nilpotency class in Lie algebras and finite p-groups, Proc. Am. Math. Soc. 133(10): 2817-2820. E-mail address: [email protected] Department of Mathematics, Jadavpur University, Kolkata-700032, West Ben- gal, India. E-mail address: [email protected] Department of Mathematics, Raiganj University, College para, Uttar Dinajpur, PIN-733134, West Bengal, India.
1704.07819	2	1704	2017-08-07T15:36:11	Notes on $G_2$: The Lie algebra and the Lie group	[ "math.RA", "math.DG" ]	These notes have been prepared for the Workshop on "(Non)-existence of complex structures on $\mathbb{S}^6$", to be celebrated in Marburg in March, 2017. The material is not intended to be original. It contains a survey about the smallest of the exceptional Lie groups: $G_2$, its definition and different characterizations joint with its relationship with $\mathbb{S}^6$ and with $\mathbb{S}^7$. With the exception of the summary of the Killing-Cartan classification, this survey is self-contained, and all the proofs are given, mainly following linear algebra arguments. Although these proofs are well-known, they are spread and some of them are difficult to find. The approach is algebraical, working at the Lie algebra level most of times. We analyze the complex Lie algebra (and group) of type $G_2$ as well as the two real Lie algebras of type $G_2$, the split and the compact one. Octonions will appear, but it is not the starting point. Also, 3-forms approach and spinorial approach are viewed and related.	math.RA	math	THE LIE ALGEBRA AND THE LIE GROUP NOTES ON G2: CRISTINA DRAPER FONTANALS∗ Abstract. These notes have been prepared for the Workshop on "(Non)- existence of complex structures on S6", celebrated in Marburg in March, 2017. The material is not intended to be original. It contains a survey about the smallest of the exceptional Lie groups: G2, its definition and different char- acterizations as well as its relationship to the spheres S6 and S7. With the exception of the summary of the Killing-Cartan classification, this survey is self-contained, and all the proofs are provided. Although these proofs are well- known, they are scattered, some of them are difficult to find, and others require stronger background, while we will usually stick to linear algebra arguments. The approach is algebraical, working at the Lie algebra level most often. We analyze the complex Lie algebra (and group) of type G2 as well as the two real Lie algebras of type G2, the split and the compact one. The octonion algebra will play its role, but it is not the starting point. Also, both the 3- forms approach and the spinorial approach are viewed and connected. Special emphasis is put on relating all the viewpoints by providing precise models. Contents 1. A summary of the Killing-Cartan classification via roots 1.1. On simple real Lie algebras 2. A first linear model of (split) G2 2.1. The model 2.2. A little bit of representation theory 3. Understanding G2 better 3.1. G2 and the cross product in F7 3.2. G2 and the octonions 4. Generic 3-forms in F7 4.1. G2 and the triple product 4.2. Generic 3-forms 4.3. Brief comments on G2-structures 5. G2 and the spheres in dimension 6 5.1. Case the norm is positive definite 5.2. A model of compact G2 5.3. Parenthesis on real forms of g2 5.4. Case the norm is not positive definite 5.5. Connectedness 6. G2 and spinors References 2 6 7 7 10 11 11 14 15 15 18 28 28 28 32 33 36 37 39 49 products, octonions, spinors. 2010 Mathematics Subject Classification. 17B25, 22E60. Key words and phrases. Exceptional Lie algebra, exceptional Lie group, generic 3-forms, cross ∗ Supported by the Spanish Ministerio de Economía y Competitividad -- Fondo Europeo de Desarrollo Regional (FEDER) MTM2016-76327-C3-1-P, and by the Junta de Andalucía grants FQM-336 and FQM-7156, with FEDER funds. 1 2 C. DRAPER 1. A summary of the Killing-Cartan classification via roots Killing essentially classified from 1888 to 1890 [Ki88] the complex finite-dimensio- nal simple Lie algebras, as a first step to classify Lie groups. An important idea for this classification is that, if g is a simple Lie algebra over C, then g is isomorphic to the Lie algebra of linear transformations {adx : x ∈ g}, where adx : g → g denotes the adjoint map given by adx(y) = [x, y]. To study this family it is natural to try to diagonalize the operators adx. So, a Cartan subalgebra is defined as a maximal subalgebra h such that adh is diagonalizable for all h ∈ h. This algebra is necessarily abelian, and, of course, it produces the eigenspace decomposition (as simultaneous diagonalization) g = ⊕α∈h∗ gα, where gα := {x ∈ g : adh(x) = α(h)x ∀h ∈ h}. The set Φ := {0 (cid:54)= α ∈ h∗ : gα (cid:54)= 0} is called a root system and its elements are linear functions called roots. As g0 coincides with h, the previous decomposition is then written as (cid:33) gα , (cid:32)(cid:88) α∈Φ g = h ⊕ called root decomposition of g relative to h. Some important properties of this decomposition are that [gα, gβ] ⊂ gα+β (clear), and that dim gα = 1 for every root α (not so clear!). This gives that the bracket of g is quite controlled by Φ. Before providing more details, recall that the simplicity of g implies the nondegeneracy of the Killing form defined by κ : g × g → C, κ(x, y) = tr(adx ady), which in particular allows to identify h with h∗ and then to define a bilinear sym- metric form also in h∗: ( , ) : h∗ × h∗ → C, α, β ∈ Φ, and (α, α) > 0, so that E := (cid:80) (α, β) := κ(tα, tβ), It turns out that (α, β) ∈ Q for all being tα ∈ h determined by α = κ(tα,−). α∈Φ Rα can be seen as an Euclidean vector space containing Φ (of real dimension equal to dimC h, also called the rank of g) for the inner product extending ( , ). Thus, we can work with angles and lengths in Φ ⊂ E. In fact, Φ satisfies: (R1) Φ is a finite subset that spans E, and 0 /∈ Φ; (R2) If α ∈ Φ, then Rα ∩ Φ = {±α}; (R3) For any α, β ∈ Φ, (cid:104)β, α(cid:105) := 2 (β,α) (R4) If α ∈ Φ, the reflection σα on the hyperplane (Rα)⊥ leaves Φ invariant To be precise, any subset of an Euclidean space satisfying the properties (R1)-(R4) is what is named a root system. One of the key points in the Killing's paper is that it reduces the difficult algebraical problem of the classification of complex simple Lie algebras to an easy geometrical problem: to classify root systems. The definition of root system is so restrictive that there exist only a few possibilities for them. The group generated by the reflections {σα : α ∈ Φ} is called the Weyl group of the root system, and it is always finite (due to being embedded in the symmetric group of Φ). (α,α) ∈ Z; (σα(β) = β − (cid:104)β, α(cid:105)α ∈ Φ for all β ∈ Φ). Take ν ∈ E such that (ν, α) (cid:54)= 0 for any α ∈ Φ. This choice allows to define as positive (resp. negative) a root α such that (ν, α) > 0 (resp. (ν, α) < 0) and so Φ = Φ+ ∪ Φ−, with Φ− = −Φ+ (R2). A positive root is called simple if it is not the sum of two positive roots. The set of simple roots, ∆ = {α1, . . . , αn}, is a basis of the Euclidean space E and it satisfies that any α ∈ Φ+ is a linear combination of the elements in ∆ where all the coefficients are nonnegative integers. NOTES ON G2 3 The square matrix C = ((cid:104)αi, αj(cid:105))1≤i≤j≤n is called Cartan matrix. The Dynkin diagram1 of Φ is the graph which consists of a node for each simple root, and where two different nodes associated to α and β in ∆ are connected by Nα,β := (cid:104)β, α(cid:105)(cid:104)α, β(cid:105) edges (it turns out that Nα,β ∈ {0, 1, 2, 3}, since it equals 4 cos2 θ < 4 for θ the angle formed by α and β, but it is a nonnegative integer!). Moreover, if Nα,β = 2 or 3, then α and β have different length2, and then an arrow is added pointing from the long to the short root. A root system is said irreducible if it is not a disjoint union of two orthogonal root systems, what happens if and only if its Dynkin diagram is connected, which is just the case if we start with a simple Lie algebra (this construction can be realized with a semisimple Lie algebra, that would correspond to nonnecessarily connected Dynkin diagrams). Some important examples of simple Lie algebras are the following: A The special Lie algebra sln+1(C) of zero trace matrices of size n + 1, whose Dynkin diagram is called of type An if n ≥ 1, αn−1 • ··· • • • α1 α2 α3 αn • B-D The orthogonal Lie algebra som(C) of skew-symmetric matrices, whose Dynkin diagram is of type Bn if m = 2n + 1 and n ≥ 2, α1 α2 α3 • ··· and of type Dn if m = 2n and n ≥ 4, • • αn−1 • αn • α1 • α2 • α3 • ··· αn−2 • (cid:26) C The symplectic Lie algebra sp2n(C) = x ∈ gl2n(C) : xt (cid:18) 0 −In (cid:19) + In 0 (cid:18) 0 (cid:19) (cid:27) x = 0 , In 0 αn−1 • αn • −In αn • whose Dynkin diagram is of type Cn if n ≥ 3, sαn−1 • ··· • • • α1 α2 α3 The Dynkin diagrams of the irreducible root systems are precisely the above ones and 1Historical remark: We do not mean that Killing exposed his results in this way. First, Cartan in his thesis [Ca94] corrected the proofs and completed them, Weyl simplified the theory in [We25] and Dynkin introduced the simple roots and the Dynkin diagrams in [Dy46]. 2The quotient between the square of the lengths is (β,β) (cid:104)β,α(cid:105) (cid:104)α,β(cid:105) , again equal to 1, 2 or 3, and it is not difficult to prove that either all the roots have the same length, or there are just two possible lengths. In such a case, the roots are called -correspondingly- long and short ones. (α,α) = + 3 k 4 C. DRAPER (E6) (E7) (E8) (F4) (G2) α1 • α1 • α1 • α1 • tα1 • α2 • α2 • α2 • sα2 • α2 • α3 • α6 α3 • • α3 • α3 • α4 • α4 • α7 α4 • • α4 • α5 • α5 • α5 • α8 • α6 • α6 • α7 • a root; i niαi) := (cid:80) on the height: ht((cid:80) Conversely, each Dynkin diagram determines a unique root system up to isomor- phism. The key is to recover first the set of positive roots by working on induction i ni, and second, to recover the inner product (fixing arbitrarily the length of a short node). We only have to take into account that: Fact 1. If α ∈ Φ+ \ ∆, there is (at least one index) i = 1, . . . , n such that α − αi is Fact 2. If α, β ∈ Φ, {β + mα : m ∈ R} ∩ Φ = {β + mα : m ∈ Z,−r ≤ m ≤ q} (1.1) for r and q positive integers such that (cid:104)β, α(cid:105) = r−q (if α, β ∈ ∆, necessarily such r = 0, since α − β is not a root). For instance, from the Dynkin diagram of type G2, a set of simple roots is given by ∆ = {α1, α2} and (cid:104)α1, α2(cid:105)(cid:104)α2, α1(cid:105) = 3 (there is a triple edge joining the two nodes). Hence, the Cartan matrix is C = (1.2) since (cid:104)αi, αj(cid:105) ∈ Z<0 if i (cid:54)= j and α1 is a short root. Thus, the proportion between the lengths is √3, and, as the cosine of the angle between α1 and α2 will be −√3/2, then the angle is 120◦. As (cid:104)α2, α1(cid:105) = −3, then α1 + α2, 2α1 + α2 and 3α1 + α2 are the only roots of the form iα1 + α2 with i (cid:54)= 0. While, as (cid:104)α1, α2(cid:105) = −1, then α1 + 2α2 is not a root, so that 2α1 + α2 is the only root of height 3 and 3α1 + α2 the only one of height 4 (2α1 + 2α2 is not a root because it is the double of a root, see R2). Now, (cid:104)3α1 + α2, α2(cid:105) = −3 + 2 = −1, so that (3α1 + α2) + α2 is the only root of height 5, and of course there is not any more, since both (3α1 + 2α2) + αi (i = 1, 2) are multiple of roots. That is, we have just obtained the root system in Figure 1: ΦG2 = ±{α1, α2, α1 + α2, 2α1 + α2, 3α1 + α2, 3α1 + 2α2}. Now, the root system Φ determines the product of g. Before recalling the ar- gument, note that the possibility of constructing g (a basis joint with the product of any pair of basic elements) does not prove the existence of a simple Lie alge- bra with such root system, but the uniqueness. In order to prove the existence, one has either to check that the constructed algebra satisfies the Jacobi identity (a straightforward but long and tedious task) or to find some concrete simple Lie (cid:19) (cid:18) 2 −1 2 −3 k j NOTES ON G2 5 3α1 + 2α2 6 α1 + α2 So S HHHHHHYα2 −α1 / −2α1 − α2 S −3α1 − α2 α1 2α1 + α2 7 - S * 3α1 + α2 HHHHHHj S Sw −α1 − α2 ? −α2 −3α1 − 2α2 Figure 1. ΦG2 = ±{α1, α2, α1 + α2, 2α1 + α2, 3α1 + α2, 3α1 + 2α2}. algebra whose related root system is isomorphic to Φ, that is, to find a model of g. Soon we will follow the second way for proving the existence of a complex Lie algebra with root system of type G2, but we previously want to emphasize that all the information of the Lie algebra is encoded in the Dynkin diagram, and with a bit of patience one could recover all the structure constants of the Lie algebra from the picture. The idea of the proof is the following. For each α ∈ Φ+, we choose hα = 2tα (α,α) ∈ h and some xα ∈ gα, x−α ∈ g−α such that [xα, x−α] = hα. In other words, the map (cid:19) (cid:18) 0 0 xα (cid:55)→ 1 0 , x−α (cid:55)→ (cid:18) 0 1 (cid:19) 0 0 (cid:18) 1 (cid:19) , , hα (cid:55)→ 0 0 −1 provides an isomorphism between Sα := gα ⊕ g−α ⊕ [gα, g−α] and sl2(C). Then {hα1, . . . , hαn} ∪ {xα : α ∈ Φ} (1.3) is a basis of g. For any α, β ∈ Φ such that β (cid:54)= ±α, we have [xα, xβ] = cα,βxα+β for some cα,β ∈ C. It is possible to scale xα in such a way that, for all α, β ∈ Φ, then c−α,−β = −cα,β. Then cα,β equals 0 if α + β /∈ Φ and otherwise equals3 ±k for k the least positive integer for which α − kβ /∈ Φ. In particular, all the structure constants for the chosen basis are in Z. The basis is called a Chevalley basis. The proof is based on representation theory of sl2(C): any irreducible represen- tation of dimension r + 1 works in the same way as the action of h = X ∂ ∂X − Y ∂ ∂Y , x = X ∂ ∂Y , y = Y ∂ ∂X on the homogeneous rth degree polynomials in two variables X and Y (each mono- mial is connected with any other). This is applied to Sα ∼= sl2(C) acting on the vector space ⊕i∈Zgβ+iα to get a "chain without holes", which allows to prove Fact 2. When Killing discovered the five exceptional complex Lie algebras, mathemati- cians were not expecting to find them. It was a surprise, and not only for Killing. 3Of course, it is possible to be more precise about the determination of the sign here. But for our purpose, it is sufficient to say that, for "some" pairs of roots, the sign can be chosen arbitrarily and that determines the signs for all remaining pairs of roots. Also, most of the textbooks about the topic contain existence theorems guaranteeing the existence of a complex semisimple Lie algebra whose root system has a prefixed Cartan matrix. The uniqueness is consequence of the Serre relations. 6 C. DRAPER He was trying to prove that sln(C), son(C) and spn(C) are the only simple finite- dimensional complex Lie algebras (at the beginning he was not even aware of spn(C)). This Killing paper has been called the Greatest Mathematical Paper of all Time in [Co89], although Killing was not very proud of his work. In part the paper contains some mistakes (corrected in [Ca94]), but probably the reason for his dis- satisfaction is that he wanted to find all the real Lie algebras, since he was interested in Geometry. The same idea may well be upheld by many potential readers, but we cannot forget the close relationship between complex and real Lie algebras. The classification of the real ones goes through the classification of the complex case: let us take into account that one of the main described tools has been the diag- onalization of the semisimple operators, which is not always possible in the real case. Such (real) classification was realized by Cartan in [Ca14], after studying and completing Killing's work. 1.1. On simple real Lie algebras. We will only mention a pair of facts about simple real Lie algebras. If L is a simple real Lie algebra, C • Either L is just a complex simple Lie algebra, but considered as a real Lie algebra; = L ⊗R C = L ⊕ iL is a complex simple Lie algebra. In this case • Or L C has a root system X, then L is called a L is said a real form of L real algebra of type X. There are 17 real simple Lie algebras of exceptional types, two of them of type G2, which will be our main goal of study today. We mentioned above that for any X ∈ {Al, Bl, Cl, Dl, E6, E7, E8, F4, G2}, there is a unique simple complex Lie algebra g with related root system X. Now we will explain why at least there exist two real forms of g, that is, two real Lie algebras of type X, for each X. If we take a Chevalley basis of g as in Eq. (1.3), then C. If L n(cid:88) Ls = Rhαj ⊕ (cid:1) Rxα (cid:0)(cid:88) α∈Φ is diagonalizable for any h ∈ h0 := (cid:80)n j=1 C s = g. This is clear by recalling that the structure is a real Lie algebra such that L constants of the Chevalley basis belong to Z. This real form is called split due to the fact of possessing a root decomposition similar to that one in the complex case (adh (cid:1) j=1 Rhαj -with real eigenvalues-). It turns out that the signature of the Killing form κs of Ls coincides with n, the rank of g, because κsh0 is positive definite and the decomposition h0⊕ is an orthogonal decomposition where the hyperbolic planes Rxα ⊕ Rx−α do not contribute to the signature. (cid:0)Rxα⊕Rx−α (cid:80) α∈Φ+ On the other hand, n(cid:88) j=1 Lc = Rihαj ⊕ (cid:88) α∈Φ+ (cid:0)R(xα − x−α) ⊕ Ri(xα + x−α)(cid:1) also satisfies [Lc, Lc] ⊂ Lc and hence it is again a real form of g. It is not difficult to check that its Killing form (obtained by restriction of the Killing form of g) is negative definite. In particular Lc cannot be isomorphic to Ls. This real form is called compact, due to the following fact: being the Killing form negative definite implies that the related Lie group is compact by Myers' Theorem [Ne, 10.24] (any connected Lie group with negative definite Killing form is compact, according to [Ne, 11.11]). (Furthermore, any other real form with negative definite Killing form is isomorphic to Lc.) If L is another real form of g, then L admits a Z2-grading L = L¯0 ⊕ L¯1, called Cartan decomposition of L, such that L¯0 ⊕ iL¯1 is a compact real form of g. From NOTES ON G2 7 this, it is possible to conclude that there is a one-to-one correspondence between isomorphism classes of real forms of g and conjugacy classes of involutive auto- morphisms of g. (Note that any Z2-grading on g is produced -as the eigenspace decomposition- by an order two automorphism of g and vice versa.) In the G2-case, there is only one conjugacy class of order two automorphisms, since there is only one symmetric space quotient of the group G2, namely, G2/SO(4). This implies that there are only two real forms of the complex Lie algebra of type G2, an interesting fact which will study in detail in Section 5.3. 2. A first linear model of (split) G2 2.1. The model. Let U = F3 be the column vectors for F ∈ {R, C}, and let × denote the usual cross product on U. For any x ∈ U, let lx denote the coordinate matrix, in the canonical basis {ei}i=1,2,3 of F3, of the map y (cid:55)→ x × y. Hence, for (2.1) x = x2 x3  x1  ⇒ lx =   0 −2yt −2xt x y a lx ly −at x3 −x2 x2 −x1 0 −x3 0 x1  0  .   : a ∈ sl3(F), x, y ∈ F3  0 −2yt −2xt x y a lx ly −at . Proposition 2.1. L = is a Lie subalgebra of gl7(F) = (Mat7(F), [ , ]) of dimension 14. Proof. The dimension of L as a vector space is clear. We only have to see that [A, B] = AB−BA ∈ L for any A, B ∈ L. Let us denote M(a,x,y) = We compute, for a, b ∈ sl3(F), and x, y, u, v ∈ U: i) [M(a,0,0), M(b,0,0)] = M([a,b],0,0); Because [at, bt] = −[a, b]t and tr([a, b]) = 0. ii) [M(a,0,0), M(0,u,0)] = M(0,au,0); Because lau = −(lua + atlu). iii) [M(a,0,0), M(0,0,y)] = M(0,0,−aty); (the same argument as in ii). iv) [M(0,x,0), M(0,u,0)] = M(0,0,2x×u); Because obviously lxu − lux = 2x × u, and not so obviously lx×u = −xut +uxt. This is a direct consequence of the identity of the cross product (x× y)× z = (x· z)y − (y · z)x (also called triple product expansion), since, relative to the canonical basis, the matrix of (x · −)y (it sends ej to xjy) is yxt (yixj in the (i, j)th place). v) [M(0,x,0), M(0,0,v)] = M(prsl3(F)(−3xvt),0,0); When doing the bracket, we get  −2xtv + 2vtx 0 0 (cid:88) There is 0 in the (1, 1)th position since xtv = x · v = vtx. The matrix in position (3, 3) is the opposite-transpose of that one in (2, 2), since lx and lv are skew-symmetric matrices. Finally, observe that lxlv = (xjvi)i,j − ( k xkvk)I3 = vxt − (vtx)I3, tr(I3) I3 = prsl3(F)(3vxt). (It is obvious that so that lxlv + 2vxt = 3vxt − tr(3vxt) tr(vxt) = vtx.) vi) [M(0,0,y), M(0,0,v)] = M(0,2y×v,0); (analogous to iv). . −2xvt − lvlx 0 0 0 0 2vxt + lxlv 8 C. DRAPER As before {ei}3 (cid:3) i=1 denotes the canonical basis of F3, and now Eij ∈ Mat3(F) denotes the unit square matrix whose entries are all of them zero except for the (i, j)th, which is 1. Take h = {M(a,0,0) : a ∈ sl3(F) diagonal matrix}, which is an abelian subalgebra of L. Define εi : h → F, for i = 1, 2, 3, as basis of L of eigenvectors for ad h. By using i), ii) and iii) above, εi (h = diag{0, s1, s2, s3,−s1,−s2,−s3}) = si Of course {ε1, ε2} is a basis of h∗, since (cid:80)3 [h, M(Eij ,0,0)] = M([(cid:80) [h, M(0,ei,0)] = M(0,(cid:80) [h, M(0,0,ei)] = M(0,0,−((cid:80) k skEkk,Eij ],0,0) = (si − sj)M(Eij ,0,0) = (εi − εj)(h)M(Eij ,0,0), k skEkkei,0) = M(0,siei,0) = εi(h)M(0,ei,0), k skEkk)tei) = M(0,0,−siei) = −εi(h)M(0,ei,0); i=1 εi = 0. We are computing now a and hence the Lie algebra L can be decomposed as 3(cid:88) i=1 ( si = 0). (cid:33) (cid:32)(cid:88) α∈Φ L = h ⊕ Lα , being Φ = {εi − εj : i (cid:54)= j} ∪ {±εi : i = 1, 2, 3} ⊂ h∗, for (i (cid:54)= j) (i = 1, 2, 3). Lεi−εj = FM(Eij ,0,0), Lεi = FM(0,ei,0), L−εi = FM(0,0,ei), Thus, L has a root decomposition (even for F = R). Does this mean that L is simple? Yes, it does: Proposition 2.2. L is a simple Lie algebra of type G2. Proof. Take 0 (cid:54)= I an ideal of L. As [h, I] ⊂ I, then I = I ∩ h ⊕ (⊕α∈ΦI ∩ Lα). Consider S = {M(a,0,0) : a ∈ sl3(F)}, which is a Lie subalgebra of L isomorphic to sl3(F). In case I∩S (cid:54)= 0, by the simplicity of S we have I∩S = S, so that h ⊂ S ⊂ I and hence L = h+[h, L] ⊂ I +[I, L] ⊂ I and I = L. The other possibility, I∩S = 0, does not occur, since in such a case there is some α ∈ ±{εi : i = 1, 2, 3} such that I ∩ Lα (cid:54)= 0. As Lα has dimension 1, then Lα ⊂ I and so [Lα, L−α] ⊂ [I, L] ⊂ I. This is a contradiction taking into account that4 0 (cid:54)= [Lα, L−α] ⊂ h ⊂ S, but I does not contain any nonzero element of S. This finishes the simplicity. Now we call α1 := ε2 and α2 := ε1 − ε2. Hence α1 + α2 = ε1, 3α1 + α2 = ε2 − ε3, 2α1 + α2 = ε1 + ε2 = −ε3, 3α1 + 2α2 = ε1 − ε3; so that Φ = ±{α1, α2, α1 + α2, 2α1 + α2, 3α1 + α2, 3α1 + 2α2} is the root system of (cid:3) type G2 described in Section 1. The above propositions are valid for both C and R: if we denote LF for distin- guishing the considered field, LR is a real form of LC. From now on, we shall denote the simple complex Lie algebra LC by g2, and the real form LR by g2,2. The second number in the notation makes reference to the signature of the Killing form, which is 2, taking into account that LR is the split real form of LC since it admits a root decomposition, or alternatively κLα⊕L−α ≡ (cid:18) 0 (cid:19) 1 0 1 . 4A well known property of the root decompositions of simple Lie algebras is 0 (cid:54)= [Lα, L−α] ⊂ 3 I3, so [Lεi , L−εi ] (cid:51) In this case, as eiet its projection on sl3(F) is Eii − 1 h. [M(0,ei,0), M(0,0,ei)] = M(−2Eii+Ei+1,i+1+Ei+2,i+2,0,0) (cid:54)= 0. i = Eii, NOTES ON G2 9 I would like to remark that finding a model (a description of the algebra inde- pendent from the roots and from the basis corresponding to such roots) has been an easy task for us now, but surprisingly, it took many years. A very suggestive paper about the exceptional group G2, which includes many references and curiosities, is [A08]. There, the references about the first results on G2 can be consulted. These results were not immediate after Killing's discovery, probably because the whole Lie theory was being developed. Both Élie Cartan and Friedrich Engel obtained, independently in 1893, that the Lie algebra of vector fields on C5 whose local flows preserve a plane distribution given by certain Pfaffian system, is the Lie algebra of type g2 (that is, g2 appears as the infinitesimal symmetries of this distribution). Both authors gave a second (this time different) geometrical realization of g2; but an algebraical realization had to wait until 1900 [En00], when Engel finally gave a realization including the missing compact form (which is besides a global version, for groups not only for algebras). His description based on 3-forms on R7 will be carefully developed in Section 4.2. The most popular description of g2 as the derivation algebra of the octonions (both split and division octonions provide the split and the compact real forms, respectively) is due to Cartan [Ca14] (in 1914, although already mentioned in a letter in 1908). We will arrive at it in Section 3.2. Remark 2.3. As a consequence of the computations made in the proof of Propo- sition 2.1, if we define L = L¯0 ⊕ L¯1 ⊕ L¯2 by L¯0 = sl3(F), L¯1 = U = F3, L¯2 = U∗ , with the product given by [L¯i, L¯j] ⊂ L¯i+¯j such that L¯0 is subalgebra and: • The actions of L¯0 on L¯1 and L¯2 are the natural one and the dual one, respectively; • [x, y] = (2x × y)t and [xt, yt] = 2x × y; • [x, yt] = −3prsl3(F)(xyt); → F given by (x, ut) (cid:55)→ utx ∈ F. then L is a Lie algebra of type G2 (just g2 if F = C and g2,2 if F = R). We are identifying the dual of F3 with the row vectors {vt : v ∈ F3}, thanks to the map U × U∗ A model equivalent to this one can be consulted, for instance, in [FuH, §22.4], where some linear models with the same philosophy are exhibited for other excep- tional Lie algebras, in particular for type E8 as sl9(F)⊕ V⊕ V∗ with V = ∧3F9. The underlying fact is that, in the complex case, any reductive subalgebra of maximal rank of a simple Lie algebra (in our case g2, the subalgebra is sl3(C)) induces a grading on this algebra by means of an abelian group (in our case, a Z3-grading), in such a way that the nontrivial components of the grading are irreducible modules (in our case, the natural one and the dual of the natural one). This philosophy is made explicit in [D08, Theorem 1], and applied for describing some linear models of the complex -consequently, also the real split- exceptional Lie algebra of type F4. This kind of models based in linear algebra turn out to be very comfortable to work with. For models on the compact algebra of type G2, one has to bear in mind that In spite of 2. compact algebras are never graded by any group different from Zr that, one can find in [BDE03] some linear models of the compact real algebra of type G2 (in a more general context, with few restrictions on the underground field), whose underlying common feature is the existence of a "nice" (simple or semisimple) subalgebra h such that (g2, h) is a reductive pair -related to reductive homogeneous spaces-. We will exhibit such a compact model in Proposition 5.4 as a consequence of the relationship between G2 and the six-dimensional sphere. Other well known works are [Vi66] on some unified models, [Kan73] on models based in Z-gradings, and specially [Tit66], which constructs all the exceptional 10 C. DRAPER if s ∈ F, u, v ∈ U = F3. In particular, for h = diag{0, s1, s2, s3,−s1,−s2,−s3} ∈ h,  ,  0 0 ei simple Lie algebras in a uniform way by using alternative and Jordan algebras, by a procedure which today is called Tits construction. 2.2. A little bit of representation theory. We consider the ground field F to be either C or R. Let V = F7 throughout this and the next sections. First we can observe that V is an irreducible representation of LF, with the natural action by matrix multiplication. Write the elements in V in blocks of sizes 1 + 3 + 3. Indeed, it is trivial that u v M(a,x,y)  s  0  =  0  ,  , Vεi = F  1 0 0 h ei 0 sx + au + y × v sy + x × u − atv  =  −2ytu − 2xtv  ,  0  = si  0  ,  = −si  ,  0  , V−εi = F  0 0 ei ei 0 h so that V = F7 decomposes as V = ⊕µ∈h∗ Vµ, for  1 0 0 h V0 = F 0 0 ei 0 0 ei i = 1, 2, 3. The set Λ = {µ ∈ h∗ : Vµ (cid:54)= 0} is called the set of weights, for Vµ := {v ∈ V : hv = µ(h)v ∀h ∈ h}. Now, any nonzero LF-submodule W ≤ V is homogeneous and Vµ ∩ W (cid:54)= 0 for some µ ∈ h∗. Then Vµ ⊂ W , and, by acting the root spaces Lα we get the whole V ⊂ W , and consequently the irreducibility of V . This behavior is not particular for LF and V , but any finite-dimensional repre- sentation W of a complex semisimple Lie algebra g decomposes as a direct sum of weight spaces W = ⊕µ∈ΛWµ, which clearly satisfy gαWµ ⊂ Wα+µ for all α ∈ Φ and µ ∈ Λ. If the module is irreducible, then it is generated by any nonzero element in Wµ for a weight µ such that α + µ /∈ Λ for all α ∈ Φ+. This weight is called max- imal and the module W is usually denoted by V (µ). In our example the weights of V are just the short roots in Φ and {0}, so that the maximal weight of V is −ε3 = 2α1 + α2 = ω1 (the short maximal root), where ωi ∈ h∗ is (always) defined of some irreducible module are precisely(cid:80)n by (cid:104)ωi, αj(cid:105) = δij. (This is possible for any semisimple g, since Cartan matrix is invertible.) The possible abstract weights µ ∈ h∗ such that µ is the maximal weight i=1 siωi, with si ∈ Z≥0, which are called dominant weights (Λ+). The representations V (ωi) are called the fundamental rep- resentations, which in our g2-case are V (ω1) ∼= C7 and V (ω2) ∼= g2, the natural and the adjoint representation respectively (3α1 + 2α2 = ω2). This implies that the nontrivial g2-representation of the least possible dimension is just C7, so that, if one wants to see the complex Lie algebra of type G2 as a subalgebra of glm(C) for some m, then m ≥ 7, what explains in part our election of model in the previous subsection. All this tells us how representation theory works for the complex Lie algebra g2 = LC. Any finite-dimensional representation is completely reducible and hence direct sum of irreducible modules. Each summand is (isomorphic to) V (s1ω1+s2ω2) for some s1, s2 ∈ Z≥0, so that it lives in V (ω1)⊗s1 ⊗ V (ω2)⊗s2, that is, it is a submodule of a tensor product of copies of the adjoint module g2 and of the natural module C7. In any case, as LF has dimension 14 but dim gl7(F) = 49 (a long distance), there is still work to do until understanding what is LF preserving, or what is characterizing LF. NOTES ON G2 11 3. Understanding G2 better 3.1. G2 and the cross product in F7. Let n : V → F be the quadratic form whose related matrix in the canonical basis {ei}i=1,...,7 of V = F7 is  = −s2−2utv−2vtu = −s2−4u·v, (3.1)  ⇒ n  −1  s N = 0 0 0 0 −2I3 0 −2I3 0 where · denotes the usual scalar product if F = R and the usual inner product if F = C. Observe that in the real case the signature of this norm n is (4, 3), since in some orthogonal basis the matrix is diag{−I4, I3}. Lemma 3.1. The algebra LF defined in Proposition 2.1 satisfies u v LF ⊂ so(V, n) = {f ∈ gl(V ) : n(f (X), Y ) + n(X, f (Y )) = 0 ∀X, Y ∈ V }.  ,  s Proof. It suffices to check n(f (X), X) = 0 for all X ∈ V , by bilinearity. With our notation, M(a,x,y) = 2s(ytu + xtv) − 2(cid:0)sxtv + (au)tv + (y × v)tv + sytu + (x × u)tu − (atv)tu(cid:1)  −2ytu − 2xtv t −s sx + au + y × v sy + x × u − atv  =  s −2v −2u  u v u v n = −2((au)tv − vt(au)) = 0, since (y × v)tv = det(y, v, v) = 0 is the mixed product in F3. (cid:18) 7 (cid:19) Thus n is LF-invariant. But so(V, n) has dimension = 21, so that not every element preserving the norm belongs to LF (g2 and g2,2 have to preserve something besides the norm). Definition 3.2. We will say that a bilinear map × : W ×W → W is a cross product if there exists a norm (nondegenerate quadratic form) q : W → F such that 2 (cid:3) • q(u × v, u) = q(u × v, v) = 0 • q(u × v, u × v) = (cid:12)(cid:12)(cid:12)(cid:12) q(u, u) q(v, u) q(u, v) q(v, v) (cid:12)(cid:12)(cid:12)(cid:12) There are no cross products in arbitrary dimension. But, for dimension 7, there for all u, v ∈ W . are: Lemma 3.3. The following is a cross product in V = F7: 2uty − 2vtx  s x y for the norm given in Eq. (3.1). u v sx − tu − 2v × y −sy + tv + 2u × x  ,  ×  s  :=  t  and Y = u v  ,  t  x y Proof. First, for X = n(X, X × Y ) = (−s − 2vt − 2ut)  2uty − 2vtx sx − tu − 2v × y −sy + tv + 2u × x = −2s(uty) + 2s(vty) − 2svtx + 2tvtu + 4vt(v × y) = 4 det(v, v, y) − 4 det(u, u, x) = 0. +2suty − 2tutv − 4ut(u × x) 12 C. DRAPER Similarly n(Y, X × Y ) = 0. Finally, recall that, by linearizing the identity (v × y) · (v × y) = (v · v)(y · y) − (v · y)(y · v), we obtain the following form of the Lagrange identity in F3: (v × y) · (u × x) = (v · u)(y · x) − (v · x)(y · u). Hence n(X, Y )2 − n(X)n(Y ) + n(X × Y ) = = (st + 2vtx + 2uty)2 − (s2 + 4u · v)(t2 + 4x · y) − 4(u · y − v · x)2 −4(sx − tu − 2v × y) · (−sy + tv + 2u × x) = s2t2 + 4(v · x)2 + 4(u · y)2 + 4stv · x + 4stu · y + 8(v · x)(u · y) −s2t2 − 4s2x · y − 4t2u · v − 16(u · v)(x · y) − 4(u · y)2 − 4(v · x)2 + 8(u · y)(v · x) +4s2x · y − 4stx · v − 4stu · y + 4t2u · v + 16(v × y) · (u × x) = 16(v · x)(u · y) − 16(u · v)(x · y) + 16(v × y) · (u × x) = 0. (cid:3) LF is precisely the Lie algebra making × invariant. Proposition 3.4. LF = Der(V,×). Proof. Denote by D = Der(V,×) = {f ∈ gl7(F) : f (X × Y ) = f (X) × Y + X × f (Y ) ∀X, Y ∈ V }. Let us prove first that if f ∈ LF, then f behaves well with respect the cross product, that is, LF ⊂ D. • For f = M(a,0,0), X =  and Y =  s  u v ,  t x y 0 f (X) × Y + X × f (Y ) = −tau + 2(atv) × y + sax + 2v × (aty) 2(au) × x − tatv + saty + 2u × (ax) f (X × Y ) = f ∗ sx − tu − 2v × y −sy + tv + 2u × x 0 a(sx − tu − 2v × y) at(sy − tv − 2u × x)  =   which coincide because for any a = (aij) ∈ sl3(F), au × x + u × ax = −at(u × x) for all u, x ∈ F3. For instance, for elements in the canonical basis u = ei, x = ej, (j = i+1, k = i+2), u×x = ek and au×x+u×ax = −akiei−akjej +(aii +ajj)ek = −akiei − akjej − akkek = −atek. (That is, U × U → U∗, (u, v) (cid:55)→ det(u, v,−), is sl3(F)-invariant. At the group level, the matrices of determinant 1 preserve the volume form, since det(P u, P v, P w) = det P det(u, v, w).)  ,  ,  , • For f = M(0,z,0), f (X) × Y + X × f (Y ) =  −2(ztv)x − 2(z × u) × y + 2(zty)u − 2v × (z × x) 2szty − 2(z × u)tx + 2ut(z × x) − 2tvtz sz × x − tz × u + 2(ztv)y − 2(zty)v  −2zt(−sy + tv + 2u × x) 2(uty − vtx)z  . z × (sx − tu − 2v × y) f (X × Y ) = The scalars in the (1)-position coincide because zt(u × x) = det(z, u, x) is alternat- ing. The vectors in positions (2) and (3) coincide due to the identity of the double cross product z × (v × y) = (z · y)v − (z · v)y. • The same arguments work to prove that f = M(0,0,z) ∈ D. Conversely, let us prove D ⊂ LF. Recall that D = Der(V,×) is the Lie algebra of the derivations of V with respect to the product given by ×. Although with NOTES ON G2 13 this product, V is neither associative nor commutative, we can obtain a lot of information looking at V as an algebra. First, (V,×) possesses a Z3-grading:   s 0 0  : s ∈ F  , V¯1 =   0 u 0  : u ∈ F3  , V¯2 =   0 0 v  : v ∈ F3  . V¯0 = This means that V¯i × V¯j ⊂ V¯i+¯j (the sum of the indices in Z3), what is an straight- forward computation. Hence the Lie algebra gl(V ) is also Z3-graded (gl(V ) = ⊕¯i∈Z3 gl(V )¯i, where f ∈ gl(V )¯i when f (V¯j) ⊂ V¯i+¯j for all ¯j ∈ Z3) and D = Der(V,×) is also Z3-graded D¯i, for D¯i = D ∩ gl(V )¯i). This helps to the proof since we have only to (D = ⊕¯i∈Z3 check that D¯i ⊂ LF ∩ gl(V )¯i (in fact, they coincide), even it suffices to prove that dim D¯0 = 8 and dim D¯1 = dim D¯2 = 3. Besides, take into account that the subspace V¯1 generates V as an algebra (since V¯1 × V¯1 = V¯2 and V¯1 × V¯2 = V¯0 -not only contained-), what implies that if d, d(cid:48) ∈ D satisfy dV¯1 = d(cid:48) • Let d ∈ D¯1. There is z ∈ U such that d V¯1, then d = d(cid:48).  =  0  1 0 0 z 0 ment belongs to V¯1). So, let us check that d = M(0,z,0), or equivalently d(X) = M(0,z,0)(X) for all X ∈ V¯1. As d(X) ∈ V¯2, there are u, w ∈ U such that X =  0  = d 0 w . Hence  × X  1 0 0 0 0 0 d 0 0 d 0 0 z 0 u 0 0 w 0 w and d(X) = z × u  =  0 so that w = 2z × u − w and d(X) = • Let d ∈ D¯0. Then there are s ∈ F, a, b ∈ gl3(F) such that  1  +  0  0  ×  ×  = M(0,z,0)(X).  0  ,  s  =  =  0  ,  0  =  0  .  0  1 since  0  = d  0  = d  1   ×  0  =  .  0  = d  2  = 2  ×  (aei) · ej + ei · (bej)  0   =  0  ×  0  = d  0  = d and (aei) · ej + ei · (bej) = aji + bij. Third, Second, b = −at, since  0 su + au 0 bw au 0 au 0 0 = d 0 ej 0 0 ei 0 0 w u 0 u 0 u 0 0 0 0 0 0 0 0 0 d 0 2be3 0 2e3 e1 0 e2 0 Let us check that a is a zero trace matrix and d = M(a,0,0). First, note that s = 0,  (since this ele-   0  ,  0 u 0  ,  . 2(ae1 × e2 + e1 × ae2) 14 (cid:3) C. DRAPER (cid:80) a3iei = −ate3 = ae1 × e2 + e1 × ae2 and the coefficient of e3 in both sides Thus − of this identity is −a33 = a11 + a22, giving zero trace. 3.2. G2 and the octonions. Define C = F ⊕ V (isomorphic to F8 as a vector space) with the product where 1 ∈ F is a unit and XY := −n(X, Y )1 + X × Y, (3.2) for all X, Y ∈ V . We extend our norm n : V → F to the norm n : C → F defined by n(1) = 1, n(1, V ) = 0 and nV = n. We use the same symbol for the norm and for its polar form n(X, Y ) = 1 Definition 3.5. An octonion algebra is • A unital algebra of dimension 8; • With a nondegenerate multiplicative quadratic form n, that is, n(XY ) = n(X)n(Y ) for all X, Y . 2 (n(X + Y ) − n(X) − n(Y )). Lemma 3.6. C is an octonion algebra. Proof. We have only to prove that n(XY ) = n(X)n(Y ). If either X or Y is equal to 1, it is clear. And if X and Y are in V , n(XY ) = n(−n(X, Y )1 + X × Y,−n(X, Y )1 + X × Y ) = (−n(X, Y ))2n(1, 1) + 0 + n(X × Y ) = n(X, Y )2 + (cid:12)(cid:12)(cid:12)(cid:12) n(X, X) n(X, Y ) n(Y, X) n(Y, Y ) (cid:12)(cid:12)(cid:12)(cid:12) = n(X, X)n(Y, Y ). (cid:3) Remark 3.7. The relationship between cross products and multiplicative norms does not exist only in the octonion case. In fact, a composition algebra is an algebra with a nondegenerate multiplicative quadratic form. According to the classical Hurwitz's Theorem in 1898, there are unital finite-dimensional composition real5 algebras only for dimensions 1, 2, 4 and 8. But whenever we have a cross product in an R-vector space W , then R ⊕ W is a composition algebra by the same argument as in the above proof. This is the way used by Brown and Gray [BrG67] to prove that cross products are only possible for dimensions 1, 3 and 7. Remark 3.8. Note that C is a quadratic algebra, that is, every element X ∈ C satisfies a quadratic equation with coefficients in F: This is a straightforward computation: Y 2 = s2 + 2sX − n(X), n(Y ) = s2 + n(X) and n(Y, 1) = s. Now it is usual to denote X 2 − 2n(X, 1)X + n(X)1 = 0. (3.3) if Y = s + X with X ∈ V , s ∈ F, then tr(X) := 2n(X, 1) ∈ F, the trace, so that V can be identified with the zero trace elements in C, i.e. those ones orthogonal to 1 (sometimes called imaginary octonions by analogy with the imaginary complex numbers). It will be useful to observe that the coefficients in the quadratic equation (3.3) are determined, i.e. if X /∈ F1 and X 2 − sX + t1 = 0 then s = tr(X) and t = n(X). 5The theory has subsequently been generalized to arbitrary quadratic forms and arbitrary fields, in particular C. NOTES ON G2 15 Proposition 3.9. LF is isomorphic to the Lie algebra of derivations of the octo- nions Der(C) = {d ∈ gl(C) : d(XY ) = d(X)Y + Xd(Y )}, by means of d ∈ LF = Der(V,×) (cid:55)→ d ∈ Der(C), (cid:26) d(1) = 0, dV = d. Furthermore, Der(C) ⊂ so(C, n). Proof. Let d ∈ LF, X, Y ∈ V . So, d(XY ) = d(X × Y ) = d(X × Y ) = d(X) × Y + X × d(Y ) is equal to d(X)Y + X d(Y ) = −n(d(X), Y )1 + d(X)× Y − n(X, d(Y ))1 + X × d(Y ), since d ∈ so(V, n) by Lemma 3.1. Also d(XY ) = d(X)Y + X d(Y ) is obviously true if either X or Y is multiple of 1, so that d ∈ Der(C). Conversely, take d ∈ Der(C). Note first that d(1) = 2d(1), so that d(1) = 0. Second, we are checking that Der(C) restricts to V . We apply d ∈ Der(C) to Eq. (3.3) getting d(X 2) − tr(X)d(X) = 0 for all X ∈ C. Thus tr(X)d(X) = d(X 2) = d(X)X + Xd(X) = (X + d(X))2 − d(X)2 − X 2 (3.3) = tr(X + d(X)) (X + d(X)) − tr(d(X)) d(X) − tr(X) X = tr(X) d(X) + tr(d(X)) X − 2n(X, d(X))1, −n(X + d(X))1 + n(d(X))1 + n(X)1 hence tr(d(X)) X = 2n(X, d(X))1 for all X ∈ C. In particular, if X ∈ V , by projecting on V , we have tr(d(X)) = 0, so that d(X) ∈ V . As d(1) = 0, then d(C) ⊂ V . Third, and finally, if X, Y ∈ V , d ∈ Der(C), d(X × Y ) = d(XY ) = d(X)Y + Xd(Y ) = −n(d(X), Y )1 + d(X) × Y − n(X, d(Y ))1 + X × d(Y ). Here we have used that the image of d is in V for applying (3.2). In particular (projecting on V ), d(X × Y ) = d(X) × Y + X × d(Y ) and d ∈ LF. As a bonus (projecting on F), we also get n(d(X), Y ) + n(X, d(Y )) = 0 if X, Y ∈ V (also consequence from Lemma 3.1); but if both X and Y are scalars the statement is clear too; and if X ∈ V , Y = 1, the equality consists of tr(d(X)) = 0. To (cid:3) summarize, Der(C) ⊂ so(C, n). 4. Generic 3-forms in F7 In [En00], Engel discovered an elegant description of the Lie algebra of type G2 which enclosed the compact form. Before giving proofs of his results about 3-forms (of course, not Engel's original proofs), we will try to go naturally to this concept from our previous situation. 4.1. G2 and the triple product. Define a triple product in V = F7 as follows: (4.1) It is an alternating (trilinear) map, taking into account that n(X × Y, Y ) = 0 and n(X × Y, X) = 0, as in Lemma 3.3. So, it can be identified with a 3-form ∧3V → F. LF ⊂ {f ∈ gl7(F) : {f (X), Y, Z} + {X, f (Y ), Z} + {X, Y, f (Z)} = 0 ∀X, Y, Z ∈ V }, which is a trivial consequence of Proposition 3.4 and the fact f ∈ LF ⊂ so(V, n): { , , } : V × V × V → F, {X, Y, Z} := n(X × Y, Z). Note that n(f (X) × Y, Z) + n(X × f (Y ), Z) + n(X × Y, f (Z)) = = n(f (X × Y ), Z) + n(X × Y, f (Z)) = 0. Our objective is to prove the equality of both algebras.6 To this aim, it is convenient to understand the 3-form better: 6 This is frequently mentioned in the literature, for instance in [SM77, Example 30], which just uses the model in 2.1. 16 C. DRAPER Remark 4.1. If {V¯i, V¯j, V¯k} (cid:54)= 0 ⇒ ¯i + ¯j + ¯k = ¯0. Denote by  1 0 0  , Ei =  0 ei 0  , Fi =  ,  0 0 ei E0 = so that the canonical basis of V = F7 is Bc = {E0, E1, E2, E3, F1, F2, F3}. Ac- cording to Lemma 3.3, the only nonzero ×-products of basic elements (taking into account skew-symmetry) are: E0 × Ei = Ei, E0 × Fi = −Fi, Ei × Fi = 2E0. Ei × Ei+1 = 2Fi+2, Fi × Fi+1 = −2Ei+2, And, according to Eq. (3.1), the only nonzero n-products of basic elements (but taking into account symmetry) are: n(E0, E0) = −1, n(Ei, Fi) = −2. Hence, the only nonzero { , , }-products of basic elements (taking into account it is alternating) are: {E0, Ei, Fi} = −2, { , , } = −2(cid:0)E∗ {F1, F2, F3} = 4, 3 ∧ F ∗ 0 ∧ E∗ (cid:1) {E1, E2, E3} = −4, 0 ∧ E∗ 1 ∧ F ∗ 3 + 4 F ∗ 1 ∧ E∗ or, in other words, the triple product considered as a 3-form is 0 ∧ E∗ −4 E∗ 1 + E∗ 2 ∧ E∗ Proposition 4.2. LF = Der(V,{ , , }) = = {f ∈ gl7(F) : {f (X), Y, Z} + {X, f (Y ), Z} + {X, Y, f (Z)} = 0 ∀X, Y, Z ∈ V }. Proof. Let f be a map preserving the triple product in Eq. (4.1). Our aim is to prove that f ∈ so(V, n), because if we knew f ∈ so(V, n) then 2 + E∗ 2 ∧ F ∗ 3 . 2 ∧ F ∗ 1 ∧ F ∗ 3 0 = n(f (X × Y ), Z) + n(X × Y, f (Z)) = n(f (X × Y ), Z) + {X, Y, f (Z)} = n(f (X × Y ), Z) − {f (X), Y, Z} − {X, f (Y ), Z} = n(f (X × Y ) − f (X) × Y − X × f (Y ), Z), and so, as n is nondegenerate, this would imply f ∈ Der(V,×) = LF (Proposi- tion 3.4). i,j=0 represents the matrix of f relative to the basis Bc, then f ∈ so(V, n) equivales to AtN + N A = 0, that is, to the fact that A has, as a block- matrix, the shape If A = (aij)6 for some x, y ∈ F3 and b, c, d ∈ gl3(F) such that b + bt = 0 = c + ct. According to Remark 4.1, 0 = {f (E0), Ei, Fi} + {E0, f (Ei), Fi} + {E0, Ei, f (Fi)} = {a00E0, Ei, Fi} + {E0, aiiEi, Fi} + {E0, Ei, ai+3,i+3Fi} = −2(a00 + aii + ai+3,i+3), (4.3) and also 0 = {f (E1), E2, E3} + {E1, f (E2), E3} + {E1, E2, f (E3)} = −4(a11 + a22 + a33), 0 = {f (F1), F2, F3} + {F1, f (F2), F3} + {F1, F2, f (F3)} = 4(a44 + a55 + a66).  0 −2yt −2xt  , x y d c b −dt (4.2) ii) {X, Y, Z} = {Fi, Ei+1, Ei+2} provides ai,i(cid:48) = 0 (zero the whole diagonal of b = X23), {X, Y, Z} = {Fi+1, Ei+1, Ei+2} implies 2ai,i+1(cid:48) = a0,i+2, {X, Y, Z} = {Fi+2, Ei+1, Ei+2} implies −2ai,i+2(cid:48) = a0,i+1, and compiling 2 a0,i+1+1(cid:48) = −ai+1,i(cid:48) (sum modulo 3) and b is the information ai,i+1(cid:48) = 1 skew-symmetric. iii) {X, Y, Z} = {E0, Fi+1, Fi+2} leads to 2ai(cid:48),0 = ai+2,i+1(cid:48) − ai+1,i+2(cid:48), so that by ii), 2ai(cid:48),0 = 2ai+2,i+1(cid:48) = 2ai+2,i+2+2(cid:48) = −a0,i+2+1 = −a0i, which con- firms the relationship between the blocks X31 and X12. The blocks (2, 1), (3, 2), (1, 3) are obtained interchanging E's and F 's in the previous series of triplets {X, Y, Z}. (In fact we have directly checked b = ly, tr(d) = 0 and (cid:3) A ∈ LF.) Remark 4.3. The proof should be essentially based on the fact that the norm n can be recovered from the triple product, although this fact seems hidden at a first sight (that is the reason why preserving the triple product is preserv- ing the cross product, but the first object is easier). Soon we will try to prove this in general, but now let us check it in our concrete example -of triple prod- uct as in Eq. (4.1) and n given by Eq. (3.1)-: If X = (s u v)t, n(X) = −s2 − (expres- 4utv = 1 4 sion too reliant on our basis), which, multiplying with {E0, Ei+2, Fi+2} = −2 and varying the possibilities, yields −{X, E1, F1}{X, E2, F2} +(cid:80)3 (cid:88) i=1{X, Ei, Ei+1}{X, Fi, Fi+1} (−1)sgσ{X, eσ(1), eσ(2)}{X, eσ(3), eσ(4)}{eσ(5), eσ(6), eσ(7)}. n(X) = −1 2632 (cid:16) (cid:17) NOTES ON G2 If we now sum Eq. (4.3) for i = 1, 2, 3, we get −6a00 − 2(cid:80)6 i=1 aii = 0, that joint Now we apply {f (X), Y, Z} + {X, f (Y ), Z} + {X, Y, f (Z)} = 0 to some various with the two above identities imply a00 = 0. triplets {X, Y, Z}: 17 i) {X, Y, Z} = {E0, Ei, Fi+1} gives 0 = ai+1,i + ai(cid:48),i+1(cid:48) for i = 1, 2, 3; -In order to simplify the notation we have replaced the indices j + 3 with j(cid:48) when j = 1, 2, 3 (so, if for instance i = 3, then i + 1(cid:48) means 4, first sum modulo 3 and second sum 3 in Z);- {X, Y, Z} = {E0, Ei, Fi+2} gives 0 = ai+2,i + ai(cid:48),i+2(cid:48), also aii + ai+3,i+3 = 0 = a00 from (4.3) with the new notation is 0 = aii + ai(cid:48),i(cid:48), that gives X33 = −X t i,j=1 of sizes as in (4.2). 22 thinking of A as a block-matrix A = (Xij)3 σ∈S7 (If we take the sum over the permutations σ ∈ S7 with σ(1) < σ(2), σ(3) < σ(4) and σ(5) < σ(6) < σ(7), then the scalar is −1 233, since −8 in our first expression appears 6 times.) This motivates Lemma 4.10. Until now we have worked at the level of the Lie algebra since, although it may seem less conceptual than working with the group, the involved equations are linear! Anyway the previous results can be read in terms of groups. Corollary 4.4. The group generated by exp(g2) ⊂ GL7(C) (resp. exp(g2,2) ⊂ GL7(R)) is a complex simple Lie group of type G2 (resp. real noncompact) contained in SO(V, n), which coincides, for F = C (resp. R), with Aut(V,×) = {f ∈ GL7(F) : f (X) × f (Y ) = f (X × Y ) ∀X, Y ∈ V }, also with the connected component of the identity of the group {f ∈ GL7(F) : {f (X), f (Y ), f (Z)} = {X, Y, Z} ∀X, Y, Z ∈ V }, (4.4) (4.5) 18 C. DRAPER Aut(C) = {f ∈ GL(C) : f (XY ) = f (X)f (Y ) ∀X, Y ∈ C}. and it is isomorphic -under the obvious extension to C = F ⊕ V , making f (1) = 1- to (4.6) This result is not completely trivial. The groups in (4.4) and (4.6) are Lie groups, subgroups of GL7(F) [Wa, Theorem 3.54] and also is the group in (4.5), a closed subgroup of GL7(F). As we have proved that all of them have the same Lie algebra, then it is true that the connected components of the identity coincide [Wa, Theorem 3.19]. So the corollary will be clear when we see that the groups in (4.4) and (4.6) are connected. But proving the connectedness is not usually an easy task. We will check it carefully in Section 5.5, obtaining it as a consequence of the diffeomorphism G2/SL3(R) ∼= H 6 3 (1) (and of its complex version). The study of this homogeneous space and of the 6-dimensional sphere S6 ∼= G2/SU(3) is one of our purposes in this paper. In the meantime, observe that the group in (4.5) has nontrivial center in the complex case because ωidV belongs to the group if ω ∈ C is a cubic root of the unit.7 Remark 4.5. S. Garibaldi suggested to me some alternative arguments. For in- stance, Corollary 4.4 is immediate using this result: a closed subgroup of SL7(C) (hence a Lie group) with identity connected component equal to G2 should be con- tained in the subgroup of SL7(C) generated by G2 and the scalar matrices. This is the case for H any of the groups in (4.4), (4.5) and (4.6), whose identity connected component coincides with G2 by the above sections, or independently by using the argument in [GGu15, Lemma 5.3] that allows to conclude that any -Zariski- closed connected subgroup lying properly between G2 and SL7(C) is simple and then, by checking the possibilities with root data, necessarily H should be the orthogo- nal group SO(V, n), for n the G2-invariant quadratic form on V . But, of course, SO(V, n) stabilizes neither ×, nor { , , }, nor the octonionic product. In order to prove the above result, note that if g is in the normalizer of G2 in SL(V ), then Ad g ∈ Aut(SL(V )) given by conjugation restricts to G2, so Ad gG2 ∈ Aut(G2). A well known fact is that every automorphism of G2 is inner, so that there is g(cid:48) G2. Hence g(g(cid:48))−1 commutes with G2 and hence it belongs to EndG2(V ) = CidV by Schur's lemma (V is G2-irreducible). Thus g is the product of an element in G2 by a scalar map. But any subgroup of SL(V ) with identity connected component equal to G2 lives in the normalizer. 4.2. Generic 3-forms. Look at the previous subsection from a different approach. Consider the triple product in Eq. (4.1) as a 3-form Ω0 ∈ ∧3V ∗ (as in Remark 4.1). The action of GL7(F) ≡ GL(7) in the vector space of 3-forms is given by the pull- back (4.7) where f∗Ω(X, Y, Z) = Ω(f−1(X), f−1(Y ), f−1(Z)). Equation (4.5) precisely says that ∈ G2 such that Ad gG2 = Ad g(cid:48) (f, Ω) (cid:55)→ f · Ω := f GL(7) × ∧3V → ∧3V Ω ∗ ∗ ∗ , GΩ0 = {f ∈ GL(7) : f · Ω0 = Ω0}, is a Lie group of type G2, the isotropy group (the stabilizer) of the element Ω0 ∈ ∧3V ∗. This implies that the orbit of Ω0 (see Remark 4.6 below) OΩ0 = {f · Ω0 : f ∈ GL(7)} is a homogeneous space diffeomorphic to GL(7)/GΩ0, in particular of dimension 49 − 14 = 35. But dim∧3V ∗ = = 35, so that OΩ0 should be an open set of ∧3V ∗. In other words, Ω0 is a generic 3-form in ∧3V ∗. 7There are connected Lie groups with nontrivial centre, for instance (cid:101)G2,2 in Remark 5.18. (cid:18) 7 But the complex group (4.5) is not that case, its discrete center giving the number of connected components. (cid:19) 3 NOTES ON G2 19 Remark 4.6. We add some references about why the orbits are submanifolds and they are open sets or not according to the dimension of the isotropy group. (Although here we have a polynomial action on a vector space, this is not necessary at all.) If G is a Lie group acting (smoothly) on a connected manifold M, then every orbit Ox0 of a point x0 ∈ M is a submanifold [Sh, Lemma 2.13]. In the proof, Sharpe takes Hx0 = {g ∈ G : g · x0 = x0} the isotropy group and he shows that it is a closed subgroup of G, and that the induced map G/Hx0 → M, gH (cid:55)→ g · x0 is an injective immersion with image Ox0. So the inclusion ι : Ox0 (cid:44)→ M is also an immersion. Let us check that Ox0 is open if and only if dim Ox0 (which coincides with dim G − dim Hx0) equals dim M. Of course, this condition is necessary. But it is sufficient too, because if dim M = dim Ox0 and x ∈ Ox0, the differential (ι∗)x : TxOx0 → TxM is not only a monomorphism but an isomorphism. Hence, Inverse Function Theorem [Wa, 1.30 Corollary (a)] says that there is a neighborhood U (⊂ Ox0) of x such that ιU : U → ι(U ) = U is a diffeomorphism onto the open set U of M. The fact that each x ∈ Ox0 possesses U an open set of M such that x ∈ U ⊂ Ox0 means obviously that Ox0 is open. Definition 4.7. A p-form Ω ∈ ∧p(Fn)∗ (p ≤ n) is said a generic form when its orbit under the action of GL(n) is open, what happens if and only if dim GΩ = dim GL(n) − dim∧p(Fn) ∗ = n2 − (cid:18) n p (cid:19) , for GΩ = {f ∈ GL(n) : f · Ω = Ω}. The existence of generic p-forms is not possible in most of the dimensions. For instance, by dimension count, if p = 3, this forces n to be at most 8. But, according to the above arguments, the existence in case (n, p) = (7, 3) is guaranteed since just Ω0 is a generic 3-form, and of course all the 3-forms in the orbit OΩ0 are generic too. Some natural questions arise: how many orbits of generic 3-forms will there be in ∧3V ∗? And, in case there are more than one, which is the isotropy group of an element in a second orbit? Of course, this group should have dimension 14, but, is it also of type G2? Although the only -complex and real- simple Lie groups of dimension 14 have type G2, nobody ensures us (a priori) that the isotropy group of an arbitrary 3-form is simple. The answers were given so soon as in 1900 [En00].8 Theorem 4.8. There is only one orbit of generic 3-forms in ∧3(C7)∗. A represen- tative is (cid:1) Ω0 = −2(cid:0)E∗ 0 ∧ E∗ −4 E∗ 1 ∧ F ∗ 1 ∧ E∗ 1 + E∗ 2 ∧ E∗ 0 ∧ E∗ 3 + 4 F ∗ 2 ∧ F ∗ 1 ∧ F ∗ 2 + E∗ 2 ∧ F ∗ 3 . 0 ∧ E∗ 3 ∧ F ∗ 3 Theorem 4.9. There are just two orbits of generic 3-forms in ∧3(R7)∗. A pair of representatives are Ω0 and (4.8) Ω1 = e147 + e257 + e367 + e123 − e156 + e246 − e345, where the used notation is eijk := e∗ i ∧ e∗ j ∧ e∗ k for {e∗ i }7 i=1 a basis of (R7)∗. Before proving these theorems, we need some auxiliary results. 8The classification of the orbits of the 3-forms appears in Walter Reichel's thesis -see references in [A08]- and later in [Sch31]. From an Algebraic Geometry approach, [SM77] classifies the connected linear algebraic groups over the complex numbers such that there is a rational irreducible representation of the group on a finite-dimensional vector space whose orbit is Zariski-dense, and also their invariants. Trilinear alternating forms on a 7-dimensional vector space are classified in [CHe88]. 20 C. DRAPER Lemma 4.10. Let Ω ∈ ∧3V ∗ be a 3-form. For any basis B = {bi : i = 1, . . . , 7} of V = F7, define nΩ,B : V × V → F the bilinear symmetric form given by nΩ,B(X, Y ) = (−1)sgσΩ(X, bσ(1), bσ(2))Ω(Y, bσ(3), bσ(4))Ω(bσ(5), bσ(6), bσ(7)). (cid:88) σ∈S7 Then, form.9 a) nΩ,B(cid:48) = (det P ) nΩ,B for P the matrix of the change of basis from B(cid:48) to B, so that the norm is determined by the 3-form up to scalar multiple; b) If Ω is a generic 3-form, then nΩ,B is a nondegenerate bilinear symmetric Proof. The symmetry comes from sg σ = sg(σ(3) σ(4) σ(1) σ(2) σ(5) σ(6) σ(7)) for any permutation σ. The bilinearity is obvious, consequence of being Ω trilinear. For proving a), if P = (aij) is the matrix of the change of basis from B(cid:48) to B, i aikbi. For any X, Y ∈ V , the trilinearity implies then the kth element in B(cid:48) is(cid:80) that nΩ,B(cid:48)(X, Y ) is equal to (−1)sgσai1σ(1) . . . ai7σ(7)Ω(X, bi1, bi2 )Ω(Y, bi3 , bi4)Ω(bi5 , bi6, bi7). (4.9) σ ∈ S7 ij = 1, . . . , 7 (cid:88) (cid:88) σ∈S7 But this sum can be considered indexed in (i1, . . . , i7) ∈ S7 because, if (i1, . . . , i7) is a fixed 7-tuple which is not a permutation of (1, . . . , 7), then (−1)sgσai1σ(1) . . . ai7σ(7)Ω(X, bi1, bi2 )Ω(Y, bi3 , bi4)Ω(bi5, bi6 , bi7) = 0. Indeed, there will be j (cid:54)= k ∈ {1, . . . , 7} such that ij = ik, and for each σ ∈ S7 the related summand is just the opposite of the summand for the permu- tation µ = σ ◦ (j k) (µ(j) = σ(k), µ(k) = σ(j) and µ(i) = σ(i) for the re- det P =(cid:80) maining indices), since (−1)sgσ = −(−1)sgµ and the rest of the expression does not change. Thus, we obtain that (4.9) equals (det P ) nΩ,B(X, Y ) by writing σ∈S7 (−1)sgσ◦iai1σ(1) . . . ai7σ(7). In particular, the nondegeneracy does not depend on the choice of B. Of course, also nondegeneracy does not depend on the choice of the 3-form in a fixed orbit, since for every f ∈ GL(V ), nf·Ω,B(X, Y ) = nΩ,f−1(B)(f −1(X), f −1(Y )). So, let us fix B and also fix ψ : V ∗ → V an isomorphism (there is no natural isomorphism, but both spaces have the same dimension, so choose a basis {e∗ i }7 of V ∗ and take ψ(e∗ i ) = ei, for instance the dual basis of B). For each 3-form Ω, define ϕΩ : V → V ∗ by ϕΩ(X)(Y ) := nΩ,B(X, Y ). Thus, nΩ,B is nondegenerate if and only if ϕΩ is an injective (linear) map, or equivalently if det ϕΩ (cid:54)= 0, where we understand by det ϕΩ := det ψϕΩ. We have all the ingredients to see that if nΩ,B degenerates, then the orbit OΩ cannot be open. Simply note that, as det ϕΩ = 0, thus i=1 OΩ = {f · Ω : f ∈ GL(V )} ⊂ M = {ω ∈ ∧3V (cid:80) but M cannot contain any open subset of the vector space ∧3V ∗, because if ω = 1≤i<j<k≤7 aijke∗ k (we identify the form with its coordinates (aijk) ∈ F35), then det ϕω is a homogeneous of degree 21 polynomial with coefficients in F and (cid:3) variables aijk. : det ϕω = 0}, i ∧e∗ j ∧e∗ ∗ 9In fact, this is an equivalence, see Remark 4.15 below. NOTES ON G2 21 Remark 4.11. It will be useful the fact that one can choose a basis B such that B is orthonormal10 for nΩ,B (when it is nondegenerate). Indeed, take B0 an arbitrary basis of V and n = nΩ,B0. Take B1 = {ei}7 i=1 an orthonormal basis for n. Take i=1, where t ∈ F is a fixed nonzero scalar which we will determine next. B2 = {tei}7 The matrix of the change of basis of B2 to B1 is tI7 with determinant t7, so nΩ,B2(tei, tei) = t7nΩ,B1(tei, tei) = t9nΩ,B1(ei, ei) = t9α, being α the determinant of the change of basis of B1 to B0. Of course B2 is orthogonal for nΩ,B2 (the orthogonality does not change by multiplying by nonzero scalars), so in order to obtain the desired basis we have only to choose t ∈ F such that t9α = 1. Of course this is possible whether F = R or C. Furthermore, an argument similar to the previous one shows that a norm n related to Ω satisfying such property (B is orthonormal for n = nΩ,B) is determined up to a ninth root of the unit (in particular, completely determined in the real case). Simply observe that if B and B(cid:48) are orthonormal for n = nΩ,B and n(cid:48) = nΩ,B(cid:48) respectively, then n(cid:48) = αn for α = det P being P = CBB(cid:48) the change of basis, so that αP tP = I7 and, taking determinants, α9 = 1. Lemma 4.12. Let Ω be a 3-form on V = F7 such that the related bilinear form n as in Lemma 4.10 is nondegenerate. Then a) (det g)9 = 1 for all g ∈ GΩ; b) G+ c) If F = R, then GΩ ⊂ SO(V, n). Ω ⊂ GΩ ∩ SL(V ) = GΩ ∩ O(V, n); Proof. a) Fix B any basis of V = F7 orthonormal for n = nΩ,B. If g ∈ GΩ, as nΩ,B(X, Y ) = ng−1·Ω,B(X, Y ) = nΩ,g(B)(g(X), g(Y )), then g(B) satisfies the same property as B, and, by the above remark, (det g)9 = 1. b) For any g ∈ GΩ, by Lemma 4.10 a), nΩ,B(X, Y ) = nΩ,g(B)(g(X), g(Y )) = (det g) nΩ,B(g(X), g(Y )), so, if det g = 1 then g ∈ O(V, n) and the converse is also true. This proves GΩ ∩ SL(V ) = GΩ ∩ O(V, n). Moreover, by a), the map det : GΩ → {ξ ∈ C : ξ9 = 1} Ω. This means that G+ is well defined. But it is a continuous map, what implies that the image of G+ constant, equal to 1 since idV ∈ G+ so that hence a subgroup of SO(V, n). Ω is Ω is a subgroup of SL(V ), (cid:3) c) This is trivial now, since 1 is the only ninth root of the unit in R. Remark 4.13. Consider the action GL7(F) × ∧3V ∗ → ∧3V ∗ as in Eq. (4.7), and a 3-form Ω such that the related bilinear form n is nondegenerate. We know that dim GΩ ≥ 14, and it is just 14 when Ω is generic (afterwards we will prove that this will be the case, since n is nondegenerate). As in the above lemma, Ω ⊂ SO(V, n). We will prove in Lemma 5.16 and in Lemma 5.15 (complex and G+ Ω = GΩ ∩ O(V, n), but, in the meantime we will real cases, respectively) that G+ Ω := GΩ ∩ O(V, n), for working with this group of dimension equal to denote by G0 dim GΩ. Thus, the orbit G0 Ω · X for a nonisotropic vector X ∈ V of the action Ω × V → V will be fully contained in {Y ∈ V : n(Y ) = n(X)}, a manifold which G0 has dimension 6 whenever n (cid:54)= 0. This implies that the isotropy group has dimension dim HX = dim G0 HX = {g ∈ G0 Ω − dim G0 Ω : g(X) = X} Ω · X ≥ 14 − 6 = 8. 10In the real case, we understand that a basis is orthonormal when the vectors are pairwise orthogonal and each of them has length ±1. 22 C. DRAPER H Define Y ∧ Z ∈ V to satisfy n(Y ∧ Z,−) = Ω(Y, Z,−), which is well defined by nondegeneracy of n. Then HX is a subgroup of (cid:48) X := {g ∈ GL(V ) : n(g(Y )) = n(Y ), g(X) = X, g(X ∧ Y ) = X ∧ g(Y ) ∀Y ∈ V }, (4.10) which will also be a group of dimension at least 8. This fact HX ⊂ H(cid:48) X is a direct consequence of n(X ∧ gY, gZ) = Ω(gX, gY, gZ) = n(X ∧ Y, Z) = n(g(X ∧ Y ), gZ) if g ∈ HX, and of n nondegenerate. Now, we can almost exhibit a proof of the theorem, using arguments which are a mixture of [pr17, Hi00, He83] since the original Engel's proof is difficult to find. (Many ideas also appear in [Ha, Chapter 6], but the proofs are not complete there.) The key of the argument is to prove first the next result. Proposition 4.14. Let Ω ∈ ∧3V ∗ be a 3-form on V = C7 such that the bilin- ear symmetric form associated by Lemma 4.10 is nondegenerate. Then one of the related bilinear forms, n, satisfies that the multiplication ∧ : V × V → V defined by is a cross product. n(X ∧ Y, Z) = Ω(X, Y, Z), Proof. Take n = nΩ,B for B a basis of V such that B is orthonormal for n as in Remark 4.11. Afterwards we will replace n with a suitable nonzero multiple. We are going to prove that ∧ is a cross product relative to n, but, as Ω is alternating, clearly n(X ∧ Y, X) = n(X ∧ Y, Y ) = 0, and our concrete purpose will be to prove the second equation in Definition 3.2. Let X ∈ V be a nonisotropic element, that is, n(X) (cid:54)= 0. Thus V decomposes as a direct sum of CX and X⊥. Let us prove that for any Y ∈ V orthogonal to → X⊥ given X, then X ∧ (X ∧ Y ) = −n(X)Y . Consider the linear map f : X⊥ by f (Y ) = X ∧ Y , which is well defined since n(X ∧ Y, X) = 0. It is linear and skew-adjoint,11 since n(f (Y ), Z) = Ω(X, Y, Z) = −Ω(X, Z, Y ) = −n(Y, f (Z)). Indeed, if Let us check first that 0 is not an eigenvalue, that is, f bijective. f (Y ) = 0 for Y (cid:54)= 0, then X ∧ Y = 0, that is, Ω(X, Y,−) = 0. As X and Y ∈ X⊥ are linearly independent, complete {X, Y } to a basis B(cid:48) = {bi}7 i=1 of V such that b1 = X and b2 = Y . As X is not isotropic, 0 (cid:54)= nΩ,B(cid:48)(X, X) (the isotropic cone does not depend on the choice of the basis, by Lemma 4.10a)), so that there is σ ∈ S7 such that all Ω(X, bσ(1), bσ(2)), Ω(X, bσ(3), bσ(4)) and Ω(bσ(5), bσ(6), bσ(7)) are nonzero complex numbers. Hence the index 1 ∈ {σ(5), σ(6), σ(7)}. But 2 has to be another of the six remaining indices, so that Ω(X, Y, bi) (cid:54)= 0 for some i, a contradiction. Also, since f ∈ so(X⊥, n), n(Y, (f + λid)(Z)) = −n((f − λid)(Y ), Z) (4.11) for all Y, Z ∈ X⊥. We start with the Jordan decomposition of the endomorphism: = ker(f − λ1id)sλ1 ⊕ ··· ⊕ ker(f − λkid)sλk X ⊥ 11Our purpose is to check that f 2 is diagonalizable, but this cannot be concluded of the fact of being f 2 selfadjoint, since we are working on C. We need some extra-information. For f we have eigenvectors, but, as they are necessarily isotropic, there is no guarantee of an invariant complementary subspace. NOTES ON G2 23 and try to prove that all sλi's are 1. Note that ker(f − λid)s is orthogonal to ker(f − αid)t if λ + α (cid:54)= 0. Indeed, for 0 (cid:54)= Y ∈ ker(f − λid)s, 0 = n(Z, (f − λid)s(Y )) = (−1)sn((f + λid)s(Z), Y ) (4.11) so that Y is orthogonal to Im(f + λid)s. But ker(f − αid)t ⊂ Im(f + λid)s, because (f + λid)sker(f−αid)t is a bijective endomorphism (if g = (f + λid)s, W = ker(f −αid)t, then g(W ) ⊂ W since g commutes with (f −αid)t so g is well defined, but ker(f − αid)t ∩ ker(f + λid)s = 0 for λ + α (cid:54)= 0, which gives the injectivity and hence W = g(W ) ⊂ Im(g)). The nondegeneracy of n implies now that sλ coincides with s−λ for any eigenvalue λ, since the corresponding blocks are dual for n. The only possibilities are sλ ∈ {1, 2, 3}. Let s be the maximum of sλ for all λ eigenvalue. s = 3 Take u3 ∈ ker(f − λid)3 \ ker(f − λid)2. Take ui = (f − λid)3−i(u3) for i = 1, 2. Choose any v1 ∈ ker(f + λid)3 such that n(u1, v1) = 1. It is not difficult to check that v1 /∈ ker(f + λid)2. Take v2 = −(f + λid)v1 and v3 = −(f + λid)v2. Thus {u1, u2, u3, v1, v2, v3} is a basis of X⊥ relative to which the matrix of f is 0 λ 1 0 λ 1 0 λ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 −λ 0 0 −1 −λ 0 0 0 0 0 0 0 −1 −λ (cid:19) .  = (cid:18) J1 (cid:18) 0 0 0 −J t 1 (cid:19) I3 0   : I3 Besides the matrix of n relative to such basis is On one hand, n(u1, vi) = n(u3, (f + λid)2(vi)) = n(u3, 0) = 0 if i = 2, 3; also n(u2, v3) = n(u3, (f + λid)(v3)) = n(u3, 0) = 0, and analogously n(vi, uj) = 0 if j > i. On the other hand, n(u1, v1) = n((f−λid)(u2), v1) = n(u2,−(f +λid)(v1)) = n(u2, v2) and similarly it is equal to n(u3, v3). X ⊂ so(X⊥, n) of the Lie group H(cid:48) Now the Lie algebra h(cid:48) X in Eq. (4.10) in (cid:19) Remark 4.13 has dimension greater than 8. An element g ∈ h(cid:48) X satisfies g(X) = X, gX⊥ ∈ so(X⊥, n) and gf = f g. The matrix related to any element g ∈ so(X⊥, n) relative to the above basis should be a block-matrix with b and c skew- symmetric matrices of size 3 (dim so(6) = 15), so commuting with f is equivalent to b c −at (cid:18) a J1b + bJ t By making these computations, we get b = c = 0 and a ∈ (cid:104){I3, J1, J 2 means that dim h(cid:48) dimension greater than 8. That is, s = 3 is not a true possibility. (4.12) 1}(cid:105). This X = 3, a contradiction with the fact of containing a subalgebra of 1c + cJ1 = 0, aJ1 = J1a. 1 = 0, J t s = 2 With analogous arguments to the previous case, it is possible to find a basis of X⊥ relative to which the matrix of f is α 0 0 0 λ 1 0 λ 0 0 0 0 0 0 0 0 0 0 0 −α 0 0 0 0 0 0 0 0 0 0 0 −λ 0 0 −1 −λ  = (cid:18) J2 0 0 −J t 2 , (cid:19) (cid:18) 0 (cid:19) I3 0 (α and λ nonzero complex numbers) and the matrix of n is this case does not occur, since any element g ∈ h(cid:48) . Again X would have as a matrix relative I3 respectively,  λ1 0 0 0 0 0 0 λ2 0 0 0 0 0 0 0 0 λ3 0 0 −λ1 0 0 0 0 0 0 0 0 −λ2 0  , 0 0 0 0 0 −λ3 (cid:18) 0 I3 (cid:19) . I3 0 24 C. DRAPER (cid:19) (cid:18) a  a11 to such basis with b and c skew-symmetric 3 × 3 matrices satisfying the relations in (4.12) for J2. This forces b and c to be zero, and a commuting with J2 makes b c −at  if α = λ,  a11 0 0 a =  if α (cid:54)= λ. 0 0 a22 a23 0 a22 a = 0 a13 a21 a22 a23 0 a22 0 In any case, dim h(cid:48) X ≤ 5 cannot be greater than 8. s = 1 In the third case (a posteriori the only possible case), we can find {u1, u2, u3, v1, v2, v3} a basis of X⊥ relative to which the matrix of f and n are, Indeed, We can assume (changing if necessary ui with vi) that λ1 = λ2 = λ3. if λ1 /∈ {±λ2,±λ3}, any g ∈ h(cid:48) X commutes with f and leaves the f-eigenspaces In particular, g leaves (cid:104)u1, v1(cid:105) and (cid:104)u2, v2, u3, v3(cid:105) invariant (again its invariant. related matrix is a block-matrix) and, as n does not degenerate in those subspaces, gX⊥ ∈ so(2)⊕ so(4). As this space has dimension 1 + 6 = 7, we would again obtain a contradiction. Then the matrix of f 2 in this basis is λ2I6 (for λ = λ1), which means that X ∧ (X ∧ Y ) = f 2(Y ) = λ2Y for all Y ∈ X⊥. We would like to prove λ2 = −n(X), and, although it is not true, we will check that λ2/n(X) does not depend on the chosen nonisotropic X. On one hand, consider the basis B1 = {X, u1, u2, u3, v1, v2, v3} of V and let us compute nΩ,B1 (X, X). As Ω(X, ui, vi) = n(X ∧ ui, vi) = n(f (ui), vi) = n(λui, vi) = λ but Ω(X, ui, vj) = 0 = Ω(X, ui, uj) if i (cid:54)= j, then nΩ,B1 (X, X) = 2 · 2 · 3! · 3!Ω(X, u1, v1)Ω(X, u2, v2)Ω(X, u3, v3) = 144λ3. , u3−v3 √ i On the other hand, B2 = { X√n(X) 2 } is an orthonormal basis for n, so that its change of basis to the original basis B has (cid:32) determinant ±1 and nΩ,B2(X, X) = ±nΩ,B(X) = ±n(X). The last ingredient is (cid:33)(cid:33)3 , u1−v1 √ i , u2−v2 √ i , u3+v3√ , u2+v2√ , u1+v1√ 2 2 2 2 2 (cid:32) 1√ so 144λ3 = det CB2B1 nΩ,B2(X) = ±i3(cid:112) 1(cid:112) det CB1B2 = n(X) det 2 1√ 2 √ 1 i −1 2 √ i 2 n(X)n(X) and 122 . = ± 1 124 n(X)3 and λ2 = −12−4/3n(X) for a cubic root of the unit. Hence λ6 = − 1 X = 1 such that We have proved that for each nonisotropic X, there is X ∈ C, 3 X∧(X∧Y ) = −X 12−4/3n(X)Y for all Y ∈ X⊥. The map {X ∈ V n(X) (cid:54)= 0} → } ⊂ C which sends X (cid:55)→ X is continuous, and the first set is connected {1, and the second one is discrete, so that the map is constant and = X does not depend on X. i√n(X) −1±i 2 √ λ 3 = i3(cid:112) (cid:18) −1 n(X) , (cid:19)3 25 (cid:48) Y, Z) = If we replace n with n(cid:48) = sn, then the wedge product defined by n(cid:48)(X ∧ Ω(X, Y, Z) is ∧ s2124/3 n(X)Y = − s3124/3 n (cid:48) (cid:48) (X)Y = −n (X)Y, (cid:48) = 1 (cid:48) (X ∧ s∧, so Y ) = − (cid:48) X ∧ choosing s ∈ C such that s3124/3 = . For the rest of the proof, we consider this n(cid:48) as our new norm n. Therefore, for any nonisotropic X, it holds X ∧ (X ∧ Y ) = n(X, Y )X − n(X)Y (4.13) for all Y ∈ V , since it is true if Y ∈ X⊥, also if Y = X and this expression is linear in Y . It is not difficult to conclude that the identity (4.13) holds for any X ∈ V by using only linear algebra arguments, but an argument using continuity is faster here, since {X ∈ V n(X) (cid:54)= 0} is an open dense set. Now, we apply n(−, Y ) to Eq. (4.13) to get n(X, Y )2 − n(X)n(Y ) = n(n(X, Y )X − n(X)Y, Y ) = n(X ∧ (X ∧ Y ), Y ) = −n(X ∧ Y, X ∧ Y ), and hence we infer that ∧ is a cross product. Proof. (Of Theorem 4.8.) Take n given by Proposition 4.14 such that the related product ∧ : V × V → V is a cross product. Linearizing12 Eq. (4.13), we get X ∧ (Y ∧ Z) + Y ∧ (X ∧ Z) = n(X, Z)Y + n(Y, Z)X − 2n(X, Y )Z (4.14) (cid:3) NOTES ON G2 for any X, Y, Z ∈ V . ⊥ X Y ∧ Z = −n(Y, Z)X. = W1 ⊕ W−1, n(W1) = n(W−1) = 0 Take a nonisotropic vector X ∈ V such that n(X) = −1. We have checked that for the eigenspaces W±1 = {Y ∈ X⊥ X ∧ Y = ±Y }. Observe that, if Y ∈ W1, Z ∈ W−1, then (4.15) Indeed, by Eq. (4.14), X ∧ (Y ∧ Z) − Y ∧ Z = n(Y, Z)X. Changing the role of Y and Z in Eq. (4.14), we also get X ∧ (Z ∧ Y ) + Z ∧ Y = n(Z, Y )X (recall X ∧ Y = Y and X ∧ Z = −Z). Summing both identities, we have −2Y ∧ Z = 2n(Y, Z)X. In particular, it holds Ω(W1, W1, W−1) = n(W1 ∧ W−1, W1) ⊂ n(X, W1) = 0. If Ω(X⊥, X⊥, X⊥) = 0, take Y ∈ X⊥ nonisotropic, which will satisfy LY (X⊥) orthog- onal to X⊥, for LY = Y ∧−. But LY : Y ⊥ → Y ⊥ is bijective (recalling the previous proof), so LY (Y ⊥ ∩ X⊥) is a 5-dimensional vector subspace of (X⊥)⊥; a contra- diction. Thus Ω(X⊥, X⊥, X⊥) (cid:54)= 0, what implies that either Ω(W1, W1, W1) (cid:54)= 0 or Ω(W−1, W−1, W−1) (cid:54)= 0. But one of the assertions implies the other one. For in- stance, take Y1, Y2, Y3 ∈ W1 such that t = Ω(Y1, Y2, Y3) (cid:54)= 0. Take Zi = sYi+1∧Yi+2 (indices modulo 3) for some 0 (cid:54)= s ∈ C which we determine next. Thus, is orthogonal to X since n(X, Yi+1 ∧ Yi+2) = n(X ∧ Yi+1, Yi+2) ⊂ i) Zi n(W1, W1) = 0; (4.14)&i) ii) Zi ∈ W−1 since X∧(Yi+1∧Yi+2) iii) n(Yi, Zi) = sΩ(Yi, Yi+1, Yi+2) = st; iv) If i (cid:54)= j, n(Yi, Zj) = sΩ(Yi, Yj+1, Yj+2) = 0 (some repeated index). v) As Z3∧ (Y2∧ Y3) = −Y2∧ (Z3∧ Y3) + n(Z3, Y3)Y2 + 0 = −Y2∧ n(Y3, Z3)X + = −Yi+1∧(X∧Yi+2) = −Yi+1∧Yi+2; In particular, Z1, Z2, Z3 is a basis of W−1. Besides, n(Y3, Z3)Y2 = 2n(Y3, Z3)Y2, then Ω(Z1, Z2, Z3) = sΩ(Y2 ∧ Y3, Z2, Z3) = sn(Z3 ∧ (Y2 ∧ Y3), Z2) = s2n(Y3, Z3)n(Y2, Z2) = 2s3t2. 12 That is, applying three times Eq. (4.13) to (X +Y )∧((X +Y )∧Z)−X∧(X∧Z)−Y ∧(Y ∧Z). 26 C. DRAPER Thus, also Ω(W−1, W−1, W−1) (cid:54)= 0. For t = −4 and s = 1 2, we get (2s3t2 = 4) Ω(X, Yi, Zi) = −2, Ω(Y1, Y2, Y3) = −4, Ω(Z1, Z2, Z3) = 4. n(X, X) = −1, n(Yi, Zi) = −2, (4.16) Hence, if we take ϕ : V → V the isomorphism given by ϕ(E0) = X, ϕ(Ei) = Yi and ϕ(Fi) = Zi, it is clear from Remark 4.1 that ϕ∗Ω0 = Ω. Thus any generic 3-form (cid:3) belongs to the orbit of Ω0. Proof. (Of Theorem 4.9.) Let Ω ∈ ∧3V ∗ be a generic 3-form on V = R7. Fix B any basis of V and n ≡ nΩ,B the nondegenerate bilinear symmetric form in Lemma 4.10 and take ∧ : V × V → V the multiplication defined by the nondegeneracy of n as n(X ∧ Y, Z) = Ω(X, Y, Z). ∈ ∧3(C7)∗ is again a generic 3-form, due to the fact that If we complexify, Ω C = dimR GΩ = 14. Thus Theorem 4.8 says that there is α ∈ C such that dimC GΩ C X ∧ (X ∧ Y ) = α(n(X, Y )X − n(X)Y ) for all X, Y ∈ V . But X ∧ (X ∧ Y ) and n(X, Y )X − n(X)Y both belong to V , so that α ∈ R. If we take s ∈ R such that s3 = α and we replace n with sn and s∧, we obtain that the new ∧ is a cross product on V = R7 consequently ∧ with 1 relative to the new n. Case n is not positive definite Take X ∈ V such that n(X) = −1. Thus X ∧ → X⊥ given by (X ∧ Y ) = Y for all Y ∈ X⊥ and the endomorphism f : X⊥ f (Y ) = X ∧ Y satisfies f 2 = id. Its minimum polynomial is x2 − 1, so that it is diagonalizable with eigenvalues ±1, that is, X⊥ = W1 ⊕ W−1 for the eigenspaces W±1 = {Y ∈ X⊥ X ∧ Y = ±Y }. Besides n(W1) = n(W−1) = 0 (f skew-adjoint for n and the eigenspaces are totally isotropic). Now the proof continues exactly in the same way than the previous one until proving that, after a change of basis, Ω = Ω0. (In particular, this proves that there is a basis relative to which n is given by (3.1) and the signature is just (4, 3).) Case n is positive definite Now let us take X ∈ V such that n(X) = 1 and, as before, the endomorphism f : X⊥ → X⊥ given by f (Y ) = X ∧ Y , which satisfies f 2 = −id and it is skew-adjoint (in particular without any real eigenvalue). We choose • X1 ∈ X⊥ such that n(X1) = 1, Y1 = f (X1); • X2 ∈ (cid:104)X, X1, Y1(cid:105) • X3 = X1 ∧ X2, Y3 = f (X3). ⊥ such that n(X2) = 1, Y2 = f (X2); Note that n(Y1, Y1) = n(f (X1), f (X1)) = −n(X1, f 2(X1)) = n(X1) and n(X1, Y1) = n(X1, X ∧ X1) = 0. Similarly n(cid:104)Xs,Ys(cid:105) ≡ I2 for s = 2, 3 and also Y2 belongs to ⊥, since n(Y2, X1) = −n(X2, Y1) and n(Y2, Y1) = n(X2, X1). We are (cid:104)X, X1, Y1(cid:105) ⊥. Indeed, n(X3, Z) = Ω(X1, X2, Z) = 0 checking now that X3 ∈ (cid:104)X, X1, Y1, X2, Y2(cid:105) for Z = X1, X2 and also for Z = X, since Ω(X, X1, X2) = n(Y1, X2) = 0. Besides X ∧ X3 = −X1 ∧ (X ∧ X2) + 0 = −X1 ∧ Y2 is perpendicular to X1, so that n(X3, Y1) = n(X3, X ∧ X1) = −n(X ∧ X3, X1) = 0. Analogously, = X2 ∧ (X ∧ X1) is orthogonal to X2 so that X ∧ X3 = −X ∧ (X2 ∧ X1) n(X3, Y2) = −n(X ∧ X3, X2) = 0. (4.14) (4.14) NOTES ON G2 27 As n(X3) = n(X1)n(X2) = 1, this implies that B is an orthonormal basis for nX⊥ such that the matrix of f is X⊥ = {X1, Y1, X2, Y2, X3, Y3}  0 −1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 −1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 −1 0 1  . We are going to prove that Ω = X∗ + X∗ 2 + X∗ 2 ∧ Y ∗ With that purpose, let us check first that 1 + X∗ 3 − X∗ ∧ (X∗ 1 ∧ X∗ 1 ∧ Y ∗ 2 ∧ X∗ 2 ∧ Y ∗ 1 ∧ Y ∗ 3 ∧ Y ∗ 3 − Y ∗ 3 ) 1 ∧ X∗ 2 ∧ Y ∗ 3 − Y ∗ 1 ∧ Y ∗ 2 ∧ X∗ 3 . (4.17) (4.18) Xs ∧ Ys = X, Xs ∧ Yk = Ys ∧ Xk if s (cid:54)= k, Xs ∧ Xk = −Ys ∧ Yk for all s, k. For instance, we can use the proof of Theorem 4.8 by complexifying, since n(iX) = −1, W1 = (cid:104)Xs + iYs s = 1, 2, 3(cid:105) and W−1 = (cid:104)Xs − iYs s = 1, 2, 3(cid:105). According to Eq. (4.15), (Xs + iYs)∧ (Xk − iYk) = −n(Xs + iYs, Xk − iYk)iX. The real part gives Xs ∧ Xk + Ys ∧ Yk = (n(Ys, Xk) + n(Xs, Yk))X = 0 and the imaginary part gives Ys ∧ Xk − Xs ∧ Yk = −(n(Xs, Xk) + n(Ys, Yk))X = −2δskX, and hence Eq. (4.18) holds. If Zi, Si, Ti ∈ {Xi, Yi}, we would like to know Ω(Zi, Sj, Tk) for all i, j, k, but it is always 0 except if (i, j, k) is a permutation of (1, 2, 3), because Ω(Xi, Xi,−) = 0 (evaluating in X⊥) and Ω(Xi, Yi,−) = n(X,−) = 0. Then we can assume i = 1, j = 2, k = 3. Thus, by using repeatedly Eq. (4.18), ◦ Ω(X1, X2, X3) = n(X3) = 1; ◦ Ω(X1, Y2, Y3) = n(X1, Y2 ∧ Y3) = −n(X1, X2 ∧ X3) = −1; ◦ Ω(X1, X2, Y3) = n(X3, Y3) = 0; ◦ Ω(X1, Y2, X3) = n(X1, Y2 ∧ X3) = n(X1, X2 ∧ Y3) = 0; ◦ Ω(Y1, X2, X3)=Ω(X1, Y2, X3) = 0; ◦ Ω(Y1, X2, Y3) = Ω(X1, Y2, Y3) = −1; ◦ Ω(Y1, Y2, X3) = −Ω(X1, X2, X3) = −1; ◦ Ω(Y1, Y2, Y3) = −Ω(X1, X2, Y3) = 0; that proves Eq. (4.17). Hence we have shown that, if we consider the basis of V i=1 the dual basis in V ∗ i=1 = {X1, X2, X3, Y1, Y2, Y3, X} and {e∗ given by {ei}7 i }7 (e∗ i (ej) = δij), then our 3-form Ω is just Ω1 in Eq. (4.8). This would complete the proof that there are at most two orbits, related respec- tively with a norm of signature (4, 3) and (0, 7). For completing that the number of orbits is just two, we have only to be sure that Ω1 as in Eq. (4.8) is generic. It C 1 is generic. But, by repeating the proof suffices to see that the complexification Ω C is a of Theorem 4.8, we conclude that the wedge-product defined by Ω C C and then Ω 1 is in the orbit of Ω0, in cross product for a suitable multiple of n C 1 is generic and Ω1 is too (14 = dimC GΩ = dimR GΩ1). Although other words, Ω C 1 the hypothesis of Theorem 4.8 was Ω being generic, the only point we really needed throughout the proof was that the related bilinear form was nondegenerate to apply (cid:3) then Proposition 4.14. C 1 and n Remark 4.15. Following the arguments in the last paragraph, the converse of Lemma 4.10b) is also true and the nondegeneracy of nB,Ω implies that Ω is generic. 28 C. DRAPER Besides, as a consequence of Theorem 4.8, in the complex case OΩ0 = {ω ∈ ∧3V ∗ : det ϕω (cid:54)= 0}, so that the complementary is the set of zeros of a polynomial (closed in the Zarisky topology) and OΩ0 is also open (hence dense) in the Zarisky topology (topology coarser than the usual one). Remark 4.16. We add just one brief comment on representation theory related again to 3-forms and quadratic forms. For Ω0 our generic 3-form on V = F7 with Bc = {bi : i = 1, . . . , 7} = {E0, E1, E2, E3, F1, F2, F3} our basis of V , the map given by ∗ ∗ ∧3V → S2V , Ω (cid:55)→ nΩ : V × V → F being (cid:88) σ∈S7 nΩ(X, Y ) = ⊕ V , as does occur. ≡ V (2ω1) ⊕ V (ω1) ⊕ V (0) ≡ S2V ∗ (−1)sgσΩ0(X, bσ(1), bσ(2))Ω0(Y, bσ(3), bσ(4))Ω(bσ(5), bσ(6), bσ(7)), is G2-invariant and g2-invariant, where by G2 we again mean G0 Ω0 = GΩ0 ∩SL(V ) = {f ∈ SL7(F) : f · Ω0 = Ω0}. Written using the notation in Section 2.2, this means that the g2-module S2V ∗ ≡ V (2ω1)⊕V (0) should be a submodule of the g2-module ∧3V ∗ 4.3. Brief comments on G2-structures. A seven-dimensional smooth manifold M is said to admit a G2-structure if there is a reduction of the structure group of its frame bundle from GL7(R) to the group G2, viewed naturally as a subgroup of SO(7). This is equivalent to fix a 3-form ω on M such that ωp ∈ ∧3(TpM )∗ is generic for all p ∈ M. So each tangent space TpM can be identified with the imaginary octonions (zero trace octonions) because ωp endows TpM with a cross product as in Proposition 4.14. Also a scalar product and an orientation in TpM are determined as a consequence of Lemmas 4.10 and 4.12 respectively. Note that the signature of the scalar product does not depend on the choice of a basis in TpM, and moreover, for connected manifolds, it does not depend on the point p ∈ M: under any identification of TpM with R7, either ωp is in the orbit of Ω1 for all p or in the orbit of Ω0 for all p ∈ M, and consequently TpM can be identified with the imaginary part of either the division octonion algebra or the split octonion algebra respectively. Now all this goes from the vector spaces to the manifold, what explains in part why the two real forms of G2 are of such eminent importance in differential geometry: any manifold with a reduction to one of the real forms of G2 carries a Riemannian metric or respectively a semi-Riemannian metric of signature (3, 4), jointly with an orientation. An introduction to modern G2-geometry and its relevance in superstring theory can be found in [A08, prA]. An accessible paper about G2-manifolds (that is, ∇gω = 0) is [Ka11]. 5. G2 and the spheres in dimension 6 The compact (resp. split) group of type G2 acts transitively on the six-dimensional sphere (resp. on certain pseudohyperbolic space). It is easy to understand this re- lationship by going on with the arguments in the previous section. 5.1. Case the norm is positive definite. First, consider Ω1 ∈ ∧3(R7)∗ in Eq. (4.8), the corresponding positive definite norm n and the related cross product × in V = R7. If we take, as in Section 3.2, O = R ⊕ V with the product where 1 ∈ R is a unit and for all X, Y ∈ V , and with the positive definite norm n : O → R defined by n(1) = 1, n(1, V ) = 0 and nV = n, we obtain that O is an octonion algebra, with a proof XY := −n(X, Y )1 + X × Y, NOTES ON G2 29 e5 e7 e2 e1 e6 e3 e4 Figure 2. Fano plane. completely similar to that one of Lemma 3.6, since × is again a cross product. This algebra O is a division algebra (C is not), since, for α ∈ R and X ∈ V , (α + X)(α − X) = (α2 + n(X))1 is nonzero if α + X (cid:54)= 0. Sometimes O is called the (division) octonion algebra and C the split octonion algebra (often denoted by Os in the real case, and by OC s ∼= OC in the complex one). The multiplication in O = R(cid:104)1, ei : 1 ≤ i ≤ 7(cid:105) can be recalled by noting that i = −1 and that the only nonzero products of basic elements (up to the unit) are e2 e1e4 = e2e5 = e3e6 = e7; e1e2 = e3; e1e6 = e5; e2e4 = e6; e3e5 = e4; taking into account that if eiej = ek, also ejek = ei and ekei = ej, and eiej = −ejei if i (cid:54)= j. Another easy way of recalling these products is Figure 2, Fano plane: in any of the seven lines (the circle is also a line) the basic elements multiply according to the arrows. Thus, each triplet {ei, ej, ek} in one of these lines spans a subalgebra isomorphic to R3 with the usual cross product. (And of course R(cid:104)1, ei, ej, ek(cid:105) is isomorphic to the -division- quaternion algebra H when ei, ej and ek are the three different elements in one line.) Then, gc := {f ∈ gl7(R) : Ω1(f (X), Y, Z) + Ω1(X, f (Y ), Z) + Ω1(X, Y, f (Z)) = 0} = {f ∈ gl7(R) : f (X × Y ) = f (X) × Y + X × f (Y ) ∀X, Y ∈ V } ∼= Der(O) = {d ∈ gl(O) ≡ gl8(R) : d(XY ) = d(X)Y + Xd(Y )} is a Lie algebra of type G2 (a real form). Corollary 5.1. The group GΩ1 : = {f ∈ GL7(R) : Ω1(f (X), f (Y ), f (Z)) = Ω1(X, Y, Z) ∀X, Y, Z ∈ V }, = {f ∈ GL7(R) : f (X) × f (Y ) = f (X × Y ) ∀X, Y ∈ V }, is a connected Lie group of type G2, subgroup of SO(7). It is isomorphic -under the obvious extension to O making f (1) = 1- to Aut(O) = {f ∈ GL(O) : f (XY ) = f (X)f (Y ) ∀X, Y ∈ O}. We will prove this corollary in Section 5.5. Recall, as in Lemma 4.12c), that the group GΩ1 ⊂ SL(V ) acts naturally in V , and the orbit of an element X of norm 1 is contained in {Y ∈ R7 : n(Y ) = 1} ≡ S6. Moreover, that orbit GΩ1 · X equals S6, that is, the action of GΩ1 on the 6-dimensional sphere is transitive. This is consequence of a more general fact that can be deduced from the proof of Theorem 4.9: if we have {X, X1, X2} and {X(cid:48), X(cid:48) 2} two pairs of three orthonormal vectors (hence linearly independent) such that Ω1(X, X1, X2) = 0 = Ω1(X(cid:48), X(cid:48) 2), then there is g ∈ GΩ1 such that 1, X(cid:48) 1, X(cid:48) 30 C. DRAPER Figure 3. Complex structure of S2 given by J(v) = x × v. g(X) = X(cid:48), g(X1) = X(cid:48) of basis from the basis of V given by 1 and g(X2) = X(cid:48) 2. To be precise, g ∈ GL(V ) is the change to × X (cid:48) 2, X × X (5.1) , X {X (cid:48) (cid:48) 2, X (cid:48) (cid:48) 1, X (cid:48) 1, X (cid:48) 2, X (cid:48) (cid:48) 1 × X {X, X1, X2, X × X1, X × X2, X1 × X2, X × (X1 × X2)} (cid:48) (cid:48) 2)}, (cid:48) × (X 1 × X which preserves Ω1. Now we compute the isotropy group of X. Proposition 5.2. Consider X ∈ R7 of norm 1 and S6 = {Y ∈ R7 : n(Y ) = 1}. Recall the usual identification TXS6 = X⊥. The linear map (5.2) J(Y ) = X × Y, satisfies J 2 = −id, so that it is a complex structure, which allows to consider the 6-dimensional real vector space W = TXS6 as a 3-dimensional complex space under C × W → W , iY := J(Y ). Besides J : TXS6 → TXS6, σ : W × W → C, is a nondegenerate Hermitian form. If HX = {g ∈ GΩ1 : g(X) = X} denotes the isotropy group of X, the map (5.3) σ(Y, Z) = n(Y, Z) − in(J(Y ), Z), g (cid:55)→ gW is well defined and a Lie group isomorphism. Therefore, HX → SU(W, σ), GΩ1 /SU(3) ∼= S6. The notation J in Eq. (5.2) reminds the usual complex structure in Figure 3. Proof. The map σ is R-linear as J is and n is bilinear. Besides σ is C-linear in the first variable, σ(iY1, Y2) = σ(J(Y1), Y2) = n(J(Y1), Y2) − in(J 2(Y1), Y2) = iσ(Y1, Y2), since J is an skew-adjoint linear map such that J 2 = −id. (Recall that J is our map f in the proof of Theorem 4.9.) Also, σ(Y1, Y2) = n(Y1, Y2) + in(J(Y1), Y2) = n(Y1, Y2) − in(J(Y2), Y1) = σ(Y2, Y1). And σ(Y, Y ) = n(Y ) is a real positive number, different from 0 if Y (cid:54)= 0. Besides, the nondegeneracy of n implies the one of σ. Now, if g ∈ HX, then g : V → V is determined by its restriction gW , which proves the injectivity of the map in Eq. (5.3). The facts n(g(Y1), g(Y2)) = n(Y1, Y2) and gJ = Jg lead to σ(g(Y1), g(Y2)) = σ(Y1, Y2), so that gW ∈ U(W, σ). In order to prove that det gW = 1, recall that an element in GL3(F) (for F either C or R) preserving an alternating nonzero 3-form in ∧3(F3)∗ has determinant 1. In our case, we can choose {X1, X2, X3, Y1, Y2, Y3} an R-basis of X⊥ such that X3 = X1 × X2 and Yi = X × Xi = f (Xi). Then B0 = {X1, X2, X3} is a C- basis of W and B(cid:48) 0 = {X1 + iY1, X2 + iY2, X3 + iY3} is a C-basis of ker(F − id) for 6Beispielenicht-integrablerGeometrienBspl1 -- FastHermitescheMfkten•(S6,gcan):S6⊂R7hatfastkomplexeStrukturJ(J2=−id),dievom"Kreuz-produkt"aufR7induziertwird.•Jistnichtintegrabel,∇gJ6=0•Problem(Hopf):BesitztS6eine(in-tegrable)komplexeStruktur?xTxS2vS2JisteinBeispieleinersog."nearlyKahler"-Struktur:∇gXJ(X)=0.Allgemeiner:(M2n,g,J)fastHermitescheMfkt:JfastkomplexeStruktur,gkompatibleRiemann'scheMetrik.DannistdieStrukturgruppeG⊂U(n)⊂SO(2n),aberHol0(∇g)=SO(2n)!Bsple:Twistorraume(CP3,F1,2)mitihrernK-Struktur,SL(2,C)R,kompaktekomplexeMfktmitb1(M)ungerade(6∃Kahler-Metrik)...NOTES ON G2 31 C → V C. As Ω C given by F (Y ) = iX×Y . Of course g F : V it commutes with F = if the matrix of gW relative to B0 coincides with the matrix of g to B(cid:48) C preserves this eigenspace since C 1 is a nonzero alternating form on ker(F − id), and ker(F−id) relative 0, then det gW = 1.13 So (5.3) provides a monomorphism of groups and hence dim HX ≤ dim SU(3) = 8. According to Remark 4.13, dim HX ≥ 8, what implies dim HX = 8 and the map (5.3) is also surjective. The transitivity of the action gives then GΩ1 /SU(W, σ) ∼= (cid:3) GΩ1 · X = S6. C Recall that topological properties of H, G and G/H are closely related in homo- geneous spaces. Thus, as a corollary of the above proposition, GΩ1 is compact and connected, due to [Ne, Lemma 11.18], and simply connected [Ne, Proposition 11.17], because both SU(3) and S6 are compact, connected and simply connected. Hence we will use the notation G2,−14 for the connected compact Lie group GΩ1 and G2,2 for the (real) split group GΩ0 (also connected as a consequence of Section 5.4), notation provided by the signature of the Killing form of the related Lie algebra. (When there is no ambiguity, we will use the notation G2 for G+ the Ω0 complex connected group of type G2, and g2 for its Lie algebra LC.) Similarly, gc deserves the name g2,−14, as we were using for LR the name g2,2. Remark 5.3. In fact, S6 is a reductive homogeneous space, that is, if h = {d ∈ gc : d(X) = 0} is the Lie algebra of the isotropy group HX, which is a compact Lie algebra isomorphic to the skew-Hermitian matrices su(W, σ) ∼= su(3), then there is a complementary subspace m such that gc = h ⊕ m, Indeed, we can take m := {ϕY : Y ∈ X⊥ [h, m] ⊂ m. } ⊂ gl(V ), for ϕY (Z) := X × (Y × Z) − Y × (X × Z) + (X × Y ) × Z, (5.4) and then check that ϕY ∈ Der(V,×) = gc (see [E92, Eq. (2.11)]). The fact ϕY (X) = 2Y implies that m is isomorphic to the vector space X⊥ = W and that m ∩ h = 0. By dimension count, gc = h ⊕ m. Moreover, this complementary invariant subspace m is unique (in this example), coinciding with the subspace orthogonal to h relative to the Killing form of gc, according to [E92, Proposition 2.3]. In general, the tangent space to the orbit, S6 in this case, is identified with gc/h, but due to the fact the homogeneous space is reductive, the identification becomes TXS6 ∼= gc/h ∼= m ∼= X⊥, which besides coincides with the usual identification TXS6 ∼= X⊥ inside R7. This allows to pass algebraical properties of (m, [ , ]m) to geometrical ones, where [ , ]m denotes the projection on m of the Lie bracket. As [ϕY1 , ϕY2]m = ϕprW (Y1Y2−Y2Y1), the (new nonassociative) algebra (m, [ , ]m) is isomorphic to (W,∗), for Y1 ∗ Y2 = prW (Y1Y2 − Y2Y1), which turns out to be the compact vector color algebra whose complexification is the vector color algebra originally introduced by Domokos and Kövesi-Domokos in the study of color symmetries arising from the Gell-Mann quark model (references in [E92]). Next we try to express the commutator, projected down to m, without passing through the identification with W . If the chosen identification is W → m ≤ gc, 13We have dealt with the complexification because Ω1 : W × W × W → R ⊂ C is not C-linear, and if we consider the restriction to Ω1 : W (cid:48) × W (cid:48) × W (cid:48) → R for W (cid:48) the real vector space spanned by {X1, X2, X3}, then Ω1 is an alternating nonzero 3-form, but g(W (cid:48)) (cid:54)⊂ W (cid:48). 32 C. DRAPER Y (cid:55)→ ϕJ(Y ), then the following identity holds (5.5) where recall that we think of J : O0 → O0, J(Y ) = X × Y , so that J 2 is the opposite of the projection on W = X⊥. Indeed, some easy computations show that [J(Y1), J(Y2)] = −X[Y1, Y2]X = −2X(Y1 × Y2)X, so that [Y1, Y2]m = 2J(Y1 × Y2), [J(Y1), J(Y2)]W = −2(X(Y1 × Y2)X)W = −2X(Y1 × Y2)W X = 2X 2(Y1 × Y2)W = −2(Y1 × Y2)W = 2J 2(Y1 × Y2), since the elements in W anticommute with X. Then, [ϕJ(Y1), ϕJ(Y2)]m = ϕ[J(Y1),J(Y2)]W = ϕJ(2J(Y1×Y2)), which obviously means (5.5) through the identification. More geometrically, S6 is a naturally reductive G2,−14-homogeneous space, that is, the canonical connection is compatible with the metric, (cid:104)[ϕY1 , ϕY2]m, ϕY3(cid:105) + (cid:104)ϕY2, [ϕY1, ϕY3 ]m(cid:105) = 0 n(J(Y1 × Y2), Y3) = n(X, (Y1 × Y2) × Y3), for all Y1, Y2, Y3 ∈ W . This is a direct consequence of the fact that (cid:104) , (cid:105) : m × m → R, the restriction of the Killing form to m, corresponds through the above identification14 between m and W with the restriction of the octonionic norm to W , up to scalar multiple. Besides, n(J(Y ), Y3) = n(−Y × X, Y3) = n(X, Y × Y3) gives, in particular, that (5.6) which provides useful information on the torsion 3-form of the canonical connection [ABoF17, Proposition 4.2]. 5.2. A model of compact G2. The reductive decomposition of gc in the above remark suggests the following linear model of the compact Lie algebra of type G2 and of its natural module. Proposition 5.4. Take W = C3 and σ : W ×W → C the Hermitian nondegenerate i=1 ui¯vi. Recall that the unitary algebra u(W ) is spanned by the maps σu,v, where σu,v(w) := σ(w, u)v − σ(w, v)u for any u, v, w ∈ W . Then the vector space form given by σ(u, v) =(cid:80)3 with anticommutative product given by L = su(W ) ⊕ W, (cid:72)(cid:35) su(W ) is subalgebra, (cid:72)(cid:35) [φ, u] = φ(u) for any φ ∈ su(W ) and u ∈ W , (cid:72)(cid:35) [u, v] = (3σu,v − tr(σu,v)idW ) + 2u × v for any u, v ∈ W , n(Y1, Y2). for × the usual vector product in W = C3, is a compact Lie algebra of type G2 isomorphic to gc. (Also tr(σu,v) = σ(v, u) − σ(u, v) = −2i Im σ(u, v).) Besides R7 ≡ R ⊕ W =: V is an irreducible L-module for the action · given by (cid:71)(cid:35) For any φ ∈ su(W ), and u ∈ W , (cid:71)(cid:35) For any u, v ∈ W , 14 With both the above identifications, Y (cid:55)→ ϕJ(Y ) and Y (cid:55)→ ϕY , since n(J(Y1), J(Y2)) = u · v = −2Im σ(u, v) − u × v. u · 1 = −2iu, φ · u = φ(u); φ · 1 = 0, NOTES ON G2 33 Finally, the 3-form on V given by Ω(s + u, t + v, r + w) = −Im(sσ(v, w) + tσ(w, u) + rσ(u, v)) + Re(det(u, v, w)) for any r, s, t ∈ R, u, v, w ∈ W , is generic, and L = Der(V, Ω) ⊂ so(V, n), for the positive definite norm The related cross product on V is given by n(s + u) = s2 + σ(u, u). (s + u) × (t + v) = −Im σ(u, v) +(cid:0)isv − itu + u × v(cid:1). Although this model is extracted from [BDE03], a difficulty when reading such paper is that the considered field k is not necessarily R, but a ground field of arbitrary characteristic, K is not necessarily C, but a two dimensional composition algebra over k (even in the case k = R, there are two possibilities for K: either C or R ⊕ R) and W in the paper is a free left K-module of rank 3 endowed with a Hermitian nondegenerate form σ : W ×W → K with trivial Hermitian discriminant. These choices produce a model valid even in the split case (considering R⊕R instead of C) but for avoiding the reader to dive in such general situation, we will sketch a proof adapted to our concrete setting. Proof. Recall that O = R ⊕ V for V = O0 = R(cid:104)ei : 1 ≤ i ≤ 7(cid:105). If we identify C ≡ R ⊕ Re7, then V = Re7 ⊕ W for W = R(cid:104)ei : 1 ≤ i ≤ 6(cid:105) = C(cid:104)e1, e2, e3(cid:105) and W is isomorphic to C3 in a natural way. This identification and the nondegenerate Hermitian form σ : W × W → C given by σ(u, v) = n(u, v) − n(e7 × u, v)e7 = −prR⊕Re7(uv) can be used to write the product in the octonion division algebra O = C ⊕ W as (5.7) for α, β ∈ C, u, v ∈ W , and the norm as n(α + u) = αα + σ(u, u). Now recall the decomposition Der(O) = {d ∈ Der(O) : d(e7) = 0} ⊕ {ϕu : u ∈ W}, where the derivations ϕu are defined in Eq. (5.4). If d(e7) = 0, then d(W ) ⊂ W and moreover σ(d(u), v) + σ(u, d(v)) = 0, so that we can consider the vector space isomorphism (α + u)(β + v) = (αβ − σ(u, v)) +(cid:0)αv + βu + u × v), Der(O) −→ su(W ) ⊕ W =: L d ϕu (cid:55)→ dW ∈ su(W ) (cid:55)→ iu. This induces on L the Lie algebra structure given in the statement of this propo- sition. Computations can be extracted from [BDE03, Theorem 5.3], where the correspondence with the notation in [BDE03] is ϕu = 1 2 De7,u (take into account De7,u(e7) = 4u = 2ϕu(e7)). The cross product is immediate from Eq. (5.7), since × = 1 The action of L on V (or on the octonions) is consequence of [BDE03, Lemma 5.1], passing through the above isomorphism, taking care of the fact that we have decided to omit e7 in the final expression. 2 [ , ] in V. The 3-form is simply obtained by taking Ω(s + u, t + v, r + w) = n((s + u) × (t + v), r + w), (cid:3) again with the caution of the e7-omission. 5.3. Parenthesis on real forms of g2. It is a well known fact of Lie theory that there are just two real forms of type G2 up to isomorphism, namely, g2,−14 and g2,2. We would like to arrive at this conclusion without using strong machinery of Lie theory but only our tools of 3-forms, although we will add some comments about more usual techniques too. First we come back to the complex setting to prove that the stabilizer of any 3-form which contains the group G2 cannot strictly contain G2. Take V = C7, Ω0 the generic 3-form in Theorem 4.8, g2 = Der(V, Ω0) and n = nΩ0,B : V × V → C 34 C. DRAPER 0 e3 the related nondegenerate form (B any fixed basis of V ). Through this subsection, C = C ⊕ V will denote the complex octonion algebra. Lemma 5.5. Up to scalar multiple, n is the unique g2-invariant symmetric bilinear map V × V → C. Proof. Giving a g2-invariant symmetric bilinear map q : V × V → C is equivalent to giving a homomorphism of g2-modules from V to V ∗, by means of q : V → V ∗, q(X) ∈ V ∗, q(X)(Y ) = q(X, Y ). Using the notations in Section 2, the module V is g2-irreducible generated by the vector X =  ∈ Vω1=−ε3 (multiplying  0 successively by L−αi, i = 1, 2, we immediately get all the weight spaces). So, a homomorphism of g2-modules from V to V ∗ is determined by providing the image of X. But the image of X must also have weight −ε3, so that it is determined up (cid:3) to scalar since dim(V ∗)ω1 = dim Vω1 = 1. (This lemma is equivalent to the fact mentioned in Remark 4.16 that S2V ∗ = Ω0 ⊂ GΩ is necessarily generic. V (2ω1) ⊕ V (0) contains only one copy of a trivial submodule type V (0).) Lemma 5.6. Any 0 (cid:54)= Ω ∈ ∧3V ∗ such that G+ Proof. Let us see that nΩ,B : V × V → C is nondegenerate, which is a sufficient condition to assure Ω generic according to Remark 4.15. Note that nΩ,B : V × V → C is GΩ ∩ SL(V )-invariant by the same argument as in Lemma 4.12b). Besides Ω0 ⊂ SL(V ), so that nΩ,B : V × V → C is G+ - Lemma 4.12b) tells us that G+ Ω0 invariant. By Lemma 5.5, nΩ,B is a nonzero scalar multiple of n = nΩ0,B. In (cid:3) particular, nΩ,B is nondegenerate (and generic by Remark 4.15). Lemma 5.7. If L1 is a real simple Lie algebra and W is an L1-irreducible (finite- dimensional) module, then one of the following possibilities happens: absolutely irreducible); a) EndL1 W ∼= R. In this case W b) EndL1 W ∼= C. In this case W c) EndL1 W ∼= H. In this case W C 1 -modules; L C C is an L 1 -irreducible module (i.e., W is C is sum of two nonisomorphic irreducible C C is sum of two isomorphic irreducible L 1 - modules. Besides dimR W is multiple of 4. Proof. On one hand, EndL1 W is a real division (associative) algebra, since for any 0 (cid:54)= f ∈ EndL1 W , the submodule ker(f ) has to be 0 and the submodule Im(f ) has to be W , so that f is invertible. Now, the classical Frobenius theorem (1877) characterizes the finite-dimensional associative division algebras over the real numbers: they are R, C and H and hence d = dimC EndL = dimR EndL1 W is either 1, 2 or 4, respectively. C 1 - modules, dimC HomL (U1, U2) is 1 if U1 and U2 are isomorphic and it is 0 otherwise. C Consequently, if U ∼= s1U1 ⊕ ··· ⊕ skUk is the decomposition of an arbitrary L 1 - module U as a sum of irreducible submodules with Ui not isomorphic to Uj if i (cid:54)= j, C. then dimC EndL C 1 On the other hand, Schur's lemma says that, for U1 and U2 irreducible L C 1 -module U = W i . Now, we apply this to the L U =(cid:80)k i=1 s2 C W C 1 C 1 a) If d = 1, then k = 1 = s1 and W b) If d = 2, then k = 2 and s1 = s2 = 1. Hence W C C is an L 1 -irreducible module. C is sum of two nonisomor- phic irreducible L C 1 -modules. c) If d = 4, then either k = 1 and s1 = 2 or k = 4 and si = 1 for all it suffices to C as a sum of irreducible i = 1, . . . , 4. We want to rule out the second possibility: realize that W C ∼= 2W is the decomposition of W NOTES ON G2 C ∼=(cid:80)4 i=1 Ui as L C 1 -module because L1-modules, so that it is not possible W in particular when considering as L1-module there would be at least 4 irreducible summands instead of just 2. Hence we have proved that U ∼= 2U1 is sum of two isomorphic irreducible L Finally, if ψ : EndL1 W → H is the isomorphism, then an action H × W → W is given by (q, w) (cid:55)→ ψ−1(q)(w), so W can be endowed with a quaternionic vector space structure, what in particular implies that dimR W is multiple of 4. C 1 -modules. 35 (cid:3) C 1 = V . C C 1 = Der(C). Then there Proposition 5.8. Let L1 be a real Lie algebra such that L exists V1 an L1-module such that V Proof. We can consider V = C7 as L1-module of (real) dimension 14. Now the complexified module V = V ⊗R C (of complex dimension 14) is not an irreducible C 1 -module but sum of two isomorphic modules. Indeed, since the arbitrary element L h = diag{0, s1, s2, s3,−s1,−s2,−s3} ∈ h acts on both C ∼= 2V (ω1) is the with the same weight −s3 = −ε3(h) = ω1(h), it follows that V C C as a sum of irreducible L decomposition of V 1 -submodules. This means that none of the situations a), b) and c) in Lemma 5.7 is satisfied (the third possibility since 14 is not multiple of 4), in other words, V is not L1-irreducible. Take V1 an irreducible C ∼= 2V (ω1), this ensures L1-submodule of V . As V (cid:3) that V ⊗ 1 and C 1 is a proper submodule of V ⊗ i  0  0 0 e3 0 e3 C 1 ∼= V (ω1). That is, V = R7 is a real representation of any real form of g2. Note that this means that not only g2,2 ⊂ gl7(R), but also g2,−14 is a subalgebra of gl7(R), as we have already emphasized in Proposition 5.4. Proposition 5.9. There are only two real forms of Der(C) up to isomorphism; namely, Der(R7, Ω0) ∼= g2,2 and Der(R7, Ω1) ∼= g2,−14. Proof. If a Lie algebra L1 is a real form of Der(C) ≡ Der(C7, Ω0), take V1 the L1- C 1 = C7. Consider the module of dimension 7 as in Proposition 5.8, i.e. such that V restriction of the C-trilinear alternating map Ω0 : C7 × C7 × C7 → C to Ω0V1 : V1 × V1 × V1 → C: thus Re(Ω0V1) and Im(Ω0V1) are both R-trilinear alternating maps, and at least one of them is nonzero, say Ω(cid:48) 1. As trivially they are L1-invariant, then L1 ⊂ Der(V1, Ω(cid:48) 1 is generic. Hence Ω(cid:48) 1 is generic (cid:3) too, so it belongs to the orbit of either Ω0 or Ω1, what finishes the proof. 1). Now Lemma 5.6 implies that Ω(cid:48)C Remark 5.10. In particular, Proposition 5.9 allows us to conclude that gc is compact without using properties of homogeneous spaces, because gc is not split since it contains a compact subalgebra isomorphic to su(3). Alternatively, we can argument without a deep detail why κ is negative definite. The computation of the Killing form is always difficult, but some clues are: h • h and m in Remark 5.3 are orthogonal for κ, since when we complexify, C ∼= L¯1 ⊕ L¯2 as in Remark 2.3, and the pieces of any group- C ∼= L¯0 and m grading (in this case a Z3-grading) satisfy κ(L¯i, L¯j) = 0 if ¯i+¯j (cid:54)= ¯0, because (adL¯iadL¯j)(L¯k) ∩ L¯k ⊂ L¯i+¯j+¯k ∩ L¯k = 0; • The restriction κh is a (positive) multiple of the Killing form of su(3); • There is α ∈ R a positive scalar such that κ(ϕY , ϕZ) = −αn(Y, Z). (The key is that both maps κ and n are h-invariant maps m × m → R, so rep- resentation theory says that one should be scalar multiple of the other. 36 C. DRAPER This allows to find α by computing κ in one concrete nonorthogonal couple (Y, Z).) Remark 5.11. The usual way of knowing that there are just two real forms of type G2 is sketched here. As mentioned in Section 1.1, the real forms (up to isomorphism) of a complex Lie algebra are in bijective correspondence with involutive (order one or two) auto- morphisms (up to conjugation) of the complex algebra. But in g2 there is only one conjugacy class of nontrivial order two automorphisms. Indeed, for C = C⊕ C7 the octonion algebra and g2 = Der(C), then is a group isomorphism, where Aut(C) = G2 −→ Aut(g2), f (cid:55)→ Ad(f ) −1. Ad(f ) : g2 → g2, Ad(f )(d) = f df (Take into consideration that Der(C) = g2 ∼= ad(g2) = Der(g2), since any derivation of a semisimple Lie algebra over F is inner. This fact is the linear version of the above.) But Ad(f )2 = idg2 means that f 2 commutes with Der(C), so that f 2 ∈ CidC, since the centralizer Centgl(C) Der(C) = CidC. As −idC is not an automorphism, then f 2 = idC. Note finally that in C there is a unique order two automorphism up to conjugation, which acts as the identity in a copy of C ⊕ C3 (quaternion algebra provided by the -essentially unique- cross product in C3) and as minus the identity in the orthogonal subspace. We follow notation in O'Neill's book [Ne, 4]. So Rn+1 5.4. Case the norm is not positive definite. Again, we are interested in the action GΩ0 × V → V , although this time the norm has signature (4, 3), so that the orbit GΩ0 · X for a nonisotropic vector X with norm n(X) = −1 will be contained in the hypercuadric {Y ∈ V : n(Y ) = −1}. (Recall that GΩ0 ⊂ SO(V, n) according to Lemma 4.12, since we are considering here the real case.) denotes Rn+1 with the : q(x) = 0} is a norm q given by diag{−Iν, In+1−ν}. The nullcone {x ∈ Rn+1 hypersurface diffeomorphic to (Rν \ 0) × Sn−ν; the pseudosphere of radius r > 0 is : q(x) = r2} and the pseudohyperbolic space of radius r > 0 is ν (r) = {x ∈ Rn+1 Sn ν (r) = {x ∈ Rn+1 H n 4 : n(Y ) = −1} ≡ H 6 Similarly to Proposition 5.2 we have, Proposition 5.12. Take X ∈ V such that n(X) = −1, and W = X⊥. Recall that W = W + ⊕ W −, for the (totally isotropic) eigenspaces W ± = {Y ∈ V : f (Y ) = ±Y }, since the map f : W → W given by f (Y ) = X × Y satisfies f 2 = id. Then, for HX = {g ∈ GΩ0 : g(X) = X} the isotropy group of X, the map (5.8) ν+1 : q(x) = −r2}. Thus, our hyperquadric is {Y ∈ R7 3 (1). ν ν ν is well defined and a Lie group isomorphism. Furthermore, HX → SL(W +), g (cid:55)→ gW + 3 (1). GΩ0 /SL3(R) ∼= H 6 (5.9) Proof. Of course we can consider the restriction of g ∈ HX to W + since g commutes with f. Now Ω0 : W + × W + × W + → R is a trilinear alternating nonzero map preserved by g, so that det gW + = 1, and the map in (5.8) is well defined. The injectivity is a consequence of the fact that gW + determines gW −, since W + is dual to W − and HX ⊂ SO(W, n). In particular, dim HX ≤ dim SL3(R) = 8. According to Remark 4.13, dim HX ≥ 8, what implies dim HX = 8 and the map (5.8) is also surjective. The last task is to prove that the orbit GΩ0 · X fills the whole {Y ∈ V : n(Y ) = 3 (1), we find Y1, Y2, Y3 ∈ ker(f − id) with 3 (1). But for any Y ∈ H 6 −1} ≡ H 6 NOTES ON G2 37 the proof of Theorem 4.8, implies that there is ϕ ∈ GΩ0 such that ϕ(cid:0) E1+F1 2 Yi+1 × Yi+2 such that Y1 + Z1 = 2Y . (This, like in Ω(Y1, Y2, Y3) = −4 and Zi = 1 3 (1).) In order to find the required basis, it so that the orbit GΩ0 · E1+F1 = H 6 ⊥ suffices to take Y1 = Y + f (Y ) ∈ ker(f − id) and Y2, Y3 ∈ ker(f − id) ∩ (cid:104)Y, f (Y )(cid:105) (adjusting the scalars to have Ω0(Y1, Y2, Y3) = −4). Then Z1 = 1 2 Y2 × Y3 is equal to Y − f (Y ) because Y − f (Y ) ∈ ker(f + id) satisfies n(Y + f (Y ), Y − f (Y )) = −2 and n(Y − f (Y ), Ys) = 0 for s = 2, 3 (properties which characterize the element (cid:3) Z1). (cid:1) = Y , 2 2 Again the connectedness of GΩ0 is an immediate consequence of this proposition. Also there is a complex version of this proposition, which will be commented in the next subsection. The homogeneous space described in (5.9) is again a reductive homogeneous space, with related reductive decomposition provided by Remark 2.3.15 This time the complementary subspace TX H 6 3 (1) is not sl3(R)-irreducible, but it breaks as a sum of the natural 3-dimensional module and its dual one. Note that H 6 3 (1) is diffeomorphic (and antiisometric) to S6 3 [Ne, Lemmas 24 and positive definite norm; norm with signature (4, 3). 25], so that in particular G2,2 acts transitively too on the pseudosphere S6 3. Remark 5.13. In fact, the same arguments can be used to obtain a stronger result, namely, our real groups of type G2 can be identified with certain Stiefel manifolds: i) G2,−14 ∼= {(X0, X1, X2) ∈ (R7)3 : n(Xi, Xj) = δij, X0 ⊥ X1 × X2}, for n a ii) G2,2 ∼= {(X0, X1, X2) ∈ (R7)3 : n(Xi, Xj) = −δij, X0 ⊥ X1 × X2}, for n a As the group preserves the product and the norm, it is clear that it also acts on the corresponding set of ordered triples. Let us see in each case why this is a transitive action with isotropy group equal to the identity. For i), the endomorphism of V as in Eq. (5.1) leads a triplet to another one. For ii), take X = X0, Y1 = X1 + X × X1 (that belongs to the eigenspace VX,1), Y2 = X2 + X × X2, Y3 ∈ VX,1 2 Yi+1 × Yi+2. Equation (4.16) says that such such that n(Y3, Y1 × Y2) = −4, Zi = 1 a set can be lead to any other set (basis, in fact) constructed in the same way. But Y1 + Z1 = 2X1 and Y2 + Z2 = 2X2 as at the end of the proof of Proposition 5.12. Observe that this remark provides a way of understanding the real groups of type G2 also related to octonions, described in [Ba02, 4.1]. For instance, for the compact case: (X0, X1, X2) is called a basic triple if Xi's are zero trace octonions whose square is −1, X1 anticommutes with X0 and X2 anticommutes with X0, X1 and X0X1 (shorter, three orthonormal vectors such that Ω1(X0, X1, X2) = 0). In this case (X1, X2, X0) is a basic triple too. Besides the subalgebras of O generated by {X0}, {X0, X1} and {X0, X1, X2} are respectively isomorphic to C, H and O. According to i) the group G2,−14 is in bijective correspondence with the set of basic triples, for the correspondence g ∈ G2,−14 ↔ (g(e1), g(e2), g(e7)). 5.5. Connectedness. We have postponed some points on the connectedness until now, namely, Corollaries 4.4 and 5.1, due to the fact that the handling of groups is more difficult than the one of algebras. First we will check the version for groups of Proposition 3.9, enclosing also the positive definite case. (In fact, Proposition 3.9 is a trivial consequence of Proposition 5.14, but not conversely.) Proposition 5.14. Let Ω be a generic 3-form on V = F7 (F either R or C), n the norm related to the corresponding cross product × and C = F ⊕ V the related 15 A reader interested in all the eight reductive homogeneous spaces of G2 (both split and compact), their invariant affine connections, holonomy algebras and so on can consult [BDE05]. 38 C. DRAPER octonion algebra. The map f ∈ Aut(V,×) (cid:55)→ f ∈ Aut(C), (cid:26) f (1) = 1, fV = f, is a Lie group isomorphism. Furthermore, Aut(V,×) ⊂ O(V, n) ∩ GΩ. Proof. Take f ∈ Aut(V,×) and linearly independent elements X, Y ∈ V . By (4.13), n(X, Y )f (X)−n(X)f (Y ) = f (X×(X×Y )) = n(f (X), f (Y ))f (X)−n(f (X))f (Y ), and consequently n(X, Y ) = n(f (X), f (Y )). As besides Ω(X, Y, Z) = n(X × Y, Z), clearly the above implies that also Aut(V,×) ⊂ O(V, n) ∩ GΩ. Now f (XY ) = f (−n(X, Y )1 + X × Y ) coincides with f (X) f (Y ) = f (X)f (Y ) = −n(f (X), f (Y ))1 + f (X) × f (Y ), so that f is an automorphism of the octonion algebra. Conversely take f ∈ Aut(C). As f (X) = f (1X) = f (1)f (X) for all X, then f (1) = 1 (f bijective). Let us check that f (V ) ⊂ V . We apply f in Eq. (3.3) to get −n(X)1 = f (−n(X)1) = f (X 2) = f (X)2 = 2n(f (X), 1)f (X) − n(f (X))1 when X ∈ V , but f (X) /∈ F1, hence n(f (X), 1) = 0 and f (X) ∈ V (Remark 3.8). We have that f (X × Y ) = f (XY + n(X, Y )1) = f (X)f (Y ) + n(X, Y )1 = (−n(f (X), f (Y )) + n(X, Y ))1 + f (X) × f (Y ) if X, Y ∈ V , so the projection on V gives f ∈ Aut(V,×). As we have commented throughout the paper, (cid:3) Lemma 5.15. Let Ω be a generic 3-form on V = R7. Then GΩ is connected. Proof. We can assume that Ω is either Ω0 or Ω1. According to [Wa, Proposi- tion 3.66], if H is a connected closed subgroup of a Lie group G and G/H is connected, then G is also connected. This immediately gives the result taking into (cid:3) consideration Propositions 5.2 and 5.12, respectively. This proves, in particular, Corollary 5.1 and the real part of Corollary 4.4. The complex case requires to be careful: By Lemma 4.12, GΩ is not connected, since det(ωidV ) = ω7 = ω (cid:54)= 1, so ωidV /∈ G+ Ω. Lemma 5.16. Let Ω be a generic 3-form on V = C7 and n a related nondegenerate bilinear form. Then Aut(V,×) = G+ Ω = GΩ ∩ SL(V ) = GΩ ∩ O(V, n) ∼= Aut(C). In particular, this finishes the complex part of Corollary 4.4. Proof. By Lemma 4.12, G+ Ω ⊂ GΩ ∩ SL(V ) = GΩ ∩ O(V, n). In order to prove that both groups are equal, it suffices to prove that GΩ∩SL(V ) is connected because the Lie algebras of both groups are the same (Der(V, Ω) ⊂ so(V, n) because Ω is in the orbit of Ω0, Lemma 3.1 and Proposition 4.2). Consider the action of this group on the connected manifold {X ∈ C7 : n(X) = −1} (since the group preserves n.) The isotropy subgroup of X is isomorphic to SL3(C) because of the same arguments as in the proof of Proposition 5.12. But the action is transitive. Indeed, if X has norm −1, there is ϕ : V → V given by ϕ(E0) = X, ϕ(Ei) = Yi and ϕ(Fi) = Zi (Yi's and Zi's have been chosen as in (4.16)), that belongs obviously to GΩ, but besides preserves n (again by (4.16)) so that ϕ ∈ GΩ ∩ O(V, n) and thus the action is transitive. Again [Wa, Proposition 3.66] gives that GΩ ∩ SL(V ) is connected. The rest of the proof is then a trivial consequence of Proposition 5.14 and the (cid:3) fact of having the same Lie algebras: Propositions 3.4 and 4.2. NOTES ON G2 39 In consequence the automorphism group of an octonion algebra is always a con- nected Lie group, both in the two real cases (corresponding to division and split octonion algebras, respectively) and also in the complex one. Instead, not all of them are simply connected. Proposition 5.17. G2,2 is not simply connected, its fundamental group is Z2. Proof. For a coset manifold M = G/H, there is an exact sequence of groups and homomorphisms [Ne, Proposition 11.17], 0 → π2(M ) → π1(H) → π1(G) → π1(M ) → π0(H) → π0(G). (5.10) We apply this sequence to G2,2/SL3(R) ∼= S6 3 is diffeomorphic (not isometric) to S3×R3, then it has trivial fundamental group as well as trivial second homotopy group, as the 3-dimensional sphere has. So π1(H) ∼= π1(G), and the result follows (cid:3) since π1(SL3(R)) = π1(SO(3)) ∼= Z2. 3. As S6 Of course Eq. (5.10) gives that G2,−14 is simply connected. In fact, the complex group G2 is also simply connected, but in order to apply the exact sequence of groups to the coset manifold G2/SL3(C) ∼= {X ∈ C7 : n(X) = 1}, one would have to understand well how is the manifold {X ∈ C7 : n(X) = 1}. Instead, we can conclude the simply connectedness in the complex case from the fact that the determinant of the Cartan matrix in (1.2) is 1, by using nonelementary facts on structure theory of Lie groups. (The determinant equal to 1 implies that the weight related with the distance between them.) Remark 5.18. The previous proposition implies that there are just three connected lattice (cid:80) Zωi and the root lattice (cid:80) Zαi coincide, but the fundamental group is real Lie groups of type G2, namely, G2,−14, G2,2 and its double covering (cid:101)G2,2. One can wish an explicit description of this simply connected Lie group (cid:101)G2,2, but note §2] exposes the basic structure theory of (cid:101)G2,2.) that this group is not a matrix Lie group: any of its linear representations is not faithful, so it cannot be seen as a closed subgroup of GLm(R) for any m.16 ([Vo94, However, there are many Lie groups of type G2 if we remove the hypothesis of connectedness. For instance [Ha, Ba02], take the group {f ∈ GL(V ) : f (x, y, z) = (f (x), f (y), f (z)) ∀x, y, z ∈ V } = {±f : f ∈ G2,2} where the associator in V = R7 is defined by (x, y, z) := (xy)z − x(yz), which is nonzero because the split octonion algebra C = R ⊕ V given by Eq. (3.2) is not associative. The above group is isomorphic to G2,2 × Z2, of course nonconnected. This example works similarly for the division octonion algebra. 6. G2 and spinors Spinors were invented by Dirac in creating his relativistic quantum theory of the electron, but today they are relevant also in quantum theory, relativity, nuclear physics, atomic and molecular physics, and condensed matter physics. Mathemati- cian readers interested in seeing how spinors relates to modern theoretic physics can consult the introduction of the Srni lecture notes [A06, Sections 1.3,1.4]. In this section we will prove that, when the group Spin(7) acts on the eight- dimensional spin representation, the subgroup stabilizing any unit spinor is again C and extend it by complexification to a representation of g2 = g 2,2. As the complex group G2 is simply connected, we may exponentiate the previous representation to one of G2. Then we restrict 16If (cid:101)G2,2 → GLm(R) is a representation, take the associated Lie algebra representation of g2,2, it to G2,2 ⊂ G2 and compose with the projection (cid:101)G2,2 → G2,2 to obtain a (cid:101)G2,2-representation. It Lie algebra representations are equal, and hence the center of (cid:101)G2,2 acts trivially. is not difficult to check that this representation coincides with the first one, since the corresponding 40 C. DRAPER our compact group of type G2. This happens not only for the definite norm and the compact group G2,−14, but for the norm of signature (4, 3) and the split group G2,2, thus providing transitive actions on the sphere (pseudosphere, respectively) of dimension 7. We are going to make all the above completely precise. We will work throughout the section with the real field, although all this works equally well in the complex setting. k Recall first that if n : V → R is a nondegenerate quadratic form, the Clifford alge- bra Cl(V, n) is a unital associative algebra that is generated by V subject to the rela- tions v2 = n(v)1 for all v ∈ V (that is, the tensor algebra T(V ) = ⊕k≥0V ⊗ ··· ⊗V quotient by the two-sided ideal generated by elements v ⊗ v − n(v)1). The prod- uct induced by the tensor product in the quotient algebra is written using jux- taposition (e.g. uv). Of course Clifford algebras satisfy the following universal property: if ρ : V → A is a linear map into a unital associative algebra A such that ρ(v)2 = n(v)1A, then there is a unique homomorphism of associative algebras ρ : Cl(V, n) → A such that ρV = ρ. (cid:19) If the dimension of V is m and {e1, ..., em} is an orthogonal basis of (V, n), then a basis of Cl(V, n) is {ei1 ei2 ··· eik 1 ≤ i1 < i2 < ··· < ik ≤ m and 0 ≤ k ≤ m}, so = 2m. The natural Z-grading on T(V ) given by assigning the degree deg(ei1 ⊗ei2 ···⊗eik ) = k, induces a Z2-grading on T(V ) which is inherited by the Clifford algebra Cl(V, n), since the quotient ideal is homogeneous. Thus, that dim Cl(V, n) =(cid:80)m (cid:18) m k=0 k Cl(V, n) = Cl¯0(V, n) ⊕ Cl¯1(V, n), being the even part of the Clifford algebra (respectively odd), denoted by Cl¯0(V, n) (resp. Cl¯1(V, n)), spanned by elements which are products of an even (resp. odd) number of vectors in V . In particular, dim Cl¯0(V, n) = 2m−1. If m is even, then Cl(V, n) is simple, and otherwise, Cl¯0(V, n) is simple (facts about real and complex Clifford algebras can be consulted in [Ha, Chapter 9]). The spin group lives in the group of invertible elements of Cl¯0(V, n), namely, 2r(cid:89) i=1 Spin(V, n) := {±a1 . . . a2r : ai ∈ V, n(ai) = 1}. n(a) a the For each a ∈ V with n(a) (cid:54)= 0, consider τa : V → V , τa(v) = v − 2n(a,v) reflection relative to the hyperplane orthogonal to a, which is an isometry with det τa = −1. It is easy to express this reflection in terms of the multiplication in the Clifford algebra: τa(v) = −ava−1, since τa(a) = −a and τa(v) = v for any v orthogonal to a. (The element a is invertible in Cl(V, n) since a2 = n(a)1 ∈ R.) This permits us to define the group homomorphism Spin(V, n) → SO(V, n), ±a1 . . . a2r (cid:55)→ τa1 . . . τa2r , (6.1) which is an epimorphism, since every orthogonal transformation is composition of reflections, by Dieudonné's theorem, and the determinant equal to 1 forces the number of reflections to be even. Besides the kernel is {±1} ∼= Z2. Indeed, if τa1 . . . τa2r = idV , then a1 . . . a2r commutes with V and hence it belongs to the center of Cl(V, n). The center of the Clifford algebra depends on m the dimension of V : If m is even, the center is R1, but if m is odd, the center is R1 ⊕ Re1 . . . em. In any case, as a1 . . . a2r is in the center, then r = 0. We will apply all the above to V = R7 and −n the opposite of the norm related to any generic 3-form in V (hence either n is positive definite or n has signature NOTES ON G2 41 (4, 3)).17 Besides, let × be the related cross product and C = R ⊕ R7 the octonion algebra (either split or division) as in Eq. (3.2). (This change of notation makes easier to deal with both cases simultaneously.) We will soon need some important properties of this (nonassociative) algebra, as for instance: It suffices to prove it if x, y ∈ V (for 1 it is evident). P1) Alternativity: x(xy) = (x2)y and (xy)x = x(yx) for all x, y ∈ C, in other words, the associator (x, y, z) = (xy)z − x(yz) is alternating (obviously trilinear). If x and y are orthogonal, x(xy) = x × (x × y) = −n(x)y = (x2)y by (4.13). And if x = y (or scalar multiple), of course (x, x, x) = −n(x)x + xn(x)1 = 0. P2) If x = s + u, s ∈ R, u ∈ V , we denote by ¯x := s − u. Then V = {x ∈ C : ¯x = −x} and tr(x) = 2n(x, 1) = x + ¯x. Also, x¯x = s2 + n(u) = n(x)1, so that every nonisotropic element is invertible. Besides, for all x, y ∈ C, ¯x¯y = yx. This is true in any alternative algebra, since This is a straightforward computation taking into account the skew- symmetry of the cross product, since, for x = s+u, y = t+v (s, t ∈ R, u, v ∈ V ), ¯x¯y = st−n(u, v)1−tu−sv+u×v and yx = st−n(v, u)1− P3) First Moufang Identity: (xax)y = x(a(xy)) for all x, y, a ∈ C. (cid:0)tu+sv+v×u(cid:1). = −(x2, a, y) − x2(ay) − (x2, y, a) − x2(ya) + x(cid:0)(xa)y + (xy)a(cid:1) = = −x2(ay + ya) + x(cid:0)(xa)y + (xy)a(cid:1) = = x(cid:0) − x(ay + ya) + (xa)y + (xy)a(cid:1) P1 = x(cid:0)(x, a, y) + (x, y, a)(cid:1) = 0. = −(x2a)y + x((xa)y) − (x2y)a + x((xy)a) = (xax)y − x(a(xy)) = (xa, x, y) + (x, a, xy) P1 = −(x, xa, y) − (x, xy, a) = P4) Second Moufang Identity: (xy)(ax) = x(ya)x for all x, y, a ∈ C, since P1 = −(x, ax, y) − x(y, a, x) (xy)(ax) − x(ya)x = (x, y, ax) − x(y, a, x) but = x(cid:0)(ax)y − a(xy)(cid:1) = x(a, x, y) P3 P1 = x(y, a, x). −(x, ax, y) = x((ax)y) − (xax)y Let ρ : V → End(C) be the map ρ(u) = Lu, for Lu the left multiplication in C given by Lu(x) = ux. The alternativity proved in (P1) provides ρ(u)2 = Lu2 = L−n(u)1 = −n(u)id. As before, the universal property gives ρ : Cl(V,−n) → End(C) a homomorphism of associative algebras and hence C is an eight-dimensional rep- resentation of Cl(V,−n) (not faithful). The restriction to the even Clifford algebra (6.2) is an isomorphism of associative algebras, since it is an injective linear map between vector spaces of the same dimension (dim End(C) = 82 = 27−1). The injectivity is a consequence of that ker(ρ) is an ideal of the simple algebra Cl¯0(V,−n) and the map is nonzero. Equation (6.2) implies that C is an irreducible Cl¯0(V,−n)- representation. The spin group Spin(V,−n) acts on C via the Clifford algebra, and this representation ρSpin(V,−n) is called the spin representation. Its elements (belonging to C, i.e. to R8) are called spinors. That is, if x ∈ C and g = ±a1 . . . a2r ∈ ∼=−→ End(C) Spin(V,−n), for ai ∈ V,(cid:81)2r i=1 n(ai) = 1, the spin action is ρ : Cl¯0(V,−n) g · x = ρ(±a1 . . . a2r)(x) = ±La1 . . . La2r (x) = ±a1(a2(. . . (a2rx) . . . )). 17The classical notation for Cl(V, n) is Clp,q(R) if (p, q) is the signature of n, and also Clp(R) if q = 0. 42 C. DRAPER 4 ≡ S7 4 when n is isotropic (S7 Let us check first that this is an action by isometries:18 ρ(Spin(V,−n)) ⊂ O(C, n). If x, y ∈ V , recall that Lx(y) = xy = −n(x, y)1 + x × y, so that n(Lx(y), z) = n(x × y, z) = −n(y, x × z) = −n(Lx(z), y) if x, y, z ∈ V . But n(Lx(y), z) = −n(Lx(z), y) is also true for any y, z ∈ C: if y = z = 1, then n(Lx(y), z) = n(x, 1) = 0 = −n(Lx(z), y); and if y = 1, z ∈ V , then −n(Lx(z), y) = −n(xz, 1) = −n(−n(x, z)1, 1) = n(x, z) = n(Lx(y), z). Consequently Spin(V,−n) acts on the manifold {x ∈ C n(x) = 1}, which is the sphere S7 when n is positive definite, and, following again O'Neill's notation, it is the pseudosphere S7 4(1)). Our goal is to check that this is a transitive action such that the isotropy subgroup of any element has type G2. For the transitivity, we use the next auxiliary result. Lemma 6.1. If x ∈ C, n(x) = 1, then there is a, b ∈ V such that ab = x. Proof. Take x = s1 + u with u ∈ V and n(x) = 1. If u = 0, then x = ±1 and If the result is clear: take any v ∈ V with n(v) (cid:54)= 0 and then v(± ¯v u (cid:54)= 0 but n(u) = 0, take v ∈ V such that n(u, v) (cid:54)= 0. Then Q = (cid:104)1, u, v, u × v(cid:105) is a 4-dimensional subalgebra of C such that nQ is nondegenerate (usually called Indeed, taking in mind (4.13), u(u × v) = n(u, v)u, a quaternion subalgebra). uv = −n(u, v)1 + u × v and v(u × v) = −n(v, u)v + n(v)u, all of them belonging to Q. If, on the contrary, n(u) (cid:54)= 0, take v ∈ V orthogonal to u but nonisotropic and again Q = (cid:104)1, u, v, u × v(cid:105) is a quaternion subalgebra of C (now uv = u × v, u(u × v) = −n(u)v and v(u × v) = n(v)u). In both cases, nondegeneracy gives C = Q ⊕ Q⊥ and there is w ∈ Q⊥ with n(w) (cid:54)= 0. Note that xQ⊥ ⊂ Q⊥, since for any y ∈ Q⊥ and z ∈ Q, n(v) ) = x. n(xy, z) = sn(y, z) − n(y, uz) = n(y, ¯xz) ∈ n(Q⊥ , QQ) ⊂ n(Q⊥ As x is invertible (P2), then xQ⊥ = Q⊥ and there is b ∈ Q⊥ with w = xb. Hence (P1) is a product of two elements in Q⊥ n(b) = n(w) (cid:54)= 0 and x = (xb) ⊂ (cid:3) 1⊥ = V . n(b) (cid:16) ¯b (cid:17) , Q) = 0. With the notations above, Proposition 6.2. The group Spin(V,−n) acts transitively on {x ∈ C n(x) = 1}. The isotropy subgroup {g ∈ Spin(V,−n) : g · 1 = 1} is isomorphic to Aut(C). Proof. Let x ∈ C be an element with n(x) = 1 and take a, b ∈ V with ab = x as in the previous lemma. Now ab ∈ Spin(V,−n) (take care with the abuse of notation, here ab denotes the product of elements of V in the Clifford algebra, while in the above line the product ab was the octonionic product) since n(a)n(b) = n(x) = 1, and besides ab · 1 = LaLb(1) = ab = x. Hence the orbit of the element 1 ∈ C is the whole {x ∈ C : n(x) = 1}. This finishes the transitivity of the action. Now take H = {g ∈ Spin(V,−n) : g·1 = 1} the isotropy subgroup of 1 ∈ C.19 We are going to prove that ρ(H) (isomorphic to H) coincides with Aut(C). Consider a triple product in C given by (cid:104) , , (cid:105) : C × C × C → C, (cid:104)x, y, z(cid:105) := (x¯y)z. (6.3) 18This is not consequence of (6.1). If x ∈ V , any element g ∈ Spin(V, −n) acts also on x by means of its image by (6.1) on the orthogonal group, but these are completely different actions: τa1 . . . τa2r (x) (cid:54)= ±a1(a2(. . . (a2rx) . . . )). From a different viewpoint, Spin(V, −n) acts on V by (6.1), but it cannot be considered a subgroup of GL(V ). The least m such that Spin(V, −n) ⊂ GLm(R) is m = 8. 19Although 1 is a remarkable element in the octonion algebra C, it is not a remarkable unit spinor, all of them are undistinguishable as a consequence of the transitivity. NOTES ON G2 43 (And observe that the octonionic product is recovered in some way from this triple product since xy = (cid:104)x, 1, y(cid:105).) Then, for any x, y, z ∈ C, a ∈ V , with(cid:81)2r (cid:104)Lax, Lay, Laz(cid:105) = ((ax)(ay))(az) = −((ax)(¯ya))(az) P1 = n(a)a((x¯y)z) = n(a)La(cid:104)x, y, z(cid:105). P2 = ((ax)(¯y¯a))(az) P4 = −(a(x¯y)a)(az) P3 = −a((x¯y)(a(az))) Hence any g ∈ Spin(V,−n) preserves the triple product: there are a1, . . . , a2r ∈ V i=1 n(ai) = 1 such that g · x = ±La1 . . . La2r (x), and then (cid:104)g · x, g · y, g · z(cid:105) = (±1)3n(a1)La1 . . . n(a2r)La2r(cid:104)x, y, z(cid:105) = ±La1 . . . La2r(cid:104)x, y, z(cid:105) = g · (cid:104)x, y, z(cid:105). Now, if g ∈ Spin(V,−n) fixes the element 1 ∈ C (consider g as an endomorphism of C by means of ρ, that is, g(x) := ρ(g)(x) = g · x), then g(xy) = g((cid:104)x, 1, y(cid:105)) = (cid:104)g · x, g · 1, g · y(cid:105) = (cid:104)g · x, 1, g · y(cid:105) = g(x)g(y), (cid:19) (cid:18) 7 2 and g ∈ Aut(C). This proves ρ(H) ⊂ Aut(C). Besides dim H = dim Spin(V,−n) − dim(Spin(V,−n) · 1) = dim SO(V,−n) − dim{x ∈ C : n(x) = 1} = − 7 = 14, what implies that the connected components of ρ(H) and of Aut(C) coincide. As we proved that Aut(C) is connected (see Lemma 5.15), we conclude that ρ(H) = (cid:3) Aut(C). This gives immediately, Corollary 6.3. Consider C = R8 endowed with a norm n either positive definite ⊥ ∼= R7. Consider the or of signature (4, 4), take x ∈ C with n(x) = 1, and V = (cid:104)x(cid:105) spin representation of Spin(V,−n) on C given by the restriction of (6.2). Then the stabilizer of the unit spinor x, (cid:26) G2,−14 G2,2 if n is definite, otherwise, {g ∈ Spin(V,−n) : g · x = x} ∼= and hence Spin(7)/G2,−14 ∼= S7, (6.4) Observe that the fundamental groups of the real Lie groups of type G2 computed in Proposition 5.17 agree with the ones we would have obtained by looking at the coset manifolds above, since Spin(3, 4)/G2,2 ∼= S7 4. π1(Spin(7)) = π1(G2,−14) = 0, π1(Spin(3, 4)) = π1(G2,2) = Z2. (Recall that the spin groups are nonnecessarily simply connected if the signature is not definite.) Both homogeneous spaces in Eq. (6.4) are naturally reductive homogeneous spaces. The algebraical structure produced in (m, 1 2 [ , ]m) is isomorphic to (V,×) for V = R7 and × the related cross product as in the above sections. We can ob- serve that this is a simple Malcev (non-Lie) algebra. Many geometrical properties 4 can be concluded from this algebraical about torsion and curvature on S7 and S7 fact. A more detailed study of these reductive homogeneous spaces, jointly with the determination of their invariant affine connections, can be found in [E93]. We can also conclude, taking into account Propositions 5.2 and 5.12, besides Corollary 6.3, that Corollary 6.4. 44 C. DRAPER i) If n is positive definite and ϕ1, ϕ2 ∈ C are two orthogonal unit spinors, {g ∈ Spin(V,−n) : g · ϕ1 = ϕ1, g · ϕ2 = ϕ2} ∼= SU(3). Thus, Spin(7)/SU(3) ∼= V2(R8) is the Stiefel manifold of orthogonal pairs of unital elements in R8. ii) If n has signature (4, 3) and ϕ1, ϕ2 ∈ C are two orthogonal spinors with n(ϕ1) = 1 = −n(ϕ2), then Thus, Spin(3, 4) {g ∈ Spin(V,−n) : g · ϕ1 = ϕ1, g · ϕ2 = ϕ2} ∼= SL3(R). (cid:16) 1 SL3(R) ∼= V1,1(R4,4) := {a ∈ Mat8,2(R) : at(cid:16) I4 (cid:17) 0 −I4 a = 0 0 0 −1 (cid:17) }. Furthermore, if we consider a third spinor, the stabilizer of the three spinors will be either SU(2) or SL2(R). Remark 6.5. When n is positive definite, we are going to see all the above in a more concrete way adapted from [prA, §2.1] to try to relate the different approaches and models. i=1 is the canonical basis of the subspace of the zero trace elements in O as in Figure 2, and we denote by κi := ρ(ei) = Lei, then we have that {κi}7 i=1 ⊂ so(O, n) is a family of skew-symmetric operators such that If {ei}7 κiκj = −κjκi If e8 denotes the identity element in O and ϕu,v := n(u,−)v − n(v,−)u ∈ so(O, n), then ϕij := ϕei,ej sends ei to ej and ej to −ei, and κ2 i = −id, if i (cid:54)= j. κ1 = −ϕ18 + ϕ23 + ϕ47 − ϕ56, κ5 = ϕ16 − ϕ27 − ϕ34 − ϕ58, κ2 = −ϕ13 − ϕ28 + ϕ46 + ϕ57, κ6 = −ϕ15 + ϕ24 − ϕ37 − ϕ68, κ7 = ϕ14 + ϕ25 + ϕ36 − ϕ78. κ3 = ϕ12 − ϕ38 − ϕ45 + ϕ67, κ4 = −ϕ17 − ϕ26 + ϕ35 − ϕ48, Note that the map so(V,−n) → Cl(V,−n)− given by ϕu,v (cid:55)→ − 1 2 [u, v] is a Lie algebra monomorphism.20 Thus we can consider so(V,−n) living in Cl(V,−n)−, and spin(V,−n) the related copy of so(V,−n) contained in ρ(Cl¯0(V,−n)) = gl(O). That is, spin(V,−n) = [ρ(V ), ρ(V )], which is then spanned by {κiκj = 1 2 [κi, κj] : 1 ≤ i < j ≤ 7}. As we have shown that Aut(O) = ρ ({g ∈ Spin(V,−n) : g · e8 = e8}), then {d ∈ spin(V,−n) : d(e8) = 0} = Der(O) ∼= Der(V, Ω1) = gc provides an isomorphism by means of the restriction map d (cid:55)→ dV . Observe that an arbitrary element in spin(V,−n) does not act on V (but on O), but it does if it annihilates (cid:104)e8(cid:105) = V ⊥. with d(u) = 0 (u ∈ O any nonzero fixed element), what provides some linear equations in the coefficients aij. To be precise, if we choose u = e8, then 0 = i=1 are d(u) = − zero, that is, Thus gc can be identified with the set of elements d =(cid:80) aijκiκj ∈ spin(V,−n)   . (cid:80) i<j aijκi(ej) ∈ V , whose coordinates relative to the basis {ei}7 (cid:80) aijκiκj ∈ spin(V,−n) : a23 + a47 − a56 = 0 a13 − a46 − a57 = 0 a12 − a45 + a67 = 0 a17 + a26 − a35 = 0 a16 − a27 − a34 = 0 a15 − a24 + a37 = 0 a14 + a25 + a36 = 0 gc = (6.5) 20If A is an associative algebra, then A− denotes the Lie algebra defined over the same vector space by the Lie bracket given by the commutator [a, b] = ab − ba. NOTES ON G2 (cid:80) Observe that the matrix of ϕi,j relative to the above basis is Eji − Eij so that i<j aijκiκj (cid:55)→ −(aij) with aji = −aij is an isomorphism of spin(V,−n) onto the Lie algebra so(7) of the skewsymmetric matrices of size 7. According to the previous sections, the subalgebra h = {d ∈ spin(V,−n) : d(e8) = 0 = d(e7)} should be isomorphic to su(W, σ). In order to understand it directly, we compute (cid:80) aijκiκj : ai7 = 0∀i a14 + a25 + a36 = 0 45 a26 = a35 a23 = a56 a16 = a34 a15 = a24 a12 = a45 a13 = a46  h =  , so that the precise isomorphism ψ : h → su(3) = {c ∈ Mat3(C) : c + ¯ct = 0} is given by (6.6) The choice of an h-invariant complementary subspace should allow us to recover ψ 0 −bt a 0 0 0 for a, b ∈ Mat3(R), a = −at, b = bt, tr(b) = 0. the compact model in Proposition 5.4. We denote by b 0  a  ly  , µx,y  = a + ib   = µax+by,ay−bx. 2x 2y 0 lx −ly lx −2xt −2yt µx,y :=  a b 0 −bt a 0 0 0 0 where lx is the coordinate matrix of the cross product in R3 as in Eq. (2.1). m :=(cid:8)µx,y : x, y ∈ R3(cid:9), clearly it is the required subspace, since gc = h ⊕ m and If Thus we can extend the Lie algebra isomorphism (6.6) to an isomorphism from because ψ : gc → L = su(W ) ⊕ W, ψ(µx,y) = −y − ix ∈ C3 = W, [µx,y, µu,v]m = 2µ y×u+x×v, x×u−y×v, [µx,y, µu,v]h = 3ψ−1(cid:0)ly×v+x×u + ipr0(uyt + yut − vxt − xvt)(cid:1), where pr0(b) = b − 1 3 tr(b)I3 ∈ sl3(R) denotes the projection on the special linear algebra. (Recall tr(uyt) = uty.) Furthermore, we can translate the natural action of gc ⊂ so(7) on the irreducible real representation R7 given by the usual matrix product. For ψ(cid:48) : R7 → V = R ⊕ C3 given by  x y s (cid:48) ψ  = s − x + iy, one easily checks that ψ(d) · ψ(cid:48)(X) = ψ(cid:48)(dX), which provides an alternative proof of Proposition 5.4. Note that if we repeat all the above for isotropic n, we would get a model analogous to the split one in Proposition 2.1 (with blocks 3 + 3 + 1 now). Remark 6.6. All these points of view are, of course, interconnected. If n is positive definite, any choice of spinor provides a 3-form in such a way that the group fixing the spinor coincides with the group fixing the related 3-form by the usual pullback. ⊥, but, by means of the ⊥ (recall n(Lvu, u) = 0 since Indeed, take 0 (cid:54)= u ∈ C. We have C = Ru ⊕ (cid:104)u(cid:105) action of Cl(V,−n), 1 · u = u and ρ(V )(u) ⊂ (cid:104)u(cid:105) 46 C. DRAPER Lv is skew-adjoint). As any (nonzero) element in V is invertible in the Clifford ⊥ by dimension count. Take Au : V × V → V the algebra, V · u = ρ(V )(u) = (cid:104)u(cid:105) multiplication defined by Au(y, x) · u = y · (x · u) + n(x, y)u. (Since n(y · (x · u) + n(x, y)u, u) = 0, it follows that there is exactly one element v ∈ V such that v· u = y· (x· u) + n(x, y)u.) This provides a skew-symmetric (1, 2)- tensor as y · (x · u) + x · (y · u) = ρ(yx + xy)(u) = ρ(−2n(x, y)1)(u) = −2n(x, y)u and so Au(y, x) = −Au(x, y). It is not difficult to check also that Au(x, y) is orthogonal to x and y in V , and that n(Au(x, y)) = n(x)n(y) − n(x, y)2, by using that n(xu, yu) = n(x, y)n(u) in C. Hence Au is a cross product for n and the map ωu : V × V × V → R given by ωu(x, y, z) = n(x, Au(y, z)) is the required 3-form. ⊥. Then If n is split we need u ∈ C nonisotropic to assure first that C = Ru⊕(cid:104)u(cid:105) ⊥ (simply take into account all works similarly to the definite case since V · u = (cid:104)u(cid:105) that we can find a basis of V formed by nonisotropic vectors), and then ωu defined as before is a 3-form such that g∗ωu = ωu if g ∈ Spin(V,−n) with g · u = u. The arguments become closed since we can define in the set of spinors R8 = ⊥ a product making it an octonion algebra such that u is its identity Ru ⊕ (cid:104)u(cid:105) element, simply by using the cross product related to the 3-form ωu. Observe that the spinorial representation C is Spin(V,−n)-irreducible, but not ⊥ is the decomposition as a sum of two irreducible G2- ⊥ is the nontrivial representation for G2 of the least possible G2-irreducible: C = Ru⊕(cid:104)u(cid:105) modules. In fact, (cid:104)u(cid:105) dimension, 7. Remark 6.7. The triple product described in Eq. (6.3) is closely related to generic 3-forms. (Again, it is not important if F is either R or C, V = F7 and C = F ⊕ V is an octonion algebra.) First, note that Λ : V × V × V × V → F, Λ(x, y, z, u) = 1 2 n(x,(cid:104)y, z, u(cid:105) − (cid:104)u, z, y(cid:105)) is multilinear and alternating, that is, a 4-form21: If one of the last three entries is repeated, then (cid:104)y, z, u(cid:105)−(cid:104)u, z, y(cid:105) = 0 is an immediate consequence of the alternativ- ity (P1). Also, 2Λ(x, x, y, z) = −n(x, (xy)z) + n(x, (zy)x) = n(xz, xy)− n(x2, zy) = n(x)n(z, y)−x2n(1, zy) = 0 since x2 = −n(x)1 and Ly is skew-adjoint for the norm. Of course the group G2 = Aut(C) preserves this 4-form because G2 preserves both the norm and the triple product. Furthermore, G2 can also be characterized as the group preserving the 4-form.22 To this purpose, simply observe that for our generic G2-invariant 3-form Ω : V × V × V → F, Ω(x, y, z) = n(xy, z), we have 0 (cid:54)= Ω∧Λ ∈ ∧7V ∗ ∼= F. We will check this fact in the definite case (hence, it will be also true in the complex case, and so in the real-split): It is a straightforward computation, using the products in Figure 2, that if 0 (cid:54)= (ei, ej, ek) (if these three elements do not associate, they are not in the same line in the Fano plane, hence), 21 Note that (cid:104)y, z, u(cid:105) − (cid:104)u, z, y(cid:105) = Ω(y, z, u) − (y, z, u) in C, so that Λ(x, y, z, u) = − 1 2 n(x, (y, z, u)) since x ∈ V . In the compact case, Λ is called the coassociative 4-form for O in [Ha, 6]. 22 To be more precise, this happens in both the real cases. In the complex case, the group × {isidV : s = 0, . . . , 3} (their identity components preserving the 4-form is not GΩ0 but G+ Ω0 coincide), since the maps of determinant 1 preserve the volume form and hence Ω0 ∧ Λ. NOTES ON G2 47 then there is l a different index such that (cid:104)ei, ej, ek(cid:105) − (cid:104)ek, ej, ei(cid:105) = ±2el, and we compute Λ = −e1245 + e1267 − e1346 − e1357 + e2347 − e2356 − e4567, with the notation used in Eq. (4.8). Thus, (With a different notation, this means that the Hodge star operator (cid:63)Ω1 = Λ, up to a scalar depending on the choice of a fixed nonzero 7-form.) Ω1 ∧ Λ = −7e1234567 (cid:54)= 0. Remark 6.8. I. Agricola pointed me out that it is not difficult to characterize the three real and complex algebras of type G2 in the language of p-forms. We will do it in the compact case. We identified so(V,−n) with ∧2V ∗ as vector spaces by means of ϕi,j (cid:55)→ eij, which allows to see gc ⊂ ∧2V ∗ as in Remark 6.5 and also to have a Lie bracket in ∧2V ∗.23 Then, gc = {α ∈ ∧2V and ∧2V ∗ = gc ⊕ m for the gc-invariant complement m = {u (cid:121) Ω1 : u ∈ V } ≤ ∧2V : (cid:63)(Ω1 ∧ α) = α}, (6.7) ∗ ∗ , which can be also characterized as ∗ m = {α ∈ ∧2V (6.8) We are now trying to provide justifications of the above facts. First note that F is diagonalizable with real eigenvalues, since it is a symmetric endomorphism relative to the scalar product (cid:104) , (cid:105) in ∧2V ∗ induced by n. Indeed, for α, β ∈ ∧2V ∗, (cid:104)F (α), β(cid:105)e1234567 = Ω1 ∧ α ∧ β = Ω1 ∧ β ∧ α = (cid:104)F (β), α(cid:105)e1234567. : (cid:63)(Ω1 ∧ α) = −2α}. Besides the sum of the eigenvalues is zero, since its matrix relative to the basis {eij : 1 ≤ i < j ≤ 7} has all the entries in the diagonal equal to zero and in particular zero trace. For easy computations, one can use the coassociative 4-form, as F (α) = −α (cid:121) Λ. All this is better understood with the help of a convenient grading of ∧2V ∗ compatible with F . Take the Z3 2-grading on the octonion algebra O given by O(¯1,¯0,¯0) = (cid:104)e1(cid:105), O(¯0,¯1,¯0) = (cid:104)e2(cid:105), O(¯0,¯0,¯1) = (cid:104)e7(cid:105). That is, the basic elements ei are all of them homogeneous, with degrees deg(e3) = (¯1, ¯1, ¯0), deg(e4) = (¯1, ¯0, ¯1), deg(e5) = (¯0, ¯1, ¯1), deg(e6) = (¯1, ¯1, ¯1) and deg(1) = (¯0, ¯0, ¯0).24 Call gi = deg ei. Any partition of a subspace C in pieces indexed by a group C = ⊕g∈GCg produces a group-grading on L = gl(C) by means of L = ⊕g∈GLg with Lg = {f ∈ gl(C) : f (Ch) ⊂ Cgh}, i.e., [Lg, Lh] ⊂ Lgh. This is our 2. The subalgebra spin(V,−n) inherits this grading (spin(V,−n)g = case for G = Z3 spin(V,−n) ∩ Lg), and the homogeneous components are j ∧ (eiej)∗ : f (cid:54)= i}(cid:105) Wi := (spin(V,−n))gi = {x ∈ spin(V,−n) : x(Cg) ⊂ Cggi ∀g} = (cid:104){e∗ ⊂ {α ∈ ∧2V ∗ : α · e8 ⊂ (cid:104)ei(cid:105)}. 23The Lie bracket in ∧2V ∗ becomes [α1 ∧ β1, α2 ∧ β2] = n(α1, α2)β1 ∧ β2 − n(β1, α2)α1 ∧ β2 − n(α1, β2)β1 ∧ α2 + n(β1, β2)α1 ∧ α2, for αi, βi 1-forms. 24 This remarkable grading exists in the split case too, where every nonzero homogeneous element is invertible. That is why, although Os is not a division algebra, it is a graded division algebra. In particular, this allows to construct both real octonion algebras as twisted group algebras for the group Z3 2. 48 C. DRAPER For instance W4 = (cid:104)e17, e26, e35(cid:105), and in all the cases Wi is spanned by the 2-forms ejk such that ei, ej and ek associate (eiej = ±ek). Besides, as the linear map Wi → (cid:104)ei(cid:105), α (cid:55)→ α · e8 is not zero, its kernel is a plane W (cid:48) i = {α ∈ Wi : α · e8 = 0} ≤ gc. The Z3 2-grading ∧2V ∗ = ⊕7 i=1Wi splits our Lie algebra of type B3 into seven Cartan subalgebras (the point is that, when we multiply two of them, we get another i .25 Now, Cartan subalgebra!), as well as happens with the Z3 the endomorphism F commutes with the three order 2 automorphisms of ∧2V ∗ producing the grading, so that Wi are F -invariant subspaces. In fact, all of them are undistinguishable. The matrix of FW4 relative to BW4 = {e17, e26, e35} is 2-grading gc = ⊕7 i=1W (cid:48)  0 −1 −1 1 0 1  , 1 1 0 so that the kernel ker(FW4 − id) coincides with W (cid:48) 4 = {a17e17 + a26e26 + a35e35 : a17 + a26 − a35 = 0}. Similarly the equations defined by ker(FWi − id) for i = 1, . . . , 7 are equal to the seven equations in (6.5). Now, the third eigenvalue of FW4 has to be −1−1 = −2, whose eigenspace is the 4, i.e. ker(FW4 + 2id) = (cid:104)e17 + e26 − e35(cid:105) = (cid:104)e4(cid:121) Ω1(cid:105). A more orthogonal one to W (cid:48) conceptual argument for proving the characterization of the spin representation is to use that Ω1 ∧ (u (cid:121) Ω1) ∧ (v (cid:121) Ω1) = 6n(u, v)e1234567, for any u, v ∈ V . Thus, (cid:104)F (u (cid:121) Ω1), v (cid:121) Ω1(cid:105) = 6n(u, v) = 2(cid:104)u (cid:121) Ω1, v (cid:121) Ω1(cid:105), which i , ej (cid:121) Ω1] ⊂ In particular gc ∩ m = 0. (cid:104)(eiej) (cid:121) Ω1(cid:105). Again, for α = a17e17 + a26e26 + a35e35 an arbitrary 2-form in W4, which belongs to m if (and only if) α ∈ gc. This computation also gives the algebraic structure provided by the reductive decomposition26, since [e4 (cid:121) Ω1, e1 (cid:121) Ω1]m = −e7 (cid:121) Ω1, [e4 (cid:121) Ω1, e1 (cid:121) Ω1]gc = 2e14 − e25 − e36, In order to prove (6.7), we check [W (cid:48) [α, e1 (cid:121) Ω1] = (−a26 + a35)e7 (cid:121) Ω1 + (a17 + a26 − a35)e14, proves (6.8). so that [ , ]m = 1 2× under the natural correspondence between m and V . Finally, some textbooks to achieve a good background for our study are: (cid:67) More about octonions in Okubo's book [Ok]; (cid:67) About exceptional algebras [Ja, Sc]; (cid:67) About exceptional groups [Ad, Ro]. Acknowledgments To my colleague and always master Alberto Elduque, for helping me with some technical points in the proof of Proposition 4.14 and for his availability; to Ilka Agricola, for reading and improving several versions of this manuscript and for 25When one works with the analogous Z3 2-grading on the split algebra g2,2, any basis formed by homogeneous elements satisfies that all its elements are semisimple. The bases with this property are important for Physics, but usually they are not easy to find in the split algebras, since all the elements in the root spaces are nilpotent. There are models of both the real forms of g2 based on this Z3 2-grading. Group gradings on the complex algebra g2 are studied in [DM06], and on its real forms in [CDM10]. 26For a more algebraical approach, this reductive decomposition is ∧2V ∗ = so(O0, n), gc = Der(O), m = ad(O0). NOTES ON G2 49 encouraging me to write it and to make it accessible; and to Skip Garibaldi, for his nice suggestions and references. References [Ad] [A08] [prA] [A06] J.F. Adams, Lectures on exceptional Lie groups. Chicago Lectures in Mathematics, University of Chicago Press, Chicago, IL, 1996. I. Agricola, Old and New on the exceptional Lie group G2, Notices of the AMS 55 (2008), 922 -- 929. I. Agricola, Lectures on the exceptional Lie group G2, Private notes about lectures at the universities of Torino and Marburg. I. Agricola, The Srní lectures on non-integrable geometries with torsion, Arch. Math. (Brno) 42 (2006), 5 -- 84. With an appendix by Mario Kassuba. [ABoF17] I. Agricola, A. Borówka, T. Friedrich, S6 and the geometry of nearly Kähler 6- manifolds, To appear in this volume. J.C. Baez, The octonions, Bull. Amer. Math. Soc. 39 (2002), 145 -- 205. [Ba02] [BDE03] P. Benito, C. Draper and A. Elduque, Models of the octonions and g2, Linear Algebra [BDE05] P. Benito, C. Draper and A. Elduque, Lie-Yamaguti algebras related to g2, J. Pure and its Applications 371 (2003) , 333 -- 359. Appl. Algebra 202 (2005), no. 1-3, 22 -- 54. [BrG67] R.B. Brown and A. Gray, Vector cross products, Comment. Math. Helv. 42 (1967), 222 -- 236. [CDM10] A.J. Calderón, C. Draper and C. Martín, Gradings on the real forms of the Albert algebra, of g2 and of f4, J. Math. Phys. 51(5) (2010), 053516, 21 pp. É. Cartan, Sur la structure des groupes de transformations finis et continus, Thèses, Nony, Paris: 2nd ed., Vuibert, Paris, 1933; OEuvres complètes: Partie I, Groupes de Lie, vols. 1 -- 2, Gauthier-Villars, Paris, 1952, pp 137 -- 287. É. Cartan, Les groupes réels simples finis et continus, Ann. Sci. École Norm. Sup. 31 (1914), 263 -- 355. [CHe88] A. Cohen and A.G. Helminck, Trilinear alternating forms on a vector space of di- [Ca94] [Ca14] [Co89] [D08] [DM06] [Dy46] [E] [E93] [E92] [En00] [FuH] [GGu15] [Ha] [He83] [Hi00] [Ja] [Kan73] [Ka11] mension 7, Comm. Algebra 16 (1988), no. 1, 1 -- 25. A.J. Coleman, The greatest mathematical paper of all time, The Mathematical Intel- ligencer 11(3) (1989), 29 -- 38. C. Draper, Models of the Lie algebra F4, Linear Algebra Appl. 428 (2008), no. 11-12, 2813 -- 2839. C. Draper and C. Martín, Gradings on g2, Linear Algebra Appl. 418 (2006), no. 1, 85 -- 111. E.B. Dynkin, The classification of simple Lie groups, Mat. Sb. 18(3) (1946), 347 -- 352. A. Elduque, Lie algebras. Course notes. http://www.unizar.es/matematicas/algebra/elduque/files/LAElduque.pdf A. Elduque and H.C. Myung, The reductive pair (B3, G2) and affine connections on S7, Journal of Pure and Applied Algebra 86 (1993), 155 -- 171. A. Elduque and H.C. Myung, Color algebras and affine connections on S6, J. Algebra 149 (1992), no. 1, 234 -- 261. F. Engel, Ein neues, dem linearen Komplexe analoges Gebilde, Leipz. Ber. 52 (1900), 63 -- 76, 220 -- 239. W. Fulton and J. Harris, Representation Theory. A first course. Springer Verlag New York, Inc. (1991). S. Garibaldi and R. Guralnick, Simple Groups stabilizing polynomials, Forum of Math- ematics, Pi 3 (2015), e3, 43 pages. doi:10.1017/fmp.2015.3 F.R. Harvey, Spinors and calibrations. Perspectives in Mathematics, vol. 9, Academic Press Inc., Boston, MA, 1990. C. Herz, Alternating 3-forms and exceptional simple Lie groups of type G2, Can. J. Math., Vol. XXXV, No. 5, 776 -- 806 (1983) N. Hitchin, The geometry of three-forms in six and seven dimensions, Preprint arXiv:math/0010054 N. Jacobson, Exceptional Lie algebras. Lecture notes in pure and applied mathematics. Marcel Dekker, Inc., New York, 1971. I.L. Kantor, Models of Exceptional Lie algebras, Soviet Math. Dokl. 14 (1973), no. 1, 254 -- 258. S. Karigiannis, What is G2-manifold?, Notices of the AMS 58 (2011), 580 -- 581. C. DRAPER [SM77] M. Sato and T. Kimura, A classification of irreducible prehomogeneous vector spaces 50 [Ki88] [Ne] [Ok] [Ro] [pr17] [Sc] [Sch31] [Sh] [Tit66] [Vi66] [Vo94] [Wa] [We25] W. Killing, Die Zusammensetzung der stetigen endlichen Transformations-gruppen, Math. Ann. 31 (1888), 252 -- 290; 33 (1888), 1 -- 48, 34 (1889), 57 -- 122, 36 (1890), 161 -- 189. B. O'Neill, Semi-Riemannian Geometry With Applications to Relativity. Academic Press, 1983. S. Okubo, Introduction to octonion and other non-associative algebras in physics. Montroll Memorial Lecture Series in Mathematical Physics, vol. 2, Cambridge Univer- sity Press, Cambridge, 1995. B. Rosenfeld, Geometry of Lie groups. Mathematics and its Applications, 393. Kluwer Academic Publishers Group, Dordrecht, 1997. xviii+393 pp. ISBN: 0-7923-4390-5 D.A. Salamon and T. Walpuski, Notes on the octonions, preprint arXiv:1005.2820v3 (February 2017) and their relative invariants, Nagoya Math. J. 65 (1977), 1 -- 155. R.D. Schafer, An introduction to nonassociative algebras. Dover Publications Inc., New York, 1995. Corrected reprint of the 1966 original. J.A. Schouten, In Klassifizierung der alternier enden Grôssen dritten Grades in 1 Dimenionen, Rendiconti del Circolo Matematico di Palermo 55 (1931), 137 -- 156. R.W. Sharpe, Differential Geometry, Cartan's Generalization of Klein's Erlangen Pro- gram. Graduate Texts in Mathematics 166, Springer Verlag. J. Tits, Algèbres alternatives, algèbres de Jordan et algèbres de Lie exceptionelles. I. Construction, Indag. Math. 28 (1966), 223 -- 237. E.B. Vinberg, A construction of exceptional simple Lie groups, Tr. Semin. Vektorn. Tensorn. Anal. 13 (1966), 7 -- 9. (Russian) D. Vogan, The unitary dual of G2, Inventiones 116 (1994), 677 -- 791. F.W. Warner, Foundations of Differentiable Manifolds and Lie groups. Graduate texts in Mathematics 94, Springer-Verlag. Reprint of the 1971 original. H. Weyl, Theorie der Darstellungen kontinuierlicher halb-einfacher Gruppen durch lineare Transformationen, Math. Zeitschrift 24 (1924), 328 -- 295; 26 (1925), 271 -- 304. Departamento de Matemática Aplicada, Universidad de Málaga, 29071 Málaga, Spain E-mail address: [email protected]
1901.01313	1	1901	2019-01-04T21:08:23	Steinberg groups for Jordan pairs - an introduction with open problems	[ "math.RA", "math.KT" ]	The paper gives an introduction to Steinberg groups for root graded Jordan pairs, a theory developed in the recent book "Steinberg groups for Jordan pairs" by Ottmar Loos and the author.	math.RA	math	STEINBERG GROUPS FOR JORDAN PAIRS -- AN INTRODUCTION WITH OPEN PROBLEMS E. NEHER Abstract. The paper gives an introduction to Steinberg groups for root graded Jordan pairs, a theory developed in the book [LN2] by Ottmar Loos and the author. Introduction The connection between Jordan structures (Jordan algebras, Jordan pairs) and Lie alge- bras and groups has a long and successful history, starting with the work of Chevalley- Schafer [CS] and continued by Jacobson [Ja1, Ja2, Ja3], Kantor [Ka1, Ka2], Koecher [Ko1, Ko2, Ko3, Ko4], Loos [Lo2, Lo3, Lo5], Springer [Sp], Springer-Veldkamp [SV] and Tits [Ti1, Ti2]. The book [LN2] by Loos and the author is a further contribution to the theme "Groups and Jordan Structures". It contains a detailed study of Steinberg groups associated with certain types of Jordan pairs. These groups generalize the classical linear and unitary Steinberg groups of a ring by, roughly speaking, replacing associative coordinates with Jordan algebras or Jordan pairs. We are able to prove the basic results on Steinberg groups (central closedness, universal central extension in the stable case) in our setting, thereby recovering all previous results, except those on groups of type E8, F4 and G2, and in addition deal with new types, not considered before. The main novelty however is our approach based on 3-graded root systems and Jordan pairs. The present paper is an introduction to the theory developed in [LN2]. In §1 we describe the linear Steinberg group St(A) of a ring A from the point of view of Jordan pairs. This is motivation for §2 where we define the Steinberg group of a root graded Jordan pair and state the main results of [LN2] regarding these groups. The final section §3 discusses some open research problems in the area of Steinberg groups and Jordan pairs. The paper does not assume any prior knowledge of linear Steinberg groups or Jordan pairs: all relevant definitions are given in the paper. We demonstrate their scope by many examples and refer the reader to [LN1] and [LN2] for most proofs. But we include the details of our discussion of the linear Steinberg group and the elementary linear group of a ring from the point of view of Jordan theory (1.7 -- 1.10 and 1.12 respectively). We also give all details of our description of the Tits-Kantor-Koecher algebra and the projective elementary group of a rectangular Jordan pair (2.12, 2.13). Notation. Throughout k is a unital commutative associative ring and A is a not neces- sarily commutative, but unital associative k-algebra. Its identity element and zero element are denoted 1A and 0A respectively. We will often simply write 1 for 1A if A is clear from the context, and analogously for 0 ∈ A. We use A× to denote the invertible elements of A. If k = Z we will refer to A as a ring. Key words and phrases. Jordan pairs, Steinberg groups. The author was partially supported by a Discovery grant from the Natural Sciences and Engineering Research Council of Canada (NSERC). 1 2 E. NEHER For non-empty sets I and J we denote by MatIJ (A) the k-module of I×J-matrices over A, i.e., maps x : I × J → A with only finitely many values different from 0. As usual, we write a matrix in the form x = (xij)(i,j)∈I×J . In case I = J we abbreviate MatI (A) = MatIJ (A). This is an associative k-algebra with respect to ordinary matrix multiplication which is unital if and only if I is finite. We put Matn(A) = MatI (A) if I = n < ∞. Here and in general I denotes the cardinality of the set I. The identity element of Matn(A) is denoted 1n, and the group Matn(A)× by GLn(A). The group commutator of elements g, h in a group G is ((g, h)) = ghg−1h−1. Acknowledgement. The author thanks Ottmar Loos for many helpful comments on an earlier version of this paper. 1. Elementary linear groups and their Steinberg groups In this section we give an introduction to elementary linear groups over a ring A (1.1) and their associated Steinberg groups (1.3). After a review of central extensions in 1.4 we state the Kervaire-Milnor-Steinberg Theorem (1.6) which says, for example, that the stable Steinberg group is the universal central extension of the stable elementary group. We also exhibit a new set of generators and relations for the Steinberg groups considered in this section (1.7 -- 1.9), which we take as axioms for a new Steinberg group defined in 1.10. The main result is Theorem 1.11: the classical and the new Steinberg groups are isomorphic. 1.1. Elementary linear groups. Let n ∈ N, n ≥ 2. As usual, Eij ∈ Matn(A) is the n × n-matrix with entry 1A at the position (ij) and 0A elsewhere. For 1 ≤ i 6= j ≤ n and a ∈ A we put The well-known multiplication rules aEij bEkl = δjk abEil for a, b ∈ A imply (E1) eij(a) eij (b) = eij(a + b). eij(a) = 1n + aEij, (a ∈ A) Hence eij(a) eij (−a) = 1n = eij(−a) eij(a), which shows that eij(a) ∈ GLn(A). The ele- mentary linear group (of rank n) is the subgroup of Matn(A)× generated by all eij(a). En(A) =(cid:10)eij(a) : 1 ≤ i 6= j ≤ n, a ∈ A(cid:11) One easily verifies two further relations of the eij(a): (E2) ((eij(a), ekl(b))) = 1n ((eij(a), ejl(b))) = eil(ab) (j 6= k, i 6= l), (E3) Taking the inverse of (E3) and using ((g, h))−1 = ((h, g)) yields the equivalent relation (i, j, l 6=). (E4) ((eij (a), eki(b))) = ekj(−ba) (i, j, k 6=). We will also need an infinite variant of Matn(A) and the group En(A). Let MatN(A) be the set of all N × N-matrices x = (xij)i,j∈N with entries from A. Recall that only finitely many xij 6= 0. The usual addition and multiplication of matrices are well-defined operations on MatN(A) satisfying all axioms of a ring, except the existence of an identity element. To remedy this, let 1N = diag(1A, 1A, . . .) be the diagonal matrix of size N × N with every diagonal entry being 1A. Then MatN(A)ex := k1N + MatN(A) is a ring with the usual addition and multiplication of matrices. Its identity element is 1N and its zero element is the zero matrix, see for example [HO, 1.2B] where this ring is denoted Mat∞(A) (its elements are the N × N-matrices with entries from A which have only finitely many non-zero entries off the diagonal and whose diagonal elements become eventually constant). We associate with x ∈ Matn(A) the matrix ιn(x) ∈ MatN(A)ex by putting x in the upper left corner and filling the diagonal outside x with 1A: 3 Then ιn maps invertible matrices in Matn(A) to invertible matrices of MatN(A)ex, in par- 0 0 diag(1A, . . .)(cid:19) ιn(x) =(cid:18)x ex. Since ιn(cid:0)eij(a)(cid:1) = ιp(cid:0)eij(a)(cid:1) for p ≥ n, we can take the ticular ιn(cid:0)eij(a)(cid:1) ∈ MatN(A)× maps ιn as identification and view all eij(a), i, j ∈ N with i 6= j, as elements of MatN(A)× ex. The (stable) elementary linear group is the subgroup E(A) of MatN(A)× ex generated by all the eij(a): It is immediate that the relations (E1) -- (E4) also hold in E(A). The group E(A) is canonically isomorphic to the limit of the inductive system (En(A), ιpn) where ιpn : En(A) → Ep(A) for p ≥ n is defined by taking the left upper (p × p)-corner of ιn(x). E(A) =(cid:10)eij(a) : i, j ∈ N, i 6= j(cid:11). 1.2. Why is En(A) important? One reason is that En(F ) = SLn(F ) in case of A = F is a field -- in other words, every matrix of determinant 1 can be reduced to the identity matrix by elementary row and column reductions. The equality En(A) = SLn(A) even holds for a noncommutative local ring, for example a division ring, if one uses the Dieudonn´e determinant ([HO, 2.2.2] or [Ro, 2.2.2 -- 2.2.6]). If A is commutative then obviously En(A) ⊂ SLn(A). Equality holds for example if A is a Euclidean ring [HO, 1.2.11]. While all of this is interesting, the real interest in En(A) and E(A) stems from their connection to Steinberg groups of A and to the K-group K2(A), defined in (1.3.2). 1.3. The Steinberg groups Stn(A) and St(A). We assume n ∈ N, n ≥ 3 (the case n = 2 is uninteresting since then the definitions below yield free products of A. The group St2(A) is defined differently, see e.g. [Mi]; it will not play a role in this paper). The Steinberg group Stn(A) is the group presented by • generators xij(a), 1 ≤ i 6= j ≤ n and a ∈ A, and • relations (E1) -- (E3) of §1.1: xij(a) xij (b) = xij(a + b) for all 1 ≤ i 6= j ≤ n and a, b ∈ A, ((xij(a), xkl(b))) = 1 if j 6= k and l 6= i, ((xij(a), xjl(b))) = xil(ab) if i, j, l 6=. The (stable) Steinberg group St(A) is the group presented by • generators xij(a), i, j ∈ N distinct, a ∈ A, and • relations (E1) -- (E3) for i, j ∈ N. Since the defining relations (E1) -- (E3) hold in En(A) and E(A) we get surjective group homomorphisms (1.3.1) ℘n : Stn(A) → En(A) and ℘ : St(A) → E(A) determined by xij(a) 7→ eij(a). The second K-group of A is then defined as (1.3.2) K2(A) := Ker(℘). 4 E. NEHER This is an important but also mysterious group, even for fields. The reader can find more about this group in the classic [Mi], and in [HO, Ch. 1], [Ma, Parts IV and V], [Ro, Ch. 4], or [We, III] (the list is incomplete). To put all of this in a bigger picture, we make a short detour on central extensions of groups. 1.4. Central extensions. Let G be a group. An extension of G is a surjective group homomorphism p : E → G. An extension is called central if Ker(p) is contained in the centre of the group G. A central extension q : X → G is a universal central extension if for all central extensions p : E → G there exists a unique homomorphism f : X → E such that q = p ◦ f : X ∃! f ❅❅❅❅❅❅❅❅ q G E ⑦⑦⑦⑦⑦⑦⑦⑦ p A group X is called centrally closed if IdX : X → X is a universal central extension. Thus, X is centrally closed if and only if every central extension p : E → X splits uniquely in the sense that there exists a unique group homomorphism f : X → E satisfying p ◦ f = IdX . The concepts defined above are related by the following facts, proved for example in [HO, 1.4C], [Mi, §5], [Ro, 4.1] or [St2, §7]. (a) A group G has a universal central extension if and only if it is perfect, i.e., generated by all commutator ((g, h)) of g, h ∈ G. In particular, a centrally closed group is perfect. (b) For two universal central extensions of a group G, say q : X → G and q′ : X ′ → G, there exists an isomorphism f : X → X ′ of groups, uniquely determined by the condition q = q′ ◦ f . (c) Let q : X → G be a universal central extension, whence G is perfect by (a). Then X is centrally closed and thus also perfect, again by (a). Obviously, G is a central quotient of X. The following fact (d) says that every universal central extension of G is obtained as a central quotient of a centrally closed group. (d) A surjective group homomorphism q : X → G is a universal central extension if and only if (i) X is centrally closed and (ii) Ker(q) is central. (e) Let f : X → G and g : G → ¯G be central extensions. Then f is a universal central extension if and only if g ◦ f is a universal central extension. To describe a universal central extension of a group G we have, by (d) and (e), two ap- proaches: (I) Find successive central extensions G1 → G0 = G, . . . , Gn → Gn−1, . . . until one of them, say Gn → Gn−1, becomes universal, and then take the composition Gn → G of these central extensions, or (II) find an extension q : X → G with X centrally closed and then find conditions for Ker(q) to be central. Although (I) seems to be the more natural approach, we will actually follow (II). But first back to Steinberg groups. In [St1] Steinberg gave a very elegant description of the universal central extension of a perfect Chevalley group over a field. "Most" Chevalley groups are perfect by [St2, Lemma 32]. In particular, for n ≥ 2 and F a field, the group SLn(F ) = En(F ) (equality by 1.2) is a Chevalley group, and it is perfect if n ≥ 4 or if n = 3 and F ≥ 3 or if n = 2 and F ≥ 4. A special case of Steinberg's result in [St1] is the following theorem. / / 5 1.5. Theorem ([St1], [St2, Thm. 10], [St3, Thm. 1.1]). Let n ∈ N, n ≥ 2 and let F be a field satisfying F > 4 if n = 3 and F 6∈ {2, 3, 4, 9} if n = 2. Then the map ℘n : Stn(F ) → En(F ) of (1.3.1) is a universal central extension. We have defined the maps ℘n : Stn(A) → En(A) and ℘ : St(A) → E(A) for any ring A. It is therefore natural to ask if Theorem 1.5 even holds for rings. An answer is given in the following Kervaire-Milnor-Steinberg Theorem. 1.6. Kervaire-Milnor-Steinberg Theorem ([Ke, Mi, St2]). For an arbitrary ring A, (a) the group Stn(A), n ≥ 5, is centrally closed. (b) The map ℘ : St(A) → E(A) is a universal central extension. An indication of the proof of (a) can be found in see [HO, 1.4.12] or [St2, Cor. 1]. The attribution of part (b) of this theorem is somewhat complicated. Kervaire cites a preliminary version of [Mi], Milnor attributes it to Steinberg and Kervaire ([Mi, p. 43]), and Steinberg says that (b), proved in [St2, Thm. 14], is "based in part on a letter from J. Milnor". In view of (a), the map ℘n : Stn(A) → En(A), n ≥ 5, is a universal central extension if and only if it is a central extension. It is known that this is not always the case, see [HO, 4.2.20]. However, by 1.4(c), both Stn(A) and St(A) are centrally closed. It is this result that we will be concentrating on, following the strategy 1.4(II). 1.7. Another look at Stn(A) and St(A): using root systems. To treat Stn(A), n ∈ N, n ≥ 3, and St(A) at the same time we use the subset N ⊂ N, which is the finite integer interval N = [1, n] or N = N. We can then put StN (A) =(Stn(A) St(A) if N = [1, n], if N = N. By definition in 1.3, StN (A) is generated by xij(a), (i, j) ∈ N × N , i 6= j, and a ∈ A. We replace this indexing set by the root system AN (notation of 2.2), which we realize in Rεi with basis (εi)i∈N and inner product ( ) defined by the Euclidean space X = Li∈N (εiεj) = δij: R = AN = {εi − εj : i, j ∈ N }, R× = R \ {0}. Thus R× = An−1 for N = [1, n] in the traditional notation, while for N = N we get an infinite locally finite root system -- a concept that we will review later in 2.1. For the moment, it suffices to use the concretely given R above. For α, β ∈ R one easily checks that (αβ) ∈ {0, ±1, ±2} with (αβ) = ±2 ⇐⇒ α = ±β. To conveniently describe the remaining cases we use the symbols (1.7.1) α ⊥ β ⇐⇒ (αβ) = 0 and α β ⇐⇒ (αβ) = 1. A straightforward analysis of the indices shows for α = εi − εj and β = εk − εl ∈ R that α ⊥ β or α β ⇐⇒ j 6= k and l 6= i, (1.7.2) α (−β) ⇐⇒ j = k, i, j, l 6= or i = l, i, j, k 6= . Hence, putting xα(a) = xij(a) for α = εi − εj ∈ R×, 6 E. NEHER the relations (E1) -- (E4) can be rewritten in terms of roots as follows. Let α, β ∈ R× and denote the relations corresponding to (Ei) by (ERi). Then the previous relations read (ER1) (ER2) (ER3) (ER4) ((xα(a), xα(b))) = 1, ((xα(a), xβ(b))) = 1, ((xα(a), xβ(b))) = xα+β(ab) ((xα(a), xβ(b))) = xα+β(−ba) β, if α ⊥ β or α if α = εi − εj, β = εj − εl, i, j, l 6=, if α = εi − εj, β = εk − εi, i, j, k 6=. In particular, the two cases for α (ER3) and (ER4). (−β) in (1.7.2) correspond precisely to the relations 1.8. Another look at StN (A): fewer generators. We continue with N and R as in 1.7. In addition we choose a nontrivial partition which we fix in the following. It induces a non-trivial partition N = I ∪ J, ∅ 6= I 6= N, (1.8.1) whose parts are R = R1 ∪ R0 ∪ R−1, R1 = {εi − εj : i ∈ I, j ∈ J}, R−1 = {εj − εi : i ∈ I, j ∈ J} = −R1, R0 = {εk − εl : (k, l) ∈ I × I or (k, l) ∈ J × J} = AI × AJ . The partition R = R1 ∪ R0 ∪ R−1 will later be seen to be an example of a 3-grading of R, but we do not need this now. Observe that every µ ∈ R0 can be written (not uniquely) as µ = α − β with α and β ∈ R1 satisfying α β. Indeed, (i) if µ = εk − εl with k, l ∈ I then µ = (εk − εj) − (εl − εj) for any j ∈ J, hence (1.8.2) xµ(a) = xkl(a) = ((xkj(a), xjl(1))) = ((xα(a), x−β(1))) by (ER3) for α = εk − εj and β = εl − εj ∈ R1, and (ii) if µ = εk − εl with k, l ∈ J then µ = (εi − εl) − (εi − εk) for any i ∈ J, hence (1.8.3) xµ(a) = xkl(a) = ((xil(−a), xki(1))) = ((xα(−a), x−β(1))) by (ER4) for α = εi − εl and β = εi − εk ∈ R1. The equations (1.8.2) and (1.8.3) show that StN (A) is already generated by (1.8.4) {xα(a) : α ∈ R1 ∪ R−1, a ∈ A}. 1.9. Another look at StN (A): fewer relations. Our next goal is to rewrite some of the relations (ER1) -- (ER4) in terms of the smaller generating set (1.8.4). Each of these relations depend on two roots ξ, τ ∈ R. Because of ((g, h))−1 = ((h, g)) we only need to consider the relations involving (ξ, τ ) lying in one of the following subsets of R × R: R1 × R1, R−1 × R−1, R1 × R−1, R0 × R1, R0 × R−1, R0 × R0. (a) Case (ξ, τ ) = (α, β) ∈ R1 × R1: Given α, β ∈ R1, exactly one of the relations α = β, α β, α ⊥ β holds. Hence only (ER1) and (ER2) apply in this case and yield (1.9.1) ((xα(a), xβ(b))) = 1 for α, β ∈ R1 and a, b ∈ A. It will now be more convenient to change notation (again) and put for α = εi − εj ∈ R1 and uα = aE+ α 7 (1.9.2) (1.9.3) E+ α = Eij, α = AE+ V + α , Because of (1.9.1) the map (1.9.4) x′ + : V + −→ StN (A), x′ +(uα) = xα(a) = xij(a) α =L(ij)∈I×J AEij. V + =Lα∈R1 V + u =Pα∈R1 uα 7→ Qα∈R1 x′ +(uα) is well-defined (independent of the chosen order for Qα∈R1 ) and satisfies +(u′) It is clear that conversely (1.9.5) implies (1.9.1). x′ +(u + u′) = x′ for u, u′ ∈ V +. +(u) x′ (1.9.5) (b) Case (ξ, τ ) = (−α, −β) ∈ R−1 × R−1: This case is analogous to Case (a). Given α = εi − εj ∈ R1 and vα = aE− α we define (1.9.6) E− α = Eji, α = AE− V − α , x−(vα) = x−α(−a) = xji(−a), V − =Lα∈R1 V − α =L(ji)∈J×I AEji. (The minus sign in the definition of x−(vα) is not significant, but has been included so that the relations below are precisely those used later on. It avoids a minus sign in the formula (1.12.3).) As in Case (a) the relations (ER1) -- (ER4) for (−α, −β) ∈ R−1 × R−1 yield ((x′ −(v′β))) = 1 and thus give rise to a well-defined map −(vα), x′ (1.9.7) satisfying (1.9.8) − : V − −→ StN (A), x′ v =Pα∈R1 aαE− α 7→ Qα∈R1 x′ −(−aαE− α ) x′ −(v + v′) = x′ −(v) x′ −(v′) for v, v′ ∈ V −. At this point we obtain a new generating set of StN (A), (1.9.9) StN (A) =(cid:10)x′ +(V +) ∪ x′ −(V −)(cid:11), (c) Case (ξ, τ ) = (α, −β) ∈ R1 × R−1: From this case we will only explicitly keep the relation (ER2), which in our new notation says (1.9.10) ((x′ +(u), x′ −(v))) = 1 for (u, v) ∈ V + α × V − β with α ⊥ β. In the following Case (d) we use the relations (ER3) and (ER4) for (α, −β) ∈ R1 × R−1 in double commutators. (d) Case (ξ, τ ) = (µ, γ) ∈ R0 × R1: To deal with this case, we view the elements of V + as I × J-matrices with only finitely many non-zero entries, as in (1.9.2). Similarly, elements in V − are J × I-matrices with finitely many non-zero entries. Matrix multiplication of matrices in V + × V − × V + is then well-defined and yields the Jordan triple product {· · · }, i.e., the map {· · · } : V + × V − × V + −→ V +, (x, y, x) 7→ {x y z} := xyz + zyx. We claim that (ER2) -- (ER4) imply −(vβ))), x′ ((((x′ (1.9.11) +(uα), x′ for α, β, γ ∈ R1 with α +(zγ))) = x′ +(−{uα vβ zγ}) β and all uα ∈ V + α , vβ ∈ V − β , zγ ∈ V + γ 8 E. NEHER We prove this by evaluating all possibilities for µ = α − β with α, β ∈ R1 satisfying α β and γ = εr − εs ∈ R1. By 1.8(i) and 1.8(ii) there are two cases for such a representation of µ, discussed below as (I) and (II). (I) α = εi − εj, β = εi − εl for i ∈ I and j, l ∈ J distinct. Thus µ = α − β = εl − εj. We let uα = aEij, vβ = bEli, zγ = cErs. Then, by (ER4) -- (E4), +(uα), x′ ((((x′ {uα vβ zγ} = {aEij bEli cErs} = δsl cbaErj =: B. −(vβ))), x′ +(zγ))) = ((((xij(a), xli(−b))), xrs(c))) = ((xlj(ba), xrs(c))) =: A, If l = s then, again by (E4), A = xrj(−cba), while B = cba Erj. Otherwise l 6= s, whence A = 1 by (E2) and clearly B = 0. This finishes the proof of (1.9.11) in case (I). (II) α = εi − εj, β = εk − εj for distinct i, k ∈ I and j ∈ J. This can be shown in the same way as (I). To obtain a slightly simpler version of (1.9.11) we apply the commutator formula ((g, h1 h2)) = ((g, h1)) · ((g, h2)) · ((((h2, g)), h1)) with g = ((x′ ((g, h1 h2)) = ((g, h1)) · ((g, h2)), which allows us to rewrite (1.9.11) in the form +(zγ) and h2 = x′ −(vβ))), h1 = x′ +(uα), x′ +(zδ) for arbitrary δ ∈ R1. We obtain ((((x′ (1.9.12) −(vβ))), x′ +(uα), x′ for α, β ∈ R1 with α +(z))) = x′ +(−{uα vβ z}) β and arbitrary uα ∈ V + α , vβ ∈ V − β , z ∈ V +. (e) Case (ξ, τ ) = (µ, −γ) ∈ R0 × R−1. We proceed as in Case (d) and define the Jordan triple product {· · · } : V − × V + × V − −→ V −, (x, y, x) 7→ {x y z} := xyz + zyx, using matrix multiplication in the definition of {· · · }. As in Case (d) one then proves the relation (1.9.13) ((((x′ +(uβ))), x′ −(vα), x′ for α, β ∈ R1 with α −(w))) = x′ −(−{vα uβ w}) β and arbitrary vα ∈ V − α , uβ ∈ V + β , w ∈ V −. (f) Case (ξ, τ ) ∈ R0 × R0: As we will see below, the relations involving these (ξ, τ ) are not needed for presenting StN (A). 1.10. The Steinberg group St(MIJ (A), R). We keep the setting of (1.7) -- (1.9). In 1.9 we defined a pair of matrix spaces, and Jordan triple products (V +, V −) =(cid:0)MatIJ (A), MatJ I(A)(cid:1) =: MIJ (A), {· · · } : V σ × V −σ × V σ → V σ, (x, y, z) 7→ {x y z} = xyz + zyx for σ ∈ {+, −}. In (1.9.3) and (1.9.6) we also introduced a family R = (Vα)α∈R1 , Vα = (V + α , V − α ) of pairs of subgroups with the property that V σ = Lα∈R1 V σ α . Furthermore, in (1.9.9) we found a new generating set for StN (A), and we rewrote some of the relations defining StN (A) in terms of this new generating set. It is then natural to define a new Steinberg group using these new generators and relations. The Steinberg group St(MIJ (A), R) is the group presented by • generators x+(u), u ∈ V +, and x−(v), v ∈ V −, and • the relations (1.9.5), (1.9.8), (1.9.10), (1.9.12) and (1.9.13). Taking σ ∈ {+, −} 9 these are (EJ1) (EJ2) (EJ3) xσ(u + u′) = xσ(u) xσ(u′) ((x+(u), x−(v))) = 1 ((((xσ(u), x−σ(v))), x−(z))) = xσ(−{u v z}) for u, u′ ∈ V σ, α × V − for (u, v) ∈ V + β , α ⊥ β, for uα ∈ V σ α , v ∈ V −σ β , z ∈ V σ with α β. (The letter "J" in (EJi) stands for "Jordan", to be explained in the next section.) From the review above, it is clear that we have a surjective homomorphism of groups Φ : St(MIJ (A), R) → StN (A), (1.10.1) where x′ σ is defined in (1.9.4) and (1.9.7). Moreover, composing Φ with the surjective group homomorphisms ℘N : StN (A) → EN (A) of (1.3.1) yields another surjective group homomorphism σ(u) xσ(u) 7→ x′ (1.10.2) pN : St(MIJ (A), R) → EN (A), x+(aEij) 7→ eij(a), x−(aEji) 7→ eji(−a) and hence a commutative diagram (1.10.3) St(MIJ (A), R) '❖❖❖❖❖❖❖❖❖❖❖ pN Φ StN (A) yttttttttt ℘N EN (A) 1.11. Theorem ([LN2]). The map Φ of (1.10.1) is an isomorphism of groups. In particular, St(MIJ (A), R) is centrally closed if N ≥ 5 and pN is a universal central extension of E(A) if N = N. Proof. To prove bijectivity of Φ is bijective, a canonical approach is to show that the family of x+(aEij) and x−(−bEij) ∈ St(MIJ (A), R) can be extended to a family of elements satisfying the defining relations (E1) -- (E3) of StN (A). As a consequence, this yields a group homomorphism Ψ : StN (A) → St(MIJ (A), R) such that Ψ ◦ Φ and Φ ◦ Ψ are the identity on the respective generators and therefore also on the corresponding groups. Another proof of the bijectivity of Φ is given in [LN2, 24.18], based on the interpretation of both groups as initial objects in an appropriate category of groups mapping onto EN (A). The second part of the theorem follows from the Kervaire-Milnor-Steinberg Theorem 1.6. (cid:3) 1.12. Another look at EN (A). It follows from the existence of the surjective group ho- momorphism pN of (1.10.2) that EN (A) is generated by pN (x+(V +) ∪ x−(V −)) and that the matrices in this image satisfy the relations (EJ1) -- (EJ3). It is instructive to verify this directly. For (u, v) ∈ MIJ (A) we define elements e+(u) and e−(v) of the ring MatN (A)ex of 1.1 by (1.12.1) Then clearly e+(u) =(cid:18)1I 0 u 1J(cid:19) , e−(v) =(cid:18) 1I −v 1J(cid:19) . 0 (1.12.2) e+(u + u′) = e+(u) e+(u′) and e−(v + v′) = e−(v) e−(v′). / / ' y 10 E. NEHER In particular, the matrices e+(u) and e−(v) are invertible with inverses e+(u)−1 = e+(−u) and e−(v)−1 = e−(−v). Since e+(aEij) = eij(a) and e−(vEji) = eji(−v) for (ij) ∈ I ×J, the equations (1.12.2) also show that e+(u) ∈ EN (A) and e−(v) ∈ EN (A). By straightforward matrix multiplication one obtains (1.12.3) In particular, taking (u, v) with vu = 0 or uv = 0, this proves vuv uvu ((e+(u), e−(v))) =(cid:18)1I − uv + uvuv and (cid:18)1I 1J(cid:19) ∈ EN (A) 1J + vu(cid:19) . 1J + vu(cid:19) ∈ EN (A). 0 0 0 (cid:18)1I − uv 0 Specifying (u, v) even more, one then easily sees that all elementary matrices ekl(a) with (k, l) ∈ I × I or (k, l) ∈ J × J lie in the subgroup of EN (A) generated by e+(V +) ∪ e−(V −). Therefore, this subgroup equals EN (A). α × V − The relation (EJ1) is (1.12.2), and the relation (EJ2) follows from (1.12.3) since for α with α ⊥ β we have uv = 0 and vu = 0. In order to prove (EJ3) in case β with α ⊥ β and let z ∈ V + arbitrary. Then uvu = 0 = uvzvu, (u, v) ∈ V + σ = +, let (u, v) ∈ V + vuv = 0 and (1J + vu)−1 = 1J − vu. Hence, by (1.12.3), α × V − ((((e+(u), e−(v))), e+(z))) =(cid:16)(cid:16)(cid:18)1I − uv 0 0 (1J − vu)−1(cid:19) , (cid:18)1I =(cid:18)1I −z + (1 − uv)z(1 − vu) 1J 0 0 z 1J(cid:19)(cid:17)(cid:17) (cid:19) = e+(−{u v z}). The relation (EJ3) for σ = − can be verified in the same way. To put this example in the general setting of the following section §2 we point out that the calculations above are not only valid for matrices of finite or countable size N , but hold for N of arbitrary cardinality. 2. Generalizations In this section we generalize the Steinberg groups considered in §1. The generalization has a combinatorial aspect, 3-graded root systems, and an algebraic aspect, root graded Jordan pairs. They are presented in 2.1 -- 2.3 and 2.4 -- 2.6 respectively. We define the Steinberg group of a root graded Jordan pair (2.7) and state the Jordan pair version of the Kervaire-Milnor-Steinberg Theorem (2.8 and 2.11). Since the elementary linear group only makes sense for special Jordan pairs, we replac it by its central quotient which can be defined for any Jordan pair: the projective elementary group PE(V ) of a Jordan pair V defined in terms of the Tits-Kantor-Koecher algebra L(V ) (2.10). We discuss L(V ) and PE(V ) for the Jordan pair of rectangular matrices over a ring in 2.12 and 2.13. 2.1. Locally finite root systems [LN1]. We use h·, ·i to denote the canonical pairing between a real vector space X of arbitrary dimension and its dual space X ∗, thus hx, ϕi = ϕ(x) for x ∈ X and ϕ ∈ X ∗. If ϕ ∈ X ∗ satisfies hα, ϕi = 2, we define the reflection sα,ϕ ∈ GL(X) by sα,ϕ(x) = x − hx, ϕiα. A locally finite root system is a pair (R, X) consisting of a real vector space X and a subset R ⊂ X satisfying the axioms (i) -- (iv) below. (i) R spans X as a real vector space and 0 ∈ R, (ii) for every α ∈ R× = R \ {0} there exists α∨ ∈ X ∗ such that hα, α∨i = 2 and sα,α∨(R) = R. 11 (iii) hα, β ∨i ∈ Z for all α, β ∈ R×. (iv) R is locally finite in the sense that R ∩ Y is finite for every finite-dimensional subspace of X. Locally finite root systems form a category RS, in which a morphism f : (R, X) → (S, Y ) is an R-linear map with f (R) ⊂ S. In this category, an isomorphism f : (R, X) → (S, Y ) is a vector space isomorphism f : X → Y with f (R) = S. Such an isomorphism necessarily satisfies hf (α), f (β)∨i = hα, β ∨i for all α, β ∈ R×. Remarks, facts and more definitions. (a) The linear form α∨ in (ii) is uniquely determined by the two conditions in (ii). Therefore, we simply write sα instead of sα,α∨ in the future. (b) Our standard reference for locally finite root systems is [LN1]. As in [LN1] we will also here abbreviate the term "locally finite root system" by root system. Then a finite root system is a root system (R, X) with R a finite set, equivalently dim X < ∞. Finite root systems are the root systems studied for example in [Bo, Ch. VI]. That [Bo] assumes 0 /∈ R does not pose any problem in applying the results developed there. The real vector space X of a root system (R, X) is usually not important. We will therefore often just refer to R rather than to (R, X) as a root system. (c) As in [LN1] and again in [LN2, §2] we assume here that 0 ∈ R, which is more natural from a categorical point of view. In [LN2, §2] the real vector space X is replaced by a free abelian group X and condition (iv) becomes that R ∩ Y be finite for every finitely generated subgroup Y of X. With the obvious concept of a morphism, this defines a category of root systems over the integers, which is equivalent to the category RS [LN2, Prop. 2.9]. (d) A locally finite root system need not be reduced in the sense that Rα ∩ R = {±α} for every α ∈ R×. The rank of a root system (R, X) is defined as the dimension of the real vector space X. (e) The direct sum of a family (R(j), X (j))j∈J of root systems is the pair (cid:0)Sj∈J R(j), Lj∈J X (j)(cid:1), which is again a root system [LN1, 3.10], traditionally written as R = Lj∈J R(j). A non- empty root system is called irreducible if it is not isomorphic to a direct sum of two non- empty root systems. Every root system uniquely decomposes as a direct sum of irreducible root systems, called its irreducible components [LN1, 3.13]. (f) Every root system (R, X) admits an inner product ( ) : X ×X → R, which is invariant in the sense that (sα(x) sα(y)) = (x y) holds for all α ∈ R× and x, y ∈ X, equivalently (2.1.1) hβ, α∨i = 2 (β α) (α α) for all α, β ∈ R× [LN1, 4.2]. If R is irreducible, ( ) is unique up to a non-zero scalar. It follows that the definition of a root system given in [Ne2] is equivalent to the definition above, and that a finite reduced root system is the same as a "root system" in [Hu], again up to 0 /∈ R. 2.2. Classification of root systems. We first present, as examples, the classical root Rεi be the systems AI , . . . , BCI . Let I be a set of cardinality I ≥ 2 and let X = Li∈I 12 E. NEHER R-vector space with basis (εi)i∈I . Define (2.2.1) (2.2.2) (2.2.3) (2.2.4) (2.2.5) AI = {εi − εj : i, j ∈ I}, DI = AI ∪ {±(εi + εj) : i 6= j}, BI = DI ∪ {±εi : i ∈ I}, CI = DI ∪ {±2εi : i ∈ I}, BCI = BI ∪ CI. Then AI is a root system in X = Ker(t) where t ∈ X ∗ is defined by t(εi) = 1, i ∈ I. Its rank is therefore I − 1. The notation A instead of the traditional A is meant to indicate this fact. Observe that AN is the root system R of 1.7. All other sets DI , . . . , BCI, are root systems in X, whence of rank I. The root systems AI , BI , CI and DI are reduced, while BCI is not. The isomorphism class of a classical root system only depends on the cardinality of the set I. In particular, when I is finite of cardinality n we will use the index n instead of I. Thus, Dn = D{1,...,n} etc. Note An+1 = A{0,1,...,n} = An in the traditional notation. The standard inner product ( ), defined by (εiεj) = δij, is an invariant inner product in the sense of 2.1(f). With the exception of D2 = A1 ⊕ A1, the root systems AI , . . . , BCI are ∼= A3, they are pairwise irreducible. Apart from the low-rank isomorphisms B2 non-isomorphic. ∼= C2, D3 The classification of root systems [LN1, Thm. 8.4] says that an irreducible root system is either isomorphic to a classical root system or to an exceptional finite root system. 2.3. 3-graded root systems. A 3-grading of a root system (R, X) is a partition R = R1 ∪ R0 ∪ R−1 satisfying the following conditions (i) -- (iii) below: (i) R−1 = −R1; (ii) (Ri + Rj) ∩ R ⊂ Ri+j for i, j ∈ {1, 0, −1}, with the understanding that Rk = ∅ for k /∈ {1, 0, −1} (iii) R1 + R−1 = R0, i.e., every root in R0 is a difference of two roots in R1. In particular (ii) says that the sum of two roots in R1 is never a root and (R1+R−1)∩R ⊂ R0, a condition which is strengthened in (iii). Since a 3-grading of a root system (R, X) is determined by the subset R1 of R, we will denote a 3-graded root system by (R, R1, X) or simply by (R, R1). A 3-graded root system is a root system equipped with a 3-grading. An isomorphism f : (R, R1, X) → (S, S1, Y ) between 3-graded root systems is a vector space isomorphism f : X → Y satisfying f (R1) = S1, hence also f (Ri) = Si for i ∈ {1, 0, −1}, and is therefore an isomorphism f : (R, X) → (S, Y ) of the underlying root systems. References for 3-graded root systems are [LN1, §17, §18], [LN2, Ch. IV] and [Ne2]. Some facts and examples. (a) The decomposition (1.8.1) of the root system R = AN is a 3-grading. The restrictions on N imposed in 1.7 are not necessary for the definition of a 3-graded root system, as we have seen in 2.2 for the root system AN . Any non-trivial partition N = I ∪ J induces a 3-grading of AN as in 1.8, denoted AI N . Thus, the 1-part of the 3-graded root system AI N is ( AI (2.3.1) N )1 = {εi − εj : i ∈ I, j ∈ N \ I}. Every 3-grading of AN is obtained in this way for a non-empty proper subset I ⊂ N . 13 (b) A 3-grading Bqf I of the root system BI is obtained by choosing a distinguished element of I, say 0 ∈ I, and putting R1 = {ε0} ∪ {ε0 ± εi : 0 6= i ∈ I}. (c) The root system R = CI has a 3-grading, denoted Cher , whose 1-part is R1 = {εi +εj : I i, j ∈ I}. Note R0 = {εi − εj : i, j ∈ I} ∼= AI . (d) The root system DI , I ≥ 4, is a subsystem of BI and CI . The 3-gradings of these two root systems, defined in (b) and (c), induce 3-gradings Dqf I of DI . The first of these is determined by R1 = {ε0 ± εi : 0 6= i ∈ I} and the second by R1 = {εi + εj : i, j ∈ I, i 6= j}. It is known that Dqf I if I = 4, but Dqf I I and Dalt if I ≥ 5. ∼= Dalt I 6∼= Dalt I (e) The 3-gradings of a root system R are determined by the 3-gradings of its irreducible components (R(j))j∈J as follows. If (R, R1) is a 3-grading, then (R(j), R1 ∩ R(j)) is a 3-grading for every j ∈ J. Conversely, given 3-gradings (R(j), R(j) 1 ) for every j, the set R1 =Sj R(j) 1 defines a 3-grading of R. These easy observations reduce the classification of 3-graded root systems to the case of irreducible root systems. Their classification is given in [LN1, 17.8, 17.9]. It turns out that an irreducible root system has a 3-grading if and only if it is not isomorphic to E8, F4 and G2. Some irreducible root systems have several non-isomorphic 3-gradings, such as AN or DI . But every 3-grading of CI is isomorphic to the 3-grading Cher of (c). I (f) The relations ⊥ and introduced in (1.7.1) in case R = AN can be defined for any root system R without using an invariant inner product. For α, β ∈ R× we put α ⊥ β ⇐⇒ hα, β ∨i = 0, equivalently hβ, α∨i = 0, α β ⇐⇒ hα, β ∨i = 1 = hβ, α∨i α → β ⇐⇒ hα, β ∨i = 2, hβ, α∨i = 1. The formula (2.1.1) shows that the definitions of ⊥ and above generalize (1.7.1). The relation → occurs for example in the root system CI : we have 2εi → εi + εj for 6= j. (In [LN1] the notation ⊤ and ⊣ is used in place of and →, respectively.) (g) Given α, β ∈ R1, exactly one of these relations holds: (2.3.2) α = β or α → β or α ← β or α β or α ⊥ β. Moreover, again for α, β ∈ R1, (2.3.3) 2α − β ∈ R ⇐⇒ 2α − β ∈ R1 ⇐⇒ α ← β or α = β. 2.4. Jordan pairs. This subsection contains a very short introduction to Jordan pairs over a commutative ring k, although for the purpose of this paper k = Z is completely sufficient. We will only present what is needed to understand this paper. A more detailed but still concise introduction to Jordan pairs is given in [LN2, §6]; the standard reference for Jordan pairs is [Lo1]. We have already seen an example of a Jordan pair in 1.9: the rectangular matrix pair MIJ (A) = (MatIJ (A), MatJ I (A)) equipped with the Jordan triple product {x y z} = xyz + zyx for (x, y, z) ∈ V σ × V −σ × V σ and σ = ±. The Jordan triple product is the linearization with respect to x of the expression Q(x)y = xyx, which did not play any role in §1, but which is the basic structure underlying Jordan pairs. A Jordan pair is a pair V = (V +, V −) of k-modules together with maps Qσ : V σ × V −σ → V σ, (x, y) 7→ Qσ(x)y, (σ = ±), which are quadratic in x and linear in y and which satisfy the identities (JP1) -- (JP3) below in all base ring extensions. To define these identities, we will simplify the notation and omit 14 E. NEHER σ, thus writing Q(x)y or simply Qxy. This does not lead to any confusion, as long as one takes care that the expressions make sense. Linearizing Qxy in x gives Qx,zy = Q(x, z)y = Qx+zy − Qxy − Qzy, which we use to define the Jordan triple product {· · · } : V σ × V −σ × V σ → V σ, (x, y, z) 7→ {x y z} = Qx,zy. To improve readability we will sometimes write {x, y, z} instead of {x y z}. If K is a commutative associative unital k-algebra we let V σ K = V σ ⊗k K and observe that there exist unique extensions of the Qσ to quadratic-linear maps Q : VK × VK → VK. The identities required to hold for x, z ∈ V σ K , σ ∈ {+, −} and any K as above are K, y, v ∈ V −σ (JP1) (JP2) (JP3) {x, y, Qxv} = Qx {y, x, v}, {Qxy, y, z} = {x, Qyx, z}, QQxyv = Qx Qy Qzv. A homomorphism f : V → W of Jordan pairs is a pair f = (f+, f−) of k-linear maps fσ : V σ → W σ satisfying fσ(Q(x)y) = Q(cid:0)fσ(x)(cid:1)f−σ(y) for all (x, y) ∈ V σ × V −σ and σ = ±. Remarks and more definitions. (a) Instead of requiring that (JP1) -- (JP3) hold for all extensions K, one can demand that (JP1) -- (JP3) as well as all their linearizations hold in V . For example, linearizing the identity (JP1) with respect to x gives the identity {z, y, Qxv} + {x, y, Qx,zv} = Qx,z {y, x, v} + Qx {y, z, v} (b) If V = (V +, V −) is a Jordan pair and S = (S+, S−) is a pair of submodules of V satisfying Q(Sσ)S−σ ⊂ Sσ for σ = ±, then S is a Jordan pair with the induced operations, called a subpair of V . (c) An idempotent in a Jordan pair V is a pair e = (e+, e−) ∈ V satisfying Q(e+)e− = e+ and Q(e−)e+ = e−. An idempotent e gives rise to the Peirce decomposition of V , V σ = V σ 2 (e) ⊕ V σ 1 (e) ⊕ V σ 0 (e), σ = ±, where the Peirce spaces Vi(e) = (V + i (e), V − i (e)), i = 0, 1, 2, are given by V σ 2 (e) = {x ∈ V σ : Q(eσ)Q(e−σ)x = x}, V σ 1 (e) = {x ∈ V σ : {eσ e−σ x} = x}, V σ 0 (e) = {x ∈ V σ : Q(eσ)x = 0 = {eσ e−σ x}}. The Peirce spaces V ± i = V ± i (e) satisfy the multiplication rules (2.4.1) (2.4.2) Q(V σ 2 V −σ j ⊂ V σ i )V −σ 0 V σ} = 0 = {V σ 2i−j, {V σ 0 V −σ 2 V σ}, {V σ i V −σ j l } ⊂ V σ V σ i−j+l, where i, j, l ∈ {0, 1, 2}, with the understanding that V σ the Vi = Vi(e) are subpairs of V . If 2 ∈ k× we have V σ i = 0, 1, 2. m = 0 if m /∈ {0, 1, 2}. In particular, i (e) = {x ∈ V : {eσ e−σ x} = ix} for 15 2.5. Examples of Jordan pairs. We now give concrete examples of Jordan pairs, to illustrate the abstract definition of 2.4. (i) Any associative, not necessarily unital or commutative k-algebra A gives rise to a Jordan pair V = (A, A) with respect to the operations Qxy = xyx. Indeed, since a base ring extension of A is again associative, it suffices to verify the identities in V , where they follows from the following calculations. {x, y, Qxv} = xy(xvx) + (xvx)yx = x(yxv + vxy)x = Qx{y, v, x}, {Qxy, y, z} = (xyx)yz + zy(xyx) = x(yxy)z + z(yxy)x = {x, Qyx, z}, QQxyv = (xyx)v(xyx) = x(y(xvx)y)x = QxQyQxv. An idempotent c of the associative algebra A, defined by c2 = c, induces an associative Peirce decomposition A = A11 ⊕A10 ⊕A01 ⊕A00 with Aij = {a ∈ A : ca = ia, ac = ja}. The pair e = (c, c) is an idempotent of the Jordan pair V whose Peirce spaces are V σ 2 (e) = A11, V σ 1 (e) = A10 ⊕ A01 and V σ 0 (e) = A00. Not only V1(e) but also (A10, A01) and (A01, A10) are subpairs of V . Not every idempotent of V has the form (c, c), c an idempotent of A. For example, if u ∈ A× and c is an idempotent of A, then (uc, cu−1) is an idempotent of V which, however, has the same Peirce spaces as (c, c). (ii) By (b) and (i), any pair (S+, S−) ⊂ (A, A) of k-submodules closed under the operation (x, y) 7→ xyx is also a Jordan pair. Jordan pairs of this form are called special. Their Jordan triple product is (2.5.1) {x y z} = xyz + zyx (x, z ∈ Sσ, y ∈ S−σ). We next describe some important cases of special Jordan pairs. (iii) Let I and J be non-empty sets. Let N = I ∪ J ′ where J ′ is a set disjoint from I and in bijection with J under j 7→ j′, and embed MatIJ (A) into the right upper corner of the associative algebra MatN (A). Similarly, we identify MatJ I(A) with the left lower corner of MatN (A). Then is a subpair of the Jordan pair (cid:0)MatN (A), MatN (A)(cid:1) and is therefore a Jordan pair, as claimed at the beginning of this subsection. If N is finite, MIJ (A) is of type (A10, A01) for A = MatN (A) and the idempotent c = 1I , see (i). MIJ (A) =(cid:0)MatIJ (A), MatJ I(A)(cid:1) (iv) Let a 7→ aJ be an involution of the associative k-algebra A. Then H(A, J) = {a ∈ A : aJ = a} is closed under the Jordan pair product, whence H(A, J) = (H(A, J), H(A, J)) is a Jordan pair. More generally, extend J to an involution of the associative k-algebra MatI (A), I ≥ 2, defined by (xij)J = (xJ ji) and again denoted by J. Then the hermitian matrix pair is a special Jordan pair. In particular, taking A = k, J = Idk and I = {1, . . . , n} we get the symmetric matrix pair Hn(k) = (Hn(k), Hn(k)). HI (A, J) =(cid:0) H(MatI (A), J), H(MatI (A), J)(cid:1) (v) Let AltI (k) be the alternating I × I-matrices over k, where x = (xij) is called alternating if xii = 0 = xij + xji for i, j ∈ I. Then the alternating matrix pair AI (k) = (AltI (k), AltI (k)) is a subpair of MI (k), whence a special Jordan pair. (vi) Let M be a k-module and let q : M → k be a quadratic form with polar form b, defined by b(x, y) = q(x + y) − q(x) − q(y). Then J(M, q) = (M, M ) is a Jordan pair with quadratic operators Qxy = b(x, y)x − q(x)y. 16 E. NEHER (vii) Let J be a unital quadratic Jordan algebra with quadratic operators Ux, x ∈ J ([Ja4, Ja5]). Then (J, J) is a Jordan pair with quadratic maps Qx = Ux. For example, if k is a field, the rectangular matrix pair Mpp(k) is of this form, but Mpq(k) for p 6= q is not. Thus, there are "more" Jordan pairs than Jordan algebras. Let C be an octonion k-algebra, see for example [SV] in case k is a field or [LPR] in general, and let J = H3(C) be the exceptional Jordan algebra of 3 × 3 matrices over C which are hermitian with respect to the standard involution of C. Then (J, J) is a Jordan pair, which is not special in the sense of (ii). Such Jordan pairs are called exceptional. 2.6. Root graded Jordan pairs. Let us first recast the Peirce decomposition 2.4(c) of an idempotent e in a Jordan pair V from the point of view of a grading. We use the 3-graded root system Cher I of 2.3(c) with I = {0, 1}. Its 1-part is R1 = {εi + εj : i, j ∈ {0, 1}} = {2ε1, ε1 + ε0, 2ε0}. Putting V σ α = V σ i+j(e) (α = εi + εj ∈ R1) we have the decomposition V σ =Lα∈R1 V σ (RG1) 2α−β, β ⊂ V σ Q(V σ α V −σ α )V −σ β V σ} = 0 {V σ α which satisfies α V −σ if α ⊥ β. {V σ β V σ γ } ⊂ V σ α−β+γ, (RG2) Here 2α − β and α − β + γ in (RG1) are calculated in X = R · ε0 ⊕ R · ε1 ∼= R2, and 2α−β = 0 if 2α − β /∈ R1 or V σ V σ α−β+γ = 0 if α − β + γ 6∈ R1. We see that, apart from the actual definition of the Peirce spaces, the rules governing the Peirce decomposition can be completely described in terms of R1. The following generalisation is then natural. is a decomposition V σ = Lα∈R1 V σ Given a 3-graded root system (R, R1) and a Jordan pair V , an (R, R1)-grading of V α , σ = ±, satisfying (RG1) and (RG2). We will use R = (Vα)α∈R1 to denote such a grading. A root graded Jordan pair is a Jordan pair equipped with an (R, R1)-grading for some 3-graded root system. In view of (2.3.3) we can make the first inclusion in (RG1) more precise: (2.6.1) Q(V σ α )V −σ β = 0 unless α → β, in which case 2α − β ∈ R1 and Q(V σ α )V −σ β ⊂ V σ 2α−β. We call R an idempotent root grading if there exists a subset ∆ ⊂ R1 and a family (eα)α∈∆ of non-zero idempotents eα ∈ Vα such that the Vβ are given by (2.6.2) Vβ = \α∈∆ Vhβ,α∨i(eα) Observe that (2.6.2) makes sense since hα, β ∨i ∈ {0, 1, 2} by (2.3.2). Neither the idempotents nor the subset ∆ ⊂ R1 are uniquely determined by an idempotent root grading, see for example (iii) below. To avoid some technicalities, we will often assume that R is a fully idempotent root grading, i.e., R is idempotent with respect to a family of idempotents with ∆ = R1. In the terminology of [Ne1] this means that V is covered by the cog (eα)α∈R1 . Examples. (i) Let R = A1 = {α, −α} equipped with the 3-grading defined by R1 = {α}. An (R, R1)-graded Jordan pair is simply a Jordan pair V = (V +, V −) for which V σ = V σ α . This root grading is idempotent if and only if V ∼= (J, J) where J is a unital Jordan algebra. To see sufficiency in case J is a Jordan algebra with identity element 1J , one uses eα = (1J , 1J ) and observes (J, J) = V2(eα). 17 (ii) Let (R, R1) = Cher potent e ∈ V can be viewed as a Cher α = 2ε1. Thus here ∆ = {α}. 2 . We have seen above that the Peirce decomposition of an idem- 2 -grading, which is idempotent with respect to e = eα, (iii) Let (R, R1) = AI I -grading of a Jordan pair V is a decomposition AN N be the 3-graded root system of 2.3(a). Put J = N \ I. An such that for all (ij) and (lm) ∈ I × J and σ = ± we have, defining V(ij) = Vεi−εj for εi − εj ∈ R1, V =L(ij)∈I×J V(ij) Q(V σ (ij) V −σ (ij))V −σ (ij) V σ (ij) ⊂ V σ (lj)} ⊂ V σ (ij), (lj), {V σ {V σ {V σ (ij) V −σ (ij) V −σ (ij) V σ (lj) V σ (im)} ⊂ V σ (lm)} ⊂ V σ (im), (im), and all other types of products vanish. An example of an AN I -graded Jordan pair V is the rectangular matrix pair MIJ (A) of an associative unital k-algebra A 6= 0, see 2.5(iii), with respect to the subpairs V(ij) = (AEij, AEji). This AI N -grading of MIJ (A) is fully idempotent with respect to the family (eα)α∈R1 , eα = (aijEij, a−1 ij Eji), where α = εi − εj and aij ∈ A×. It is also idempotent with respect to the following smaller family: fix i0 ∈ I and j0 ∈ J and consider (eα)α∈∆ where ∆ = {εi − εj0 : i ∈ I} ∪ {εi0 − εj : j ∈ J}. Let a, b ⊂ A be k-submodules with aba ⊂ a and bab ⊂ b. Then S = (MatIJ (a), MatJ I(b)) N -grading from MIJ (A) by putting S(ij) = is a Jordan subpair of MIJ (A). It inherits the AI (aEij , bEji). This AI N -grading of S is in general not idempotent, e.g., if a is a nil ideal. (iv) The hermitian matrix pair V = HI (A, J) of 2.5(iv) has an idempotent Cher -grading. I Indeed, define hij(a) =(aEij + aJ Eji aEii if a ∈ A and i 6= j, if i = j and a ∈ H(A, J). Then V =Lα∈R1 Vα with Vα = Vεi+εj =((cid:0)hij(A), hij (A)(cid:1), (cid:0)hii(H(A, J)), hii(H(A, J))(cid:1), if i 6= j if i = j. is a Cher (eα)α∈R1 for which eα = (hij (1), hij (1)), α = εi + εj. -grading of V . I It is fully idempotent, for example with respect to the family (v) The remaining examples in 2.5 all have idempotent root gradings. The alternating matrix pair AI (k) of 2.5(v) has an idempotent Dalt I -grading ([LN2, 23.24]). The Jordan pair J(M, q) associated with a quadratic form q in 2.5(v) has an idempotent Bqf if q contains I a hyperbolic plane, or even a Dqf I -grading if q is hyperbolic ([LN2, 23.25]). If C is a split octonion algebra in the sense of [SV] or [LPR] the exceptional Jordan pair (H3(C), H3(C)) has an idempotent root grading with R of type E7 ([Ne1, III, §3]). 2.7. The Steinberg group St(V, R). Let (R, R1) be a 3-graded root system and let V be a Jordan pair with a root grading R = (Vα)α∈R1 , not necessarily idempotent. The Steinberg group St(V, R) is the group with the following presentation: 18 E. NEHER • The generators are x+(u), u ∈ V +, and x−(v), v ∈ V −. To formulate the relations, we first introduce, for α 6= β ∈ R1 and (u, v) ∈ V + the element b(u, v) in the free group with the above generators by the equation α × V − β , (2.7.1) x+(u) x−(v) = x−(v + Qvu) b(u, v) x+(u + Quv) • Then the relations are xσ(u + u′) = xσ(u) xσ(u′) ((x+(u), x−(v))) = 1 for u, u′ ∈ V σ, for (u, v) ∈ V + α × V − β , α ⊥ β, (((b(u, v), x+(z))) = x+(−{u v z} + QuQvz), ((b(u, v)−1, x−(y))) = x−(−{v u y} + QvQuy) (St1) (St2) (St3) for all (u, v) ∈ V + α × V − β with α 6= β and all (z, y) ∈ V . Remarks. (a) Let us have a closer look at the element b(u, v) in (2.7.1). By (2.3.2) the β, α → β and α ← β and by (2.6.1), Qvu = 0 unless possibilities for α, β are α ⊥ β, α β → α, and Quv = 0 unless α → β. Therefore, by (St2), (2.7.2) 1, ((x−(−v), x+(u))), x−(−Qvu) ((x−(−v), x+(u))), ((x−(−v), x+(u))) x+(−Quv), if α ⊥ β, β, if α if α → β, if α ← β. b(u, v) =  In general, the factors x−(−Qvu) and x+(−Quv) in the last two cases are 6= 1. The reader may be puzzled by the definition of b(u, v): why not take "b(u, v) = ((x−(−v) , x+(u)))"? We will give a justification for this in 2.11. (b) We claim that (St3) follows from (St2) in case α ⊥ β. Indeed, the left hand sides of the two equations (St3) are 1 because b(u, v) = 1, but also the right hand sides are 1, since, say for σ = +, we have {u v z} = 0 by (RG2) and QuQvz = 0 by (2.7.3) below, : (2.7.3) Q(V σ α )Q(V −σ β )V σ γ 6= 0 =⇒ α = β or α ← β = γ or α β ← γ ⊥ α, which can be shown by repeated application of (2.3.3). (c) Let (V, R) = (MIJ (A), R). Comparing the definition of St(MIJ (A), R) with the one in 1.10, it is clear that the first two relations coincide: (EJ1) = (St1) and (EJ2) = (St2). We claim that the relations (EJ3) coincide with the two relations in (St3). Indeed, since we do not have a relation α ← β in AI N , it follows from (b) and the assumption α 6= β β → St(V, R), in (St3) that we only need to consider the case α v 7→ ((x+(u), x−(v))) is homomorphism of groups, whence, by (2.7.2), β. Then the map V − b(u, v) = ((x−(−v), x+(u))) = ((x+(u), x−(−v)))−1 = ((x+(u), x−(v))). Thus (EJ3) = (St3) for σ = + because QuQxz = 0 by (2.7.3). The equality of the two relations for σ = − can be established in the same way. We can now state the generalization of part (a) of the Kervaire-Milnor-Steinberg Theo- rem 1.6 in the setting of this section. 19 2.8. Theorem A. Let (R, R1) be a 3-graded root system whose irreducible components all have rank ≥ 5, and let V be a Jordan pair with a fully idempotent root grading R. Then the Steinberg group St(V, R) is centrally closed. This theorem is one of the main results of [LN2]; its proof takes up all of Chapter VI of [LN2]. It is shown there in greater generality. First, it also true when R has connected components of rank 4, but not of type D4. Moreover, it is not necessary to assume that the root grading R is fully idempotent. For the irreducible components of type BI or CI , I ≥ 5, one only needs idempotents eα ∈ Vα in case α is a long root in type B and α a short root in type C. This generality allows us to consider groups defined in terms of hermitian matrices associated with form rings in the sense of [Ba]. 2.9. Highlights of our approach. The novel aspect of our approach is the consistent use of the theory of 3-graded root systems and Jordan pairs, which introduces new methods in the theory of elementary and Steinberg groups. For example, instead of first dealing with the case of finite root systems and then taking a limit to get the stable (= infinite rank) case, we deal with both cases at the same time. Moreover, our approach avoids having to deal with concrete matrix realizations of the groups in question, as is traditionally done, see e.g. [Ba] or [HO]. It allows for a concise description of the defining relations, independent of the types of root systems involved. Finally, as the discussion of the linear Steinberg group in 1.9 -- 1.11 shows, we need fewer relations than in previous work, for example no relations involving two roots in R0. With the exception of groups defined in terms of root systems of type E8, F4 and G2, which are not amenable to a Jordan approach, cf. 2.3(e), our Theorem 2.8 covers all types of Steinberg groups considered before. In addition, it also presents some new types, e.g., for elementary orthogonal groups. A detailed comparison of our Theorem 2.8 with previously known results is given in [LN2, 27.11]. At this point it is natural to ask if there also exists a generalization of part (b) of Theorem 1.6, stating that the map ℘ : St(A) → E(A) is a universal central extension. While the group St(V, R) gives a satisfactory replacement for the linear Steinberg group St(A), recasting the elementary linear group E(A) in the framework of Jordan pairs is limited to special Jordan pairs in the sense of 2.5(ii). While this can be done, see [Lo4], we will instead replace the elementary group E(A) by the projective elementary group PE(V ), see 2.10, that can be defined for any Jordan pair V . From the point of view of universal central extensions, this is harmless since, as we will see in 2.13, the group PE(V ) is isomorphic to the central quotient PE(A) = E(A)/Z(E(A)) and universal central extensions of a group and its central quotients are essentially the same by 1.4(e). 2.10. The Tits-Kantor-Koecher algebra and the projective elementary group of a Jordan pair. Let V be a Jordan pair, defined over a commutative ring k of scalars. It is fundamental (and well-known) that V gives rise to a Z-graded Lie k-algebra (2.10.1) L(V ) = L(V )1 ⊕ L(V )0 ⊕ L(V )−1, introduced at about the same time by Tits, Kantor and Koecher in [Ti1, Ti2, Ka1, Ka2, Ka3, Ko1, Ko3] and called the Tits-Kantor-Koecher algebra of V . Various versions of L(V ) exist, but all agree that (cid:0)L(V )1, L(V )−1(cid:1) = (V +, V −) as k-modules. For our purposes, the most appropriate choice for L(V )0 is (2.10.2) where ζ = (IdV +, IdV −) and δ(x, y) = (D(x, y), −D(y, x)) ∈ End(V +) × End(V −), defined by D(x, y)z = {x y z}. We let gl(V σ) be the Lie algebra defined by End(V σ) with the L(V )0 = kζ + Spank{δ(x, y) : (x, y) ∈ V }, 20 E. NEHER commutator as the Lie product. By definition, the Lie product of L(V ) is determined by the conditions that it be alternating, that L(V )0 be a subalgebra of the Lie algebra gl(V +) × gl(V −) and that [V σ, V σ] = 0, [D, z] = Dσ(z), [x, y] = −δ(x, y) for D = (D+, D−) ∈ L(V )0, z ∈ V σ and (x, y) ∈ V . It follows from the identity (JP15) in [Lo1], [D(x, y), D(u, v)] = D({x y u}, v) − D(u, {y x v}). that L(V )0 is indeed a subalgebra. As a k-Lie algebra, L(V ) is generated by ζ, V + and V −, and it has trivial centre. For a Jordan pair V with a fully idempotent root grading R a description of the derived algebra [L(V ), L(V )] is given in [Ne3]. The Tits-Kantor-Koecher algebra of a special Jordan pair is described in [Lo4, §2]. We will work out L(V ) for a rectangular matrix pair in 2.12. An automorphism f of V gives rise to an automorphism L(f ) of L, defined by x ⊕ D ⊕ y 7→ f+(x) ⊕ (f ◦ D ◦ f −1) ⊕ f−(y). The map f 7→ L(f ) is an embedding of the automorphism group Aut(V ) of V into the automorphism group of L(V ). Any (x, y) ∈ V gives rise to automorphisms exp+(x) and exp−(y) of L(V ), defined in terms of the decomposition (2.10.1) by the formal 3 × 3-matrices (2.10.3) exp+(x) =  1 ad x Qx 0 ad x 0 1 0 1   , exp−(y) =  1 ad y Qy 0 1 0 0 ad y 1  . The map expσ, σ = ±, is an injective homomorphism from the abelian group (V σ, +) to the automorphism group of L(V ), whose image is denoted U σ. The projective elementary group of V is the subgroup PE(V ) of Aut(cid:0)L(V )(cid:1) generated by U + ∪ U −, introduced in [Lo4] and studied further in [LN2, §7, §8]. We have now explained all the concepts used in the generalization of part (b) of the Kervaire-Milnor-Steinberg Theorem 1.6. 2.11. Theorem B. Let (R, R1) be a 3-graded root system and let V be a Jordan pair with a root grading R = (Vα)α∈R1. (a) There exists a group homomorphism π : St(V, R) → PE(V ), uniquely determined by π(cid:0)xσ(u)(cid:1) = expσ(u), (u ∈ V σ). (b) If all irreducible components of R have infinite rank and R is fully idempotent with respect to a family (eα)α∈R1 , the homomorphism π is a universal central extension. Theorem B is established in [LN2]. Part (a) follows from [LN2, Cor. 21.12]. By Fact 1.4(d) and Theorem 2.8, the proof of (b) boils down to showing that Ker π is cen- tral, which we do in [LN2, Cor. 27.6]. As for Theorem 2.8, it is not necessary to assume that R is fully idempotent. In the setting of (a) let (u, v) ∈ V + β with α 6= β and let b(u, v) be the element of St(V, R) defined in (2.7.1). Then π(b(u, v)) = L(f ) for some f ∈ Aut(V ) (for the experts: f is the inner automorphism (B(u, v), B(v, u)−1) of [Lo1, 3.9]). That π(b(u, v)) ∈ α × V − L(cid:0) Aut(V )(cid:1) ⊂ Aut(L(V )) is the motivation for the perhaps surprising definition of b(u, v). We finish this section by describing L(V ) and PE(V ) for V = MIJ (A). 2.12. The Tits-Kantor-Koecher algebra of a rectangular matrix pair. Let V = MIJ (A) = (cid:0)MatIJ (A), MatJ I(A)(cid:1) be the rectangular matrix pair of 2.5(iii). In this sub- section we present a model for the Tits-Kantor-Koecher algebra L = L(V ) in terms of elementary matrices which will be used in 2.13 to link the elementary group of V and the abstractly defined group PE(V ). Let 1I = diag(1A, . . .) be the diagonal matrix of size I × I, define 1J analogously and let A be the unital associative k-algebra 21 whose operations are given by matrix addition and matrix multiplication. In particular, MatJ I (A) A = A(V ) =(cid:18)k 1I + MatI (A) 0(cid:19) e1 =(cid:18)1I 0 0 MatIJ (A) k 1J + MatJ (A)(cid:19) =(cid:18)MatI (A)ex MatIJ (A) MatJ I (A) MatJ (A)ex(cid:19) 0 1J(cid:19) e2 =(cid:18)0 and 0 are orthogonal idempotents of A. We consider A rather than its subalgebra MatN (A)ex, N = I ∪ J, since this will allow us to model the element ζ of (2.10.2). The Peirce decomposition of A with respect to the idempotent e1 is 0 A11 =(cid:18)MatI (A)ex 0 0(cid:19) , A01 =(cid:18) MatJ I (A) 0(cid:19) , 0 0 0 A10 =(cid:18)0 MatIJ (A) (cid:19) , A00(cid:18)0 0 MatJ (A)ex(cid:19) . 0 0 Let A(−) be the Lie algebra associated with A. Thus, A(−) is defined on the k-module underlying A and its Lie algebra product is [x, y] = xy − yx for x, y ∈ A. The Lie algebra A(−) is Z-graded, A(−) =Ln∈Z A A (−) n with (−) 1 = A10, A (−) 0 = A11 ⊕ A00, A (−) −1 = A01 (−) n = 0 for n /∈ {1, 0, −1}. We define e = e(V ) as the subalgebra of A− generated by and A e1, e2 and e1 =(cid:18)0 MatIJ (A) 0 0 (cid:19) = A (−) 1 and e−1 =(cid:18) 0 0 MatJ I (A) 0(cid:19) = A (−) −1 . Lie algebra. We now relate e to the Tits-Kantor-Koecher algebra L = L(V ) of V . First, for a = Put e0 = k e1 + k e2 + [e1, e−1] and ei = 0 for i /∈ {−1, 0, 1}. Then e =Li∈Z ei is a Z-graded 0 d(cid:1) ∈ e0 define ∆(a) = (∆(a)+, ∆(a)−) ∈ Endk(V +) × Endk(V −) by (cid:0) a 0 ∆(a)+(u) = au − ud, ∆(a)−(v) = dv − va, so that (2.12.1) We claim: the map ha,(cid:18)0 b c 0(cid:19)i =(cid:18) 0 dv − va au − ud 0 (cid:19) =(cid:18) 0 ∆+(a)(u) ∆−(a)(v) 0 (cid:19) . Ψ : e → L, (cid:18)a b c d(cid:19) 7→ b ⊕ ∆(a, d) ⊕ (−c) is a surjective Lie algebra homomorphism whose kernel is z(e), the centre of e, and thus induces an isomorphism (2.12.2) e/z(e) ∼= L 22 E. NEHER of Lie algebras ([Lo4, 2.6], [LN2, 7.2]). Indeed, Ψ is surjective since ∆(e1) = ζ = −∆(e2) and for (x, y) ∈ V (2.12.3) by (2.5.1). To see that Ker Ψ = z(e), observe for m =(cid:0) a b [m, e1] = 0 ⇐⇒ b = 0 = c ⇐⇒ [m, e2] = 0, (2.12.4) ∆(cid:0)(cid:2)(cid:0) 0 x 0 0(cid:1), (cid:0) 0 0 y 0(cid:1)(cid:3)(cid:1) = ∆(cid:0) xy 0 0 −yx(cid:1) = δ(x, y) c d(cid:1) ∈ A that whence by (2.12.1), m ∈ Ker Ψ ⇐⇒ b = 0 = c, ∆(a, d) = 0, (cid:0) a 0 0 d(cid:1) ∈ e0, ⇐⇒ [m, e1] = 0 = [m, e2], [m, e1] = 0 = [m, e−1], m ∈ e, ⇐⇒ m ∈ z(e) because e1, e2, e1 and e−1 generate e as Lie algebra. Finally, since both e and L are Z-graded and Ψ preserves this grading, Ψ is a Lie algebra homomorphism as soon as Ψ preserves products of type [xi, yj] for (i, j) = (0, ±1), (1, −1) and (0, 0). For (0, ±1) and (1, −1) this follows from (2.12.1) and (2.12.3) respectively; the case (0, 0) is left to the reader. In the remainder of this subsection we will give a more precise description of e and its centre, see (2.12.5) and (2.12.7). Let [A, A] = Span{ab − ba : a, b ∈ A}, the derived algebra of the Lie algebra A−, and let slN (A) := {x = (xkl) ∈ MatN (A) : Pn∈N xnn ∈ [A, A]}. From [aEkl, bErs] = abδlrEks − baδksErl one then gets e1 ⊕ [e1, e−1] ⊕ e−1 = slN (A), whence (2.12.5) k e1 + k e2 + slN (A) = e. The description of the centre z(e) depends on the cardinality of N because A (−) 0 ∩ A1N =(A 1N if N < ∞, if N = ∞. k 1N Denoting by Z(A) = {z ∈ A : [z, A] = 0} the centre of A (= centre of A(−)), a straightfor- ward calculation shows (−) 0 ∩ Z(A)1N =(Z(A)1N , k1N , N < ∞, N = ∞. Since e is generated by e1, e2, e1 and e−1, (2.12.4) and (2.12.6) imply (2.12.6) (−) 0 (cid:8)x ∈ A (2.12.7) : [x, e1] = 0 = [x, e−1](cid:9) = A z(e) = e0 ∩ (Z(A) 1N ) =(e0 ∩ (Z(A)1N ), k1N , N < ∞, N = ∞. . For example, if A = k is a field of characteristic 0 and N = n is finite, we get e = gln(k), z(e) = k1n and L ∼= sln(k). 2.13. The projective elementary group of a rectangular matrix pair. We use the notation of 2.12 and let V = MIJ (A). The goal of this subsection is to show that the group PE(V ) is isomorphic to a central quotient of the elementary group E(A). We put and call it the elementary group of V . Since by 1.12 the group EN (A) is generated by e+(V +) ∪ e−(V −) this agrees with the definition of the elementary group of an arbitrary E(V ) = EN (A) special Jordan pair in [Lo4, §2] or [LN2, 6.2]. We will identify the Tits-Kantor-Koecher algebra L(V ) = L with e/z(e) via the isomorphism (2.12.2) induced by Ψ : e → L. Any g ∈ A× gives rise to an automorphism Ad g of A− defined by (Ad g)(x) = gxg−1. If Ad g stabilizes the subalgebra e of A−, it also stabilizes z(e), and therefore descends to an automorphism Ad(g) of e/z(e) = L satisfying Ψ ◦ (Ad ge) = (Ad g) ◦ Ψ. The map Ad : {g ∈ A× : (Ad g)(e) = e} → Aut(L), g 7→ Ad g 23 is a group homomorphism. For g = e+(x) = (cid:0) 1 x Ad e+(x) acts as follows: 0 1(cid:1) ∈ A× as in (1.12.1), the automorphism (cid:0) Ad e+(x)(cid:1) (cid:18)0 b (cid:0) Ad e+(x)(cid:1) (cid:18)a 0 (cid:0) Ad e+(x)(cid:1) (cid:18) 0 0 0(cid:19) =(cid:18)0 b 0 0(cid:19) , 0 d(cid:19) =(cid:18)a −ax + xd −c 0(cid:19) =(cid:18)−xc xcx cx(cid:19) −c d 0 0 (cid:19) =(cid:18)a 0 0 d(cid:19) +h(cid:18)0 x 0 0(cid:19) , (cid:18)a 0 0 d(cid:19)i, 0 =(cid:18) 0 −c 0(cid:19) +h(cid:18)0 x 0 0(cid:19) , (cid:18) 0 −c 0(cid:19)i +(cid:18)0 Qxc 0 (cid:19) . 0 0 These equations show that the automorphism Ad e+(x) stabilizes e and, by comparison with (2.10.3), that Ad e+(x) = exp+(x). One proves in the same way that Ad e−(y), y ∈ V −, stabilizes e and that Ad e−(y) = exp−(y). Since PE(V ) is generated by exp+(V +) ∪ exp−(V −), the homomorphism Ad restricts to a surjective group homomorphism AdE : E(V ) → PE(V ), g 7→ Ad g. We claim that its kernel is the centre Z(E(V )) of E(V ): (2.13.1) (2.13.2) Z(E(V )) = Ker Ad E(V ) = Ker AdE, whence E(V )/ Z(E(V )) ∼= PE(V ). Proof of (2.13.1): Clearly, Z(E(V )) = Ker Ad E(V ) ⊂ Ker AdE, so it remains to show that g ∈ Ker AdE is central in E(V ). Let g =(cid:0) a b c d(cid:1) and g−1 =(cid:0) a′ b′ c′ d′(cid:1). Then 0 (2.13.3) (2.13.4) (2.13.5) (2.13.6) 0 0 bd′ ca′ g(cid:18)1I 0 1J(cid:19) g−1 =(cid:18)bc′ g(cid:18)0 g−1(cid:18)0 x g−1(cid:18)0 0 0(cid:19) g−1 =(cid:18)aa′ ab′ cb′(cid:19) , dc′ dd′(cid:19) , 0 0(cid:19) g =(cid:18)a′xc a′xd c′xd(cid:19) , y 0(cid:19) g =(cid:18)b′ya b′yb d′ya d′yp(cid:19) . c′xc Since (Ad g) (m) ≡ m ≡ (Ad g−1)(m) mod z(e) for all m ∈ e and since z(e) is diagonal by (2.12.7), it follows from (2.13.3), (2.13.4) and (2.13.6) that ab′ = 0 = ca′ = bd′, d′ya = y for y ∈ V −. Applied to c ∈ V −, this proves cb′ = (d′ca) · b′ = d′c · ab′ = 0 and then bc′ = b · (d′c′a) = bd′ · c′a = 0. From 1N = gg−1 we obtain (cid:18)1I 0 0 1J(cid:19) =(cid:18)aa′ + bc′ ∗ ∗(cid:19) =(cid:18)aa′ ∗ ∗(cid:19) . ∗ ∗ 24 E. NEHER Together with the already established equations this shows, using (2.13.3), that (Ad g)(e1) = e1. Because g ∈ A, we get b = 0 = c from (2.12.4). Thus also b′ = 0 = c′, so that (2.13.5) and (2.13.6) prove that Ad g fixes e±1. Since e±(V ±) = 1N + e±1, we see that Ad g fixes the generators e±(V ±) of E(V ), i.e. g is central. (cid:3) A similar result holds for any special Jordan pair V : there always exists a surjective group homomorphism from the elementary group E(V ) (which we have not defined) onto the projective elementary group PE(V ), whose kernel is central, but not necessarily the centre of E(V ), see [Lo4, Thm. 2.8]. 3. Some open problems We describe some open problems for Steinberg and projective elementary groups of Jor- dan pairs. Our list is very much limited by the author's taste and knowledge. This section requires some expertise in Jordan pairs. 3.1. The normal subgroup structure of PE(V ). The problem is quite easily stated: Given a Jordan pair V , describe all normal subgroups of PE(V ). As stated, this may be too general. We therefore discuss some special cases. (a) In view of the results of [Lo5] it is natural to ask: when is PE(V ) a perfect group, when is it simple? Indeed, [Lo5, Thm. 2.6] says that, for a nondegenerate Jordan pair V with dcc on principal ideals, PE(V ) is a perfect group if and only if V has no simple elements. Also, by [Lo5, Thm. 2.8], PE(V ) is a simple (abstract) group if and only if V factors isomorphic to (F2, F2), (F3, F3) or (cid:0) H2(F2), H2(F2)(cid:1). Here Fq is the field with q is simple and not isomorphic to (F2, F2), (F3, F3) or (cid:0) H2(F2), H2(F2)(cid:1) ([Lo5, Thm. 2.8]). (the alternating group on four letters), and PE(cid:0) H2(F2), H3(F2)(cid:1) ∼= S6. Thus, the problem That the exceptional cases have to be excluded in these two theorems is evident from the isomorphisms PE(F2, F2) ∼= S3 (the symmetric group on three letters), PE(F3, F3) ∼= A4 is to find out if these two theorems of [Lo5] hold for more general Jordan pairs. It follows from (b) that is natural to assume simplicity of V for the second theorem. (b) Every ideal I of the Jordan pair V gives rise to a normal subgroup of PE(V ). Indeed, one can show ([LN2, 7.5]) that the canonical map can : V → V /I induces a surjective group homomorphism PE(can) : PE(V ) → PE(V /I). We let PE(V, I) be its kernel: 1 / PE(V, I) / PE(V ) PE(can) / PE(V /I) / 1 Problem: describe PE(V, I) by generators and relations. For elementary linear groups over rings this is a standard result, see for example [HVZ, Lemma 3]. The paper [CK] shows that even in case SL2(A) one needs methods from Jordan algebras. 3.2. Central closedness of St(V, R) in low ranks. We have excluded low rank cases in Theorem 2.8 for the simple reason that it is not true without further assumptions in low ranks. We discuss 2 ≤ rank R ≤ 4 in (a) and rank R = 1 in (b). (a) One knows ([LN2, 27.11]) that St(V, R) is a classical linear or unitary Steinberg group. Let us first consider the case that V is defined over a field F and that dimF Vα = 1 for all α ∈ R1. Then [St3, Thm. 1.1] applies and yields that St(V, R) is not centrally closed if and only if (R, R1) and F satisfy one of the following conditions. (3.2.1) (R, R1) A1 2 A1 3 or A2 F 2, 4 2 2 3 Cher 2 Bqf 3 3 Cher 3 Dalt 4 2 2 / / / / 25 In this table we use the abbreviation A1 A1 St(V, R) is centrally closed, as mentioned in 2.8. N for N = 2, I = 1 and analogously for 4 . The cases R = A4, B4, C4 do not appear in the table because in these cases 3, . . . , Dalt 2 = AI Still assuming that V is defined over a field F , it is natural to replace the assumption dimF Vα = 1 by the requirement that the fully idempotent root grading R of V is a division grading in the sense that all root spaces Vα are Jordan division pairs, which means that for every non-zero x ∈ V σ is invertible. Preliminary investigations lead us to conjecture: α the endomorphism Q(x)V −σ α (C) If St(V, R) is not centrally closed then dimF Vα = 1 for all α ∈ R1 and F satisfies the restrictions of table (3.2.1). (b) R = A1: As in (a) we assume that R is a division grading, i.e., V is a division pair and is therefore isomorphic to the Jordan pair (J, J) of a division Jordan algebra J. By [LN2, 9.13] this is equivalent to PE(V ) being a rank one group in the sense of [Lo6]. Since the grading is trivial, St(V, R) is the free product of the abelian groups V + and V −, which is not perfect in general, a necessary condition for a group to be centrally closed (1.4(a)). Following the example of Chevalley groups [St2], it seems more promising to consider the group St(J) defined by the following presentation: • generators xσ(a), a ∈ J, σ = ± and, putting wb = x−(b−1) x+(b) x−(b−1) for 0 6= b ∈ J, • relations xσ(a + b) = xσ(a) xσ(b) for a, b ∈ J and wb x−(a) w−1 b = x+(cid:0)U (b)a) for all a ∈ J and all 0 6= b ∈ J. We remark that St(J) is the Steinberg group St(V, S) of [LN2, 13.1], where S is the set of all non-zero idempotents of V . By [LN2, 13.6], St(J) is the classical Steinberg group St(A) in case V = (A, A) and A an associative division algebra. To motivate our conjecture in this case, let us first consider the special case J = Fq. Since by 3.1(a) the group PE(V ) is not perfect in case J = Fq, q = 2, 3, these cases have to be excluded. Moreover, by [St3, Th. 1.1], St(J) is not centrally closed in case V = (Fq, Fq) and q ∈ {4, 9}, but these values of q are the only exceptions for V = (F, F ), F a field. This leads us to ask: (Q) Is St(J) centrally closed whenever J 6= Fq with q ∈ {2, 3, 4, 9} ? There exists an example of an associative unital F5-algebra A for which St(A) is not centrally closed [Str, Ex. 4], but A is not a division algebra. 3.3. Centrality of Ker(π). Let π : St(V, R) → PE(V ) be the homomorphism of Theo- rem 2.11(a). For simplicity, let us assume that R is irreducible. If R has infinite rank, part (b) of 2.11 says that π is a universal central extension. The problem here is: find sufficient conditions for Ker(π) to be central if R has finite rank. Some special cases are known. For example, if V is split in the sense of [Ne4], centrality of Ker(π) is established in [vdK], [La], [LS] and [Si] for rank ≥ 3. The quoted papers all use the same method, pioneered by [vdK], namely a "basis-free presentation of St(V, R)". Can the method of [vdK] be generalized to treat St(V, R), V split root graded, in a case-free manner? For a slightly different type of Steinberg group and a unit regular V , centrality of Ker(π) is shown in [Lo5, Th. 1.12]. 26 E. NEHER References [Ba] A. Bak, K-theory of forms, Ann. of Math. Studies, vol. 98, Princeton University Press, 1981. [Bo] N. Bourbaki, Groupes et alg`ebres de Lie, chapitres 4 -- 6, Masson, Paris, 1981. [CS] C. Chevalley and R. D. Schafer, The simple Lie algebras F4 exceptional and E6, Proc. Nat. Acad. Sci. U.S.A. 36 (1950), 137 -- 141. [CK] D. L. Costa and G. E. Keller, The E(2, A) sections of SL(2, A), Ann. of Math. (2) 134 (1991), no. 1, 159 -- 188. [HO] A. J. Hahn and O. T. OMeara, The classical groups and K-theory, Grundlehren, vol. 291, Springer- Verlag, 1989. [HVZ] R. Hazrat, N. Vavilov, and Z. Zhang, The commutators of classical groups, Zap. Nauchn. Sem. S.- Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 443 (2016), no. Voprosy Teorii Predstavleniı Algebr i Grupp. 29, 151 -- 221. J. Humphreys, Introduction to Lie algebras and Representation Theory, Graduate Texts in Mathe- matics vol. 9, Springer-Verlag, New York, 1972. [Hu] [Ja1] N. Jacobson, Some groups of transformations defined by Jordan algebras. I, J. Reine Angew. Math. 201 (1959), 178 -- 195. , Some groups of transformations defined by Jordan algebras. II. Groups of type F4, J. Reine Angew. Math. 204 (1960), 74 -- 98. , Some groups of transformations defined by Jordan algebras. III, J. Reine Angew. Math. 207 (1961), 61 -- 85. , Lectures on quadratic Jordan algebras, Tata Institute of Fundamental Research, 1969. , Structure theory of Jordan algebras, Lecture Notes in Math., vol. 5, The University of [Ja2] [Ja3] [Ja4] [Ja5] Arkansas, 1981. [Ka1] I. L. Kantor, Classification of irreducible transitive differential groups, Dokl. Akad. Nauk SSSR 158 (1964), 1271 -- 1274. [Ka2] [Ka3] , Non-linear groups of transformations defined by general norms of Jordan algebras, Dokl. Akad. Nauk SSSR 172 (1967), 779 -- 782. , Certain generalizations of Jordan algebras, Trudy Sem. Vektor. Tenzor. Anal. 16 (1972), 407 -- 499. [Ke] M. Kervaire, Multiplicateurs de Schur et K-th´eorie, Essays on Topology and Related Topics (M´emoires d´edi´es `a Georges de Rham), Springer, New York, 1970, pp. pp 212 -- 225. [Ko1] M. Koecher, Imbedding of Jordan algebras into Lie algebras. I, Amer. J. Math. 89 (1967), 787 -- 816. [Ko2] [Ko3] [Ko4] [La] A. Lavrenov, Another presentation for symplectic Steinberg groups, J. Pure Appl. Algebra 219 (2015), , Uber eine Gruppe von rationalen Abbildungen, Invent. Math. 3 (1967), 136 -- 171. , Imbedding of Jordan algebras into Lie algebras. II, Amer. J. Math. 90 (1968), 476 -- 510. , Gruppen und Lie-Algebren von rationalen Funktionen, Math. Z. 109 (1969), 349 -- 392. no. 9, 3755 -- 3780. [LS] A. Lavrenov and S. Sinchuk, On centrality of even orthogonal K2, J. Pure Appl. Algebra 221 (2017), no. 5, 1134 -- 1145. [Lo1] O. Loos, Jordan pairs, Springer-Verlag, Berlin, 1975, Lecture Notes in Mathematics, vol 460. [Lo2] [Lo3] [Lo4] [Lo5] , Homogeneous algebraic varieties defined by Jordan pairs, Mh. Math. 86 (1978), 107 -- 129. , On algebraic groups defined by Jordan pairs, Nagoya Math. J. 74 (1979), 23 -- 66. , Elementary groups and stability for Jordan pairs, K-Theory 9 (1995), 77 -- 116. , Steinberg groups and simplicity of elementary groups defined by Jordan pairs, J. Algebra 186 (1996), no. 1, 207 -- 234. [Lo6] [LN1] [LN2] , Rank one groups and division pairs, Bull. Belg. Math. Soc. Simon Stevin 21 (2014), no. 3, 489 -- 521. , Locally finite root systems, Mem. Amer. Math. Soc. 171 (2004), no. 811. , version August Steinberg groups for Jordan pairs, 2018, available at http://agt2.cie.uma.es/ loos/homepage/. [LPR] O. Loos, H. P. Petersson and M. L. Racine, Inner derivations of alternative algebras over commutative rings, Algebra & Number Theory, 2, (2008), 927 -- 968. [Ma] B. Magurn, An algebraic introduction to K-Theory, Encyclopedia of Mathematics and Its Applications [Mi] 87, Cambridge University Press 2002. J. Milnor, Introduction to algebraic K-theory, Ann. of Math. Studies, vol. 72, Princeton University Press, 1971. 27 [Ne1] E. Neher, Jordan triple systems by the grid approach, Lecture Notes in Math., vol. 1280, Springer- Verlag, 1987. [Ne2] E. Neher, Syst`emes de racines 3-gradu´es, C. R. Acad. Sci. Paris S´er. I 310 (1990), 687 -- 690. [Ne3] , Lie algebras graded by 3-graded root systems and Jordan pairs covered by a grid, Amer. J. Math 118 (1996), 439 -- 491. [Ne4] [Ro] [Si] , Polynomial identities and nonidentities of split Jordan pairs, J. Algebra 211 (1999, 206 -- 224. J. Rosenberg, Algebraic K-theory and its applications, Graduate Texts in Mathematics, vol. 147, Springer-Verlag, New York, 1994. S. Sinchuk, On centrality of K2 for Chevalley groups of type Eℓ, J. Pure Appl. Algebra 220 (2016), 857 -- 875. [Sp] T. A. Springer, Jordan algebras and algebraic groups, Springer-Verlag, New York, 1973, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol 75. [SV] T. A. Springer and F. D. Veldkamp, Octonions, Jordan algebras and exceptional groups. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2000 [St1] R. Steinberg, G´en´erateurs, relations, et revetements de groupes alg´ebriques, Colloq. Th´eorie des Groupes Alg´ebriques (Bruxelles), 1962, 113-127. , Lectures on Chevalley groups, Yale University Lecture Notes, New Haven, Conn., 1967. , Generators, relations and coverings of algebraic groups II, J. Algebra 71 (1981), 527-543. J. R. Strooker, The fundamental group of the general linear group, J. Algebra 48 (1977), 477 -- 508. [St2] [St3] [Str] [Ti1] J. Tits, Une classe dalg`ebres de Lie en relation avec les alg`ebres de Jordan, Nederl. Akad. Weten- sch. Proc. Ser. A 65 = Indag. Math. 24 (1962), 530 -- 535. [Ti2] J. Tits, Alg`ebres alternatives, alg`ebres de Jordan et alg`ebres de Lie exceptionnelles. I. Construction, Nederl. Akad. Wetensch. Proc. Ser. A 69 = Indag. Math. 28 (1966), 223 -- 237. [vdK] W. van der Kallen, Another presentation for Steinberg groups, Nederl. Akad. Wetensch. Proc. Ser. A 80 = Indag. Math. 39 (1977), 304 -- 312. [We] C. Weibel, The K-bbok: an introduction to algebraic K-theory, Graduate Studies in Mathematics 145, AMS 2013. Department of Mathematics and Statistics, University of Ottawa, Ottawa, Ontario K1N 6N5, Canada E-mail address: [email protected]
1512.04640	1	1512	2015-12-15T04:16:59	Nil-good and nil-good clean matrix rings	[ "math.RA" ]	The notion of clean rings and 2-good rings have many variations, and have been widely studied. We provide a few results about two new variations of these concepts and discuss the theory that ties these variations to objects and properties of interest to noncommutative algebraists. A ring is called nil-good if each element in the ring is the sum of a nilpotent element and either a unit or zero. We establish that the ring of endomorphisms of a module over a division is nil-good, as well as some basic consequences. We then define a new property we call nil-good clean, the condition that an element of a ring is the sum of a nilpotent, an idempotent, and a unit. We explore the interplay between these properties and the notion of clean rings.	math.RA	math	NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS ALEXI BLOCK GORMAN AND WING YAN SHIAO Abstract. The notion of clean rings and 2-good rings have many variations, and have been widely studied. We provide a few results about two new variations of these concepts and discuss the theory that ties these variations to objects and properties of interest to noncommutative algebraists. A ring is called “nil-good” if each element in the ring is the sum of a nilpotent element and either a unit or zero. We establish that the ring of endomorphisms of a module over a division is nil-good, as well as some basic consequences. We then define a new property we call “nil-good clean,” the condition that an element of a ring is the sum of a nilpotent, an idempotent, and a unit. We explore the interplay between these properties and the notion of clean rings. 1. Introduction In 1977, W. K. Nicholson defined a ring R to be clean if for every a ∈ R there is u a unit in R and e an idempotent in R such that a = u + e [5]. The interest in the clean property of rings stems from its close connection to exchange rings, since clean is a concise property that implies exchange. Properties of rings related to the clean and exchange properties have been largely expanded and researched, and some generalizations closely relate to other properties of interest to algebraists. The study of rings generated by their units dates back to the 1950’s, when it was proved that any endomorphism of a module over a division ring is equal to the sum of two units, unless the dimension of the module is 1 or the division ring is F2, as established in [7] and [8]. This motivated algebraists to make extensive study of rings generated by their units. Later, Peter V´amos defined an element a in R to be 2-good if it can be expressed as the sum of two units in R, and defined a ring R to be 2-good if every element in R is 2-good [6]. In general, a ring is n-good if each element can be written as the sum of n units, and these properties have distinct applications from those of clean, but have also led to a diverse line of inquiry in ring theory. P. Danchev defined a property in [2] related to 2-good in the following way: an element a in R is nil-good if a = n + u where n is a nilpotent element of R and u is either 0 or a unit in R. The ring R is called nil-good if every element of R is nil-good. In this paper, we prove that if R is a division ring, then Mn(R) is nil-good for all n ∈ N. We then establish some basic properties of nil-good rings in general. We extend these results to specifically characterize local rings and artinian rings that are nil-good. We then relate this property to clean rings in a new way by defining the property nil-good clean. We say a ring R is nil-good clean if for all r ∈ R there is a unit u, a nilpotent n and an idempotent e in R such that r = u + n + e. This property holds for all clean and all nil-good rings, yet we show it includes a larger class of rings than only those that satisfy one property or the other. Understanding how the nil-good clean property generalizes to include rings that are neither nil-good nor clean may reveal more about the interaction of those two properties within unital rings. We provide an example of a nil-good clean ring that is not exchange, and therefore 1 2 BLOCK GORMAN AND SHIAO not clean. The example has properties similar to the ring provided by G. M. Bergman in [3] of a nonclean exchange ring. Throughout this paper rings are associative with unity. We denote the Jacobson radical J(R) for a ring R and write Mn(R) for the ring of n × n matrices over R. 2. Matrices over a Field We first prove that Mn(R) is a nil-good ring when R is a field for illustrative purposes. The linear algebra over Mn(R) required in the case where R is a field is more accessible, and the subsequent proof in the case where R is a division ring is more concise and intuitive as a result. We can write a nil-good decomposition for any element of Mn(k) where k is a field by putting all matrices in rational canonical form. Theorem 2.1. For all n ≥ 1 the ring Mn(k) is nil-good if k is a field. Proof. Note that suitable rearrangement of the basis elements allows us to rearrange the companion matrix blocks in a matrix’s rational canonical form without altering the nilpotence or invertibility of that matrix. If the coefficients of any matrix are all zero then its rational canonical form is exclusively zero except possibly on the subdiagonal, making it nilpotent, in which case we let U be the zero matrix, and let N be the matrix in question. Similarly, if the matrix A we wish to decompose is the n × n zero matrix we let both U and N be the zero matrix. If the matrix A in question is a unit, we let N be the zero matrix, and let U be the original matrix, the rational canonical form of A. Now suppose A is a non-nilpotent, non-unit matrix. Then, it will have zero as its −c0 coefficient for some companion matrix block, as well as some non-zero −c0 coefficients. We may choose an arrangement of the companion matrix basis that allows us to place the companion matrix blocks corresponding to the nonzero first coefficient in the upper left corner, and to place the blocks for which the −c0 coefficient is zero in the lower right hand corner, ordered amongst themselves by size of block. Thus, we consider a matrix of the form   Cg1 0 0 ... 0 Cg2 . . . · · · 0 · · · ... . . . . . . 0 0 Cgk   where each Cgi is a companion matrix that has some nonzero element for all i < j for some j > 1. We call an r × r block consisting of zeros everywhere except the subdiagonal, which consists entirely of ones, an Nr block. Therefore, each Cgi for i < k has some nonzero coefficient −ci or is an Nr block of size 2 × 2 or greater. If the size m × m companion matrix block Cgi of A is invertible, then we let the corresponding diagonal block of U be Cgi and the corresponding diagonal block of N be the m × m zero block. If A contains any companion matrix block with zero as the −c0 coefficient, then in the unit U of its decomposition we add a −1 in the entry corresponding to the −c0-coefficient of the companion matrix block, so that the block becomes invertible. Correspondingly, we place a 1 in the entry of the nilpotent matrix corresponding to the entry of the −c0 coefficient of that companion matrix block. Otherwise we leave the corresponding block in the nilpotent summand entirely zero, so that the NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS 3 direct sum of these blocks will be the direct sum of nilpotents, making the overall matrix nilpotent as well. Note that the method described works for Nr blocks as well as those with nonzero coefficents. In general, the decomposition of the companion matrix block that has zero as its −c0 coefficient looks like: 0 · · ·  . . . 1 . . . . . .  0 ... · · · ... 0 0 −c1 ... 1 −cm−1   = 0 · · ·  . . . 1 . . . . . .  0 ... · · · ... 0 −1 −c1 ... 1 −cm−1   +   0 · · · ... . . . ... . . . 0 · · · 0 . . . · · ·  1 0 ... 0  Since these matrices will be useful later, the first matrix on the right side of the equation we will gi is invertible gi and the second matrix on the right we will denote N ∗. Observe that C ∗ denote C ∗ and N ∗ is nilpotent. Whether the coefficients −c1 through −cm−1 are zero or nonzero does not affect the validity of the decomposition, since the addition of a one in the first entry of the last column makes the columns of the A − N block necessarily linearly independent, ensuring that this decomposition is in fact the sum of an invertible matrix and a nilpotent one. The direct sum of such m × m blocks will respectively be invertible and nilpotent as well. Suppose the rational canonical form of the matrix is a series of companion matrix blocks followed by a zero block:   Cg1 0 ... 0 0 . . . 0 · · · Cgm 0 0 ... ... [0]   We will decompose the zero block of size n × n augmented with the last two rows and columns of the Cgm block of size r × r to ensure that there is at least one nonzero entry, which is the one on the last row of Cgm and the (m − 1)th column. Thus the augmented matrix that we then decompose is of the form 0 −cs−2 1 −cs−1 A =   0 ... 0 0 0 . . . . . . · · · · · · · · · . . . . . . · · ·  0 0 ... ... 0  ... ... 0 where A is an (n + 2) × (n + 2) matrix. To find a suitable (n + 2) × (n + 2) nilpotent matrix of rank n − 1 for the decomposition, we will conjugate Nn+2 by a suitable invertible matrix since conjugation will result in another nilpotent matrix. Choose 4 BLOCK GORMAN AND SHIAO  0 0 ... ... 0 1 −1  0 ... · · · · · · 1 −1 · · · 0 1 −1 . . . . . . 0 · · · 0 1 0 0 · · · ... . . . . . . 0 1 −1 · · · 0   and P −1 =   1 1 1 1 1 0 ... ... ... 1 1 0 1 1 1 . . . · · · · · · . . . . . . 0 · · · 1 ... ... 1 1 · · · 1 0 ... ... ... 0   P = Then, P Nn+2P −1 =   1 0 0 ... ... 0 −1 · · · · · · 0 0 1 · · · · · · . . . 0 0 ... . . . ... 0 · · · −1 −1 −1 1 · · · · · · 1 0 0 ... ... 0 0 1 0   will be the nilpotent matrix involved in the decomposition. Now let U =   1 1 −cs−2 1 −cs−1 0 ... ... ... 0 −1 · · · 0 0 0 . . . · · · 0 0 0 · · · · · · · · · 0 1 0 ... ... ... 0 · · · −1 −1 −1 1 0 0 ... ... ... 0 . . . 1   and N =  0 −1 0 0 0 −1 ... ... 0 1  · · · · · · · · · . . . −1 · · · 1 0 −1 0  0 0 −1 0 0 ... ... ... ... 0 0 1 1  To see why U is invertible, note that regardless of the values of cs−2 and cs−1 the second through the last column form a linearly independent set of vectors because of the negative ones on different rows. Then we only need to determine if the first column can be written as a linear combination of vectors from this set. If indeed there were scalars such that the first column could be written as a linear combination of the others in U , then having zeros everywhere except the first, second and last row will restrict the coefficients of the columns to be zero except possibly the (n + 1)th and (n + 2)th column. However, a routine calculation shows such a linear combination is not possible either. Thus the matrix is invertible. NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS 5 Now we consider A as the direct sum of Cg1 through Cg(m−1) and Cgm with the zero block, as illustrated: · · · . . . . . . 0  1  0 −c0 ... ... 0 −cs−2 1 −cs−1 0 ... ... ...   ⊕ [0] where Cgm ⊕ [0] is an (n + r) × (n + r) matrix. Then choose U = C ∗ g1 ⊕ · · · ⊕ C ∗ g(m−1) ⊕ Then we have N = N ∗ ⊕ · · · ⊕ 0 · · · . . . . . . 1    0  1 ... 0 1 −c0 ... ... 0 ... ... 1 −cs−2 1 −cs−1 0 ... ... ... 0 −1 · · · 1 0 0 0 . . . · · · · · · · · · . . . · · · · · ·   0 · · · 0 · · · · · · · · · 0 1 0 ... ... ... 0 · · · −1 −1 −1 1 0 0 ... ... ... 0 . . . 1 · · · −1 ... . . . 0 0 −1 0 0 0 −1 ... ... 0 1 · · · 0 −1 0   0 0 −1 0 0 ... ... ... ... 0 0 1 1 · · · · · · . . . −1 · · · 1 which is nilpotent. Thus A = U + N where U is a unit and N is a nilpotent. One can check that in the last direct summand the first (r − 2) columns and the last (n + 1) columns are linearly independent because of the ones in different rows. Due to the 1’s the rth row and (r + 1)th column, the (r + 1)th column cannot be written as a linear combination of the others, thus the only concern is the (r − 1)th column. Writing it as a linear combination of other columns requires constructing the linear combination using only the (n + r − 1)th and (r + n)th columns, because to use the (r − 2)th column requires also using the rth column due to the 1 in the first row and (r − 2)th column. However, the 1 in the (r + 1)th row and rth column prevents this. Yet we’ve 6 BLOCK GORMAN AND SHIAO observed the (r − 1)th, (n + r − 1)th, and (r + n)th columns are linearly independent. Thus U is invertible. Having addressed all possible cases for an n × n matrix’s rational canonical form, we again recall that conjugation by an invertible matrix and its inverse preserves both unity and nilpotence, so we may conclude that Mn(k) is nil-good for any field k and dimension n. (cid:3) 3. Matrices over Division Rings We now consider linear operators on division ring modules of dimension n. Since the modules we consider are finite-dimensional, there exists a basis with respect to which we may express linear operators as n × n matrices. Theorem 3.1. For all n ≥ 1 the ring Mn(D) is nil-good if D is a division ring. Proof. Given an n × n matrix A ∈ Mn(D) there exists an invertible matrix Q ∈ Mn(D) such that A = QAdQ−1 where Ad is a matrix of the form UA ⊕ NA. Here UA is an m × m invertible block on the diagonal and NA is an (n − m) × (n − m) nilpotent block. Although a matrix over a division ring does not necessarily have a rational canonical form, there exists a primary rational canonical form [1] similar to the standard rational canonical form for certain matrices. Suppose a matrix A ∈ Mn(R) is algebraic over the center of the division ring R and that it has a single elementary divisor α. P. Cohn proved that if α = c1c2 . . . cs then A may be put into the form C1 0 N ∗ C2 . . . · · · . . . . . . 0 ... 0   0 ...   . . . N ∗ Cs where N ∗ is a matrix with a 1 in the entry in the upper right corner and zeros everywhere else, and C is the companion matrix of ci for each i. We note that the matrix NA may be put in primary rational ranonical form since it is algebraic over the center of any ring and its minimal polynomial xk = 0 has a single elementary divisor. Since the companion matrix of any power of nilpotent NA is a direct sum of Nr blocks, the primary rational canonical form of NA is zero everywhere except the subdiagonal, which, as in the case of matrices over a field, makes the primary rational canonical form a direct sum of Nr blocks as well. If the Nr blocks have dimension 2 or higher, then we can write NA in the form  Nr1 0  0 Nr2 ... . . . · · · 0 · · · 0 ... . . . . . . 0 0 Nrk   NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS 7 which, by subtracting a nilpotent of the form  N ∗ r1 0  0 N ∗ r2 ... . . . · · · 0 0 · · · ... . . . . . . 0 0 N ∗ rk   ri is an N ∗ matrix with the dimension of the corresponding Nri matrix) (in which the matrix N ∗ results in an invertible matrix. The details of this proccess are outlined more explicitly in the case of a vector space over a field. If any companion matrix in the primary rational canonical form of NA has a k × k zero block in its companion matrix, then we consider the augmented diagonal block given by the zero block and the k + 1 entries of the column immediately to the left and the k + 1 entries furthest to the right in row immediately above. If the zero block in the companion matrix is not the first diagonal block in the first companion matrix on the diagonal of the primary rational canonical form of NA, then the augmented block in question will be of the form  0 1  0 ... 0 0 ... . . . · · · · · · . . . · · · 0 0 ... 0   for which, if we substract the nilpotent matrix · · · · · · . . .   0 −1 0 0 0 −1 ... ... 0 1 · · · 0 −1 0  0 0 −1 0 0 ... ... ... ... 0 0 1 1  −1 · · · 1 the result is an invertible block. Just as is the case for matrices over a field, this redistribution of columns and rows into slightly smaller or larger blocks does not change the block martices used to decompose Ad into the sum of an invertible matrix and a nilpotent one. In the case that the first k × k block of the first companion matrix on the diagonal of the primary rational canonical form of NA is a zero block, we treat the first (m + k) × (m + k) block, which is comprised of UA and the zero block in question, as described in the case detailed below, then treat the bottom (n − m − k) × (n − m − k) block as described above. In the special case that the NA block is an (n − m) × (n − m) zero block, a little more work is required than for the field case to find a decomposition of the matrix since the invertible block is not in a nice normal form. We first consider the case when amm is nonzero. Then we augment the zero block by the last n − m entries of the last column and row of UA. 8 BLOCK GORMAN AND SHIAO The augmented matrix A =   amm ... ... 0 0 . . . . . . · · · · · · . . . . . . · · ·  0 ... ... 0  can be written as the sum of the following: U = amm + 1   0 ... ... ... 0 −1 0 0 1 · · · · · · . . . · · · · · · 0 0 · · · · · · 0 1 ... ... ... 0 · · · −1 −1 −1 1 0 ... ... ... 0 . . . 1   and N =   0 −1 0 0 0 −1 ... ... 0 1 · · · · · · · · · . . . −1 1 · · · 0 −1 0  0 0 −1 0 0 ... ... ... ... 0 0 1 1  In both matrices, the last (n − m) columns of U are linearly independent because of the ones on exclusively different rows. If writing the first column as a linear combination of the others is possible, only the (n − m)th column can be included. Since amm 6= 0, the first column is not a scalar multiple of the (n − m)th column, and thus U is invertible. Therefore, Ad = U + N , where U =   a11 · · · . . . a1m ... amm + 1 0 0 ... ... 0 −1 0 0 1 · · · · · · 0 . . . · · · · · ·   · · · · · · · · · 0 · · · · · · 0 1 0 ... ... 0 · · · −1 −1 −1 1 0 0 ... ... 0 . . . 1 NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS 9 and N = 0 ... 0 0 · · ·  . . .  0 −1 0 0 0 −1 ... ... 0 1 · · ·   . 0 0 −1 0 0 ... ... ... ... 0 0 1 1 0 −1 0 · · · · · · . . . −1 · · · 1 One can see that the first (m − 1) and the last n columns are linearly independent. So if U is not invertible, then the mth column can be written as a linear combination of the other columns. Note that because of the zeros, the coefficients of the last m columns, except possibly the (m + 1)th column, have to be zero. Since there is a negative one on the last row of the mth column, the coefficient of the (n − 1)th column has to be one. Then the difference of the mth column and the (n − 1)th column would be a linear combination of the first (m − 1) columns. The existence of such a linear combination would imply that UA is not invertible, a contradiction. Therefore U is invertible. Now consider the case in which amm = 0. If there is a nonzero entry on the diagonal, say aii, then we can conjugate Ad by a permutation matrix P that swaps the ith row with the mth row. Then the (m, m)th entry in P AdP −1 will be nonzero, and we can apply the above method to decompose the matrix. If all of the entries on the diagonal are zero, but a(m−1)m 6= 0, then we can conjugate Ad by an invertible matrix S that subtracts the (m − 1)th row from the mth row. Note that conjugation does not change the invertibility of the unit block and the nilpotence of the nilpotent block. We define S and its inverse S−1 as the following (m + n) × (m + n) matrices:   1  1 . . . S =  and S−1 =  . . . 1 1 1 In    · · · . . . · · · 1 −1 1 In 0 ... So A′ d = SAdS−1 =   a(m−1)1 a(m−1)m am1 − a(m−1)1 am(m−1) am(m−1) − a(m−1)m −a(m−1)m a(m−1)m a1(m−1) + a1m ... a1m ...   ⊕ [0] Now the (m, m)th entry of UA′ If all of the entries on the diagonal are zero and but am−1(m) = 0, there is a nonzero entry on the mth column because UA is invertible. Let akm be nonzero. Then we can conjugate Ad by a permutation matrix P that swaps the kth row with the (m − 1)th row. is nonzero, so we can use the first case for decomposition. d 10 BLOCK GORMAN AND SHIAO Then let A′′ d = P AdP −1 =   0 ... a(m−1)1 ... ak1 am1 · · · · · · · · · · · · a1(m−1) ... 0 ... ak(m−1) am(m−1) · · · ... · · · ... · · · · · · a1k ... a(m−1)k ... 0 amk a1m ... 0 ... akm 0   ⊕ [0] Now in the mth column, the entry on the (m − 1)th row is nonzero, we can conjugate A′ d by S as introduced above and follow the above method for decomposition. Therefore, solely by applying a certain series of invertible linear transformations, one may find a nil-good decomposition of any square matrix over a division ring. (cid:3) 4. General Properties of Nil-good Rings Having established the essential fact that the ring of n × n matrices over a division ring is nil- good, we may observe some sufficient or necessary conditions for some types of widely used rings to be nil-good. In particular we give a necessary and sufficient condition for artinian rings and matrices over a local ring. For completeness, proofs of other elementary facts are provided. The following four remarks also appear in [2], but we briefly provide our own proofs of these elementary results for completeness. Remark 4.1. A ring R is nil-good if and only if there exists a nil ideal A such that R/A is nil-good. Proof. The forward direction is trivial, simply consider the ideal (0). Suppose now that A is a nil ideal of R and every element of R/A has a nil-good decomposition. If ¯a is nilpotent in R/A then ¯ak = 0 in R/A for some k ∈ N so ak ∈ R/A. If R/A is a nil ideal, this implies ak = n for some nilpotent element n in R/A. Then if ak is nilpotent, it is immediate that a is nilpotent. Since any nil ideal is contained in the radical and units lift modulo J(R), we conclude any unit ¯u in R/A lifts to a unit u in R. So the nil-good decomposition of any element in R/A lifts to the sum of a unit and a nilpotent in R. (cid:3) As a corollary to this, we know a ring R is nil-good if and only if for any nil ideal A the quotient ring R/A is nil-good. The essence of the proof is similar to that of the above proposition, with the added observation that if R is a nil-good ring and A is a nil ideal, then given a = n + u for a nilpotent n in R and a unit u in R the image ¯a in R/A has the decomposition u + n = ¯u + ¯n. Since A is a nil ideal it is contained in J(R) so ¯u is a unit in the quotient ring. Moreover, the fact that nk = 0 for some k means that nk ∈ A so ¯nk = ¯0. Remark 4.2. If R is nil-good then J(R) is a nil ideal. Proof. If R is nil-good then for all y ∈ J(R) we know y = n + u where n is nilpotent and u is a unit or zero. Suppose for contradiction that u is a unit. Then if U (R) denotes the set of units in R, we know 1 − yu−1 ∈ U (R) by definition of the Jacobson radical. However 1 − yu−1 = 1 − nu−1 − 1 which implies −nu−1 ∈ U (R), a contradiction. So we must have that u = 0 which implies y = n + 0 is nilpotent. (cid:3) NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS 11 If J(R) is a nil ideal, then by the above two remarks we know R is nil-good if and only if R/J(R) is nil-good. Therefore, if J(R) is a nil ideal, we wish to know when R/J(R) is nil-good. The following result will prove useful to that end. If J(R) is the unique ideal that is maximal both as a left ideal and as a right ideal, then we say R is a local ring. Remark 4.3. A local ring R is nil-good if and only if J(R) is nil. Proof. If R is nil-good then J(R) is nil by Remark 4.2. If J(R) is a nil ideal then since J(R) is equal to the unique maximal ideal of R, any element not in J(R) is a unit. Therefore for any element a in R either has the decomposition a = n + 0 or a = u + 0 which implies R is nil-good. (cid:3) Remark 4.4. If R is nil-good then R has no nontrivial central idempotents. Proof. Suppose R is nil-good and e is a central idempotent. Then e = u+n which implies u = e−n. If u is a unit then e − n is a unit that commutes with nilpotent n, which implies u + n is a unit. This implies e = 1. If u = 0 then e is nilpotent, but e = e2 so e = 0. Therefore e is trivial if it is central. (cid:3) The above remark allows us to conclude that a semisimple ring, which is always isomorphic to the direct product of matrix rings over division rings of various shapes and sizes, is nil-good if and only if it is simple. Observing that if R is a left artinian ring then R/J(R) is semisimple [4], we arrive at the following result. Proposition 4.5. If R is a left artinian ring, then R is nil-good if and only if J(R) is maximal. Proof. Suppose R is left artinian. Then R/J(R) is semisimple, and therefore R ≃ Mn1 (D1) × · · · × Mnr (Dr) for some division rings D1, ..., Dr and positive intgeres n1, ..., nr. However, any nontrivial direct product contains central idempotents. So if R is nil-good it must be that R/J(R) is isomorphic to a matrix ring over a division ring, which is a simple semisimple ring. The quotient R/J(R) is simple if and only if the ideal J(R) is maximal. Conversely, J(R) is nil if R is artinian, and if J(R) is maximal then we know R/J(R) is a simple semisimple ring. As shown for the first direction, we can conclude R/J(R) ≃ Mn(D) for some division ring D and natural number n. Then by Remark 4.1, the nil-good decompositions of R/J(R) lift modulo J(R). Thus R is a nil-good ring. (cid:3) We may also establish a few facts about matrices over nil-good rings. If R is a simple artinian ring, then Mn(R) ≃ Mn(Mk(D)) and since Mn(Mk(D)) ≃ Mnk(D) we conclude that any matrix ring over a simple artinian ring is nil-good. Corollary 4.6. If R is a local ring such that J(R) is nil, then Mn(R) is nil-good if and only if Mn(J(R)) the maximal ideal of Mn(R) is a nil ideal. Proof. Since R is local R/J(R) is a division ring, thus Mn(R/J(R)) is nil-good. Since Mn(J(R)) the maximal ideal of Mn(R) is nil, by above remarks Mn(R) is nil-good. Conversely, if Mn(R) is nil-good then by above proposition, its maximal ideal Mn(J(R)) is nil. (cid:3) The above remark is included because although we would like to say that R local and J(R) nil implies that Mn(R) is nil-good, this only holds if the Kothe conjecture does as well. 12 BLOCK GORMAN AND SHIAO 5. The Nil-Good Clean Property We now define a new property “nil-good clean,” which is slightly weaker than clean or nil-good in general. An element is nil-good clean if it can be written as the sum of a nilpotent, idempotent and a unit. A ring is nil-good clean if all its elements are nil-good clean. Observe that in the commutative case, nil-good clean is equivalent to the property clean. Let R be a commutative ring that is nil-good clean. Then any element a ∈ R can be written as the sum of a nilpotent n, an idempotent e and a unit u. Since u′ = u + n is always a unit in a commutative ring, a = u′ + e. Therefore, a is clean. For the opposite direction, if R is a clean commutative ring, then any element a can be written as the sum of a unit u and an idempotent e. It follows that by letting the nilpotent n be zero, we have the nil-good clean decomposition a = e + u + 0. We have found one example of a nil-good clean ring that is not clean. It is a subring of lower- triangular column-finite matrices. We say a matrix is “diagonal-finite” if there is some fixed n such that only the first n sub-diagonals below the main diagonal contain nonzero entries for some non- negative integer n. The set of lower-triangular matrices with this property form a ring. We denote the ring of column-finite matrices as CFMN(R) and denote the element of this ring by A = (aij)∞ for which (aij)j are the rows, and (aij )i are the columns. i,j=1 Definition 5.1. We denote the ring of lower-triangular diagonal-finite matrices over a ring R by LTDFMN(R) = {A = (aij ) ∈ CFMN(R) there exists n ∈ N such that aij = 0 for all i ≥ j+n, j ≥ 1}. To see that this is indeed a subring, note that if A is a lower triangular matrix with n nonzero subdiagonals and B is a lower triangular matrix with m nonzero subdiagonals then in the product AB, each column (abij)i will have abij = 0 above the diagonal, and abij = 0 if i ≥ j + n + m for all j ≥ 1. Therefore AB can only have at most n + m nonzero diagonals below the main diagonal, so the set is closed under multiplication. Clearly it is also a group with addition, and satisfies the usual ring axioms. Lemma 5.2. For every idempotent E in LTDFMN(R) and i ∈ N, eii is idempotent. Moreover, eii = 1 for all i implies E = I Proof. As E is idempotent, E2 = E and eiieii = eii. Thus eii is idempotent. We will show by induction on subdiagonls that eii = 1 for all i implies that the elements on the nth subdiagonal are zero. Consider the first subdiagonal as the base case. By the idempotence of E, when we consider the (i − 1)th elment on the ith row, we have the equation ei,i−1(1) + (1)ei,i−1 = ei,i−1. Then 2ei,i−1 = ei,i−1 and ei,i−1 = 0. This shows that all the elements on the first subdiagonal are zero. Suppose that all elements on or above the kth diagonal, except the main diagonal, are zero. We will now show that all elements on the (k + 1)th diagonal are also zero, i.e. ei,i−l = 0 for any natural number l ≤ k and any i. Consider the (i − k − 1)th element on the ith row. By the idempotence of E, ei,i−k−1 = ei,i−k−1ei−k−1,i−k−1 + ei,i−kei−k,i−k−1 + · · · + ei,i−1ei−1,i−k−1 + ei,iei,i−k−1 = ei,i−k−1(1) + (0)ei−k,i−k−1 + · · · + (0)ei−1,i−k−1 + (1)ei,i−k−1 = 2ei,i−k−1 and thus ei,i−k−1 = 0. Therefore every element on the (k + 1)th subdiagonal is zero. By induction, all the elements below the main diagonal are zero and E = I. (cid:3) Lemma 5.3. For every unit U in LTDFMN(R) and i ∈ N, uii is a unit. Proof. Since U is invertible, there exists A ∈ LTDFMN(R) such that U A = I. Then uiiaii = 1 for any i ∈ N and thus each uii is a unit. (cid:3) NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS 13 Theorem 5.4. The ring LTDFMN(R) is not exchange. Proof. For contradiction suppose that LTDFMN(R) is exchange. Consider the matrix A = 1 0  1 1 0 1 ... ... ...  0 ... ... · · · 0 1 1 0 ... · · · 0 1 1 . . . · · · 0 1 . . .   · · · . . . . . . We assume there exists an idempotent matrix E such that E ∈ (LTDFMN(R))A and I − E ∈ (LTDFMN(R))(I − A). Note that I − E = B(I − A) for some B ∈ LTDFMN(R). Since I − A =  0 1 0 ... ... ...  0 · · · 0 0 1 0 0 ... ... 1 0 ... · · · 0 0 1 . . . · · · 0 0 . . .   · · · . . . . . . and (I − A)ii =0 for all i, (I − E)ii = Bii((I − A)ii) = Bii(0) = 0. Therefore Eii = 1 for all i. As proved in the lemma earlier that E with main diagonal of ones implies that E = I and therefore I ∈ (LTDFMN(R))A. This means that there exists C ∈ LTDFMN(R) such that CA = I and so A is invertible. The inverse of A has to have ones and negative ones alternating infinitely in each column, and thus is not in LTDFMN(R). This is a contradiction so LTDFMN(R) is not exchange. (cid:3) Corollary 5.5. LTDFMN(R) is not clean for any unital ring R Proposition 5.6. The ring LTDFMN(R) is nil-good clean if R is a clean ring. Proof. Let A = (aij) ∈ LTDFMN(R) be an infinite matrix that has n nonzero diagonals on or below the main diagonal. We can write A as the sum of two block-diagonal elements of LTDFMN(R). We define the first summand D = L Dk. It is the direct sum of 2n × 2n blocks, thus for all k ∈ N define Dk = (dij )2n(k+1) i,j=1+2kn where dij = aij if 2kn < i, j ≤ 2(k + 1)n. We define the second summand N as the direct sum of the n × n zero matrix and L Nk where for all k ∈ N we define Nk = (nij )(2k+3)n i,j=1+(2k+1)n where nij = aij if 2(k + 1)n < i ≤ (2k + 3)n and (2k + 1)n < j < 2(k + 1)n. 14 BLOCK GORMAN AND SHIAO ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ A = 2n 2n                   . . . ∗ ∗ ∗ ∗ ∗   ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ D =   2n ∗ ∗ ∗ ∗ ∗     ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗   ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗     . . . 2n ∗ ∗ ∗ ∗ ∗     ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ NIL-GOOD AND NIL-GOOD CLEAN MATRIX RINGS 15   n  0   2n     ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗   N =   2n     ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗     . . . Note that N is a direct sum of finite nilpotent matrices, and therefore is nilpotent itself, and D is the direct sum of finite matrices over a clean ring, so each Dk has a clean decomposition, and therefore the direct sum of the units that decompose each Dk is a unit, the direct sum of the idempotents that decompose each Dk is an idempotent, and the sum of those two direct sums form a clean decomposition of D. Therefore LTDFMN(R) is nil-good clean. (cid:3) Question 1: Is there a nil-good clean ring that is exchange but not clean? We suspect that there is little overlap between nil-good clean rings that are not clean and exchange rings. Acknowledgement The authors would like to thank Alexander J. Diesl for his guidance and contributions to our research. References 1. Paul M. Cohn, The similarity reduction of matrices over a skew field, Math. Z. 132 (1973), 151–163. 2. Peter Danchev, Nil-good unital rings, pre-print, 2015. 3. David Handelman, Perspectivity and cancellation in regular rings, Journal of Algebra 48 (1977), no. 1, 1–16. 4. Tsit-Yuen Lam, A first course in noncommutative rings, vol. 131, Springer Science & Business Media, 2001. 5. W. K. Nicholson, Lifting idempotents and exchange rings, Trans. Amer. Math. Soc. 229 (1977), 269–278. 6. Peter V´amos, 2-good rings, Q. J. Math. 56 (2005), no. 3, 417–430. 7. Kenneth G. Wolfson, An ideal-theoretic characterization of the ring of all linear transformations, Amer. J. Math. 75 (1953), 358–386. 8. Daniel Zelinsky, Every linear transformation is a sum of nonsingular ones, Proc. Amer. Math. Soc. 5 (1954), 627–630. 16 BLOCK GORMAN AND SHIAO Department of Mathematics, Wellesley College, Wellesley, MA 02481 E-mail address: [email protected] Department of Mathematics, Wellesley College, Wellesley, MA 02481 E-mail address: [email protected]
1907.06942	1	1907	2019-07-16T11:23:51	Spectral properties for a type of heptadiagonal symmetric matrices	[ "math.RA" ]	In this paper we express the eigenvalues of a sort of real heptadiagonal symmetric matrices as the zeros of explicit rational functions establishing upper and lower bounds for each of them. From these prescribed eigenvalues we compute also eigenvectors for these type of matrices. A formula not depending on any unknown parameter for the determinant and the inverse of these heptadiagonal matrices is still provided.	math.RA	math	Spectral properties for a type of heptadiagonal symmetric matrices Joao Lita da Silva1 Department of Mathematics and GeoBioTec Faculty of Sciences and Technology NOVA University of Lisbon Quinta da Torre, 2829-516 Caparica, Portugal Abstract In this paper we express the eigenvalues of a sort of real heptadiagonal symmetric matrices as the zeros of explicit rational functions establishing upper and lower bounds for each of them. From these prescribed eigenvalues we compute also eigenvectors for these type of matrices. A formula not depending on any unknown parameter for the determinant and the inverse of these heptadiagonal matrices is still provided. Key words: Heptadiagonal matrix, eigenvalue, eigenvector, determinant, inverse matrix 2010 Mathematics Subject Classification: 15A18, 15A15, 15A09. 1 Introduction The main goal of this paper is to express the eigenvalues of the following n×n real heptadiagonal matrix 9 1 0 2 l u J 6 1 ] . A R h t a m [ 1 v 2 4 9 6 0 . 7 0 9 1 : v i X r a Hn = (1.1) ξ η c d 0 ... ... ... ... ... 0 η a b c d . . . c b a b c . . . . . . d c b a b . . . . . . . . . 0 d c b a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . a b c d 0 . . . . . . . . . b a b c d . . . . . . c b a b c . . . d c b a η 0 ... ... ... ... ... 0 d c η ξ   as the zeros of explicit rational functions giving also upper and lower bounds non-depending of any unknown parameter to each of them. Further, we shall compute eigenvectors for these sort of matrices at the expense of the prescribed eigenvalues. The matrices of the form (1.1) fall into a general class of matrices called band matrices (see [13], page 13) which are widely used in several 1E-mail address: [email protected] 1 areas of science and engineering such as numerical solution of ordinary and partial differential equations (ODE and PDE), interpolation problems, boundary value problems among others (see, for instance, [2], [5], [7], [8], [11], [14]). To accomplish our purpose and in a first stage, we shall exploit the so-called modification tech- nique founded by Fasino in [6] for matrices of the type (1.1) in order to decompose them into an orthogonal block diagonalization and, at a second stage, use results concerning to a secular equa- tion of diagonal matrices perturbed by the addition of rank-one matrices developed by Anderson in the nineties (see [1]). Our decomposition will also lead us to explicit formulae for the determinant and the inverse of complex heptadiagonal matrices (1.1) assuming, of course, its nonsingularity. 2 Auxiliary tools Consider the class of matrices defined by An :=nAn ∈ Mn(C) : [An]k−1,ℓ + [An]k+1,ℓ = [An]k,ℓ−1 + [An]k,ℓ+1 , k, ℓ = 1, . . . , no where it is assumed [An]n+1,ℓ = [An]k,n+1 = [An]0,ℓ = [An]k,0 = 0 for all k, ℓ. We begin by gather two results announced in [3], presenting them for complex matrices. The proofs do not suffer any changes from the original ones and so we omit the details. Lemma 1 (a) The class An is the algebra generated over C by the n × n matrix Ωn = . (2.1)  0 1 1 0 1 . . . 0 ... ... ... 0 . . . 0 1 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 1 0 . . . 1 0 1 0 ... ... ... 0 1 0  (b) If An ∈ An and a⊤ is its first row then An = n−1Xk=0 ωk+1Ωk n where Ωn is the n×n matrix (2.1) and ω⊤ = (ω1, ω2, . . . , ωn) is the solution of the upper triangular system Unω = a, with [Un]0,0 := 1, [Un]0,ℓ := 0 for all 1 6 ℓ 6 n, [Un]k,ℓ :=(cid:26) [Un]k−1,ℓ−1 + [Un]k+1,ℓ−1 , otherwise 0, 1 6 k 6 ℓ 6 n . Throughout, n will denote an integer greater or equal to 5 and Sn will be the n× n symmetric, involutory and orthogonal matrix defined by [Sn]k,ℓ :=r 2 n + 1 n + 1(cid:19) . sin(cid:18) kℓπ (2.2) 2 Our second auxiliary result provide us an orthogonal diagonalization for the following n×n complex heptadiagonal symmetric matrix bHn = a − c b − d b − d a b c d . . .  c d 0 ... ... ... ... ... 0 c b a b c . . . . . . d c b a b . . . . . . . . . 0 d c b a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . a b c d 0 . . . . . . . . . b a b c d . . . . . . c b a b c . . . d c b . . . . . . . . . . . . 0 ... ... ... ... ... 0 d c . (2.3)  b − d a b − d a − c k = 1, . . . , n. (2.4) Lemma 2 Let n > 5 be an integer, a, b, c, d ∈ C and λk = a + 2b cos(cid:16) kπ If bHn is the n × n matrix (2.3) then n+1(cid:17) + 2c cos(cid:16) 2kπ n+1(cid:17) + 2d cos(cid:16) 3kπ n+1(cid:17) , where Λn = diag (λ1, . . . , λn) and Sn is the matrix (2.2). bHn = SnΛnSn Proof. Suppose an integer n > 5 and a, b, c, d ∈ C. Since bHn ∈ An and its first row is a⊤ = (a − c, b − d, c, d, 0, . . . , 0) we have, from Lemma 1, n + d Ω3 n. where Using the spectral decomposition Ωn = bHn = (a − 2c)In + (b − 3d)Ωn + c Ω2 2 cos(cid:18) ℓπ n + 1(cid:19) sℓ s⊤ℓ , nXℓ=1 q 2 n+1(cid:17) n+1 sin(cid:16) ℓπ n+1(cid:17) n+1 sin(cid:16) 2ℓπ q 2 q 2 n+1(cid:17) n+1 sin(cid:16) nℓπ n + 1(cid:19) + 4c cos2(cid:18) ℓπ nXℓ=1(cid:20)(a − 2c) + 2(b − 3d) cos(cid:18) ℓπ nXℓ=1   sℓ = ... bHn = = λksℓ s⊤ℓ = SnΛnSn (i.e. the ℓth column of Sn), it follows n + 1(cid:19) + 8d cos3(cid:18) ℓπ n + 1(cid:19)(cid:21) sℓ s⊤ℓ where Λn = diag (λ1, . . . , λn) and Sn is the matrix (2.2). The proof is completed. (cid:3) 3 The following statement is an orthogonal block diagonalization for matrices Hn of the form (1.1) extending Proposition 3.1 in [3] which is valid only for heptadiagonal symmetric Toeplitz matrices. Lemma 3 Let n > 5 be an integer, a, b, c, d, ξ, η ∈ C, λk, k = 1, . . . , n be given by (2.4) and Hn the n × n matrix (1.1). (a) If n is even, x =  v = and then    ... ... ... ... , y =     n+1(cid:17) 2√n+1 sin(cid:16) π n+1(cid:17) 2√n+1 sin(cid:16) 3π 2√n+1 sinh (n−1)π n+1 i n+1(cid:17) 2√n+1 sin(cid:16) 2π n+1(cid:17) 2√n+1 sin(cid:16) 4π n+1(cid:17) 2√n+1 sin(cid:16) nπ Hn = SnPn(cid:20) Φ n 2√n+1 sin(cid:16) 2π n+1(cid:17) n+1(cid:17) 2√n+1 sin(cid:16) 6π 2√n+1 sinh (2n−2)π n+1 i n+1(cid:17) 2√n+1 sin(cid:16) 4π n+1(cid:17) 2√n+1 sin(cid:16) 8π n+1(cid:17) 2√n+1 sin(cid:16) 2nπ 2 (cid:21) P⊤n Sn [Pn]k,ℓ =( 1 if k = 2ℓ − 1 or k = 2ℓ − n O O Ψ n 0 otherwise , w = 2 where Sn is the n × n matrix (2.2), Pn is the n × n permutation matrix defined by (2.5a) (2.5b) (2.5c) and Φ n 2 Ψ n 2 = diag (λ1, λ3, . . . , λn−1) + (c + ξ − a)xx⊤ + (d + η − b)xy⊤ + (d + η − b)yx⊤ = diag (λ2, λ4, . . . , λn) + (c + ξ − a)vv⊤ + (d + η − b)vw⊤ + (d + η − b)wv⊤. (2.5d) (2.5e) (b) If n is odd, and x = v =  ...  n+1(cid:17) 2√n+1 sin(cid:16) π n+1(cid:17) 2√n+1 sin(cid:16) 3π n+1(cid:17) 2√n+1 sin(cid:16) nπ n+1(cid:17) 2√n+1 sin(cid:16) 2π 2√n+1 sin(cid:16) 4π n+1(cid:17) n+1 i 2√n+1 sinh (n−1)π ...   , y = , w = 4   ...  n+1(cid:17) 2√n+1 sin(cid:16) 2π n+1(cid:17) 2√n+1 sin(cid:16) 6π n+1(cid:17) 2√n+1 sin(cid:16) 2nπ n+1(cid:17) 2√n+1 sin(cid:16) 4π 2√n+1 sin(cid:16) 8π n+1(cid:17) n+1 i 2√n+1 sinh 2(n−1)π ... (2.6a) (2.6b)  then Hn = SnPn" Φ n+1 2 O 2 # P⊤n Sn O Ψ n−1 where Sn is the n × n matrix (2.2), Pn is the n × n permutation matrix defined by [Pn]k,ℓ =( 1 if k = 2ℓ − 1 or k = 2ℓ − n − 1 0 otherwise and Φ n+1 2 Ψ n−1 2 = diag (λ1, λ3, . . . , λn) + (c + ξ − a)xx⊤ + (d + η − b)xy⊤ + (d + η − b)yx⊤ = diag (λ2, λ4, . . . , λn−1) + (c + ξ − a)vv⊤ + (d + η − b)vw⊤ + (d + η − b)wv⊤. (2.6c) (2.6d) (2.6e) Proof. Consider an integer n > 5, a, b, c, d, ξ, η ∈ C, λk, k = 1, . . . , n given by (2.4) and Hn the n × n matrix (1.1). Setting θ := c + ξ − a, ϑ := d + η − b, . . . 0 n + 1(cid:19)(cid:21)(cid:2)1 + (−1)k+ℓ(cid:3) . Since [Gn]k,ℓ = [Kn]k,ℓ = 0 whenever k + ℓ is odd, we can permute rows and columns of Λn + Gn + Kn according to the permutation matrices (2.5c) and (2.6c), yielding: for n even, Hn = SnPn(cid:20) Υ n 2 + θxx⊤ + ϑxy⊤ + ϑyx⊤ O + θvv⊤ + ϑvw⊤ + ϑwv⊤ (cid:21) P⊤n Sn, O ∆ n 2 5  bEn = c + ξ − a 0 ... ... 0 0 . . . . . . . . . . . . . . . . . . . . . . . . 0 0 0 ... ... 0 c + ξ − a  and bFn =  0 d + η − b d + η − b 0 ... ... 0 0 0 . . . . . . 0 0 . . . . . . . . . . . . . . . . . . . . . . . . 0 0 . . . . . . 0 0 d + η − b 0 ... ... 0 d + η − b 0  we have, from Lemma 2, SnHnSn = Sn(cid:16)bHn +bEn +bFn(cid:17) Sn = Λn + Gn + Kn where Sn is the n × n matrix (2.2), bHn is the n × n matrix (2.3), Λn = diag (λ1, . . . , λn) , [Gn]k,ℓ = 2θ n + 1 n + 1(cid:19) sin(cid:18) ℓπ sin(cid:18) kπ n + 1(cid:19) + sin(cid:18) 2kπ n + 1(cid:19) sin(cid:18) 2ℓπ n + 1(cid:19)(cid:2)1 + (−1)k+ℓ(cid:3) n + 1(cid:19) sin(cid:18) ℓπ and [Kn]k,ℓ = 2ϑ n + 1(cid:20)sin(cid:18) kπ where Pn is the matrix (2.5c), Υ n are given by (2.5a); for n odd, 2 = diag(λ1, λ3, . . . , λn−1), ∆ n 2 = diag(λ2, λ4, . . . , λn) and x, y Hn = SnPn" Υ n+1 2 + θxx⊤ + ϑxy⊤ + ϑyx⊤ O + θvv⊤ + ϑvw⊤ + ϑwv⊤ # P⊤n Sn, O ∆ n−1 2 with Pn defined in (2.6c), Υ n+1 defined by (2.6a). The proof is completed. 2 = diag(λ1, λ3, . . . , λn), ∆ n−1 2 = diag(λ2, λ4, . . . , λn−1) and v, w (cid:3) Remark Let us point out that the decomposition for real heptadiagonal symmetric Toeplitz ma- trices established in Proposition 3.1 of [3] at the expense of the bordering technique is no more useful for matrices having the shape (1.1). As consequence, some results stated by these authors will be necessarily extended, particularly, the referred decomposition and a formula to compute the determinant of real heptadiagonal symmetric Toeplitz matrices (Corollary 3.1 of [3]). 3 Main results 3.1 Determinant of Hn The orthogonal block diagonalization established in Lemma 3 will lead us to an explicit formula for the determinant of the matrix Hn. Theorem 1 Let n > 5 be an integer, a, b, c, d, ξ, η ∈ C, λk, k = 1, . . . , n be given by (2.4), n+1(cid:17), k = 1, . . . , 2n and Hn the n× n matrix (1.1). If θ := c + ξ − a, ϑ := d + η − b and (a) n is even then xk = sin(cid:16) kπ det(Hn) =" n 2Yj=1 " n 2Yj=1 λ2j−1 + n 2Xk=1 λ2j + n 2Xk=1 2Yj=1 n j6=k λ2j 4θx2 2k+8ϑx2kx4k (n+1) n 2Yj=1 j6=k 4θx2 2k−1+8ϑx2k−1x4k−2 (n+1) λ2j−1 λ2j n 2 2Yj=1 − X16k<ℓ6 n − X16k<ℓ6 n 2Yj=1 j6=k,ℓ n 2 j6=k,ℓ 16ϑ2(x2kx4ℓ−x2ℓx4k)2 (n+1)2 #× λ2j−1 16ϑ2(x2k−1x4ℓ−2−x2ℓ−1x4k−2)2 (n+1)2 #. (b) n is odd then 2 n−1 λ2j 4θx2 (n+1) (n+1) n+1 n+1 n−1 n−1 (n+1)2 4θx2 λ2j j6=k,ℓ n+1 λ2j + λ2j−1 2k+8ϑx2k x4k #× λ2j−1 + 2Xk=1 2Yj=1 j6=k 2Yj=1 j6=k 2k−1+8ϑx2k−1x4k−2 16ϑ2(x2kx4ℓ−x2ℓx4k)2 2Yj=1 2Yj=1 − X16k<ℓ6 n−1 − X16k<ℓ6 n+1 det(Hn) =" n−1 2Yj=1 " n+1 2Xk=1 2Yj=1 integer n > 5, a, b, c, d, ξ, η ∈ C, xk = sin(cid:16) kπ block-triangular matrices (see [9], page 185) and Lemma 3 ensure det(Hn) = det(cid:0)Φ n θ := c + ξ − a, ϑ := d + η − b and the notations used in Lemma 3. The determinant formula for We shall first assume λk 6= 0 for all k = 1, . . . , n, x2 2k−1 λ2k−1 6= −1 n+1(cid:17), k = 1, . . . , 2n, λk, k = 1, . . . , n given by (2.4), 2(cid:1) det(cid:0)Ψ n 2(cid:1). Proof. Since both assertions can be proven in the same way, we only prove (a). Consider an even 16ϑ2(x2k−1x4ℓ−2−x2ℓ−1x4k−2)2 λ2j−1 (3.1a) n + 1 2 j6=k,ℓ (n+1)2 4θ n 2Xk=1 #. 6 4θ n + 1 n 2Xk=1 4θx2 2k−1 + 8ϑx2k−1x4k−2 (n + 1)λ2k−1 − n + 4ϑ n + 1 x2k−1x4k−2 x2 2k−1 λ2k−1 16ϑ2 2Xk=1 (x2k−1x4ℓ−2 − x2ℓ−1x4k−2)2 (n + 1)2 X16k<ℓ6 n λ2k−1λ2ℓ−1 6= −1 λ2k−1 2 (3.1b) 6= −1 (3.1c) n 2Xk=1 and 4θ n + 1 n 2Xk=1 x2 2k λ2k 6= −1 n + 4ϑ x2 2k λ2k n + 1 2Xk=1 (n + 1)2 X16k<ℓ6 n 16ϑ2 2 x2kx4k 6= −1 λ2k (x2kx4ℓ − x2ℓx4k)2 λ2kλ2ℓ (3.2a) (3.2b) (3.2c) 6= −1. 4θ n + 1 n 2Xk=1 2k + 8ϑx2kx4k (n + 1)λ2k − n 4θx2 2Xk=1 2(cid:1) = det(cid:0)Υ n =(cid:20)1 + θx⊤Υ−1 =" n 2Xk=1 2Yj=1 λ2j−1 + n 2 n 2 n 2Yj=1 j6=k Putting Υ n 2 := diag (λ1, λ3, . . . , λn−1) and ∆ n 2 := diag (λ2, λ4, . . . , λn), we have det(cid:0)Φ n + θxx⊤ + ϑxy⊤ + ϑyx⊤(cid:1) x + 2ϑx⊤Υ−1 y + ϑ2(cid:16)x⊤Υ−1 n 2 y(cid:17)2 − ϑ2(cid:16)x⊤Υ−1 n 2 x(cid:17)(cid:16)y⊤Υ−1 n 2 n 2 y(cid:17)(cid:21) det(cid:0)Υ n 2(cid:1) 4θx2 2k−1+8ϑx2k−1x4k−2 (n+1) + λ2j−1 and − X16k<ℓ6 n 2 λ2j−1 16ϑ2(x2k−1x4ℓ−2−x2ℓ−1x4k−2)2 (n+1)2 # n 2Yj=1 j6=k,ℓ 2 det(cid:0)Ψ n 2(cid:1) = + θvv⊤ + ϑvw⊤ + ϑwv⊤(cid:1) = det(cid:0)∆ n =(cid:20)1 + θv⊤∆−1 =" n 2Xk=1 2Yj=1 v + 2ϑv⊤∆−1 2Yj=1 2k+8ϑx2kx4k λ2j + (n+1) 4θx2 λ2j n 2 n 2 n n j6=k n 2 w + ϑ2(cid:16)v⊤∆−1 − ϑ2(cid:16)v⊤∆−1 w(cid:17)2 2Yj=1 − X16k<ℓ6 n λ2j n 2 n 2 j6=k,ℓ n 2 w(cid:17)(cid:21) det(cid:0)∆ n v(cid:17)(cid:16)w⊤∆−1 2(cid:1) # 16ϑ2(x2kx4ℓ−x2ℓx4k)2 (n+1)2 (see [12], page 69 and 70), i.e. λ2j + n 2Xk=1 λ2j n 2Yj=1 j6=k 4θx2 2k+8ϑx2k x4k (n+1) det(Hn) =" n 2Yj=1 " n 2Xk=1 2Yj=1 λ2j−1 + n n 2Yj=1 j6=k 2 − X16k<ℓ6 n − X16k<ℓ6 n 2 n j6=k,ℓ 2Yj=1 2Yj=1 n j6=k,ℓ λ2j 16ϑ2(x2kx4ℓ−x2ℓx4k)2 (n+1)2 #· 4θx2 2k−1+8ϑx2k−1x4k−2 (n+1) λ2j−1 λ2j−1 16ϑ2(x2k−1x4ℓ−2−x2ℓ−1x4k−2)2 (n+1)2 #. (3.3) Since both sides of (3.3) are polynomials in the variables a, b, c, d, ξ, η, conditions (3.1a), (3.1b), (3.1c), (3.2a), (3.2b), (3.2c) as well as λk 6= 0 can be dropped and (3.3) is valid more generally. (cid:3) 7 3.2 Eigenvalue localization for Hn The next lemma will allow us to express the eigenvalues of key matrices in this paper as the zeros of explicit rational functions providing, additionally, explicit upper and lower bounds for each one. Throughout, k · k will denote the Euclidean norm. Lemma 4 Let n > 5 be an integer, a, b, c, d, ξ, η ∈ R and λk, k = 1, . . . , n be given by (2.4). (a) If n is even, i. x, y are given by (2.5a) and the eigenvalues of (3.4a) diag (λ1, λ3, . . . , λn−1) + (c + ξ − a)xx⊤ + (d + η − b)xy⊤ + (d + η − b)yx⊤ are not of the form a + 2b cosh (2k−1)π n+1 i + 2c cosh 2(2k−1)π then the eigenvalues of (3.4a) are precisely the zeros of the rational function n+1 2 n+1 i + 2d cosh 3(2k−1)π n+1 i + 2(d + η − b) sinh (2k−1)π n+1 i − sinh (4k−2)π n+1 i sinh (4ℓ−2)π i, k = 1, . . . , n n+1 i sinh (4k−2)π n+1 i n+1 i sinh (2ℓ−1)π n+1 io2 λ2k−1 − t + . (λ2k−1 − t)(λ2ℓ−1 − t) (c + ξ − a) sin2h (2k−1)π nsinh (2k−1)π f (t) = 1 + 4 n + 1 n 2Xk=1 − 16(d + η − b)2 (n + 1)2 X16k<ℓ6 n 2 are the eigenvalues of (3.4a) and λτ (1) 6 λτ (3) 6 . . . 6 Moreover, if µ1 6 µ2 6 . . . 6 µ n λτ (n−1) are arranged in non-decreasing order by some bijection τ defined in {1, 3, . . . , n − 1} then (c+ξ−a)+√(c+ξ−a)2+4(d+η−b)2 (c+ξ−a)−√(c+ξ−a)2+4(d+η−b)2 2 6 µk 6 λτ (2k−1) + 2 λτ (2k−1) + for each k = 1, . . . , n 2 . 2 ii. v, w are given by (2.5b) and the eigenvalues of eigenvalues of (3.5a) are precisely the zeros of the rational function are not of the form a + 2b cos(cid:16) 2kπ n+1(cid:17) + 2c cos(cid:16) 4kπ diag (λ2, λ4, . . . , λn) + (c + ξ − a)vv⊤ + (d + η − b)vw⊤ + (d + η − b)wv⊤ n+1(cid:17), k = 1, . . . , n n+1(cid:17) n+1(cid:17) sin(cid:16) 4kπ n+1(cid:17) sin(cid:16) 2ℓπ n+1(cid:17)i2 n+1(cid:17) + 2d cos(cid:16) 6kπ (c + ξ − a) sin2(cid:16) 2kπ n+1(cid:17) + 2(d + η − b) sin(cid:16) 2kπ hsin(cid:16) 2kπ n+1(cid:17) sin(cid:16) 4ℓπ n+1(cid:17) − sin(cid:16) 4kπ 2Xk=1 (n + 1)2 X16k<ℓ6 n (λ2k − t)(λ2ℓ − t) 16(d + η − b)2 λ2k − t g(t) = 1 + n + 1 − + 4 n 2 2 then the (3.5b) . are the eigenvalues of (3.5a) and λσ(2) 6 λσ(4) 6 . . . 6 Furthermore, if ν1 6 ν2 6 . . . 6 ν n λσ(n) are arranged in non-decreasing order by some bijection σ defined in {2, 4, . . . , n} then 2 λσ(2k) + (c+ξ−a)−√(c+ξ−a)2+4(d+η−b)2 2 (c+ξ−a)+√(c+ξ−a)2+4(d+η−b)2 2 (3.5c) (3.4b) (3.4c) (3.5a) for every k = 1, . . . , n 2 . (b) If n is odd, 6 νk 6 λσ(2k) + 8 i. x, y are given by (2.6a) and the eigenvalues of (3.6a) diag (λ1, λ3, . . . , λn) + (c + ξ − a)xx⊤ + (d + η − b)xy⊤ + (d + η − b)yx⊤ are not of the form a+2b cosh (2k−1)π n+1 i+2c cosh 2(2k−1)π n+1 then the eigenvalues of (3.6a) are precisely the zeros of the rational function 2 n+1 i+2d cosh 3(2k−1)π n+1 i + 2(d + η − b) sinh (2k−1)π n+1 i − sinh (4k−2)π n+1 i sinh (4ℓ−2)π i, k = 1, . . . , n+1 n+1 i sinh (4k−2)π n+1 i n+1 i sinh (2ℓ−1)π n+1 io2 λ2k−1 − t + (λ2k−1 − t)(λ2ℓ−1 − t) . (c + ξ − a) sin2h (2k−1)π nsinh (2k−1)π X16k<ℓ6 n+1 2 f (t) = 1 + 4 n + 1 n+1 2Xk=1 16(d + η − b)2 (n + 1)2 − (3.6b) 2 Moreover, if µ1 6 µ2 6 . . . 6 µ n+1 are the eigenvalues of (3.6a) and λτ (1) 6 λτ (3) 6 . . . 6 λτ (n) are arranged in non-decreasing order by some bijection τ defined in {1, 3, . . . , n} then (c+ξ−a)+√(c+ξ−a)2+4(d+η−b)2 λτ (2k−1) + for any k = 1, . . . , n+1 2 . (c+ξ−a)−√(c+ξ−a)2+4(d+η−b)2 6 µk 6 λτ (2k−1) + (3.6c) 2 2 ii. v, w are given by (2.6b) and the eigenvalues of eigenvalues of (3.7a) are precisely the zeros of the rational function n+1(cid:17) + 2c cos(cid:16) 4kπ are not of the form a + 2b cos(cid:16) 2kπ diag (λ2, λ4, . . . , λn−1) + (c + ξ − a)vv⊤ + (d + η − b)vw⊤ + (d + η − b)wv⊤ n+1(cid:17), k = 1, . . . , n−1 n+1(cid:17) sin(cid:16) 4kπ n+1(cid:17) n+1(cid:17) sin(cid:16) 2ℓπ n+1(cid:17)i2 n+1(cid:17) + 2(d + η − b) sin(cid:16) 2kπ n+1(cid:17) − sin(cid:16) 4kπ hsin(cid:16) 2kπ (c + ξ − a) sin2(cid:16) 2kπ n+1(cid:17) + 2d cos(cid:16) 6kπ n+1(cid:17) sin(cid:16) 4ℓπ 16(d + η − b)2 2Xk=1 λ2k − t g(t) = 1 + n + 1 + n−1 4 2 (n + 1)2 − . (3.7a) then the (λ2k − t)(λ2ℓ − t) X16k<ℓ6 n−1 2 (3.7b) Furthermore, if ν1 6 ν2 6 . . . 6 ν n−1 are the eigenvalues of (3.7a) and λσ(2) 6 λσ(4) 6 . . . 6 λσ(n−1) are arranged in non-decreasing order by some bijection σ defined in {2, 4, . . . , n− 1} then 2 λσ(2k) + (c+ξ−a)−√(c+ξ−a)2+4(d+η−b)2 2 6 νk 6 λσ(2k) + (c+ξ−a)+√(c+ξ−a)2+4(d+η−b)2 2 (3.7c) for all k = 1, . . . , n−1 2 . Proof. Suppose n > 5 an even integer, a, b, c, d, ξ, η ∈ R, λk, k = 1, . . . , n given by (2.4) and put θ := c + ξ − a, ϑ := d + η − b. We shall denote by S(k, m) the collection of all k-element subsets of {1, 2, . . . , m} written in increasing order; additionally, for any rectangular matrix M, we shall indicate by det (M[I, J]) the minor determined by the subsets I = {i1 < i2 < . . . < ik} and J = {j1 < j2 < . . . < jk}. Supposing θ 6= 0, X = 2q θ n+1(cid:17) n+1 sin(cid:16) π 2q θ n+1(cid:17) n+1 sin(cid:16) π n+1(cid:17) sin(cid:16) 2π 2ϑ√θ(n+1) 2q θ n+1 sin(cid:16) 3π 2q θ n+1 sin(cid:16) 3π sin(cid:16) 6π n+1(cid:17) . . . n+1(cid:17) . . . n+1(cid:17) . . . 2ϑ√θ(n+1) 2q θ n+1(cid:17) n+1 sin(cid:16) nπ 2q θ n+1(cid:17) n+1 sin(cid:16) nπ n+1 i sinh (4n−2)π 2ϑ√θ(n+1)  9 and Y = 2ϑ√θ(n+1) 2q θ n+1(cid:17) n+1 sin(cid:16) π sin(cid:16) 2π n+1(cid:17) 2q θ n+1 sin(cid:16) π n+1(cid:17) 2ϑ√θ(n+1) 2q θ n+1 sin(cid:16) 3π sin(cid:16) 6π 2q θ n+1 sin(cid:16) 3π n+1(cid:17) . . . n+1(cid:17) . . . n+1(cid:17) . . . 2ϑ√θ(n+1) 2q θ n+1(cid:17) n+1 sin(cid:16) nπ sinh (4n−2)π n+1 i 2q θ n+1 sin(cid:16) nπ n+1(cid:17) ,  Theorem 1 of [1] ensures that ζ is an eigenvalue of (3.4a) if and only if 1 + n 2Xk=1 XJ∈S(k, n 2 ) XI∈S(k,3) det (X[I, J]) det (Y[I, J]) Qj∈J (λ2j−1 − ζ) = 0 provided that ζ is not an eigenvalue of diag (λ1, λ3, . . . , λn−1). Since 1 + n n 4 − 16ϑ2 n + 1 = 1 + 2Xk=1 XJ∈S(k, n 2 ) XI∈S(k,3) 2Xk=1 (n + 1)2 X16k<ℓ6 n X = Y = 2q ϑ n+1 sin(cid:16) π 2q ϑ n+1 sin(cid:16) 2π 2q ϑ n+1 sin(cid:16) 2π 2q ϑ n+1 sin(cid:16) π 2 = det (X[I, J]) det (Y[I, J]) Qj∈J (λ2j−1 − ζ) θ sin2h (2k−1)π nsinh (2k−1)π n+1 i + 2ϑ sinh (2k−1)π n+1 i sinh (4ℓ−2)π λ2k−1 − ζ n+1 i n+1 i sinh (4k−2)π n+1 i sinh (2ℓ−1)π n+1 i − sinh (4k−2)π n+1 io2 + (λ2k−1 − ζ)(λ2ℓ−1 − ζ) n+1(cid:17) 2q ϑ n+1(cid:17) 2q ϑ n+1(cid:17) 2q ϑ n+1(cid:17) 2q ϑ n+1 sin(cid:16) 3π n+1 sin(cid:16) 6π n+1 sin(cid:16) 6π n+1 sin(cid:16) 3π 2q ϑ n+1(cid:17) . . . n+1(cid:17) n+1 sin(cid:16) nπ n+1(cid:17) . . . 2q ϑ n+1 i n+1 sinh (4n−2)π n+1(cid:17) . . . 2q ϑ n+1 i n+1 sinh (4n−2)π 2q ϑ n+1(cid:17) . . . n+1(cid:17) n+1 sin(cid:16) nπ  ,  we obtain (3.4b). Considering now θ = 0 and setting we still have that ζ is an eigenvalue of (3.4a) if and only if 1 + n 2Xk=1 XJ∈S(k, n 2 ) XI∈S(k,2) det (X[I, J]) det (Y[I, J]) Qj∈J (λ2j−1 − ζ) = 0 assuming that ζ is not an eigenvalue of diag (λ1, λ3, . . . , λn−1). Hence, 1 + n 2Xk=1 XJ∈S(k, n 2 ) XI∈S(k,2) det (X[I, J]) det (Y[I, J]) Qj∈J (λ2j−1 − ζ) nsinh (2k−1)π n 8ϑ = 1 + 2Xk=1 n+1 i sinh (4ℓ−2)π sinh (2k−1)π n+1 i − sinh (4k−2)π n+1 i sinh (4k−2)π n+1 i n+1 i sinh (2ℓ−1)π n+1 io2 λ2k−1 − ζ n + 1 + (λ2k−1 − ζ)(λ2ℓ−1 − ζ) − 16ϑ2 (n + 1)2 X16k<ℓ6 n 2 and (3.4b) is established. Let µ1 6 µ2 6 . . . 6 µ n be the eigenvalues of (3.4a) and λτ (1) 6 λτ (3) 6 . . . 6 λτ (n−1) be arranged in non-decreasing order by some bijection τ defined in {1, 3, . . . , n− 1}. Thus, 2 λτ (2k−1) + λmin(cid:0)θxx⊤ + ϑxy⊤ + ϑyx⊤(cid:1) 6 µk 6 λτ (2k−1) + λmax(cid:0)θxx⊤ + ϑxy⊤ + ϑyx⊤(cid:1) (3.8) 10 where α± := . From the identities, n 2 −4ϑ2[(x⊤y)2−kxk2kyk2] Spec(cid:0)θxx⊤ + ϑxy⊤ + ϑyx⊤(cid:1) =(cid:8)0, α−, α+(cid:9) θkxk2+2ϑx⊤y±q(θkxk2+2ϑx⊤y)2 n + 1 (cid:21) = n + 1 (cid:21) , sin2(cid:20) (4k − 2)π sin2(cid:20) (2k − 1)π 2Xk=1 n + 1 (cid:21) = 0 sin(cid:20) (2k − 1)π 2Xk=1 n + 1 (cid:21) sin(cid:20) (4k − 2)π 2Xk=1 n + 1 = 4 n n for each k = 1, . . . , n ϑyx⊤ is 2 (see [10], page 242). Since the characteristic polynomial of θxx⊤ + ϑxy⊤ + det(cid:2)tI n 2 − θxx⊤ − ϑxy⊤ − ϑyx⊤(cid:3) = t we have that its spectrum is = t n 2 −2ht2 −(cid:0)θx⊤x + ϑy⊤x + ϑx⊤y(cid:1) t 2 −2nt2 −(cid:16)θ kxk2 + 2ϑ x⊤y(cid:17) t + ϑ2h(cid:0)x⊤y(cid:1)2 + ϑ2(cid:0)x⊤y(cid:1)(cid:0)y⊤x(cid:1) − ϑ2(cid:0)x⊤x(cid:1)(cid:0)y⊤y(cid:1)i − kxk2 kyk2io n (3.9) it follows kxk = kyk = 1 and x⊤y = 0. Hence, (3.8) and (3.9) yields (3.4c). The proofs of the remaining assertions are performed in the same way and so will be omitted. (cid:3) The next statement allows us to locate the eigenvalues of Hn providing also explicit bounds for each of them. Theorem 2 Let n > 5 be an integer, a, b, c, d, ξ, η ∈ R, λk, k = 1, . . . , n be given by (2.4) and Hn the n × n matrix (1.1). 2 and the (a) If n is even, the eigenvalues of Φ n 2 2 then the eigenvalues of Hn are eigenvalues of Ψ n 2 precisely the zeros of the rational functions f (t) and g(t) given by (3.4b) and (3.5b), respectively. Moreover, if µ1 6 µ2 6 . . . 6 µ n are the zeros of g(t) (counting multiplicities in both cases) then µk, k = 1, . . . , n 2 satisfy (3.4c) and (3.5c), respectively. in (2.5d) are not of the form λ2k−1, k = 1, . . . , n in (2.5e) are not of the form λ2k, k = 1, . . . , n are the zeros of f (t) and ν1 6 ν2 6 . . . 6 ν n 2 and νk, k = 1, . . . , n 2 2 in (2.6e) are not of the form λ2k, k = 1, . . . , n−1 2 in (2.6d) are not of the form λ2k−1, k = 1, . . . , n+1 (b) If n is odd, the eigenvalues of Φ n+1 and 2 then the eigenvalues the eigenvalues of Ψ n−1 2 of Hn are precisely the zeros of the rational functions f (t) and g(t) given by (3.6b) and (3.7b), respectively. Furthermore, if µ1 6 µ2 6 . . . 6 µ n+1 are the zeros of f (t) and ν1 6 ν2 6 . . . 6 ν n−1 are the zeros of g(t) (counting multiplicities in both cases) then µk, k = 1, . . . , n+1 and νk, k = 2 1, . . . , n−1 2 Proof. Suppose an integer n > 5, a, b, c, d, ξ, η ∈ R and λk, k = 1, . . . , n be given by (2.4). (a) According to Lemma 3 and the determinant formula for block-triangular matrices (see [9], page 185), the characteristic polynomial of Hn, for n even, is satisfy (3.6c) and (3.7c), respectively. 2 2 2 det (tIn − Hn) = det(cid:0)tI n 2 − Φ n 2(cid:1) det(cid:0)tI n 2 − Ψ n 2(cid:1) are given by (2.5d) and (2.5e), respectively, so that the thesis is a direct where Φ n consequence of Lemma 4. and Ψ n 2 2 (b) For n odd, we obtain det (tIn − Hn) = det(cid:16)tI n+1 2 − Φ n+1 2 (cid:17) det(cid:16)tI n−1 2 − Ψ n−1 2 (cid:17) where Φ n+1 Lemma 4. 2 and Ψ n−1 2 are given by (2.6d) and (2.6e), respectively. The conclusion follows from (cid:3) 11 3.3 Eigenvectors of Hn The decomposition presented in Lemma 3 allows us also to compute eigenvectors for Hn in (1.1). Theorem 3 Let n > 5 be an integer, a, b, c, d, ξ, η ∈ R, λk, k = 1, . . . , n be given by (2.4) and Hn the n × n matrix (1.1). (a) If n is even, Sn is the n× n matrix (2.2), Pn is the n× n permutation matrix (2.5c), the zeros of (3.5b) are µ1, . . . , µ n not of the form λ2k, k = 1, . . . , n 2 , of (3.4b) are not of the form λ2k−1, k = 1, . . . , n 2 , the zeros ν1, . . . , ν n 2 2 n 2Xj=1 (c + ξ − a) sin2h (2j−1)π  (c + ξ − a) sin(cid:16) 2jπ  2Xj=1 n+1 i + (d + η − b) sinh (2j−1)π n+1(cid:17) sin(cid:16) 4jπ n+1 i sinh (4j−2)π n+1 i  6= n+1(cid:17) + (d + η − b) sin2(cid:16) 4jπ n+1(cid:17)  6= µk − λ2j−1 νk − λ2j 4 n n + 1 n + 1 4 , and b 6= d + η then n+1 ) 2 sin( 2π √n+1(µk−λ1) + n+1 ) 2 sin( 6π √n+1(µk−λ3) + 2 sin[ (2n−2)π √n+1(µk−λn−1) + n+1 ] SnPn  n+1−4 8 8 n 2 n 2 n+1 n+1 Pj=1( (c+ξ−a) sin[ (2j−1)π Pj=1( (c+ξ−a) sin2[ (2j−1)π Pj=1( (c+ξ−a) sin[ (2j−1)π Pj=1( (c+ξ−a) sin2[ (2j−1)π n+1 n+1 n 2 n 2 n+1−4 ] sin[ (4j−2)π n+1 −λ2j−1 µk ]+(d+η−b) sin2[ (4j−2)π n+1 ]+(d+η−b) sin[ (2j−1)π n+1 ] sin[ (4j−2)π n+1 ] µk −λ2j−1 ] sin[ (4j−2)π n+1 −λ2j−1 µk ]+(d+η−b) sin2[ (4j−2)π n+1 ]+(d+η−b) sin[ (2j−1)π n+1 ] sin[ (4j−2)π n+1 ] µk −λ2j−1 ... 8 n 2 n+1 Pj=1( (c+ξ−a) sin[ (2j−1)π Pj=1( (c+ξ−a) sin2[ (2j−1)π n+1 n 2 n+1−4 ] sin[ (4j−2)π n+1 −λ2j−1 µk ]+(d+η−b) sin2[ (4j−2)π n+1 ]+(d+η−b) sin[ (2j−1)π n+1 ] sin[ (4j−2)π n+1 ] µk −λ2j−1 ] ) ] ) ] ) ) ) ) sin( π n+1 ) √n+1(µk−λ1) sin( 3π n+1 ) √n+1(µk−λ3) sin[ (n−1)π n+1 ] √n+1(µk−λn−1) (3.10a)  0 ... 0 12 is an eigenvector of Hn associated to µk, k = 1, . . . , n 2 and SnPn  0 ... 0 ... n+1−4 8 8 n 2 n 2 Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ n 2 n 2 n+1−4 n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk 8 n 2 Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ n 2 n+1−4 n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk # # # sin( 2π n+1 ) √n+1(νk−λ2) sin( 4π n+1 ) √n+1(νk−λ4) sin( nπ n+1 ) √n+1(νk−λn) # # # n+1 ) 2 sin( 4π √n+1(νk−λ2) + n+1 ) 2 sin( 8π √n+1(νk−λ4) + n+1 ) 2 sin( 2nπ √n+1(νk−λn) + (3.10b) is an eigenvector of Hn associated to νk, k = 1, . . . , n 2 . (b) If n is odd, Sn is the n× n matrix (2.2), Pn is the n× n permutation matrix (2.5c), the zeros of (3.7b) µ1, . . . , µ n+1 are not of the form λ2k, k = 1, . . . , n−1 2 , of (3.6b) are not of the form λ2k−1, k = 1, . . . , n+1 2 , the zeros ν1, . . . , ν n−1 2 2 n+1 2Xj=1 (c + ξ − a) sin2h (2j−1)π  (c + ξ − a) sin(cid:16) 2jπ  2Xj=1 n+1 i + (d + η − b) sinh (2j−1)π n+1(cid:17) sin(cid:16) 4jπ n+1 i sinh (4j−2)π n+1 i  6= n+1(cid:17) + (d + η − b) sin2(cid:16) 4jπ n+1(cid:17)  6= µk − λ2j−1 νk − λ2j n−1 4 n + 1 n + 1 4 , and b 6= d + η then n+1 ) 2 sin( 2π √n+1(µk−λ1) + n+1 ) 2 sin( 6π √n+1(µk−λ3) + 2 sin( 2nπ n+1 ) √n+1(µk−λn−1) + SnPn  ] sin[ (4j−2)π n+1 −λ2j−1 µk ]+(d+η−b) sin2[ (4j−2)π n+1 ]+(d+η−b) sin[ (2j−1)π n+1 ] sin[ (4j−2)π n+1 ] µk −λ2j−1 ] sin[ (4j−2)π n+1 −λ2j−1 µk ]+(d+η−b) sin2[ (4j−2)π n+1 ]+(d+η−b) sin[ (2j−1)π n+1 ] sin[ (4j−2)π n+1 ] µk −λ2j−1 ] sin[ (4j−2)π n+1 −λ2j−1 µk ]+(d+η−b) sin2[ (4j−2)π n+1 ]+(d+η−b) sin[ (2j−1)π n+1 ] sin[ (4j−2)π n+1 ] µk −λ2j−1 ] ) ] ) ] ) sin( π n+1 ) √n+1(µk−λ1) sin( 3π n+1 ) √n+1(µk−λ3) sin( nπ n+1 ) √n+1(µk−λn−1) ) ) ) n+1−4 n+1−4 8 8 n+1 n+1 2 2 n+1 n+1 n+1 Pj=1( (c+ξ−a) sin[ (2j−1)π Pj=1( (c+ξ−a) sin2[ (2j−1)π Pj=1( (c+ξ−a) sin[ (2j−1)π Pj=1( (c+ξ−a) sin2[ (2j−1)π ... n+1 n+1 n+1 2 2 n+1 8 2 2 n+1 n+1 Pj=1( (c+ξ−a) sin[ (2j−1)π Pj=1( (c+ξ−a) sin2[ (2j−1)π 0 ... 0 n+1 n+1−4 13 (3.11a)  is an eigenvector of Hn associated to µk, k = 1, . . . , n+1 2 and 0 ... 0 SnPn  n+1 ) 2 sin( 4π √n+1(νk−λ2) + n+1 ) 2 sin( 8π √n+1(νk−λ4) + 2 sin[ 2(n−1)π √n+1(νk−λn) + n+1 ] n+1−4 8 8 n−1 n−1 2 2 2 n−1 Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ ... n−1 2 n+1−4 8 n−1 2 Pj=1" (c+ξ−a) sin( 2jπ Pj=1" (c+ξ−a) sin( 2jπ n−1 2 n+1−4 n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk n+1 ) sin( 4jπ n+1 )+(d+η−b) sin2( 4jπ −λ2j n+1 ) νk (3.11b) sin( 2π n+1 ) √n+1(νk−λ2) sin( 4π n+1 ) √n+1(νk−λ4) sin[ (n−1)π n+1 ] √n+1(νk−λn) # # # # # # is an eigenvector of Hn associated to νk, k = 1, . . . , n−1 2 . Proof. Since both assertions can be proven in the same way, we only prove (a). Let n > 5 be even. We can rewrite the matricial equation (µkIn − Hn)q = 0 as SnPn" µkI n 2 − Φ n O 2 O 2 − Ψ n µkI n 2 # P⊤n Snq = 0 (3.12) where Sn is the matrix (2.2), Pn is the permutation matrix (2.5c), Φ n and (2.5e), respectively. Thus, 2 and Ψ n 2 are given by (2.5d) (cid:2)µkI n (cid:2)µkI n 2 − diag (λ1, λ3, . . . , λn−1) − (c + ξ − a)xx⊤ − (d + η − b)xy⊤ − (d + η − b)yx⊤(cid:3) q1 = 0, 2 − diag (λ2, λ4, . . . , λn) − (c + ξ − a)vv⊤ − (d + η − b)vw⊤ − (d + η − b)wv⊤(cid:3) q2 = 0, (cid:20) q1 q2 (cid:21) = P⊤n Snq that is, q1 = α(cid:2)µkI n 2 − diag (λ1, λ3, . . . , λn−1) − (c + ξ − a)xx⊤ − (d + η − b)xy⊤(cid:3)−1 q2 = 0 y, for α 6= 0 (see [4], page 41) and q = SnPn(cid:20) α(cid:2)µkI n 2 − diag (λ1, λ3, . . . , λn−1) − (c + ξ − a)xx⊤ − (d + η − b)xy⊤(cid:3)−1 0 y (cid:21) is a nontrivial solution of (3.12). Thus, choosing α = 1 we conclude that (3.10a) is an eigenvector of Hn associated to the eigenvalue µk. Similarly, from (νkIn − Hn)q = 0 we have SnPn" νkI n 2 − Φ n O 2 O 2 − Ψ n νkI n 2 # P⊤n Snq = 0 14 and q = SnPn(cid:20) α(cid:2)νkI n 2 − diag (λ2, λ4, . . . , λn) − (c + ξ − a)vv⊤ − (d + η − b)vw⊤(cid:3)−1 for α 6= 0, which is an eigenvector of Hn associated to the eigenvalue νk. 0 w (cid:21) , (cid:3) 3.4 Expression of H−1 n The orthogonal block diagonalization presented in Lemma 3 and Miller's formula for the inverse of the sum of nonsingular matrices lead us to an explicit expression for the inverse of Hn. Theorem 4 Let n > 5 be an integer, a, b, c, d, ξ, η ∈ C, λk, k = 1, . . . , n be given by (2.4) and Hn the n × n matrix (1.1). If λk 6= 0 for every k = 1, . . . , n, Hn is nonsingular and: (a) n is even then H−1 n = SnPn(cid:20) Q n O O R n 2 2 (cid:21) P⊤n Sn where Sn is the n × n matrix (2.2), Pn is the n × n permutation matrix (2.5c), Q n 2 = Υ−1 2 − n (d+η−b)+(d+η−b)2y⊤Υ −1 n 2 x ρ (d+η−b)2y⊤Υ −1 n 2 y−(c+ξ−a) ρ Υ−1 n 2 (cid:0)yx⊤ + xy⊤(cid:1) Υ−1 n 2 Υ−1 xx⊤Υ−1 + + (d+η−b)2x⊤Υ −1 n 2 n 2 n 2 x Υ−1 n 2 yy⊤Υ−1 n 2 , (3.13a) (b) n is odd then n 2 n 2 v+ v + 2(d + η − b)w⊤∆−1 w(cid:1)2 (d + η − b)2h(cid:0)v⊤∆−1 n = SnPn" Q n+1 H−1 −(cid:0)v⊤∆−1 2 # P⊤n Sn R n−1 2 O O n 2 v(cid:1)(cid:0)w⊤∆−1 n 2 (3.13d) w(cid:1)i . where Sn is the n × n matrix (2.2), Pn is the n × n permutation matrix (2.6c), Q n+1 2 =Υ−1 n+1 2 − (d+η−b)+(d+η−b)2y⊤Υ −1 n+1 x 2 (d+η−b)2y⊤Υ −1 n+1 2 ρ ρ y−(c+ξ−a) Υ−1 n+1 2 Υ−1 n+1 2 (cid:0)yx⊤ + xy⊤(cid:1) Υ−1 (d+η−b)2x⊤Υ 2 −1 n+1 n+1 2 xx⊤Υ−1 n+1 + ρ 2 15 + x Υ−1 n+1 yy⊤Υ−1 n+1 , 2 2 (3.14a) with Υ n 2 := diag (λ1, λ3, . . . , λn−1), x, y given by (2.5a), x + 2(d + η − b)y⊤Υ−1 ρ = 1 + (c + ξ − a)x⊤Υ−1 y(cid:1)2 (d + η − b)2h(cid:0)x⊤Υ−1 n 2 n 2 and R n 2 = ∆−1 2 − n (d+η−b)+(d+η−b)2w⊤∆ −1 n 2 v ρ ∆−1 n ρ x+ n 2 n 2 −(cid:0)x⊤Υ−1 2 (cid:0)wv⊤ + vw⊤(cid:1) ∆−1 n 2 + with ∆ n 2 := diag (λ2, λ4, . . . , λn), v, w given by (2.6a) and = 1 + (c + ξ − a)v⊤∆−1 n 2 (d+η−b)2w⊤∆ −1 n 2 w−(c+ξ−a) ρ ∆−1 vv⊤∆−1 + n 2 n 2 (d+η−b)2v⊤∆ ρ v −1 n 2 ∆−1 n 2 ww⊤∆−1 n 2 , (3.13c) x(cid:1)(cid:0)y⊤Υ−1 n 2 y(cid:1)i , (3.13b) with Υ n+1 2 := diag (λ1, λ3, . . . , λn), x, y given by (2.6a), ρ = 1 + (c + ξ − a)x⊤Υ−1 2 2 2 n+1 n+1 n+1 x+ (d + η − b)2h(cid:0)x⊤Υ−1 −(cid:0)x⊤Υ−1 x + 2(d + η − b)y⊤Υ−1 y(cid:1)2 2 (cid:0)wv⊤ + vw⊤(cid:1) ∆−1 ∆−1 (d+η−b)2v⊤∆ w−(c+ξ−a) −1 n−1 n−1 n+1 v ρ 2 2 2 −1 n−1 2 ∆−1 n−1 vv⊤∆−1 n−1 + 2 2 + n−1 (d+η−b)+(d+η−b)2w⊤∆ and R n−1 2 =∆−1 n−1 2 − (d+η−b)2w⊤∆ −1 n−1 2 ρ x(cid:1)(cid:0)y⊤Υ−1 2 n+1 (3.14b) y(cid:1)i , v ∆−1 n−1 ww⊤∆−1 n−1 , 2 2 (3.14c) with ∆ n−1 2 := diag (λ2, λ4, . . . , λn−1), v, w in (2.6b), = 1 + (c + ξ − a)v⊤∆−1 2 n−1 v + 2(d + η − b)w⊤∆−1 w(cid:1)2 (d + η − b)2h(cid:0)v⊤∆−1 n−1 2 2 n−1 −(cid:0)v⊤∆−1 2 n−1 v(cid:1)(cid:0)w⊤∆−1 2 n−1 w(cid:1)i . (3.14d) Proof. Consider an even integer n > 5, a, b, c, d, ξ, η ∈ C, λk 6= 0, k = 1, . . . , n be given by (2.4) and Hn in (1.1) nonsingular. Recall that if Hn is nonsingular then ρ and in (3.13b) and (3.13d), respectively, are both nonzero. Setting θ := c + ξ − a, ϑ := d + η − b and assuming that conditions (3.1a) and (3.1b) are satisfied (note that (3.1c) corresponds to ρ 6= 0) we have, from the main result of [12] (see [12], pages 69 and 70), ρ v+ + θxx⊤(cid:1)−1 2 (cid:0)Υ n + θxx⊤ + ϑxy⊤(cid:1)−1 + θxx⊤(cid:1)−1 =(cid:0)Υ n = 2 = Υ−1 n 2 − 1+θx⊤Υ −1 n 2 (cid:0)Υ n 2 and = Υ−1 n 2 − θ 1+θx⊤Υ x −1 n 2 Υ−1 n 2 xx⊤Υ−1 n 2 , ϑ − 1+ϑy⊤(cid:0)Υ n 2 θ x+ϑy⊤Υ x +θxx⊤(cid:1)−1 xx⊤Υ−1 n Υ−1 n 2 −1 n 2 x(cid:0)Υ n 2 2 − + θxx⊤(cid:1)−1 xy⊤(cid:0)Υ n 2 1+θx⊤Υ −1 n 2 ϑ x+ϑy⊤Υ x −1 n 2 + θxx⊤(cid:1)−1 xy⊤Υ−1 Υ−1 n 2 n 2 = 2 + θxx⊤ + ϑxy⊤ + ϑyx⊤(cid:1)−1 (cid:0)Υ n + θxx⊤ + ϑxy⊤(cid:1)−1 =(cid:0)Υ n 1+ϑx⊤(cid:0)Υ n ϑ+ϑ2y⊤Υ − + −1 n 2 x 2 Υ−1 n = Υ−1 n 2 − ρ ϑ 2 +θxx⊤+ϑxy⊤(cid:1)−1 2 (cid:0)yx⊤ + xy⊤(cid:1) Υ−1 n 2 y(cid:0)Υ n 2 + θxx⊤ + ϑxy⊤(cid:1)−1 xy⊤(cid:0)Υ n 2 + θxx⊤ + ϑxy⊤(cid:1)−1 ϑ2y⊤Υ −1 n 2 y−θ ρ + Υ−1 n 2 xx⊤Υ−1 n 2 + ϑ2x⊤Υ x −1 n 2 ρ Υ−1 n 2 yy⊤Υ−1 (3.15) n 2 , with Υ n supposing (3.2a) and (3.2b) (observe that (3.2c) is 6= 0) we obtain := diag (λ1, λ3, . . . , λn−1), x, y given by (2.5a) and ρ in (3.13b). 2 In the same way, = ∆−1 n 2 − θ 1+θv⊤∆ v −1 n 2 ∆−1 n 2 vv⊤∆−1 n 2 , (cid:0)∆ n 2 + θvv⊤(cid:1)−1 2 (cid:0)∆ n + θvv⊤ + ϑvw⊤(cid:1)−1 + θvv⊤(cid:1)−1 =(cid:0)∆ n = 2 = ∆−1 n 2 − 1+θv⊤∆ −1 n 2 ϑ − 1+ϑw⊤(cid:0)∆ n θ v+ϑw⊤∆ v 2 +θvv⊤(cid:1)−1 ∆−1 n 2 vv⊤∆−1 n −1 n 2 v(cid:0)∆ n 2 2 − + θvv⊤(cid:1)−1 vw⊤(cid:0)∆ n 2 1+θv⊤∆ −1 n 2 ϑ v+ϑw⊤∆ v −1 n 2 + θvv⊤(cid:1)−1 ∆−1 vw⊤∆−1 n 2 n 2 16 and (cid:0)∆ n 2 = + θvv⊤ + ϑvw⊤ + ϑwv⊤(cid:1)−1 + θvv⊤ + ϑvw⊤(cid:1)−1 =(cid:0)∆ n 1+ϑv⊤(cid:0)∆ n 2 ϑ+ϑ2w⊤∆ − +θvv⊤+ϑvw⊤(cid:1)−1 −1 n 2 + ϑ v 2 = ∆−1 n 2 − w(cid:0)∆ n 2 + θvv⊤ + ϑvw⊤(cid:1)−1 ϑ2w⊤∆ vw⊤(cid:0)∆ n 2 + θvv⊤ + ϑvw⊤(cid:1)−1 ∆−1 n 2 (cid:0)wv⊤ + vw⊤(cid:1) ∆−1 n 2 + −1 n 2 w−θ ∆−1 n 2 vv⊤∆−1 n 2 ϑ2v⊤∆ v −1 n 2 + ∆−1 n 2 ww⊤∆−1 (3.16) n 2 , 2 := diag (λ2, λ4, . . . , λn), v, w given by (2.6a) and in (3.13d). Since the nonsingularity where ∆ n of Hn and λk 6= 0, for all k = 1, . . . , n are sufficient for the both sides of (3.15) and (3.16) to be well-defined, conditions (3.1a), (3.1b), (3.2a) and (3.2b) previously assumed can be dropped. Hence, the block diagonalization provided in (a) of Lemma 3 together with 8.5b of [9] (see page 88) establish the thesis in (a). The proof of (b) is analogous, so that we will omit the details. (cid:3) Acknowledgements. This work is a contribution to the Project UID/GEO/04035/2013, funded by FCT - Funda¸cao para a Ciencia e a Tecnologia, Portugal. References [1] J. Anderson, A secular equation for the eigenvalues of a diagonal matrix perturbation, Linear Algebra Appl. 246 (1996) 49 -- 70. [2] S.O. Asplund, Finite boundary value problems solved by Green's matrix, Math. Scand. 7 (1959) 49 -- 56. [3] D. Bini and M. Capovani, Spectral and computational properties of band symmetric Toeplitz matrices, Linear Algebra Appl. 52/53 (1983) 99 -- 126. [4] J.R. Bunch, C.P. Nielsen, D.C. Sorensen, Rank-one modification of the symmetric eigen- problem, Numer. Math. 31 (1978) 31 -- 48. [5] S. Demko, Inverses of band matrices and local convergence of spline projections, SIAM J. Numer. Anal. 14(4) (1977) 616 -- 619. [6] D. Fasino, Spectral and structural properties of some pentadiagonal symmetric matrices, Calcolo 25(4) (1988) 301 -- 310. [7] C.F. Fischer and R.A. Usmani, Properties of some tridiagonal matrices and their application to boundary value problems, SIAM J. Numer. Anal. 6(1) (1969) 127 -- 142. [8] S. Haley, Solution of band matrix equations by projection-recurrence, Linear Algebra Appl. 32 (1980) 33 -- 48. [9] D.A. Harville, Matrix Algebra From a Statistician's Perspective, Springer-Verlag, 1997. [10] R.A. Horn, C.R. Johnson, Matrix Analysis (second edition), Cambridge University Press, 2013. 17 [11] A.J. Keeping, Band matrices arising from finite difference approximations to a third order partial differential, SIAM J. Numer. Anal. 7(1) (1970) 142 -- 156. [12] K.S. Miller, On the inverse of the sum of matrices, Math. Mag. 54(2) (1981) 67 -- 72. [13] S. Pissanetsky, Sparse Matrix Technology, Academic Press, 1984. [14] R.A. Usmani, T.H. Andres, D.J. Walton, Error estimation in the integration of ordinary differential equations, Int. J. Comput. Math. 5 (1976) 241 -- 256. 18
1208.4087	2	1208	2013-03-01T17:37:02	Classification of the direct limits of involution simple associative algebras and the corresponding dimension groups	[ "math.RA" ]	A classification of (countable) direct limits of finite dimensional involution simple associative algebras over an algebraically closed field of arbitrary characteristic is obtained. This also classifies the corresponding dimension groups. The set of invariants consists of two supernatural numbers and two real parameters.	math.RA	math	Classification of the direct limits of involution simple associative algebras and the corresponding dimension groups A. A. Baranov Department of Mathematics University of Leicester Leicester LE1 7RH e-mail: [email protected] Abstract A classification of the (countable) direct limits of finite dimensional involution simple associative algebras over an algebraically closed field of arbitrary characteristic is obtained. This also classifies the corresponding dimension groups. The set of invariants consists of two supernatural numbers and two real parameters. 1 Introduction The ground field F is algebraically closed of arbitrary characteristic. Let A be an associative algebra over F (not necessarily containing an identity element). Assume A has an involution, that is, a linear transformation ∗ of A such that (a∗)∗ = a and (ab)∗ = b∗a∗ for all a, b ∈ A. We will sometimes denote this algebra by (A, ∗) to reflect the fact that A is an algebra with involution. Note that our involution is F-linear, i.e. we consider involutions of the first kind only. The algebra A is called involution simple if A2 6= 0 and it has no non-trivial ∗-invariant ideals. We say that an infinite dimensional algebra A is locally (semi)simple if any finite subset of A is contained in a finite dimensional (semi)simple subalgebra. Note that we do not require A to have an identity element. If A has an involution and these subalgebras can be chosen involution simple with respect to the inherited involution then A is called locally involution simple. Observe that A itself is involution simple in that case. The aim of this paper is to classify locally involution simple associative algebras over F of countable dimension. Let A be a locally simple associative algebra of countable dimension over F. It follows from the definition that there is a chain of simple subalgebras A1 ⊂ A2 ⊂ A3 ⊂ . . . of A such that A = ∪∞ i=1Ai. One can also view A as the direct limit lim−→ Ai for the sequence A1 → A2 → A3 → . . . (1) of injective homomorphisms of finite dimensional simple associative algebras Ai. Since F is algebraically closed, each Ai can be identified with the algebra Mni(F) of all ni × ni matrices over F for some ni. Moreover, each embedding Ai → Ai+1 can be written in the following matrix form M 7→ diag(M, . . . , M, 0, . . . , 0), M ∈ Mni(F). (2) 1 Therefore in order to describe locally simple associative algebras of countable dimension one needs to classify the direct limits of the sequences of matrix algebras (1). Elliot [6] did this in terms of systems of idempotents. It has been shown later that Elliot's invariant can be interpreted in terms of the K0-functor (see Theorem 3.8). As a particular case of our main results we get another parametrization of these algebras. Assume now that the algebra A is locally involution simple, i.e. we have a sequence (1) of involution simple finite dimensional algebras Ai and A = lim−→ Ai. Note that all homomorphisms in (1) respect the involution but do not necessarily preserve the identity element. It is well known that every involution simple finite dimensional F-algebra is either a full matrix algebra or the direct sum of two isomorphic matrix algebras. Therefore the combinatorial picture is much more complicated than in (2). However it is still possible to provide an explicit parametrization (see our main Theorems 4.1 and 5.2). In Section 3 we prove that two locally involution simple algebras of the same type (orthogonal, symplectic or special) are isomorphic if and only if they are isomorphic as associative algebras (Theorem 3.4). This partially reduces the classification problem to locally semisimple associative algebras. These algebras are normally classified by ordered dimension groups (see Theorem 3.8). However, as it is pointed out in [4], although dimension groups are relatively easy objects, their isomorphism classes are not, and general classification is not available. Some examples of known isomorphism classes of the dimension groups can be found in [4]. Our main Theorem 4.1 gives complete classification of the dimension groups which correspond to the locally involution simple algebras. These are the direct limits of the sequences Z3 → Z3 → · · · → Z3 → . . . where the embeddings are given by the Bratteli diagrams (21). Another approach (K-theoretical in nature) to the classification of locally involution simple associative algebras can be found within the general theory of compact group actions on locally semisimple algebras, see [11, 3]. In the case of order 2 automorphisms this was done by Fack and Mar´echal [9] (unital embeddings given by the Bratteli diagrams (20)) and Elliott and Su [7] (in terms of K-theoretical invariants). Our approach uses some technique developed by Baranov and Zhilinskii for the classification of the diagonal direct limits of finite dimensional simple Lie algebras over an algebraically closed field of characteristic zero [2]. It is shown in [1] that there is a natural bijective correspondence between such Lie algebras and locally involution simple associative algebras, so the classification should be similar. Unfortunately the proofs in [1, 2] are very dependent on characteristic zero and fail to work in positive characteristic. In the present paper we provide new, characteristic free, proofs. However, the case of characteristic 2 still requires special attention and the classification is slightly different in that case. Note that our results do not exhaust the problem of classification of all involution simple locally finite dimensional associative algebras (of countable dimension), since there are examples of such algebras which are not locally semisimple (see [8, 15]). 2 Preliminaries Recall that an associative algebra A with involution is called involution simple if A2 6= 0 and it has no non-trivial ∗-invariant ideals. The following is well-known. Proposition 2.1 Let A be an involution simple associative algebra. Then either A is simple as an algebra or A has exactly two non-zero proper ideals B1 and B2. Moreover both B1 and B2 2 are simple algebras, B∗ 1 = B2 and A = B1 ⊕ B2. Proof. Assume A is not simple. Let B1 be a non-zero proper ideal of A. Then B2 = B∗ 1 is also an ideal of A. Since B1 + B2 and B1 ∩ B2 are ∗-invariant ideals of A and A is involution simple, one has B1 + B2 = A and B1 ∩ B2 = 0, i.e. A = B1 ⊕ B2. Now, if B is a non-zero proper ideal of B1 then B ⊕ B∗ is a non-zero proper ∗-invariant ideal of B1 ⊕ B2 = A. Therefore B = B1 and both B1 and B2 are simple algebras. Assume now that C is another non-zero proper ideal of A. Then by the above argument, A = C ⊕ C ∗. If B1 ⊆ C or B2 ⊆ C then it is easy to see that C = B1 or B2. Assume this is not the case. Let B = B1 ∩ C. Then B + B∗ is a proper ∗-invariant ideal of A, so B = 0. In particular, B1C ⊆ B1 ∩ C = 0. Similarly, B2C = 0 and B1C ∗ = B2C ∗ = 0. This implies AA = (B1 + B2)(C + C ∗) = 0, which is a contradiction. Let A be a finite dimensional associative algebra over F with involution ∗. Assume that A is involution simple. Then by Proposition 2.1, A is either simple or A = B ⊕ B∗ the sum of two (anti)isomorphic simple subalgebras. Thus, we can identify A with either End V or End V1 ⊕ End V2 for some finite dimensional vector spaces V , V1, and V2 over F with dim V1 = dim V2. By fixing bases of V , V1, and V2, one can represent the algebras End V and End V1 ⊕ End V2 in the matrix forms Mn(F) and Mm(F) ⊕ Mm(F), respectively, where n = dim V and m = dim V1 = dim V2. We say that these are their matrix realizations. We say that a matrix realization of (End V, ∗) is canonical if the involution in the chosen basis has one of the following two forms: X 7→ X t, X 7→ X τ , X ∈ Mn(F) X ∈ Mn(F) (transpose); (symplectic transpose). (3) (4) In the latter case n is even and X τ = −JX tJ where J = diag(cid:18)(cid:18) 0 −1 0 (cid:19) , . . . ,(cid:18) 0 (n/2 blocks). We say that a matrix realization of (End V1⊕End V2, ∗) is canonical if the involution in the chosen basis has the following form: 1 −1 0 (cid:19)(cid:19) 1 (X1, X2) 7→ (X t 2, X t 1), X1, X2 ∈ Mm(F). (5) It is well known that any finite dimensional involution simple algebra over an algebraically closed field has a canonical matrix realization. Indeed, let us first consider the algebra End V . Let b : V × V → F be a nondegenerate symmetric or skew-symmetric bilinear form on V . For each x ∈ End V define αb(x) by the following property b(αb(x)v, w) = b(v, xw) for all v, w ∈ V. Then the map αb : End V → End V is an involution of the algebra End V , called the adjoint involution with respect to b. More exactly we have the following fact. Theorem 2.2 ([12, Ch.1, Introduction]) The map b 7→ αb induces a one-to-one correspon- dence between the equivalence classes of nondegenerate symmetric and skew-symmetric bilinear forms on V modulo multiplication by a factor in F× and involutions (of the first kind) on End V . 3 Recall that a bilinear form is called alternating if b(v, v) = 0 for all v ∈ V . Obviously, if char F 6= 2, then the form b is alternating if and only if it is skew-symmetric. If char F = 2, then b is alternating if and only if it is symmetric and for any choice of basis of V , all diagonal entries of the matrix of b are zeros. An involution α of End V is called symplectic (resp. orthogonal) if it is adjoint to an alternating (resp. symmetric non-alternating) bilinear form on V . Recall that each finite dimensional orthogonal (resp. symplectic) vector space over an algebraically closed field has an orthonormal (resp. hyperbolic) basis. That is, the matrix of b in this basis is either the identity (in the orthogonal case) or J (see above) in the symplectic case (see for example [14, Theorems 11.10 and 11.14]). It is easy to see that the adjoint involution in this basis is canonical, i.e. of the forms (3) and (4), respectively. Thus, we get the following well-known fact. Proposition 2.3 Let V be a vector space of dimension n over F and let ∗ be an involution of End V . Then the algebra (End V, ∗) has a canonical matrix realization. To prove a similar result for the algebra End V1 ⊕ End V2, we need the following simple fact. Proposition 2.4 Each involution of the matrix algebra Mn(F) is of the following form: X 7→ CX tC −1 where C is an invertible matrix. Proof. The matrix transpose X 7→ X t is a natural involution of Mn(F). Thus the map X 7→ (X ∗)t is an automorphism of Mn(F). By Skolem-Noether theorem each automorphism of Mn(F) is inner, i.e. there exists an invertible matrix K such that (X ∗)t = K −1XK. Therefore X ∗ = K tX t(K −1)t = K tX t(K t)−1, as required. Proposition 2.5 Let V1 and V2 be vector spaces of dimension m and let ∗ be an involution of the algebra End V1 ⊕ End V2 such that (End V1)∗ = End V2. Then for every matrix realization of End V1 there is a matrix realization of End V2 such that the corresponding matrix realization of (End V1 ⊕ End V2, ∗) is canonical. Proof. Fix any matrix realizations of End V1 and End V2, i.e. identify these algebras with the algebra Mm(F). Then the map ∗ : End V1 → End V2 gives an involution X 7→ X ∗ of Mm(F). By Proposition 2.4, X ∗ = CX tC −1. It remains to change basis of V2 (i.e. matrix realization of End V2), to eliminate C. Let A be an involution simple finite dimensional algebra over F. We say that A is of type S, or of symplectic type, if A is simple as an algebra and the involution is symplectic. Similarly we define the orthogonal type O. If A is not simple, then we say that A is of type A, or of special type. Note that algebras of type S are not isomorphic to those of type O (as algebras with involution). Thus the canonical matrix realizations (as in Propositions 2.3 and 2.5) give a complete classification of finite dimensional involution simple algebras over an algebraically closed field. Remark 2.6 We will also use other canonical forms for involutions. Let n be even. Define the following n × n matrices: J+ = diag(cid:18)(cid:18) 0 1 1 0 (cid:19) , . . . ,(cid:18) 0 1 1 0 (cid:19)(cid:19) , Q± =(cid:18) 0 I ±I 0 (cid:19) 4 where I is the identity n/2 × n/2 matrix. Then J+ and Q+ define nondegenerate bilinear symmetric forms on the natural Mn(F)-module. If char F 6= 2, these forms are non-alternating, so induce orthogonal involutions on Mn(F): τ+ : X 7→ J+X tJ+ and θ+ : X 7→ Q+X tQ+. In view of Proposition 2.3, by choosing an appropriate basis, these involutions can be represented as matrix transpose. Thus each orthogonal involution (for char F 6= 2 and algebras of even degree) can be represented as τ+ (resp. θ+) in a suitable basis. Similarly, the involution θ− : X 7→ −Q−X tQ− is symplectic and each symplectic involution (for any characteristic) can be represented in this form. The following simple fact will be used later. Lemma 2.7 Let B be a finite dimensional involution simple algebra of even dimension. Let e be the identity element of B. If char F = 2, assume that B is not of type O. Then B has idempotents f and g such that e = f + g, f g = gf = 0, and f ∗ = g. Proof. This is obvious if B is of type A. Assume that B is symplectic. Represent B as in Proposition 2.3. Then one can easily check that f = diag(1, 0, 1, 0, . . . , 1, 0) and g = diag(0, 1, 0, 1, . . . , 0, 1) are the required idempotents. If the involution ∗ of B is orthogonal, then by Remark 2.6, it can be represented as X ∗ = J+X tJ+, X ∈ Mn(F). Then it is easy to check that the same idempotents f and g as in the symplectic case satisfy the required conditions. Now we are going to study embeddings of involution simple algebras, i.e. injective homomor- phisms ε : A1 → A2 which respect involution. We do not require these embeddings to preserve the identity element. Since the embeddings respect involution we often use the same symbol "∗" to denote the involution of A1 and A2. We usually identify A1 with its image ε(A1) in A2. If Ai is of type A, we denote by Bi and Ci its simple components (so Ai = Bi ⊕ Ci and Bi ∼= Ci). It is convenient to assume Bi = Ai if Ai is of type S or O. We denote by ei, fi, and gi the identities of Ai, Bi, and Ci, respectively. Thus ei = fi + gi if Ai is of type A, and ei = fi otherwise. Note that f ∗ i = gi if Ai is of type A. Recall that Bi ∼= Mni(F) for some ni ∈ N. We say that ni is the degree of Ai. Denote by Vi the natural Bi-module of dimension ni and by Wi the natural module for Ci (if Ci 6= 0). We consider these modules as Ai-modules in a natural way. If Ai is not of type A, we denote by bi a nondegenerate bilinear form on Vi corresponding to the involution ∗ on Ai (see Theorem 2.2). Denote by Ti the trivial one-dimensional Ai-module (with zero action). Now the restriction of the A2-module V2 to A1 is completely reducible, so can be described as follows. V2↓A1 = V1 ⊕ · · · ⊕ V1 ⊕ W1 ⊕ · · · ⊕ W1 ⊕ T1 ⊕ · · · ⊕ T1 (6) where l, r, z ∈ N ∪ {0} and r = 0 if A1 is not of type A. l {z } r {z } z {z } Definition 2.8 The triple (l, r, z) in (6) is called the signature of the embedding ε : A1 → A2. Remark 2.9 If both A1 and A2 are of type A, then the signature depends on the choice of the simple components of A1 and A2, e.g. by swapping B1 and C1 (or B2 and C2, see (8) below), the signature (l, r, z) is replaced by (r, l, z). Thus we can and will assume that l ≥ r. 5 Definition 2.10 We say that a homomorphism ε : Mn1 → Mn2 of signature (l, 0, z) of two matrix algebras is canonical if ε(M ) = diag(M, . . . , M , 0, . . . , 0 ), M ∈ Mn1(F). (7) l {z z {z } } {z } z , 0, . . . , 0 } l {z } r {z We say that a homomorphism ε : Mn1 ⊕ Mn1 → Mn2 ⊕ Mn2 of signature (l, r, z) is canonical if ε(M, N ) = (diag(M, . . . , M , N, . . . , N ), diag(N, . . . , N , M, . . . , M , 0, . . . , 0 )) (8) l {z } r {z } z {z } for all M, N ∈ Mn1(F). We say that an embedding ε : A1 → A2 of finite dimensional involution simple algebras over F of the same type (A, O, or S) is (canonically) representable if for every canonical matrix realization of A1 there exists a canonical matrix realization of A2 such that the matrix embedding ε is canonical. Remark 2.11 (1) It is easy to see that canonical matrix homomorphisms (7)-(8) commute with the canonical matrix involutions (3)-(5) (e.g. in type O the canonical involution is just matrix transpose). (2) Note that compositions of canonical matrix homomorphisms are canonical. We are going to show that all embeddings of involution simple algebras of the same type are representable, except for types O and S in characteristic 2. Proposition 2.12 Let ε : A1 → A2 be an embedding of finite dimensional involution simple algebras over F of type A. Then ε is representable. Proof. Let ni be the degree of Ai. Fix any bases of V1 and W1 such that the corresponding matrix realization Mn1(F) ⊕ Mn1(F) of A1 is canonical (i.e. the involution has the form (5)). Let πB (resp. πC) denote the projection A2 → B2 (resp. A2 → C2). Fix any basis of V2 which agree with the bases of V1 and W1 and the decomposition (6), i.e. the projection πBε(A1) has the following matrix form. πBε(M, N ) = diag(M, . . . , M , N, . . . , N , 0, . . . , 0 ), M, N ∈ Mn1(F). l {z } r {z } z {z } Fix a basis of W2 such that the corresponding matrix realization of (A2, ∗) is canonical (see Proposition 2.5). Then ε(M, N ) = ε((N t, M t)∗) = (ε(N t, M t))∗ = ((πCε(N t, M t))t, (πBε(N t, M t))t), so Therefore as required. πCε(M, N ) = (πBε(N t, M t))t = diag(N, . . . , N , M, . . . , M , 0, . . . , 0 ). ε(M, N ) = (diag(M, . . . , M , N, . . . , N ), diag(N, . . . , N , M, . . . , M , 0, . . . , 0 )) {z } z , 0, . . . , 0 } l {z } {z l r {z } } z {z } } {z r z {z } l {z } r {z Our aim now is to prove a similar result for orthogonal and symplectic algebras in charac- teristic 6= 2. We need some auxiliary lemmas. 6 Lemma 2.13 ([12, 2.23]) Let D1 and D2 be finite dimensional simple algebras over F with involutions α1 and α2, respectively. Then α = α1 ⊗ α2 is an involution of D1 ⊗F D2. (i) If α1 and α2 are orthogonal, then α is orthogonal. (ii) If α1 is orthogonal and α2 is symplectic, then α is symplectic. (iii) If α1 and α2 are symplectic, then α is orthogonal in the case of char F 6= 2 and symplectic otherwise. Recall that e1 is the identity element of A1. We will use the notation ¯A1 = e1A2e1 and ¯V1 = e1V2. Let ¯b1 be the restriction of the form b2 to ¯V1. Lemma 2.14 Assume that A2 is not of type A. Then (i) ¯A1 is a ∗-invariant simple subalgebra of A2; (ii) ¯V1 is an irreducible ¯A1-module and ¯V1 = ¯A1V2; (iii) the form ¯b1 on ¯V1 is nondegenerate and corresponds to the involution ∗ on ¯A1; moreover, ¯b1 has the same type as b2 except in the case when char F = 2 and A2 is of type O. Proof. Note that e1 is an idempotent of A2 and e∗ 1 = e1, so (i) and (ii) are clear. Now assume that ¯b1 is degenerate, i.e. there exists v ∈ V2 such that e1v 6= 0 and b2(e1v, e1w) = 0 for all w ∈ V2. Then b2(e1v, w) = b2(e1e1v, w) = b2(e1v, e1w) = 0 for all w ∈ V2, which contradicts to nondegeneracy of b2. symmetric), then ¯b1 is alternating (resp. symmetric). It remains to note that if b2 is alternating (resp. Lemma 2.15 Let ε : A1 → A2 be an embedding of involution simple algebras of types different from A and let αi denotes the involution of Ai. Fix any canonical matrix realization (Mn1(F), α1) of A1. Then there exists a matrix realization (Mn2(F), α2) of A2 such that the following hold. (i) The embedding ε is the composition of the following embeddings of algebras with involution. (Mn1(F), α1) η −→ (Mn1(F) ⊗F Mk(F), α1 ⊗ β1) ι−→ (Mkn1(F), β2) ζ −→ (Mn2 (F), α2) where η(X) = X ⊗ e with e the identity element of Mk(F), ι is the natural isomorphism, and ζ is a natural embedding (i.e. of signature (1, 0, z)). (ii) If char F 6= 2 and A1 and A2 are both of type O, then α1 = β1 = β2 = α2 = t (matrix transpose) (iii) If char F 6= 2 and A1 and A2 are both of type S, then α1 = β2 = α2 = τ (symplectic transpose) and β1 = t. 7 Proof. Let ¯A1 be as in Lemma 2.14 and let β2 be the restriction of α2 to ¯A1. Let B be the centralizer of A1 in ¯A1. Note that A1 and ¯A1 are simple and have the same identity element. Therefore B is simple and ¯A1 = A1B ∼= A1 ⊗F B (see e.g. [12, 1.5]). Clearly, B is β2-invariant. Denote by β1 the restriction of β2 to B. Then ( ¯A1, β2) ∼= (A1 ⊗F B, α1 ⊗ α2). We get the following chain of embeddings of algebras with involution: A1 −→ A1 ⊗F B ≃ ¯A1 −→ A2. Identifying A1 with Mn1(F), B with Mk(F) for some k, ¯A1 with Mkn1(F), and A2 with Mn2(F), we prove (i). Assume now that char F 6= 2 and A1 and A2 are of the same type S (resp. O), i.e. α1 and α2 are of type S (resp. O). Then by Lemma 2.14, β2 is of type S (resp. O). Therefore by Lemma 2.13, β1 is of type O. Fixing an appropriate isomorphism B ∼= Mk(F), by Lemma 2.3, we can assume that β1 is a matrix transpose. Using the same lemma we get that β2 can t). Now by Lemma 2.14, the restriction of the form b1 to ¯V1 is be represented as τ (resp. nondegenerate. Thus V2 = ¯V1 ⊕ ¯V ⊥ 1 , we can easily represent α2 as τ (resp. t). 1 . By choosing a suitable basis in ¯V ⊥ As a corollary we get the following analogue of Proposition 2.12 for symplectic and orthogonal algebras. Proposition 2.16 Let ε : A1 → A2 be an embedding of finite dimensional involution simple algebras over F of the same type S or O. Assume that char F 6= 2. Then ε is representable. That is, for every canonical matrix realization of A1 there exists a canonical matrix realization of A2 such that the embedding ε is of the form (7). Proposition 2.19 below shows that the case of characteristic 2 is exceptional indeed. We will also need the following result, which describes embeddings of involution simple algebras of different types. Recall that (l, r, z) is the signature of the embedding ε : A1 → A2. Proposition 2.17 Let ε : A1 → A2 be an embedding of finite dimensional involution simple algebras over F. Assume that char F 6= 2. (i) If A1 is of type A and A2 is not of type A, then l = r. (ii) If A1 is of type S (resp., O) and A2 is of type O (resp., S), then l is even. (iii) If A1 and A2 are both not of type A and l is even, then there exist an algebra D of type A, an embedding η : A1 → D with the signature (l/2, 0, 0) and an embedding ζ : D → A2 with the signature (1, 1, z) such that ε = ζη. (iv) If A1 and A2 are of type A and l = r, then there exist an algebra D of type O (resp., S), embeddings η : A1 → D of signature (l, l, 0) and ζ : D → A2 of signature (1, 0, z) such that ε = ζη. Proof. (i) Recall that Ai = Bi⊕Ci where Bi and Ci are the simple components of Ai and B∗ i = Ci. And fi and gi = f ∗ i are the identities of Bi and Ci, respectively. Obviously, l = (dim f1A2f1)/n1 and r = (dim g1A2g1)/n1 where n1 = dim V1 = dim W1. Since (f1A2f1)∗ = g1A2g1, we get that l = r. Note that this is valid for the case of char F = 2 as well. 8 (ii) Represent the embedding A1 → A2 as in Lemma 2.15(i). Note that k = l. By Lemma 2.14(iii), β2 has the same type as α2. Thus the types of α1 and β2 are different. By Lemma 2.13, β1 must be symplectic. Therefore k = l is even. (iii) Represent the embedding A1 → A2 as in Lemma 2.15(i). Denote by B the algebra Mk(F). By assumption, k = l is even. Let e be the identity element of B. By Lemma 2.7, B has two idempotents f and g such that e = f + g, f g = gf = 0, and f ∗ = g. Then Bf = f Bf and Bg = gBg are simple subalgebras of B, Bf ∩ Bg = 0, Bf Bg = BgBf = 0, and B∗ f = Bg. Thus B′ = Bf ⊕ Bg is an involution simple subalgebra of B of type A. Therefore D = A1 ⊗F B′ is an involution simple subalgebra of A1 ⊗F B of type A. Since e = f + g, D contains A1. Clearly the signature of the embedding A1 → D is (l/2, 0, 0) and the signature of the embedding D → A2 is (1, 1, z). (iv) Let A = Mn(F) ⊕ Mn(F) be an involution simple algebra with standard involution (X, Y )∗ = (Y t, X t). Let k ≤ n and let the algebra D = Mk(F) have an involution α. Define a "corner" embedding ϕ : D → A via ϕ(Z) = ( ¯Z, (Z α)t) where ¯Z = diag(Z, 0 . . . , 0). Since t ◦ α is an automorphism of D, ϕ is an algebra homomorphism. Moreover, one can easily check that ϕ respects involution: ϕ(Z α) = (Z α, Z t) = ( ¯Z, (Z α)t)∗ = ϕ(Z)∗ By Proposition 2.12, the embedding ε can be represented as in (8) with l = r and involution ∗ acting as (X, Y )∗ = (Y t, X t) on both algebras. Now let α be either symplectic involution θ− or orthogonal involution θ+ (see Remark 2.6) of the algebra D = M2l(F). Let ζ = ϕ : D → A2 be the corner embedding of algebras with involution described above. Note that it is an embedding of signature (1, 0, z). Observe that (cid:18) a b c d (cid:19)θ±!t =(cid:18) dt ±bt at (cid:19)t ±ct =(cid:18) d ±c a (cid:19) , ±b for (cid:18) a b c d (cid:19) ∈ M2l(F), where a, b, c, d are square matrices of size l. Therefore, it is easy to see from formula (8) that ε(A1) ⊂ ζ(D). Define by η the following composition of embeddings: A1 −→ ε(A1) −→ ϕ(D) ζ −1 −→ D. Then η is of signature (l, r, 0) and ε = ζη, as required. It remains to consider the case of characteristic 2, which is a bit more complicated. Lemma 2.18 Let char F = 2 and let ε : A1 → A2 be an embedding of involution simple algebras preserving the identity element (i.e. ε(e1) = e2). Assume that A2 is of type O. Then A1 is of type O. Proof. Assume that A1 is of type A or S. Then by Lemma 2.7, A1 has idempotents f and g such that e1 = f + g, f g = gf = 0, and f ∗ = g. Let b be a symmetric nondegenerate form on V2 corresponding to the involution. Then for all v ∈ V2 we have b(v, v) = b((f + g)v, v) = b(f v, v) + b(v, f v) = 0, as b is symmetric. Therefore b is alternating, so A2 is symplectic, which contradicts the assump- tion. 9 Proposition 2.19 Let char F = 2 and let ε : A1 → A2 be an embedding of involution simple algebras of the same type X = O or S. Then the following conditions are equivalent. (i) The embedding ε is representable. (ii) Each ∗-invariant involution simple subalgebra D of A2 containing A1 is of type X. Moreover, if the embedding ε is not representable, then there exists a ∗-invariant involution simple subalgebra D of A2 which is of type A and contains A1. Proof. By Lemma 2.15(i), the embedding ε can be represented as the composition of embeddings A1 → C → A2 where C = e1A2e1 ∼= A1 ⊗ Mk(F) is involution simple of type O or S and has the same identity element e1 as A1, and the embedding C → A2 is natural (of signature (1, 0, z)). (i) ⇒ (ii) (X = O): Assume that ε is representable and there exists a ∗-invariant involution simple subalgebra D of A2 containing A1 which is not of type O. The matrix presentation (7) shows that C is of type O. Let eD be the identity element of D. Since eD is an idempotent, the algebra F = eDA2eD is a ∗-invariant simple subalgebra of A2 containing D. By Lemma 2.18, it cannot be orthogonal. Therefore F is of type S. Note that F contains C and e1F e1 = C. Therefore by Lemma 2.14(iii), C must be of the same type S, which is a contradiction. (i) ⇒ (ii) (X = S): Assume that ε is representable and there exists a ∗-invariant involution simple subalgebra D of A2 containing A1 which is not of type S. First assume that D is of type A. Then e1De1 is an involution simple subalgebra of C = e1A2e1 of type A containing A1. Recall that C ∼= A1 ⊗ Mk(F). Therefore e1De1 ∼= A1 ⊗ E where E is an involution simple subalgebra of Mk(F) of type A with the same identity element. Since ε is representable, the involution on Mk(F) is orthogonal, which contradicts to Lemma 2.18. Suppose now that D is of type O. As in the case X = O, the algebra F = eDA2eD is a ∗- invariant simple subalgebra of A2 containing C. By Lemma 2.14(iii), F is symplectic. Therefore F ∼= D⊗F Mq(F) with a symplectic involution on Mq(F) (Lemma 2.13(i)). By Lemma 2.7, Mq(F) has two idempotents f and g such that f + g is the identity element of Mq(F), f g = gf = 0, and f ∗ = g. Therefore D′ = D ⊗ f ⊕ D ⊗ g is an involution simple subalgebra of A2 of type A containing A1. However the case of type A subalgebra containing A1 has been already considered in the previous paragraph. (ii) ⇒ (i) and "Moreover" part: Assume that the embedding ε is not representable. We are going to show that A2 contains an involution simple subalgebra of type A containing A1. Recall that C ∼= A1 ⊗ Mk(F) and the restriction of the involution ∗ on C has the form α1 ⊗ α2 where α1 is the involution of A1 and α2 is an involution of Mk(F). Clearly if α2 is orthogonal, then ε is representable (see the proof of Lemma 2.15). Therefore α2 is symplectic. Then, as above, Mk(F) has two idempotents f and g such that f + g is the identity element of Mk(F), f g = gf = 0, and f ∗ = g. Therefore D = A1 ⊗ f ⊕ A1 ⊗ g is an involution simple subalgebra of A2 of type A containing A1. The proposition follows. The following results show how embedding signatures behave under compositions. Proposition 2.20 Let ε1 : A1 → A2 and ε2 : A2 → A3 be embeddings of involution simple algebras of the same type with the signatures (l1, r1, z1) and (l2, r2, z2), respectively. Denote by (l, r, z) the signature of ε = ε2ε1. Then l = l1l2 + r1r2, r = r1l2 + l1r2, z = z1(l2 + r2) + z2. 10 (9) (10) Proof. For types S and O one has r = r1 = r2 = 0, so the statement immediately follows from (6). For type A, the embeddings are representable so one can use (8). Note that l + r = (l1 + r1)(l2 + r2) and l − r = (l1 − r1)(l2 − r2). Thus, the following is true. Corollary 2.21 Let A1 → · · · → Ak be a sequence of embeddings of involution simple algebras of the same type. Let (li, ri, zi) be the signature of Ai → Ai+1, (l, r, z) the signature of A1 → Ak, si = li + ri, ci = li − ri, s = l + r, c = l − r. Then s = s1 . . . sk−1 and c = c1 . . . ck−1. Recall that ni is the degree of Ai (so Ai ∼= Mni(F) or Mni(F) ⊕ Mni(F)). Lemma 2.22 Let ε1 : A1 → A2 and ε : A1 → A3 be representable embeddings of involution sim- ple algebras of the same type (A, S or O) with the signatures (l1, r1, z1) and (l, r, z), respectively. Assume that a triple of non-negative integers (l2, r2, z2) satisfies the following conditions l + r = (l1 + r1)(l2 + r2), l − r = (l1 − r1)(l2 − r2), n3 = n2(l2 + r2) + z2 (11) (12) (13) where ni is the degree of Ai. Then there exists a representable embedding ε2 : A2 → A3 with the signature (l2, r2, z2) such that ε = ε2ε1. Proof. Fix any canonical matrix realizations of A1, A2, A3 such that the matrix embeddings ε1 and ε become canonical (see Definition 2.10). Consider the canonical matrix embedding ε2 : A2 → A3 with signature (l2, r2, z2). The embedding ε2 is well-defined because of (13) and respects the involution (see Remark 2.11(1)). By Remark 2.11(2), the matrix homomorphism ε2ε1 is canonical. By rewriting the conditions (11) and (12) in the form (9) and (10) we see that both canonical homomorphisms ε and ε2ε1 have the same signature, so ε = ε2ε1. 1 pα2 Our classification will be given in terms of so-called supernatural (or Steinitz) numbers. They are defined as follows. Let (p1, p2, . . . ) be the increasing sequence of all prime numbers. The set of all mappings from {p1, p2, . . . } into the set {0, 1, 2, . . . } ∪ {∞} is called the set of supernatural (or Steinitz) numbers. If a supernatural number takes a value α1 at p1, α2 at p2,. . . , this element will be denoted by pα1 2 . . . . The supernatural numbers can be regarded as formal products of powers of primes where infinity is permitted as a power. The set of natural numbers N can be identified in an evident way with a subset of supernatural numbers. If Π = pα1 2 . . . and Π′ = p . . . . We say that Π divides Π′ if and only if α1 ≤ α′ Π′ ) if there exists n ∈ N such that nq ∈ N and nΠ = nqΠ′. If there exists non-zero q ∈ Q such that Q Π = qΠ′, then we say that Π and Π′ are Q-equivalent and denote this relation by Π ∼ Π′. Let S = (s1, s2, . . . ) be a sequence of natural numbers. Denote by Π(S) the supernatural number s1s2s3 . . . . We will use the following simple observation. α′ 2 . . . are two supernatural numbers, we set ΠΠ′ = p 2 2 . . . . Let q ∈ Q. We write Π = qΠ′ (or q ∈ Π 1, α2 ≤ α′ α1+α′ 1 1 α′ 1 p 1 1 pα2 α2+α′ 2 2 p Proposition 2.23 ([2, Proposition 3.2]) Let S = (si)i∈I and S ′ = (s′ natural numbers. Then q ∈ Π(S) and l = l(k) ∈ I such that s1 . . . si divides qs′ Z). j)j∈J be sequences of Π(S ′) if and only if for each i ∈ I and k ∈ J there exist j = j(i) ∈ J k divides s1 . . . sl (over j (over Z) and qs′ 1 . . . s′ 1 . . . s′ 11 3 Bratteli diagrams and dimension groups Let A1 → A2 → A3 → · · · → Ai → Ai+1 → . . . (14) be a sequence of embeddings of finite dimensional involution simple algebras over F. Assume that all Ai are of the same type and char F 6= 2. Then, as we proved in Propositions 2.12 and 2.16, all embeddings Ai → Ai+1 are representable, so one can assume that all Ai are matrix (or double matrix) algebras and the embeddings and involutions are canonical. This justifies the following definition. Definition 3.1 Let A be a locally involution simple associative algebra of countable dimension. We say that A is canonically representable if it is isomorphic to the direct limit of the sequence (14) where all Ai are matrix (resp. double matrix) algebras with canonical involutions of the same type X (= A, S or O) and all embeddings are canonical. In that case we say that the sequence (14) is a canonical representation for A and A is of type X. Note that the type X of the algebra A may not be unique. Proposition 2.19 shows that some of the embeddings Ai → Ai+1 may not be representable in characteristic 2. Fortunately, there is a way to modify the sequence (14), without changing the limit algebra, in order to get representable embeddings even in characteristic 2. Theorem 3.2 Let A be a locally involution simple associative algebra over F of countable di- mension. Then A is canonically representable. Proof. Let A1 → A2 → A3 → . . . be a sequence of embeddings of involution simple finite dimensional associative algebras such that A = lim−→ Ai. Choose an infinite subsequence of al- gebras of the same type. If char F 6= 2, or char F = 2 and all algebras are of type A, then all embeddings are representable by Propositions 2.12 and 2.16. Assume char F = 2. If there is an infinite number of non-representable embeddings, then by Proposition 2.19, we can replace the subsequence by a sequence of embeddings of algebras of type A. Otherwise, we get the result by removing a finite number of algebras in the beginning of the sequence. Theorem 3.2 reduces classification of locally involution simple algebras to the following two problems: (a) classification of the direct limits of canonical sequences of the same type; (b) classification of intertype isomorphisms. We are going to simplify Problem (a) even further and reduce it to the algebras without involution. We need the following trivial observation. Proposition 3.3 Let A1 → A2 → A3 → . . . and A′ 3 → . . . be two sequences of embeddings of algebras (or algebras with involution). Then lim−→ Ai ∼= lim−→ A′ j if and only if there exist sequences of indices i1 < i2 < . . . and j1 < j2 < . . . and homomorphisms ϕk : Aik → A′ jk and ϕ′ jk → Aik+1 (k = 1, 2, . . . ) such that the following diagram commutes. 2 → A′ 1 → A′ k : A′ Ai1 −→ Ai2 −→ . . . Aik −→ Aik+1 −→ . . . ↓ϕ1 րϕ′ A′ j1 −→ A′ ↓ϕk րϕ′ jk −→ A′ j2 −→ . . . A′ 1 ↓ϕ2 րϕ′ −→ . . . jk+1 2 ↓ϕk+1 րϕ′ k+1 k (15) 12 Proof. Set A = lim−→ Ai and A′ = lim−→ A′ Fix any index i1. Then there exists j1 such that ϕ(Ai1 ) ⊆ A′ ϕ−1(A′ of ϕ−1 to A′ obvious. j. Assume that there exists an isomorphism ϕ : A → A′. j1. Similarly, there exists i2 such that k the restriction jk , k = 1, 2, . . . . Then the diagram above commutes. The converse statement is j1) ⊆ Ai2, and so on. Denote by ϕk the restriction of ϕ to Aik , and by ϕ′ Theorem 3.4 Two locally involution simple associative algebras of the same type over F of countable dimension are isomorphic if and only if they are isomorphic as associative algebras. jk and ϕ′ k : A′ jk+1 Proof. Let A and A′ be two locally involution simple associative algebras and let A = lim−→ Ai j be their canonical representations. Assume that A and A′ are isomorphic and A′ = lim−→ A′ as associative algebras. Using Proposition 3.3, we get a commutative diagram (15), where ϕk : Aik → A′ jk → Aik+1 are algebra homomorphisms, not necessarily respecting the involution. Let εk : Aik → Aik+1 and ε′ be the horizontal maps. Note that they are canonical and respect the involution. Denote by ψk (resp. ψ′ k) the canonical map jk → Aik+1) of the same signature as ϕk (resp. ϕ′ Aik → A′ k). Then by Remark 2.11(1) these maps respect the involution. It remains to show that they make the diagram (15) commutative. Note that the signature of ψ′ kϕk = εk. Since both k = ε′ kψk and εk are canonical, we get that ψ′ ψ′ k. Therefore the diagram (15) commutes with respect to the maps ψk and ψ′ k. Proposition 3.3 implies that A and A′ are isomorphic as algebras with involution. kψk equals to the signature of ϕ′ kψk = εk. Similarly, one proves that ψk+1ψ′ jk (resp. A′ k : A′ jk → A′ The converse statement is trivial. Theorem 3.4 reduces Problem (a) to classifying the direct limits of finite dimensional semisim- ple algebras. This is usually done in terms of Bratteli diagrams, the K0-functor and dimension groups. To make the statements of the results a little bit clearer it is best to work in the category of unital algebras (i.e. algebras with identity elements and with identity preserving homomorphisms). In our case this can be easily achieved by adjoining an external identity element. Definition 3.5 Let A be an associative algebra. Define the algebra A as follows. If A has an identity element, put A = A. Otherwise, put A = A + F1 A where 1 A is the identity element of A. Note that if A has an involution then this involution trivially extends to A. Lemma 3.6 Let A be a locally semisimple associative algebra. Then A is locally semisimple in the category of unital algebras. Proof. Let A = lim−→ Ai with Ai finite dimensional semisimple. If A contains an identity element 1A, then A is the direct limit of those Ai which contain 1A, as required. If A has no identity element, then A = A + F1 A = lim−→ Bi where Bi = Ai + F1 A are obviously finite dimensional and semisimple. Proposition 3.7 Let A and A′ be involution simple associative algebras. Then A ∼= A′ as associative algebras if and only if A ∼= A′ as associative algebras. 13 Proof. By construction, A ∼= A′ implies A ∼= A′. Assume now that A ∼= A′. We need to show that A ∼= A′. Denote by Soc(A) the sum of all minimal ideals of A. Then by Proposition 2.1, Soc(A) = A. Obviously, Soc(A) ⊆ Soc( A). We claim that Soc(A) = Soc( A). Indeed, this is obvious if A = A. Assume A 6= A, i.e. A has no identity element. Let M be a minimal ideal of A such that M 6⊆ Soc(A). Then M ∩ Soc(A) = 0. But Soc(A) = A is an ideal of codimension 1 in A. Therefore M is one-dimensional and A = A ⊕ M . Write 1 A = a + m where a ∈ A and m ∈ M . Then obviously a is an identity element of A, which is a contradiction. Therefore, A = Soc( A) ∼= Soc( A′) = A′, as required. Locally semisimple algebras are best described in terms of their Bratteli diagrams. These are defined as follows. Let B be the direct limit of the infinite sequence B1 → B2 → B3 → . . . (16) i ⊕ S2 i . Let V j where the Bi are finite dimensional semisimple algebras over F. Let S1 i ⊕ · · · ⊕ Ski components of Bi, i.e. Bi = S1 be considered as an Bi-module. Denote by mjq i to Sj the sequence (16) consists of the vertices V = {V j i vertices V j labelled by the number mjq i+1 are connected by an edge if and only if mjq i be the degree of Sj i+1 i+1). The Bratteli diagram of i = 1, 2, 3, . . . ; 1 ≤ j ≤ ki} and edges. Two i > 0. In that case the edge is i . Then obviously i , . . . Ski i -module. Then V j in the restriction of V q i be the simple i can i , S2 i be the natural Sj is the number of copies of Sj i that are mapped to Sq the multiplicity of V j i i (i.e. mjq i i and V q i . Let nj i = dim V j mjq i nj i ≤ nq i+1 ki Xj=1 (17) Moreover, if all homomorphisms in (16) are unital then we have equality in (17) for all i and q, so the whole sequence (16) can be reconstructed from its Bratteli diagram provided the degrees of the simple components of the first term B1 are known (in the case of non-unital embeddings extra data is needed). Now let A be a locally involution simple associative algebra of type X (= A, S or O) over F of countable dimension. By Theorem 3.2, A is the direct limit of the sequence (14) where all Ai are matrix (resp. double matrix) algebras with canonical involutions of the same type X and all embeddings are canonical. We will denote by (li, ri, zi) the signature of the embedding Ai → Ai+1 and by ni the degree of Ai (i.e. Ai = Mni(F) and ri = 0 for X = S, O and Ai = Mni(F) ⊕ Mni(F) for X = A). By Remark 2.9, for type A algebras we can and will assume that li ≥ ri for all i. It is convenient to add to the sequence an algebra of degree 1 (the 1-dimensional algebra F is considered to be of both types O and S), so we will assume that n1 = 1, l1 = n2 and r1 = z1 = 0. Denote by T the triple sequence (li, ri, zi)i∈N. Since ni+1 = (li + ri)ni + zi for all i, the canonical sequence (14) is uniquely determined by the triple sequence T and type X. We will denote by A(T , X) the corresponding locally involution simple associative algebra over F, by A(T ) the corresponding locally semisimple algebra (i.e. the direct limit of the associative algebras (14) disregarding the involution) and by A(T ) the corresponding algebra with an identity element (see Definition 3.5). Recall that by Theorem 3.4 and Proposition 3.7, A(T , X) ∼= A(T ′, X) if and only if A(T ) ∼= A(T ′) (equivalently, A(T ) ∼= A(T ′)). If A has an identity element 1A (i.e. A = A) then we can and will assume that 1A ∈ Ai for all i. Put Bi = Ai if 1A ∈ A and Bi = Ai + F1 A otherwise, see the proof of Lemma 3.6. Then Bi is 14 semisimple, with possibly one extra 1-dimensional simple component. Moreover, all embeddings Bi → Bi+1 are unital and A = lim−→ Bi. Recall that X is the type of A. We will denote by B(T ) the Bratteli diagram B( A) of the algebra A with respect to the sequence B1 → B2 → B3 → . . . . If 1A ∈ A and the type X = S, O, then all zi = 0 and it is easy to see that B(T ) is • l1 • l2 • l3 . . . (18) The locally semisimple algebras of this type are just the limits of "pure diagonal" matrix em- beddings Mni → Mni+1 given by M 7→ diag(M, . . . , M ) (li blocks), M ∈ Mni(F). They were first classified by Glimm [10] (in C∗-algebras setting). It is easy to see that two algebras of this type are isomorphic if and only if their corresponding supernatural numbers Π = l1l2l3 . . . are equal. If 1A 6∈ A and the type X = S, O, then B(T ) is • • l1 ♣♣♣♣♣♣♣♣ ♣♣♣ z1 1 • • l2 ♣♣♣♣♣♣♣♣ ♣♣♣ z2 1 • • l3 ♦♦♦♦♦♦♦♦ z3♦♦♦ 1 . . . . . . (19) The corresponding locally semisimple algebras A(T ) are the direct limits of matrix embeddings of the shape (2). They were first classified by Dixmier [5] (in C∗-algebras setting). Dixmier's parametrization consists of the supernatural number Π = l1l2l3 . . . and one real parameter θ, which is in fact the inverse of our density index δ, see below. The diagrams of this shape also parametrize so-called "diagonal" direct limits of finite symmetric and alternating groups [13]. If 1A ∈ A and the type X = A, then B(T ) is • • ◆◆◆ l1 r1 r1♣♣♣ l1 ♣♣♣♣♣♣♣♣ ◆◆◆◆◆◆◆◆ • • ◆◆◆ l2 r2 r2♣♣♣ l2 ♣♣♣♣♣♣♣♣ ◆◆◆◆◆◆◆◆ • • ❖❖❖ l3 r3 r3♦♦♦ l3 ♦♦♦♦♦♦♦♦ ❖❖❖❖❖❖❖❖❖ . . . . . . (20) The corresponding algebras were first classified by Fack and Mar´echal [9] (in C∗-algebras setting). If 1A 6∈ A and the type X = A, then B(T ) is • • • ✁ ✁ ✁ ✁ ✁ ✁ ♣♣♣♣♣♣♣♣ ◆◆◆◆◆◆◆◆ ♣♣♣♣♣♣♣♣ ✁ ✁ ✁ ✁ l1 ✁ ◆◆◆ l1 r1 r1♣♣♣ ✁ z1 ✁ ✁ ✁ z1♣♣♣ 1 • • • ✁ ✁ ✁ ✁ ✁ ✁ ♣♣♣♣♣♣♣♣ ◆◆◆◆◆◆◆◆ ♣♣♣♣♣♣♣♣ ✁ ✁ ✁ ✁ l2 ✁ ◆◆◆ l2 r2 r2♣♣♣ ✁ z2 ✁ ✁ z2♣♣♣ 1 ✁ • • • ❖❖❖ l3 r3 r3♦♦♦ l3 ⑧ ⑧ z3 ⑧ z3♦♦♦ 1 ⑧ ⑧ . . . ⑧ . . . ⑧ ⑧ ⑧ ⑧ ⑧ ⑧ ♦♦♦♦♦♦♦♦ ❖❖❖❖❖❖❖❖❖ ♦♦♦♦♦♦♦♦ ⑧ ⑧ ⑧ ⑧ . . . (21) This is the most general case. We parametrize the corresponding algebras by two supernatural numbers and two real parameters (see Theorem 4.1). Let B = lim−→ Bi be a unital locally semisimple algebra and let K0(B) be its Grothendieck group with positive cone K0(B)+. Note that the homomorphism Bi → Bi+1 induces the homo- morphism of the abelian groups K0(Bi) → K0(Bi+1) and K0(B) can be obtained as the direct limit lim−→ K0(Bi). Since Bi are finite dimensional and semisimple, one has (K0(Bi), K0(Bi)+) = (Zki, Zki + ) where ki is the number of the simple components of Bi. Therefore the abelian group K0(B) is the direct limit of the sequence Zk1 → Zk2 → · · · → Zki → Zki+1 → . . . 15 the ith level is given by the matrix   li ri 0 ri li 0 zi zi 1   . Moreover, the embedding on the ith level is given by the adjacency (or multiplicities) matrix of the ith level of the Bratteli diagram of B. For example, for the algebra A in (21) the group K0( A) is the direct limit of the sequence Z3 → Z3 → · · · → Z3 → . . . were the embedding on Let 1B be the identity element of B and let [1B] be the corresponding element of K0(B)+. The triple (K0(B), K0(B)+, [1B]) is called the dimension group of B and is a complete invariant for unital locally semisimple algebras. More exactly the following is true. Theorem 3.8 [6] Let B1 and B2 be unital locally semisimple algebras. Then B1 ∼= B2 if and only if there is an order-isomorphism ϕ : K0(B1) → K0(B2) such that ϕ([1B1 ]) = [1B2]. A similar result holds for non-unital algebras if one replaces [1B] by the scale of K0(B). For a triple sequence T we denote by G(T ) the dimension group of the unital locally semisim- ple algebra A(T ). 4 The classification of algebras of the same type and the corre- sponding dimension groups In this section, T = (li, ri, zi)i∈N is the triple sequence of the canonically represented locally involution simple algebra A = A(T , X) of type X. Recall that li ≥ ri and li + ri ≥ 1 for all i. The degrees ni of the subalgebras Ai satisfy the following: n1 = 1 and ni+1 = (li + ri)ni + zi for all i ≥ 1. Set si = li + ri, ci = li − ri (i = 1, 2, . . . ), S = (si)i∈N, C = (ci)i∈N, sk i = si . . . sk−1 and i = ci . . . ck−1. Put δi = si ck 1/ni. Then δi+1 = si+1 1 ni+1 = si 1si nisi + zi = si 1 ni + (zi/si) ≤ δi. (22) The limit δ = lim i→∞ δi is called the density index of T and is denoted by δ(T ). Since δ2 = s1/n2 = 1, we have 0 ≤ δ ≤ 1. If δ = 0, then the triple sequence is called sparse. If there exists i such that for all j > i we have δj = δi 6= 0, then the triple sequence is called pure. In view of (22) this is equivalent to the following. There exists i such that for all j ≥ i we have zj = 0. In this case, by removing a finite number of terms from the canonically represented sequence without changing the limit algebra, we may and will assume that zi = 0 for all i. We say that the triple sequence is dense if and only if 0 < δ < δi for all i. If there exists i such that cj = sj (equivalently, rj = 0) for all j ≥ i, then T is called one-sided. Otherwise, it is called two-sided. If for each i there exists j > i such that cj = 0 (equivalently, lj = rj), then T is called (two-sided) symmetric. Otherwise it is called non-symmetric. In the latter case we may and will assume that ci > 0 for all i ∈ N. Set σi = c1...ci s1...si . The limit σ = lim i→∞ σi 16 is called the symmetry index of T and is denoted by σ(T ). Observe that 0 ≤ σ ≤ 1. Two-sided non-symmetric triple sequences with σ = 0 are called weakly non-symmetric, and those with σ 6= 0 are called strongly non-symmetric. Thus all triple sequences can be partitioned into three classes with respect to density and into four classes with respect to symmetry. Density types (D1) Sparse (δ = 0). (D2) Dense (δi > δ > 0 for all i). (D3) Pure (δi = δ > 0 for some i). Symmetry types (S1) One-sided (rj = 0 for all j ≫ 1). (S2) Two-sided symmetric (lj = rj for an infinite set of j). (S3) Two-sided weakly non-symmetric (rj > 0 for an infinite set of j, lk > rk for all k ≫ 1, and σ = 0). (S4) Two-sided strongly non-symmetric (rj > 0 for an infinite set of j, lk > rk for all k ≫ 1, and σ 6= 0). Now we are ready to prove our main classification result for algebras of the same type. Theorem 4.1 Let T = {(li, ri, zi) i ∈ N} and T ′ = {(l′ i) i ∈ N} be triple sequences and let X = A, S or O. Set δ = δ(T ), σ = σ(T ), δ′ = δ(T ′) and σ′ = σ(T ′). Then the locally involution simple algebras A(T , X) and A(T ′, X) (respectively, the locally semisimple algebras A(T ) and A(T ′); respectively, the dimension groups G(T ) and G(T ′)) are isomorphic if and only if the following conditions hold. i, z′ i, r′ (A1) The triple sequences T and T ′ have the same density type. (A2) Π(S) Q ∼ Π(S ′). (A3) δ δ′ ∈ Π(S) Π(S ′) for dense and pure triple sequences (types (D2) and (D3)). (B1) The triple sequences T and T ′ have the same symmetry type. (B2) Π(C) Q ∼ Π(C ′) for two-sided non-symmetric triple sequences (types (S3) and (S4)). (B3) There exists α ∈ Π(S) sequences (type (S4)). Moreover, α = δ pure (types (D2) and (D3)). Π(S ′) such that α σ σ′ ∈ Π(C) Π(C′) for two-sided strongly non-symmetric triple δ′ if in addition the triple sequences are dense or 17 Proof. The proof is similar to that in the case of Lie algebras in characteristic zero (see [2]). First we will prove necessity. By Theorem 3.4 and Propositions 3.7 and 3.8, it is enough to prove the result for the locally semisimple algebras A(T ) and A(T ′). We will prove the following more general statement (which will later be used for intertype isomorphisms, Theorem 5.2). If A(T , X) ∼= A(T ′, X ′) (we do not demand that X = X ′), then T and T ′ satisfy the conditions (A1), (A2), (A3). Moreover, if X = X ′ = A, then the conditions (B1), (B2), (B3) hold. Let j)j∈J (I ∼= J ∼= N) be canonically represented sequences of involution simple (Ai)i∈I and (A′ algebras of types X and X ′, corresponding to the triple sequences T and T ′, respectively. We have A ∼= A′ where A = lim−→ Ai, A′ = lim−→ A′ i. By Proposition 3.3, there exist subsequences i1 < i2 < . . . of I, j1 < j2 < . . . of J, and embeddings εk : Aik → A′ jk → Aik+1 (k = 1, 2, . . . ) such that the following diagram is commutative. k : A′ jk , ε′ Ai1 −→ . . . −→ Aik −→ Aik+1 −→ . . . −→ Aim −→ . . . ↓ε1 րε′ A′ j1 −→ . . . −→ A′ −→ . . . −→ A′ jk −→ A′ jm −→ . . . ր ↓εk րε′ ր ↓εm ր ↓εk+1 ր 1 k jk+1 (23) Let (pk, qk, uk) (resp., (p′ Set si = li + ri, ci = li − ri, δi = si are defined similarly. We have k, u′ k, q′ k)) be the signature of εk (resp., ε′ 1/ni, δ = limi→∞ δi. The numbers n′ k). Let ni be the degree of Ai. j,... for the algebra A′ j, s′ jm = (pm + qm)nim + um = (pm + qm)sim n′ 1 δ−1 im + um = (pm + qm)sik 1 sim ik δ−1 im + um. On the other hand, jm = s′jm n′ 1 (δ′ jm)−1 = s′jk 1 s′jm jk (δ′ jm)−1. In view of commutativity of the diagram and by Corollary 2.21 we have ik (pm + qm) = (pk + qk)s′jm sim jk . Dividing (24) and (25) by s′jm jk , we get (pk + qk)sik 1 δ−1 im + um/s′jm jk = s′jk 1 (δ′ jm)−1, so (pk + qk)sik 1 δ′ jm ≤ s′jk 1 δim. Taking m → ∞, we obtain (pk + qk)sik Corollary 2.21, we have (pk + qk)(p′ k + q′ k) = sik+1 ik . Hence 1 δ′ ≤ s′jk 1 δ. Similarly, we get (p′ k + q′ k)s′jk 1 δ ≤ sik+1 1 (pk + qk)sik 1 δ′ ≤ s′jk 1 δ ≤ (p′ k + q′ k)−1sik+1 1 δ′ = (pk + qk)sik 1 δ′. Therefore (pk + qk)sik k)s′jk (p′ k + q′ 1 δ′ = s′jk 1 δ, 1 δ = sik+1 1 δ′. (24) (25) (26) (27) δ′. By (28) (29) Clearly δ = 0 if and only if δ′ = 0. Therefore T is sparse if and only if T ′ is so. If the triple sequence T is pure, then δ = δim for some m. Subtracting (28) from (27), we get 0 ≤ (pk + qk)sik 1 (δ′ jm − δ′) ≤ s′jk 1 (δim − δ) = 0. 18 Therefore δ′ holds. jm = δ′, so T ′ is also pure. By symmetry, T is pure if and only if T ′ is pure. So (A1) By (26), sim ik divides (pk + qk)s′jm jk for all m > k. On the other hand, in view of commutativity of the diagram we have so (pk + qk)s′jm jk divides sim+1 ik sim+1 ik = (pk + qk)s′jm jk (p′ m + q′ m), . Therefore by Proposition 2.23, Π(Sik ) = (pk + qk)Π(S ′ jk ), (30) (31) where Sik = (sik , sik+1, . . . ), S ′ jk = (s′ jk , s′ Finally, if δ and δ′ are nonzero (dense or pure sequences), then by (28) and (29), sik jk+1, . . . ). It follows that Π(S) Q ∼ Π(S ′), so (A2) holds. 1 di- for any k. Therefore by Proposition 2.23, Π(S) = vides (δ/δ′)s′jk 1 and (δ/δ′)s′jk (δ/δ′)Π(S ′), and (A3) holds. 1 divides sik+1 1 Assume now that X = X ′ = A. By Corollary 2.21, one can write down equalities for "differences" similar to (26) and (30): ik (pm − qm) = (pk − qk)c′jm cim jk ; cim+1 ik = (pk − qk)c′jm jk (p′ m − q′ m). (32) (33) If T ′ is symmetric, then by definition, for each k there exists m such that c′jm from (33) that cim+1 Assume that T is non-symmetric. Recall that in this case one can suppose that all ci and c′ nonzero. Dividing (33) by (30), we get jk = 0. It follows ik = 0, so T is symmetric. Therefore, T is symmetric if and only if T ′ is so. j are cim+1 ik sim+1 ik = (pk − qk) (pk + qk) c′jm jk s′jm jk (p′ (p′ m − q′ m + q′ m) m) , or equivalently, Taking m → ∞, we get σim+1 1 · sik 1 cik 1 = σ′jm 1 · (pk − qk) (pk + qk) s′jk 1 c′jk 1 (p′ (p′ m − q′ m + q′ m) m) σ · sik 1 cik 1 ≤ σ′ · (pk − qk) (pk + qk) s′jk 1 c′jk 1 . Similarly, dividing (32) by (26) and taking m → ∞, we get Combining with (36), we obtain σ · sik 1 cik 1 ≥ σ′ · (pk − qk) (pk + qk) s′jk 1 c′jk 1 . σ · sik 1 cik 1 = σ′ · (pk − qk) (pk + qk) s′jk 1 c′jk 1 . 19 . (34) (35) (36) (37) (38) It follows that σ = 0 if and only if σ′ = 0. That is, T is weakly non-symmetric if and only if T ′ is so. Assume that T ′ is one-sided. Then σ′ = σ′jm for some m. Subtracting (38) from (35), we 1 ≤ 0. Therefore σim+1 have 0 ≤ (σim+1 = σ, i.e. T is one-sided. So (B1) holds. − σ)sik 1 /cik 1 1 1 Similarly to (31), one can get Π(Cik ) = (pk − qk)Π(C ′ jk ). It follows that Π(C) Q ∼ Π(C ′), so (B2) holds. Assume now that T and T ′ are strongly non-symmetric, i.e. σ 6= 0 and σ′ α = (pk + qk)sik 1 /s′jk 1 . Then (38) can be rewritten in the form σ σ′ αc′jk 1 = (pk − qk)cik 1 . Observe that α ∈ Π(S) Π(S ′) . Indeed, using (31), we have αΠ(S ′) = (pk + qk)sik 1 Π(S ′ jk ) = sik 1 Π(Sik ) = Π(S). (39) 6= 0. Set (40) Moreover, if T and T ′ are dense or pure, then by (28), α = δ/δ′. It follows from (40) and (39) that σ σ′ αΠ(C ′) = (pk − qk)cik 1 Π(C ′ jk ) = cik 1 Π(Cik ) = Π(C). Therefore, σ σ′ α ∈ Π(C) Π(C′) . This proves (B3). (cid:3) To prove the sufficiency in Theorem 4.1, we need the following lemma. Lemma 4.2 Let T and T ′ satisfy the conditions (A1), (A2), (A3), (B1), (B2), (B3) of the theo- rem. Fix α ∈ Π(S) Π(C′) for the case of two-sided non-symmetric triple sequences (β/α = σ/σ′ if T and T ′ are strongly non-symmetric). Let i, j, a, b be integers such that Π(S ′) (α = δ/δ′ if T and T ′ are dense or pure), β ∈ Π(C) (a) αs′j (b) βc′j 1 = asi 1, 1 = bci 1 (for two-sided non-symmetric T and T ′). Then there exists k > i such that a′ = sk even and ck i = 0 for the case of symmetric T and T ′), a′ ≥ b′ and nk ≥ a′n′ j. i /a and b′ = ck i /b are integers of the same parity (a′ is If otherwise is not specified we assume that T and T ′ are two-sided non-symmetric. Proof. The case of one-sided and symmetric sequences can be settled by removing from the proof the arguments with ci, β, b. Since α ∈ Π(S) Π(S ′) and αs′j 1 = asi 1, we have Π(Si) = α(si 1)−1s′j 1 Π(S ′ j) = aΠ(S ′ j). Similarly, we get Π(Ci) = bΠ(C ′ j). (41) (42) Therefore there exists k1 > i such that a′ = sk for each m the integers s′ if and only if 2 divides Π(C ′ m and c′ m + r′ m = l′ i /a and b′ = ck m = l′ m − r′ i /b are integers for all k ≥ k1. Since m have the same parity, 2 divides Π(S ′ j) j) always). Therefore by j) (for symmetric sequences 2 divides Π(S ′ 20 (41) and (42), there exists k2 ≥ k1 such that the integers a′ and b′ have the same parity (a′ is even and ck i = 0 for the case of symmetric T and T ′) for all k ≥ k2. Set γk = b′/a′. In view of (a) and (b), we have 1ck ci i βc′j 1 If T and T ′ are weakly non-symmetric, then σk 1 → 0 as k → ∞, so γk → 0. If T and T ′ are strongly non-symmetric, then by assumption β/α = σ/σ′, so σk 1 σ′j 1 a sk i γk = ck i b α β = = · · αs′j 1 si 1sk i · . γk → α β · σ σ′j 1 = σ′ σ′j 1 < 1 as k → ∞. In both cases there exists k3 ≥ k2 such that γk ≤ 1 (i.e. a′ ≥ b′) for all k ≥ k3. Set νk = nk/a′ − n′ j. We have to show that νk ≥ 0 for sufficiently large k. One has νk = nk a′ − n′ j = sk 1 a′δk − s′j 1 δ′ j = asi 1 δk − s′j 1 δ′ j = s′j 1 ( α δk − 1 δ′ j ). (The last equality follows from (a).) If T and T ′ are sparse, then δk → 0 as k → ∞, so νk → +∞. Therefore there exists k4 ≥ k3 such that νk ≥ 0 for all k ≥ k4. Let T and T ′ be dense. Then α = δ/δ′ and δ′ j > δ′. Therefore νk = s′j 1 ( δ δk · 1 δ′ − 1 δ′ j ) → s′j 1 ( 1 δ′ − 1 δ′ j ) > 0, as k → ∞. Hence there exists k4 ≥ k3 such that νk ≥ 0 for all k ≥ k4. Let T and T ′ be pure. Then there exists k4 ≥ k3 such that δ = δk for all k ≥ k4. Therefore νk = s′j 1 ( 1 δ′ − 1 δ′ j ) ≥ 0, for all k ≥ k4. So each k ≥ k4 satisfies the assumptions of the theorem. Proof of sufficiency in Theorem 4.1. According to Proposition 3.3 we have to construct sequences i1 < i2 < . . . , j1 < j2 < . . . , and embeddings εk : Aik → A′ jk → Aik+1 (k = 1, 2, . . . ) such that the diagram (23) is commutative. Fix α ∈ Π(S) Π(S ′) (α = δ/δ′ if T and T ′ are dense or pure) and β ∈ Π(C) Π(C′) for the case of two-sided non-symmetric triple sequences (β/α = σ/σ′ if T and T ′ are strongly non-symmetric). Fix also j0 ∈ J. Since Π(S ′) = α−1Π(S) and Π(C ′) = β−1Π(C), by Proposition 2.23, there exists i1 ∈ I such that k : A′ jk , ε′ (a0) α−1si1 (b0) β−1ci1 1 = a0s′j0 1 , 1 = b0c′j0 1 (for two-sided non-symmetric T and T ′) /a0 and b1 = c′j1 j0 where a0, b0 ∈ N. Applying Lemma 4.2 (interchanging T and T ′), we find j1 such that a1 = s′j1 /b0 are integers of the same parity (a1 is even if T and T ′ are symmetric), j0 a1 ≥ b1 and n′ j1 − a1ni1 (p1 = q1 = a1/2 for symmetric sequences). Consider the canonical embedding ε1 : Ai1 → A′ j1 with the signature (p1, q1, u1). We have j1 ≥ a1ni1. Set p1 = (a1 + b1)/2, q1 = (a1 − b1)/2, u1 = n′ (a1) αs′j1 1 = a1si1 1 , 21 (b1) βc′j1 1 = b1ci1 1 . Proceed by induction. Assume that sequences i1 < · · · < ik, j1 < · · · < jk and embeddings ε1, ε′ 1, . . . , εk have been constructed, and the following conditions hold. (ak) αs′jk (bk) βc′jk 1 = aksik 1 , 1 = bkcik 1 where ak = pk + qk, bk = pk − qk. Construct an embedding ε′ there exists ik+1 > ik such that a′ k is even if T and T ′ are symmetric), a′ (a′ k = (a′ q′ k = q′ /ak and b′ k ≥ b′ k = a′ k = nik+1 − a′ k = sik+1 k = cik+1 k)/2, u′ k − b′ jk (p′ kn′ ik ik k as follows. By Lemma 4.2, /bk are integers of the same parity k)/2, k + b′ k = (a′ k and nik+1 ≥ a′ k/2 for symmetric sequences). Since jk . Set p′ kn′ (pk + qk)(p′ (pk − qk)(p′ k + q′ k − q′ k) = aka′ k) = bkb′ ik k = sik+1 k = cik+1 , ik , k ≥ 0, by Lemma 2.22 there exists an embedding ε′ and u′ where ιk denotes the embedding Aik → Aik+1. Observe that k : A′ jk → Aik+1 such that ιk = ε′ kεk (a′ (b′ k) α−1sik+1 k) β−1cik+1 1 = a′ 1 = b′ ks′jk 1 , ks′jk 1 . Therefore Lemma 4.2 can be applied once more (interchanging T and T ′). So the result follows by induction. (cid:3) Remark 4.3 It is not difficult to see that for pure triple sequences one can always assume that all zi = 0 (by removing a finite number of terms in the sequences). In this case δ = 1, so the condition (A3) can be rewritten in the form Π(S) = Π(S ′). 5 Isomorphisms of algebras of different types In this section we find conditions under which A(T , X) ∼= A(T ′, X ′) where T and T ′ are triple sequences and X 6= X ′. We also give a general parametrization of countable locally involution simple algebras. Lemma 5.1 Let T be a two-sided symmetric triple sequence, S = S(T ). Then 2∞ divides Π(S). Proof. By definition, li = ri (in particular, si = li + ri is even) for an infinite set of i. Therefore 2∞ divides Π(S). Theorem 5.2 Let T , T ′ be triple sequences. (i) Let char F 6= 2. Then A(T , A) ∼= A(T ′, O) (resp.,A(T , A) ∼= A(T ′, S)) if and only if T is two-sided symmetric, 2∞ divides Π(S ′) and the conditions (A1), (A2), (A3) of Theorem 4.1 hold. (ii) Let char F 6= 2. A(T , O) ∼= A(T ′, S) if and only if 2∞ divides both Π(S) and Π(S ′), and the conditions (A1), (A2), (A3) of Theorem 4.1 hold. (iii) Let char F = 2. If X, X ′ ∈ {A, O, S} are different, then A(T , X) is not isomorphic to A(T ′, X ′). 22 (i). Set A = A(T , A) and A′ = A(T ′, O). Assume that A ∼= A′. Then, as it was Proof. established in the proof of Theorem 4.1, the conditions (A1), (A2) and (A3) hold. Now denote by (xk, yk, zk) the signature of Aik → Aik+1 (see diagram (23)). Since the diagram is commutative, we have xk = pkp′ k) are the signatures of εk and k, respectively. By Proposition 2.17(i), pk = qk, so xk = yk. Therefore cik+1 ε′ ik = xk − yk = 0, so T is two-sided symmetric. By Lemma 5.1, 2∞ divides Π(S). Therefore in view of condition (A2), 2∞ divides Π(S ′). k where (pk, qk, uk) and (p′ k and yk = qkp′ k, 0, u′ Conversely. Let A = A(T , A) and A′ = A(T ′, O) be such that T is two-sided symmetric, 2∞ divides Π(S ′) and the conditions (A1), (A2) and (A3) hold. Then there exists a sequence of is even for all k = 1, 2, . . . . By Proposition 2.17(iii), there indices j1 < j2 < . . . such that s exists an algebra A′′ such that the diagram k of type A and representable embeddings A′ k and A′′ jk → A′′ ′jk+1 jk k → A′ jk+1 A′ jk ց −→ A′′ k ր A′ jk+1 is commutative. Set A′′ = lim−→ A′′ k. Let T ′′ be the corresponding triple sequence. We have A′′ = A(T ′′, A). By construction, A′′ ∼= A′. Moreover, by the above arguments (the proof of necessity) T ′′ is symmetric and the conditions (A1), (A2) and (A3) (for T ′ and T ′′) hold. Since the same is true for the pair T , T ′, we conclude that the pair T , T ′′ also satisfies these Q conditions. Indeed, (A1) trivially holds. Further, since Π(S ′) ∼ Π(S ′), we Π(S ′) , then have Π(S) ∼ Π(S ′′). Finally, if δ′ Q ∼ Π(S ′′) and Π(S) Π(S ′′) and δ δ′′ ∈ Π(S ′) δ′ ∈ Π(S) Q Π(S ′) = δ′ δ′′ Π(S ′′) = δ′ δ Π(S), δ′′ ∈ Π(S) so δ A ∼= A′. The proof for the case A′ = A(T ′, S) is similar. Π(S ′′) . Consequently, by Theorem 4.1, A(T , A) ∼= A(T ′′, A), i.e. A ∼= A′′. Therefore (ii). Let A(T , O) ∼= A(T ′, S). Using Proposition 2.17 (ii), (iii), it is not difficult to con- struct an algebra A(T ′′, A) ∼= A(T , O) ∼= A(T ′, S). The claim now follows from Theorem 5.2 (i). To prove the converse statement we construct A(T ′′, A) isomorphic to A(T , O) and use Theo- rem 5.2 (i). (iii). By definition of A(T , X), all the corresponding embeddings are representable. Thus the claim follows from Proposition 2.19. It remains to discuss the general parametrization. Let A be a locally involution simple associative algebra over F of countable dimension. Then by Theorem 3.2, A is canonically representable, i.e. A is the direct limit of a sequence (Ai)i∈N of subalgebras of the same type X = A, O, or S such that all embeddings are canonical. Fix any such system of subalgebras. This gives the triple sequence T = ((li, ri, zi))i∈N, the sequences of "sums" S = (li + ri)i∈N and (for X = A only) "differences" C = (li − ri)i∈N. Now we can determine the density type D=(D1), (D2) or (D3), the density index δ = δ(T ), supernatural number ΠS = Π(S), and (for X = A only) the symmetry type S=(S1), (S2), (S3), or (S4), the symmetry index σ = σ(T ), and supernatural number ΠC = Π(C). So one can associate with any algebra A a tuple P(A) = (X, D, S, δ, σ, ΠS , ΠC) 23 where X, D, S describe a type of A; δ and σ are real numbers (0 ≤ δ, σ ≤ 1); ΠS and ΠC are supernatural numbers. For X = S, O (and X = A with one-sided or symmetric T ) we use a shorter list of invariants: A 7→ (X, D, δ, ΠS ). By Theorem 4.1, the tuples associated with two nonisomorphic algebras are distinct. The question under what conditions A and A′ with tuples P(A) and P(A′) are isomorphic has been resolved in Theorems 4.1 and 5.2. References [1] Y.A. Bahturin, A.A. Baranov, and A.E. Zalesskii, Simple Lie subalgebras of locally finite associative algebras, J. Algebra 281 (2004), no. 1, 225-246. [2] A.A. Baranov and A.G. Zhilinskii, Diagonal direct limits of simple Lie algebras, Commun. in Algebra 27 (1999), 2749-2766. [3] O. Bratteli, G. Elliott, D. Evans, A. Kishimoto, On the classification of inductive limits of inner actions of a compact group, in Current topics in operator algebras (Nara, 1990), 13 -- 24, World Sci. Publishing, River Edge, NJ, 1991 [4] O. Bratteli, P. Jorgensen, V. Ostrovskyi, Representation theory and numerical AF- invariants. The representations and centralizers of certain states on Od. Mem. Amer. Math. Soc. 168 (2004), no. 797, xviii+178 pp. [5] J. Dixmier, On some C ∗-algebras considered by Glimm, J. Functional Analysis 1(1967), 182-203. [6] G. Elliott, On classification of inductive limits of sequences of semisimple finite dimensional algebras, J. Algebra 38 (1976), 29-44. [7] G. Elliott, H. Su, K-theoretic classification for inductive limit Z2 actions on AF-algebras, Canad. J. Math. 48 (1996), no. 5, 946-958. [8] D. Farkas and R. Snider, Locally finite-dimensional algebras, Proc. Amer. Math. Soc. 81 (1981), 369 -- 372. [9] T. Fack and O. Mar´echal, Sur la classification des sym´etries des C ∗-algebres UHF, Canad. J. Math. 31 (1979), no. 3, 496 -- 523. [10] J. Glimm, On a certain class of operator algebras. Trans. Amer. Math. Soc. 95 (1960), 318-340. [11] D. Handelman, Classification of compact group actions on locally semisimple algebras, in Group actions on rings (Brunswick, Maine, 1984), 137 -- 153, Contemp. Math., 43, Amer. Math. Soc., Providence, RI, 1985. [12] M.-A. Knus, A. Merkurjev, M. Rost, J.-P. Tignol, The book of involutions, Amer. Math. Soc., Providence, RI, 1998. 24 [13] Y. Lavrenyuk, V. Nekrashevych, On classification of inductive limits of direct products of alternating groups. J. Lond. Math. Soc. (2) 75 (2007), no. 1, 146-162. [14] S. Roman, Advanced linear algebra, Springer, New York, 1992. [15] A. E. Zalesskii, Direct limits of finite dimensional algebras and finite groups, Trends in ring theory (Miskolc, 1996), 221239, CMS Conf. Proc., 22, Amer. Math. Soc., Providence, RI, 1998. 25
1502.02354	1	1502	2015-02-09T04:34:05	Homological Dimensions Relative to Preresolving Subcategories	[ "math.RA", "math.CT", "math.KT" ]	We introduce relative preresolving subcategories and precoresolving subcategories of an abelian category and define homological dimensions and codimensions relative to these subcategories respectively. We study the properties of these homological dimensions and codimensions and unify some important properties possessed by some known homological dimensions. Then we apply the obtained properties to special subcategories and in particular to module categories. Finally we propose some open questions and conjectures, which are closely related to the generalized Nakayama conjecture and the strong Nakayama conjecture.	math.RA	math	Homological Dimensions Relative to Preresolving Subcategories ∗† Zhaoyong Huang‡ Department of Mathematics, Nanjing University, Nanjing 210093, Jiangsu Province, P.R. China Abstract We introduce relative preresolving subcategories and precoresolving subcategories of an abelian category and define homological dimensions and codimensions relative to these subcategories respectively. We study the properties of these homological dimen- sions and codimensions and unify some important properties possessed by some known homological dimensions. Then we apply the obtained properties to special subcategories and in particular to module categories. Finally we propose some open questions and conjectures, which are closely related to the generalized Nakayama conjecture and the strong Nakayama conjecture. 1. Introduction In classical homological theory, homological dimensions are important and fundamental invariants and every homological dimension of modules is defined relative to some certain subcategory of modules. For example, projective, flat and injective dimensions of modules are defined relative to the categories of projective, flat and injective modules respectively. When projective, flat and injective modules are generalized to Gorenstein projective, Gorenstein flat and Gorenstein injective modules respectively in relative homological theory, Gorenstein projective, Gorenstein flat and Gorenstein injective dimensions emerge; and in particular, they share many nice properties of projective, flat and injective dimensions respectively (e.g. [AB, C, CFH, CI, DLM, EJ1, EJ2, EJL, GD, GT, HI, H2, HuH, LHX, MD, SSW, Z]). Then a natural question is: if two homological (co)dimensions relative to a category and its subcat- egory are defined, what is the relation between these two homological (co)dimensions? The purpose of this paper is to study this question. We introduce relative preresolving subcate- gories and precoresolving subcategories and define homological dimensions and codimensions ∗2010 Mathematics Subject Classification: 18G25, 18G20, 18G10, 16E10. †Keywords: abelian categories, (pre)resolving subcategories, (pre)coresolving subcategories, homological dimension, homological codimension, (Gorenstein) projective dimension, (Gorenstein) injective dimension. ‡E-mail address: [email protected] 1 relative to these subcategories respectively. Then we study their properties and unify some important properties possessed by some known homological dimensions. This paper is organized as follows. In Section 2, we give some terminology and some preliminary results; in particular, we give the definition of homological (co)dimension relative to a certain full and additive subcategory of an abelian category. In Section 3, we first give the definition of (pre)resolving subcategories of an abelian category. Then we give some criteria for computing and comparing homological dimensions relative to different preresolving subcategories. Let E and T be additive and full subcate- gories of an abelian category A such that T is E -preresolving with an E -proper generator C . Assume that 0 → M → T1 → T0 → A → 0 is an exact sequence in A with both T0 and T1 objects in T . Then there exists an exact sequence 0 → M → T → C → A → 0 in A with T an object in T and C an object in C ; and furthermore, if the former exact sequence is HomA (X, −)-exact for some object X in A , then so is the latter one. As applications of this result, we get that an object in A is an n-C -cosyzygy if and only if it is an n-T -cosyzygy; and also get that the T -dimension of an object A in A is at most n if and only if there exists an exact sequence 0 → Kn → Cn−1 → Cn−2 → · · · → C0 → A → 0 in A with all Ci objects in C and Kn an object in T . In addition, we give some sufficient conditions under which the T -dimension and the C -dimension of an object in A are identical. Section 4 is completely dual to Section 3. In Section 5, we apply the results in Sections 3 and 4 to special subcategories and in particular to module categories. Some known results are generalized. Finally we propose some questions and conjectures concerning the obtained results, which are closely related to the generalized Nakayama conjecture and the strong Nakayama conjecture. Throughout this paper, A is an abelian category and all subcategories of A are full and additive. 2. Preliminaries In this section, we give some terminology and some preliminary results. Definition 2.1. ([Hu]) Let C be a subcategory of A and n ≥ 0. (1) If there exists an exact sequence 0 → M → Cn−1 → Cn−2 → · · · → C0 → A → 0 in A with all Ci objects in C , then M is called an n-C -syzygy object (of A), and A is called an n-C -cosyzygy object (of M ); in this case, we denote by M = Ωn We denote by Ωn C (M ). C (A )) the subcategory of A consisting of n-C -syzygy (resp. C (A) and A = Ω−n C (A ) (resp. Ω−n 2 n-C -cosyzygy) objects. (2) For an object A in A , the C -dimension (resp. C -codimension), denoted by C - dim A (resp. C - codim A), is defined as inf{n ≥ 0 there exists an exact sequence 0 → Cn → · · · → C1 → C0 → A → 0 (resp. 0 → A → C 0 → C 1 → · · · → C n → 0) in A with all Ci (resp. C i) objects in C }. Set C - dim A (resp. C - codim A) = ∞ if no such integer exists. Let C be a subcategory of A . We denote by C ⊥ = {A is an object in A Exti A (C, A) = 0 for any object C in C and i ≥ 1} and ⊥C = {A is an object in A Exti A (A, C) = 0 for any object C in C and i ≥ 1}. Lemma 2.2. Let C and D be subcategories of A , and let M be an object in ⊥C and M C (M ′) is an object in D ⊥. If D- dim M ≤ n(< ∞), C (A ) such that some Ωn ′ an object in Ω−n then Exti A (M, M ′ ) = 0 for any i ≥ 1. Proof. By assumption, there exists an exact sequence: ′′ 0 → M → Cn−1 → · · · → C1 → C0 → M ′ → 0 in A with all Ci objects in C and M Exti ) ∼= Extn+i A (M, M ′′ A (M, M ′ ′′ an object in D ⊥. Let M be an object in ⊥C . Then ) for any i ≥ 1. If D- dim M ≤ n(< ∞), then there exists an exact sequence: 0 → Dn → · · · → D1 → D0 → M → 0 in A with all Di objects in D. So Extn+i hence Exti A (M, M ′ ) = 0 for any i ≥ 1. A (M, M ′′ ) ∼= Exti A (Dn, M ′′ ) = 0 for any i ≥ 1 and (cid:3) Let E be a subcategory of A . Recall from [EJ2] that a sequence: S : · · · → S1 → S2 → S3 → · · · in A is called HomA (E , −)-exact (resp. HomA (−, E )-exact) if HomA (E, S) (resp. HomA (S, E)) is exact for any object E in E . An epimorphism (resp. a monomorphism) f in A is called E -proper (resp. E -coproper) if it is HomA (E , −)-exact (resp. HomA (−, E )-exact). Proposition 2.3. Let C and E be subcategories of A and let C be closed under kernels of (E -proper) epimorphisms. If 0 → A1 → A2 → A3 → 0 (2.1) is a (HomA (E , −)-exact) exact sequence in A with A3 an object in C , then C - dim A1 ≤ C - dim A2. 3 Proof. Let C - dim A2 = n(< ∞) and 0 → Cn → · · · → C1 → C0 → A2 → 0 be an exact sequence in A with all Ci objects in C . By [Hu, Theorem 3.2], there exist exact sequences: and 0 → Cn → · · · → C1 → C → A1 → 0 0 → C → C0 → A3 → 0 (2.2) From the proof of [Hu, Theorem 3.2] we see that if (2.1) is HomA (E , −)-exact, then so is (2.2). Because C is closed under kernels of (E -proper) epimorphisms and A3 is an object in C by assumption, C is an object in C and C - dim A1 ≤ n. (cid:3) Let C be a subcategory of A . We denote by C ⊥C if Exti A (C1, C2) = 0 for any objects C1, C2 in C and i ≥ 1, and denote by C - dim<∞ (resp. C - codim<∞) the subcategory of A consisting of objects with finite C -dimension (resp. C -codimension). Lemma 2.4. Let C be a subcategory of A such that C ⊥C and C - dim<∞ is closed under direct summands, and let 0 → K → C → A → 0 be an exact sequence in A with C - dim A < ∞ and C an object in C . If K is an object in C ⊥, then C - dim K < ∞. Proof. Because C - dim A < ∞, there exists an exact sequence: 0 → M → C0 → A → 0 in A with C0 an object in C and C - dim M < ∞. Consider the following pull-back diagram: 0 M 0 M 0 0 / K / N / C0 / K / C / A / 0 0 0 0 Because C ⊥C and C - dim M < ∞, it is easy to get that M ∈ C ⊥ by dimension shifting. So the middle column in the above diagram splits, and hence C - dim N ≤ C - dim M < ∞ by 4 / / / / / / / / / [Hu, Lemma 3.1]. Because K is an object in C ⊥ by assumption, the middle row in the above diagram also splits and K is isomorphic to a direct summand of N . Thus C - dim K < ∞. (cid:3) Definition 2.5. Let C ⊆ T be subcategories of A . (1) (cf. [SSW]) C is called a generator (resp. cogenerator) for T if for any object T in T , there exists an exact sequence 0 → T ′ → C → T → 0 (resp. 0 → T → C → T ′ → 0) in T with C an object in C . (2) Let E be a subcategory of A . C is called an E -proper generator (resp. E -coproper cogenerator) for T if for any object T in T , there exists a HomA (E , −) (resp. HomA (−, E ))- exact exact sequence 0 → T ′ C is an object in C and T ′ → C → T → 0 (resp. 0 → T → C → T ′ → 0) in A such that is an object in T . Lemma 2.6. Let C ⊆ T be subcategories of A such that C is a cogenerator for T , and let 0 → A1 → A2 → A3 → 0 be an exact sequence in A such that both A2 and A3 are objects in T ⊥. If A1 is an object in C ⊥, then A1 is an object in T ⊥. Proof. Let 0 → A1 → A2 → A3 → 0 be an exact sequence in A such that both A2 and A3 are objects in T ⊥. Then Exti is a cogenerator for T by assumption, there exists an exact sequence: A (T, A1) = 0 for any object T in T and i ≥ 2. Because C in A with C an object in C and T ′ an object in T , which yields an exact sequence: 0 → T → C → T ′ → 0 Exti A (C, A1) → Exti A (T, A1) → Exti+1 A (T ′ , A1) for any i ≥ 1. Note that Exti+1 is an object in C ⊥, then Exti A (T ′, A1) = 0 for any i ≥ 1 by the above argument. So, if A1 A (T, A1) = 0 for any i ≥ 1 and A1 is an object in T ⊥. (cid:3) Lemma 2.7. Let C ⊆ T be subcategories of A such that C is a cogenerator for T and C is closed under direct summands. Then T T T ⊥ ⊆ C . Proof. Let T be an object in T T T ⊥. Then there exists a split exact sequence: 0 → T → C → T ′ → 0 in A with C an object in C and T ′ an object in T . So T is isomorphic to a direct summand of C. Because C is closed under direct summands by assumption, T is an object in C . (cid:3) Sather-Wagstaff, Sharif and White introduced the Gorenstein category G(C ) as follows. Definition 2.8. ([SSW]) Let C be a subcategory of A . The Gorenstein subcategory G(C ) of A is defined as G(C ) = {A is an object in A there exists an exact sequence 5 · · · → C1 → C0 → C 0 → C 1 → · · · HomA (C , −)-exact and HomA (−, C )-exact, such that A ∼= Im(C0 → C 0)}. in A with all terms objects in C , which is both The Gorenstein category unifies the following notions: modules of Gorenstein dimen- sion zero ([AB]), Gorenstein projective modules, Gorenstein injective modules ([EJ1]), V - Gorenstein projective modules, V -Gorenstein injective modules ([EJL]), W-Gorenstein mod- ules ([GD]), and so on (see [Hu] for the details). 3. Computation and Comparison of Homological Dimensions In this section, we introduce the notion of (pre)resolving subcategories of A . Then we give some criteria for computing and comparing homological dimensions relative to different preresolving subcategories. Definition 3.1. Let E and T be subcategories of A . Then T is called E -preresolving in A if the following conditions are satisfied. (1) T admits an E -proper generator. (2) T is closed under E -proper extensions, that is, for any HomA (E , −)-exact exact sequence 0 → A1 → A2 → A3 → 0 in A , if both A1 and A3 are objects in T , then A2 is also an object in T . An E -preresolving subcategory T of A is called E -resolving if the following condition is satisfied. (3) T is closed under kernels of E -proper epimorphisms, that is, for any HomA (E , −)- exact exact sequence 0 → A1 → A2 → A3 → 0 in A , if both A2 and A3 are objects in T , then A1 is also an object in T . The following list shows that the class of E -(pre)resolving subcategories is rather large. Example 3.2. (1) Let A admit enough projective objects and E the subcategory of A consisting of projective objects. Then a subcategory of A closed under E -proper extensions is just a subcategory of A closed under extensions. Furthermore, if C = E in the above definition, then an E -preresolving subcategory is just a subcategory which contains all projective objects and is closed under extensions, and an E -resolving subcategory is just a projectively resolving subcategory in the sense of [H2]. (2) Let C be a subcategory of A with C ⊥C . Then by [SSW, Corollary 4.5], the Goren- stein subcategory G(C ) of A is a C -preresolving subcategory of A with a C -proper generator C ; furthermore, if C is closed under kernels of epimorphisms, then G(C ) is a C -resolving subcategory of A by [SSW, Theorem 4.12(a)]. 6 (3) Let R be a ring, Mod R the category of left R-modules and P(Mod R) the subcategory of Mod R consisting of projective modules. Recall from [EJ2] that a pair of subcategories (X , Y ) of Mod R is called a cotorsion pair if X = {X ∈ Mod R Ext1 any Y ∈ Y } and Y = {Y ∈ Mod R Ext1 R(X, Y ) = 0 for R(X, Y ) = 0 for any X ∈ X }. If (X , Y ) is a cotorsion pair in Mod R, then X is a P(Mod R)-preresolving subcategory of Mod R with a P(Mod R)-proper generator P(Mod R) ([EJ2]). (4) Let R be a ring and F(Mod R) the subcategory of Mod R consisting of flat modules. Then by [Hu, Lemma 3.1 and Theorem 3.2], it is not difficult to see that the subcategory of Mod R consisting of strongly Gorenstein flat modules (see [DLM] or Section 5 below for the definition) is an F(Mod R)-resolving subcategory of Mod R with an F(Mod R)-proper generator P(Mod R). (5) Let R be a ring. Then, the subcategory of Mod R consisting of the modules A satisfy- ing Exti R(A, P ) = 0 for any P ∈ P(Mod R) and i ≥ 1, is a P(Mod R)-resolving subcategory of Mod R with a P(Mod R)-proper generator P(Mod R). Let R be a left noetherian ring, mod R the category of finitely generated left R-modules and P(mod R) the subcategory of mod R consisting of projective modules. Then the subcategory of mod R consisting of the modules A satisfying Exti R(A, R) = 0 for any i ≥ 1 is a P(mod R)-resolving subcategory of mod R with a P(mod R)-proper generator P(mod R). Unless stated otherwise, in the rest of this section, we fix a subcategory E of A and an E -preresolving subcategory T of A admitting an E -proper generator C . We will give some criteria for computing the T -dimension of a given object A in A , and then compare it with the C -dimension of A. The following two propositions play a crucial role in this section. Proposition 3.3. Let 0 → M → T1 f −→ T0 → A → 0 be an exact sequence in A with both T0 and T1 objects in T . Then we have (1) There exists an exact sequence: 0 → M → T → C → A → 0 in A with T an object in T and C an object in C . (2) If (3.1) is HomA (X, −)-exact for some object X in A , then so is (3.2). Proof. (1) Let (3.1) (3.2) 0 → M → T1 f −→ T0 → A → 0 7 be an exact sequence in A with both T0 and T1 objects in T . Because there exists a HomA (E , −)-exact exact sequence: ′ 0 → T 0 → C → T0 → 0 in A with C an object in C and T 0 an object in T , we have the following pull-back diagram: ′ 0 ′ T 0 0 ′ T 0 0 0 / W / C / Im f / T0 / A / A / 0 / 0 Then consider the following pull-back diagram: 0 0 0 ′ T 0 0 ′ T 0 0 0 / M / T / W / M / T1 / Im f / 0 0 0 0 Because the middle column in the first diagram is HomA (E , −)-exact, the first column in the first diagram (that is, the third column in the second diagram) and the middle column ′ in the second diagram are also HomA (E , −)-exact by [Hu, Lemma 2.4(1)]. Because both T 0 and T1 are objects in T , T is also an object in T . Connecting the middle rows in the above two diagrams we get the desired exact sequence. (2) If (3.1) is HomA (X, −)-exact for some object X in A , then so are the third rows in the above two diagrams. So the middle rows in the above two diagrams and (3.2) are also HomA (X, −)-exact by [Hu, Lemma 2.4(1)]. (cid:3) 8 / / / / / / / / / / / / / / / / / As an application of Proposition 3.3, we get the following Proposition 3.4. Let n ≥ 1 and 0 → M → Tn−1 → Tn−2 → · · · → T0 → A → 0 be an exact sequence in A with all Ti objects in T . Then there exist an exact sequence: 0 → N → Cn−1 → Cn−2 → · · · → C0 → A → 0 and a HomA (E , −)-exact exact sequence: 0 → T → N → M → 0 in A with all Ci objects in C and T an object in T . In particular, an object in A is an n-C -cosyzygy if and only if it is an n-T -cosyzygy. Proof. We proceed by induction on n. The case for n = 1 has been proved in the proof of Proposition 3.3. Now suppose that n ≥ 2 and we have an exact sequence: 0 → M → Tn−1 → Tn−2 → · · · → T0 → A → 0 in A with all Ti objects in T . Put K = Ker(T1 → T0). By Proposition 3.3, we get an exact sequence: 0 → K → T 1 → C0 → A → 0 ′ in A with T ′ 1 an object in T and C0 an object in C . Put A′ = Im(T ′ 1 → C0). Then we get an exact sequence: 0 → M → Tn−1 → Tn−2 → · · · → T2 → T ′ ′ 1 → A → 0 in A . Thus the assertion follows from the induction hypothesis. (cid:3) The following corollary is an immediate consequence of Proposition 3.4. Corollary 3.5. Let M be an object in A and n ≥ 0. If T - codim M = n. Then there exists a HomA (E , −)-exact exact sequence 0 → T → N → M → 0 in A with C - codim N ≤ n and T an object in T . Proof. Let M be an object in A with T - codim M = n. Applying Proposition 3.4 with A = 0 we get a HomA (E , −)-exact exact sequence 0 → T → N → M → 0 in A with C - codim N ≤ n and T an object in T . (cid:3) We give a criterion for computing the T -dimension of an object in A as follows. 9 Theorem 3.6. The following statements are equivalent for any object A in A and n ≥ 0. (1) T - dim A ≤ n. (2) There exists an exact sequence: 0 → Kn → Cn−1 → Cn−2 → · · · → C0 → A → 0 in A with all objects Ci in C and Kn an object in T . Proof. (2) ⇒ (1) is trivial. (1) ⇒ (2) We proceed by induction on n. The case for n = 0 is trivial. If n = 1, then there exists an exact sequence: in A with both T0 and T1 objects in T . Applying Proposition 3.3 with M = 0, we get an 0 → T1 → T0 → A → 0 exact sequence: 0 → K1 → C0 → A → 0 in A with C0 an object in C and K1 an object in T . Now suppose n ≥ 2. Then there exists an exact sequence: 0 → Tn → Tn−1 → · · · → T1 → T0 → A → 0 in A with all Ti objects in T . Put M = Ker(T1 → T0). By Proposition 3.3, we get an exact sequence: 0 → M → T 1 → C0 → A → 0 ′ in A with C0 an object in C and T 1 an object in T . Put B = Im(T 1 → C0). Then we get ′ ′ an exact sequence: 0 → Tn → Tn−1 → · · · → T 1 → B → 0. ′ By the induction hypothesis, we get the following exact sequence: 0 → Kn → Cn−1 → · · · → C1 → B → 0 in A with all Ci objects in C and Kn an object in T . Thus we get the desired exact sequence: 0 → Kn → Cn−1 → Cn−2 → · · · → C1 → C0 → A → 0. (cid:3) The following result gives a criterion for computing the T -codimension of an object in A . To some extent, the proof of this result is dual to that of Theorem 3.6, so we omit it. 10 Theorem 3.7. The following statements are equivalent for any object M in A and n ≥ 0. (1) T - codim M ≤ n. (2) There exists an exact sequence: 0 → M → K 0 → C 1 → · · · → C n−1 → C n → 0 in A with K 0 an object in T and all C i objects in C , that is, there exists an exact sequence: in A with T an object in T and C - codim H ≤ n − 1. 0 → M → T → H → 0 The following result gives a sufficient condition such that the n-C -syzygy of an object in A with T -dimension at most n is in T , in which the first assertion generalizes [AB, Lemma 3.12]. Theorem 3.8. Let T be closed under kernels of (E -proper) epimorphisms and T ⊆ C ⊥, and let n ≥ 0. Then for any object A in A with T - dim A ≤ n we have (1) For any (HomA (E , −)-exact) exact sequence 0 → Kn → Cn−1 → · · · → C1 → C0 → A → 0 in A with all Ci objects in C , Kn is an object in T . (2) If 0 → K → C → A → 0 is a (HomA (E , −)-exact) exact sequence in A with C an object in C , then T - dim K ≤ n − 1. Proof. Let T - dim A ≤ n and T ⊆ C ⊥. Then there exists a HomA (C , −)-exact exact sequence: 0 → Tn → C ′ n−1 → · · · → C ′ 1 → C ′ 0 → A → 0 in A with all C ′ i objects in C and Tn an object in T by Proposition 3.4. (1) Let 0 → Kn → Cn−1 → · · · → C1 → C0 → A → 0 be a (HomA (E , −)-exact) exact sequence in A with all Ci objects in C . Then by [Hu, Theorem 3.2] we get a (HomA (E , −)-exact) exact sequence: 0 → Kn → TnM Cn−1 → C ′ n−1 M Cn−2 · · · → C ′ 1M C0 → C ′ 0 → 0. Because T is closed under kernels of (E -proper) epimorphisms by assumption, Kn is an object in T . (2) Put T1 = Im(C ′ 1 → C ′ 0). Then we have a HomA (C , −)-exact exact sequence: 0 → T1 → C ′ 0 → A → 0 11 in A with T - dim T1 ≤ n − 1. Let 0 → K → C → A → 0 be a (HomA (E , −)-exact) exact sequence in A with C an object in C . Consider the following pull-back diagram: 0 T1 / K / X 0 T1 / C ′ 0 / K / C / A 0 0 0 0 / 0 0 Because the third column in this diagram is HomA (C , −)-exact, the middle column is also HomA (C , −)-exact by [Hu, Lemma 2.4(1)]. So the middle column splits and X ∼= T1 L C. Then the middle row yields an exact sequence: 0 → K → T1 M C → C ′ 0 → 0. By Proposition 2.3, T - dim K ≤ T - dim T1L C ≤ n − 1. (cid:3) We use T - dim≤n to denote the subcategory of A consisting of objects with T -dimension at most n. Corollary 3.9. Let T be a C -resolving subcategory of A with a C -proper generator C and T ⊆ C ⊥. If T is closed under direct summands, then so is T - dim≤n for any n ≥ 0. Proof. The case for n = 0 follows from the assumption. Now Let n ≥ 1 and let A be an object in A with T - dim A ≤ n and A = A1 L A2. Because T ⊆ C ⊥ by assumption, there exists a HomA (C , −)-exact exact sequence: 0 → Kn → Cn−1 → Cn−2 → · · · → C0 f0→ A → 0 in A with all Ci objects in C and Kn an object in T by Theorem 3.6. Note that both and 0 → A2 ) ( 0 1A2 −→ A (1A1 ,0) −→ A1 → 0 0 → A1 (1A1 0 ) −→ A (0,1A2 ) −→ A2 → 0 12 / / / / / / / / / are exact and split. So both and (1A1 ,0)f0 −→ A1 → 0 C0 (0,1A2 )f0 −→ A2 → 0 C0 are HomA (C , −)-exact exact sequences. By [Hu, Theorem 3.6], we get the following HomA (C , −)- exact exact sequences: and C0M C1 → C0 → A1 → 0 C0 M C1 → C0 → A2 → 0. Again by [Hu, Theorem 3.6], we get the following HomA (C , −)-exact exact sequences: C0M C1M C2 → C0M C1 → C0 → A1 → 0 and C0 M C1 M C2 → C0 M C1 → C0 → A2 → 0. Continuing this procedure, we finally get the following HomA (C , −)-exact exact sequences: 0 → Xn → n−1M i=0 Ci → n−2M i=0 Ci → · · · → C1M C0 → C0 → A1 → 0 and 0 → Yn → n−1M i=0 Ci → n−2M i=0 Ci → · · · → C1 M C0 → C0 → A2 → 0. Put Uj = Lj i=0 Ci for any 0 ≤ j ≤ n − 1. Then we get a HomA (C , −)-exact exact sequence: 0 → XnM Yn → Un−1M Un−1 → Un−2 M Un−2 → · · · → U1 M U1 → U0M U0 → A → 0. By Theorem 3.8, XnL Yn is an object in T . So both Xn and Yn are objects in T and hence T - dim A1 ≤ n and T - dim A2 ≤ n. (cid:3) The following result gives some sufficient conditions such that the T -dimension and the C -dimension of an object in A are identical. Theorem 3.10. Let T ⊆ C ⊥T ⊥C and C be closed under direct summands. Then for an object A in A , T - dim A = C - dim A if one of the following conditions is satisfied. (1) C - dim A < ∞, E = C and T is closed under kernels of C -proper epimorphisms. (2) C - dim A < ∞, E = C and C - dim<∞ is closed under direct summands. 13 (3) A is an object in T ⊥ and C is a cogenerator for T . Proof. It is trivial that C - dim A ≥ T - dim A. In the following we prove C - dim A ≤ T - dim A. Suppose T - dim A = n < ∞. (1) Let C - dim A = t(< ∞). If n < t, then consider the following HomA (C , −)-exact exact sequence: 0 → Ct → · · · → Cn → Cn−1 → · · · → C1 → C0 → A → 0 in A with all Ci objects in C . Put Kn = Im(Cn → Cn−1). So Kn is an object in T by Theorem 3.8(1), and hence an object in ⊥C by assumption. It yields that the exact sequence: 0 → Ct → · · · → Cn → Kn → 0 splits and Kn is isomorphic to a direct summand of Cn. Because C is closed under direct summands by assumption, Kn is an object in C and C - dim A ≤ n, which is a contradiction. So n ≥ t. In the following, we prove (2) and (3). Let 0 → Tn → Tn−1 → · · · → T1 → T0 → A → 0 be an exact sequence in A with all Ti objects in T . By Proposition 3.4, we get an exact sequence: 0 → Kn → Cn−1 → · · · → C1 → C0 → A → 0 and a HomA (E , −)-exact exact sequence: 0 → T → Kn → Tn → 0 in A with all Ci objects in C and T an object in T . So Kn is an object in T (⊆ C ⊥T ⊥C ). (2) Because C - dim A < ∞, C - dim Kn < ∞ by Lemma 2.4. By assumption T ⊆ ⊥C , it is easy to see that Kn is isomorphic to a direct summand of some object in C . Since C is closed under direct summands by assumption, Kn is an object in C and C - dim A ≤ n. (3) Let A be an object in T ⊥ and Ki = Im(Ci → Ci−1) for any 1 ≤ i ≤ n − 1. Then all Ki are objects in C ⊥. By Lemma 2.6, all Ki are objects in T ⊥, and in particular Kn is an object in T ⊥. So Kn is an object in C by Lemma 2.7, and hence C - dim A ≤ n. (cid:3) The following result gives a sufficient condition such that the T -codimension and the C -codimension of an object in A are identical. 14 Theorem 3.11. Let D be a subcategory of A such that T ⊆ ⊥C T D ⊥, and let C - codim≤n be closed under direct summands for any n ≥ 0. If M is an object in A with D- dim M < ∞, then T - codim M = C - codim M . Proof. It is clear that C - codim M ≥ T - codim M . In the following we prove T - codim M ≥ C - codim M . Without loss of generality, assume T - codim M = n < ∞. If n = 0, then M is an object in T and there exists a HomA (E , −)-exact exact sequence: ′ 0 → M → C → M → 0 (3.3) in A with C an object in C and M ′ an object in T . Notice that T ⊆ ⊥C T D ⊥ by assumption, so Exti ) = 0 for any i ≥ 1 by Lemma 2.2. It follows that the exact A (M, M ′ sequence (3.3) splits, which implies that M is isomorphic to a direct summand of C. Because C is closed under direct summands by assumption, M is an object in C . Now suppose n ≥ 1. By Theorem 3.7, there exists an exact sequence: 0 → M → T → H → 0 in A with T an object in T and C - codim H ≤ n − 1. It is easy to see that M is an object in ⊥C . Because there exists a HomA (E , −)-exact exact sequence: 0 → T ′ ′ → C → T → 0 in A with C ′ an object in C and T ′ an object in T , we have the following pull-back diagram: 0 ′ T / N / M 0 0 ′ T ′ / C / T 0 0 0 / H / H / 0 / 0 By the exactness of the middle row in the above diagram, C - codim N ≤ n. Because A (M, T ′) = 0 by Lemma 2.2, the first column in the above diagram splits. So M is Ext1 15 / / / / / / / / isomorphic to a direct summand of N . Because C - codim≤n is closed under direct sum- mands by assumption, C - codim M ≤ n. (cid:3) In the following, we fix a subcategory C of A . The following two corollaries give some sufficient conditions such that the G(C )-dimension and the C -dimension of an object in A are identical. The first one is a generalization of [Z, Theorem 2.3]. Corollary 3.12. Let C ⊥C and let C be closed under direct summands. Then for any object A in G(C )⊥, G(C )- dim A = C - dim A. Proof. Let C ⊥C . It is clear that C is a C -proper generator and a C -coproper cogenerator for G(C ). By [SSW, Corollary 4.5], G(C ) is closed under extensions. By [Hu, Lemma 5.7], G(C ) ⊆ C ⊥T ⊥C . Now the assertion follows from Theorem 3.10(3). (cid:3) The following is a generalization of [H1, Theorem 2.2] and [Z, Corollary 2.5]. Corollary 3.13. Let C ⊥C and let C be closed under direct summands, and let D be a subcategory of G(C )⊥. Then for any object A in C ⊥ with D- codim A < ∞, G(C )- dim A = C - dim A. Proof. Let A be an object in C ⊥ with D- codim A < ∞. Because D is a subcategory of G(C )⊥, it is easy to see that A is an object in G(C )⊥ by Lemma 2.6. Then the assertion follows from Corollary 3.12. (cid:3) The following result gives a sufficient condition such that the G(C )-dimension and the ⊥C -dimension of an object in A are identical. Theorem 3.14. Let C ⊥C and let A an object in A with G(C )- dim A < ∞. Then G(C )- dim A = ⊥C - dim A. Proof. By [Hu, Lemma 5.7], G(C )- dim A ≥ ⊥C - dim A. In the following we prove G(C )- dim A ≤ ⊥C - dim A. Suppose ⊥C - dim A = n < ∞ and G(C )- dim A = m < ∞. If n = 0, then A is an object in ⊥C . So by [Hu, Theorem 5.8], A is an object in G(C ) and m = 0. Let n ≥ 1. Then Extn+i A (A, C) = 0 for any object C in C and i ≥ 1. If m > n, then consider the following exact sequence: 0 → Gm → · · · → Gn → Gn−1 → G1 → G0 → A → 0 in A with all Gi objects in G(C ). Putting Kn = Im(Gn → Gn−1), then Kn is an object in ⊥C and G(C )- dim Kn ≤ m − n < ∞. By the above argument, Kn is an object in G(C ). 16 So G(C )- dim A ≤ n, which is a contradiction. Thus m ≤ n. It follows that G(C )- dim A ≤ ⊥C - dim A. (cid:3) 4. Dual Results In this section, we introduce the notion of (pre)coresolvmg subcategories of A . Then we give some criteria for computing and comparing homological codimensions relative to differ- ent precoresolving subcategories. The results and their proofs in this section are completely dual to that in Section 3, so we only list the results without proofs. Definition 4.1. Let E and T be subcategories of A . Then T is called E -precoresolving in A if the following conditions are satisfied. (1) T admits an E -coproper cogenerator. (2) T is closed under E -coproper extensions, that is, for any HomA (−, E )-exact exact sequence 0 → A1 → A2 → A3 → 0 in A , if both A1 and A3 are objects in T , then A2 is also an object in T . An E -precoresolving subcategory T of A is called E -coresolving if the following condition is satisfied. (3) T is closed under cokernels of E -coproper monomorphisms, that is, for any HomA (−, E )- exact exact sequence 0 → A1 → A2 → A3 → 0 in A , if both A1 and A2 are objects in T , then A3 is also an object in T . The following list shows that the class of E -(pre)coresolving subcategories is rather large. Example 4.2. (1) Let A admit enough injective objects and E the subcategory of A consisting of injective objects. Then a subcategory of A closed under E -coproper extensions is just a subcategory of A closed under extensions. Furthermore, if C = E in the above definition, then an E -precoresolving subcategory is just a subcategory which contains all injective objects and is closed under extensions, and an E -coresolving subcategory is just an injectively coresolving subcategory in the sense of [H2]. (2) Let C be a subcategory of A with C ⊥C . Then by [SSW, Corollary 4.5], the Goren- stein subcategory G(C ) of A is a C -precoresolving subcategory of A with a C -coproper cogenerator C ; furthermore, if C is closed under cokernels of monomorphisms, then G(C ) is a C -coresolving subcategory of A by [SSW, Theorem 4.12(b)]. (3) Let R be a ring and I(Mod R) the subcategory of Mod R consisting of injective modules. If (X , Y ) is a cotorsion pair in Mod R, then Y is an I(Mod R)-precoresolving subcategory of Mod R with an I(Mod R)-coproper cogenerator I(Mod R) ([EJ2]). 17 (4) Let R be a ring. Recall that a module E in Mod R is called FP-injective if Ext1 R(M, E) = 0 for any finitely presented left R-module M . FP-injective modules are also known as abso- lutely pure modules. We use FI(Mod R) to denote the subcategory of Mod R consisting of FP-injective modules. Then by [Hu, Lemma 3.1 and Theorem 3.4], it is not difficult to see that the subcategory of Mod R consisting of Gorenstein FP-injective modules (see [MD] or Section 5 below for the definition) is an FI(Mod R)-coresolving subcategory of Mod R with an FI(Mod R)-coproper cogenerator I(Mod R). (5) Let R be a ring. We denote by cores ^P(Mod R) = {M ∈ Mod R there exists a HomR(−, P(Mod R))-exact exact sequence 0 → M → P 0 → P 1 → · · · → P i → · · · in Mod R with all P i projective}. Then by [Hu, Lemma 3.1 and Theorem 3.8], it is easy to see that cores ^P(Mod R) is a P(Mod R)-coresolving subcategory of Mod R with a P(Mod R)-coproper cogenerator P(Mod R). Let R be a left and right noetherian ring. Then by [AB, Theorem 2.17] and [Hu, Lemma 3.1], the subcategory of mod R consisting of ∞-torsionfree modules (see [HuH] or Section 5 below for the definition) is a P(mod R)-coresolving subcategory of mod R with a P(mod R)-coproper cogenerator P(mod R). Unless stated otherwise, in the rest of this section, we fix a subcategory E of A and an E -precoresolving subcategory T of A admitting an E -coproper cogenerator C . We will give some criteria for computing the T -codimension of a given object A in A , and then compare it with the C -codimension of A. The following two propositions play a crucial role in this section. Proposition 4.3. Let 0 → M → T 0 → T 1 → A → 0 be an exact sequence in A with both T 0 and T 1 objects in T . Then we have (1) There exists an exact sequence: 0 → M → C → T → A → 0 in A with T an object in T and C an object in C . (2) If (4.1) is HomA (−, X)-exact for some object X in A , then so is (4.2). (4.1) (4.2) As an application of Proposition 4.3, we get the following Proposition 4.4. Let n ≥ 1 and 0 → M → T 0 → T 1 → · · · → T n−1 → A → 0 be an exact sequence in A with all T i objects in T . Then there exist an exact sequence: 0 → M → C 0 → C 1 → · · · → C n−1 → B → 0 18 and a HomA (−, E )-exact exact sequence: 0 → A → B → T → 0 in A with all C i objects in C and T an object in T . In particular, an object in A is an n-C -syzygy if and only if it is an n-T -syzygy. The following corollary is an immediate consequence of Proposition 4.4. Corollary 4.5. Let A be an object in A and n ≥ 0. If T - dim A = n. Then there exists a HomA (−, E )-exact exact sequence 0 → A → B → T → 0 in A with C - dim B ≤ n and T an object in T . We give a criterion for computing the T -codimension of an object in A as follows. Theorem 4.6. The following statements are equivalent for any object M in A and n ≥ 0. (1) T - codim M ≤ n. (2) There exists an exact sequence: 0 → M → C 0 → C 1 → · · · → C n−1 → K n → 0 in A with all objects C i in C and K n an object in T . The following result gives a criterion for computing the T -dimension of an object in A . Theorem 4.7. The following statements are equivalent for any object A in A and n ≥ 0. (1) T - dim A ≤ n. (2) There exists an exact sequence: 0 → Cn → Cn−1 → · · · → C1 → K0 → A → 0 in A with K0 an object in T and all Ci objects in C , that is, there exists an exact sequence: 0 → H → T → A → 0 in A with T an object in T and C - dim H ≤ n − 1. The following result gives a sufficient condition such that the n-C -cosyzygy of an object in A with T -codimension at most n is in T . Theorem 4.8. Let T be closed under cokernels of (E -coproper) monomorphisms and T ⊆ ⊥C , and let n ≥ 0. Then for any object M in A with T - codim M ≤ n we have 19 (1) For any (HomA (−, E )-exact) exact sequence 0 → M → C 0 → C 1 → · · · → C n−1 → K n → 0 in A with all C i objects in C , K n is an object in T . (2) If 0 → M → C → K → 0 is a (HomA (−, E )-exact) exact sequence in A with C an object in C , then T - codim K ≤ n − 1. We use T - codim≤n to denote the subcategory of A consisting of objects with T - codimension at most n. Corollary 4.9. Let T be a C -coresolving subcategory of A with a C -coproper cogenera- tor C and T ⊆ ⊥C . If T is closed under direct summands, then so is T - codim≤n for any n ≥ 0. The following result gives some sufficient conditions such that the T -codimension and the C -codimension of an object in A are identical. Theorem 4.10. Let T ⊆ C ⊥T ⊥C and C be closed under direct summands. Then for an object M in A , T - codim M = C - codim M if one of the following conditions is satisfied. (1) C - codim M < ∞, E = C and T is closed under cokernels of C -coproper monomor- phisms. (2) C - codim M < ∞, E = C and C - dim<∞ is closed under direct summands. (3) M is an object in ⊥T and C is a generator for T . The following result gives a sufficient condition such that the T -dimension and the C - dimension of an object in A are identical. Theorem 4.11. Let D be a subcategory of A such that T ⊆ C ⊥T ⊥D, and let C - dim≤n be closed under direct summands for any n ≥ 0. If A is an object in A with D- codim A < ∞, then T - dim A = C - dim A. In the following, we fix a subcategory C of A . The following two corollaries give some sufficient conditions such that the G(C )-codimension and the C -codimension of an object in A are identical. Corollary 4.12. Let C ⊥C and let C be closed under direct summands. Then for any object M in ⊥G(C ), G(C )- codim M = C - codim M . Corollary 4.13. Let C ⊥C and let C be closed under direct summands, and let D be a subcategory of ⊥G(C ). Then for any object M in ⊥C with D- dim M < ∞, G(C )- codim M = C - codim M . The following result gives a sufficient condition such that the G(C )-codimension and the 20 C ⊥-codimension of an object in A are identical. Theorem 4.14. Let C ⊥C and let M an object in A with G(C )- codim M < ∞. Then G(C )- codim M = C ⊥- codim M . 5. Applications and Questions In this section, we will apply the results in Sections 3 and 4 to special subcategories and in particular to module categories. Finally we propose some open questions and conjectures concerning the obtained results. 5.1. Special subcategories We define res eC = {A is an object in A there exists a HomA (C , −)-exact exact sequence · · · → Ci → · · · → C1 → C0 → A → 0 in A with all Ci objects in C }. Dually, we define cores eC = {M is an object in A there exists a HomA (−, C )-exact exact sequence 0 → M → C 0 → C 1 → · · · → C i → · · · in A with all C i objects in C } (see [SSW]). We have the following Fact 5.1. (1) Note that C is a C -proper generator for res eC and res eC T ⊥C . By [Hu, Lemma 3.1(1)], both res eC and res eC T ⊥C are closed under C -proper extensions. So both res eC and res eC T ⊥C are C -preresolving. We remark that if C is a C -proper generator for A , then res eC = A and res eC T ⊥C = ⊥C . (2) If C is closed under kernels of epimorphisms, then so are both res eC and res eC T ⊥C ([Hu, Proposition 4.7(1)]). Dually, we have the following (3) Note that C is a C -coproper cogenerator for cores eC and C ⊥T cores eC . By [Hu, Lemma 3.1(2)], both cores eC and C ⊥T cores eC are closed under C -coproper extensions. So both cores eC and C ⊥T cores eC are C -precoresolving. We also remark that if C is a C -coproper cogenerator for A , then cores eC = A and C ⊥T cores eC = C ⊥. (4) If C is closed under cokernels of monomorphisms, then so are both cores eC and C ⊥T cores eC ([Hu, Proposition 4.7(2)]). T = res eC and T = res eC T ⊥C respectively, and apply the results in Section 4 in the cases for T = cores eC and T = C ⊥T cores eC respectively. We will not list these consequences in Application 5.2. By Fact 5.1, we can apply the results in Section 3 in the cases for details. 5.2. Module categories 21 In this subsection, R is a ring and all subcategories of Mod R are full and additive. For a module A in Mod R, we denote the projective, injective and flat dimensions of A by pdR A, idR A and fdR A respectively. We first give the following Proposition 5.3. Let T be a subcategory of Mod R. (1) If T is closed under extensions and P(Mod R) ⊆ T ⊆ ⊥P(Mod R), then pdR A = T - dim A for any A ∈ Mod R with pdR A < ∞. (2) If T is closed under extensions and I(Mod R) ⊆ T ⊆ I(Mod R)⊥, then idR A = T - codim A for any A ∈ Mod R with idR A < ∞. Proof. (1) Because T ⊆ ⊥P(Mod R) = P(Mod R)⊥T ⊥P(Mod R) by assumption, we get the assertion by Theorem 3.10(2). (2) It is dual to (1). (cid:3) Let (X , Y ) be a cotorsion pair in Mod R. Then X T Y is called the heart of (X , Y ). A cotorsion pair (X , Y ) is called hereditary if X = ⊥Y and Y = X ⊥; in this case, X is projectively resolving and Y is injectively coresolving ([GT, Lemma 2.2.10]). By Proposition 5.3 we immediately have the following Corollary 5.4. Let (X , Y ) be a hereditary cotorsion pair in Mod R with the heart C (= X T Y ). (1) If C = P(Mod R), then for any A ∈ Mod R with pdR A < ∞, pdR A = X - dim A. (2) If C = I(Mod R), then for any A ∈ Mod R with idR A < ∞, idR A = Y - codim A. Note that a module in G(P(Mod R)) (resp. G(I(Mod R))) is just a Gorenstein projective (resp. injective) module in Mod R. So G(P(Mod R))- dimR A (resp. G(I(Mod R))- codimR A) is just the Gorenstein projective (resp. injective) dimension of a module A in Mod R. We denote the Gorenstein projective (resp. injective) dimension of a module A in Mod R by GpdR A (resp. GidR A). Definition 5.5. Let A be a module in Mod R. (1) ([DLM]) A is called strongly Gorenstein flat if there exists a HomR(−, F(Mod R))- exact exact sequence · · · → P1 → P0 → P 0 → P 1 → · · · in Mod R with all terms projective, such that A ∼= Im(P0 → P 0). We use SGF(Mod R) to denote the subcategory of Mod R consisting of strongly Gorenstein flat modules. The strongly Gorenstein flat dimension SGfdR A of A is defined to be inf{n there exists an exact sequence 0 → Gn → · · · → G1 → G0 → A → 0 in Mod R with all Gi in 22 SGF (Mod R)}. Set SGfdR A = ∞ if no such n exists. (2) ([MD]) A is called Gorenstein FP-injective if there exists a HomR(FI(Mod R), −)- exact exact sequence · · · → I1 → I0 → I 0 → I 1 → · · · in Mod R with all terms injective, such that A ∼= Im(I0 → I 0). We use GFI(Mod R) to denote the subcategory of Mod R consisting of Gorenstein FP-injective modules. The Gorenstein FP-injective dimension GFidR A of A is defined to be inf{n there exists an exact sequence 0 → A → H 0 → H 1 → · · · → H n → 0 in Mod R with all H i in GFI(Mod R)}. Set GFidA = ∞ if no such n exists. It is trivial that there exist the following inclusions: P(Mod R) ⊆ S GF(Mod R) ⊆ G(P(Mod R))( ⊆ cores ^ P(Mod R), ⊆ ⊥(P(Mod R)) ⊇ ⊥(F(Mod R)) ⊇ S GF(Mod R), and I(Mod R) ⊆ GF I(Mod R) ⊆ G(I(Mod R))( ⊆ res ^I(Mod R), ⊆ (I(Mod R))⊥ ⊇ (F I(Mod R))⊥ ⊇ GF I(Mod R). So for any module A in Mod R, we have pdR A ≥ SGfdR A ≥ GpdR A( ≥ cores ^ P(Mod R)- dim A, ≥ ⊥(P(Mod R))- dim A ≤ ⊥(F(Mod R))- dim A ≤ SGfdR A, and idR A ≥ GFidR A ≥ GidR A( ≥ res ^ I(Mod R)- codim A, ≥ (I(Mod R))⊥- codim A ≤ (F I(Mod R))⊥- codim A ≤ GFidR A. Theorem 5.6. Let A be a module in Mod R. (1) If A ∈ (SGF (Mod R))⊥, then pdR A = SGfdR A. (2) If A ∈ (G(P(Mod R)))⊥, then pdR A = SGfdR A = GpdR A. (3) If idR A < ∞, then pdR A = SGfdR A = GpdR A = cores ^P(Mod R)- dim A. (4) If pdR A < ∞, then pdR A = SGfdR A = GpdR A = ⊥(P(Mod R))- dim A = ⊥(F(Mod R))- dim A. (5) If SGfdR A < ∞, then SGfdR A = GpdR A = ⊥(P(Mod R))- dim A = ⊥(F(Mod R))- dim A. (6) If GpdR A < ∞, then GpdR A = ⊥(P(Mod R))- dim A. (7) If fdR A < ∞, then pdR A = SGfdR A. Proof. (1) (resp. (2)) It is clear that P(Mod R) is both a P(Mod R)-proper generator and a P(Mod R)-coproper cogenerator for SGF (Mod R) (resp. G(P(Mod R))). Note that SGF (Mod R) (resp. G(P(Mod R))) is closed under extensions by [Hu, Lemma 3.1] (resp. [H2, Theorem 2.5]). Then by putting T = SGF (Mod R) (resp. T = G(P(Mod R))) and 23 C = P(Mod R) in Theorem 3.10(3), we have pdR A = SGfdR A (resp. pdR A = GpdR A) if A ∈ (SGF (Mod R))⊥ (resp. A ∈ (G(P(Mod R)))⊥). (3) It is clear that P(Mod R) is a P(Mod R)-coproper cogenerator for cores ^P(Mod R). Note that cores ^P(Mod R) is closed under P(Mod R)-coproper extensions by [Hu, Lemma 3.1]. Then by putting T = cores ^P(Mod R), C = P(Mod R) and D = I(Mod R) in Theorem 4.11, we have pdR A = cores ^P(Mod R)- dim A if idR A < ∞. (4) By Proposition 5.3(1), we have pdR A = ⊥(P(Mod R))- dim A if pdR A < ∞. (5) Note that SGF (Mod R) is closed under extensions (see the proof of (1)). Let SGfdR A = n < ∞. Then by Theorem 3.6, there exists an exact sequence: 0 → Gn → Pn−1 → Pn−2 → · · · → P0 → A → 0 in Mod R with all Pi projective and Gn strongly Gorenstein flat. Put Ki = Im(Pi → Pi−1) for any 1 ≤ i ≤ n − 1. Suppose ⊥(P(Mod R))- dim A = m < ∞. It suffices to show m ≥ n. If m < n, then Km ∈ ⊥(P(Mod R)) and Kn−1 ∈ ⊥(P(Mod R)) by Theorem 3.8(1). Because there exists a HomR(−, F(Mod R))-exact exact sequence 0 → Gn → P → G → 0 in Mod R with P projective and G strongly Gorenstein flat, we have the following push-out diagram: 0 0 0 0 / Gn / Pn−1 / Kn−1 / P G 0 / G′ / Kn−1 G 0 / 0 / 0 By using an argument similar to that in the proof of [H2, Theorem 2.5], we get that SGF (Mod R) is closed under direct summands. Because both the middle column and the middle row in the above diagram split, G′ ∼= Pn−1 ⊕ G is strongly Gorenstein flat and Kn−1 ′ , which implies that Kn−1 is strongly Gorenstein is isomorphic to a direct summand of G flat and SGfdR A ≤ n − 1. It is a contradiction. (6) We get the assertion by putting C = P(Mod R) in Theorem 3.14 or C = P(Mod R) and T = G(P(Mod R)) in Theorem 3.10. (7) By the definition of strongly Gorenstein flat modules, it is easy to see that A ∈ (SGF (Mod R))⊥ if fdR A < ∞. Then the assertion follows from (1). (cid:3) 24 / / / / / / / / Remark 5.7. Theorem 5.6(2) is [Z, Theorem 2.3]. Theorem 5.6(3) generalizes [H1, The- orem 2.2] which states that for a module A in Mod R, if idR A < ∞, then pdR A = GpdR A. Notice that a module in Mod R with finite injective dimension is in (G(P(Mod R)))⊥, so we may also get [H1, Theorem 2.2] by Theorem 5.6(2) ([Z, Corollary 2.5]). Theorem 5.6(6) is well known ([H2, Theorem 2.20]). Let A be a module in Mod R. Recall that A is called Gorenstein flat if there exists an exact sequence: · · · → F1 → F0 → F 0 → F 1 → · · · in Mod R with all terms flat, such that A ∼= Im(F0 → F 0) and the sequence remains still exact after applying the functor I NR − for any injective right R-module I. The Gorenstein flat dimension of A, denoted by GfdR A, is defined as inf{n there exists an exact sequence 0 → Hn → · · · → H1 → H0 → A → 0 with all Hi Gorenstein flat}. Set GfdR A = ∞ if no such n exists. ([EJT, H2]). The following is an open question: whether is every Gorenstein projective module over any ring Gorenstein flat? Holm proved in [H2, Proposition 2.4] that if R is a right coherent ring with finite left finitistic projective dimension, then every Gorenstein projective module in Mod R is Gorenstein flat. As an immediate consequence of Theorem 5.6, we have the following Corollary 5.8. Let R be a right coherent ring and A a module in Mod R. Then GpdR A ≥ GfdR A if either of the following conditions is satisfied: (1) A ∈ (G(P(Mod R)))⊥ (in particular, if pdR A < ∞ or idR A < ∞), (2) SGfdR A < ∞. Proof. Let R be a right coherent ring and A a module in Mod R. Then SGfdR A ≥ GfdR A by [DLM, Proposition 2.3]. So the assertions follow from Theorem 5.6(2)(5) respectively. (cid:3) Recall from [B1] that R is called left GF-closed if the subcategory of Mod R consisting of Gorenstein flat modules is closed under extensions. Corollary 5.9. Let A be a module in Mod R and n a non-negative integer. (1) ([CFH, Lemma 2.17]) If GpdR A = n, then there exists an exact sequence 0 → A → B → T → 0 in Mod R with pdB A = n and T Gorenstein projective. (2) If SGfdR A = n, then there exists an exact sequence 0 → A → B → T → 0 in Mod R with pdB A = n and T strongly Gorenstein flat. (3) If R is left GF-closed and GfdR A = n, then there exists an exact sequence 0 → A → B → T → 0 in Mod R with fdR B = n and T Gorenstein flat. 25 Proof. (1) (resp. (2)) Let GpdR A = n (resp. SGfdR A = n). By Corollary 4.5, there exists an exact sequence 0 → A → B → T → 0 in Mod R with pdR B ≤ n and T Gorenstein strongly Gorenstein flat). Then by [H2, Theorem 2.5] (resp. Example projective (resp. 3.2(4)) and Proposition 2.3, we have GpdR B ≥ GpdR A = n (resp. SGfdR B ≥ SGfdR A = n). So pdR B = n by Theorem 5.6(4). (3) Let R be left GF-closed and GfdR A = n. By Corollary 4.5, there exists an exact sequence 0 → A → B → T → 0 in Mod R with fdR B ≤ n and T Gorenstein flat. Then by [B1, Theorem 2.3] and Proposition 2.3, we have GfdR B ≥ GfdR A = n. So fdR B = n by [B2, Theorem 2.2]. (cid:3) Recall that the FP-injective dimension FP - idR A of A in Mod R is defined to be inf{n there exists an exact sequence 0 → A → E0 → E1 → · · · → En → 0 in Mod R with all Ei in FI(Mod R)}. Set FP - idR A = ∞ if no such n exists. The following result is the dual of Theorem 5.6. Theorem 5.10. Let A be a module in Mod R. (1) If A ∈ ⊥(GFI(Mod R)), then idR A = GFidR A. (2) If A ∈ ⊥(G(I(Mod R))), then idR A = GFidR A = GidR A. (3) If pdR A < ∞, then idR A = GFidR A = GidR A = res ^I(Mod R)- codim A. (4) If idR A < ∞, then idR A = GFidR A = GidR A = (I(Mod R))⊥- codim A = (FI(Mod R))⊥- codim A. (5) If GFidR A < ∞, then GFidR A = GidR A = (I(Mod R))⊥- codim A = (FI(Mod R))⊥- codim A. (6) If GidR A < ∞, then GidR A = (I(Mod R))⊥- codim A. (7) If FP - idR A < ∞, then idR A = GFidR A. Remark 5.11. Theorem 5.10(3) generalizes [H1, Theorem 2.1] which states that for a module A in Mod R, if pdR A < ∞, then idR A = GidR A. Notice that a module in Mod R with finite projective dimension is in ⊥(G(I(Mod R))), so we may also get [H1, Theorem 2.1] by Theorem 5.10(2). Theorem 5.10(6) is well known ([H2, Theorem 2.22]). 5.3. Questions In view of the assertions (3) and (4) in Theorem 5.6, it is natural to ask the following Question 5.12. If A is a module in Mod R with idR A < ∞, does then pdR A = ⊥(P(Mod R))- dim A hold? Question 5.13. If A is a module in Mod R with pdR A < ∞, does then pdR A = 26 cores ^P(Mod R)- dim A hold? by ⊥ Exti From now on, R is a left and right Noetherian ring (unless stated otherwise). We denote RR = {M ∈ mod R Exti R(RM, RR) = 0 for any i ≥ 1} (resp. ⊥RR = {N ∈ mod Rop Rop(NR, RR) = 0 for any i ≥ 1}). For any module A in mod R, there exists a projective presentation: P1 f −→ P0 → A → 0 of A in mod R (note: if R is an artinian algebra, then this projective presentation of A is chosen to be the minimal one). Then we get an exact sequence: 0 → A∗ → P ∗ 0 f ∗ −→ P ∗ 1 → Tr A → 0 in mod Rop, where (−)∗ = Hom(−, R) and Tr A = Coker f ∗ is the transpose of A. Auslander and Bridger generalized the notions of finitely generated projective modules and the projec- tive dimension of finitely generated modules as follows. A module A in mod R is said to have RR and Tr A ∈ ⊥RR ([AB]). It is well known that over Gorenstein dimension zero if A ∈ ⊥ a left and right Noetherian ring, a finitely generated module is Gorenstein projective if and only if it has Gorenstein dimension zero ([EJ2, Proposition 10.2.6]). Let A be a module in mod R. Recall from [HuH] that A is called ∞-torsionfree if Tr A ∈ ⊥RR. We use T (mod R) to denote the subcategory of mod R consisting of ∞- torsionfree modules. The torsionfree dimension of A, denoted by T - dimR A, is defined as inf{n there exists an exact sequence 0 → Xn → · · · → X1 → X0 → A → 0 in mod R with all Xi in T (mod R)}. Set T - dimR A = ∞ if no such n exists. By [AB, Theorem 2.17], a module is in cores ^P(mod R) if and only if it is in T (mod R). So cores ^P(mod R)- dim A = T - dimR A for any module A in mod R. By Example 4.2(5) and Corollary 4.5, we immediately have the following Corollary 5.14. ([HuH, Corollary 3.5]) Let A be a module in mod R with T - dimR A = n. Then there exists an exact sequence 0 → A → B → T → 0 in Mod R with pdR B ≤ n and T ∞-torsionfree. The following result is analogous to Theorem 5.6(3)(4). Theorem 5.15. Let R be a left and right Noetherian ring and A a module in mod R. (1) If idR A < ∞, then pdR A = GpdR A = T - dimR A. (2) If pdR A < ∞, then pdR A = GpdR A = ⊥ RR- dim A. 27 In view of the assertions in Theorem 5.15, it is natural to ask the following questions, which are finitely generated versions of Questions 5.12 and 5.13 respectively. Question 5.16. If A is a module in mod R with idR A < ∞, does then pdR A = ⊥ RR- dim A hold? (equivalently, does then GpdR A = ⊥ RR- dim A hold?) Question 5.17. If A is a module in mod R with pdR A < ∞, does then pdR A = T - dimR A hold? (equivalently, does then GpdR A = T - dimR A hold?) Let R be an artinian algebra and C(R) the center of R, and let J be the Jacobson radical of C(R) and I(C(R)/J) the injective envelope of C(R)/J. Then the Matlis duality D(−) = HomC(R)(−, I(C(R)/J)) between mod R and mod Rop induces a duality between projective (resp. injective) modules in mod R and injective (resp. projective) modules in mod Rop. As a special case of Question 5.16, we propose the following Conjecture 5.18. Let R be an artinian algebra. (1) A module A in mod R is projective if A is injective and A ∈ ⊥ (2) R is selfinjective if D(RR) ∈ ⊥ RR. RR. The generalized Nakayama conjecture (GNC for short) states that over any artinian algebra R, a module A in mod R is projective if Exti R(A ⊕ R, A ⊕ R) = 0 for any i ≥ 1 ([AR]). The strong Nakayama conjecture (SNC for short) states that over any artinian algebra R, for any 0 6= A in mod R there exists an i ≥ 0 such that Exti R(A, R) 6= 0 ([CoF]). These two conjectures remain still open. Observe that an equivalent version of GNC states that over any artinian algebra R, for any simple module S in mod R there exists i ≥ 0 such that Exti R(S, R) 6= 0 ([AR]). So SNC⇒GNC. It is easy to see that GNC⇒Conjecture 5.18(1)⇒Conjecture 5.18(2). The following result shows that Question 5.17 is closely related to SNC. Proposition 5.19. Let R be an artinian algebra. Then the following statements are equivalent. (1) SNC holds for Rop. (2) A module in mod R is projective if A ∈ T (mod R) and pdR A ≤ 1. Proof. (1) ⇒ (2) Let A ∈ T (mod R) and pdR A ≤ 1. Then Tr A ∈ ⊥RR and there exists a minimal projective presentation: 0 → P1 → P0 → A → 0 28 in mod R, which induces an exact sequence: 0 → A∗ → P ∗ 0 → P ∗ 1 → Tr A → 0 in mod Rop. So we get the following commutative diagram with exact rows: 0 0 / (Tr A)∗ P1 ∼= / P ∗∗ 1 P0 ∼= / P ∗∗ 0 A / 0 Thus (Tr A)∗ = 0 and Exti R(Tr A, R) = 0 for any i ≥ 0. Then Tr A = 0 by (1), which implies that A is projective by [ARS, Chapter IV, Proposition 1.7(b)]. (2) ⇒ (1) Let B be a module in mod Rop such that Exti Rop(B, R) = 0 for any i ≥ 0. Then B has no non-zero projective summands. So B ∼= Tr Tr B by [ARS, Chapter IV, Proposition 1.7(c)] and hence Exti Rop(Tr Tr B, R) = 0 for any i ≥ 0. So Tr B ∈ T (mod R). From a minimal projective presentation Q1 → Q0 → B → 0 of B in mod Rop, we get an exact sequence: 0 → B∗ → Q∗ 0 → Q∗ 1 → Tr B → 0 0, Q∗ 1 projective. Because B∗ = 0, pdR Tr B ≤ 1. Then Tr B is projective in mod R with Q∗ by (2), which implies that B is projective. Again because B∗ = 0, B = 0. Therefore SNC holds for Rop. (cid:3) Acknowledgements. This research was partially supported by the Specialized Research Fund for the Doctoral Program of Higher Education (Grant No. 20100091110034), NSFC (Grant No. 11171142), NSF of Jiangsu Province of China (Grant No. BK2010007) and a Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions. The author thanks the referee for the useful suggestions. References [A] M. Auslander, Coherent functors, in: Proc. of the Conf. on Categorial Algebra, La Jolla, 1965, Springer-Verlag, Berlin, 1966, pp.189 -- 231. [AB] M. Auslander and M. Bridger, Stabe Module Theory, Memoirs Amer. Math. Soc. 94, Amer. Math. Soc., Providence, RI, 1969. [AR] M. Auslander and I. Reiten, On a generalized version of the Nakayama conjecture, Proc. Amer. Math. Soc. 52 (1975), 69 -- 74. 29 / / / / / / / / / / [ARS] M. Auslander, I. Reiten and S.O. Smolφ, Representation Theory of Artin Algebras, Corrected reprint of the 1995 original, Cambridge Studies in Adv. Math. 36, Cambridge Univ. Press, Cambridge, 1997. [B1] D. Bennis, Rings over which the class of Gorenstein flat modules is closed under ex- tensions, Comm. Algebra 37 (2009), 855 -- 868. [B2] D. Bennis, A note on Gorenstein flat dimension, Algebra Colloq. 18 (2011), 155 -- 162. [C] L.W. Christensen, Gorenstein Dimension, Lect. Notes in Math. 1747, Springer-Verlag, Berlin, 2000. [CFH] L.W. Christensen, A. Frankild and H. Holm, On Gorenstein projective, injective and flat dimensions -- A functorial description with applications, J. Algebra 302 (2006), 231 -- 279. [CI] L.W. Christensen and S. Iyengar, Gorenstein dimension of modules over homomor- phisms, J. Pure Appl. Algebra 208 (2007), 177 -- 188. [CoF] R.R. Colby and K.R. Fuller, A note on the Nakayama conjectures, Tsukaba J. Math. 14 (1990), 343 -- 352. [DLM] N.Q. Ding, Y.L. Li and L.X. Mao, Strongly Gorenstein flat modules, J. Aust. Math Soc. 86 (2009), 323 -- 338. [EJ1] E.E. Enochs and O.M.G. Jenda, Gorenstein injective and projective modules, Math. Z. 220 (1995), 611 -- 633. [EJ2] E.E. Enochs and O.M.G. Jenda, Relative Homological Algebra, De Gruyter Exp. in Math. 30, Walter de Gruyter, Berlin, New York, 2000. [EJL] E.E. Enochs, O.M.G. Jenda and J. A. L´opez-ramos, Covers and envelopes by V - Gorenstein modules, Comm. Algebra 33 (2005), 4705 -- 4717. [EJT] E.E. Enochs, O.M.G. Jenda and B. Torrecillas, Gorenstein flat modules, J. Nanjing Univ. Math. Biquarterly 10 (1993), 1 -- 9. [GD] Y.X. Geng and N.Q. Ding, W-Gorenstein moduless, J. Algebra 325 (2011), 132 -- 146. [GT] R. Gobel and J. Triifaj, Approximations and Endomorphism Algebras of Modules, De Gruyter Exp. in Math. 41, Walter de Gruyter, Berlin, New York, 2006. 30 [H1] H. Holm, Rings with finite Gorenstein injective dimension, Proc. Amer. Math. Soc. 132 (2003), 1279 -- 1283. [H2] H. Holm, Gorenstein homological dimensions, J. Pure Appl. Algebra 189 (2004), 167 -- 193. [Hu] Z.Y. Huang, Proper resolutions and Gorenstein categories, J. Algebra 393 (2013), 142 -- 169. [HuH] C.H. Huang and Z.Y. Huang, Torsionfree dimension of modules and self-injective dimension of rings, Osaka J. Math. 49 (2012), 21 -- 35. [LHX] Z.F. Liu, Z.Y. Huang and A.M. Xu, Gorenstein projective dimension relative to a semidualizing bimodule, Comm. Algebra 41 (2013), 1 -- 18. [MD] L.X. Mao and N.Q. Ding, Gorenstein FP-injective and Gorenstein flat modules, J. Algebra Appl. 7 (2008), 491 -- 606. [SSW] S. Sather-Wagstaff, T. Sharif and D. White, Stability of Gorenstein categories, J. London Math. Soc. 77 (2008), 481 -- 502. [Z] P. Zhang, Gorensteinness and Buchweitz theorem, Preprint 2012. 31
1012.2844	1	1012	2010-12-13T19:59:26	Hopf-like Algebras and Extended P-B-W Theorems	[ "math.RA" ]	Based on invariant algebras, we introduce representations$^{6-th}$ of Lie algebras and representations$^{< 4-th>}$ of Leibniz algebras, give the extended P-B-W Theorems in the context of the new representations of Lie algebras and Leibniz algebras, and generalize the Hopf-algebra structure on the enveloping algebras of Lie Algebras.	math.RA	math	Hopf-like Algebras and Extended P-B-W Theorems Department of Mathematics, The University of British Columbia Keqin Liu Vancouver, BC, Canada, V6T 1Z2 November, 2010 Abstract Based on invariant algebras, we introduce representations6−th of Lie algebras and representationsh4−thi of Leibniz algebras, give the extended P-B-W The- orems in the context of the new representations of Lie algebras and Leibniz algebras, and generalize the Hopf-algebra structure on the enveloping algebras of Lie Algebras. Throughout, an associative algebra always means an associative algebra hav- ing an identity, a homomorphism from an associative algebra to an associative algebra always preserves the identity, and all vector spaces are vector spaces over a field k. Let V be a vector space, and let End(V ) be the associative algebra of all It is well-known that End(V ) is a Lie linear transformations from V to V . algebra with respect to the following bracket [ , ] [f, g] := f g − gf for f , g ∈ End(V ). (1) This Lie algebra is denoted by gℓ(V ). In this paper, the bracket defined by (1) will be called the ordinary bracket, and a homomorphism from a Lie algebra L to the Lie algebra gℓ(V ) will be called an ordinary representation of the Lie algebra L on V . This paper is concerned with the new ways of representing Lie algebras and Leibniz algebras by using linear transformations. The key idea in this new ways is to replace the ordinary bracket by new brackets and replace the associative algebra End(V ) by a subalgebra of End(V ). Let W be a non-zero subspace of a vector space V , and let EndW (V ) be the set of all linear transformations from V to V which have W as an invariant subspace, i.e., EndW (V ) := { f f ∈ End(V ) and f (W ) ⊆ W }. Clearly, EndW (V ) is a subalgebra of the associative algebra End(V ). Although the way of making End(V ) into a Leibniz algebra has not been found, there are 1 several ways of making the subalgebra EndW (V ) into a Leibniz algebra. Also, there are several new brackets which are different from the ordinary bracket and make EndW (V ) into a Lie algebra. In fact, if we fix a scalar k ∈ k and a linear transformation q satisfying q(W ) = 0 and q(v) − v ∈ W for all v ∈ V , (2) then EndW (V ) becomes a Leibniz algebra with respect to the following 4-th angle bracket h , i4,k: hf, gi4,k = f g − gf + gqf − f gq + kf qg − kqgf for f , g ∈ EndW (V ), and EndW (V ) becomes a Lie algebra with respect to the following 6-th square bracket [ , ]6,k: [f, g]6,k := f g − gf − f gq + gf q + kf qg − kgqf for f , g ∈ EndW (V ). Replacing gℓ(V ) by the Leibniz algebra (EndW (V ), h , i4,k), we get the notion of representationsh4−thi of Leibniz algebras. Replacing gℓ(V ) by the Lie algebra (EndW (V ), [ , ]6,k), we get the notion of representations6−th of Lie algebras. The Poincar´e-Birkhoff-Witt Theorem (P-B-W Theorem) and the Hopf alge- bra structure for the enveloping algebras of Lie algebras are of great importance to the ordinary representations of Lie algebras. To initiate the study of the new representations of Lie algebras and Leibniz algebras, it is natural to study the counterparts of the P-B-W Theorem and the Hopf algebra structure for the enveloping algebras of Lie algebras in the context of the new representations of Lie algebras and Leibniz algebras. The purpose of this paper is to present our results in this study. This paper consists of six sections. In section 1, we introduce the notion of invariant algebras, discuss some basic properties of invariant algebras, and define representations6−th of Lie algebras and representations4−th of Leibniz algebras. In section 2, we introduce free invariant algebras and construct a basis for a free invariant algebras. In section 3, we introduce enveloping6−th algebras of Lie algebras and give the extended6−th P-B-W theorem. In section 4, we introduce bialgebras with σ-counit and Hopf-like6−th algebras. The main results of this section is that the enveloping6−th algebra of a Lie algebra is a Hopf-like6−th algebra. In section 5, we introduce envelopingh4−thi algebras of Leibniz algebras and give the extendedh4−thi P-B-W theorem. In section 6, we introduce Hopf-likeh4−thii algebras with i = 1, 2 and study Hopf-likeh4−thii algebra structures on the envelopingh4−thi algebra of a Leibniz algebra. 1 Invariant Algebras We begin this section by introducing the notion of invariant algebras. 2 Definition 1.1 Let A be an associative algebra with an idempotent q. The set (A, q) := { x, x ∈ A and qxq = qx } (3) is called the (right) invariant algebra induced by the idempotent q. The most important example of invariant algebras is the linear invariant al- gebra over a vector space. Let V be a vector space, let End(V ) be the associative algebra of all linear transformations from V to V , and let I be the identity linear transformation of V . If W is a subspace of V , then an linear transformation q satisfying (2) is an idempotent. An element q of End(V ) satisfying (2) is called a W -idempotent. Definition 1.2 If W is a subspace of a vector space V and q is a W -idempotent, then the invariant algebra (End(V ), q) induced by q is called the linear invari- ant algebra over V induced by the W -idempotent q. Clearly, the linear invariant algebra (End(V ), q) over V induced by a W - idempotent q consists of all linear transformations of V having W as an invariant subspace, i.e., (End(V ), q) = EndW (V ). Definition 1.3 Let (A, qA) and (B, qB ) be invariant algebras. A linear map φ : (A, qA ) → (B, qB ) is called an invariant homomorphism if φ(xy) = φ(x)φ(y) for x, y ∈ (A, qA ) and φ(1A ) = 1B , φ(qA ) = qB , where 1A and 1B are the identities of A and B, respectively. A bijective invariant homomorphism is called an invariant isomorphism. The next proposition shows that any invariant algebra can be imbedded in a linear invariant algebra. Proposition 1.1 If (A, q) is an invariant algebra, then the map φ : a 7→ aL for a ∈ (A, q) is an injective invariant homomorphism from (A, q) to the linear invariant al- gebra (End(A, q), qL ) over (A, q) induced by the (A, q)ann-idempotent qL , where aL is the left multiplication defined by aL (x) := ax for x ∈ (A.q) and (A, q)ann := { qx − x x ∈ (A, q) }. There are six ways of making an invariant algebra into a Lie algebra. One of the six ways is given in the following 3 Proposition 1.2 If q is an idempotent of an associative algebras A, then (A, q) becomes a Lie algebra under the following 6-th square bracket [x, y]6,k := xy − yx − xyq + yxq + kxqy − kyqx, (4) where x, y ∈ (A, q), and k is a fixed scalar in the field k. This Lie algebra is denoted by Lie(cid:16)(A, q), [ , ]6,k(cid:17). Following [2], a vector space L is called a (right) Leibniz algebra if there exists a binary operation h , i: L × L → L such that the (right) Leibniz identity holds: hhx, yi, zi = hx, hy, zii + hhx, zi, yi for x, y, z ∈ L. There are four ways of making an invariant algebra into a Leibniz algebra. One of the four ways is given in the following Proposition 1.3 If q is an idempotent of an associative algebras A, then (A, q) becomes a Leibniz algebra under the following 4-th angle bracket hx, yi4 = xy − yx + yqx − xyq + kxqy − kqyx, (5) where x, y ∈ (A, q), and k is a fixed scalar in the field k. This Leibniz algebra is denoted by Leib(cid:16)(A, q), h , i4,k(cid:17). The Lie algebra Lie(cid:16)(End(V ), q), [ , ]6,k(cid:17) is denoted by gℓ bra Lie(cid:16)(End(V ), q), h , i4,k(cid:17) is denoted by gℓ Let W be a subspace of a vector space V , and let q be a W -idempotent. [ , ]6,k q,W (V ) and is called the 6-th general linear Lie algebra induced by (q, W ). The Leibniz alge- h , i4,k q,W (V ) and is called the 4-th general linear Leibniz algebra induced by (q, W ). We finish this section with the following Definition 1.4 Let W be a subspace of a vector space V over a field k and let q be a W -idempotent. (i) A Lie algebra homomorphism ϕ from a Lie algebra (L, [ , ]) to the 6-th [ , ]6,k q,W (V ) is called a representation6−th of L general linear Lie algebra gℓ on V induced by (q, W ). (ii) A Leibniz algebra homomorphism ϕ from a Leibniz algebra (L, h , i) to the 4-th general linear Leibniz algebra gℓ of L on V induced by (q, W ). h , i4,k q,W (V ) is called a representation4−th 4 2 Free Invariant Algebras Let ( A, q) be an invariant algebra, and let i be a map from a set X to ( A, q). The pair (cid:16)( A, q),i(cid:17) is called the free invariant algebra generated by the set X if the following universal property holds: given any invariant algebra (A, q) and any map θ : X → (A, q) there exists a unique invariant homomorphism θ : ( A, q) → (A, q) such that θi = θ; that is, the following digram is commutative ( A, q) ix X = θ−→ (A, q) θ x X The free invariant algebra generated by a set is clearly unique. We now construct free invariant algebras over a field k. Let X := { xj j ∈ J } be a set, and let q be a symbol which is not an element of X. Let T (V ) be the tensor algebra based on a vector space V , where V =Mj∈J kxjM kq is the vector space over k with a basis XS{q}. Let I be the ideal of T (V ) generated by { q ⊗ q − q, q ⊗ a ⊗ q − q ⊗ a a ∈ T (V ) }. Let A := T (V ) I and q := q + I. Then ( A, q) is an invariant algebra. Proposition 2.1 The pair (cid:16)( A, q),i(cid:17) is the free invariant algebra generated by the set X = { xj j ∈ J }, where the map i : X →(cid:16)( A, q),i(cid:17) is defined by i(xj ) := xj + I for j ∈ J. Let xj := i(xj) = xj + I for j ∈ J. The product of two elements a and b of ( A, q) is simply denoted by ab. The next proposition gives a basic property of the free invariant algebra(cid:16)( A, q), i(cid:17) generated by the set X. Proposition 2.2 The following subset of (cid:16)( A, q),i(cid:17) S :=(cid:26) 1, q, xj1 · · · xjm , xj1 · · · xjt q xjt+1 · · · xjm (cid:12)(cid:12)(cid:12)(cid:12) is a k-basis of the vector space ( A, q). { xj1 , · · · , xjm } ⊆ X, m ∈ Z≥1, m ≥ t ≥ 0 (cid:27) 5 3 Enveloping6−th Algebras of Lie Algebras In the remaining part of this paper, we assume that the scalar k is non-zero. Using invariant algebras and invariant homomorphisms in section 1, we in- troduce enveloping6−th algebras of Lie algebras in the following Definition 3.1 Let (L, [ , ]) be a Lie algebra (arbitrary dimensionality and char- acteristic). By a enveloping6−th algebra of (L, [ , ]) we will understand a i : L → (U , ¯q) satisfying the following two conditions: pair (cid:16)(U , ¯q), i(cid:17) composed of an invariant algebra (U , ¯q) together with a map (i) the map i : (L, [ , ]) → Lie(cid:16)(U , ¯q), [ , ]6,k(cid:17) is a Lie algebra homomorphism; that is, i is linear and i([x, y]) = [i(x), i(y)]6,k = i(x)i(y) − i(y)i(x) + −i(x)i(y)¯q + i(y)i(x)¯q + ki(x)¯qi(y) − ki(y)¯qi(x) for x, y ∈ L, (ii) given any invariant algebra (A, q) and any Lie algebra homomorphism f : (L, [ , ]) → Lie(cid:16)(A, q), [ , ]6,k(cid:17) there exists a unique invariant homomor- phism f ′ : (U , ¯q) → (A, q) such that f = f ′i; that is, the following diagram is commutative: (U , ¯q) ix L = f ′ −→ (A, q) f x L. Clearly, the enveloping6−th algebra of a Lie algebra L, which is also denoted by (U 6−th k (L), ¯q), is unique up to an invariant isomorphism. We now construct the enveloping6−th algebra of a Lie algebra (L, [ , ]) over a field k. Let X = { xj j ∈ J } be a basis of (L, [ , ]). Let (cid:16)( A, q),i(cid:17) be the free invariant algebra generated by the set X. By Proposition 2.2, X := { xj := i(xj ) j ∈ J } is a linearly independent subset of ( A, q). Hence, i can be extended to an injective linear map i : L → ( A, q). Let x := i(x) for all x ∈ L. Let R be the ideal of ( A, q) which is generated by all the elements of the form where x, y ∈ L := i(L). Let d[x, y] − xy + y x + xy q − y xq − kxq y + ky qx, U := A R , ¯q := q + R. 6 (6) Define a map i : L → (U , ¯q) by i(x) := x + R for x ∈ L. (7) We have the following Proposition 3.1 The pair (cid:16)(U , ¯q), i(cid:17) defined by (6) and (7) enveloping6−th algebra for the Lie algebra (L, [ , ]) is an The following proposition gives the counterpart of P-B-W Theorem in the context of representations6−th of Lie algebras. Proposition 3.2 (The Extended6−th P-B-W Theorem) Let L be a Lie algebra with a basis X := { xj j ∈ J }. If the set J of indices is ordered, then the following set of cosets   (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) T := qxj1 · · · xjm + R, xi1 · · · xit xj0 qxj1 · · · xjm + R, xj1 · · · xjm + R xi1 , · · · , xit , xj0 , xj1 , · · · , xjm ∈ X, it ≥ · · · ≥ i1, jm ≥ · · · ≥ j1 ≥ j0, t, m ∈ Z≥0 is a basis for the enveloping6−th algebra (U , ¯q)) of the Lie algebra L. .   4 Hopf-like6−th Algebras Let C be a vector space over a field k. The canonical map c 7→ 1 ⊗ c is denoted by C → k ⊗ C, and the canonical map c 7→ c ⊗ 1 is denoted by C → C ⊗ k respectively, where c ∈ C. If V and W are vector spaces over a field k, then the twist map τ : V ⊗ W → W ⊗ V is defined by τ (v ⊗ w) := w ⊗ v, where v ∈ V and w ∈ W . First, we introduce the concept of a bialgebra with σ-counit. Definition 4.1 An associative algebra H over a field k is called a bialgebra with σ-counit if there exist five linear maps m : H ⊗ H → H, u : k → H, ∆ : H → H ⊗ H, ε : H → k and σ : H → H such that the following eight diagrams are commutative: • associativity H ⊗ H ⊗ H m⊗id−→ H ⊗ H id ⊗ my H ⊗ H m−→ 7 m y H • unit • coassociativity • σ-counit u⊗id−→ H ⊗ H id⊗u←− H ⊗ k y H id= id= ∆−→ H ⊗ H id ⊗ ∆ y H ⊗ H ∆⊗id−→ H ⊗ H ⊗ H k ⊗ H ←− H −→ H ⊗ k id ⊗ ε x ∆←− H ∆−→ H ⊗ H k ⊗ H y H H ∆y ε ⊗ idx H ⊗ H m y H σ x • ∆ is an algebra homomorphism H ⊗ H ∆ ⊗ ∆y H ⊗ H ⊗ H ⊗ H m−→ H ∆−→ id⊗τ ⊗id−→ H ⊗ H m ⊗ m x H ⊗ H ⊗ H ⊗ H ∆−→ H ⊗ H u ⊗ u x k −→ k ⊗ k H ux • ε is an algebra homomorphism H ⊗ H ε⊗ε−→ k ⊗ k H = H my H ux tively. A bialgebra H with σ-counit is also denoted by H(cid:0)m,u,∆ y ε,σ (cid:1) if the five linear The maps ∆ and ε are called the comultiplication and the σ-counit respec- maps have to be indicated explicitly. ε−→ k k = k y ε Based on bialgebras with σ-counit, we now introduce Hopf-like algebras in the following 8 Definition 4.2 A bialgebra H = H(cid:0)m,u,∆ ε,σ (cid:1) with σ-counit over a field k is called a Hopf-like6−th algebra if there exists a linear map S : H → H such that the following diagram is commutative: ∆−→ H ⊗ H S⊗id−→ H ⊗ H S−→ H ε−→ k ∆−→ H ⊗ H id⊗S−→ H ⊗ H H H (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) H m−→ H u−→ H (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) m−→ H (8) The linear map S satisfying (8) is called the antipode6−th-like of the Hopf- like6−th algebra. The first generalization of the Hopf algebra structure for the ordinary en- veloping algebra of a Lie algebra is given in the following Proposition 4.1 The enveloping6−th algebra U 6−th (L) of a Lie algebra L is a Hopf-like6−th algebra, where the comultiplication ∆, the σ-counit ε and the antipode6−th-like S satisfy k ∆(q) = q ⊗ q, ∆(x) = (x + kqx − xq) ⊗ 1 + 1 ⊗ (x + kqx − xq) + +(1 − k)qx ⊗ q + (1 − k)q ⊗ qx for x ∈ L, ε(q) = 1, ε(x) = 0 for x ∈ L, S(q) = 1 − q, S(x) = − 1 k x − kqx + 1 k xq for x ∈ L and the linear map σ : U 6−th k (L) → U 6−th k (L) is defined by σ(a) := a + qa − aq for a ∈ U 6−th k (L). 5 Envelopingh4−thi algebras of Leibniz algebras The concept of the envelopingh4−thi algebra of a Leibniz Algebra is introduced in the following Definition 5.1 Let (L, h, i) be a Leibniz algebra (arbitrary dimensionality and characteristic). By a envelopingh4−thi algebra of (L, h, i) we will understand a pair (cid:16)(U , ¯q), i(cid:17) composed of an invariant algebra (U , ¯q) together with a map i : L → (U , ¯q) satisfying the following two conditions: 9 (i) the map i : (L, h, i) → Leib(cid:16)(U , ¯q), h, i4,k(cid:17) is a Leibniz algebra homomor- phism; that is, i is linear and i(hx, yi) = hi(x), i(y)i4,k = i(x)i(y) − i(y)i(x) + +i(y)¯qi(x) − i(x)i(y)¯q + ki(x)¯qi(y) − k ¯qi(x)i(y) for x, y ∈ L, (ii) given any invariant algebra (A, q) and any Leibniz algebra homomorphism f : (L, h, i) → Leib(cid:16)(A, q), h, i4,k(cid:17) there exists a unique invariant homo- morphism f ′ : (U , ¯q) → (A, q) such that f = f ′i; that is, the following diagram is commutative: (U , ¯q) ix L = f ′ −→ (A, q) f x L. Clearly, the envelopingh4−thi algebra of a Leibniz algebra L, which is also denoted by (U h4−thi k (L), ¯q), is unique up to an invariant isomorphism. We now construct the enveloping invariant algebra of a Leibniz algebra (L, h, i) over a field k. Let X = { xj j ∈ J } be a basis of (L, h, i). Let (cid:16)( A, q),i(cid:17) be the free invariant algebra generated by the set X. By Propo- sition 2.2, X := { xj := i(xj ) j ∈ J } is a linearly independent subset of ( A, q). Hence, i can be extended to an injective linear map i : L → ( A, q). Let x := i(x) for all x ∈ L. Let R be the ideal of ( A, q) which is generated by all the elements of the form where x, y ∈ L and we identify L with i(L). Let dhx, yi − xy + y x − y q x + xy q − kxq y + k q y x, U := A R , ¯q := q + R. Define a map i : L → (U , ¯q) by i(x) := x + R for x ∈ L. (9) (10) Proposition 5.1 The pair(cid:16)(U , ¯q), i(cid:17) defined by (9) and (10) is the envelopingh4−thi algebra of the Leibniz algebra (L, h, i) The following proposition gives the counterpart of P-B-W Theorem in the context of representationsh4−thi of Leibniz algebras. 10 Proposition 5.2 (The Extendedh4−thi P-B-W Theorem) Let L be a Leib- niz algebra with a basis X := { xj j ∈ J }. If the set J of indices is ordered, then the following set of cosets of model monomials T :=  qxj1 · · · xjm + R, xj qxj1 · · · xjm + R, xj1 · · · xjm + R (cid:12)(cid:12)(cid:12)(cid:12) { xj, xj1 , · · · , xjm } ⊆ X, jm ≥ · · · ≥ j1 and m ∈ Z≥0 is a basis for the envelopingh4−thi algebra of the Leibniz algebra L.   6 Hopf-likeh4−thi Algebras We begin this section by introducing two generalizations of Hopf algebras. Definition 6.1 Let H = (H, m, u, ∆, ε) be a bialgebra, where (H, m, u) gives the associative algebra structure of H, and (H, ∆, ε) gives the coalgebra structure of H. (i) H is called a Hopf-likeh4−thi1 algebra if there exists a linear map S : H → H such that the following diagram is commutative: ∆−→ H ⊗ H S⊗id−→ H ⊗ H S−→ H ε−→ k ∆−→ H ⊗ H id⊗S−→ H ⊗ H H H (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) H u−→ H (11) m−→ H (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) m−→ H The linear map S satisfying (11) is called the antipodeh4−thi1-like of the Hopf-likeh4−thi1 algebra H. (ii) H is called a Hopf-likeh4−thi2 algebra if there exist an algebra homomor- phism σ : H → H and a linear map S : H → H such that the following diagram is commutative: H H (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) H ∆−→ H ⊗ H S⊗id−→ H ⊗ H m−→ H σ−→ H S−→ H ε−→ k ∆ −→ H ⊗ H id⊗S −→ H ⊗ H u−→ m −→ H (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) H (12) The linear map S satisfying (12) is called the antipodeh4−thi2-like of the Hopf-likeh4−thi2 algebra H. 11 Clearly, a Hopf algebra is both a Hopf-likeh4−thi1 algebra and a Hopf-likeh4−thi2 algebra with σ = id. The other two generalizations of the Hopf algebra structure for the ordinary enveloping algebra of a Lie algebra is given in the following Proposition 6.1 Let (U h4−thi niz algebra L. k (L), q) be the envelopingh4−thi algebra of a Leib- (i) If k = 1, then (U h4−thi 1 (L), q) is a Hopf-likeh4−thi1 algebra; (ii) If k 6= 0, then (U h4−thi k (L), q) is a Hopf-likeh4−thi2 algebra, where the comultiplication ∆, the counit ε, the antipode-likeh4−thii S with i = 1, 2 satisfy ∆(q) = q ⊗ q, ∆(x) = (x + kqx − xq) ⊗ 1 + 1 ⊗ (x + kqx − xq) + +(−kqx + xq) ⊗ q + q ⊗ (−kqx + xq) for x ∈ L, ε(q) = 1, ε(x) = 0 for x ∈ L, S(q) = 1 − q, S(x) = − k and the algebra homomorphism σ is defined by 1 k x +(cid:16) 1 − k(cid:17)xq for x ∈ L. σ(a) := a + qa − aq for a ∈ U h4−thi k (L). References [1] Sorin Dascalescu, Constantin Nastasescu and Serban Raianu, Hopf Alge- bras, an introduction, Monographs and textbooks in pure and applied math- ematics; 235, Marcel Dekker (2001). [2] Jean-Louis Loday, Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz, Enseign. Math. (2) 39 (1993), no. 3-4, 269 -- 293. [3] Jean-Louis Loday and Teimuraz Pirashvili, Universal enveloping algebras of Leibniz algebras and (co)homology, Math. Ann. 296 (1993), 139-158. [4] Susan Montgomery, Hopf Algebras and Their Actions on Rings, American Mathematical Society, 1993. [5] Moss E. Sweedler, Hopf Algebras, Benjamin, New York, 1969. 12
1503.07022	1	1503	2015-03-24T13:15:06	Co-Moufang deformations of the universal enveloping algebra of the algebra of traceless octonions	[ "math.RA" ]	By means of graphical calculus we prove that, over fields of characteristic zero, any bialgebra deformation of the universal enveloping algebra of the algebra of traceless octonions satisfying the dual of the left and right Moufang identities must be coassociative and cocommutative.	math.RA	math	CO-MOUFANG DEFORMATIONS OF THE UNIVERSAL ENVELOPING ALGEBRA OF THE ALGEBRA OF TRACELESS OCTONIONS JOS´E M. P´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV Abstract. By means of graphical calculus we prove that, over fields of char- acteristic zero, any bialgebra deformation of the universal enveloping algebra of the algebra of traceless octonions satisfying the dual of the left and right Moufang identities must be coassociative and cocommutative. 1. Introduction Throughout this paper the base field k will be assumed to be of characteristic zero The only spheres that are H-spaces are S0, S1, S3 and S7. While S0, S1 and S3 are Lie groups, the product on S7, inherited from the octonions, is no longer associative but satisfies the left, middle and right Moufang identities a(x(ay)) = ((ax)a)y, (ax)(ya) = (a(xy))a and ((xa)y)a = x(a(ya)). A Moufang loop is a set (M, xy) with a binary product xy such that: 1) it satisfies any of the Moufang identities 2) the left and right multiplication operators by any x ∈ M , Lx : y 7→ xy and Rx : y 7→ yx, are bijective and 3) M contains a unit element for the product xy. The tangent space m at the unit element of any analytic Moufang loop M can be endowed with a skew-symmetric product [a, b] that satisfies a generalization of the Jacobi identity: J(a, b, [a, c]) = [J(a, b, c), a] where J(a, b, c) := [[a, b], c] + [[b, c], a] + [[c, a], b] is the Jacobian of a, b and c. Such algebraic structures (m, [a, b]) are called Malcev algebras [18, 28] and they locally classify analytic Moufang loops up to isomorphism. The proof of this connection between Moufang loops and Malcev algebras initiated the development of a non- associative Lie correspondence finally achieved by Mikheev and Sabinin in 1987 [27] with techniques of differential geometry. The tangent space of S7 at the unit element is the Malcev algebra of traceless octonions. Given any not necessarily associative algebra (A, xy), the generalized alternative nucleus of A is Nalt(A) := {a ∈ A (a, x, y) = −(x, a, y) = (x, y, a) ∀x,y∈A}. 2010 Mathematics Subject Classification. 17D10, 17B37. Key words and phrases. Malcev algebras and Quantized enveloping algebras and Deformations. The first author thanks support from the Spanish Ministry of Science and Innovation (MTM2010-18370-C04-03) and from FAPESP (2011/51553-4). He also thanks the Instituto de Matem´atica e Estat´ıstica/USP of Sao Paulo for its kind hospitality during his stay in 2011. The second author thanks support from CNPq, Proc. 456698/2014-0 and FAPESP, Proc. 2014/09310- 5. Both authors also thank Bodo Pareigis for sharing his beautiful "diagrams" package. 1 2 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV where (x, y, z) := (xy)z − x(yz) stands for the associator of x, y and z. The generalized alternative nucleus is always closed under the commutator product [x, y] := xy − yx and it becomes a Malcev algebra with this product. Alternative algebras are those algebras A such that, like the octonions, satisfy Nalt(A) = A, so they provide natural examples of Malcev algebras. Unfortunately, it remains an open problem whether or not any Malcev algebra is a subalgebra of (A, [x, y]) for some alternative algebra (A, xy). In [24] it was proved that any Malcev algebra appears as a Malcev subalgebra of the generalized alternative nucleus of some alge- bra. This result is similar to the corresponding result for Lie algebras, namely, the Poincar´e-Birkhoff-Witt Theorem: any Lie algebra appears as a Lie subalgebra of an associative algebra with the commutator product. As in the case of Lie algebra, the result is based on the explicit construction of a universal enveloping algebra U (m) for the Malcev algebra (m, [a, b]). In the case that (m, [a, b]) is a Lie algebra then U (m) turns out to be the usual associative universal enveloping algebra of m. The universal enveloping algebras of Malcev algebras are quite close to Hopf algebras. U (m) admits a cocommutative and coassociative coalgebra structure (U (m), ∆, ǫ) with the additional property that the comultiplication, or coproduct, ∆ and the counit ǫ are homomorphisms of algebras. The Malcev algebra m is recov- ered as the space of primitive elements, i.e., elements a such that ∆(a) = a⊗1+1⊗a. However, in general, U (m) is non-associative but it satisfies some Hopf version of the Moufang identities, the left, middle and right Hopf-Moufang identities: X a(1)(x(a(2)y)) = X((a(1)x)a(2))y, X(a(1)x)(ya(2)) = X(a(1)(xy))a(2) and X((xa(1))y)a(2) = X x(a(1)(ya(2))). The study of U (m) led in [24] to a generalization for Malcev algebras of the well- known Ado-Iwasawa Theorem and the Chevalley and Konstant Theorems [25, 29]. Explicit formulas for the product of U (m) in some low-dimensional cases are also known [2 -- 5]. Another approach to the construction of the product of U (m) was carried out in [1]. This approach unified the connections between groups with triality and Moufang loops [8, 10, 13, 20], and Lie algebras with triality and Malcev algebras [12,19] by means of the notion of Hopf algebra with triality. For an account of universal enveloping algebras of generalizations of Lie and Malcev algebras see [21 -- 23]. In [17] a study of the possible coassociative bialgebra deformations of U (m) was carried out for the Malcev algebra of traceless octonions m = M(α, β, γ) [30]. The reason for focusing on these algebras is that any central simple Malcev algebra is either a Lie algebra or isomorphic to an algebra M(α, β, γ). Bialgebra deformations of U (m) should be an analogue of quantized enveloping algebras of Lie algebras in a non-associative setting. The development of quantized enveloping algebras and quantum groups during the last two decades has been spectacular so the search for bialgebra deformations of enveloping algebras of Malcev is challenging. Given a Malcev algebra (m, [a, b]) over a field k, a bialgebra deformation of U (m) over the ring of formal power series K = k[[h]] is a topologically free K-module Uh(m) endowed with four maps of K-modules ph : Uh(m) ⊗Uh(m) → Uh(m), (product) (coproduct) ∆h : Uh(m) → Uh(m) ⊗Uh(m), ιh : K → Uh(m), 1 7→ 1, ǫh : Uh(m) → K, (unit) (counit) CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 3 where ⊗ stands for the completed tensor product in the h-adic topology, such that (1) (Uh(m), ιh, ph, ǫh, ∆h) satisfies the axioms of bialgebra over the commuta- tive ring K (see Remarks to Definition 4.1.3 in [7]) but with the algebraic tensor product replaced by its completion and without assuming the coas- sociativity, (2) Uh(m)/hUh(m) ∼= U (m) as a k-vector space and, with this identification, (3) ph ≡ p (mod h) and ∆h ≡ ∆ (mod h) with p and ∆ the multiplication and comultiplication of U (m) respectively. Since Uh(m) is topologically free and Uh(m)/hUh(m) ∼= U (m), we can identify Uh(m) with U (m)[[h]] as a K-module and we will do so. The unit and counit are assumed to be the natural extensions to U (m)[[h]] of the unit and counit of U (m). The null defor- mation of U (m) is obtained by extending K-linearly the structure maps of U (m). Trivial deformations are those that are isomorphic to the null deformation under a K-linear bialgebra isomorphism which is the identity modulo h. To avoid confusion in this context where two multiplications, p and ph, and two comultiplications, ∆ and ∆h, appear we will stick to the following notation: xy := p(x ⊗ y), x • y := ph(x ⊗y) X x(1) ⊗ x(2) := ∆(x) for all x ∈ U (m), and X x[1] ⊗x[2] := ∆h(x) for all x ∈ Uh(m). Bialgebra deformations of U (m) are rather general objects so some extra restric- tions are imposed either on the multiplication or on the comultiplication. Quan- tized universal enveloping algebras of Lie algebras are assumed to be associative and coassociative [7], i.e., the associativity and coassociativity are properties that one wants to preserve when deforming these structures. However, U (m) is no longer associative so, it is more appropriate to preserve the Hopf-Moufang identities. In [17] it was proved that, in contrast to the Lie case, any coassociative but possibly non-cocommutative bialgebra deformation of U (M(α, β, γ)) satisfying the Hopf-Moufang identities is trivial. The coassociativity allowed an approach to the study of bialgebra deformations similar to that carried out for Lie algebras. How- ever, since the product of U (m) is not associative this restriction on the coalgebra structure seems artificial. According to the self-dual structure of Hopf algebras, the dual version of the Hopf-Moufang identities, that we will call co-Moufang identi- ties, is probably the natural restriction on the coalgebra structure of any bialgebra deformation of U (m). Co-Moufang identities are satisfied, for instance, by the co- ordinate algebra k[S7] of the seven-dimensional sphere where they have been used to develop differential geometry [15]. Unfortunately, when dealing with non-coassociative coalgebras Sweedler's nota- tion for the comultiplication turns out to be very obscure and cumbersome. Graph- ical calculus is preferable in this case: x y ✝✆:= xy, x ✞☎:= ∆(x), x r := ǫ(x) x y := x ⊗ y 7→ y ⊗ x, and := x x Compositions of maps are written from top to bottom. For instance, the equalities (1.1) X ǫ(x(1))x(2) = x =X ǫ(x(2))x(1) are encoded in the following equalities of diagrams (1.2) x = x ✞☎ r = x ✞☎ r 4 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV The left and right Hopf-Moufang identities are represented by x y z ✞☎ x y z ✞☎ x y z ✞☎ x y z ✞☎ ✝✆ = ✝✆ and ✝✆ = ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ The left and right co-Moufang identities are obtained by turning upside-down the diagrams corresponding to the left and right Moufang identities: ✞☎ ✞☎ ✞☎ ✞☎ ✞☎ ✞☎ ✞☎ ✞☎ (left co-Moufang) ✞☎ = ✞☎ (right co-Moufang) ✞☎ = ✞☎ ✝✆ ✝✆ ✝✆ ✝✆ In other words [15], and (1.3) X x(1)x(2)(2)(1) ⊗ x(2)(1) ⊗ x(2)(2)(2) =X x(1)(1)(1)x(1)(2) ⊗ x(1)(1)(2) ⊗ x(2) (1.4) X x(1) ⊗ x(2)(2)(1) ⊗ x(2)(1)x(2)(2)(2) =X x(1)(1)(1) ⊗ x(1)(2) ⊗ x(1)(1)(2)x(2). In this paper we will prove that any bialgebra deformation Uh(m) of U (m), where m is a finite-dimensional central simple Malcev algebra, satisfying the left and right co-Moufang identities is coassociative. In the case that m = M(α, β, γ) these deformations are also cocommutative. Hence, any bialgebra deformation of U (M(α, β, γ)) satisfying the left and right Moufang and co-Moufang identities is trivial. This result reveals an extraordinary rigidity of the universal enveloping algebra of non-Lie tangent algebras. 2. Cocommutativity Some consequences of the co-Moufang identities will be frequently used in our arguments, where we will always assume that m is a Malcev algebra. We collect them in the following lemma. Lemma 2.1. Any bialgebra deformation of U (m) satisfying the left and right co- Moufang identities also satisfies 1) ✞☎ ✞☎ = ✞☎ ✝✆ ✞☎ ✝✆ ✞☎ ✞☎ 4) ✞☎ = ✝✆ ✞☎ ✝✆ 2) 5) ✞☎ ✞☎ ✞☎ = ✝✆ ✞☎ ✝✆ ✞☎ ✞☎ 3) ✞☎ = ✞☎ ✝✆ ✝✆ ✞☎ ✞☎ = ✞☎ ☛ ✟ ✞☎ ☛ ✟ ✞☎ ✞☎ 6) ✞☎ ✞☎ = ✞☎ ✞☎ ✝✆ ✝✆ ✝✆ ✝✆ CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 5 Proof. By evaluating Id ⊗ǫ ⊗ Id on both sides of the left co-Moufang identity and using the property (1.2) of the counit we get 1): ✞☎ ✞☎ − ✞☎ ✝✆r ✞☎ ✞☎ ✞☎ = − ✞☎ ✝✆ ✞☎ ✝✆ ✞☎ ✞☎ ✝✆r The proof of equalities 2), 3) and 4) is similar. Equality 5) is a consequence of 3) and the left co-Moufang identity. Identity 6) is a consequence of 4) and the right co-Moufang identity. (cid:3) (cid:3) All the results can be established without diagrams, and we will do so sometimes to compare both approaches. However, it will become more and more apparent that diagrams are a valuable notation for dealing with non-coassociative bialgebras. For instance, part 1) in Lemma 2.1 could be rewritten as and, as we have seen, it is derived from the left co-Moufang identity by X x[1] • x[2][1] ⊗x[2][2] =X x[1][1] • x[1][2] ⊗x[2] X x[1] • x[2][1] ⊗1 ⊗x[2][2] = X x[1] • x[2][2][1] ⊗ǫ(x[2][1])1 ⊗x[2][2][2] = X x[1][1][1] • x[1][2] ⊗ǫ(x[1][1][2])1 ⊗x[2] = X x[1][1] • x[1][2] ⊗1 ⊗x[2]. then Q is semisimple with eigenvalues {2n n ≥ 0} -- notice that there is no am- biguity in the notation an when a ∈ m since ka is an abelian Lie subalgebra of m, so the subalgebra generated by a in U (m) is isomorphic to the associative and commutative algebra U (ka). The primitive elements m form the subspace spanned by the eigenvectors of eigenvalue 2. When more intensive calculus is needed this approach is less useful. The following lemma is well-known and its proof is obvious. Lemma 2.2. Let (Ah, •) and (Bh, •) be topologically free algebras, that we iden- tify with A[[h]] and B[[h]] as k[[h]]-modules for some k-algebras A and B, and ϕn, ψn : A → B be linear maps (n ≥ 0) such that the maps ϕh = Pn≥0 ϕnhn, ψh =Pn≥0 ψnhn satisfy ϕh(x • y) = ϕh(x) • ϕh(y) and ψh(x • y) = ψh(x) • ψh(y) for all x, y ∈ Ah. If ϕi = ψi i = 0, . . . , n − 1 then (ϕn − ψn)(xy) = (ϕn − ψn)(x)ϕ0(y) + ψ0(x)(ϕn − ψn)(y) for al x, y ∈ A. ϕnm = ψnm then ϕn = ψn. In particular, if Ah = Uh(m), ϕi = ψi i = 0, . . . , n − 1 and The map Q : U (m) → U (m) given by will play an important role. Since U (m) is spanned by {an a ∈ m, n ≥ 0} [24] and Q = ✞☎ ✝✆ Q(x) :=X x(1)x(2) Q(an) =X(cid:18)n i(cid:19)aian−i = 2nan 6 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV The comultiplication ∆hU(m) and the multiplication phU(m) can be developed as a power series on h ∆hU(m) = ∆0 + h∆1 + h2∆2 + · · · , phU(m) = p0 + hp1 + h2p2 + · · · where pi : U (m) ⊗ U (m) → U (m) and ∆i : U (m) → U (m) ⊗ U (m) are linear maps and ∆0 = ∆, p0 = p are the comultiplication and the multiplication on U (m). The following notation will be very helpful: 0✞☎:= ∆0, +✞☎:= h∆1 + h2∆2 + · · · so, when restricted to U (m), (2.1) ✞☎= 0✞☎++✞☎. We will use a similar notation for phU(m) too. Theorem 2.3. Any bialgebra deformation of U (M(α, β, γ)) satisfying the left and right co-Moufang identities is cocommutative. Proof. We will prove by induction that ∆n = ∆op then ∆0 = ∆op we have 0 . Let us assume that ∆1 = ∆op n . Since U (m) is cocommutative n−1. By Lemma 2.1 1 , . . . , ∆n−1 = ∆op ✞☎ ✞☎ ✞☎ ✞☎ ✞☎ ✞☎ ✞☎ 0✞☎ +✞☎ D := − ✞☎ ✝✆ = ✞☎ − ✝✆ = ✞☎ ✝✆ ✞☎ ✝✆ ✞☎ − ✝✆ ✞☎ = ✝✆ ✞☎ − ✝✆ ✞☎ ✞☎ − ✝✆ ✝✆ Since ∆i = ∆op i i = 0, . . . , n − 1 then, modulo hn+1, ✞☎ 0✞☎ +✞☎ ✞☎ 0✞☎ 0✞☎ ✞☎ D ≡ By Lemma 2.1 ✞☎ − ✝✆ ✞☎ − ✝✆ = ✞☎ ✝✆ ✞☎ − ✝✆ ✞☎ + ✝✆ ✞☎ ✝✆ − ✞☎ ✝✆ ✞☎ 0✞☎ 0✞☎ ✞☎ ✞☎ 0✞☎ 0✞☎ 0✞☎ +✞☎ D ≡ ✞☎ − ✝✆ ✞☎ + ✝✆ ✞☎ ✝✆ − ✞☎ = ✝✆ ✞☎ − ✝✆ ✞☎ + ✝✆ ✞☎ ✝✆ − ✞☎ − ✞☎ ✝✆ ✝✆ Since ∆i = ∆op i i = 0, . . . , n − 1 then, modulo hn+1, ✞☎ 0✞☎ 0✞☎ 0✞☎ +✞☎ 0✞☎ 0✞☎ 0✞☎ 0✞☎ D ≡ ✞☎ − ✝✆ ✞☎ + ✝✆ ✞☎ ✝✆ − ✞☎ − ✝✆ ✞☎ = ✝✆ ✞☎ − ✝✆ ✞☎ + ✝✆ ✞☎ ✝✆ ✞☎ − ✝✆ CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 7 Any a ∈ m is primitive, i.e., ∆0(a) = a ⊗ 1 + 1 ⊗ a thus a 0✞☎ ✞☎ = ✝✆ a 0✞☎ ✞☎ = ✝✆ a ✞☎+ a ⊗ 1, a ✞☎+ a ⊗ 1 a ✞☎(cid:19) (mod hn+1) a 0✞☎ = ✞☎ ✝✆ a 0✞☎ ✞☎ = ✝✆ a ✞☎ − ✞☎ ✝✆ a ✞☎ − 0✞☎ 0✝✆ a ✞☎+ a ⊗ 1, a ✞☎+ a ⊗ 1, a ✞☎ ✞☎ ✝✆ a ✞☎ and Since ∆i = ∆op i ≡ 2(cid:18) a ✞☎− i = 0, . . . , n − 1 we finally get ≡ 2(cid:18) a ✞☎− a ✞☎(cid:19) (mod hn+1) 0✞☎ 0✝✆ n )(a) is an eigenvector of Q of eigenvalue 2 so This shows that (∆n − ∆op for any a ∈ m. The result follows from [17]. (cid:3) (cid:3) (∆n − ∆op n )(a) ∈ m ∧ m 3. Coassociativity Let Uh(m) be a bialgebra deformation of U (m) satisfying the left and right co- Moufang identities. The coassociator Ch := (∆h ⊗ Id)∆h − (Id ⊗∆h)∆h of Uh(m) measures the lack of coassociativity. It can be developed into a power series ChU(m) = hC1 + h2C2 + · · · . The constant term vanishes due to the coassociativity of U (m). We will prove that Cn = 0 using induction on n. Lemma 3.1. Assume that C1 = · · · = Cn−1 = 0. Then for any a ∈ m a ✞☎ a ✞☎ ✞☎✞☎ ✝✆ ≡ ✞☎ − ✞☎ ✝✆ a ✞☎ − ✞☎ a ✞☎ ✞☎ + a ✞☎ − ✞☎ a ✞☎ ✞☎ (mod hn+1) Proof. By Lemma 2.1, ✞☎ ✞☎ ☛ ✟ ✞☎ ☛ ✟ ✞☎ D := ✞☎ − ✞☎ ✝✆ ✞☎✞☎ ✝✆ = ✞☎ ✞☎ ✝✆ − ✞☎✞☎ ✝✆ = − ✞☎ ✝✆✞☎ ✞☎ ✞☎ ✝✆ 8 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV Decomposing the first summand and using ∆0 = ∆op 0 we get D = ☛ ✟ ☛ ✟ ✞☎ ☛ ✟ ☛ ✟ ✞☎ ✞☎ +✞☎ ✝✆ ✞☎ + − 0✞☎ ✝✆ ✞☎ = ✝✆✞☎ ✞☎ +✞☎ ✝✆ ☛ ✟ ☛ ✟ ☛ ✟ ✞☎ + ✞☎ 0✞☎ − ✞☎ ✝✆✞☎ ✝✆ = ✞☎ +✞☎ ✝✆ − ✞☎ +✞☎ + ✞☎ ✞☎ − ✞☎ ✝✆✞☎ ✝✆ ✝✆ Lemma 2.1 implies that ☛ ✟ ☛ ✟ ✞☎ ✞☎ D = ✞☎ +✞☎ ✝✆ − ✞☎ +✞☎ + ✞☎ − ✞☎ ✞☎ ✝✆✞☎ ☛ ✟ ✝✆ ☛ ✟ ✝✆ 0✞☎ +✞☎ ✞☎ = ✞☎ +✞☎ ✝✆ − ✞☎ +✞☎ + ✞☎ + ✞☎ ✞☎ − ✞☎ ✞☎ ✝✆✞☎ ✝✆ ✝✆ ✝✆ Since C1 = · · · = Cn−1 = 0 then, modulo hn+1, ☛ ✟ ☛ ✟ 0✞☎ +☛ ✟ ✞☎ D ≡ ✞☎ +✞☎ ✝✆ − ✞☎ +✞☎ + ✞☎ + ✞☎ ✞☎ − ✞☎ ✞☎ ✝✆✞☎ ☛ ✟ ✝✆ ☛ ✟ ✝✆ 0✞☎ ✝✆ +☛ ✟ +☛ ✟ ✞☎ = ✞☎ +✞☎ ✝✆ − ✞☎ +✞☎ + ✞☎ + ✞☎ ✞☎ + +✞☎ ✞☎ − 0✞☎ ✞☎ ✝✆✞☎ ✝✆ ✝✆ ✝✆ ✝✆ We again use that C1 = · · · = Cn−1 = 0 to obtain, modulo hn+1, ☛ ✟ ☛ ✟ 0✞☎ +☛ ✟ +☛ ✟ ✞☎ D ≡ = ✞☎ +✞☎ ✝✆ − ☛ ✟ ✞☎ +✞☎ ✝✆ − ✞☎ + +✞☎ ✞☎ + ✞☎ ✞☎ + +✞☎ ✞☎ − 0✞☎ ✞☎ ✝✆✞☎ ✝✆ 0☛ ✟ ✝✆ 0✞☎ ✝✆ +☛ ✟ ✝✆ ✞☎ ✞☎ + +✞☎ ✞☎ + ✞☎ ✞☎ − 0✞☎ ✞☎ ✝✆✞☎ ✝✆ ✝✆ ✝✆ CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 9 ☛ ✟ 0☛ ✟ 0✞☎ +☛ ✟ +✞☎ 0✞☎ = ✞☎ +✞☎ ✝✆ − ✞☎ + +✞☎ ✞☎ + ✞☎ ✞☎ − 0✞☎ ✞☎ − ✝✆✞☎ ✞☎ ✝✆✞☎ ✝✆ ✝✆ ✝✆ We use that C1 = · · · = Cn−1 = 0 once more to get modulo hn+1 ☛ ✟ 0☛ ✟ 0✞☎ +☛ ✟ +☛ ✟ 0✞☎ D ≡ − ✞☎ +✞☎ ✝✆ ☛ ✟ ✞☎ + +✞☎ ✞☎ + ✞☎ ✞☎ − 0✞☎ ✞☎ − ✞☎ ✞☎ ✝✆✞☎ ✝✆ 0☛ ✟ ✝✆ 0✞☎ ✝✆ +☛ ✟ ✝✆ +☛ ✟ 0✞☎ ≡ ≡ ✞☎ − +✞☎ ✞☎ + +✞☎ ✞☎ + ✞☎ ✞☎ − 0✞☎ ✝✆ 0☛ ✟ ✝✆ 0☛ ✟ ✝✆ 0✞☎ ✝✆ 0✞☎ ✞☎ ✝✆✞☎ ✞☎ − ✞☎ ✝✆ ✞☎ − +✞☎ ✞☎ + +✞☎ ✞☎ − ✞☎ ✞☎ ✝✆✞☎ ✝✆ ✝✆ ✝✆ Since ∆0 = ∆op 0 we finally obtain 0☛ ✟ 0☛ ✟ 0☛ ✟ 0☛ ✟ 0✞☎ 0✞☎ D ≡ = ✞☎ − ✞☎ ✞☎ + 0✞☎ ✞☎ − 0✞☎ ✞☎ + ✞☎ ✞☎ − ✞☎ ✞☎ ✝✆✞☎ ✝✆ 0☛ ✟ ✝✆ 0☛ ✟ ✝✆ 0✞☎ ✝✆ 0✞☎ ✝✆ ✞☎ − ✞☎ ✞☎ + ✞☎ ✞☎ − ✞☎ ✞☎ ✝✆✞☎ ✝✆ ✝✆ ✝✆ Evaluating at a ∈ m and simplifying it follows that a ✞☎ a ✞☎ ✞☎✞☎ ✝✆ ≡ ✞☎ − ✞☎ ✝✆ a ✞☎ − ✞☎ a ✞☎ ✞☎ + a ✞☎ − ✞☎ a ✞☎ ✞☎ (mod hn+1) (cid:3) (cid:3) Lemma 3.2. If C1 = · · · = Cn−1 = 0 then for any a ∈ m we have Cn(a) ∈ m ∧ m ⊗ U (m). 10 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV Proof. Lemma 3.1 implies that, modulo hn+1, the element a ✞☎ a ✞☎ ✞☎✞☎ ✝✆ ✞☎ − ✞☎ ✝✆ is skew-symmetric with respect to the first and second slots. Therefore a ✞☎ a ✞☎ ✞☎✞☎ ✝✆ ≡ − ✞☎ − ✞☎ ✝✆ a ☛ ✟ a ✞☎ ✞☎ + ☛ ✟ ✞☎ ✝✆ ✞☎ ✝✆ Using Lemma 3.1 and this equality we get a a a ☛ ✟ ☛ ✟ ☛ ✟ a ☛ ✟ a ✞☎ a ✞☎ ☛ ✟ ✞☎✞☎ ✝✆✝✆ − ✞☎☛ ✟ ✝✆✞☎ ✝✆ ≡ ✞☎ ✞☎ − + ☛ ✟ − ☛ ✟ ✞☎ ✝✆ a ☛ ✟ ✞☎ ✝✆ a ✞☎ ✞☎ ✝✆ a ✞☎ ✞☎ ✝✆ a ✞☎ ≡ ✞☎ + ✞☎ − ✞☎ ✝✆ ✞☎ ✝✆ ✞☎✞☎ ✝✆ − ☛ ✟ ✞☎ ✝✆ Since C1 = · · · = Cn−1 = 0 this implies that a ☛ ✟ a ☛ ✟ a ☛ ✟ a ✞☎ a ✞☎ a ✞☎ ☛ ✟ − 0✞☎0✞☎ 0✝✆0✝✆ 0✞☎☛ ✟ ≡ 0✝✆ 0✞☎ 0✝✆ ✞☎ + ✞☎ − 0✞☎ 0✝✆ 0✞☎ 0✝✆ 0✞☎✞☎ 0✝✆ − ☛ ✟ 0✞☎ 0✝✆ i.e., Cn(a) ∈ ker(Q ⊗ Q ⊗ Id −Q ⊗ Id ⊗ Id − Id ⊗Q ⊗ Id). The space U (m) ⊗ U (m) ⊗ U (m) is spanned by {ai ⊗ bj ⊗ x a, b ∈ m, i, j ≥ 0, x ∈ U (m)} and (Q ⊗ Q ⊗ Id −Q ⊗ Id ⊗ Id − Id ⊗Q ⊗ Id)(ai ⊗ bj ⊗ x) = (2i+j − 2i − 2j)ai ⊗ bj ⊗ x. This shows that Cn(a) ∈ m ⊗ m ⊗ U (m). Now, Lemma 3.1 implies that the element 2 a ✞☎ ✞☎ − a ✞☎ ✞☎! = a ✞☎ a ✞☎ a ✞☎ a ✞☎ 0✞☎✞☎ 0✝✆ = ✞☎ − 0✞☎ 0✝✆ ✞☎✞☎ ✝✆ ✞☎ − ✞☎ ✝✆ is skew-symmetric with respect to the first and second slot. Thus Cn(a) ∈ m ∧ m ⊗ U (m). (cid:3) (cid:3) CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 11 Lemma 3.3. If C1 = · · · = Cn−1 = 0 then for any a ∈ m we have a ✞☎ a ✞☎ ✞☎✞☎ ✝✆ − ✞☎ ≡ ✞☎ ✝✆ a ✞☎ − ✞☎ a ✞☎ + ✞☎ a ✞☎ − ✞☎ a ✞☎ ✞☎ (mod hn+1) ✞☎ ✞☎ ✞☎✞☎ ✞☎ ✝✆− Proof. Let D := ✞☎ ✝✆ comultiplication of U (m) imply that . Lemma 2.1 and the coassociativity of the ✞☎ ☛ ✟ ✞☎ ☛ ✟ ☛ ✟ D = = ✞☎ − ✝✆✞☎ ✞☎ ✞☎ − ✝✆✞☎ ✞☎ = ✞☎ ✝✆ ☛ ✟ ✞☎ + +✞☎ ✞☎ − ✝✆✞☎ ✞☎ − +✞☎ ✝✆ ☛ ✟ ☛ ✟ ✞☎ 0✞☎ ✝✆ ✞☎ − +✞☎ ✞☎ ✞☎ ✝✆ ✝✆ ✝✆ Using Lemma 2.1 again, ✞☎ ☛ ✟ ☛ ✟ ✞☎ − ✝✆ ✞☎ ✞☎ + +✞☎ ✞☎ − +✞☎ ✞☎ ✞☎ ✞☎ D = ≡ = ✝✆ ✞☎ ☛ ✟ ✝✆ ☛ ✟ ✝✆ 0✞☎ +☛ ✟ ✞☎ − ✝✆ ✞☎ ✞☎ + +✞☎ ✞☎ − +✞☎ ✞☎ ✞☎ − ✞☎ ✞☎ (mod hn+1) ✝✆ ✞☎ ☛ ✟ ✝✆ ☛ ✟ ✝✆ 0✞☎ ✝✆ +☛ ✟ +☛ ✟ ✞☎ − ✝✆ ✞☎ ✞☎ + +✞☎ ✞☎ − +✞☎ ✞☎ ✞☎ − ✞☎ +✞☎ − ✞☎ 0✞☎ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ where the second congruence follows from the fact that C1 = · · · = Cn−1 = 0 and the last equality is a consequence of the coassociativity of U (m). Again, using C1 = · · · = Cn−1 = 0 we deduce that modulo hn+1 0✞☎ +☛ ✟ ☛ ✟ ☛ ✟ 0✞☎ +☛ ✟ D ≡ ✞☎ + ✝✆✞☎ ✞☎ +✞☎ − ✞☎ + +✞☎ ✞☎ − +✞☎ ✞☎ ✞☎ − ✞☎ +✞☎ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ 12 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV 0✞☎ +☛ ✟ ☛ ✟ ☛ ✟ 0✞☎ +☛ ✟ ≡ ✞☎ + ✝✆✞☎ ✞☎ +✞☎ − ✞☎ +✞☎ + ✞☎ +✞☎ − ✞☎ ✞☎ − ✞☎ +✞☎ ✝✆ ✝✆ ✝✆ ✝✆ Simplifying we get 0✞☎ 0☛ ✟ 0☛ ✟ 0✞☎ D ≡ ✞☎ − ✝✆✞☎ ✞☎ +✞☎ + ✞☎ +✞☎ − ✞☎ ✞☎ (mod hn+1) Since ∆0 = ∆op 0 then ✝✆ ✝✆ 0✞☎ 0☛ ✟ 0☛ ✟ 0✞☎ D ≡ ✞☎ − ✝✆✞☎ ✞☎ ✞☎ + ✞☎ ✞☎ − ✞☎ ✞☎ (mod hn+1) The result follows by evaluating this expression at a. (cid:3) ✝✆ ✝✆ Lemma 3.4. If C1 = · · · = Cn−1 = 0 then for any a ∈ m we have Cn(a) ∈ m ∧ m ∧ m Proof. Similar to the proof of Lemma 3.2. (cid:3) (cid:3) (cid:3) Theorem 3.5. Any bialgebra deformation of U (M(α, β, γ)) satisfying the left and right co-Moufang identities is coassociative and cocommutative. Proof. We have proved that U (M(α, β, γ)) is cocommutative, so we only have to prove that it is coassociative. We will proceed by induction. Assume that C1 = · · · = Cn−1 = 0. Since Cn(a) ∈ m ∧ m ∧ m then it is an eigenvector of Q ⊗ Id ⊗ Id and Id ⊗ Id ⊗Q of eigenvalue 2. Hence, Lemma 3.1 and Lemma 3.3 imply that a ✞☎ − ✞☎ a ✞☎ − ✞☎ 2  2  a ✞☎ ✞☎  ✞☎  a ✞☎ a ✞☎ − ✞☎ ≡ a ✞☎ − ✞☎ ≡ a ✞☎ ✞☎ + a ✞☎ − ✞☎ a ✞☎ + ✞☎ a ✞☎ − ✞☎ a ✞☎ ✞☎ a ✞☎ ✞☎ (mod hn+1) (mod hn+1) and ✞☎ and From the cocommutativity of ∆h we get a ✞☎ a ✞☎ a ✞☎ a ✞☎ − ≡ ✞☎ − ✞☎ ✞☎ − ✞☎ a ✞☎ , a ✞☎ ≡ ✞☎ a ✞☎ − ✞☎ a ✞☎ ✞☎ (mod hn+1) a ✞☎ ≡ ✞☎ a ✞☎ ≡ ✞☎ a ✞☎ ✞☎ (mod hn+1) CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 13 so Therefore, Lemma 2.2. 2  − a ✞☎ ✞☎ a ✞☎ − ✞☎ a ✞☎ ✞☎ a ✞☎ + ✞☎ a ✞☎ ✞☎ (mod hn+1) a ✞☎ ✞☎  ≡ − = 0, i.e., Cn(a) = 0 for all a ∈ m. The result follows from (cid:3) (cid:3) Theorem 3.6. Let Uh(m) be a bialgebra deformation of U (m) satisfying the left and right co-Moufang identities. Then, for any x ∈ Uh(m) the element Ch(x) belongs to the kernel of the map Uh(m) ⊗Uh(m) ⊗Uh(m) → Uh(m) ⊗Uh(m) ⊗Uh(m) x ⊗y ⊗z 7→ X x[1] • y ⊗x[2] ⊗z + x • z[1] ⊗y ⊗z[2] Proof. Let R : x ⊗y ⊗z 7→P x[1] • y ⊗x[2] ⊗z and S : x ⊗y ⊗z ⊗P x • z[1] ⊗y ⊗z[2]. The left co-Moufang identity ✞☎ ✞☎ ✞☎ ✞☎ ✝✆ ✞☎ ✞☎ = ✝✆ can be written as R(Cl(x)) = S(Cr(x)) where Cl(x) = (∆h ⊗ Id)∆h(x) and Cr(x) = (Id ⊗∆h)∆h(x). We also observe that ✞☎ ✞☎ ✞☎ ✞☎ = ✝✆ ✞☎ ✞☎ ✝✆ so S(Cl(x)) = R(Cr(x)). Therefore, (R + S)(Cl(x)) = (R + S)(Cr(x)) and (R + S)(Ch(x)) = (R + S)(Cl(x) − Cr(x)) = 0. (cid:3) (cid:3) Theorem 3.7. Let g be a finite-dimensional central simple Lie algebra and Uh(g) a bialgebra deformation of U (g) satisfying the left and right co-Moufang identities. Then Uh(g) is coassociative. Proof. Let us assume that C1 = · · · = Cn−1 = 0. We will prove that Cn = 0. The following Sweedler like notation for Cn(x) will be very useful: With this notation Theorem 3.6 gives (3.1) Cn(x) =X x′ ⊗ x′′ ⊗ x′′′, (2) ⊗ x′′′ +X x′x′′′ (1)x′′ ⊗ x′ X x′ x ∈ U (g) (1) ⊗ x′′ ⊗ x′′′ (2) = 0. Lemma 2.2 with ϕn = Pi+j=n(∆j ⊗ Id)∆i and ψn = Pi+j=n(Id ⊗∆j)∆i implies that Cn(xy) = X Cn(x)(y(1) ⊗ y(2) ⊗ y(3)) +X(x(1) ⊗ x(2) ⊗ x(3))Cn(y) 14 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV = X x′y(1) ⊗ x′′y(2) ⊗ x′′′y(3) +X x(1)y′ ⊗ x(2)y′′ ⊗ x(3)y′′′. We can apply Theorem 3.6 to the element Ch(xy) to get X(x′ (1)y(1))(x′′y(2)) ⊗ x′ (2)y(3) ⊗ x′′′y(4) (1))(x(2)y′′) ⊗ x(3)y′ (2) ⊗ x(4)y′′′ +X(x(1)y′ +X(x′y(1))(x′′′ +X(x(1)y′)(x(2)y′′′ (1)y(2)) ⊗ x′′y(3) ⊗ x′′′ (2)y(4) (1)) ⊗ x(3)y′′ ⊗ x(4)y′′′ (2) = 0. In fact, since Cn(x) and Cn(y) also satisfy (3.1) then all the summands where y(1) or x(2) are scalars vanish. Moreover, since Cn(g) ⊆ g ∧ g ∧ g then substituting x and y for elements a and b of m we get X(a′ba′′ ⊗ 1 ⊗ a′′′ + ba′′ ⊗ a′ ⊗ a′′′ + b′ab′′ ⊗ 1 ⊗ b′′′ +ab′′ ⊗ b′ ⊗ b′′′ + a′ba′′′ ⊗ a′′ ⊗ 1 + a′b ⊗ a′′ ⊗ a′′′ +b′ab′′′ ⊗ b′′ ⊗ 1 + b′a ⊗ b′′ ⊗ b′′′) = 0. Projecting onto ker ǫ ⊗ ker ǫ ⊗ ker ǫ this equality leads to X ba′′ ⊗ a′ ⊗ a′′′ + ab′′ ⊗ b′ ⊗ b′′′ + a′b ⊗ a′′ ⊗ a′′′ + b′a ⊗ b′′ ⊗ b′′′ = 0. The skew-symmetry of the tensors in Cn(g) implies that This equality with the substitution b = a is X[a′, b] ⊗ a′′ ⊗ a′′′ − [a, b′] ⊗ b′′ ⊗ b′′′ = 0. We can also obtain X[a, a′] ⊗ a′′ ⊗ a′′′ = 0. X a′ ⊗ [a, a′′] ⊗ a′′′ = 0, X a′ ⊗ a′′ ⊗ [a, a′′′] = 0 by the skew-symmetry of Cn(g). Therefore, if λ : g → End(g ∧ g ∧ g) denotes the representation of g on the g-module g ∧ g ∧ g then (3.2) λa · Cn(a) = 0. Lemma 2.2 implies that Cn is a 1-cocycle of g with values in g ∧ g ∧ g. Since g is simple this cocycle is a cobundary so there exists r ∈ g∧g∧g such that Cn(a) = λa·r. By (3.2), we finally obtain that λ2 a · r = 0 for all a ∈ g. In particular, the Casimir operator of g kills the element r. However, the Casimir operator acts as a non-zero scalar multiple of the identity on any non-trivial irreducible module of g. It follows that r spans a trivial submodule and that Cn(a) = λa · r = 0 for all a ∈ g. Hence Cn = 0 by Lemma 2.2. (cid:3) (cid:3) Theorem 3.8. Let g be a finite-dimensional central simple Lie algebra. If Uh(g) is a bialgebra deformation that satisfies the left and right Moufang identities then Uh(g) is associative. Proof. We will write the associator in Uh(g) of elements x, y, z ∈ U (g) as (x, y, z)• = (x • y) • z − x • (y • z) = A1(x, y, z)h + A2(x, y, z)h2 + · · · CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 15 and we will prove by induction that An = 0. Assume that A1 = · · · = An−1 = 0, then 0✞☎ +✞☎ 0✞☎ +✞☎ implies that ✝✆ + ✝✆ ✝✆ = ✝✆ ✝✆ + ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ ✝✆ 0✞☎ 0✞☎ ✝✆ ≡ ✝✆ ✝✆ ✝✆ (mod hn+1). Hence, modulo hn+1, the element a ∈ g verifies that ✝✆ ✝✆ a • (y • z) + y • (a • z) ≡ (z • y) • z + (y • a) • z i.e., (a, y, z)• ≡ −(y, a, z)•. Having also considered the right Moufang identity, we would have obtained that ¯a ∈ Nalt(Uh), where ¯x and Uh stand for x + hn+1Uh(g) and Uh(g)/hn+1Uh(g) respectively. Thus, the jacobian of ¯a, ¯b, ¯c ∈ g + hn+1Uh is J(¯a, ¯b, ¯c) = 6(¯a, ¯b, ¯c)• = 6An(a, b, c)hn Let m be the Malcev subalgebra generated by g + hn+1Uh inside Uh but where the product is understood to be the commutator product x • y − y • x. Since m/( m ∩ hUh) ∼= ( m + hUh)/hUh ⊆ Uh/hUh ∼= U (g) then m/( m ∩ hUh) is a Lie algebra isomorphic to g. Since m ∩ hUh is a nilpotent ideal of m, the radical in fact, we can find a Lie subalgebra s of m such that m = s ⊕ ( m ∩ hUh), a direct sum of vector spaces [11, 16]. Let S be the subalgebra of Uh generated by {1} ∪ s ∪ {h}. Clearly S/(S ∩ hUh) ∼= (S + hUh)/hUh ∼= U (g) which implies that hnS = hnUh since hn+1Uh = 0 = hn+1S. We recursively get hiS = hiUh for i = n, n − 1, . . . , 1, 0 so S = Uh. Since s ⊆ Nalt(Uh) and s is a Lie algebra then S is associative. Thus An(x, y, z) = 0. (cid:3) (cid:3) Appendix A. Proof of Theorem2.3 We include a non-graphical proof of Theorem 2.3 so that the reader can check the correspondence between the movements performed in the diagrams and the algebraic manipulations of the equalities. Recall that ∆h = Pn hn∆n and ph = Pn hnpn stand for the comultiplication and multiplication of Uh(m) respectively. The map x ⊗y 7→ y ⊗x will be denoted by τ . We rewrite part of the statement of Lemma 2.1 for convenience: ✞☎ ✞☎ corresponds to (ph ⊗ Id)(Id ⊗∆h)∆h = ((ph∆h) ⊗ Id)∆h = ✞☎ ✝✆ ✞☎ ✝✆ and ✞☎ ✞☎ ✞☎ = ✞☎ corresponds to (ph ⊗ Id)(Id ⊗∆op h )∆h = (ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆h. ✝✆ ✝✆ 16 JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV Part 2) of Lemma 2.1 establishes that (Id ⊗ph)(∆h ⊗ Id)∆h = (Id ⊗(ph∆h))∆h. Af- ter composing with τ we get τ (Id ⊗(ph∆h))∆h = τ (Id ⊗ph)(∆h ⊗ Id)∆h. The left- hand side of this equality is ((ph∆h) ⊗ Id)∆op h . The right-hand side of the equality is τ (Id ⊗ph)(τ ⊗ Id)(∆op ⊗ Id)∆h. Therefore, h ⊗ Id)∆h = (ph ⊗ Id)(Id ⊗τ )(∆op h = (ph ⊗ Id)(Id ⊗τ )(∆op h h ((ph∆h) ⊗ Id)∆op ⊗ Id)∆h The proof of Theorem 2.3 goes as follows. By Lemma 2.1 we have D := (ph∆h ⊗ Id)∆h − (ph∆h ⊗ Id)∆op h = (ph ⊗ Id)(Id ⊗∆h)∆h − (ph∆h ⊗ Id)∆op h = (ph ⊗ Id)(Id ⊗∆h)∆h − (ph ⊗ Id)(Id ⊗τ )(∆op j = 0, . . . , n − 1 then, modulo hn+1, h Since ∆j = ∆op j ⊗ Id)∆h D ≡ (ph ⊗ Id)(Id ⊗∆h)∆h − (ph ⊗ Id)(Id ⊗τ )(∆op ⊗ Id)∆0 hi+j+k(pk ⊗ Id)(Id ⊗τ )(∆j ⊗ Id)∆i − Xi+j+k≤n 1≤i ≡ (ph ⊗ Id)(Id ⊗∆h)∆h − (ph ⊗ Id)(Id ⊗τ )(∆op h ⊗ Id)∆0 +(ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆0 − (ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆h By Lemma 2.1 D = (ph ⊗ Id)(Id ⊗∆h)∆h − (ph ⊗ Id)(Id ⊗τ )(∆op h ⊗ Id)∆0 +(ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆0 − (ph ⊗ Id)(Id ⊗∆op ⊗ Id)∆0 +(ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆0 − (ph ⊗ Id)(Id ⊗∆op ≡ (ph ⊗ Id)(Id ⊗∆h)∆h − (ph ⊗ Id)(Id ⊗τ )(∆op h h )∆h h )∆0 hi+j+k(pk ⊗ Id)(Id ⊗∆op j )∆i. − Xi+j+k≤n 1≤i Since ∆j = ∆op j j = 0, . . . , n − 1 then, modulo hn+1, D ≡ (ph ⊗ Id)(Id ⊗∆h)∆h − (ph ⊗ Id)(Id ⊗τ )(∆op ⊗ Id)∆0 +(ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆0 − (ph ⊗ Id)(Id ⊗∆op h h )∆0 hi+j+k(pk ⊗ Id)(Id ⊗∆j)∆i − Xi+j+k≤n 1≤i ≡ (ph ⊗ Id)(Id ⊗∆h)∆0 − (ph ⊗ Id)(Id ⊗τ )(∆op ⊗ Id)∆0 +(ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆0 − (ph ⊗ Id)(Id ⊗∆op h h )∆0. Any a ∈ m is primitive, i.e., ∆0(a) = a ⊗ 1 + 1 ⊗ a thus (ph ⊗ Id)(Id ⊗∆h)∆0(a) = (ph ⊗ Id)(Id ⊗∆h)(a ⊗ 1 + 1 ⊗ a) = ∆h(a) + a ⊗ 1 (ph ⊗ Id)(Id ⊗τ )(∆op h ⊗ Id)∆0(a) = (ph ⊗ Id)(Id ⊗τ )(∆op h ⊗ Id)(a ⊗ 1 + 1 ⊗ a) = ∆op h (a) + a ⊗ 1 (ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)∆0(a) = (ph ⊗ Id)(Id ⊗τ )(∆h ⊗ Id)(a ⊗ 1 + 1 ⊗ a) = ∆h(a) + a ⊗ 1 (ph ⊗ Id)(Id ⊗∆op h )∆0(a) = CO-MOUFANG DEFORMATIONS OF U(M(α, β, γ)) 17 (ph ⊗ Id)(Id ⊗∆op h )(a ⊗ 1 + 1 ⊗ a) = ∆op h (a) + a ⊗ 1 and (ph∆h ⊗ Id)∆h(a) − (ph∆h ⊗ Id)∆op h (a) ≡ 2(∆h(a) − ∆op h (a)) (mod hn+1) Since ∆i = ∆op i i = 0, . . . , n − 1 we finally get (p0∆0 ⊗ Id)∆n(a) − (p0∆0 ⊗ Id)∆op This shows that (∆n − ∆op n (a)). n )(a) is an eigenvector of Q of eigenvalue 2 so n (a) = 2(∆n(a) − ∆op for any a ∈ m. The result follows from [17]. (∆n − ∆op n )(a) ∈ m ∧ m References [1] G. Benkart, S. Madariaga, and J. M. P´erez-Izquierdo, Hopf algebras with triality, to appear in Trans. Amer. Math. Soc. [2] Murray R. Bremner, Irvin R. Hentzel, Luiz A. Peresi, and Hamid Usefi, Universal enveloping algebras of the four-dimensional Malcev algebra, Algebras, representations and applications, Contemp. Math., vol. 483, Amer. Math. Soc., Providence, RI, 2009, pp. 73 -- 89. MR2497952 (2010g:17039) [3] Murray R. Bremner, Irvin R. Hentzel, Luiz A. Peresi, Marina V. Tvalavadze, and Hamid Usefi, Enveloping algebras of Malcev algebras, Comment. Math. Univ. Carolin. 51 (2010), no. 2, 157 -- 174. MR2682471 (2011j:17056) [4] Marina V. Tvalavadze and Murray R. Bremner, Enveloping algebras of solvable Malcev alge- bras of dimension five, Comm. Algebra 39 (2011), no. 8, 2816 -- 2837. MR2834132 [5] Murray R. Bremner and Hamid Usefi, Enveloping algebras of the nilpotent Malcev alge- bra of dimension five, Algebr. Represent. Theory 13 (2010), no. 4, 407 -- 425. MR2660854 (2011k:17055) [6] Renate Carlsson, On the exceptional central simple non-Lie Mal′cev algebras, Trans. Amer. Math. Soc. 244 (1978), 173 -- 184. MR506614 (80a:17001) [7] Vyjayanthi Chari and Andrew Pressley, A guide to quantum groups, Cambridge University Press, Cambridge, 1995. Corrected reprint of the 1994 original. MR1358358 (96h:17014) [8] Stephen Doro, Simple Moufang loops, Math. Proc. Cambridge Philos. Soc. 83 (1978), no. 3, 377 -- 392. MR0492031 (58 #11195) [9] V. G. Drinfel′d, Quantum groups, (Berkeley, Calif., 1986), Amer. Math. Soc., Providence, RI, 1987, pp. 798 -- 820. MR934283 (89f:17017) [10] George Glauberman, On loops of odd order. II, J. Algebra 8 (1968), 393 -- 414. MR0222198 (36 #5250) [11] A. N. Griskov, An analogue of Levi's theorem for Mal′cev algebras, Algebra i Logika 16 (1977), no. 4, 389 -- 396, 493 (Russian). MR0573915 (58 #28114) [12] Alexander Grishkov, Lie algebras with triality, J. Algebra 266 (2003), no. 2, 698 -- 722. MR1995132 (2004h:17019) [13] Alexander N. Grishkov and Andrei V. Zavarnitsine, Groups with triality, J. Algebra Appl. 5 (2006), no. 4, 441 -- 463. MR2239539 (2007g:20062) [14] Michio Jimbo, A q-difference analogue of U (g) and the Yang-Baxter equation, Lett. Math. Phys. 10 (1985), no. 1, 63 -- 69. MR797001 (86k:17008) [15] J. Klim and S. Majid, Hopf quasigroups and the algebraic 7-sphere, preprint, available at arXiv:0906.5026v3[math.QA]. [16] E. N. Kuz′min, Levi's theorem for Mal′cev algebras, Algebra i Logika 16 (1977), no. 4, 424 -- 431, 493 (Russian). MR0573914 (58 #28113) [17] S. Madariaga and J. M. P´erez-Izquierdo, Non-existence of coassociative quantized universal enveloping algebras of the traceless octonions, to appear in Comm. Algebra. [18] A. I. Mal′cev, Analytic loops, Mat. Sb. N.S. 36(78) (1955), 569 -- 576 (Russian). MR0069190 (16,997g) [19] P. O. Mikheev, On the embedding of Mal′tsev algebras into Lie algebras, Algebra i Logika 31 (1992), no. 2, 167 -- 173, 221 (Russian, with Russian summary); English transl., Algebra and Logic 31 (1992), no. 2, 106 -- 110 (1993). MR1289030 18 [20] JOS ´E M. P ´EREZ-IZQUIERDO AND IVAN P. SHESTAKOV , Groups that envelop Moufang loops, Uspekhi Mat. Nauk 48 (1993), no. 2(290), 191 -- 192 (Russian, with Russian summary); English transl., Russian Math. Surveys 48 (1993), no. 2, 195 -- 196. MR1239875 (94g:20098) [21] J. Mostovoy and J. M. P´erez-Izquierdo, Formal multiplications, bialgebras of distributions and nonassociative Lie theory, Transform. Groups 15 (2010), no. 3, 625 -- 653. MR2718940 (2011i:20104) [22] J. Mostovoy, J. M. Perez-Izquierdo, and I. P. Shestakov, Hopf algebras in non-associative Lie theory, Bull. Math. Sci. 4 (2014), no. 1, 129 -- 173. MR3174282 [23] Jos´e M. P´erez-Izquierdo, Algebras, hyperalgebras, nonassociative bialgebras and loops, Adv. Math. 208 (2007), no. 2, 834 -- 876. [24] Jos´e M. P´erez-Izquierdo and Ivan P. Shestakov, An envelope for Malcev algebras, J. Algebra 272 (2004), no. 1, 379 -- 393. MR2029038 (2004j:17040) [25] , On the center of the universal enveloping algebra of the central simple non-Lie Malcev algebras in characteristic p, Proceedings of Jordan Structures in Algebra and Analysis Meeting, Editorial C´ırculo Rojo, Almer´ıa, 2010, pp. 227 -- 242. MR2648360 (2011j:17058) [26] N. Reshetikhin, Quantization of Lie bialgebras, Internat. Math. Res. Notices 7 (1992), 143 -- 151. MR1174619 (93h:17041) [27] L. V. Sabinin and P. O. Mikheev, Infinitesimal theory of local analytic loops, Dokl. Akad. Nauk SSSR 297 (1987), no. 4, 801 -- 804 (Russian); English transl., Soviet Math. Dokl. 36 (1988), no. 3, 545 -- 548. MR924255 (89g:22003) [28] Arthur A. Sagle, Malcev algebras, Trans. Amer. Math. Soc. 101 (1961), 426 -- 458. MR0143791 (26 #1343) [29] V. N. Zhelyabin and I. P. Shestakov, Chevalley and Kostant theorems for Mal′tsev algebras, Algebra Logika 46 (2007), no. 5, 560 -- 584, 664 (Russian, with Russian summary); English transl., Algebra Logic 46 (2007), no. 5, 303 -- 317. MR2378631 (2009e:17065) [30] K. A. Zhevlakov, A. M. Slin′ko, I. P. Shestakov, and A. I. Shirshov, Rings that are nearly associative, Pure and Applied Mathematics, vol. 104, Academic Press Inc. [Harcourt Brace Jovanovich Publishers], New York, 1982. Translated from the Russian by Harry F. Smith. MR668355 (83i:17001) Departamento de Matem´aticas y Computaci´on, Universidad de La Rioja, 26004, Lo- grono, Spain E-mail address: [email protected] Instituto de Matematica e Estat´ıstica, Universidade de Sao Paulo, Caixa postal 66281, CEP 05315-970, Sao Paulo, Brazil E-mail address: [email protected]
1101.2239	3	1101	2011-06-02T22:47:11	Obstructing extensions of the functor Spec to noncommutative rings	[ "math.RA", "math.AG", "math.OA" ]	In this paper we study contravariant functors from the category of rings to the category of sets whose restriction to the full subcategory of commutative rings is isomorphic to the prime spectrum functor Spec. The main result reveals a common characteristic of these functors: every such functor assigns the empty set to M_n(C) for n >= 3. The proof relies, in part, on the Kochen-Specker Theorem of quantum mechanics. The analogous result for noncommutative extensions of the Gelfand spectrum functor for C*-algebras is also proved.	math.RA	math	OBSTRUCTING EXTENSIONS OF THE FUNCTOR SPEC TO NONCOMMUTATIVE RINGS MANUEL L. REYES Abstract. This paper concerns contravariant functors from the category of rings to the category of sets whose restriction to the full subcategory of commutative rings is isomorphic to the prime spectrum functor Spec. The main result reveals a common characteristic of these functors: every such functor assigns the empty set to Mn(C) for n ≥ 3. The proof relies, in part, on the Kochen-Specker Theorem of quantum mechanics. The analogous result for noncommutative extensions of the Gelfand spectrum functor for C ∗-algebras is also proved. 1. Introduction The prime spectrum of commutative rings and the Gelfand spectrum of commutative C ∗-algebras play a foundational role in the classical link between algebra and geometry, since these spectra form the underlying point-sets of the spaces attached to a commutative ring or C ∗-algebra. It is tempting to hope that one could extend these spectra to the noncommutative setting in order to construct the "underlying set of a noncommutative space." The main results of this paper (Theorems 1.1 and 1.2 below) hinder naive attempts to do so by obstructing the existence of functors that extend these spectra. In order to produce an obstruction, one must first fix the desired properties of the "non- commutative spectrum" in question. Consider the prime spectrum Spec. From the viewpoint of Spec as an underlying point-set, two facts of key importance are (1) the spectrum of every nonzero commutative ring is nonempty, and (2) the prime spectrum construction can be regarded as a contravariant functor from the category of commutative rings to the category of sets, Spec : CommRing → Set . (For commutative rings, Spec is easily made into a functor because the inverse image of a prime ideal under a ring homomorphism is again prime.) Over the years, many different extensions of the prime spectrum to noncommutative rings have been studied. Let F be a rule assigning to each ring R a set F (R), such that for every Date: June 2, 2011. 2010 Mathematics Subject Classification. Primary: 16B50, 14A22, 46L85; Secondary: 81P15. Key words and phrases. spectrum functor, matrix algebra, commutative subring, partial algebra, prime partial ideal, Kochen-Specker Theorem. The author was supported by a Ford Foundation Predoctoral Diversity Fellowship at the University of California, Berkeley, and a University of California President's Postdoctoral Fellowship at the University of California, San Diego. 1 2 MANUEL L. REYES commutative ring C one has F (C) ∼= Spec(C). There are two desirable properties that such an invariant may possess. Property A: For every nonzero ring R, the set F (R) is nonempty. Property B: The invariant F can be made into a set-valued functor extending Spec, in the sense that the assignment R 7→ F (R) is the object part of a functor F whose restriction to the category of commutative rings is isomorphic to Spec. Examples of invariants that satisfy Property A include the set of prime ideals of a noncommu- tative ring, Goldman's prime torsion theories [7], and the "left spectrum" of Rosenberg [16]. (These invariants satisfy Property A because they all have elements corresponding to maxi- mal one- or two-sided ideals.) Some invariants that satisfy Property B are the spectrum of the "abelianization" R 7→ Spec(R/[R, R]), the set of completely prime ideals, and the "field spectrum" of Cohn [4]. Each of the different "noncommutative spectra" listed above possess only one of the two properties. Our first main result states that this situation is inevitable. Theorem 1.1. Let F be a contravariant functor from the category of rings to the category of sets whose restriction to the full subcategory of commutative rings is isomorphic to Spec. Then F (Mn(C)) = ∅ for any n ≥ 3. Next we state the analogous result in the context of C ∗-algebras. For our purposes, we define the Gelfand spectrum of a commutative unital C ∗-algebra A to be the set Max(A) of maximal ideals of A; these are necessarily closed in A. The set Max(A) is in bijection with the set of characters of A, which are the nonzero multiplicative linear functionals (equivalently, unital algebra homomorphisms) A → C; the correspondence associates to each character its kernel (see [5, Thm. I.2.5]). This is easily given the structure of a contravariant functor Max : CommCAlg → Set . With appropriate topologies taken into account, the Gelfand spectrum functor provides a contravariant equivalence between the category of commutative unital C ∗-algebras and the category of compact Hausdorff spaces. The following analogue of Theorem 1.1 provides a similar obstruction to any noncommu- tative extension of the Gelfand spectrum functor. Theorem 1.2. Let F be a contravariant functor from the category of unital C ∗-algebras to the category of sets whose restriction to the full subcategory of commutative unital C ∗-algebras is isomorphic to Max. Then F (Mn(C)) = ∅ for any n ≥ 3. Of course, the statements of Theorems 1.1 and 1.2 with the category of sets replaced by the category Top of topological spaces follow as immediate corollaries. There are plenty of results stating that a particular spectrum of a ring or algebra is empty. For instance, it is easy to find examples of noncommutative C ∗-algebras that have no characters. In the realm of algebra, one can think of rings that have no homomorphisms to any division ring as having empty spectra. For one more example, S. P. Smith suggested a notion of "closed point" such that every infinite dimensional simple C-algebra has no closed points [19, p. 2170]. Notice that each of these examples assumes a fixed notion of spectrum. OBSTRUCTING EXTENSIONS OF SPEC 3 The main feature setting Theorem 1.1 apart from the arguments mentioned above is that it applies to any notion of spectrum satisfying Properties A and B mentioned above, and similarly for Theorem 1.2. Indeed, these spectra need not be defined in terms of ideals (either one-sided or two-sided) or modules at all. Outline of the proof. The proofs of Theorems 1.1 and 1.2 proceed roughly as follows: (1) construct a functor that is "universal" among all functors whose restriction to the com- mutative subcategory is the spectrum functor; (2) show that this functor assigns the empty set to Mn(C); (3) by universality, conclude that every such functor does the same. It is perhaps surprising that a key tool used for step (2) above is the Kochen-Specker Theorem [11] of quantum mechanics, which forbids the existence of certain hidden variable theories. Recently this result has surfaced in the context of noncommutative geometry in the Bohrification construction introduced by C. Heunen, N. Landsman, and B. Spitters in [8, Thm. 6]. Those authors use the Kochen-Specker Theorem to show that a certain "space" associated to the C ∗-algebra of bounded operators on a Hilbert space of dimension ≥ 3 has no points. This is obviously close in spirit to Theorems 1.1 and 1.2. A common theme between that paper and the present one is the focus on commutative subalgebras of a given algebra, and we acknowledge the inspiration and influence of that work on ours. In the ring-theoretic case, step (1) is achieved in Section 2. The universal functor p-Spec is defined in terms of prime partial ideals, which requires an exposition of partial algebras along with their ideals and morphisms. Step (2) is carried out in Section 3, where we establish a connection between prime partial ideals and the Kochen-Specker Theorem. The proof of Theorem 1.1 (basically Step (3) above) is given in Section 4, and it is accompanied by some corollaries. In Section 5 we prove Theorem 1.2 by quickly following Steps (1)–(3) in the context of C ∗-algebras, and we state a few of its corollaries. Generalizations and positive implications. Since the present results were announced, stronger obstructions to spectrum functors have been proved by B. van den Berg and C. He- unen in [3]. Those results hinder the extension of Spec and Max even when they are con- sidered as functors whose codomains are over-categories of Top, such as the categories of locales and toposes. However, one can view these obstructions in a positive light: it seems that the actual construction of contravariant functors extending the classical spectra neces- sitates a creative choice of target category C that contains Top (or Set, if one forgets the topology). From this perspective, the construction of "useful" noncommutative spectrum functors extending the classical ones seems to remain an interesting issue. Conventions. All rings are assumed to have identity and ring homomorphisms are as- sumed to preserve the identity, except where explicitly stated otherwise. The categories of unital rings and unital commutative rings are respectively denoted by Ring and CommRing. We will consider Spec as a contravariant functor from the category of commutative rings to the category of sets, instead of topological spaces, unless indicated otherwise. A con- travariant functor F : C1 → C2 can also be viewed as a covariant functor out of the opposite category F : Cop 1 → C2. For the most part, we will view contravariant functors as functors that reverse the direction of arrows, in order to avoid dealing with "opposite arrows." But 4 MANUEL L. REYES when it is convenient we will occasionally change viewpoint and consider contravariant func- tors as covariant functors out of the opposite category. Given a category C, we will often write C ∈ C to mean that C is an object of C. When there is danger of confusion, we will write the more precise expression C ∈ Obj(C). 2. A universal Spec functor from prime partial ideals In this section we will define a functor p-Spec that is universal among all candidates for a "noncommutative Spec." We set the stage for its construction by describing the universal property that we seek. Given categories C and C ′, we let Fun(C, C ′) denote the category of (covariant) functors from C to C ′ whose morphisms are natural transformations. (This category need not have small Hom-sets.) The inclusion of categories CommRing ֒→ Ring induces a restriction functor r : Fun(Ringop, Set) → Fun(CommRingop, Set) F 7→ F CommRingop, which is defined in the obvious way on morphisms (i.e., natural transformations). Now we define the "fiber category" over Spec ∈ Fun(CommRingop, Set) to be the category r −1(Spec) ∼−→ Spec an isomor- whose objects are pairs (F, φ) with F ∈ Fun(Ringop, Set) and φ : r(F ) phism of functors, in which a morphism ψ : (F, φ) → (F ′, φ′) is a morphism ψ : F → F ′ of functors such that φ′ ◦ r(ψ) = φ, i.e. the following commutes: r(F ) r(ψ) r(F ′) #FFFFFFFF φ {wwwwwwww φ′ Spec . −1 is slightly different from other (Our use of the terminology "fiber category" and notation r instances in the literature. The main difference is that we are considering objects that map to Spec under r up to isomorphism, rather than "on the nose.") The category r −1(Spec) is of fundamental importance to us; we are precisely interested in those contravariant functors from Ring to Set whose restriction to CommRing is isomorphic to Spec. The "universal Spec functor" p-Spec that we seek is a terminal object in this category. The rest of this section is devoted to defining this functor and proving its universal property. The functor p-Spec to be constructed is best understood in the context of partial algebras, whose definition we recall here. The notion of a partial algebra was defined in [11, §2]. (A more precise term for this object would probably be partial commutative algebra, but we retain the historical and more concise terminology in this paper.) Definition 2.1. A partial algebra over a commutative ring k is a set R with a reflexive symmetric binary relation ⊥ ⊆ R × R (called commeasurability), partial addition and mul- tiplication operations + and · that are functions ⊥ → R, a scalar multiplication operation k × R → R, and elements 0, 1 ∈ A such that the following axioms are satisfied: (1) For all a ∈ R, a ⊥ 0 and a ⊥ 1; / / # { OBSTRUCTING EXTENSIONS OF SPEC 5 (2) The relation ⊥ is preserved by the partial binary operations: for all a1, a2, a3 ∈ R with ai ⊥ aj (1 ≤ i, j ≤ 3) and for all λ ∈ k, one has (a1 + a2) ⊥ a3, (a1a2) ⊥ a3, and (λa1) ⊥ a2; (3) If ai ⊥ aj for 1 ≤ i, j ≤ 3, then the values of all (commutative) polynomials in a1, a2, and a3 form a commutative k-algebra. A partial ring is a partial algebra over k = Z. The third axiom of a partial algebra appears as stated in [11, p. 64]. While the axiom is succinct, it can be instructive to unravel its meaning. The third axiom is equivalent to the following collection of axioms: (3.0) The element 0 ∈ R is an additive identity and 1 ∈ R is a multiplicative identity; (3.1) Addition and multiplication are commutative when defined: if a ⊥ b in R, then a + b = b + a and ab = ba; (3.2) Addition and multiplication are associative on commeasurable triples: if a ⊥ b, a ⊥ c, and b ⊥ c in R, then (a + b) + c = a + (b + c) and (a · b) · c = a · (b · c); (3.3) Multiplication distributes over addition on commeasurable triples: if a ⊥ b, a ⊥ c, and b ⊥ c in R, then a · (b + c) = a · b + a · c; (3.4) Each element a ∈ R is commeasurable to an element −a ∈ R that is an additive inverse to a and such that a ⊥ r =⇒ −a ⊥ r for all r ∈ R (see the paragraph before Lemma 3.3 for a discussion of uniqueness of inverses); (3.5) Multiplication is k-bilinear. Definition 2.2. A commeasurable subalgebra of a partial k-algebra R is a subset C ⊆ R consisting of pairwise commeasurable elements that is closed under k-scalar multiplication and the partial binary operations of R. (Thus the operations of R restricted to C endow C with the structure of a commutative k-algebra.) In particular, given any a ∈ R one can evaluate every polynomial in k[x] at x = a to obtain commeasurable k-subalgebra k[a] ⊆ R. More generally, any set of pairwise commeasurable elements of R is contained in a commeasurable k-subalgebra of R. Notice also that R is the union of its commeasurable k-subalgebras. When we need to distinguish between a k-algebra and a partial k-algebra, we shall refer to the former as a "full" algebra. As the following example shows, every full algebra can be considered as a partial algebra in a standard way. Example 2.3. Let R be a (full) algebra over a commutative ring k. We may define a relation ⊥ ⊆ R × R by a ⊥ b if and only if ab = ba (i.e., [a, b] = 0). This relation along with the addition, multiplication, and scalar multiplication inherited from R make R into a partial algebra over k. For us, this is the prototypical example of a partial algebra. We will refer to this as the "standard partial algebra structure" on R. Considering a full algebra R as a partial algebra is, in effect, a way to restrict our attention to only the commutative subalgebras of R. This is further amplified when one applies the notions (defined below) of morphisms of partial algebras and partial ideals to the algebra R. Example 2.4. Another important example of a partial algebra is considered in [11]. Let A be a unital C ∗-algebra, and let Asa denote the set of self-adjoint elements of A. Notice that 6 MANUEL L. REYES the sum and product of two commuting self-adjoint elements is again self-adjoint, and that real scalar multiplication preserves Asa. So if ⊥ ⊆ Asa × Asa is the relation of commutativity (as in the previous example), then Asa forms a partial algebra over R. Just as one may study ideals of a k-algebra, we will consider "partial ideals" of a partial k-algebra. Definition 2.5. Let R be a partial algebra over a commutative ring k. A subset I ⊆ R is a partial ideal of R if, for all a, b ∈ R such that a ⊥ b, one has: • a, b ∈ I =⇒ a + b ∈ I; • b ∈ I =⇒ ab ∈ I. Equivalently, a partial ideal of R is a subset I ⊆ R such that, for every commeasurable subalgebra C ⊆ R, the intersection I ∩ C is an ideal of C. If R is a (full) k-algebra, then a partial ideal of R is a partial ideal of the standard partial algebra structure on R. To better understand the set of partial ideals of an arbitrary (full or partial) algebra, it helps to consider some general examples. Let R be an algebra over a commutative ring k. If I is a left, right, or two-sided ideal of R, then I is a clearly a partial ideal of R. Furthermore, when R is commutative the partial ideals of R are precisely the ideals of R. Lemma 2.6. Let I be a partial ideal of a partial k-algebra R. Then I = R if and only if 1 ∈ I. Proof. ("If" direction.) If 1 ∈ I, then 1 ⊥ R gives R = (R · 1) ⊆ I. Hence I = R. (cid:3) Proposition 2.7. Let D be a division ring. Then the only partial ideals of D are 0 and D. Proof. Suppose that I ⊆ D is a nonzero partial ideal, and let 0 6= a ∈ I. Then a ⊥ a−1, so 1 = a−1 · a ∈ I. It follows from Lemma 2.6 that I = D. (cid:3) Yet another example of a partial ideal in an arbitrary ring R is the set N ⊆ R of nilpotent elements of R. Indeed, for any commutative subring C of R, N ∩ C is the nilradical of C and hence is an ideal of C. It is well-known that the set of nilpotent elements of a ring R is not even closed under addition for many noncommutative rings R. In fact, it is hard to find any structural properties that this set possesses for a general ring R, making this observation noteworthy. (This example also illustrates that ring theorists must take particular care not to impose their usual mental images of ideals upon the notion of a partial ideal.) We now introduce a notion of prime partial ideal, which will provide a type of "spectrum." Definition 2.8. A partial ideal P of a partial k-algebra R is prime if P 6= R and whenever x ⊥ y in A, xy ∈ P implies that either x ∈ P or y ∈ P . Equivalently, a partial ideal P of R is prime if P ( R and for every commeasurable subalgebra C ⊆ R, P ∩ C is a prime ideal of C. The set of prime partial ideals of a (full) k-algebra R is denoted p-Spec(R). If R is a commutative k-algebra, then the prime partial ideals of R are precisely the prime ideals of R. Now the fact that Spec : CommRing → Set defines a (contravariant) functor depends on the fact prime ideals behave well under homomorphisms of commutative rings. It turns out that prime partial ideals behave just as well, provided that one uses the "correct" OBSTRUCTING EXTENSIONS OF SPEC 7 notion of a morphism of partial algebras. This is proved in Lemma 2.10 below. The following definition was given in [11, §2]. Definition 2.9. Let R and S be partial algebras over a commutative ring k. A morphism of partial algebras is a function f : R → S such that, for every λ ∈ k and all a, b ∈ R with a ⊥ b, • f (a) ⊥ f (b), • f (λa) = λf (a), • f (a + b) = f (a) + f (b), • f (ab) = f (a)f (b), • f (0) = 0 and f (1) = 1. (In other words, f preserves the commeasurability relation and its restriction to every com- measurable subalgebra C ⊆ R is a homomorphism of commutative k-algebras f C : C → f (C).) Of course, any algebra homomorphism R → S of k-algebras is also a morphism of partial algebras when R and S are considered as partial algebras. Lemma 2.10. Let f : R → S be a morphism of partial k-algebras, and let I be a partial ideal of S. (1) The set f −1(I) ⊆ R is a partial ideal of R. (2) If I is prime, then f −1(I) is also prime. This holds, in particular, when R and S are (full) algebras, f is a k-algebra homomorphism, and I is a (prime) partial ideal of S. Proof. Let a, b ∈ R be such that a ⊥ b. Then f (a) ⊥ f (b). If a, b ∈ f −1(I) then f (a), f (b) ∈ I. Thus f (a + b) = f (a) + f (b) ∈ I, so that a + b ∈ f −1(I). On the other hand if a ∈ R and b ∈ f −1(I), then f (a) ∈ S and f (b) ∈ I. This means that f (ab) = f (a)f (b) ∈ I, whence ab ∈ f −1(I). Thus f −1(I) is a partial ideal of R. Now suppose that I is prime. The fact that I 6= S implies that f −1(I) 6= R, thanks to Lemma 2.6. If a ⊥ b in R are such that ab ∈ f −1(I), then f (a) ⊥ f (b) and f (a)f (b) = f (ab) ∈ I. Because I is prime, either f (a) ∈ I or f (b) ∈ I. In other words, either a ∈ f −1(I) or b ∈ f −1(I). This proves that f −1(I) is prime. (cid:3) Definition 2.11. The rule assigning to each ring R the set p-Spec(R) of prime partial ideals of R, and to each ring homomorphism f : R → S the map of sets p-Spec(S) → p-Spec(R) P 7→ f −1(P ), is a contravariant functor from the category of rings to the category of sets. We denote this functor by p-Spec : Ring → Set, extending the notation introduced in Definition 2.8. Notice immediately that the restriction of p-Spec to CommRing is equal to Spec, and −1(Spec) defined earlier in this therefore the functor p-Spec gives an object of the category r section. Of course, this functor could be defined on the category of all partial algebras and partial algebra homomorphisms. But because our primary interest is in the category of rings, we have chosen to restrict our definition to that category. 8 MANUEL L. REYES Example 2.12. Recall that an ideal P ⊳ R is completely prime if R/P is a domain; that is, P 6= R and for a, b ∈ R, ab ∈ P implies that either a ∈ P or b ∈ P . Certainly every completely prime ideal of a ring is a prime partial ideal. Thus every domain has a prime partial ideal: its zero ideal. Recalling Proposition 2.7 we conclude that the zero ideal of a division ring D is its unique prime partial ideal, so that p-Spec(D) is a singleton. The universal property of p-Spec will be established in Theorem 2.15 below. In prepa- ration, we observe that a partial ideal of a ring is equivalent to a choice of ideal in every commutative subring. For a partial algebra R over a commutative ring k, we let Ck(R) denote the partially ordered set of all commeasurable subalgebras of R. (In case R is a ring, C (R) := CZ(R) is the poset of commutative subrings of R.) Recall that a subset S of a partially ordered set X is cofinal if for every x ∈ X there exists s ∈ S such that x ≤ s. Lemma 2.13. Each of the following data uniquely determines a partial ideal of a partial algebra R: (1) A rule I that associates to each commeasurable subalgebra C ⊆ R an ideal I(C) ⊳ C such that, if C ⊆ C ′ are commeasurable subalgebras of R, then I(C) = I(C ′) ∩ C; (2) A rule I that associates to each commeasurable subalgebra C ⊆ R an ideal I(C) ⊳ C such that, if C1 and C2 are commeasurable subalgebras of R, then I(C1) ∩ C2 = C1 ∩ I(C2); (3) For a cofinal set S of commeasurable subalgebras of R, a rule I that associates to each C ∈ S an ideal I(C)⊳C such that, if C1 and C2 are in S, then I(C1)∩C2 = C1∩I(C2); (4) A rule I that associates to each maximal commeasurable subalgebra C ⊆ R an ideal I(C) ⊳ C such that, if C1 and C2 are maximal commeasurable subalgebra of R, then I(C1) ∩ C2 = C1 ∩ I(C2). Proof. First notice that the rules described in (1) and (2) are equivalent. For if I satisfies (1), then for any C1, C2 ∈ Ck(R) we have I(C1) ∩ C2 = I(C1) ∩ (C1 ∩ C2) = I(C1 ∩ C2) = I(C2) ∩ (C1 ∩ C2) = I(C2) ∩ C1. Thus I satisfies (2). Conversely, if I satisfies (2) and if C, C ′ ∈ C (R) are such that C ⊆ C ′, then I(C) = I(C) ∩ C ′ = C ∩ I(C ′), proving that I satisfies (1). The equivalence of the rules described in (2)–(4) is straightforward to verify. To complete the proof, we show that the data described in (1) uniquely determines a partial ideal of R. Given a rule I as in (1), the set J = SC∈Ck(R) I(C) ⊆ R is certainly a partial ideal of R. Conversely, given a partial ideal J of R, the assignment I sending C 7→ I(C) := J ∩ C satisfies (1). Clearly these maps I 7→ J and J 7→ I are mutually inverse. (cid:3) OBSTRUCTING EXTENSIONS OF SPEC 9 A choice of a prime ideal in each commutative subring of a ring R can be viewed as an element of the product QC∈C (R) Spec(C). The above characterization (1) of partial ideals says that the prime partial ideals can be identified with those elements (PC)C∈C (R) of this product such that for every C, C ′ ∈ C (R) with C ⊆ C ′, one has PC ′ ∩ C = PC. This fact is used below. The last step before the main result of this section is to give an alternative description of p-Spec as a certain limit. We recall the "product-equalizer" construction of limits in the category of sets (see [14, V.2]). Let D : J → Set be a diagram (i.e., D is a functor and J is a small category). Then the limit of D can be formed explicitly as lim←− J (xj) ∈ Y j∈Obj(J) D(j)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) D =   with the morphisms lim←− D → D(j) defined for each j ∈ Obj(J) via projection. D(f )(xi) = xj for all i, j ∈ Obj(J) and all f : i → j in J  For a ring R, we view the partially ordered set C (R) defined above as a category by considering each element of C (R) as an object and each inclusion as a morphism. (The appropriate analogue of this category in the context of C ∗-algebras makes a key appearance in the definition of the Bohrification functor [8, Def. 4] of Heunen, Landsman, and Spitters.) The functor that is shown to be isomorphic to p-Spec in the following proposition is very close to one defined by van den Berg and Heunen in [2, Prop. 5]. , Proposition 2.14. The contravariant functor p-Spec : Ring → Set is isomorphic to the functor defined, for a given ring R, by R 7→ lim←− C∈C (R)op Spec(C). This isomorphism preserves the isomorphism of functors p-Spec CommRing ∼= Spec. Proof. For any ring R, we have the following isomorphisms of sets: lim←−C∈C (R)op Spec(C) =   =   (PC) ∈ Y C∈C (R) (PC) ∈ Y C∈C (R) ∼= p-Spec(R), Spec(C)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Spec(C)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)   for all inclusions i : C ֒→ C ′, Spec(i)(PC ′) = PC for all inclusions C ⊆ C ′, PC ′ ∩ C = PC   where the last isomorphism comes from Lemma 2.13 (and the discussion that followed). These isomorphisms are natural in R and thus provide an isomorphism of functors. (cid:3) We will now show that p-Spec is our desired "universal Spec functor." In fact, we prove a stronger result stating that it is universal among all contravariant functors Ring → Set whose restriction to CommRing has a natural transformation to Spec that is not necessarily an isomorphism. This is made precise below. Given functors K : A → B and S : A → C, we recall that the (right) Kan extension of S along K is a functor R : B → C along with a natural transformation ε : RK → S such 10 MANUEL L. REYES that for any other functor F : B → C with a natural transformation η : F K → S there is a unique natural transformation δ : F → R such that η = ǫ ◦ (δK). (The "composite" δK : F K → RK of a functor with a natural transformation is a common shorthand for the horizontal composite δ ◦ 1K of the identity natural transformation 1K : K → K with δ, so that δK(X) = δ(K(X)) : F K → RK for any X ∈ A; see [14, II.5] for information on horizontal composition.) When K : A → B is an inclusion of a subcategory A ⊆ B, notice that F K = F A is the restriction. In this case, the natural transformation δK : F K → RK is the induced natural transformation of the restricted functors δA : F A → RA. Theorem 2.15. The functor p-Spec : Ringop → Set, along with the identity natural transfor- mation p-Spec CommRingop → Spec, is the Kan extension of the functor Spec : CommRingop → Set along the embedding CommRingop ⊆ Ringop. In particular, p-Spec is a terminal object in the category r −1(Spec). Proof. Let F : Ring → Set be a contravariant functor with a fixed natural transformation η : F CommRing → Spec. We need to show that there is a unique natural transformation δ : F → p-Spec that induces η upon restriction to CommRing ⊆ Ring. To construct δ, fix a ring R. For every commutative subring C of R, the inclusion C ⊆ R gives a morphism of sets F (R) → F (C) and η provides a morphism ηC : F (C) → Spec(C); these compose to give morphisms F (R) → Spec(C). By naturality of the morphisms involved, these maps out of F (R) collectively form a cone over the diagram obtained by applying Spec to the diagram C (R) of commutative subrings of R. By the universal property of the limit, there exists a unique dotted arrow making the square below commute for all C ∈ C (R): δR F (R) lim←− C∈C (R) Spec(C) ∼ p-Spec(R) ηC F (C) Spec(C). These morphisms δR form the components of a natural transformation δ : F → p-Spec. By construction, δ induces η when restricted to CommRing. Uniqueness of δ is guaranteed by the uniqueness of dotted arrow used to define δR above. The second sentence of the theorem follows from the first by applying the universal prop- erty of the Kan extension in the special case where F : Ringop → Set is a functor with a natural transformation η : F CommRingop → Spec that is an isomorphism. (cid:3) 3. Morphisms of partial algebras and the Kochen-Specker Theorem Having defined our universal functor p-Spec extending Spec, we must now determine its value on the algebra Mn(C). The first result of this section establishes a connection between the partial prime ideals of this algebra and certain morphisms of partial algebras. We recall a relevant fact from commutative algebra. Let C be a finite dimensional com- mutative algebra over an algebraically closed field k. Such an algebra is artinian, so all of OBSTRUCTING EXTENSIONS OF SPEC 11 its prime ideals are maximal. Given a maximal ideal m ⊆ C, the factor k-algebra C/m is a finite dimensional field extension of the algebraically closed field k and thus is is isomorphic to k. Hence Spec(C) is in bijection with the set of k-algebra homomorphisms C → k. This situation is generalized below. Proposition 3.1. Let R be partial algebra over an algebraically closed field k such that every element of R is algebraic over k (e.g., R is a finite dimensional k-algebra). Then there is a bijection between the set p-Spec(R) and the set of all morphisms of partial k-algebras f : R → k, which associates to each such morphism f the inverse image f −1(0). Proof. Because R consists of algebraic elements, every element of R generates a finite dimen- sional commeasurable subalgebra. In other words, R is the union of its finite dimensional commeasurable subalgebras. Given a morphism f : R → k of partial k-algebras, the set Pf := f −1(0) ⊆ R is a prime partial ideal of R according to Lemma 2.10. Furthermore, for each finite dimensional com- measurable subalgebra C ⊆ R, the prime ideal C ∩ Pf ⊳ C is maximal. Thus the restriction f C must be equal to the canonical homomorphism C ։ C/(Pf ∩ C) ∼−→ k. Conversely, suppose that P ⊆ R is a prime partial ideal. We define a function f : R → k as follows. As before, for each finite dimensional commeasurable subalgebra C ⊆ R the prime ideal P ∩ C of C is a maximal ideal. Thus we may define gC : C → k via the quotient ∼−→ k. Notice that for finite dimensional commeasurable subalgebras map C ։ C/(P ∩ C) C ⊆ C ′, the following diagram commutes: C C ′ / C/(P ∩ C) ∼ / k / C ′/(P ∩ C ′) ∼ / k. Thus there is a well-defined function fP : R → k given, for any r ∈ R, by fP (r) = gC(r) for any finite dimensional commeasurable subalgebra C of R containing r (such as C = k[r] ⊆ R). It is clear from the construction of fP that f −1 P (0) = P . We have defined maps P 7→ fP and f 7→ Pf . The last sentences of the previous two (cid:3) paragraphs show that these assignments are mutually inverse, completing the proof. Thus the proof of Theorem 1.1 is reduced to understanding the morphisms of partial C-algebras Mn(C) → C. The Kochen-Specker Theorem provides just the information that we need. This theorem, due to S. Kochen and E. Specker [11], is a "no-go theorem" from quantum mechanics that rules out the existence of certain types of hidden variable theories. Probability is an inherent feature in the mathematical formulation of quantum physics; only the evolution of the probability amplitude of a system is computed. A hidden variable theory is, roughly speaking, a theory devised to explain quantum mechanics by predicting outcomes of all measurements with certainty. The observable quantities of a quantum system are mathematically represented by self- adjoint operators in a C ∗-algebra. The Heisenberg Uncertainty Principle implies that if two such operators do not commute, then the exact values of the corresponding observables cannot be simultaneously determined. On the other hand, commuting observables have no _ / / / / / / / / 12 MANUEL L. REYES uncertainty restriction imposed upon them by Heisenberg's principle. In [11] Kochen and Specker argued that a hidden variable theory should assign a real value to each observable of a quantum system in such a way that values of the sum or product of commuting observables is equal to the sum or product of their corresponding values. That is to say, Kochen and Specker's notion of a hidden variable theory is a morphism of partial R-algebras from the partial algebra of observables to R. With this motivation, Kochen and Specker showed that no such morphism exists. Kochen-Specker Theorem 3.2. Let n ≥ 3, and for A := Mn(C) let Asa ⊆ A denote the subset of self-adjoint elements of A. There does not exist a morphism of partial R-algebras f : Asa → R. Actually, [11] establishes this result for n = 3, but it is often cited in the literature for n ≥ 3. Because the reduction to the case n = 3 is straightforward, we include it below. Proof for n > 3. We assume that the result holds for n = 3, as proved in [11]. Let n > 3, and assume for contradiction that there is a morphism of partial algebras f : Asa → R. Let Pi = Eii ∈ Asa be the orthogonal projection onto the ith basis vector. Then P Pi = I In particular, because f is a morphism of partial algebras we have and PiPj = δijPi. P f (Pi) = f (P Pi) = 1. Furthermore, each f (Pi) = f (P 2 i ) = f (Pi)2 must equal either 0 or 1. So the values f (Pi) are all equal to 0, except for one Pj with f (Pj) = 1. Choose two of the other projections Pi to get a set of three distinct projections Pj, Pk, and Pℓ. Then E := Pj + Pk + Pℓ is an orthogonal projection, so there is an isomorphism of the corner algebra EAE ∼= M3(C) that preserves self-adjoint elements. Now the restriction of f to (EAE)sa = EAE ∩ Asa satisfies all properties of a morphism of partial R-algebras, except possibly the preservation of the multiplicative identity. But the multiplicative identity of (EAE)sa is E and f (E) = f (Pj) + f (Pk) + f (Pℓ) = 1, proving that f is a morphism of partial algebras. This contradicts the Kochen-Specker Theorem in dimension 3. (cid:3) In Corollary 3.4 below, we will establish an analogue of the Kochen-Specker Theorem that is more suitable for our purposes. First we require one preparatory result. Given an element x of a partial ring R, we will say that another element y ∈ R is an inverse of x if x ⊥ y and xy = 1. (Such an element need not be unique! An example of an element with two inverses is easily constructed by taking two copies of a Laurent polynomial ring k[x1, x−1 1 ], k[x2, x−1 2 ], "gluing" them by identifying k[x1] with k[x2], and declaring x−1 i ⊥ x1 = x2 but with the x−1 i not commeasurable to one another. An inverse y of x is unique if y is commeasurable to all elements of R that are commeasurable to x. We thank George Bergman for these observations.) The following argument is a standard one. It basically appeared in [11, pp.81–82], and it even has roots in the theory of the Gelfand spectrum of C ∗-algebras. Lemma 3.3. Let R be a partial algebra over a commutative ring k 6= 0, and let f : R → k be a morphism of partial k-algebras. Then for any r ∈ R, the element r − f (r) ∈ R does not have an inverse. In particular, if k is a field and R = Mn(k), then f (r) ∈ k is an eigenvalue of r. Proof. If r − f (r) has an inverse u ∈ R, then k = 0 by the following equation: OBSTRUCTING EXTENSIONS OF SPEC 13 1 = f (1) = f ((r − f (r))u) = f (r − f (r))f (u) = (f (r) − f (f (r) · 1))f (u) = (f (r) − f (r))f (u) = 0. (cid:3) We now have the following reformulation of the Kochen-Specker Theorem that is more appropriate to our needs. (One could think of it as a "complex-valued," rather than "real- valued," Kochen-Specker Theorem.) Together with Proposition 3.1, this constitutes the final "key result" used in the proof of Theorem 1.1. Corollary 3.4. For any n ≥ 3, there is no morphism of partial C-algebras Mn(C) → C. Proof. Let A = Mn(C). Every self-adjoint matrix in A has real eigenvalues, so Lemma 3.3 implies that a morphism of partial C-algebras A → C restricts to a morphism of partial R- algebras Asa → R. But such morphisms are forbidden by the Kochen-Specker Theorem 3.2. (cid:3) It is natural to ask what is the status of Corollary 3.4 in the case n = 2. Regarding their original theorem, Kochen and Specker demonstrated the existence of a morphism of partial R-algebras M2(C)sa → R in [11, §6], showing that the Kochen-Specker Theorem does not extend to n = 2. Similarly, Corollary 3.4 does not extend to n = 2. There exist morphisms of partial algebras M2(C) → C, and we can describe all of them as follows. Incidentally, this result also shows that the statement of Theorem 1.1 is not valid in the case n = 2; the functor F = p-Spec assigns a nonempty set (of cardinality 22ℵ0 , in fact!) to M2(C). Proposition 3.5. Let k be an algebraically closed field, and let I ⊆ A := M2(k) be any set of idempotents such that the set of all idempotents of A is partitioned as {0, 1} ⊔ I ⊔ {1 − e : e ∈ I}. Then for every function α : I → {0, 1} there is a morphism of partial k-algebras fα : A → k such that the restriction of f to I is α : I → {0, 1} ⊆ k. Moreover, there are bijective correspondences between: • the set of functions α : I → {0, 1}; • the set of morphisms of partial k-algebras A → k; and • the set of prime partial ideals of A; given by α ↔ fα ↔ f −1 α (0). Proof. First we construct a commutative k-algebra B with a morphism of partial k-algebras h : A → B. Let N be a set of nonzero nilpotent elements of A such that every nonzero nilpotent matrix in A has exactly one scalar multiple in N . Let B be the commutative k-algebra B := k[xe, xn : e ∈ I, n ∈ N ] with relations x2 n = 0 for n ∈ N . e = xe for e ∈ I and x2 14 MANUEL L. REYES A result of Schur [18] (also proved more generally by Jacobson [9, Thm. 1]) implies that ev- ery maximal commutative subalgebra of A is has k-dimension 2. Thus the intersection of two distinct commutative subalgebras of A is the scalar subalgebra k ⊆ A. This makes it easy to see that a function h : A → B is a morphism of partial k-algebras if and only if its restriction to every 2-dimensional commutative subalgebra of A is a k-algebra homomorphism. Now define a function h : A → B as follows. For each scalar λ ∈ k ⊆ A, we set h(λ) = λ ∈ k ⊆ B. Now assume a ∈ A \ k. Then k[a] is a 2-dimensional commutative subalgebra of A. Because the only 2-dimensional algebras over the algebraically closed field k are k × k and k[ε]/(ε2), there exists b ∈ I ⊔ N such that k[a] = k[b]. The careful choice of the sets I and N ensures that this b is unique. Thus it suffices to define h on each k[b]. But for b ∈ I ⊔ N , the map k[b] → B defined by sending b 7→ xb is clearly a homomorphism of k-algebras. We define the restriction of h to k[a] = k[b] to be this homomorphism, which in particular defines the value h(a). Certainly h is well-defined, and it is a morphism of partial algebras because its restriction to every 2-dimensional subalgebra is an algebra homomorphism. Thus we have constructed a morphism of partial algebras h : A → B. Given a function α : I → {0, 1}, there exists a k-algebra homomorphism gα : B → k given by sending xe 7→ α(e) ∈ k for e ∈ I and xn 7→ 0 for n ∈ N . So the composite fα := gα ◦ h is a morphism of partial k-algebras whose restriction to I is equal to α. The bijection between the three sets in the statement of the proposition follows directly from Proposition 3.1 above and Lemma 3.6 below. (cid:3) Lemma 3.6. Let R be a partial algebra over an algebraically closed field k in which every el- ement is algebraic (e.g., R is a finite dimensional k-algebra). A morphism of partial algebras R → k is uniquely determined by its restriction to the set of idempotents of R. Proof. Let f : R → k be a morphism of partial k-algebras, and let C ⊆ R be a finite dimensional commeasurable subalgebra of R. Because R is the union of its finite dimensional commeasurable subalgebras, it is enough to show that the restriction of f to C, which is a k-algebra homomorphism C → k, is uniquely determined by its values on the idempotents of C. Because C is finite dimensional it is artinian and thus is a finite direct sum of local k- algebras. Write C = A1 ⊕ · · · ⊕ An where each (Ai, Mi) is local and the identity element of Ai is ei, an idempotent of C. Since k is algebraically closed, each of the residue fields Ai/Mi is isomorphic to k as a k-algebra. Thus each Ai = kei ⊕ Mi. Because Ai is finite dimensional, its Jacobson radical Mi is nilpotent and hence is in the kernel of f C. It now follows easily that the restriction of f to C is determined by the values f (ei). (cid:3) 4. Proof and consequences of the main result We are now prepared to prove Theorem 1.1, the main ring-theoretic result of the paper. Proof of Theorem 1.1. Fix n ≥ 3 and let A = Mn(C)). According to Theorem 2.15 there exists a natural transformation F → p-Spec. By Proposition 3.1, p-Spec(A) is in bijection with the set of morphisms of partial C-algebras A → C. No such morphisms exist according to Corollary 3.4 of the Kochen-Specker Theorem, so p-Spec(A) = ∅. The existence of a function F (A) → p-Spec(A) = ∅ now implies that F (A) = ∅. (cid:3) OBSTRUCTING EXTENSIONS OF SPEC 15 It seems appropriate to mention some partial positive results that contrast with The- orem 1.1. One might hope that restricting to certain well-behaved ring homomorphisms could allow the functor Spec to be "partially extended." In this vein, Procesi has shown [15, Lem. 2.2] that if f : R → S is a ring homomorphism such that S is generated over f (R) by elements centralizing f (R), then for every prime ideal Q ⊳ S the inverse image f −1(Q) is again prime. Furthermore, he showed in [15, Thm. 3.3] that if R is a Jacobson PI ring and f : R → S is a ring homomorphism such that S is generated by the image f (R) and finitely many elements that centralize f (R), then for every maximal ideal M ⊳ S the inverse image f −1(M) is a maximal ideal of R. (Although he only stated these results for injective f , they are easily seen to hold more generally.) On the other hand, one may try to replace functions between prime spectra by "multi- valued functions," which may send a single element of one set to many elements of another set. For instance, one might consider a functor that maps each homomorphism R → S of noncommutative rings to a correspondence Spec(S) → Spec(R), which sends a single prime ideal of S to some nonempty finite set of prime ideals of R. This notion was introduced by Artin and Schelter in [1, §4] and studied in further detail by Letzter in [13]. There is an appropriate notion of "continuity" of a correspondence, and it is shown in [13, Cor. 2.3] (see also [1, Prop. 4.6]) that if f : R → S is a ring homomorphism and S is a PI ring, then the associated correspondence is continuous. However, there exist homomorphisms between noetherian rings whose correspondence is not continuous [13, §2.5]. We now present a few corollaries of Theorem 1.1. The first is a straightforward general- ization of that theorem replacing Mn(C) with Mn(R) where R is any ring containing a field isomorphic to C. Corollary 4.1. Let F : Ring → Set be a contravariant functor whose restriction to the full subcategory of commutative rings is isomorphic to Spec. If R is any ring with a homomor- phism C → R, then F (Mn(R)) = ∅ for n ≥ 3. Proof. The homomorphism C → R induces a homomorphism Mn(C) → Mn(R). Thus we have a set map F (Mn(R)) → F (Mn(C)). If n ≥ 3 then by Theorem 1.1 the latter set is empty; hence the former set must also be empty. (cid:3) In the corollary above, R can be any complex algebra. But rings that contain C as a non-central subring, such as the real quaternions, are also allowed. On the other hand, suppose that R is a complex algebra such that R ∼= Mn(R) for some n ≥ 2. It follows that R ∼= Mn(R) ∼= Mn(Mn(R)) ∼= Mn2(R), so we may assume that n ≥ 4. Then the corollary implies that for functors F as above, F (R) ∼= F (Mn(R)) = ∅. For instance, if V is an infinite dimensional C-vector space and R is the algebra of C-linear endomorphisms of V , then the existence of a vector space isomorphism V ∼= V ⊕n (any n ≥ 2) implies the existence of an algebra isomorphism R ∼= Mn(R). An attempt to extend the ideas above suggests one possible algebraic generalization of the Kochen-Specker Theorem. Suppose that p-Spec(Mn(Z)) = ∅ for some integer n ≥ 3. For any ring R the canonical ring homomorphism Z → R induces a morphism Mn(Z) → Mn(R). Then one would have p-Spec(Mn(R)) = ∅. It would follow that any contravariant functor 16 MANUEL L. REYES F : Ring → Set whose restriction to CommRing is isomorphic to Spec must assign the empty set to Mn(R) for any ring R. This highlights the importance of the following question. Question 4.2. Do there exist integers n ≥ 3 such that p-Spec(Mn(Z)) = ∅? If p-Spec(Mn(Z)) were in fact empty for all sufficiently large values of n, then this would be a sort of "integer-valued" Kochen-Specker Theorem. The next corollary of Theorem 1.1 concerns certain functors sending rings to commutative rings. Consider the functor Ring → CommRing that sends each ring R to its "abelianization" R/[R, R]. Rings whose abelianization is zero are easy to find, and this functor necessarily destroys all information about these rings. One could try to abstract this functor by con- sidering any functor Ring → CommRing whose restriction to CommRing is isomorphic to the identity functor. The following result says that every such "abstract abelianization functor" necessarily destroys matrix algebras. Corollary 4.3. Let α : Ring → CommRing be a functor such that the restriction of α to CommRing is isomorphic to the identity functor. Then for any ring R with a homomorphism C → R and any n ≥ 3, one has α(Mn(R)) = 0. In particular, α is not faithful. Proof. Because α restricts to the identity functor on CommRing, the contravariant functor ∼= Spec. For n ≥ 3, Corollary 4.1 implies that F := Spec ◦α : Ring → Set satisfies F CommRing Spec(α(Mn(R)) = F (Mn(R)) = ∅. Hence the commutative ring α(Mn(R)) is zero. To see that α is not faithful, fix n ≥ 3 and consider that α induces a function HomRing(Mn(C), Mn(C)) → HomCommRing(α(Mn(C)), α(Mn(C)) = HomCommRing(0, 0). The latter set is a singleton, while the former set is not a singleton (because Mn(C) has nontrivial inner automorphisms). So the function above is not injective, proving that α is not faithful. (cid:3) Interestingly, this result does not hold in the case n = 2; we thank George Bergman for this observation. Let α : Ring → CommRing be the functor sending each ring to the colimit of the diagram of its commutative subrings. Certainly αCommRing is isomorphic to the identity functor on CommRing. One can check that for an algebraically closed field k, the commutative ring α(M2(k)) is isomorphic to the algebra B constructed in the proof of Proposition 3.5; in particular, α(M2(k)) 6= 0. (At the very least, it is not hard to verify from the universal property of the colimit that there exists a homomorphism α(M2(k)) → B, confirming that α(M2(k)) 6= 0.) Furthermore, one can show that this functor is initial among all "abstract abelianization functors," but the details will not be presented here. The final corollary of Theorem 1.1 to be presented in this section is a rigorous proof that the rule that assigns to each (not necessarily commutative) ring R the set Spec(R) of prime ideals of R is "not functorial." (Recall that an ideal P ⊳ R is prime if, for all ideals I, J ⊳ R, IJ ⊆ P implies that either I ⊆ P or J ⊆ P .) The fact that this assignment "is not a functor" seems to be common wisdom. (Specific mention of this idea in the literature is not widespread, but see [20, pp. 1 and 36] or [13, §1] for examples.) It is easy to verify that this assignment is not a functor in the natural way; that is, if f : R → S a ring homomorphism and P ⊳ S is prime, one can readily see that the ideal f −1(P ) ⊳ R need not be prime. OBSTRUCTING EXTENSIONS OF SPEC 17 However, we are unaware of any rigorous statement or proof in the literature of the precise result below. Corollary 4.4. There is no contravariant functor F : Ring → Set whose restriction to the full subcategory CommRing is isomorphic to Spec and such that, for every ring R, the set F (R) is in bijection with the set of prime ideals of R. Proof. Assume for contradiction that such F exists. Fix n ≥ 3. Because the zero ideal of Mn(C) is (its unique) prime, the assumption on F implies F (Mn(C)) 6= ∅, violating Theorem 1.1. (cid:3) This corollary can also be derived from an elementary argument that avoids using Theo- rem 1.1. In fact, the statement can even be strengthened as follows. Proposition 4.5. There is no contravariant functor F : Ring → Set whose restriction to CommRing is isomorphic to Spec and such that F satisfies either of the following conditions: (1) For some field k and some integer n ≥ 2, the set F (Mn(k)) is a singleton; (2) F is Morita invariant in the following sense: for any Morita equivalent rings R and S, one has F (R) ∼= F (S). Proof. First notice that if F satisfies condition (2) above, then it satisfies condition (1) because Mn(k) is Morita equivalent to k, which would mean that F (Mn(k)) ∼= F (k) ∼= Spec(k) is a singleton. So assume for contradiction that there exists a functor F as above satisfying (1). Fix k and n as in condition (1). Define π := (1 2 · · · n) ∈ Sn, a permutation of the set {1, 2, . . . , n}. Let ρ be the automorphism of kn given by (ai) 7→ (aπ(i)), let P ∈ Mn(k) be the permutation matrix whose ith row is the π(i)th standard basis row vector, and let σ be the inner automorphism of Mn(k) given by σ(A) = P AP −1. For the final piece of notation, let ι : kn ֒→ Mn(k) be the diagonal embedding. The following equality of algebra homomorphisms kn → Mn(k) is elementary: ι ◦ ρ = σ ◦ ι. Applying the contravariant functor F to this equation gives F (ρ) ◦ F (ι) = F (ι) ◦ F (σ). By hypothesis the set F (Mn(k)) is a singleton. Hence the automorphism F (σ) of F (Mn(k)) is the identity. It follows that (4.6) On the other hand F (kn) ∼= Spec(kn) = {1, . . . , n}, and under this isomorphism F (ρ) acts as Spec(ρ) = π−1 which has no fixed points. Thus the image of the unique element of F (Mn(k)) under F (ι) is distinct from its image under F (ρ) ◦ F (ι), contradicting (4.6) above. (cid:3) F (ρ) ◦ F (ι) = F (ι). Because the set of prime ideals of a noncommutative ring is Morita invariant (for instance, see [12, (18.45)]) the proposition above implies Corollary 4.4. Notice that Proposition 4.5 with k = C and n = 2 cannot be derived from Theorem 1.1 because that theorem does not apply to the algebra M2(C), as explicitly shown in Proposition 3.5. Of course, there are many important examples of invariants of rings extending Spec of a commutative ring that respect Morita equivalence, aside from the set of prime two-sided ideals of a ring. Two examples are the prime torsion theories introduced by O. Goldman 18 MANUEL L. REYES in [7] and the spectrum of an abelian category defined by A. Rosenberg in [17]. (Incidentally, both of these spectra arise from the theory of noncommutative localization.) Each of these invariants is certainly useful in the study of noncommutative algebra, and they have appeared in different approaches to noncommutative algebraic geometry. Thus we emphasize that Proposition 4.5 does not in any way suggest that such invariants should be avoided. It simply reveals that we cannot hope for such invariants to be functors to Set. 5. The analogous result for C ∗-algebras In this section we will prove Theorem 1.2, which obstructs extensions of the Gelfand spectrum functor to noncommutative C ∗-algebras. We begin by reviewing some facts and setting some conventions about the category of C ∗-algebras. (Many of these basics can be found in [5, §I.5] and [10, §4.1].) We emphasize that all C ∗-algebras considered in this section are assumed to be unital. Let CAlg denote the category whose objects are unital C ∗-algebras and whose morphisms are identity-preserving ∗-homomorphisms. Such morphisms do not increase the norm and are norm-continuous. The only topology on a C ∗-algebra to which we will refer is the norm topology. A closed ideal of a C ∗-algebra is always ∗-invariant, and the resulting factor algebra is a C ∗-algebra. A C ∗-subalgebra of a C ∗-algebra A is a closed subalgebra C ⊆ A that is invariant under the involution of A; such a subalgebra inherits the structure of a Banach algebra with involution from A and is itself a C ∗-algebra with respect to this inherited structure. If f : A → B is a ∗-homomorphism, then the image f (A) ⊆ B is always a C ∗-subalgebra. The full subcategory of CAlg consisting of commutative unital C ∗-algebras is denoted by CommCAlg. Finally, the reader may wish to see Section 1 for the definition of the (contravariant) Gelfand spectrum functor Max : CommCAlg → Set. As in the ring-theoretic case, we define an appropriate category of functors in which we seek a universal functor. The inclusion of categories CommCAlg ֒→ CAlg induces a restriction functor between functor categories r : Fun(CAlg op , Set) → Fun(CommCAlg op , Set) F 7→ F CommCAlg op. Again we define the "fiber category" over Max ∈ Fun(CommCAlg , Set) to be the category −1(Max) of pairs (F, φ) where F ∈ Fun(CAlg −→ Max is an isomor- −1(Max) is a natural transformation phism of functors; a morphism ψ : (F, φ) → (F ′, φ′) in r ψ : F → F ′ such that φ′ ◦ r(ψ) = φ. Our first goal is to locate a final object of the category −1(Max), which we view as a "universal noncommutative Gelfand spectrum functor." , Set) and φ : r(F ) op r r op ∼ First we define the analogue of the spectrum p-Spec of prime partial ideals. To do so, it will be useful to think in terms of partial C ∗-algebras, as defined by van den Berg and Heunen in [2, §4]. Definition 5.1. A partial C ∗-algebra P is a partial C-algebra with an involution ∗ : P → P and a function k · k : P → R such that any set S ⊆ P of pairwise commeasurable elements is contained in a set T ⊆ P such that the restricted operations of P endows T with the OBSTRUCTING EXTENSIONS OF SPEC 19 structure of a commutative C ∗-algebra. Such a subset T ⊆ P is called a commeasurable C ∗- subalgebra of P . A ∗-morphism of partial C ∗-algebras f : P → Q is a morphism of partial C-algebras satisfying f (a∗) = f (a)∗ for all a ∈ P . It is very important to note that, unlike the ring-theoretic case, a C ∗-algebra with the commeasurability relation of commutativity is generally not a partial C ∗-algebra. What is true is that for any C ∗-algebra A, the set N(A) = {a ∈ A : aa∗ = a∗a} of normal elements (with commutativity as commeasurability) is always a partial C ∗-algebra. This makes use of the fact that any normal element of a C ∗-algebra has the property that its centralizer is a ∗-subalgebra; see [6]. The assignment A 7→ N(A) defines a functor from the category of C ∗-algebras to the category of partial C ∗-algebras, defined on a morphism f : A → B by restricting and corestricting f to N(f ) : N(A) → N(B); see [2, Prop. 3]. (Because of this subtlety regarding normal elements, we will typically use P, Q to denote partial C ∗-algebras and A, B to denote full C ∗-algebras.) Definition 5.2. A partial closed ideal of a partial C ∗-algebra P is a subset I ⊆ P such that, for every commeasurable C ∗-subalgebra C ⊆ P , the intersection I ∩ C is a closed ideal of C. If, for every commeasurable C ∗-subalgebra C one has that I ∩ C is a maximal ideal of C, then I is a partial maximal ideal of N. Let A be a C ∗-algebra. We say that a subset I ⊆ A is a partial closed (resp. maximal) ideal of A if I ⊆ N(A) and I is a partial closed (resp. maximal) ideal of the partial C ∗-algebra N(A). Because we require a partial closed ideal I of a C ∗-algebra A to consist of normal elements, I is completely determined by its intersection with all commutative subalgebras in the sense that I = SC(I ∩ C), where C ranges over all commutative C ∗-subalgebras of A. This is true because an element of A is normal if and only if it is contained in a commutative C ∗-subalgebra of A. As in the ring-theoretic case, partial closed ideals behave well under homomorphisms. Lemma 5.3. Let f : P → Q be a ∗-homomorphism of C ∗-algebras, and let I be a partial closed (resp. maximal) ideal of Q. The set f −1(I) ⊆ P is a partial closed (resp. maximal) partial ideal of P . In particular, if f : A → B is a ∗-homomorphism between C ∗-algebras and I ⊆ N(B) ⊆ B is a partial closed (resp. maximal) ideal, then f −1(I) ∩ N(A) ⊆ A is a partial closed (resp. maximal) ideal of A. Proof. Let C ⊆ P be a commeasurable C ∗-subalgebra. We wish to show that f −1(I) ∩ C is a closed (resp. maximal) ideal of C. First notice that since C consists of pairwise commeasurable elements, so does f (C) ⊆ Q. Thus there is a commeasurable C ∗-subalgebra D ⊆ Q such that f (C) ⊆ D. But then since f (co)restricts to a ∗-homomorphism of (full) C ∗-algebras C → D, its image f (C) ⊆ D is a C ∗-subalgebra. It follows that f (co)restricts to a ∗-homomorphism of (full, commutative) C ∗-algebras f C : C → f (C). Now I ∩ f (C) is a closed (resp. maximal) ideal in f (C) by hypothesis, so its preimage under f C is a closed (resp. maximal) ideal of C. On the other hand, (f C)−1(I ∩ f (C)) is easily seen to be equal to f −1(I) ∩ C. Hence the latter is a closed (resp. maximal) ideal, as desired. (cid:3) 20 MANUEL L. REYES This allows us to define a "partial Gelfand spectrum" functor. Definition 5.4. We define a contravariant functor p-Max : CAlg → Set by assigning to every C ∗-algebra A the set p-Max(A) of partial maximal ideals of A, and by assigning to each morphism f : A → B in CAlg the function p-Max(B) → p-Max(A) M 7→ f −1(M) ∩ N(A). (The only potential hindrance to functoriality is the preservation of composition of mor- phisms, but this is easily verified. Alternatively, this is seen to be a functor because it is the composite of the functor from C ∗-algebras to partial C ∗-algebras A 7→ N(A) with the contravariant functor from partial C ∗-algebras to sets that sends a partial algebra to the set of its partial maximal ideals.) Notice that the restriction of p-Max to CommCAlg is equal to the Gelfand spectrum functor Max, so that p-Max is an object of the category r −1(Max). As in Proposition 2.14, the functor p-Max can be recovered through a limit construction. For a C ∗-algebra A, we let C ∗(A) denote the partially ordered set of its commutative C ∗- subalgebras, viewed as a category in the usual way. Proposition 5.5. The contravariant functor p-Max : CAlg → Set is isomorphic to the functor defined, for a given C ∗-algebra A, by A 7→ lim←− C∈C ∗(A)op Max(C). This isomorphism preserves the isomorphism of functors p-Max CommCAlg = Spec. We will not include a proof of this fact, but we will mention the main subtlety. The appropriate analogue of Lemma 2.13 (replacing each occurrence of "commeasurable subal- gebra" with "commeasurable C ∗-subalgebra") still holds, and is used as before to prove the present result. Here it is crucial to recall that a partial closed ideal I ⊆ A consists of nor- mal elements, for this ensures that I is determined by its intersection with all commutative C ∗-subalgebras of A. Just as before, this allows one to show that p-Max is a "universal Gelfand spectrum functor." Theorem 5.6. The functor p-Max : CAlg p-Max CommCAlg → Max, is the Kan extension of the functor Max : CommCAlg along the embedding CommCAlg the category r op → Set, with the identity natural transformation op → Set . In particular, p-Max is a terminal object in −1(Max). op ⊆ CAlg op Our next goal is to connect the functor p-Max to the Kochen-Specker Theorem, in a manner similar to that of Section 3. We have the following analogue of Proposition 3.1. Its proof is completely analogous, and thus is omitted. OBSTRUCTING EXTENSIONS OF SPEC 21 Proposition 5.7. Let P be a partial C ∗-algebra. There is a bijection between the set of partial maximal ideals of P and the set of ∗-morphisms of partial C ∗-algebras f : P → C, which associates to each such morphism f the inverse image f −1(0). In particular, if A is a C ∗-algebra then there is a bijection between p-Max(A) and the set of ∗-morphisms of partial C ∗-algebras f : N(A) → C. We have effectively reduced the proof of Theorem 1.2 to a question of the existence of morphisms of partial C ∗-algebras. Thus we are in a position to apply the Kochen-Specker Theorem. The following corollary to Kochen-Specker is proved just like Corollary 3.4, relying upon Lemma 3.3. Corollary 5.8 (A corollary of the Kochen-Specker Theorem). For any n ≥ 3, there is no ∗-morphism of partial C ∗-algebras N(Mn(C)) → C. We are now ready to give a proof of Theorem 1.2, obstructing extensions of the Gelfand spectrum functor. Proof of Theorem 1.2. Fix n ≥ 3 and let A = Mn(C) ∈ CAlg. By Theorem 5.6 there is a natural transformation F → p-Max. The set p-Max(A) is in bijection with the set of ∗- morphisms of partial C ∗-algebras N(A) → C according to Proposition 5.7. By Corollary 5.8 of the Kochen-Specker Theorem there are no such ∗-morphisms. Thus p-Max(A) = ∅, and the existence of a function F (A) → p-Max(A) gives F (A) = ∅. (cid:3) The corollaries to Theorem 1.1 given in Section 4 all have analogues in the setting of C ∗-algebras. For the most part we will omit the proofs of these results because they are such straightforward adaptations of those given in Section 4. First we provide an analogue of Corollary 4.1, and we include its proof only to illustrate how our restriction to unital C ∗-algebras comes into play. Corollary 5.9. Let F : CAlg → Set be a contravariant functor whose restriction to the full subcategory of commutative C ∗-algebras is isomorphic to Max. Then for any C ∗-algebra A and integer n ≥ 3, one has F (Mn(A)) = ∅. Proof. Because A is unital, there is a canonical morphism of C ∗-algebras C → C · 1A ⊆ A. This induces a ∗-morphism Mn(C) → Mn(A). Thus there is a function of sets F (Mn(A)) → F (Mn(C)), and the latter set is empty by Theorem 1.2. Hence F (Mn(A)) = ∅. (cid:3) As in the discussion following Corollary 4.1, this result shows that if A is a unital C ∗- algebra for which there is an isomorphism A ∼= Mn(A) for some n ≥ 2, then for any functor F as above, F (A) = ∅. As an example, we may take A to be the algebra of bounded operators on an infinite-dimensional Hilbert space. Next is the appropriate analogue of Corollary 4.3. Corollary 5.10. Let α : CAlg → CommCAlg be a functor whose restriction to CommCAlg is isomorphic to the identity functor. Then for every C ∗-algebra A and every n ≥ 3, one has α(Mn(A)) = 0 Finally, there is the following analogue of Corollary 4.4. 22 MANUEL L. REYES Corollary 5.11. There is no contravariant functor F : CAlg → Set whose restriction to the full subcategory CommC*Alg is isomorphic to Max and such that, for every C ∗-algebra A, the set F (R) is in bijection with the set of primitive ideals of A. This corollary can be obtained as a consequence either of Theorem 1.2 or of the obvious analogue of Proposition 4.5. In fact, the proof of the latter proposition (with k = C) extends directly to the setting of C ∗-algebras because all of the homomorphisms used in its proof are actually ∗-homomorphisms. Acknowledgments I am grateful to Andre Kornell and Matthew Satriano for several stimulating conversations in the early stages of this work, George Bergman for many insightful comments, Lance Small for helpful references to the literature, and Theo Johnson-Freyd for advice on creating Tik z diagrams. Finally, I thank the referee for a number of suggestions and corrections that improved the readability of the paper. References 1. M. Artin and W. Schelter, Integral ring homomorphisms, Adv. in Math. 39 (1981), no. 3, 289–329. MR 614165 (83e:16015) 2. Benno van den Berg and Chris Heunen, Noncommutativity as a colimit, to appear in Appl. Categor. Struct. (2011), available at http://dx.doi.org/10.1007/s10485-011-9246-3. 3. , No-go theorems for functorial localic spectra of noncommutative rings, arXiv:1101.5924, preprint. 4. P. M. Cohn, The affine scheme of a general ring, Applications of sheaves (Proc. Res. Sympos. Appl. Sheaf Theory to Logic, Algebra and Anal., Univ. Durham, Durham, 1977), Lecture Notes in Math., vol. 753, Springer, Berlin, 1979, pp. 197–211. MR 555546 (81e:16002) 5. Kenneth R. Davidson, C ∗-Algebras by Example, Fields Institute Monographs, vol. 6, American Mathe- matical Society, Providence, RI, 1996. MR 1402012 (97i:46095) 6. Bent Fuglede, A commutativity theorem for normal operators, Proc. Nat. Acad. Sci. U. S. A. 36 (1950), 35–40. MR 0032944 (11,371c) 7. Oscar Goldman, Rings and modules of quotients, J. Algebra 13 (1969), 10–47. MR 0245608 (39 #6914) 8. Chris Heunen, Nicolaas P. Landsman, and Bas Spitters, A topos for algebraic quantum theory, Comm. Math. Phys. 291 (2009), no. 1, 63–110. MR 2530156 9. N. Jacobson, Schur's theorems on commutative matrices, Bull. Amer. Math. Soc. 50 (1944), 431–436. MR 0010540 (6,33b) 10. Richard V. Kadison and John R. Ringrose, Fundamentals of the Theory of Operator Algebras, Vol. I: Elementary Theory, Graduate Studies in Mathematics, vol. 15, American Mathematical Society, Provi- dence, RI, 1997, Reprint of the 1983 original. MR 1468229 (98f:46001a) 11. Simon Kochen and E. P. Specker, The problem of hidden variables in quantum mechanics, J. Math. Mech. 17 (1967), 59–87. MR 0219280 (36 #2363) 12. T. Y. Lam, Lectures on Modules and Rings, Graduate Texts in Mathematics, vol. 189, Springer-Verlag, New York, 1999. MR 1653294 (99i:16001) 13. Edward S. Letzter, On continuous and adjoint morphisms between non-commutative prime spectra, Proc. Edinb. Math. Soc. (2) 49 (2006), no. 2, 367–381. MR 2243792 (2007d:16003) 14. Saunders Mac Lane, Categories for the Working Mathematician, second ed., Graduate Texts in Mathe- matics, vol. 5, Springer-Verlag, New York, 1998. MR 1712872 (2001j:18001) 15. C. Procesi, Non commutative Jacobson-rings, Ann. Scuola Norm. Sup. Pisa (3) 21 (1967), 281–290. MR 0224652 (37 #251) OBSTRUCTING EXTENSIONS OF SPEC 23 16. Alexander L. Rosenberg, The left spectrum, the Levitzki radical, and noncommutative schemes, Proc. Nat. Acad. Sci. U.S.A. 87 (1990), no. 21, 8583–8586. MR 1076775 (92d:14001) 17. , Noncommutative local algebra, Geom. Funct. Anal. 4 (1994), no. 5, 545–585. MR 1296568 (95m:14003) 18. I. Schur, Zur Theorie vertauschbaren Matrizen, J. Reine Angew. Math. 130 (1905), 66–76. 19. S. Paul Smith, Subspaces of non-commutative spaces, Trans. Amer. Math. Soc. 354 (2002), no. 6, 2131– 2171. MR 1885647 (2003f:14002) 20. Freddy M. J. Van Oystaeyen and Alain H. M. J. Verschoren, Non-commutative Algebraic Geometry: An Introduction, Lecture Notes in Mathematics, vol. 887, Springer-Verlag, Berlin, 1981. MR 639153 (85i:16006) Department of Mathematics, University of California, San Diego, 9500 Gilman Drive #0112, La Jolla, CA 92093-0112 E-mail address: [email protected] URL: http://math.ucsd.edu/~m1reyes/
1704.00191	1	1704	2017-04-01T15:59:00	On $(\sigma,\delta)$-skew McCoy modules	[ "math.RA" ]	Let $(\sigma,\delta)$ be a quasi derivation of a ring $R$ and $M_R$ a right $R$-module. In this paper, we introduce the notion of $(\sigma,\delta)$-skew McCoy modules which extends the notion of McCoy modules and $\sigma$-skew McCoy modules. This concept can be regarded also as a generalization of $(\sigma,\delta)$-skew Armendariz modules. Some properties of this concept are established and some connections between $(\sigma,\delta)$-skew McCoyness and $(\sigma,\delta)$-compatible reduced modules are examined. Also, we study the property $(\sigma,\delta)$-skew McCoy of some skew triangular matrix extensions $V_n(M,\sigma)$, for any nonnegative integer $n\geq 2$. As a consequence, we obtain: (1) $M_R$ is $(\sigma,\delta)$-skew McCoy if and only if $M[x]/M[x](x^n)$ is $(\overline{\sigma},\overline{\delta})$-skew McCoy, and (2) $M_R$ is $\sigma$-skew McCoy if and only if $M[x;\sigma]/M[x;\sigma](x^n)$ is $\overline{\sigma}$-skew McCoy.	math.RA	math	ON (σ, δ)-SKEW MCCOY MODULES MOHAMED LOUZARI AND L'MOUFADAL BEN YAKOUB Abstract. Let (σ, δ) be a quasi derivation of a ring R and MR a right R-module. In this paper, we introduce the notion of (σ, δ)-skew McCoy modules which extends the notion of McCoy modules and σ-skew McCoy modules. This concept can be regarded also as a generalization of (σ, δ)- skew Armendariz modules. Some properties of this concept are estab- lished and some connections between (σ, δ)-skew McCoyness and (σ, δ)- compatible reduced modules are examined. Also, we study the property (σ, δ)-skew McCoy of some skew triangular matrix extensions Vn(M, σ), for any nonnegative integer n ≥ 2. As a consequence, we obtain: (1) MR is (σ, δ)-skew McCoy if and only if M [x]/M [x](xn) is (σ, δ)-skew McCoy, and (2) MR is σ-skew McCoy if and only if M [x; σ]/M [x; σ](xn) is σ-skew McCoy. 1. Introduction Throughout this paper, R denotes an associative ring with unity and MR a right R-module. For a subset X of a module MR, rR(X) = {a ∈ RXa = 0} and ℓR(X) = {a ∈ RaX = 0} will stand for the right and the left annihilator of X in R respectively. An Ore extension of a ring R is denoted by R[x; σ, δ], where σ is an endomorphism of R and δ is a σ-derivation, i.e., δ : R → R is an additive map such that δ(ab) = σ(a)δ(b) + δ(a)b for all a, b ∈ R (the pair (σ, δ) is also called a quasi-derivation of R). Recall that elements of R[x; σ, δ] are polynomials in x with coefficients written on the left. Multiplication in R[x; σ, δ] is given by the multiplication in R and the condition xa = σ(a)x + δ(a), for all a ∈ R. In the next, S will stand for the Ore extension R[x; σ, δ]. On the other hand, we have a natural functor − ⊗R S from the category of right R-modules into the category of right S-modules. For a right R-module M , the right S-module M ⊗R S is called the induced module [16]. Since R[x; σ, δ] is a free left R-module, elements of M ⊗R S can be seen as polynomials in x with coefficients in M with natural addition and right S-module multiplication. For any 0 ≤ i ≤ j (i, j ∈ N), f j i ∈ End(R, +) will denote the map which is the sum of all possible words in σ, δ built with i factors of σ and j − i factors i (a)xi for of δ (e.g., f n n = σn and f n i=0 f j 0 = δn, n ∈ N). We have xja = Pj Date: March 14, 2018. 2010 Mathematics Subject Classification. 16S36, 16U80. Key words and phrases. McCoy module, (σ, δ)-skew McCoy module, semicommutative module, Armendariz module, (σ, δ)-skew Armendariz module, reduced module. 1 2 M. LOUZARI AND L. BEN YAKOUB all a ∈ R, where i, j are nonnegative integers with j ≥ i (see [14, Lemma 4.1]). Following Lee and Zhou [15], we introduce the notation M [x; σ, δ] to write the S-module M ⊗R S. Consider M [x; σ, δ] :=( n Xi=0 mixi n ≥ 0, mi ∈ M) ; which is an S-module under an obvious addition and the action of monomials of R[x; σ, δ] on monomials in M [x; σ, δ]R[x;σ,δ] via (mxj)(axℓ) = mPj for all a ∈ R and j, ℓ ∈ N. The S-module M [x; σ, δ] is called the skew poly- nomial extension related to the quasi-derivation (σ, δ). i=0 f j i (a)xi+ℓ A module MR is semicommutative, if for any m ∈ M and a ∈ R, ma = 0 implies mRa = 0 [6]. Let σ an endomorphism of R, MR is called an σ- semicommutative module [17] if, for any m ∈ M and a ∈ R, ma = 0 implies mRσ(a) = 0. For a module MR and a quasi-derivation (σ, δ) of R, we say that MR is σ-compatible, if for each m ∈ M and a ∈ R, we have ma = 0 ⇔ mσ(a) = 0. Moreover, we say that MR is δ-compatible, if for each m ∈ M and a ∈ R, we have ma = 0 ⇒ mδ(a)=0. If MR is both σ-compatible and δ-compatible, we say that MR is (σ, δ)-compatible (see [3]). In [17], a module MR is called σ-skew Armendariz, if m(x)f (x) = 0 j=0 ajxj ∈ R[x; σ] implies miσi(aj) = 0 for all i, j. According to Lee and Zhou [15], MR is called σ- Armendariz, if it is σ-compatible and σ-skew Armendariz. i=0 mixi ∈ M [x; σ] and f (x) =Pm where m(x) =Pn j=0 bjxj ∈ R[x; σ, δ], we have mixibjxj = 0 for all i, j. Following Alhevas and Moussavi [1], a module MR is called (σ, δ)-skew Ar- i=0 mixi ∈ M [x; σ, δ] mendariz, if whenever m(x)g(x) = 0 where m(x) = Pp and g(x) =Pq In this paper, we introduce the concept of (σ, δ)-skew McCoy modules which is a generalization of McCoy modules and σ-skew McCoy modules. This concept can be regarded also as a generalization of (σ, δ)-skew Armen- dariz modules and rings. We study connections between reduced modules, (σ, δ)-compatible modules and (σ, δ)-skew McCoy modules. Also, we show that (σ, δ)-skew McCoyness passes from a module MR to its skew triangular matrix extension Vn(M, σ). In this sens, we complete the definition of skew triangular matrix rings Vn(R, σ) given by Isfahani [12], by introducing the notion of skew triangular matrix modules. Moreover, we give some results on (σ, δ)-skew McCoyness for skew triangular matrix modules. 2. (σ, δ)-skew McCoy modules Cui and Chen [7, 8], introduced both concepts of McCoy modules and σ-skew McCoy modules. A module MR is called McCoy if m(x)g(x) = 0, j=0 bjxj ∈ R[x] \ {0} implies that there exists a ∈ R \ {0} such that m(x)a = 0. A module MR is called i=0 mixi ∈ M [x; σ] and j=0 bjxj ∈ R[x; σ] \ {0} implies that there exists a ∈ R \ {0} such where m(x) =Pp σ-skew McCoy if m(x)g(x) = 0, where m(x) = Pp g(x) = Pq i=0 mixi ∈ M [x] and g(x) =Pq ON (σ, δ)-SKEW MCCOY MODULES 3 that m(x)a = 0. With the same manner, we introduce the concept of (σ, δ)- skew McCoy modules which is a generalization of McCoy modules, σ-skew McCoy modules and (σ, δ)-skew Armendariz modules. Definition 2.1. Let MR be a module and M [x; σ, δ] the corresponding (σ, δ)- skew polynomial module over R[x; σ, δ]. (1) The module MR is called (σ, δ)-skew McCoy if m(x)g(x) = 0, where j=0 bjxj ∈ R[x; σ, δ] \ {0}, im- ℓ (a) = m(x) =Pp plies that there exists a ∈ R \ {0} such that m(x)a = 0 (i.e., Pp i=0 mixi ∈ M [x; σ, δ] and g(x) =Pq (2) The ring R is called (σ, δ)-skew McCoy if R is (σ, δ)-skew McCoy as 0, for all ℓ = 0, 1, · · · , p). i=ℓ mif i a right R-module. Remark 2.2. (1) If MR is an (σ, δ)-skew Armendariz module then it is (σ, δ)-skew McCoy (Proposition 2.4). But the converse is not true (Example 2.5). (2) If σ = idR and δ = 0 we get the concept of McCoy module, if only δ = 0, we get the concept of σ-skew McCoy module. (3) A module MR is (σ, δ)-skew McCoy if and only if for all m(x) ∈ M [x; σ, δ], rR[x;σ,δ](m(x)) 6= 0 ⇒ rR[x;σ,δ](m(x)) ∩ R 6= 0. An ideal I of a ring R is called (σ, δ)-stable, if σ(I) ⊆ I and δ(I) ⊆ I. Proposition 2.3. (1) Let I be a nonzero right ideal of R. If I is (σ, δ)-stable then R/I is an R-module (σ, δ)-skew McCoy. (2) For any index set I, if Mi is an (σi, δi)-skew McCoy as Ri-module for (σ, δ) = (σi, δi)i∈I . each i ∈ I, then Qi∈I Mi is an (σ, δ)-skew McCoy as Qi∈I Ri-module, where (3) Every submodule of an (σ, δ)-skew McCoy module is (σ, δ)-skew Mc- Coy. In particular, if I is a right ideal of an (σ, δ)-skew McCoy ring then IR is (σ, δ)-skew McCoy module. (4) A module MR is (σ, δ)-skew McCoy if and only if every finitely gen- erated submodule of MR is (σ, δ)-skew McCoy. ℓ=0 f i ℓ = 0, 1, · · · , i. Hence m(x)r = ¯0. ℓ(r)xℓ ∈ I[x; σ, δ], because f i i=0 mixi ∈ (R/I)[x; σ, δ], where mi = ri + I ∈ R/I for all i = 0, 1, · · · , p and r an arbitrary nonzero element of I. We have ℓ (r) ∈ I for all Proof. (1) Let m(x) =Pp m(x)r = Pp i=0(ri + I)Pi (2) Let M = Qi∈I Mi and R = Qi∈I Ri such that each Mi is an (σi, δi)- and f (x) = (fi(x))i∈I ∈ R[x; σ, δ] \ {0}, where mi(x) = Pp Mi[x; σi, δi] and fi(x) = Pq skew McCoy as Ri-module for all i ∈ I. Take m(x) = (mi(x))i∈I ∈ M [x; σ, δ] s=0 mi(s)xs ∈ t=0 ai(t)xt ∈ Ri[x; σi, δi] for each i ∈ I. Suppose that m(x)f (x) = 0, then mi(x)fi(x) = 0 for each i ∈ I. Since Mi is (σi, δi)- skew McCoy, there exists 0 6= ri ∈ Ri such that mi(x)ri = 0 for each i ∈ I. Thus m(x)r = 0 where 0 6= r = (ri)i∈I ∈ R. (3) and (4) are obvious. (cid:3) 4 M. LOUZARI AND L. BEN YAKOUB Proposition 2.4. If MR is an (σ, δ)-skew Armendariz module then it is (σ, δ)-skew McCoy. i=0 mixi ∈ M [x; σ, δ] and g(x) =Pq j=0 bjxj ∈ R[x; σ, δ]\ {0}. Suppose that m(x)g(x) = 0, then mixibjxj = 0 for all i, j. Since g(x) 6= 0 then bj0 6= 0 for some j0 ∈ {0, 1, · · · , p}. Thus mixibj0xj0 = 0 for ℓ (bj0))xℓ+j0 = 0, and ℓ(bj0) = 0 for all ℓ = 0, 1, · · · , p. Thus m(x)bj0 = 0, therefore (cid:3) Proof. Let m(x) =Pp all i. On the other hand mixibj0xj0 =Pp so Pp MR is (σ, δ)-skew McCoy. ℓ=0(Pp i=ℓ mif i i=ℓ mif i By the next example, we see that the converse of Proposition 2.4 does not hold. Example 2.5. Let R be a reduced ring. Consider the ring a a12 a13 a14 a23 a24 0 0 a a34 a 0 0 a 0 0 R4 =      , a, aij ∈ R  Since R is reduced then it is right McCoy and so R4 is right McCoy, by [18, Proposition 2.1]. But R4 is not Armendariz by [13, Example 3]. A module (σ, δ)-skew McCoy need not to be McCoy by [8, Example 2.3(2)]. Also, the following example shows that, there exists a module which is McCoy but not (σ, δ)-skew McCoy. Example 2.6. Let Z2 be the ring of integers modulo 2, and consider the ring R = Z2 ⊕ Z2 with the usual addition and multiplication. Let σ be an endomorphism of R defined by σ((a, b)) = (b, a) and δ an σ-derivation of R defined by δ((a, b)) = (a, b) − σ((a, b)). The ring R is commutative reduced then it is McCoy. However, for p(x) = (1, 0)x and q(x) = (1, 1) + (1, 0)x ∈ R[x; σ, δ]. We have p(x)q(x) = 0, but p(x)(a, b) 6= 0 for any 0 6= (a, b) ∈ R. Therefore, R is not (σ, δ)-skew McCoy. Also, R is not (σ, δ)- compatible, because (0, 1)(1, 0) = (0, 0), but (0, 1)σ((1, 0)) = (0, 1)2 6= (0, 0) and (0, 1)δ((1, 0)) = (0, 1)(1, 1) = (0, 1) 6= (0, 0). Lemma 2.7. Let MR be an (σ, δ)-compatible module. For any m ∈ MR, a ∈ R and nonnegative integers i, j. We have the following: (1) ma = 0 ⇒ mσi(a) = mδj(a) = 0. (2) ma = 0 ⇒ mσi(δj (a)) = mδi(σj(a)) = 0. Proof. The verification is straightforward. (cid:3) If MR is an (σ, δ)-compatible module then ma = 0 ⇒ mf j i (a) = 0 for any nonnegative integers i, j such that i ≥ j, where m ∈ MR and a ∈ R. For a subset U of MR and (σ, δ) a quasi-derivation of R, the set of all skew polynomials with coefficients in U is denoted by U [x; σ, δ]. ON (σ, δ)-SKEW MCCOY MODULES 5 Lemma 2.8. Let MR be a module and (σ, δ) a quasi-derivation of R. The following are equivalent: (1) For any U ⊆ M [x; σ, δ], (rR[x;σ,δ](U ) ∩ R)[x; σ, δ] = rR[x;σ,δ](U ). j=0 ajxj ∈ ℓ=i mℓf ℓ i=0 mixi ∈ M [x; σ, δ] and f (x) = Pq i=0 mixi ∈ M [x; σ, δ] and f (x) =Pq i (aj) = 0 for all i, j. j=0 ajxj ∈ R[x; σ, δ]. If m(x)f (x) = 0, we have f (x) ∈ rR[x;σ,δ](m(x)) = (rR[x;σ,δ](m(x))∩ R)[x; σ, δ]. Then aj ∈ rR[x;σ,δ](m(x)) for all j, so that m(x)aj = 0 for i (aj) = 0 for all 0 ≤ i ≤ p. Thus ℓ=i mℓf ℓ (2) For any m(x) = Pp R[x; σ, δ]. If m(x)f (x) = 0 implies Pp Proof. (1) ⇒ (2). Let m(x) =Pp all j. But m(x)aj = 0 ⇔ Pp Pp i (aj) = 0 for all i, j. ℓ=i mℓf ℓ (2) ⇒ (1). Let U ⊆ M [x; σ, δ], we have always (rR[x;σ,δ](U ) ∩ R)[x; σ, δ] ⊆ rR[x;σ,δ](U ). Conversely, let f (x) ∈ rR[x;σ,δ](U ) then by (2), we have U aj = 0 for all j and so aj ∈ rR[x;σ,δ](U ) ∩ R. Therefore f (x) ∈ (rR[x;σ,δ](U ) ∩ R)[x; σ, δ]. (cid:3) Theorem 2.9 (McCoy's Theorem for module extensions). Let MR be a module and N a nonzero submodule of M [x; σ, δ]. If one of the equivalent conditions of Lemma 2.8 is satisfied. Then rR[x;σ,δ](N ) 6= 0 implies rR(N ) 6= 0. Proof. Suppose that rR[x;σ,δ](N ) 6= 0, then there exists 0 6= f (x) =Pp rR[x;σ,δ](N ). But rR[x;σ,δ](N ) = (rR[x;σ,δ](N ) ∩ R)[x; σ, δ] by Lemma 2.8. Therefore all ai are in rR[x;σ,δ](N ), so ai ∈ rR(N ) for all i. Since f (x) 6= 0 then there exists i0 ∈ {0, 1, · · · p} such that 0 6= ai0 ∈ rR(N ). So that rR(N ) 6= 0. (cid:3) i=0 aixi ∈ Definition 2.10. Let MR be a module and σ an endomorphism of R. We say that MR satisfies the condition (Cσ) if whenever mσ(a) = 0 with m ∈ M and a ∈ R, then ma = 0. Proposition 2.11. Let m(x) =Pp i=0 mixi ∈ M [x; σ, δ] and f (x) =Pq ∈ R[x; σ, δ] such that m(x)f (x) = 0. If one of the following conditions hold: (a) MR is (σ, δ)-skew Armendariz and satisfy the condition (Cσ). (b) MR is reduced and (σ, δ)-compatible. Then miaj = 0 for all i, j. j=0 ajxj Proof. (a) Since MR is (σ, δ)-skew Armendariz then from m(x)f (x) = 0, ℓ(aj)xj+ℓ = miσi(aj)xi+j + Q(x) = 0 where Q(x) is a polynomial in M [x; σ, δ] of degree strictly less than i + j. Thus miσi(aj) = 0, therefore miaj = 0 for all i, j. we get mixiajxj = 0 for all i, j. But mixiajxj = miPi ℓ=0 f i (b) We will use freely the fact that, if ma = 0 then mσi(a) = mδj(a) = i (a) = 0 for any nonnegative integers i, j with j ≥ i. From m(x)f (x) = 0, mf j we have the following system of equations: (0) (1) mpσp(aq) = 0, mpσp(aq−1) + mp−1σp−1(aq) + mpf p p−1(aq) = 0, 6 M. LOUZARI AND L. BEN YAKOUB (2) mpσp(aq−2)+mp−1σp−1(aq−1)+mpf p p−1(aq−1)+mp−2σp−2(aq)+mp−1f p−1 p−2 (aq) (3) mpσp(aq−3) + mp−1σp−1(aq−2) + mpf p +mpf p p−2(aq) = 0, p−1(aq−2) + mp−2σp−2(aq−1) +mp−1f p−1 p−2 (aq−1) + mpf p p−2(aq−1) + mp−3σp−3(aq) + mp−2f p−2 p−3 (aq) + mpf p p−3 (aq) +mp−1f p−1 p−3(aq) = 0, ... Xj+k=ℓ p Xi=0 q (mi Xk=0 i Xj=0 f i j(ak)) = 0, ... p miδi(a0) = 0. Xi=0 (ℓ) (p + q) From equation (0), we have mpaq = 0 by σ-compatibility. Multiplying equation (1) on the right hand by aq, we get (1′) mpσp(aq−1)aq + mp−1σp−1(aq)aq + mpf p Since MR is semicommutative, then p−1(aq)aq = 0, mpaq = 0 ⇒ mpσp(aq−1)aq = mpf p p−1(aq)aq = 0. By Lemma 2.12, equation (1′) gives mp−1aq = 0. Also, by (σ, δ)-compatibility, equation (1) implies mpσp(aq−1) = 0, because mpaq = mp−1aq = 0. Thus mpaq−1 = 0. Summarizing at this point, we have (α) mpaq = mp−1aq = mpaq−1 = 0 Now, multiplying equation (2) on the right hand by aq, we get (2′) mpσp(aq−2)aq+mp−1σp−1(aq−1)aq+mpf p p−1(aq−1)aq+mp−2σp−2(aq)aq +mp−1f p−1 p−2 (aq)aq + mpf p p−2(aq)aq = 0, With the same manner as above, equation (2′) gives mp−2σp−2(aq)aq = 0 and thus mp−2aq = 0 (β). Also, multiplying equation (2) on the right hand by aq−1, we get (2′′) mpσp(aq−2)aq−1 + mp−1σp−1(aq−1)aq−1 + mpf p p−1(aq−1)aq−1 +mp−2σp−2(aq)aq−1 + mp−1f p−1 p−2 (aq)aq−1 + mpf p p−2(aq)aq−1 = 0 Equations (α) and (β) implies 0 = mpσp(aq−2)aq−1 = mpf p p−1(aq−1)aq−1 = mp−2σp−2(aq)aq−1 ON (σ, δ)-SKEW MCCOY MODULES 7 = mp−1f p−1 p−2 (aq)aq−1 = mpf p p−2(aq)aq−1 Hence, equation (2′′) gives mp−1σp−1(aq−1)aq−1 = 0 and by Lemma 2.12, we get mp−1aq−1 = 0 (γ). Now, by equations (α),(β) and (γ), we get mp−1σp−1(aq−1) = mpf p p−2 (aq) = mpf p 0. Therefore equation (2) implies mpσp(aq−2) = 0, so that mpaq−2 = 0. p−1(aq−1) = mp−2σp−2(aq) = mp−1f p−1 p−2(aq) = Summarizing at this point, we have miaj = 0 with i + j ∈ {p + q, p + q − (cid:3) 1, p + q − 2}. Continuing this procedure yields miaj = 0 for all i, j. Lemma 2.12. Let MR be an (σ, δ)-compatible module, if ma2 = 0 implies ma = 0 for any m ∈ M and a ∈ R. Then (1) mσ(a)a = 0 implies ma = mσ(a) = 0. (2) maσ(a) = 0 implies ma = mσ(a) = 0. Proof. The proof is straightforward. (cid:3) According to Lee and Zhou [15], a module MR is called σ-reduced, if for any m ∈ M and a ∈ R. We have (1) ma = 0 implies mR ∩ M a = 0. (2) ma = 0 if and only if mσ(a) = 0. The module MR is called reduced if MR is idR-reduced. Lemma 2.13 ([15, Lemma 1.2]). The following are equivalent for a module MR: (1) MR is σ-reduced. (2) The following three conditions hold: For any m ∈ M and a ∈ R, (a) ma = 0 implies mRa = mRσ(a) = 0. (b) maσ(a) = 0 implies ma = 0. (c) ma2 = 0 implies ma = 0. By Lemma 2.13, a module MR is reduced if and only if it is semicommu- tative with ma2 = 0 implies ma = 0 for any m ∈ M and a ∈ R. Corollary 2.14 ([1, Theorem 2.19]). Every (σ, δ)-compatible and reduced module is (σ, δ)-skew Armendariz. Proof. Clearly from Proposition 2.11(b). (cid:3) Let MR be a module and (σ, δ) a quasi derivation of R. We say that MR satisfies the condition (∗), if for any m(x) ∈ M [x; σ, δ] and f (x) ∈ R[x; σ, δ], m(x)f (x) = 0 implies m(x)Rf (x) = 0. A module MR which satisfies the condition (∗) is semicommutative. But the converse is not true, by the next example. Example 2.15. Take the ring R = Z2 ⊕ Z2 with (σ, δ) as considered in Example 2.6. Since R is commutative then the module RR is semicom- mutative. However, it does not satisfy the condition (∗). For p(x) = (1, 0)x and q(x) = (1, 1) + (1, 0)x ∈ R[x; σ, δ]. We have p(x)q(x) = 0, but p(x)(1, 0)q(x) = (1, 0) + (1, 0)x 6= 0. Thus p(x)Rq(x) 6= 0. 8 M. LOUZARI AND L. BEN YAKOUB Theorem 2.16. If a module MR is (σ, δ)-compatible and reduced, then it satisfies the condition (∗). Proof. Let m(x) =Pp j=0 ajxj ∈ R[x; σ, δ], such that m(x)f (x) = 0. By Proposition 2.11(b) and semicommutativity of MR, we have miRaj = 0 for all i and j. Moreover, compatibility implies mif ℓ (cid:3) i=0 mixi ∈ M [x; σ, δ] and f (x) =Pq k(Raj) = 0 for all i, j, k, ℓ. Therefore m(x)Rf (x) = 0. Since the ring R = Z2 ⊕Z2 is reduced, then from Example 2.15, we can see that the condition "(σ, δ)-compatible" in Theorem 2.16 is not superfluous. Proposition 2.17. Let MR be an (σ, δ)-compatible module which satisfies i=0 mixi ∈ M [x; σ, δ] and f (x) = = 0 for all j=0 ajxj ∈ R[x; σ, δ] \ {0}, m(x)f (x) = 0. Then miap+1 q i = 0, 1, · · · , p. (∗). Suppose that for any m(x) = Pp Pq Proof. Let m(x) =Pp i=0 mixi ∈ M [x; σ, δ] and f (x) =Pq j=0 ajxj ∈ R[x; σ, δ]\ {0}, such that m(x)f (x) = 0. We can suppose that aq 6= 0. From m(x)f (x) = 0, we get mpσp(aq) = 0. Since MR is (σ, δ)-compatible, we have mpaq = 0 which implies mpxpaq = 0. Since m(x)f (x) = 0 implies m(x)aqf (x) = 0. Then 0 = (mpxp+mp−1xp−1+· · ·+m1x+m0)(a2 qxq+aqaq−1xq−1+· · ·+aqa1x+aqa0) = (mp−1xp−1+· · ·+m1x+m0)(a2 qxq +aqaq−1xq−1+· · ·+aqa1x+aqa0). If we put f ′(x) = aqf (x) and m′(x) = Pp−1 0. Continuing this procedure yields miap+1−i Consequently miap+1 q = 0 for all i = 0, 1, · · · , p. i=0 mixi then we get mp−1a2 q = = 0 for all i = 0, 1, · · · , p. (cid:3) q Corollary 2.18. Let MR be an (σ, δ)-compatible module over a reduced ring R. If MR satisfies (∗), then it is (σ, δ)-skew McCoy. Proof. Let m(x) =Pp i=0 mixi ∈ M [x; σ, δ] and f (x) =Pq {0}, such that m(x)f (x) = 0. We can suppose that aq 6= 0. By Proposi- tion 2.17, we have miap+1 q = 0 for all i = 0, 1, · · · , p. Since MR is (σ, δ)- compatible, we get mixiap+1 )xℓ = 0 for all i. Hence m(x)ap+1 6= 0, because R is reduced. Consequently MR is (σ, δ)-skew McCoy. (cid:3) q = miPi q = 0 where ap+1 ℓ (ap+1 ℓ=0 f i q j=0 ajxj ∈ R[x; σ, δ]\ q Example 2.19. Consider a ring of polynomials over Z2, R = Z2[x]. Let σ : R → R be an endomorphism defined by σ(f (x)) = f (0). Then (1) R is not σ-compatible. Let f = 1 + x, g = x ∈ R, we have f g = (1 + x)x 6= 0, however f σ(g) = (1 + x)σ(x) = 0. (2) R is σ-skew Armendariz [10, Example 5]. From Example 2.19, we see that the ring R = Z2[x] is σ-skew McCoy because it is σ-skew Armendariz, but it is not σ-compatible. Thus the (σ, δ)- compatibility condition is not essential to obtain (σ, δ)-skew McCoyness. ON (σ, δ)-SKEW MCCOY MODULES 9 Example 2.20 ([5, Example 2.5]). Let R be a ring, σ an endomorphism of R and δ be a σ-derivation of R. Suppose that R is σ-rigid. Consider the ring V3(R) =    a b c 0 a b 0 0 a   a, b, c ∈ R  . The ring V3(R) is (σ, δ)-skew McCoy, reduced and (σ, δ)-compatible, and by Theorem 2.16, it satisfies the condition (∗). 3. (σ, δ)-skew McCoyness of some matrix extensions For a nonnegative integer n ≥ 2, let R be a ring and M a right R-module. Consider and Sn(R) := a a12 a13 a23 0 a 0 ... ... 0 0 a 0 ... 0 . . . a1n . . . a2n . . . a3n ... . . . a . . .     a, aij ∈ R       0 ... 0 Sn(M ) := m m12 m13 0 m m23 0 ... 0 . . . m1n . . . m2n m . . . m3n ... ... . . . . . . m 0     Sn(M ) and A = (aij) ∈ Sn(R), U A = (mij) ∈ Sn(M ) with mij =Pn Clearly, Sn(M ) is a right Sn(R)-module under the usual matrix addi- tion operation and the following scalar product operation. For U = (uij) ∈ k=1 uikakj for all i, j. A quasi derivation (σ, δ) of R can be extended to a quasi deriva- tion (σ, δ) of Sn(R) as follows: σ((aij)) = (σ(aij )) and δ((aij)) = (δ(aij )). We can easily verify that δ is a σ-derivation of Sn(R). m, mij ∈ M   Theorem 3.1. A module MR is (σ, δ)-skew McCoy if and only if Sn(M ) is (σ, δ)-skew McCoy as an Sn(R)-module for any nonnegative integer n ≥ 2. Proof. The proof is similar to [4, Theorem 14]. (cid:3) Now, for n ≥ 2. Consider Vn(R) := a0 a1 a2 a3 a0 a1 a2 0 a0 a1 0 0 ... ... ... ... 0 0 0 0 0 0 0 0 . . . an−1 . . . an−2 . . . an−3 ... . . . a1 . . . a0 . . .       a0, a1, a2, · · · , an−1 ∈ R   m0, m1, m2, · · · , mn−1 ∈ M   10 and Vn(M ) := M. LOUZARI AND L. BEN YAKOUB m0 m1 m2 m3 0 m0 m1 m2 0 m0 m1 0 ... ... ... 0 0 0 0 0 0 ... 0 0 . . . mn−1 . . . mn−2 . . . mn−3 ... . . . . . . m1 . . . m0       With the same method as above, Vn(M ) is a right Vn(R)-module, and a quasi derivation (σ, δ) of R can be extended to a quasi derivation (σ, δ) of Vn(R). Note that Vn(M ) ∼= M [x]/M [x](xn) where M [x](xn) is a submodule of M [x] generated by xn and Vn(R) ∼= R[x]/(xn) where (xn) is an ideal of R[x] generated by xn. Proposition 3.2. A module MR is (σ, δ)-skew McCoy if and only if Vn(M ) is (σ, δ)-skew McCoy as an Vn(R)-module for any nonnegative integer n ≥ 2. Proof. The proof is similar to that of [4, Theorem 14] or [8, Proposition 2.27]. (cid:3) Corollary 3.3. For a nonnegative integer n ≥ 2, we have: (1) MR is (σ, δ)-skew McCoy if and only if M [x]/M [x](xn) is (σ, δ)-skew McCoy. (2) R is (σ, δ)-skew McCoy if and only if R[x]/(xn) is (σ, δ)-skew McCoy. (3) R is McCoy if and only if R[x]/(xn) is McCoy. In the next, we define skew triangular matrix modules Vn(M, σ), based on the definition of skew triangular matrix rings Vn(R, σ) given by Isfahani [12]. Let σ be an endomorphism of a ring R and MR a right R-module. For n ≥ 2. Consider Vn(R, σ) := and Vn(M, σ) :=         a0 a1 a2 a3 a0 a1 a2 0 a0 a1 0 0 ... ... ... ... 0 0 0 0 0 0 0 0 . . . an−1 . . . an−2 . . . an−3 ... . . . a1 . . . . . . a0   a0, a2, · · · , an−1 ∈ R   m0 m1 m2 m3 0 m0 m1 m2 0 m0 m1 0 ... ... ... 0 0 0 0 0 0 ... 0 0 . . . mn−1 . . . mn−2 . . . mn−3 ... . . . . . . m1 . . . m0   m0, m2, · · · , mn−1 ∈ M   ON (σ, δ)-SKEW MCCOY MODULES 11 Clearly Vn(M, σ) is a right Vn(R, σ)-module under the usual matrix ad- dition operation and the following scalar product operation. a0 a1 a2 a3 a0 a1 a2 0 a0 a1 0 0 ... ... ... ... 0 0 0 0 0 0 0 0 . . . an−1 . . . an−2 . . . an−3 ... . . . a1 . . . . . . a0 =   , where cn−1 cn−2 cn−3 ... c1 c0   m0 m1 m2 m3 0 m0 m1 m2 0 m0 m1 0 ... ... ... 0 0 0 0 0 0 ... 0 0   . . . mn−1 . . . mn−2 . . . mn−3 ... . . . . . . m1 . . . m0 c1 c0 0 ... 0 0 c2 c1 c0 ... 0 0 c3 c2 c1 ... 0 0     . . . . . . . . . . . . . . . . . . c0 0 0 ... 0 0   ci = m0σ0(ai) + m1σ1(ai−1) + m2σ2(ai−2) + · · · + miσi(a0) for each 0 ≤ i ≤ n − 1. We denote elements of Vn(R, σ) by (a0, a1, · · · , an−1) and elements of Vn(M, σ) by (m0, m1, · · · , mn−1). There is a ring isomorphism ϕ : R[x; σ]/(xn) → Vn(R, σ) given by ϕ(a0+a1x+a2x2+· · ·+an−1xn−1+(xn)) = (a0, a1, a2, · · · , an−1), and an abelian group isomorphism φ : M [x, σ]/M [x, σ](xn) → Vn(M, σ) given by φ(m0+m1x+m2x2+· · ·+mn−1xn−1+(xn)) = (m0, m1, m2, · · · , mn−1) such that φ(N (x)A(x)) = φ(N (x))ϕ(A(x)) for any N (x) = m0+m1x+m2x2+· · ·+mn−1xn−1+(xn) ∈ M [x, σ]/M [x, σ](xn) and A(x) = a0 + a1x + a2x2 + · · · + an−1xn−1 + (xn) ∈ R[x; σ]/(xn). The endomorphism σ of R can be extended to Vn(R, σ) and R[x; σ], and we will denote it in both cases by σ. Theorem 3.4. A module MR is σ-skew McCoy if and only if Vn(M, σ) is σ-skew McCoy as an Vn(R, σ)-module for any nonnegative integer n ≥ 2. Proof. We shall adapt the proof of [4, Theorem 14] to this situation. Note that Vn(R, σ)[x, σ] ∼= Vn(R[x, σ], σ) and Vn(M, σ)[x, σ] ∼= Vn(M [x, σ], σ). We only prove when n = 2, because other cases can be proved with the same manner. Suppose that MR is σ-skew McCoy. Let 0 6= m(x) ∈ V2(M, σ)[x, σ] and 0 6= f (x) ∈ V2(R, σ)[x, σ] such that m(x)f (x) = 0, where i=0 m(i) i=0 m(i) 11 m(i) 12 0 m(i) i=0 m(i) m(x) = 0 0 p Xi=0 m(i) Xj=0 a(j) 11 0 q 11 ! xi = Pp 11 ! xj = Pq a(j) 12 a(j) j=0 a(j) 0 11 xi Pp Pp 11 xj Pq Pq 12 xi 11 xi ! =(cid:18) α11 α12 α11 (cid:19) 11 xj ! =(cid:18) β11 β12 β11 (cid:19) 12 xj 0 j=0 a(j) j=0 a(j) f (x) = 12 M. LOUZARI AND L. BEN YAKOUB 0 0 Then(cid:18) α11 α12 α11 (cid:19)(cid:18) β11 β12 β11 (cid:19) = 0, which gives α11β11 = 0 and α11β12+ α12σ(β11) = 0 in M [x; σ]. If α11 6= 0, then there exists 0 6= β ∈ {β11, β12} such that α11β = 0. Since MR is σ-skew McCoy then there exists 0 6= c ∈ R which satisfies α11c = 0, thus (cid:18) α11 α12 If α11 = 0 then (cid:18) 0 α12 0 0 (cid:19) =(cid:18) 0 α11c 0 (cid:19) = 0. 0 0 (cid:19) = 0, for any 0 6= c ∈ R. Therefore, α11 (cid:19)(cid:18) 0 c 0 (cid:19)(cid:18) 0 c 0 0 0 V2(M, σ) is σ-skew McCoy. Conversely, suppose that V2(M, σ) is an σ-skew McCoy module. Let 0 6= m(x) = m0 +m1x+· · ·+mpxp ∈ M [x; σ] and 0 6= f (x) = a0 +a1x+· · ·+ 0 aqxq ∈ R[x; σ], such that m(x)f (x) = 0. Then(cid:18) m(x) (cid:18) m(x)f (x) such that (cid:18) m(x) m(x) (cid:19)(cid:18) f (x) 0 a (cid:19) ∈ V2(R, σ) 0 a (cid:19) = 0, because V2(M, σ) is σ-skew Mc- m(x)f (x) (cid:19) = 0, so there exists 0 6= (cid:18) a b m(x) (cid:19)(cid:18) a b Coy. Thus m(x)a = m(x)b = 0, where a 6= 0 or b 6= 0. Therefore, MR is σ-skew McCoy. (cid:3) 0 0 0 0 0 0 0 f (x) (cid:19) = Corollary 3.5. For a nonnegative integer n ≥ 2, we have: (1) MR is σ-skew McCoy if and only if M [x; σ]/M [x; σ](xn) is σ-skew McCoy. (2) R is σ-skew McCoy if and only if R[x; σ]/(xn) is σ-skew McCoy. (3) MR is McCoy if and only if M [x]/M [x](xn) is McCoy. (4) R is McCoy if and only if R[x]/(xn) is McCoy. References [1] A. Alhevaz and A. Moussavi, On skew Armendariz and skew quasi-Armendariz mod- ules, Bull. Iran. Math. Soc. 1 (2012), 55-84. [2] S. Annin, Associated primes over skew polynomials rings, Comm. Algebra 30 (2002), 2511-2528. [3] S. Annin, Associated primes over Ore extension rings, J. Algebra appl. 3 (2004), 193-205. [4] M. Ba¸ser, T. K. Kwak and Y. Lee, The McCoy condition on skew polynomial rings, Comm. Algebra 37 (2009), 4026-4037. [5] L. Ben Yakoub and M. Louzari, Ore extensions of extended symmetric and reversible rings, Inter. J. of Algebra 3, 2009, No.9, 423-433. [6] A. M. Buhphang and M. B. Rege, semicommutative modules and Armendariz mod- ules, Arab J. Math. Sciences 8 (2002), 53-65. [7] J. Cui and J. Chen, On McCoy modules, Bull. Korean Math. Soc. 48 (2011), No. 1, 23-33. [8] J. Cui and J. Chen, On α-skew McCoy modules, Turk. J. Math. 36 (2012), 217-229. [9] E. Hashemi, Extensions of Baer and quasi-Baer modules, Bull. of the Iran. Math. Soc. 37 (2011), No. 1, 1-13. [10] C. Y. Hong, N. K. Kim and T. K. Kwak,On Skew Armendariz Rings, Comm. Algebra 31(1) (2003), 103-122. ON (σ, δ)-SKEW MCCOY MODULES 13 [11] N.H. McCoy, Annihilators in polynomial rings, Amer. math. monthly 64 (1957), 28- 29. [12] A. R. Nasr-Isfahani, On Skew Triangular Matrix Rings, Comm. Algebra 39 (2011), 4461-4469 [13] N. K. Kim and Y. Lee, Armendariz rings and reduced rings, J. Algebra 223 (2000), 477-488. [14] T. Y. Lam, A. Leroy and J. Matczuk, Primeness, semiprimeness and the prime radical of Ore extensions, Comm. Algebra 25 (8) (1997), 2459-2506. [15] T. K. Lee and Y. Lee, Reduced Modules, Rings, modules, algebras and abelian groups, 365-377, Lecture Notes in Pure and App. Math. 236 Dekker, New york, (2004). [16] A. Leroyy and J. Matczuk, On induced modules over Ore extensions, Comm. Algebra 32(7) (2004), 2743-2766. [17] C. P. Zhang and J. L. Chen, σ-skew Armendariz modules and σ-semicommutative modules, Taiwanese J. Math. 12(2) (2008), 473-486. [18] R. Zhao and Z. Liu, Extensions of McCoy Rings, Algebra Colloq. 16 3 (2009), 495- 502. Depertment of Mathematics, Faculty of sciences, Abdelmalek Essaadi Uni- versity, BP. 2121 Tetouan, Morocco E-mail address: [email protected] Depertment of Mathematics, Faculty of sciences, Abdelmalek Essaadi Uni- versity, BP. 2121 Tetouan, Morocco E-mail address: [email protected]
1210.0494	3	1210	2013-09-18T17:53:00	Lamination exact relations and their stability under homogenization	[ "math.RA" ]	Relations between components of the effective tensors of composites that hold regardless of composite's microstructure are called exact relations. Relations between components of the effective tensors of all laminates are called lamination exact relations. The question of existence of sets of effective tensors of composites that are stable under lamination, but not homogenization was settled by Milton with an example in 3D elasticity. In this paper we discuss an analogous question for exact relations, where in a wide variety of physical contexts it is known (a posteriori) that all lamination exact relations are stable under homogenization. In this paper we consider 2D polycrystalline multi-field response materials and give an example of an exact relation that is stable under lamination, but not homogenization. We also shed some light on the surprising absence of such examples in most other physical contexts (including 3D polycrystalline multi-field response materials). The methods of our analysis are algebraic and lead to an explicit description (up to orthogonal conjugation equivalence) of all representations of formally real Jordan algebras as symmetric $n\times n$ matrices. For each representation we examine the validity of the 4-chain relation\|a 4th degree polynomial identity, playing an important role in the theory of special Jordan algebras.	math.RA	math	Lamination exact relations and their stability under homogenization Yury Grabovsky November 21, 2018 Abstract Relations between components of the effective tensors of composites that hold regardless of composite's microstructure are called exact relations. Relations between components of the effective tensors of all laminates are called lamination exact relations. The question of existence of sets of effective tensors of composites that are stable under lamination, but not homogenization was settled by Milton with an example in 3D elasticity. In this paper we discuss an analogous question for exact relations, where in a wide variety of physical contexts it is known (a posteriori) that all lamination exact relations are stable under homogenization. In this paper we consider 2D polycrystalline multi-field response materials and give an example of an exact relation that is stable under lamination, but not homogenization. We also shed some light on the surprising absence of such examples in most other physical contexts (including 3D polycrystalline multi-field response materials). The methods of our analysis are algebraic and lead to an explicit description (up to orthogonal conjugation equivalence) of all representations of formally real Jordan algebras as symmetric n × n matrices. For each representation we examine the validity of the 4-chain relation -- a 4th degree polynomial identity, playing an important role in the theory of special Jordan algebras. 3 1 0 2 p e S 8 1 ] . A R h t a m [ 3 v 4 9 4 0 . 0 1 2 1 : v i X r a 1 Introduction The study of effective behavior of composite materials abounds with beautiful formulas linking the components of the effective tensor and the tensor of material properties of its constituents, when virtually nothing is known about the microstructure of the composite [22, 27, 31, 43, 20, 47, 34, 35, 12, 13, 11, 5, 6, 33] (see also a review by Milton [39]). The general theory of such formulas was developed in [15, 18, 16]. The idea was to identify all relations that are preserved in all laminate microstructures, where two constituent materials are combined in layers perpendicular to a given unit vector (lamination direction). According to the general theory, reviewed in Section 2, each relation preserved under lamination corresponds to a Jordan multialgebra -- a subspace of the space of symmetric matrices closed with respect to several Jordan multiplications. Mutations of Jordan algebras [28] are particular examples of Jordan multialgebras, where all multiplications are "internal", i.e. coming from a single Jordan product in a classical Jordan algebra. In the context of exact relations not every Jordan multialgebra is a mutant of a classical one. The fundamental problem in the theory of exact relations is whether every Jordan multialgebra gives rise to a microstructure-independent relation. If this is the case, then all lamination exact relations are stable under homogenization. In [18, 16] we obtained an algebraic condition on Jordan multialgebras sufficient for stability under homogenization. In order to describe it we define completion of a Jordan multialgebra to be the set of all symmetric matrices in the smallest associative multialgebra containing the original Jordan multialgebra. If the Jordan multialgebra is complete, i.e. equal to its completion, then the corresponding lamination exact relation is stable under homogenization. Mutations of classical special Jordan algebras do not seem to posses superior completion properties compared to the general Jordan multialgebras, and hence, will not be investigated as a special subclass in this paper. The completeness for classical special Jordan algebras is related to the 4th degree polynomial identity, called the 4-chain relation, via Cohn's theorem [7]. It is well-known that classical special Jordan algebras do not have to be complete. However, from the point of view of the theory of composite materials the situation may be reversed if we restrict our attention only to those Jordan multialgebras that arise in physics. In particular, 1 we will focus on the Jordan multialgebras corresponding to the polycrystalline composites [15, 19, 18]. These possess additional structure of a representation space of SO(3) or SO(2) (for 2D or fiber-reinforced composites, whose microgeometry is completely determined by any 2D cross-section). The interaction between the group representation structure of SO(d) and the multiplicative structure of Jordan multialgebras has been studied in a series of papers [44, 46, 45, 30] inspired by the theory of exact relations. While there is no general theorem that guarantees completeness for SO(3) or SO(2)-invariant Jordan multialgebras, all known physically relevant examples of SO(3) or SO(2)-invariant Jordan multialgebras are complete. Is it a coincidence or a new fundamental property specific to rotationally invariant Jordan multialgebras? In this paper we exhibit a physically motivated example in 2D featuring an SO(2)-invariant incomplete Jordan multialgebra and the corresponding lamination exact relation that is not stable under homogenization (see (8.4)). This needs to be compared with an earlier result in [18] that all SO(3)-invariant Jordan multialgebras in the 3D analog of our example are complete, giving further support for the conjecture that all physically relevant SO(3)-invariant Jordan multialgebras are complete (see Conjecture 2.5 below). We remark that our example of SO(2)-invariant incomplete Jordan multialgebra could lead to another example of a rank-1 convex function that is not quasiconvex. While such an example has already been produced both in Calculus of Variations [48] and the theory of composite materials [40, Section 39.9], our example, coming from the theory of exact relations would possess additional SO(2) symmetry absent in the existing examples. While the 3D version of our example, described in Section 3, can be completely analyzed (see Theorem 3.1), the 2D version is more involved. In the more difficult 2D case we only exhibit a specific family of Jordan multialgebras generated by the subalgebras of Sym(Rn) -- the Jordan algebra of all real symmetric n × n matrices. The question of completeness of these subalgebras has not been studied. The answer to this question represents an algebraic contribution of this paper, which includes a complete characterization of all faithful representations of all formally real Jordan algebras in Sym(Rn) up to orthogonal conjugation. We note that all formally real Jordan algebras, up to a Jordan algebra isomorphism, were described in [26], where an attempt was made to build an algebraic foundation of quantum mechanics, where the observables would be defined axiomatically, and not as operators on a Hilbert space with the extraneous structure of associative multiplication. The explicit characterization of Jordan subalgebras of Sym(Rn), reveals that completeness can indeed fail, but only in a few exceptional cases. Adding more multiplications with respect to which the algebras have to be closed makes these cases even more exceptional, so that in the relatively low dimensions completeness holds in general. Specifically, all Jordan subalgebras of Sym(Rn) are complete, if n < 8. Even for general n completeness is generic. A generic symmetric n × n matrix has n distinct eigenvalues. Any subalgebra of Sym(Rn) containing such a matrix must necessarily be complete. This sheds light on the conspicuous absence of incomplete Jordan multialgebras from all physically relevant examples so far, since all of them, with the exception of the example in Section 3, are relatively low dimensional. 2 General theory of exact relations The standard references [4, 25] for the mathematical theory of composite materials emphasize homogenization theorems and deal primarily with conducting and elastic composites. In the case of periodic composites there is an abstract Hilbert space framework [29, 10, 41, 36, 37, 38] that treats all coupled field composites (e.g. piezo-electric, thermo-elastic, etc.) from the common point of view. We refer to [40] for a systematic treatment of composites and exact relations based on these ideas. In this paper we adopt the abstract Hilbert space formalism. Since our specific question is algebraic in nature we give here an algebra-centered overview of the general theory of exact relations. We refer the reader to [18, 40] for a composite material-centered exposition. A linear material response to applied fields (electric, elastic, thermal, etc.) is described by a symmetric positive definite operator L on a finite dimensional inner product space T . If we are interested in polycrys- talline composites the space T is also a representation space of the rotation group SO(3) or SO(2), if we deal with fiber-reinforced composites, whose microstructure is determined by any cross-section perpendicular to 2 the chosen "fiber direction". In this case the inner product on T is SO(d)-invariant, d = 2, 3. In making a composite we choose a compact set of admissible materials U ⊂ Sym+(T ), where Sym+(T ) is the set of symmetric positive definite operators on T . In the case of polycrystalline composites the set U must be SO(d)-invariant. The effective tensor L∗ of a composite depends on the microstructure. Varying the microstructure over all possible ones we obtain the set G(U ) of corresponding effective tensors L∗ of composites, called the G-closure of U . Generically the set G(U ) has a non-empty interior in Sym+(T ), even if U consists of only 2 points. We are interested in special situations, where G(U ) is a submanifold of Sym+(T ) of non-zero co-dimension. Definition 2.1. The submanifold M of Sym+(T ) is called an exact relation if the effective tensor L∗ of a composite will lie in M, whenever constituent materials lie in M, regardless of the microstructure. 2.1 Lamination exact relations and Jordan multialgebras In order to identify these special exact relations cases we test G-closure set by taking two arbitrary points {L1, L2} ⊂ G(U ) and forming a laminate -- a composite consisting of layers of material L1 alternating with layers of material L2. Such a laminate is described by the direction of the lamination n ∈ Sd−1 (d = 2 or 3) and the volume fractions θ1, θ2 = 1 − θ1 of L1 and L2, respectively. For the simple laminate microstructure there exists a beautiful explicit formula for the effective tensor [38, 40] Wn(L∗; L0) = θ1Wn(L1; L0) + θ2Wn(L2; L0), (2.1) where L0 is an arbitrary "reference material" and the map Wn(L, L0) has the form Wn(L; L0) = [(L0 − L)−1 − ΓL0(n)]−1 = [IT − (L0 − L)ΓL0(n)]−1(L0 − L), where IT is the identity map on T , and ΓL0(n) ∈ Sym(T ) is an explicitly known function of n ∈ Sd−1, whose specific definition is not important for our purposes. We remark, that even though the formula (2.1) involves L0, the effective tensor L∗ defined by it is independent of L0. Restricting attention only to laminate microstructures we formulate the notion of lamination exact relation. Definition 2.2. A lamination exact relation is a submanifold M of positive co-dimension in Sym+(T ) such that the effective tensor L∗ of a laminate made with {L1, L2} ⊂ M is in M for any choice of lamination direction n ∈ Sd−1 and volume faction θ1 ∈ [0, 1]. It is clear from (2.1) that Wn(M; L0) is a convex subset of Sym(T ) and at the same time a submanifold of Sym(T ) of the same co-dimension as M. Therefore, Wn(M; L0) is an affine subspace of Sym(T ). However, choosing L0 ∈ M and observing that W (L0; L0) = 0 implies that the affine subspaces Πn = Wn(M; L0) are linear subspaces of Sym(T ). The function Λm,n = Wm ◦ W −1 n maps Πn into Πm diffeomorphically (at least in the neighborhood of 0). We easily compute Λm,n(K) = [K −1 + ΓL0(n) − ΓL0(m)]−1 = [IT − (ΓL0 (m) − ΓL0(n))K]−1K. We see that the differential of Λm,n at K = 0 is the identity map: dΛm,n(0)ξ = ξ for any ξ ∈ T0Πn. That means that the tangent spaces of Πn and Πm are the same. It follows that Πn = Πm, since the submanifolds Πn and Πm, being subspaces of Sym(T ), coincide with their tangent spaces. Thus, Π = Wn(M; L0) is a well-defined subspace of Sym(T ), independent of n. Expanding the map Λn,m : Π → Π in powers of K we obtain Λm,n(K) = KAm,nK + KAm,nKAm,nK + KAm,nKAm,nKAm,nK + . . . , where Am,n = ΓL0(m) − ΓL0(n). We conclude that for M to be a lamination exact relation it is necessary and sufficient that KAK ∈ Π, K ∈ Π, A ∈ A = Span{ΓL0(m) − ΓL0(n) : n = 1}. (2.2) 3 It is easy to check that the subspace A does not depend on m (regardless of what ΓL0(m) is). The sub- spaces Π satisfying (2.2) are Jordan multialgebras, since they are closed with respect to a family of Jordan multiplications K1 ∗A K2 = (K1AK2 + K2AK1), A ∈ A. 1 2 To avoid any ambiguity we give the following definition. Definition 2.3. A subspace Π of an associative algebra B is called a Jordan A-multialgebra if it is closed with respect to a collection of Jordan multiplications1 X ∗A Y = 1 2 (XAY + Y AX), A ∈ A, (2.3) where A is a subspace of B. A classical definition of a special Jordan algebra corresponds to 1-dimensional subspaces A. A subspace Π′ of B is called an associative A-multialgebra if it is closed with respect to a collection of multiplications (X, Y ) 7→ XAY , A ∈ A. The associative A-multialgebra Π′ is called symmetric if X ∈ Π′ implies X T ∈ Π′. We remark that the notion of mutation of a Jordan algebra [28] provides a family of examples of Jordan multialgebras (they can be called "inner multialgebras"). If Π0 is closed with respect to a single product X ∗A0 Y then it will always be closed with respect to a family of multiplications (2.3), where A = {A0BA0 : B ∈ Π0}. The question of comparing algebraic structures of an algebra and its mutation is studied in [28], and is not the object of our investigation. 2.2 Exact relations and complete Jordan multialgebras Recall that Jordan multialgebras Π correspond to the lamination exact relations (see Definition 2.2). The fundamental question in the theory of exact relations is whether or not all lamination exact relations are exact relations in the sense of Definition 2.1. This question was investigated in [15, 18, 16, 40], where simple algebraic sufficient conditions have been formulated. To state this condition we define the notion of complete Jordan A-multialgebra. Starting with a Jordan A-multialgebra Π we define A(Π) to be the smallest associative A-multialgebra in the sense of Definition 2.3 containing Π. The associative A-multialgebra A(Π) is necessarily symmetric in the sense of Definition 2.3. We then define a completion Π of Π by Π = A(Π)sym, where Vsym for a subspace V ⊂ End(T ), such that V T = V . def= V \ Sym(T ) = {K + K T : K ∈ V } Definition 2.4. A Jordan A-multialgebra Π ⊂ Sym(T ) is called complete if We will say that Π is incomplete, if Π is not complete. Π = Π, (2.4) (2.5) It is easy to see that the completion Π of Π is the smallest complete Jordan A-multialgebra containing Π. According to the general theory [18, 16], complete Jordan A-multialgebras Π correspond to exact relations M = W −1 n (Π; L0). It is easy to show that Π is complete if and only if 3 and 4-chain identities K1A1K2A2K3 + K3A2K2A1K1 ∈ Π, K1A1K2A2K3A3K4 + K4A3K3A2K2A1K1 ∈ Π (2.6) 1It would be more proper to use the term special Jordan A-multialgebra. In this paper we omit the qualifier "special", since we do not deal with the most general Jordan algebras over a general field. 4 hold for all {K1, K2, K3, K4} ⊂ Π and all {A1, A2, A3} ⊂ A. The proof is a straightforward modification of the proof of Cohn's theorem [7, 24]. We remark that in the case of classical Jordan algebras (one-dimensional subspace A) the 3-chain relations hold automatically, while this is not so for Jordan multialgebras. This is a crucial distinction between classical special Jordan algebras and Jordan multialgebras, since a significant part of the classical Jordan algebra theory is based on the triple product [24, 28, 32] {K1, K2, K3} = K1K2K3 + K3K2K1. It is well-known that not every classical special Jordan algebra is complete. However, the examples given in textbooks are not subalgebras of Sym(Rn) -- the space of all symmetric n × n matrices. Additionally, even if there are incomplete subalgebras of Sym(Rn), it is still not clear if the presence of several multiplications would not change the situation. In fact, we have already shown [18] that all Jordan SO(2) and SO(3)- invariant multialgebras in important physical contexts, such as conductivity, elasticity, piezo-electricity, etc. are complete. In an attempt to resolve this apparent contradiction between classical theory of Jordan algebras and physically relevant examples we describe a class of SO(2) and SO(3)-invariant Jordan multialgebras. 2.3 SO(d)-invariant Jordan multialgebras In this section it is important that T be a real finite dimensional representation of SO(d) (d = 2 or 3). The irreducible representations (irreps) are parametrized by non-negative integers, called weights. In the theory of composite materials, where one is interested in coupled thermal, electric and elastic properties of materials, the space T can contain only the irreps of weights 0, 1 and 2. The weight 0 irrep W0 is a 1-dimensional space with trivial group action. The weight 1 irrep W1 is Rd with the natural action of SO(d), while the weight 2 irrep W2 is the space Sym0(Rd) of symmetric trace-free d × d matrices with SO(d) acting by conjugation A 7→ RAR−1. Thus, all conceivable coupled field physical problems would be accommodated by T = W0 ⊗ Rn0 ⊕ W1 ⊗ Rn1 ⊕ W2 ⊗ Rn2 , n0 ≥ 0, n1 ≥ 0, n2 ≥ 0 (2.7) for an appropriate choice of n0, n1 and n2. The notation in (2.7) emphasizes that the group acts trivially on the second factors in tensor products above. We assume that the representation T is equipped with a fixed SO(d)-invariant inner product. The space Sym(T ) of symmetric maps on T has a natural action of SO(d) and splits into the direct sum of irreps up to weight 4. In addition to the representation T the algebraic structure of our problem is determined by the choice of a subrepresentation A ⊂ Sym(T ), which must be isomorphic to W2 ⊕ W4 as an SO(d) module (or only W2, if n2 = 0). The choice of such a subrepresentation is highly non-unique, and we assume that a generic choice is made. The goal is to answer the fundamental question of completeness for SO(d)-invariant Jordan A-multialgebras. Conjecture 2.5. Assume that T is given by (2.7) and A ⊂ Sym(T ) is isomorphic to W2 ⊕ W4 as an SO(3) module. We conjecture that any rotationally invariant A-multialgebra is complete. The rationale behind this conjecture is a positive result established in [18] (see also Section 3), when n0 = n2 = 0 and A = W2 ⊗ RIn1 , where In denotes the n × n identity matrix. The conjecture is stated only for SO(3)-invariant Jordan A-multialgebras, since this paper settles the SO(2)-invariant version of the conjecture in the negative (see Section 8). In addition to the results of Section 3 we have also computed a complete list of Jordan A-multialgebras for n0 = n1 = n2 = 1 in [18]. The conjecture was then verified by hand for each individual Jordan A-multialgebra. 3 Case study: Multifield response composite materials Multifield linear response materials were considered in [35, 34]. E = (∇φ1, . . . , ∇φn) induce n conjugate fluxes J = (j1, . . . , jn) satisfying In this context n coupled potential fields ∇ · j1 = . . . = ∇ · jn = 0. 5 Thermoelectric properties of composites would fit in this context with n = 2. For general values of n the space T is a direct sum of n copies of Rd: T = Rd ⊕ . . . ⊕ Rd = V1 ⊗ Rn, d = 2, 3, n {z } where V1 is the SO(d) irrep of weight 1, i.e. V1 = Rd with standard action of SO(d). The space A was computed in [18]. A = W2 ⊗ In ⊂ Sym(T ), where W2 ⊂ Sym(V1) is the SO(d) irrep of weight 2, denoting the space of symmetric, trace-free d×d matrices where SO(d) acts by conjugation. For the case d = 3 all SO(3)-invariant Jordan A-multialgebras have been computed in [18]. The result is given in Theorem 3.1 below. Theorem 3.1. Let Π be an SO(3)-invariant Jordan A-multialgebra in Sym(V1 ⊗ Rn), where A = W2 ⊗ In. Then there exists an associative subalgebra B of End(Rn), such that BT = B and Π = (End(W1) ⊗ B)sym. An immediate consequence of this theorem is that all SO(3)-invariant Jordan A-multialgebras are com- plete. When d = 2 we can label points in V1 = R2 with standard action of SO(2), by complex numbers, so that the action of Rθ ∈ SO(2) on z ∈ C is given by eiθz. Here Rθ denotes the rotation through the angle θ in counterclockwise (positive) direction. We write T = C ⊗ Rn = Cn, Rθ · u = eiθu, u ∈ Cn, Rθ ∈ SO(2). The SO(2)-invariant inner product on T is (u, v)T = ℜehu, vi, where h·, ·i is the standard Hermitean inner product on Cn. Every K ∈ End(T ) = EndR(Cn) is uniquely determined by two complex n × n matrices X and Y via its action on u ∈ Cn: Ku = Xu + Y u. Therefore, we will write K(X, Y ) to identify elements of End(T ). We easily compute Sym(T ) = {K(X, Y ) : X ∈ H(Cn), Y ∈ Sym(Cn)}, where H(Cn) denotes the set of all complex Hermitean n × n matrices. In this notation A = {K(0, zIn) : z ∈ C}, Rθ · K(X, Y ) = K(X, e2θY ). Therefore, an arbitrary SO(2) submodule of Sym(T ) is given by Π = ΠL,M = {(X, Y ) : X ∈ L ⊂ H(Cn), Y ∈ M ⊂ Sym(Cn)}, where L can be any real subspace of H(Cn and M can be any complex subspace of Sym(Cn). If an SO(2) submodule Π is also an A-multialgebra, then the subspaces L and M have to satisfy Y 2 + XX T ∈ M, Y X + XY ∗ ∈ L for all X ∈ L, Y ∈ M. The 3-chain condition is equivalent to cX1X T 2 X3 + cX3X T 2 X1 ∈ L, for all c ∈ C, and all {X1, X2, X3} ⊂ L. (3.1) (3.2) The 4-chain condition is a lot more complicated. The complete characterization of all solutions of (3.1) is unknown. However, some families of solutions can be easily identified. We will focus on one such family, since it will lead us to incomplete SO(2)-invariant Jordan A-multialgebras, in contrast to Theorem 3.1 in 3D. 6 Let Π0 ⊂ Sym(Rn) be a Jordan subalgebra of Sym(Rn), i.e. Π0 is a subspace of Sym(Rn) closed with respect to the Jordan product Then A ∗ B = 1 2 (AB + BA). L = {0}, M = Π0 ⊗ C (3.3) (3.4) solves (3.1). The 3-chain condition (3.2) is therefore trivially satisfied. The 4-chain condition from (2.6) reduces to the classical 4-chain condition Y1Y2Y3Y4 + Y4Y3Y2Y1 ∈ Π0 for all {Y1, Y2, Y3, Y4} ⊂ Π0 (3.5) Our next goal is to characterize all Jordan subalgebras Π0 of Sym(Rn) up to orthogonal conjugation. We observe that all Jordan subalgebras of Sym(Rn) are formally real, i.e. if X 2 1 + . . . + X 2 m = 0, m ≥ 1, {X1, . . . , Xm} ⊂ Sym(Rn) then X1 = . . . = Xm = 0. The complete classification of all formally real Jordan algebras, up to an isomorphism, has been achieved by Jordan, von Neumann and Wigner in [26]. Our goal is to go a little further and classify all faithful representations of formally real Jordan algebras in Sym(Rn). We note that completeness in the sense of Definition 2.4 is a property of the representation of an algebra, not of the algebra itself. In fact, our results will produce an example of two different faithful representations of the same algebra in Sym(Rn), one of which is complete, while the other is not. 4 Formally real Jordan algebras In this section we review the complete classification by Jordan, von Neumann and Wigner of all formally real Jordan algebras for the convenience of the reader. In addition to the classification itself, we will also need many of their intermediate results in [26]. The first set of statements (Theorem 4.1) refers to an arbitrary formally real Jordan algebra. Theorem 4.1. (a) There is a unique element IΠ ∈ Π such that IΠ ∗ A = A for all A ∈ Π. (b) The subalgebra hIΠ, Ai generated by A ∈ Π and IΠ contains pairwise orthogonal non-zero idempotents2 E1, . . . , Es such that for some {λ1, . . . , λs} ⊂ R. E1 + . . . + Es = IΠ, λ1E1 + . . . + λsEs = A (4.1) (c) An idempotent is called unresolvable, if it cannot be written as a sum of two orthogonal non-zero idem- potents. There exist unresolvable, pairwise orthogonal idempotents E1, . . . , Er such that E1 + . . . + Er = IΠ. (4.2) The resolution of unity (4.2) is not unique but the number r of unresolvable pairwise orthogonal idem- potents in it is always the same. (d) If A ⊂ Π is a proper ideal then Π = A ⊕ B as a direct sum of algebras, where B = {B ∈ Π : A ∗ B = 0, for all A ∈ A} is the complementary ideal. 2i.e. Eν ∗ Eσ = δνσ Eν , ν = 1, . . . , s, σ = 1, . . . , s. 7 (e) Any algebra Π is a direct (orthogonal) sum of simple algebras Π = Π1 ⊕ . . . ⊕ Πm, (4.3) where each component Πj is identified with a minimal proper ideal in Π. The second set of results (Theorem 4.2) refers to a simple formally real Jordan algebra Π, in which the decomposition of unity (4.2) has been chosen. Let Mρσ =(cid:26)A ∈ Π : Eτ ∗ A = 1 2 (δρτ + δστ )A for all τ = 1, . . . , r(cid:27) , ρ, σ = 1, . . . , r. Clearly, Mρσ = Mσρ. Theorem 4.2. (a) Π = M1≤ρ≤σ≤r Mρσ, as a direct sum of vector spaces. (b) dim Mσσ = 1, p = dim Mρσ is independent of σ and ρ, as long as σ 6= ρ. (c) The spaces Mρσ have bases {X ρσ 1 , . . . , X ρσ p } such that X ρσ µ ∗ X ρσ ν = δµν (Eρ + Eσ), µ, ν = 1, . . . , p, 1 ≤ ρ < σ ≤ r. The third result (Theorem 4.3) is the well-known characterization of all formally real simple Jordan algebras. Theorem 4.3. If Π is a formally real simple Jordan algebra then it is isomorphic to one of the following algebras classified according to the number of indecomposable idempotents in the resolution of unity (4.2) • r = 1, R • r = 2, "spin factors" Sn, n ≥ 3 defined by Sn = Span{ISn , s1, . . . , sn−1}, where ISn ∗ ISn = ISn , ISn ∗ sj = sj, si ∗ sj = δijISn . • r ≥ 3, Sym(Rr), H(Cr), H(Hr), denoting real symmetric r × r matrices, complex Hermitean r × r matrices and quaternionic Hermitean r × r matrices, respectively. • 27-dimensional exceptional Albert algebra M8 3 (r = 3). It has no non-trivial representations in Sym(Rn). It was shown by Albert [3] that M8 3 cannot be identified as a subspace of an associative algebra that is closed under the multiplication (3.3). Remark 4.4. It will be useful to list the value of the invariant p = dim Mρσ, defined in Theorem 4.2, for all the simple algebras from Theorem 4.3. • p = 1 for Sym(Rr), r ≥ 1 (Sym(R2) ∼= S3), • p = 2 for H(Cr), r ≥ 2 (H(C2) ∼= S4), • p = 3 for S5 • p = 4 for H(Hr), r ≥ 2 (H(H2) ∼= S6), • p ≥ 5 for Sp+2. For M8 3 p = 8. 8 5 The structure of Jordan subalgebras of Sym(Rn) 5.1 The non-singular Jordan algebras From now on Π refers to a subalgebra of Sym(Rn), i.e. a subspace closed with respect to the multiplication (3.3). Let Then, in the basis which is the union of the bases for V = N (Π)⊥ and N (Π) N (Π) = {u ∈ Rn : Au = 0 for all A ∈ Π}. Π =(" A 0 0 # : A ∈ Π0 ⊂ Sym(V )) , 0 (5.1) where Π0 is a Jordan subalgebra of Sym(V ), for which N (Π0) = {0}. Definition 5.1. We say that the Jordan algebra Π is non-singular if N (Π) = {0}. Any algebra Π is non-singular on V = N (Π)⊥. Hence, without loss of generality, we may assume that Π is non-singular. Lemma 5.2. Suppose that the Jordan algebra Π ⊂ Sym(Rn) is non-singular. Then, In ∈ Π, where In is the n × n identity matrix. Proof. By Theorem 4.1 there exists the algebra identity IΠ ∈ Π. Then I2 Π = IΠ. Therefore, the symmetric matrix IΠ may have eigenvalues that are either 1 or 0. Suppose u ∈ Rn is an eigenvector of IΠ with eigenvalue zero. Then, by assumption, there exists A ∈ Π, such that v = Au 6= 0. Applying the equality 2A = AIΠ +IΠA to the vector u we obtain 2v = IΠv. This contradicts the fact that 2 is not an eigenvalue for IΠ. Thus, IΠ may not have an eigenvalue 0. Therefore, IΠ = In. Let A ∈ Π. Suppose that its eigenvalues are {λ1, . . . , λs} ⊂ R and the corresponding eigenspaces are Vα, α = 1, . . . , s. By Theorem 4.1 the idempotents Eα in (4.1) must be orthogonal projections PVα onto the eigenspaces Vα of A. We will call them spectral projections. Thus, for any A ∈ Π all of its spectral projections PVα must also be in Π, and A = λαPVα , V1 ⊕ . . . ⊕ Vs = Rn, (5.2) sXα=1 5.2 Splitting of Jordan algebras Definition 5.3. We say that the Jordan algebra Π ⊂ Sym(Rn) splits over the orthogonal decomposition Rn = V1 ⊕ . . . ⊕ Vm, if Π = Π1 ⊕ . . . ⊕ Πm, is a direct sum of Jordan algebras, where Πα = {A ∈ Π : Aw = 0 for all w ∈ V ⊥ α }, α = 1, . . . , m. If Π splits over Rn = V1 ⊕ . . . ⊕ Vm, then in the basis, which is the union of bases for Vα, α = 1, . . . , m the algebra Π has the form Π =  A1 0 . . . 0 Am  : Aα ∈ Π0 α ⊂ Sym(Vα), α = 1, . . . , m . In other words all matrices in Π have block-diagonal structure, with independent blocks Aα taken from subalgebras Π0 α ⊂ Sym(Vα) that are isomorphic to Πα. Clearly, each Πα ⊂ Π is an ideal. Conversely, if A ⊂ Π is an ideal, then, according to Theorem 4.1, there exists a complementary ideal such that Π = A ⊕ B, as a direct sum of algebras. B = {B ∈ Π : A ∗ B = 0 for all A ∈ A}, 9 Lemma 5.4. Let A and B be the complementary pair of ideals in Π. Then there exists V ⊂ Rn such that Π = A ⊕ B is the splitting of Π over Rn = V ⊕ V ⊥. Proof. Let V = N (B). The subspace V cannot be all of Rn, since otherwise B = {0} and A = Π. If V = {0}, then B is a non-singular algebra and In ∈ B, by Lemma 5.2. But then B = Π and A = {0}. The subspace V ⊥ is invariant for B, since all matrices in B are symmetric. Thus, B is isomorphic to a subalgebra B0 of Sym(V ⊥) defined by the restriction of B on V ⊥. By our construction N (B0) = {0} and, applying Lemma 5.2 to B0 we conclude that PV ⊥ ∈ B. Now, let B ∈ Π be such that Bv = 0 for all v ∈ V . Then PV ⊥ ∗ B = B. Thus, B ∈ B, since B is an ideal. We have now proved that B = {B ∈ Π : Bv = 0 for all v ∈ N (B)}. To finish the proof of the lemma we need to show that N (A) = V ⊥. First we observe that V (and therefore V ⊥) is an invariant subspace for A. Indeed, for any v ∈ V and any A ∈ A we have since PV ⊥ ∈ B. Also, for any w ∈ V ⊥ and any A ∈ A we have PV ⊥(Av) = 2(PV ⊥ ∗ A)v − A(PV ⊥v) = 0, Aw = PV ⊥(Aw) = 2(PV ⊥ ∗ A)w − A(PV ⊥w) = −Aw. Thus, V ⊥ ⊂ N (A). To prove the reverse inclusion we observe that PV ∗ B = 0 for all B ∈ B. Indeed, B ∗ PV ⊥ = B for all B ∈ B, therefore B ∗ PV = B ∗ (In − PV ⊥ ) = B − B = 0. It follows that PV ∈ A. Thus, if Ax = 0 for all A ∈ A, then PV x = 0 and x ∈ V ⊥. Corollary 5.5. Let (4.3) be the decomposition of Π into a direct sum of simple algebras. Then, there exist pairwise orthogonal subspaces V1, . . . , Vm of Rn such that (4.3) is the splitting of Π over the orthogonal decomposition Rn = V1 ⊕ . . . ⊕ Vm. Thus, , (5.3) Π =  A1 0 . . . 0 Am  : Aα ∈ Πα ⊂ Sym(Vα), α = 1, . . . , m where Πα ⊂ Sym(Vα), α = 1, . . . , m are simple Jordan algebras. 5.3 Irreducible Jordan algebras According to Corollary 5.5 we need to understand the structure of simple non-singular Jordan algebras Π ⊂ Sym(Rn). As in the theory of associative algebras it is important whether or not there is a common invariant subspace for all matrices A ∈ Π. Any Jordan algebra is completely reducible in the sense that there exists an orthogonal decomposition Rn = V1 ⊕ . . . ⊕ Vk, such that all subspaces Vα are invariant for all matrices A ∈ Π and they do not contain any smaller proper invariant subspaces for Π. Definition 5.6. A Jordan subalgebra of Sym(Rn) is called irreducible if it does not have any proper invariant subspaces in Rn. We remark that an irreducible Jordan subalgebra of Sym(Rn) is just a faithful irreducible representation of a simple formally real Jordan algebra. Our goal is to describe the structure of an arbitrary simple non-singular subalgebra of Sym(Rn) in terms of irreducible algebras. Lemma 5.7. Let Π be a simple non-singular subalgebra of Sym(Rn) and Rn = V1 ⊕ . . . ⊕ Vk be an orthogonal decomposition of Rn into the sum of irreducible invariant subspaces. Let Πα = {PVαAPVα : A ∈ Π} ⊂ Sym(Vα), α = 1, . . . , k. 10 Then Πα ⊂ Sym(Vα), α = 1, . . . , k are irreducible Jordan algebras in the sense of Definition 5.6. Moreover, the maps Tαβ : Πα → Πβ defined as are Jordan algebra isomorphisms. TαβA = PVβ KPVβ , A = PVαKPVα, K ∈ Π Proof. Let A ∈ Πα and suppose {K1, K2} ⊂ Π are such that A = PVα K1PVα = PVα K2PVα. Then PVα(K1 − K2)PVα = 0. Let A = {K ∈ Π : PVαKPVα = 0}. Let us show that A is an ideal in Π. Let X ∈ Π and K ∈ A be arbitrary. For any x ∈ Rn we have x = v + w, where v ∈ Vα and w ∈ V ⊥ α . Obviously PVα (X ∗ K)PVαw = 0. We also have 2PVα(X ∗ K)PVα v = PVα KXv, since Kv = 0 for any v ∈ Vα. But Xv ∈ Vα, since Vα is an invariant subspace for Π, and thus, KXv = 0. It follows that X ∗ K ∈ A. The ideal A cannot be Π, since in that case Vα ⊂ N (Π), contrary to the assumption that Π is a non-singular algebra. Hence, A = {0}. Hence, for any A ∈ Πα there is a unique K ∈ Π for which A = PVα KPVα. Moreover, for any {K1, K2} ⊂ Π we have (PVα K1PVα ) ∗ (PVα K2PVα ) = PVα (K1 ∗ K2)PVα . Indeed, the invariance of Vα implies that for any v ∈ Vα PVα K1PVαPVα K2v = PVα K1K2v = K1K2v. It follows that Πα ⊂ Sym(Vα) are irreducible Jordan algebras. The maps Tαβ are well-defines and preserve Jordan multiplication Tαβ(A1 ∗ A2) = PVβ (K1 ∗ K2)PVβ = (PVβ K1PVβ ) ∗ (PVβ K2PVβ ) = (TαβA1) ∗ (TαβA2). By construction, the maps Tαβ are surjective. Let us show that they are also injective. If A ∈ Πα and A 6= 0, then there is a unique K ∈ Π such that A = PVα KPVα. Clearly, K 6= 0. If TαβA = PVβ KPVβ = 0, then K ′ = 0 and K 6= K ′ satisfy PVβ KPVβ = PVβ K ′PVβ in contradiction of uniqueness. Hence, we have proved the structure theorem for simple non-singular Jordan subalgebras of Sym(Rn). Theorem 5.8. Let Π ⊂ Sym(Rn) be a non-singular simple Jordan algebra. Then there exists an orthonormal basis (o.n.b.) of Rn in which the algebra Π has the form , (5.4) Π =  A 0 T1A . . . 0 Tk−1A  : A ∈ Π0 ⊂ Sym(RN ) where Π0 ⊂ Sym(RN ) is an irreducible Jordan algebra and T1, . . . , Tk−1 are Jordan algebra isomorphisms. Thus, we have reduced the problem of description of all subalgebras of Sym(Rn) to the problem of char- acterization of all irreducible representations of simple Jordan algebras and their isomorphisms. The problem of completeness of Π can also be restated. It is now clear that for a Jordan algebra to be complete in the sense of Definition 2.4 it is necessary and sufficient for each of its simple components Πα, to be complete. The latter condition will be satisfied if and only if the irreducible algebras Π0 and Πα = TαΠ0, α = 1, . . . , k − 1, in the representation (5.4) are complete and Jordan isomorphisms Tα satisfy Tα(A1A2A3A4 + A4A3A2A1) = TαA1TαA2TαA3TαA4 + TαA4TαA3TαA2TαA1. (5.5) 11 6 The characterization of irreducible Jordan algebras To complete the characterization of all Jordan subalgebras Sym(Rn) we need to describe all irreducible Jordan algebras up to an orthogonal equivalence and compute their Jordan isomorphisms. In this section we take on the former problem. 6.1 Block decomposition Let J be a simple formally real Jordan algebra. Let Φ : J → Sym(Rn) be an irreducible representation of J. According to Theorem 4.2 there exist r indecomposable orthogonal idempotents {E1, . . . , Er} ⊂ J. Their images under the representation Φ must necessarily be orthogonal projectors onto the mutually orthogonal subspaces W1, . . . , Wα such that Rn = W1 ⊕ . . . ⊕ Wr. The algebra Π = Φ(J) can be represented as a direct sum Π =L1≤ρ≤σ≤r Mρσ (block decomposition), where Mαα = RPWα, PWα = Φ(Eα), α = 1, . . . , r, and Mαβ = {PWαAPWβ + PWβ APWα : A ∈ Π}, 1 ≤ α < β ≤ r. The spaces Mαβ have the same dimension p for all 1 ≤ α < β ≤ r and have the basis X αβ µ = PWα AµPWβ + PWβ AµPWα such that In particular, It follows that X αβ µ ∗ X αβ ν = δµν(PWα + PWβ ), µ, ν = 1, . . . , p. (X αβ µ )2 = PWα AµPWβ AµPWα + PWβ AµPWα AµPWβ = PWα + PWβ . From the first equality we obtain that PWα AµPWβ AµPWα = PWα , PWβ AµPWα AµPWβ = PWβ . dim(Wα) = rank(PWα ) = rank(PWα AµPWβ AµPWα ) ≤ dim(Wβ), while from the second one we obtain dim(Wβ) = rank(PWβ ) = rank(PWβ AµPWα AµPWβ ) ≤ dim(Wα). Thus, dim(Wα) = d = n/r for all α = 1, . . . , r. 6.2 Structure of Mαβ-blocks (6.1) (6.2) In order to continue, it will be convenient to fix arbitrary orthonormal bases for spaces Wα and work with the space U αβ of d × d matrices representing the αβ-block of Mαβ. The n × n symmetric matrices X αβ µ ∈ Mαβ µ ∈ U αβ. When α 6= β the relation (6.2) for X αβ µ µ )T = 0, µ 6= ν. ν (bX αβ p )T , µ = 1, . . . , p. Then Y αβ p = Id and ν + Y αβ ν Y αβ µ = 0, 1 ≤ µ < ν ≤ p − 1. (6.3) given by (6.1) can be described by their αβ d × d blocks bX αβ can be written in terms of matrices bX αβ ν )T + bX αβ µ (bX αβ It follows that {bX αβ 1 , . . . , bX αβ µ (bX αβ bX αβ µ = bX αβ µ (bX αβ bX αβ 1 , . . . , bX αβ µ )T = 0, Y αβ µ Y αβ p } ⊂ O(d). Let Y αβ (Y αβ µ )2 = −Id, Y αβ µ + (Y αβ µ )T = Id, as follows p We consider two cases r = 2 and r > 2. 12 6.2.1 Case r > 2 When r > 2 Remark 4.4 implies that J can only be Sym(Rr), for which p = 1, H(Cr), for which p = 2, or H(Hr), for which p = 4. The explicit description of the irreducible representations of these three algebras will be written in terms of the canonical matrix representations of complex numbers and quaternions. Specifically we define the maps ϕ, ψ : C → EndR(R2) and Q : H → EndR(R4) as follows y −x # , x # , ϕ(x + iy) =" x −y ϕ(q0 + iq1) # . Q(q0 + iq1 + jq2 + kq3) =" ϕ(q0 + iq1) −ψ(q2 + iq3) ψ(x + iy) =" x ψ(q2 + iq3) y y (6.4) (6.5) These functions have the properties ϕ(a)z = az, ψ(a)z = az, Q(q)h = qh. Additionally ϕ(a)T = ϕ(a), ψ(a)T = ψ(a), Q(q)T = Q(q). For a collection of d × d matrices Mαβ, 1 ≤ α ≤ r, 1 ≤ β ≤ r the notation (Mαβ) stands for the dr × dr matrix given in block-form by the r × r block-matrix, whose αβ-blocks are d × d matrices Mαβ. Theorem 6.1. Up to an orthogonal conjugation a simple formally real special Jordan algebra J with r ≥ 3 has a unique irreducible representation by matrices in Sym(Rn), given explicitly by the following formulas (a) J = Sym(Rr). Then n = r and Φ(J) = J, J ∈ Sym(Rr) (b) J = H(Cr). Then n = 2r and H(Cr) ∋ J 7→ Φ(J) = (ϕ(Jαβ )). (c) J = H(Hr). Then n = 4r and H(Hr) ∋ J 7→ Φ(J) = (Q(Jαβ)). Proof. Let K denote the division algebra R, C or H in the definition of the algebra J. For an irreducible representation Φ : J → Sym(Rn) (n = dr) and α 6= β let Φαβ : K → U αβ, while Φαα(1) = Id, so that Φ(J) = (Φαβ(Jαβ )). The maps Φαβ satisfy Φαβ(q)T = Φβα(q), Φαβ(q)Φαβ (q)T = q2Id, Φαβ(q)Φβγ(h) = Φαγ(qh), α 6= γ In particular, Φαβ(1) ∈ O(d). Let R = Id 0 Φ12(1) . . . 0 Φ1r(1) .  Then R ∈ O(n) and Ψ(J) = RΦ(J)RT is an orthogonally equivalent irreducible representation of J in Sym(Rn), such that Ψ1α(1) = Id, α = 1, . . . , r. Let α 6= β and β 6= 1. Then If in addition, α 6= 1 (which is possible if and only if r ≥ 3), then Ψαβ(q) = Ψ1α(1)Ψαβ(q) = Ψ1β(q). Ψαβ(q) = Ψβα(q)T = Ψ1α(q)T = Ψα1(q) Hence, when α 6= β, α 6= 1, β 6= 1 we have Ψαβ(q)Ψαβ(h) = Ψα1(q)Ψ1β(h) = Ψαβ(qh). 13 For any β 6= 1 there is α 6= 1 and α 6= β, since r ≥ 3. Then Ψ1β(q) = Ψαβ(q). Similarly, for any α 6= 1 there is β 6= 1 and α 6= β, and it follows that Ψα1(q) = Ψαβ(q). Hence, for any α 6= β the functions Ψαβ are algebra homomorphisms. In particular, if α 6= β we have Ψβα(q) = Ψαβ(q)T = q2(Ψαβ(q))−1 = Ψαβ(q2q−1) = Ψαβ(q). In particular, Ψ21(q) = Ψ12(q). Also, for any α 6= 1, 2 Ψ1α(q) = Ψα1(q) = Ψα2(q) = Ψ12(q). Finally, if α 6= β, α 6= 1, β 6= 1 then Ψαβ(q) = Ψ1β(q) = Ψ12(q). Thus, there is a division algebra homomorphism Ψ12 : K → EndR(Rd) such that Ψ12(q)T = Ψ12(q) and Ψαβ(q) = Ψ12(q) for all α 6= β and all q ∈ K. The homomorphism Ψ12 is non-zero and hence injective (since K is a simple algebra over R). If the associative algebra U = Ψ12(K) ⊂ EndR(Rd) has a proper invariant subspace V ⊂ Rd then the subspace V ⊕ . . . ⊕ V ⊂ Rd ⊕ . . . ⊕ Rd = Rn r times r times } {z } {z is a proper invariant subspace of the representation Ψ(J). We conclude that Ψ12 must be an irreducible representation of K for an irreducible representation Ψ of J. It remains to note that there is a unique3 (up to an orthogonal conjugation) irreducible representation Ψ12 of K on Rd satisfying Ψ12(q)T = Ψ12(q). It is given explicitly by the following formulas • If K = R then d = 1 and Ψ12(q) = q. • If K = C then d = 2 and Ψ12(q) = ϕ(q), given by (6.4). • If K = H then d = 4 and Ψ12(q) = Q(q), given by (6.5). 6.2.2 Case r = 2 When r = 2 we have Π =(" λIn/2 AT A µIn/2 # : {λ, µ} ⊂ R, A ∈ U) . The conditions (6.3) on the basis (I, Y1, . . . , Yp−1) of U 12 = U are both necessary and sufficient for Π to be closed with respect to the Jordan product (3.3). The irreducibility condition is equivalent to the requirement that all matrices in U have no common proper invariant subspace. Indeed, if U has a proper invariant subspace V then the space V = {(v1, v2) : {v1, v2} ⊂ V } is a proper invariant subspace for Π. If U has no proper invariant subspaces and if V is a proper invariant subspace of Π then (v, w) ∈ V implies that (v, 0) = K1(v, w) ∈ V and (0, w) = K2(v, w) ∈ V, where K1 ∈ Π corresponds to λ = 1, µ = 0, A = 0 and K2 ∈ Π corresponds to λ = 0, µ = 1, A = 0. Then there are subspaces V1 and V2 of Rn/2 such that V = {(v1, v2) : v1 ∈ V1, v2 ∈ V2}. The invariance of V is then equivalent to the condition that U maps V1 into V2 and V2 into V1. However, In/2 ∈ U and hence, if v1 ∈ V1 then v1 ∈ V2 and conversely, if v2 ∈ V2 then v2 ∈ V1. It follows that V1 = V2 = V . But then V must be an invariant subspace for U. Therefore, V is either {0} or Rn/2, in which case V is also either {0} or Rn. The problem of identifying subspaces U as described above is called a real Radon-Hurwitz problem [23, 42], who studied it in connection with the question of composition of quadratic forms. This problem has 3For the sake of completeness this statement will be a consequence of our analysis of the case r = 2. See Remarks 6.4 and 6.5 below. 14 connections to self-dual 2-forms [8], linearly independent vector fields on spheres [1], system of hyperbolic conservation laws [2], Clifford algebras [9], etc. In particular the matrices Y1, . . . , Yp−1 in (6.3) are the generators of the 2p−1-dimensional Clifford algebra Cℓp−1,0(R). We are interested in an explicit and complete characterization of all possible subspaces U as above. While Hurwitz solved this problem in the complex case, his solution is difficult to adapt to the real case. Here we present the alternative solution based on the representation theory of finite groups [14]. Following [14] we let Gp be a finite group with generators ε, a1, . . . , ap−1, p ≥ 2, satisfying the relations ε2 = 1, a2 k = ε, εak = akε, akal = εalak, k, l = 1, . . . , p − 1, k 6= l. Then, the matrices Y1, . . . , Yp−1 satisfying (6.3) describe an irreducible orthogonal representation of the group Gp that sends ε to −In/2. Eckmann [14] has computed the number, type, and dimensions of such irreps of Gp (see Appendix A). His results are summarized in the following theorem. Theorem 6.2 (Eckmann). All non-isomorphic irreducible real representations of Gp that map ε to −I are characterized as follows. • p = 1, 7 mod 8. Then the unique real irrep Vp is a representation of real type and d(p) = dim Vp = 2 p−1 2 . • p = 3, 5 mod 8. Then the unique real irrep Vp is a representation of quaternionic type and d(p) = dim Vp = 2 p+1 2 . • p = 2, 6 mod 8. Then the unique real irrep Vp is a representation of complex type and d(p) = dim Vp = p 2 . 2 • p = 0 mod 8. Then there are two distinct irreps V + p and V − p both of real type and d(p) = dim V ± p = p−2 2 . 2 • p = 4 mod 8. Then there are two distinct irreps V + p and V − p dim V ± p = 2 p 2 . both of quaternionic type and d(p) = We note that the function d(p) given explicitly in Theorem 6.2 can also be defined recursively by d(p + 8) = 16d(p), d(1) = 1, d(2) = 2, d(3) = d(4) = 4, d(5) = d(6) = d(7) = d(8) = 8. (6.6) It remains to construct the representations Vp explicitly. It is sufficient to indicate the images of a1, . . . , ap−1 (since it is required that ε 7→ −Id(p)). To describe the answer we need to introduce the following notation. where ϕ was given in (6.4). The function bQ satisfies bQ(q0 + iq1 + jq2 + kq3) =" ϕ(q0 + iq1) bQ(q1)bQ(q2) = bQ(q1q2), For {q, h} ⊂ H we also define where the maps Q and bQ are defined in (6.5) and (6.7), respectively. The map O has the following properties O(q, h)T = O(q, −h), O(q, h)O(q, h)T = (q2 + h2)I8. For 2 ≤ p ≤ 9 we have the following explicit representations, which are slightly modified versions of the ones in [8]. 15 −ϕ(q2 − iq3) ϕ(q0 − iq1) # , ϕ(q2 + iq3) bQ(q1)Q(q2) = Q(q2)bQ(q1). bQ(q)T = bQ(q), −bQ(h) Q(q) # , O(q, h) =" Q(q) bQ(h) (6.7) (6.8) List 6.3. • p = 2, d(p) = 2, ρ2(a1) = ϕ(i). Remark 6.4. The uniqueness of the irreducible representation V2 implies that up to an orthogonal conjugation the irreducible representation of C on EndR(Rd) satisfies d = 2 and is given by C ∋ c 7→ ϕ(c). • p = 3, d(p) = 4, ρ3(a1) = Q(i), ρ3(a2) = Q(j) • p = 4, d(p) = 4. There are two non-isomorphic representations: ρ± 4 (a1) = ±Q(i), ρ± 4 (a2) = ±Q(j), ρ± 4 (a3) = ±Q(k). Indeed, ρ+ 4 (a1a2) = ρ+ 4 (a3), ρ− 4 (a1a2) = −ρ− 4 (a3). Remark 6.5. Let Ψ0 : H → EndR(Rd) be a representation of H. If we define ρ(ε) = −Id, ρ(a1) = Ψ0(i), ρ(a2) = Ψ0(j), ρ(a3) = Ψ0(k) then ρ will be a representation of G4 on Rd. Clearly, ρ is irreducible if and only if Ψ0 is irreducible. Thus, d = 4 and, up to the orthogonal conjugation, either Ψ0(i) = Q(i), Ψ0(j) = Q(j), Ψ0(k) = Q(k), or Ψ0(i) = −Q(i), Ψ0(j) = −Q(j), Ψ0(k) = −Q(k). However, the latter choice results in Ψ0(ij) = −Ψ0(i)Ψ0(j). Hence, up to an orthogonal conjugation Ψ0(q) = Q(q). • p = 5, d(p) = 8, ρ5(a1) = O(0, 1), ρ5(a2) = O(0, i), ρ5(a3) = O(0, j), ρ5(a4) = O(0, k). • p = 6, d(p) = 8, • p = 7, d(p) = 8, ρ6(a1) = O(0, i), ρ6(a2) = O(0, j), ρ6(a3) = O(0, k), ρ6(a4) = O(i, 0), ρ6(a5) = O(j, 0). ρ7(a1) = O(0, i), ρ7(a2) = O(0, j), ρ7(a3) = O(0, k), ρ7(a4) = O(i, 0), ρ7(a5) = O(j, 0), ρ7(a6) = O(k, 0). • p = 8, d(p) = 8: There are two non-isomorphic representations: ρ− 8 (ai) = −ρ+ 8 (ai), i = 1, . . . , 7 ρ− 8 (a1) = O(0, 1), ρ− 8 (a2) = O(0, i), ρ− 8 (a3) = O(0, j), ρ− 8 (a4) = O(0, k), ρ− 8 (a5) = O(i, 0), ρ− 8 (a6) = O(j, 0), ρ− 8 (a7) = O(k, 0). The irrep ρ− 8 is not isomorphic to ρ+ 8 because whereas ρ+ 8 (a2a3a4a5a6a7) = ρ+ 8 (a1), ρ− 8 (a2a3a4a5a6a7) = −ρ− 8 (a1), 16 • p = 9, d(p) = 16. ρ9(a1) = ψ(i) ⊗ O(i, 0), ρ9(a2) = ψ(i) ⊗ O(j, 0), ρ9(a3) = ψ(i) ⊗ O(k, 0), ρ9(a4) = ψ(i) ⊗ O(0, i), ρ9(a5) = ψ(i) ⊗ O(0, j), ρ9(a6) = ψ(i) ⊗ O(0, k), ρ9(a7) = ψ(i) ⊗ O(0, 1), ρ9(a8) = ϕ(i) ⊗ O(1, 0). For m1 × n1 matrix A and m2 × n2 matrix B the tensor product notation A ⊗ B denotes m1m2 × n1n2 matrix written in block-form as A ⊗ B = a11B . . . . . . . . . am11B . . . am1n1B  . a1n1B . . . Suppose that representations ρp of Gp have been constructed for 2 ≤ p ≤ 9. For p ≥ 10 we define ρp(ai) = ρ9(ai) ⊗ Id(p−8), i = 1, . . . , 8, ρp(ai) = B ⊗ ρp−8(ai), i = 9, . . . , p − 1, (6.9) where B = ψ(1) ⊗ I8 ∈ Sym(R16). We have B2 = I16. Therefore, ρp(ai)2 = −Ip and ρp(ai)ρp(aj) = −ρp(aj)ρp(ai) if i 6= j and both i and j are either below 9 or above 8. We need to check the anticommutativity property for i ≤ 8 and j ≥ 9: ρ9(ai)B ⊗ ρp−8(aj ) = −Bρ9(ai) ⊗ ρp−8(aj). This will be satisfied if We can verify (6.10) explicitly via the formulas for ρ9 below. In addition to the explicit representations Vp it is also convenient to have an explicit form of the subspaces ρ9(ai)B = −Bρ9(ai). (6.10) Wp = Span{ρp(a1), . . . , ρp(ap−1)} ⊂ Skew(Rd(p)), U = Up = RId(p) ⊕ Wp. List 6.6. • p = 1, W1 = {0} ⊂ R, U1 = R • p = 2, • p = 3, • p = 4, • p = 5, • p = 6, W2 = Rϕ(i) = Skew(R2), U2 = {ϕ(z) : z ∈ C}. W3 = {Q(q) : q ∈ H, q = q1i + q2j}. U3 = {Q(q) : q ∈ H, q = q0 + q1i + q2j}. W4 = {Q(q) : q ∈ H, ℜe(q) = 0}, U4 = {Q(q) : q ∈ H}. W5 = {O(0, h) : h ∈ H}, U5 = {O(λ, h) : h ∈ H, λ ∈ R}. W6 = {O(q, h) : {q, h} ⊂ H, ℜe(h) = 0, q = q1i + q2j} . U6 = {O(q, h) : {q, h} ⊂ H, ℜe(h) = 0, q = q0 + q1i + q2j} . 17 • p = 7, • p = 8, W7 = {O(q, h) : {q, h} ⊂ H, ℜe(q) = 0, ℜe(h) = 0} , U7 = {O(q, h) : {q, h} ⊂ H, ℜe(h) = 0} . W8 = {O(q, h) : {q, h} ⊂ H, ℜe(q) = 0} , U8 = {O(q, h) : {q, h} ⊂ H} . The formula (6.9) results in Wp =(" I8 ⊗ A O(q, h) ⊗ Id(p−8) −O(q, h)T ⊗ Id(p−8) −I8 ⊗ A # : A ∈ Wp−8, {q, h} ⊂ H) , p ≥ 9, (6.11) where the spaces W1, . . . W8 are given in the List 6.6 and O(q, h) is defined in (6.8). We remark that the formulas for Wp and Up, p = 1, . . . , 8 are somewhat arbitrary, since conjugation by any orthogonal transformation would produce equally valid formulas. However, when p = 1, 2, 4 and 8 the spaces Wp and Up are O(d(p))-invariant and therefore represent the canonical forms. The formula (6.11) implies that all spaces W2s and U2s are canonical. It is important to note that the two different representations of Gp when p = 0 mod 4 result in different representations of the spin factors SN , when N = 2 mod 4 (N ≥ 6), even though the images of ΠSN of SN under both representations are the same: ΠSN =(" λId(N −2) AT A µId(N −2) # : {λ, µ} ⊂ R, A ∈ UN −2 = RId(N −2) ⊕ WN −2) , (6.12) where Wp is given by (6.11). The existence of the two non-isomorphic representations of Gp is reflected in the existence of the map T : ΠSN → ΠSN , which maps ρ+ to ρ−. As such it is a Jordan algebra automorphism that cannot be written as an orthogonal conjugation. Conversely, every Jordan algebra automorphism maps one representation of Gp into another. For those p for which such a representation is unique the automorphism must be an orthogonal conjugation. The Jordan algebra automorphism T : ΠSN → ΠSN can be written explicitly as T" λId(N −2) AT A µId(N −2) # =" λId(N −2) A AT µId(N −2) # , A ∈ UN −2, {λ, µ} ⊂ R. (6.13) Thus, we have proved the following characterization of all irreducible representations of spin factors SN , N ≥ 3 (S1 = R and S2 is not simple) in Sym(Rn). Theorem 6.7. Each spin factor SN , N ≥ 3 is represented by a unique (up to the orthogonal conjugation) irreducible subalgebra ΠSN ⊂ Sym(R2d(N −2)) given by (6.12). Moreover, each Jordan algebra automorphisms of ΠSN can be represented by an orthogonal conjugation, unless N = 2 mod 4 (N ≥ 3), in which case each Jordan algebra automorphisms of ΠSN can be represented by a composition of the map T given by (6.13) and an orthogonal conjugation. 7 Structure theorem and completeness We can now summarize all our results and describe explicitly, up to an orthogonal conjugation all Jordan subalgebras of Sym(Rn). Theorem 7.1 (Jordan subalgebras of Sym(Rn)). 18 Π =  A1 0 . . . Am 0 0  : A1 ∈ Π1, . . . , Am ∈ Πm , (i) Let Π be a Jordan subalgebra of Sym(Rn). Then there exists an o.n.b. of Rn in which where each Πα is a simple non-singular Jordan subalgebra of Sym(Rdα), for some dα ≥ 1 for which d1 + . . . + dm ≤ n. (ii) Let Π be a simple non-singular Jordan subalgebra of Sym(Rn). Then there exits an o.n.b. in Rn in which Π has one of the following forms (a) Π = {In/r ⊗ A : A ∈ Sym(Rr)}, r ≥ 1; (b) Π = {In/2r ⊗ A : A ∈ H(Cr)}, r ≥ 2; (c) Π = {In/4r ⊗ A : A ∈ H(Hr)}, r ≥ 2; (d) Π = {In/2d(N −2) ⊗ A : A ∈ ΠSN }, N = 5, 7, 8, . . ., where ΠSN is given by (6.12) and function d(·) is defined in (6.6); (e) Π =(" Is1 ⊗ A 0 0 Is2 ⊗ T A # : A ∈ ΠSN), s1 +s2 = n/2d(N −2), s1 > 0, s2 > 0, N = 2 mod 4, N ≥ 6, where the Jordan automorphism T : ΠSN → ΠSN is given by (6.13). The explicit characterization of all Jordan subalgebras of Sym(Rn) in Theorem 7.1 allows us to answer a question about completeness of a subalgebra Π ⊂ Sym(Rn). By part (i) of Theorem 7.1 we can write Π = Π1 ⊕ . . . ⊕ Πm, where Πα is a simple non-singular subalgebra of Sym(Rdα). The algebra Π will be complete if and only if each of the algebras Πα is orthogonally equivalent to one of the algebras in cases (a), (b) or (c) in part (ii) of Theorem 7.1. We remark on a single exception to the statement that completeness is determined by the isomorphism class of Π. The algebras Π in part (ii)(e) N = 6 and part (ii)(c) r = 2 are isomorphic to H(H2). However, the algebras in part (ii)(c) r = 2 are complete, while the ones in part (ii)(e) N = 6 are not. Suppose now that we are looking for all subalgebras of Sym(Rn) that are closed not only with respect to the product (3.3) but also with respect to the products (2.3), where the subspace A of multiplications is spanned by finitely many matrices A1 = In, A2, . . . , As. Let Π be a non-singular A-multialgebra. Then, by Lemma 5.2, In ∈ Π. Therefore, Aj = InAjIn ∈ Π, j = 2, . . . , s, and hence, each multiplication X ∗Aj Y is a mutation of the standard multiplication (3.3) [28]. We first choose a basis in which Π has the form described in the Structure Theorem 7.1. In that basis the submatrices of A2, . . . , As corresponding to N (Π)⊥ must be block-diagonal, with each diagonal block being a member of a simple component Πα of Π. That is also sufficient, since the triple product KαAαLα + LαAαKα always belongs to Πα. If we exclude the special case when the αα-block Aα of A has the form Aα = Ik ⊗ A0 we will exclude all cases in which the simple algebra Πα is reducible. In particular, this would exclude all part (ii)(e) cases. Hence, the question of completeness of A-multialgebras reduces to the question of completeness of A-multialgebras on the invariant subspaces of A. If A has no invariant subspaces then Π must be irreducible. An irreducible algebra is always complete if n is not a power of 2 or is less than 8. Even though possibilities like Aα = Ik ⊗ A0 cannot be ruled out in general, we may try to understand when incomplete algebras can arise generically. The subspace of Sym(Rn) of all matrices whose upper left 8 × 8 submatrix is in ΠS5 has co-dimension 31. Therefore, the space V0 of s-tuples of symmetric matrices (one of which is In), all of whose upper left 8 × 8 submatrices is in ΠS5 has co-dimension 31(s − 1). The dimension of the O(n) orbit of a generic s-tuple of matrices is n(n − 1)/2. The subspace V0 will intersect this O(n) orbit generically only if n(n − 1) ≥ 62(s − 1). For example, for a generic 5-dimensional space A (containing In), all Jordan multialgebras will be complete, if n < 17. These dimensional considerations explain the reason for completeness in all of the examples in [18, 16, 17, 21]. Conjecture 2.5 suggests that another way to eliminate failure of completeness may be to restrict the Jordan multialgebras to SO(3)-invariant ones. 19 8 Lamination exact relation that is not closed under homogeniza- tion Returning to the physical example of multifield response composite materials in Section 3 we see that the smallest n for which the incomplete Jordan subalgebra of Sym(Rn) provides an example on incomplete Jordan A-multialgebra is n = 8. However, the quaternionic formalism of Section 6 suggests an example of incomplete Jordan A-multialgebra when n = 44. Let L = {iQ(q) : q ∈ H, ℜe(q) = 0} ⊂ H(C4), M = {aI4 : a ∈ C} ⊂ Sym(C4), (8.1) where Q(q) is defined in (6.5). Let us verify that the 3-chain condition (3.2) fails. Indeed, let c = i, X1 = iQ(q1), X2 = iQ(q2), X2 = iQ(q3), where it will be convenient to identify the purely imaginary quaternions q1, q2 and q3 with vectors in R3. Let us assume that {q1, q2, q3} form a basis of R3, so that the mixed product (q1, q2, q3) = q1 · (q2 × q3) is non-zero. We compute Then X1X T 2 X3 = iQ(q1q2q3). iX1X T 2 X3 − iX3X T 2 X1 = Q(2(q1, q2, q3)) = 2(q1, q2, q3)I4 6∈ L. If we add real multiples of I4 to L, we will obtain a completion of our incomplete Jordan A-multialgebra: L = {αI4 + iQ(q) : α ∈ R, q ∈ H, ℜe(q) = 0}, M = M = {aI4 : a ∈ C}, (8.2) Indeed, the Jordan A-multialgebra (8.2) consists of all elements in Sym(T ) that belong to the associative A-multialgebra L′ = M′ = {aQ(q) : a ∈ C, q ∈ H}. (8.3) It is easy to check that the associative A-multialgebra (8.3) is symmetric in the sense of Definition 2.3. In [16] it was shown that the 3-chain relation property is necessary for the SO(2)-invariant Jordan multi- algebra to correspond to an exact relation. Hence, the incomplete Jordan multialgebra (8.1) provides the first example of a rotationally invariant lamination exact relation in 2D that is not stable under homogenization. For the sake of reference we compute the image of both (8.1) and its completion (8.2) in physical variables. In order to formulate the results it will be convenient to identify T = R2 ⊗ R4 with H2, via the isomorphisms T = R2 ⊗ R4 ∼= R4 ⊕ R4 ∼= H ⊕ H ∼= H2, where we have identified R4 with H. The quaternionic materials are identified with H+(H2) -- the set of positive definite quaternionic-Hermitean 2 × 2 matrices If then λ > 0, µ > 0, h ∈ H, det L = λµ − h2 > 0. L =" λ h h µ # , E =" q1 q2 # ∈ H2 = T , J =" p1 p2 # ∈ H2 = T , hq1 + µq2 # . " p1 p2 # =" λq1 + hq2 J = LE, 4We have identified all solutions of (3.1) for n = 2 by hand and verified that all the Jordan A-multialgebras are complete in this case. The case n = 3 remains unexplored. 20 We may also represent the multifield response of quaternionic materials by the conventional 8 × 8 real sym- metric matrix µI4 # , L =" λI4 Q(h) Q(h) where Q(q) is defined in (6.5). Finally, the lamination exact relation that is not stable under homogenization corresponding to (8.1) is given by Obviously, det L can be any positive constant in the definition of M. The constant is set to 1 for simplicity. M = {L ∈ H+(H2) : det L = 1}. (8.4) 9 Acknowledgement This material is based upon work supported by the National Science Foundation under Grant No. 1008092. A Summary of [14] Let Gp be a finite group with generators a1, . . . , ap−1, p ≥ 2 and ε satisfying the relations ε2 = 1Gp , a2 k = ε, εak = akε, akal = εalak, k, l = 1, . . . , p − 1, k 6= l. Let S = i1, . . . , ik be a subset of {1, . . . , p − 1}, where 1 ≤ i1 < i2 < . . . < ik ≤ p − 1. Let aS = ai1 ai2 . . . aik , where a∅ = 1Gp. Then Gp = {aS, εaS : S ⊂ {1, . . . , p−1}}. Hence, Gp = 2p. The first observation is that the commutator subgroup of Gp is K = {1Gp, ε}. It follows that Gp has exactly Gp/K = 2p−1 non-isomorphic complex 1D representations. In each of them ε gets sent to 1, since ε belongs to a commutator subgroup. The second observation is that it is easy to list all conjugacy classes of Gp explicitly. They are ({1Gp}, {ε}, {aS, εaS}, S $ {1, . . . , p − 1}, {a{1,...,p−1}}, {εa{1,...,p−1}}, p is even, {1Gp}, {ε}, {aS, εaS}, S ⊂ {1, . . . , p − 1}, S 6= ∅, {a{1,...,p−1}, εa{1,...,p−1}}, p is odd. The total number of non-isomorphic complex irreps of Gp equals to the number of conjugacy classes of Gp. We see that when p is odd there is exactly one non-1D irrep, while when p is even there are exactly two. The sum of squares of the dimensions of all the irreps of Gp equals to the order of Gp. Hence, when p is odd the dimension d of the non-1D irrep is d = 2 2 . When p is even we denote the dimensions of the two non-1D irreps by d1 and d2. Recall that d1 and d2 must divide the order of the group. Hence, d1 = 2α, d2 = 2β and 22α + 22β + 2p−1 · 12 = 2p. Thus, d1 = d2 = 2 2 . We remark that ε cannot get sent to the identity matrix I in a non-1D irrep, since in that case the commutator of Gp gets mapped to I and all images of elements of Gp would commute with one another. Since the image of ε must commute with the images of all elements in the group, it must be mapped (by Schur's lemma) into a multiple of the identity λI for some λ ∈ C. However, λ2 = 1, since ε2 = 1Gp. Therefore, ε must be sent to −I. p−2 p−1 The third observation is that g2 = 1Gp or ε for any g ∈ Gp. In fact, (εaS)2 = a2 S = εr(r+1)/2, where S gets mapped to (−1)r(r+1)/2I. This allows an explicit computation of the Frobenius- r = S. Therefore, a2 Schur indicator where χ is the character of the representation. The result is 1 S = χ(g2), Gp Xg∈Gp S = sign(cid:16)cos(cid:16) πp 4 (cid:17)(cid:17) . 21 This shows that the non-1D irreps of Gp are of real type, when p = 0, 1, 7 mod 8, complex type, when p = 2, 6 mod 8 and quaternionic type when p = 3, 4, 5 mod 8. In particular, when p = 2, 6 mod 8 the representations U and U are not isomorphic and hence exhaust the list of two non-isomorphic irreps for Gp. Thus, in order to obtain real irreps of Gp in which ε gets sent to −I we take real parts of the representations U for p = 0, 1, 7 mod 8, real parts of the representations U ⊕ U for p = 3, 4, 5 mod 8 and real parts of the representations U ⊕ U for p = 2, 6 mod 8. Theorem 6.2 summarizes the results. References [1] J. F. Adams. Vector fields on spheres. The Annals of Mathematics, 75(3):603 -- 632, 1962. [2] J. F. Adams, P. D. Lax, and R. S. Phillips. On matrices whose real linear combinations are nonsingular. Proceedings of the American Mathematical Society, 16(2):318 -- 322, 1965. [3] A. A. Albert. On a certain algebra of quantum mechanics. The Annals of Mathematics, 35(1):65 -- 73, Jan 1934. [4] A. Bensoussan, J. L. Lions, and G. Papanicolaou. Asymptotic analysis of periodic structures. North- Holland Publ., 1978. [5] Y. Benveniste. Exact results in the micromechanics of fibrous piezoelectric composites exhibiting pyro- electricity. Proc. Roy. Soc. Lond. A, 441:59 -- 81, 1993. [6] Y. Benveniste. Exact connections between polycrystal and crystal properties in two-dimensional poly- crystalline aggregates. Proc. Roy. Soc. Lond. A, 447:1 -- 22, 1994. [7] P. M. Cohn. On homomorphic images of special Jordan algebras. Canadian J. Math., 6:253 -- 264, 1954. [8] N. Degirmenci and S¸. Ko¸cak. Generalized self-duality of 2-forms. Advances in applied Clifford algebras, 13(1):107 -- 113, 2003. [9] N. Degirmenci and N. Ozdemir. The construction of maximum independent set of matrices via clifford algebras. TURKISH JOURNAL OF MATHEMATICS, 31(2):193 -- 205, 2007. [10] G. F. Dell'Antonio, R. Figari, and E. Orlandi. An approach through orthogonal projections to the study of inhomogeneous random media with linear response. Ann. Inst. Henri Poincar´e, 44:1 -- 28, 1986. [11] M. L. Dunn. Exact relations between the thermoelectroelastic moduli of heterogeneous materials. Proc. Roy. Soc. London Ser. A, 441(1913):549 -- 557, 1993. [12] G. J. Dvorak. On uniform fields in heterogeneous media. Proc. Roy. Soc. London Ser. A, 431(1881):89 -- 110, 1990. [13] G. J. Dvorak. On some exact results in thermoplasticity of composite materials. J. Thermal Stresses, 15(2):211 -- 228, 1992. Sixty-fifth Birthday of Bruno A. Boley Symposium, Part 2 (Atlanta, GA, 1991). [14] B. Eckmann. Gruppentheoretischer beweis des satzes von hurwitz-radon uber die komposition quadratis- cher formen. Commentarii Mathematici Helvetici, 15(1):358 -- 366, 1942. [15] Y. Grabovsky. Exact relations for effective tensors of polycrystals. I: Necessary conditions. Arch. Ration. Mech. Anal., 143(4):309 -- 330, 1998. [16] Y. Grabovsky. Algebra, geometry and computations of exact relations for effective moduli of composites. In G. Capriz and P. M. Mariano, editors, Advances in Multifield Theories of Continua with Substructure, Modeling and Simulation in Science, Engineering and Technology, pages 167 -- 197. Birkhauser, Boston, 2004. 22 [17] Y. Grabovsky. Exact relations for effective conductivity of fiber-reinforced conducting composites with the Hall effect via a general theory. SIAM J. Math Anal., 41(3):973 -- 1024, 2009. [18] Y. Grabovsky, G. W. Milton, and D. S. Sage. Exact relations for effective tensors of polycrystals: Necessary conditions and sufficient conditions. Comm. Pure. Appl. Math., 53(3):300 -- 353, 2000. [19] Y. Grabovsky and D. S. Sage. Exact relations for effective tensors of polycrystals. II: Applications to elasticity and piezoelectricity. Arch. Ration. Mech. Anal., 143(4):331 -- 356, 1998. [20] Z. Hashin. Thermal expansion of polycrystalline aggregates: I. Exact results. J. Mech. Phys. Solids, 32:149 -- 157, 1984. [21] M. Hegg. Exact Relations And Links For Fiber-Reinforced Elastic Composites. PhD thesis, Temple University, Philadelphia, Pennsylvania, May 2012. [22] R. Hill. Elastic properties of reinforced solids: Some theoretical principles. J. Mech. Phys. Solids, 11:357 -- 372, 1963. [23] A. Hurwitz. Uber die komposition der quadratischen formen. Mathematische Annalen, 88(1):1 -- 25, 1922. [24] N. Jacobson. Structure and representations of Jordan algebras. American Mathematical Society, Provi- dence, R.I., 1968. American Mathematical Society Colloquium Publications, Vol. XXXIX. [25] V. V. Jikov, S. M. Kozlov, and O. A. Oleınik. Homogenization of differential operators and integral functionals. Springer-Verlag, Berlin, 1994. Translated from the Russian by G. A. Yosifian. [26] P. Jordan, J. v. Neumann, and E. Wigner. On an algebraic generalization of the quantum mechanical formalism. The Annals of Mathematics, 35(1):29 -- 64, 1934. [27] J. B. Keller. A theorem on the conductivity of a composite medium. J. Math. Phys., 5:548 -- 549, 1964. [28] M. Koecher, A. Krieg, and S. Walcher. The Minnesota notes on Jordan algebras and their applications. Springer, 1999. [29] W. Kohler and G. C. Papanicolaou. Bounds for effective conductivity of random media. In R. Burridge, S. Childress, and G. Papanicolaou, editors, Macroscopic properties of disordered media, pages 111 -- 130, Berlin, 1982. Springer. [30] N. Kwon and D. S. Sage. Subrepresentation semirings and an analog of 6j-symbols. J. Math. Phys., 49(6):063503, 21, 2008. [31] V. M. Levin. Thermal expansion coefficients of heterogeneous materials. MTT, 2(1):88 -- 94, 1967. [32] K. McCrimmon. A taste of Jordan algebras. Springer, 2004. [33] M. Milgrom. Some more exact results concerning multifield moduli of two-phase composites. J. Mech. Phys. Solids, 45(3):399 -- 404, 1997. [34] M. Milgrom and S. Shtrikman. Linear response of polycrystals to coupled fields: Exact relations among the coefficients. Physical Review B (Solid State), 40(9):5991 -- 5994, 1989. [35] M. Milgrom and S. Shtrikman. Linear response of two-phase composites with cross moduli: Exact universal relations. Physical Review A (Atomic, Molecular, and Optical Physics), 40(3):1568 -- 1575, 1989. [36] G. W. Milton. Multicomponent composites, electrical networks and new types of continued fraction. I. Comm. Math. Phys., 111(2):281 -- 327, 1987. [37] G. W. Milton. Multicomponent composites, electrical networks and new types of continued fraction. II. Comm. Math. Phys., 111(3):329 -- 372, 1987. 23 [38] G. W. Milton. On characterizing the set of possible effective tensors of composites: the variational method and the translation method. Comm. Pure Appl. Math., 43:63 -- 125, 1990. [39] G. W. Milton. Composites: A myriad of microstructure independent relations. In T. Tatsumi, E. Watan- abe, and T. Kambe, editors, Theoretical and applied mechanics (Proc. of the XIX International Congress of Theoretical and Applied mechanics, Kyoto, 1996), pages 443 -- 459. Elsevier, Amsterdam, 1997. [40] G. W. Milton. The theory of composites, volume 6 of Cambridge Monographs on Applied and Computa- tional Mathematics. Cambridge University Press, Cambridge, 2002. [41] G. W. Milton and R. V. Kohn. Variational bounds on the effective moduli of anisotropic composites. J. Mech. Phys. Solids, 36(6):597 -- 629, 1988. [42] J. Radon. Lineare scharen orthogonaler matrizen. Abhandlungen aus dem Mathematischen Seminar der Universitat Hamburg, 1:1 -- 14, 1922. [43] B. W. Rosen and Z. Hashin. Effective thermal expansion coefficients and specific heats of composite materials. Int. J. Engng. Sci, 8:157 -- 173, 1970. [44] D. S. Sage. Group actions on central simple algebras. J. Algebra, 250:18 -- 43, 2002. [45] D. S. Sage. Quantum Racah coefficients and subrepresentation semirings. J. Lie Theory, 15(1):321 -- 333, 2005. [46] D. S. Sage. Racah coefficients, subrepresentation semirings, and composite materials. Adv. in Appl. Math., 34(2):335 -- 357, 2005. [47] K. Schulgasser. Thermal expansion of polycrystalline aggregates with texture. J. Mech. Phys. Solids, 35(1):34 -- 42, 1987. [48] V. Sver´ak. Rank-one convexity does not imply quasiconvexity. Proc. Roy. Soc. Edinburgh Sect. A, 120(1-2):185 -- 189, 1992. 24
1003.0585	2	1003	2011-10-29T22:57:42	Scalars, Monads, and Categories	[ "math.RA", "math.CT" ]	This chapter describes interrelations between: (1) algebraic structure on sets of scalars, (2) properties of monads associated with such sets of scalars, and (3) structure in categories (esp. Lawvere theories) associated with these monads. These interrelations will be expressed in terms of "triangles of adjunctions", involving for instance various kinds of monoids (non-commutative, commutative, involutive) and semirings as scalars. It will be shown to which kind of monads and categories these algebraic structures correspond via adjunctions.	math.RA	math	Scalars, Monads, and Categories Dion Coumans Bart Jacobs Inst. for Mathematics, Astrophysics Inst. for Computing and and Particle Physics (IMAPP) Information Sciences (ICIS) www.math.ru.nl/~coumans www.cs.ru.nl/~bart Radboud University Nijmegen, The Netherlands December 31, 2013 Abstract The paper describes interrelations between: (1) algebraic structure on sets of scalars, (2) properties of monads associated with such sets of scalars, and (3) structure in categories (esp. Lawvere theories) associated with these monads. These interrelations will be expressed in terms of “triangles of adjunctions”, involving for instance various kinds of monoids (non-commutative, commutative, involutive) and semirings as scalars. It will be shown to which kind of monads and categories these algebraic structures correspond via adjunctions. 1 Introduction Scalars are the elements s used in scalar multiplication s · v , yielding for in- stance a new vector for a given vector v . Scalars are elements in some algebraic structure, such as a field (for vector spaces), a ring (for modules), a group (for group actions), or a monoid (for monoid actions). A categorical description of scalars can be given in a monoidal category C, with tensor ⊗ and tensor unit I , as the homset C(I , I ) of endomaps on I . In [15] it is shown that such homsets C(I , I ) always form a commutative monoid; in [2, §3.2] this is called the ‘miracle’ of scalars. More recent work in the area of quantum computation has led to renewed interest in such scalars, see for instance [1, 2], where it is shown that the presence of biproducts makes this homset C(I , I ) of scalars a semiring, and that daggers † make it involu- tive. These are first examples where categorical structure (a category which is monoidal or has biproducts or daggers) gives rise to algebraic structure (a set with a commutative monoid, semiring or involution structure). Such cor- respondences form the focus of this paper, not only those between categorical and algebraic structure, but also involving a third element, namely structure on endofunctors (especially monads). Such correspondences will be described in terms of triangles of adjunctions. 1 Sets A 7→ A×(−) ⊣ (−)(1) A 7→ A×(−) ⊣ (−)(1) SetsSets left Kan ⊥ restrict Setsℵ0 Figure 1: Basic triangle of adjunctions. To start, we describe the basic triangle of adjunctions that we shall build on. At this stage it is meant as a sketch of the setting, and not as an exhaustive explanation. Let ℵ0 be the category with natural numbers n ∈ N as ob jects. Such a number n is identified with the n-element set n = {0, 1, . . . , n − 1}. Morphisms n → m in ℵ0 are ordinary functions n → m between these finite sets. Hence there is a full and faithful functor ℵ0 ֒→ Sets. The underline notation is useful to avoid ambiguity, but we often omit it when no confusion arises and write the number n for the set n. Now consider the triangle in Figure 1, with functor categories at the two bottom corners. We briefly explain the arrows (functors) in this diagram. The downward arrows Sets → SetsSets and Sets → Setsℵ0 describe the functors that map a set A ∈ Sets to the functor X 7→ A × X . In the other, upward direction right adjoints are given by the functors (−)(1) describing “evaluate at unit 1”, that is F 7→ F (1). At the bottom the inclusion ℵ0 ֒→ Sets induces a functor SetsSets → Setsℵ0 by restriction: F is mapped to the functor n 7→ F (n). In the reverse direction a left adjoint is obtained by left Kan extension [17, Ch. X]. Explicitly, this left adjoint maps a functor F : ℵ0 → Sets to the functor L(F ) : Sets → Sets given by: L(F )(X ) = (cid:16) `i∈N F (i) × X i(cid:17)/ ∼, where ∼ is the least equivalence relation such that, for each f : n → m in ℵ0 , where a ∈ F (n) and v ∈ X m . κm (F (f )(a), v) ∼ κn (a, v ◦ f ), The adjunction on the left in Figure 1 is then in fact the composition of the other two. The adjunctions in Figure 1 are not new. For instance, the one at the bottom plays an important role in the description of analytic functors and species [14], see also [10, 3, 6]. The category of presheaves Setsℵ0 is used to provide a semantics for binding, see [7]. What is new in this paper is the systematic organisation of correspondences in triangles like the one in Figure 1 for various kinds of algebraic structures (instead of sets). • There is a triangle of adjunctions for monoids, monads, and Lawvere the- ories, see Figure 2. 2 ' ' 1 1 @ @ h h q q • This triangle restricts to commutative monoids, commutative monads, and symmetric monoidal Lawvere theories, see Figure 3. • There is also a triangle of adjunctions for commutative semirings, commu- tative additive monads, and symmetric monoidal Lawvere theories with biproducts, see Figure 4. • This last triangle restricts to involutive commutative semirings, involutive commutative additive monads, and dagger symmetric monoidal Lawvere theories with dagger biproducts, see Figure 5 below. These four figures with triangles of adjunctions provide a quick way to get an overview of the paper (the rest is just hard work). The triangles capture fundamental correspondences between basic mathematical structures. As far as we know they have not been made explicit at this level of generality. The paper is organised as follows. It starts with a section containing some background material on monads and Lawvere theories. The triangle of adjunc- tions for monoids, much of which is folklore, is developed in Section 3. Sub- sequently, Section 4 forms an intermezzo; it introduces the notion of additive monad, and proves that a monad T is additive if and only if in its Kleisli cate- gory Kℓ(T ) coproducts form biproducts, if and only if in its category Alg(T ) of algebras products form biproducts. These additive monads play a crucial role in Sections 5 and 6 which develop a triangle of adjunctions for commutative semirings. Finally, Section 7 introduces the refined triangle with involutions and daggers. The triangles of adjunctions in this paper are based on many detailed veri- fications of basic facts. We have chosen to describe all constructions explicitly but to omit most of these verifications, certainly when these are just routine. Of course, one can continue and try to elaborate deeper (categorical) structure underlying the triangles. In this paper we have chosen not to follow that route, but rather to focus on the triangles themselves. 2 Preliminaries We shall assume a basic level of familiarity with category theory, especially with adjunctions and monads. This section recalls some basic facts and fixes notation. For background information we refer to [4, 5, 17]. In an arbitrary category C we write finite products as ×, 1, where 1 ∈ C is the final ob ject. The pro jections are written as πi and tupling as hf1 , f2 i. Finite coproducts are written as + with initial ob ject 0, and with copro jections κi and cotupling [f1 , f2 ]. We write !, both for the unique map X → 1 and the unique map 0 → X . A category is called distributive if it has both finite products and finite coproducts such that functors X × (−) preserve these coproducts: the canonical maps 0 → X × 0, and (X × Y ) + (X × Z ) → X × (Y + Z ) are isomorphisms. Monoidal products are written as ⊗, I where I is the tensor ∼=→ (X ⊗ Y ) ⊗ Z for unit, with the familiar isomorphisms: α : X ⊗ (Y ⊗ Z ) 3 ∼=→ X for unit, and in the symmetric ∼=→ X and λ : I ⊗ X associativity, ρ : X ⊗ I ∼=→ Y ⊗ X for swap. case also γ : X ⊗ Y We write Mnd(C) for the category of monads on a category C. For con- venience we write Mnd for Mnd(Sets). Although we shall use strength for monads mostly with respect to finite products (×, 1) we shall give the more general definition involving monoidal products (⊗, I ). A monad T is called strong if it comes with a ‘strength’ natural transformation st with components st : T (X ) ⊗ Y → T (X ⊗ Y ), commuting with unit η and multiplication µ, in the sense that st ◦ η ⊗ id = η and st ◦ µ ⊗ id = µ ◦ T (st) ◦ st. Additionally, for the familiar monoidal isomorphisms ρ and α, st T (Y ) ⊗ I T (Y ⊗ I ) $IIIIIIIII ρ T (Y ) T (ρ) T (X ) ⊗ (Y ⊗ Z ) st T (X ⊗ (Y ⊗ Z )) α T (α) (T (X ) ⊗ Y ) ⊗ Z st⊗id / T (X ⊗ Y ) ⊗ Z st / T ((X ⊗ Y ) ⊗ Z ) Also, when the tensor ⊗ is a cartesian product × we sometimes write these ρ and α for the obvious maps. The category StMnd(C) has monads with strength (T , st) as ob jects. Mor- phisms are monad maps commuting with strength. The monoidal structure on C is usually clear from the context. Lemma 1 Monads on Sets are always strong w.r.t. finite products, in a canon- ical way, yielding a functor Mnd(Sets) = Mnd → StMnd = StMnd(Sets). Proof For every functor T : Sets → Sets, there exists a strength map st : T (X )× Y → T (X ×Y ), namely st(u, y ) = T (λx. hx, y i)(u). It makes the above diagrams commute, and also commutes with unit and multiplication in case T is a monad. Additionally, strengths commute with natural transformations σ : T → S , in the sense that σ ◦ st = st ◦ (σ × id). (cid:3) Given a general strength map st : T (X ) ⊗ Y → T (X ⊗ Y ) in a symmetric monoidal category one can define a swapped st′ : X ⊗ T (Y ) → T (X ⊗ Y ) as ∼=→ Y ⊗ X is the swap map. There are now st′ = T (γ ) ◦ st ◦ γ , where γ : X ⊗ Y in principle two maps T (X ) ⊗ T (Y ) ⇒ T (X ⊗ Y ), namely µ ◦ T (st′ ) ◦ st and µ ◦ T (st) ◦ st′ . A strong monad T is called commutative if these two composites T (X ) ⊗ T (Y ) ⇒ T (X ⊗ Y ) are the same. In that case we shall write dst for this (single) map, which is a monoidal transformation, see also [16]. The powerset monad P is an example of a commutative monad, with dst : P (X ) × P (Y ) → P (X × Y ) given by dst(U, V ) = U × V . Later we shall see other examples. We write Kℓ(T ) for the Kleisli category of a monad T , with X ∈ C as ob jects, and maps X → T (Y ) in C as arrows. For clarity we sometimes write a fat dot • for composition in Kleisli categories, so that g • f = µ ◦ T (g ) ◦ f . The inclusion functor C → Kℓ(T ) is written as J , where J (X ) = X and J (f ) = η ◦ f . A map of monads σ : T → S yields a functor Kℓ(σ) : Kℓ(T ) → Kℓ(S ) which is the identity on ob jects, and maps an arrow f to σ ◦ f . This functor Kℓ(σ) 4 / / $ / / / / commutes with the J ’s. One obtains a functor Kℓ : Mnd(C) → Cat, where Cat is the category of (small) categories. We will use the following standard result. Lemma 2 For T ∈ Mnd(C), consider the generic statement “if C has ♦ then so does Kℓ(T ) and J : C → Kℓ(T ) preserves ♦’s”, where ♦ is some property. This holds for: (i). ♦ = (finite coproducts +, 0), or in fact any colimits; (ii). ♦ = (monoidal products ⊗, I ), in case the monad T is commutative; f ⊗g −→ T (U ) ⊗ T (V ) Proof Point (i) is obvious; for (ii) one defines the tensor on morphisms in Kℓ(T ) as: f g dst −→ T (U ⊗ V )(cid:1). (cid:0)X → T (U )(cid:1) ⊗ (cid:0)Y → T (V )(cid:1) = (cid:0)X ⊗ Y Then: J (f ) ⊗ J (g ) = dst ◦ ((η ◦ f ) ⊗ (η ◦ g )) = η ◦ (f ⊗ g ) = J (f ⊗ g ). (cid:3) As in this lemma we sometimes formulate results on monads in full gener- ality, i.e. for arbitrary categories, even though our main results—see Figures 2, 3, 4 and 5—only deal with monads on Sets. These results involve algebraic structures like monoids and semirings, which we interpret in the standard set- theoretic universe, and not in arbitrary categories. Such greater generality is possible, in principle, but it does not seem to add enough to justify the addi- tional complexity. Often we shall be interested in a “finitary” version of the Kleisli construction, corresponding to the Lawvere theory [18, 12] associated with a monad. For a monad T ∈ Mnd on Sets we shall write KℓN (T ) for the category with natural numbers n ∈ N as ob jects, regarded as finite sets n = {0, 1, . . . , n − 1}. A map f : n → m in KℓN (T ) is then a function n → T (m). This yields a full inclusion KℓN (T ) ֒→ Kℓ(T ). It is easy to see that a map f : n → m in KℓN (T ) can be identified with an n-cotuple of elements fi ∈ T (m), which may be seen as m-ary terms/operations. By the previous lemma the category KℓN (T ) has coproducts given on ob jects simply by the additive monoid structure (+, 0) on natural numbers. There are obvious copro jections n → n + m, using n + m ∼= n + m. The identities n + 0 = n = 0 + n and (n + m) + k = n + (m + k) are in fact the familiar monoidal isomorphisms. The swap map is an isomorphism n + m ∼= m + n rather than an identity n + m = m + n. In general, a Lawvere theory is a small category L with natural numbers n ∈ N as ob jects, and (+, 0) on N forming finite coproducts in L. It forms a categorical version of a term algebra, in which maps n → m are understood as n-tuples of terms ti each with m free variables. Formally a Lawvere theory involves a functor ℵ0 → L that is the identity on ob jects and preserves finite coproducts “on the nose” (up-to-identity) as opposed to up-to-isomorphism. A morphism of Lawvere theories F : L → L′ is a functor that is the identity on ob jects and strictly preserves finite coproducts. This yields a category Law . 5 Mon A ⊣ E ∼=HKℓN KℓNA H ⊣ Mnd T ⊥ KℓN Law A(M ) = M × (−) action monad E (T ) = T (1) evaluation at singleton set 1 H(L) = L(1, 1) endo-homset of 1 ∈ L   Figure 2: Basic relations between monoids, monads and Lawvere theories. Kleisli category restricted to ob jects n ∈ N monad associated with Lawvere theory L. where KℓN (T ) T (L) = TL Corollary 3 The finitary Kleisli construction KℓN for monads on Sets, yields a functor KℓN : Mnd → Law. (cid:3) 3 Monoids The aim of this section is to replace the category Sets of sets at the top of the triangle in Figure 1 by the category Mon of monoids (M , ·, 1), and to see how the corners at the bottom change in order to keep a triangle of adjunctions. Formally, this can be done by considering monoid ob jects in the three categories at the corners of the triangle in Figure 1 (see also [7, 6]) but we prefer a more concrete description. The results in this section, which are summarised in Fig- ure 2, are not claimed to be new, but are presented in preparation of further steps later on in this paper. We start by studying the interrelations between monoids and monads. In principle this part can be skipped, because the adjunction on the left in Figure 2 between monoids and monads follows from the other two by composition. But we do make this adjunction explicit in order to completely describe the situation. The following result is standard. We only sketch the proof. Lemma 4 Each monoid M gives rise to a monad A(M ) = M × (−) : Sets → Sets. The mapping M 7→ A(M ) yields a functor Mon → Mnd. Proof For a monoid (M , ·, 1) the unit map η : X → M × X = A(M ) is x 7→ (1, x). The multiplication µ : M × (M × X ) → M × X is (s, (t, x)) 7→ (s · t, x). The standard strength map st : (M × X ) × Y → M × (X × Y ) is given by st((s, x), y ) = (s, (x, y )). Each monoid map f : M → N gives rise to a map of 6 ) ) 1 1 ? ? i i q q monads with components f × id : M × X → N × X . These components commute with strength. (cid:3) The monad A(M ) = M × (−) is called the ‘action monad’, as its cate- gory of Eilenberg-Moore algebras consists of M -actions M × X → X and their morphisms. The monoid elements act as scalars in such actions. Conversely, each monad (on Sets) gives rise to a monoid. In the following lemma we prove this in more generality. For a category C with finite products, we denote by Mon(C) the category of monoids in C, i.e. the category of ob jects M in C carrying a monoid structure 1 → M ← M ×M with structure preserving maps between them. Lemma 5 Each strong monad T on a category C with finite products, gives rise to a monoid E (T ) = T (1) in C. The mapping T 7→ T (1) yields a functor StMnd(C) → Mon(C) Proof For a strong monad (T , η , µ, st), we define a multiplication on T (1) by µ ◦ T (π2 ) ◦ st : T (1) × T (1) → T (1), with unit η1 : 1 → T (1). Each monad map σ : T → S gives rise to a monoid map T (1) → S (1) by taking the component of σ at 1. (cid:3) The swapped strength map st′ gives rise to a swapped multiplication on T (1), namely µ ◦ T (π1 ) ◦ st′ : T (1) × T (1) → T (1), again with unit η1 . It corresponds to (a, b) 7→ b · a instead of (a, b) 7→ a · b like in the lemma. In case T is a commutative monad, the two multiplications coincide as we prove in Lemma 10. The functors defined in the previous two Lemmas 4 and 5 form an adjunction. This result goes back to [19]. Lemma 6 The pair of functors A : Mon ⇄ Mnd : E forms an adjunction A ⊣ E , as on the left in Figure 2. Proof For a monoid M and a (strong) monad T on Sets there are (natural) bijective correspondences: A(M ) σ / T in Mnd M f / T (1) in Mon ρ−1 ∼= / M × 1 = A(M )(1) Given σ one defines a monoid map σ : M → T (1) as: / T (1)(cid:17), σ = (cid:16)M where ρ−1 = hid, !i in this cartesian case. Conversely, given f one gets a monad map f : A(M ) → T with components: f X = (cid:16)M × X / T (X )(cid:17), T (λ) ∼= σ1 f ×id / T (1) × X st / T (1 × X ) 7 / / / / / / / ∼=→ X . Straightforward computations show that these where λ = π2 : 1 × X assignments indeed give a natural bijective correspondence. (cid:3) Notice that, for a monoid M , the counit of the above adjunction is the ∼=→ M . Hence the adjunction is a pro jection (E ◦ A)(M ) = A(M )(1) = M × 1 reflection. We now move to the bottom of Figure 2. The finitary Kleisli construction yields a functor from the category of monads to the category of Lawvere theories (Corollary 3). This functor has a left adjoint, as is proven in the following two standard lemmas. Lemma 7 Each Lawvere theory L, gives rise to a monad TL on Sets, which is defined by TL (X ) = (cid:16) `i∈N L(1, i) × X i(cid:17)/ ∼, where ∼ is the least equivalence relation such that, for each f : i → m in ℵ0 ֒→ L, where g ∈ L(1, i) and v ∈ X m . κm (f ◦ g , v) ∼ κi (g , v ◦ f ), (1) Final ly, the mapping L → TL yields a functor T : Law → Mnd. Proof For a Lawvere theory L, the unit map η : X → TL (X ) = (cid:16) `i∈N L(1, i)× X i(cid:17)/ ∼ is given by x 7→ [κ1 (id1 , x)]. The multiplication µ : T 2 L (X ) → TL(X ) is given by: µ([κi (g , v)]) = [κj ((g0 + · · · + gi−1 ) ◦ g , [v0 , . . . , vi−1 ])] where g : 1 → i, and v : i → TL (X ) is written as v(a) = κja (ga , va ), for a < i, and j = j0 + · · · + ji−1 . It is straightforward to show that this map µ is well-defined and that η and µ indeed define a monad structure on TL . For each morphism of Lawvere theories F : L → K, one may define a monad morphism T (F ) : TL → TK with components T (F )X : [κi (g , v)] 7→ [κi (F (g ), v)]. This yields a functor T : Law → Mnd. Checking the details is left to the reader. (cid:3) Lemma 8 The pair of functors T : Law ⇄ Mnd : KℓN forms an adjunction T ⊣ KℓN , as at the bottom in Figure 2. Proof For a Lawvere theory L and a monad T there are (natural) bijective correspondences: T (L) σ / T in Mnd L F / KℓN (T ) in Law 8 / / Given σ , one defines a Law-map σ : L → KℓN (T ) which is the identity on ob jects and sends a morphism f : n → m in L to the morphism λi<n. [κm (f ◦κi ,idm )] n / T (L)(m) σm / / T (m) in KℓN (T ). Conversely, given F , one defines a monad morphism F with components F X : T (L)(X ) → T (X ) given, for i ∈ N, g : 1 → i ∈ L and v ∈ X i , by: [κi (g , v)] 7→ (T (v) ◦ F (g ))(∗), where ∗ is the unique element of 1. (cid:3) Finally, we consider the right-hand side of Figure 2. For each category C and ob ject X in C, the homset C(X, X ) is a monoid, where multiplication is given by composition with the identity as unit. The mapping L 7→ H(L) = L(1, 1), defines a functor Law → Mon. This functor is right adjoint to the composite functor KℓN ◦ A. Lemma 9 The pair of functors KℓN ◦ A : Mon ⇄ Law : H forms an adjunc- tion KℓN ◦ A ⊣ H, as on the right in Figure 2. Proof For a monoid M and a Lawvere theory L there are (natural) bijective correspondences: KℓNA(M ) F / / L in Law M f / H(L) in Mon Given F one defines a monoid map F : M → H(L) = L(1, 1) by s 7→ F (1 hλx. s,!i −−−−−→ M × 1). hλx. s,!i Note that 1 −−−−−→ M × 1 = A(M )(1) is an endomap on 1 in KℓNA(M ). Since F is the identity on ob jects it sends this endomap to an element of L(1, 1). Conversely, given a monoid map f : M → L(1, 1) one defines a Law -map f : KℓNA(M ) → L. It is the identity on ob jects and sends a morphism h : n → m in KℓNA(M ), i.e. h : n → M × m in Sets, to the morphism f (h) = (cid:16)n (cid:2)κh2 (i) ◦f (h1 (i))(cid:3)i<n / / m(cid:17) . Here we write h(i) ∈ M × m as pair (h1 (i), h2 (i)). We leave further details to the reader. (cid:3) 9 / / Given a monad T on Sets, HKℓN (T ) = KℓN (T )(1, 1) = Sets(1, T (1)) is a monoid, where the multiplication is given by µ (1 a−→ T (1)) · (1 b−→ T (1)) = (cid:0)1 a−→ T (1) −→ T (1)(cid:1). The functor E : Mnd(C) → Mon(C), defined in Lemma 5 also gives a multi- plication on Sets(1, T (1)) ∼= T (1), namely µ ◦ T (π2 ) ◦ st : T (1) × T (1) → T (1). These two multiplications coincide as is demonstrated in the following diagram, T (b) −−−→ T 2(1) 1 a T (1) ha,bi ρ−1 T (1) × 1 &MMMMMMMMMM T (ρ−1 ) T (1 × 1) st T (b) T (1) × T (1) id×b st · / T (1 × T (1)) T (id×b) T (λ) T 2(1) µ / T (1) In fact, E ∼= HKℓN , which completes the picture from Figure 2. 3.1 Commutative monoids In this subsection we briefly summarize what will change in the triangle in Fig- ure 2 when we restrict ourselves to commutative monoids (at the top). This will lead to commutative monads, and to tensor products. The latter are induced by Lemma 2. The new situation is described in Figure 3. For the adjunc- tion between commutative monoids and commutative monads we start with the following basic result. Lemma 10 Let T be a commutative monad on a category C with finite prod- ucts. The monoid E (T ) = T (1) in C from Lemma 5 is then commutative. Proof Recall that the multiplication on T (1) is given by µ ◦ T (λ) ◦ st : T (1) × T (1) → T (1) and commutativity of the monad T means µ ◦ T (st′ ) ◦ st = µ ◦ T (st) ◦ st′ where st′ = T (γ ) ◦ st ◦ γ , for the swap map γ , see Section 2. Then: µ ◦ T (λ) ◦ st ◦ γ = µ ◦ T (T (λ) ◦ st′ ) ◦ st ◦ γ = T (λ) ◦ µ ◦ T (st′ ) ◦ st ◦ γ = T (ρ) ◦ µ ◦ T (st) ◦ st′ ◦ γ by commutativity of T , and because ρ = λ : 1 × 1 → 1 = µ ◦ T (T (ρ) ◦ st ◦ γ ) ◦ st = µ ◦ T (ρ ◦ γ ) ◦ st = µ ◦ T (λ) ◦ st. 10 (cid:3) / / + + / / & / / / / / / The proof of the next result is easy and left to the reader. Lemma 11 A monoid M is commutative (Abelian) if and only if the associ- ated monad A(M ) = M × (−) : Sets → Sets is commutative (as described in Section 2). (cid:3) Next, we wish to define an appropriate category SMLaw of Lawvere theories with symmetric monoidal structure (⊗, I ). In order to do so we need to take a closer look at the category ℵ0 described in the introduction. Recall that ℵ0 has n ∈ N as ob jects whilst morphisms n → m are functions n → m in Sets, where, as described earlier n = {0, 1, . . . , n − 1}. This category ℵ0 has a monoidal structure, given on ob jects by multiplication n × m of natural numbers, with 1 ∈ N as tensor unit. Functoriality involves a (chosen) coordinatisation, in the following way. For f : n → p and g : m → q in ℵ0 one obtains f ⊗g : n×m → p×q as a function: f ⊗ g = co−1 p,q ◦ (f × g ) ◦ con,m : n × m −→ p × q , where co is a coordinatisation function n × m = {0, . . . , (n × m) − 1} con,m ∼= / {0, . . . , n − 1} × {0, . . . , m − 1} = n × m, given by co(c) = (a, b) ⇔ c = a · m + b. (2) co / n × m γ Sets / m × n co−1 We may write the inverse co−1 : n × m → n × m as a small tensor, as in a ⊗ b = co−1 (a, b). Then: (f ⊗ b)(a ⊗ b) = f (a) ⊗ g (b). The monoidal isomorphisms in ℵ0 are then obtained from Sets, as in γ ℵ0 = (cid:16)n × m / m × n(cid:17). Thus γ ℵ0 (a ⊗ b) = b ⊗ a. Similarly, the associativity map αℵ0 : n ⊗ (m ⊗ k) → (n ⊗ m) ⊗ k is determined as αℵ0 (a ⊗ (b ⊗ c)) = (a ⊗ b) ⊗ c. The maps ρ : n × 1 → n in ℵ0 are identities. This tensor ⊗ on ℵ0 distributes over sum: the canonical distributivity map (n ⊗ m) + (n ⊗ k) → n ⊗ (m + k) is an isomorphism. Its inverse maps a ⊗ b ∈ n ⊗ (m + k) to a ⊗ b ∈ n × m if b < m, and to a ⊗ (b − m) ∈ n × k otherwise. We thus define the ob jects of the category SMLaw to be symmetric monoidal Lawvere theories L ∈ Law for which the map ℵ0 → L strictly preserves the monoidal structure that has just been described via multiplication (×, 1) of natural numbers; additionally the coproduct structure must be preserved, as in Law. Morphisms in SMLaw are morphisms in Law that strictly preserve this tensor structure. We note that for L ∈ SMLaw we have a distributivity ∼=→ n ⊗ (m + k), since this isomorphism lies in the range of the n ⊗ m + n ⊗ k functor ℵ0 → L. 11 / / / / CMon A ⊣ E ∼=HKℓN KℓNA H ⊣ CMnd T ⊥ KℓN SMLaw Figure 3: Commutative version of Figure 2, with commutative monoids, com- mutative monads and symmetric monoidal Lawvere theories. By Lemma 2 we know that the Kleisli category Kℓ(T ) is symmetric monoidal if T is commutative. In order to see that also the finitary Kleisli category KℓN (T ) ∈ Law is symmetric monoidal, we have to use the coordinatisation map described in (2). For f : n → p and g : m → q in KℓN (T ) we then obtain f ⊗ g : n × m → p × q as co f ×g dst / n × m / T (p) × T (q) T (co−1 ) / T (p × q)(cid:17). f ⊗ g = (cid:16)n × m / T (p × q) We recall from [15] (see also [1, 2]) that for a monoidal category C the homset C(I , I ) of endomaps on the tensor unit forms a commutative monoid. This applies in particular to Lawvere theories L ∈ SMLaw, and yields a functor H : SMLaw → Mon given by H(L) = L(1, 1), where 1 ∈ L is the tensor unit. Thus we almost have a triangle of adjunctions as in Figure 3. We only need to check the following result. Lemma 12 The functor T : Law → Mnd defined in (1) restricts to SMLaw → CMnd. Further, this restriction is left adjoint to KℓN : CMnd → SMLaw. Proof For L ∈ SMLaw we define a map dst / T (L)(X × Y ) / [κi×j (g ⊗ h, (v × w) ◦ coi,j )], T (L)(X ) × T (L)(Y ) (cid:0)[κi (g , v)], [κj (h, w)](cid:1) where g : 1 → i and h : 1 → j in L yield g ⊗ h : 1 = 1 ⊗ 1 → i ⊗ j = i × j , and co is the coordinatisation function (2). Then one can show that both µ ◦ T (L)(st′ ) ◦ st and µ ◦ T (L)(st) ◦ st′ are equal to dst. This makes T (L) a commutative monad. In order to check that the adjunction T ⊣ KℓN restricts, we only need to verify that the unit L → KℓN (T (L)) strictly preserves tensors. This is easy. (cid:3) 12 ~ ~ * * 1 1 > > j j q q / / / / / / 4 Additive monads Having an adjunction between commutative monoids and commutative monads (Figure 3) raises the question whether we may also define an adjunction between commutative semirings and some specific class of monads. It will appear that so-called additive commutative monads are needed here. In this section we will define and study such additive (commutative) monads and see how they relate to biproducts in their Kleisli categories and categories of algebras. We consider monads on a category C with both finite products and coprod- ucts. If, for a monad T on C, the ob ject T (0) is final—i.e. satisfies T (0) ∼= 1— then 0 is both initial and final in the Kleisli category Kℓ(T ). Such an ob ject that is both initial and final is called a zero object. Also the converse is true, if 0 ∈ Kℓ(T ) is a zero ob ject, then T (0) is final in C. Although we don’t use this in the remainder of this paper, we also mention a related result on the category of Eilenberg-Moore algebras. The proofs are simple and are left to the reader. Lemma 13 For a monad T on a category C with finite products (×, 1) and coproducts (+, 0), the fol lowing statements are equivalent. (i). T (0) is final in C; (ii). 0 ∈ Kℓ(T ) is a zero object; (iii). 1 ∈ Alg(T ) is a zero object. (cid:3) A zero ob ject yields, for any pair of ob jects X, Y , a unique “zero map” 0X,Y : X → 0 → Y between them. In a Kleisli category Kℓ(T ) for a monad T on C, this zero map 0X,Y : X → Y is the following map in C / T (Y )(cid:17). 0X,Y = (cid:16)X For convenience, we make some basic properties of this zero map explicit. / 1 ∼= T (0) (3) !X / T (!Y ) Lemma 14 Assume T (0) is final, for a monad T on C. The resulting zero maps 0X,Y : X → T (Y ) from (3) make the fol lowing diagrams in C commute X T (0) 0 T 2(Y ) T (X ) #GGGGGGG yttttttt 0 0 T (Y ) µ X f Y 0 T (Y ) "FFFFFFF 0 / T (Z ) 0 T (f ) X 0 T (Y ) "FFFFFFF 0 S (Y ) σY where f : Y → Z is a map in C and σ : T → S is a map of monads. (cid:3) Still assuming that T (0) is final, the zero map (3) enables us to define a canonical map bc def= (cid:16)T (X + Y ) hµ◦T (p1 ),µ◦T (p2 )i / T (X ) × T (Y )(cid:17), (4) 13 / / / # o o y / / " / / " / / p1 p2 [η,0Y ,X ] [0X,Y ,η] where def= (cid:16)X + Y def= (cid:16)X + Y / T (X )(cid:17), / T (Y )(cid:17). Here we assume that the underlying category C has both finite products and finite coproducts. The abbreviation “bc” stands for “bicartesian”, since this maps connects the coproducts and products. The auxiliary maps p1 , p2 are sometimes called pro jections, but should not be confused with the (proper) pro jections π1 , π2 associated with the product × in C. We continue by listing a series of properties of this map bc that will be useful in what follows. (5) Lemma 15 In the context just described, the map bc : T (X + Y ) → T (X ) × T (Y ) in (4) has the fol lowing properties. (i). This bc is a natural transformation, and it commutes with any monad map σ : T → S , as in: T (X + Y ) T (f +g) T (U + V ) bc bc T (X ) × T (Y ) T (X + Y ) T (f )×T (g) σX+Y / T (U ) × T (V ) S (X + Y ) bc bc T (X ) × T (Y ) σX ×σY / S (X ) × S (Y ) (ii). It also commutes with the monoidal isomorphisms (for products and co- products in C): bc / T (X + 0) T (X ) × T (0) &MMMMMMMMMMM ∼= T (ρ) ρ∼= T (X ) T (X + Y ) bc / T (X ) × T (Y ) T ([κ2 ,κ1 ]) ∼= ∼= hπ2 ,π1 i T (Y + X ) bc / T (Y ) × T (X ) T ((X + Y ) + Z ) T (α) ∼= T (X + (Y + Z )) bc bc T (X + Y ) × T (Z ) bc×id / (T (X ) × T (Y )) × T (Z ) / T (X ) × T (Y + Z ) id×bc / T (X ) × (T (Y ) × T (Z )) α∼= (iii). The map bc interacts with η and µ in the fol lowing manner: X + Y η 'OOOOOOOOOOO hp1 ,p2 i / T (X ) × T (Y ) T (X + Y ) bc 14 / / / / / / / / / & / / / / / / / ' / T 2 (X + Y ) µ T (X + Y ) T (T (X ) + T (Y )) bc T 2(X ) × T 2(Y ) T (bc) T ([T (κ1 ),T (κ2 )]) T (T (X ) × T (Y )) bc T 2(X + Y ) µ×µ hT (π1 ),T (π2 )i µ T 2 (X ) × T 2(Y ) µ×µ / T (X ) × T (Y ) T (X + Y ) bc / T (X ) × T (Y ) (iv). If C is a distributive category, bc commutes with strength st as fol lows: T (X + Y ) × Z bc×id (T (X ) × T (Y )) × Z dbl / (T (X ) × Z ) × (T (Y ) × Z ) st st×st T ((X + Y ) × Z ) ∼= / / T ((X × Z ) + (Y × Z )) bc / T (X × Z ) × T (Y × Z ) where dbl is the “double” map hπ1 × id, π2 × idi : (A × B ) × C → (A × C ) × (B × C ). Proof These properties are easily verified, using Lemma 14 and the fact that the pro jections pi are natural, both in C and in Kℓ(T ). (cid:3) The definition of the map bc also makes sense for arbitrary set-indexed (co)products (see [13]), but here we only consider finite ones. Such generalised bc-maps also satisfy (suitable generalisations of ) the properties in Lemma 15 above. We will study monads for which the canonical map bc is an isomorphism. Such monads will be called ‘additive monads’. Definition 16 A monad T on a category C with finite products (×, 1) and finite coproducts (+, 0) wil l be cal led additive if T (0) ∼= 1 and if the canonical map bc : T (X + Y ) → T (X ) × T (Y ) from (4) is an isomorphism. We write AMnd(C) for the category of additive monads on C with monad morphism between them, and similarly ACMnd(C) for the category of additive and commutative monads on C. A basic result is that additive monads T induce a commutative monoid structure on ob jects T (X ). This result is sometimes taken as definition of additivity of monads (cf. [9]). Lemma 17 Let T be an additive monad on a category C and X an object of C. There is an addition + on T (X ) given by = (cid:16)T (X ) × T (X ) def where ∇ = [id, id]. Then: / T (X )(cid:17), T (∇) + bc−1 / T (X + X ) 15 / / / / / / / / / / / / (i). this + is commutative and associative, (ii). and has unit 01,X : 1 → T (X ); (iii). this monoid structure is preserved by maps T (f ) as wel l as by multiplica- tion µ; (iv). the mapping (T , X ) 7→ (T (X ), +, 01,X ) yields a functor Ad : AMnd(C) × C → CMon(C). Proof The first three statements follow by the properties of bc from Lemma 15. For instance, 0 is a (right) unit for + as demonstrated in the following diagram. T (X ) ρ−1 T (X ) × T (0) ∼= )RRRRRRRRRRRRRRR ∼= T (ρ−1 ) T (X + 0) bc−1 id×T (!) T (X ) × T (X ) bc−1 T (id+ !) T (X + X ) TTTTTTTTTTTTTTTTT ∼= T (ρ) T (∇) T (X ) + Regarding (iv) we define, for a pair of morphisms σ : T → S in AMnd(C) and f : X → Y in C, Ad((σ, f )) = σ ◦ T (f ) : T (X ) → S (Y ), which is equal to S (f ) ◦ σ by naturality of σ . Preservation of the unit by Ad((σ, f )) follows from Lemma 14. The following diagram demonstrates that addition is preserved. T (X ) × T (X ) T (f )×T (f ) T (Y ) × T (Y ) σ×σ S (Y ) × S (Y ) bc−1 T (X + X ) T (∇) bc−1 T (f +f ) TTTTTTTTTTTTTTTT σ bc−1 T (Y + Y ) (PPPPPPPPPPPP σ S (f +f ) S (X + X ) S (Y + Y ) T (X ) σ S (∇) / S (X ) S (∇) S (f ) / S (Y ) where we use point (i) of Lemma 15 and the naturality of σ . It is easily checked that this mapping defines a functor. (cid:3) By Lemma 2, for a monad T on a category C with finite coproducts, the Kleisli construction yields a category Kℓ(T ) with finite coproducts. Below we 16 / / ) / / x x / / * / / / / / / * ( / / / / will prove that, under the assumption that C also has products, these coproducts form biproducts in Kℓ(T ) if and only if T is additive. Again, as in Lemma 13, a related result holds for the category Alg(T ). Definition 18 A category with biproducts is a category C with a zero object 0 ∈ C, such that, for any pair of objects A1 , A2 ∈ C, there is an object A1 ⊕A2 ∈ C that is both a product with projections πi : A1 ⊕ A2 → Ai and a coproduct with coprojections κi : Ai → A1 ⊕ A2 , such that πj ◦ κi = ( idAi 0Ai ,Aj Theorem 19 For a monad T on a category C with finite products (×, 1) and coproducts (+, 0), the fol lowing are equivalent. if i 6= j. if i = j (i). T is additive; (ii). the coproducts in C form biproducts in the Kleisli category Kℓ(T ); (iii). the products in C yield biproducts in the category of Eilenberg-Moore al- gebras Alg(T ). Here we shall only use this result for Kleisli categories, but we include the result for algebras for completeness. hf ,gi bc−1 hf , g iKℓ / T (X ) × T (Y ) Proof First we assume that T is additive and show that (+, 0) is a product in Kℓ(T ). As pro jections we take the maps pi from (5). For Kleisli maps f : Z → T (X ) and g : Z → T (Y ) there is a tuple via the map bc, as in / T (X + Y )(cid:17). def= (cid:16)Z One obtaines p1 • hf , g iKℓ = µ ◦ T (p1) ◦ bc−1 ◦ hf , g i = π1 ◦ bc ◦ bc−1 ◦ hf , g i = π1 ◦ hf , g i = f . Remaining details are left to the reader. Conversely, assuming that the coproduct (+, 0) in C forms a biproduct in Kℓ(T ), we have to show that the bicartesian map bc : T (X + Y ) → T (X ) × T (Y ) is an isomorphism. As + is a biproduct, there exist pro jection maps qi : X1 + X2 → Xi in Kℓ(T ) satisfying qj • κi = ( idXi 0Xi ,Xj From these conditions it follows that qi = pi , where pi is the map defined in (5). The ordinary pro jection maps πi : T (X1) × T (X2) → T (Xi ) are maps T (X1) × T (X2) → Xi in Kℓ(T ). Hence, as + is a product, there exists a unique map h : T (X1) × T (X2) → X1 + X2 in Kℓ(T ), i.e. h : T (X1) × T (X2 ) → T (X1 + X2 ) in C, such that p1 • h = π1 and p2 • h = π2 . It is readily checked that this map h is the inverse of bc. if i 6= j. if i = j 17 / / β To prove the equivalence of (i) and (iii), first assume that the monad T is additive. In the category Alg(T ) of algebras there is the standard product hα◦T (π1 ),β◦T (π2 )i (cid:16)T (X ) α / / X × Y (cid:17). / Y (cid:17) def= (cid:16)T (X × Y ) / X (cid:17) × (cid:16)T (Y ) In order to show that × also forms a coproduct in Alg(T ), we first show that for an arbitrary algebra γ : T (Z ) → Z the ob ject Z carries a commutative monoid structure. We do so by adapting the structure (+, 0) on T (Z ) from Lemma 17 to (+Z , 0Z ) on Z via def= (cid:16)Z × Z / T (Z ) × T (Z ) def= (cid:16)1 / Z (cid:17) 0 This monoid structure is preserved by homomorphisms of algebras. Now, we can form copro jections k1 = hid, 0Y ◦ !i : X → X × Y , and a cotuple of alge- f γ β g γ bra homomorphisms (T X α→ X ) −→ (T Z → Z ) and (T Y → X ) −→ (T Z → Z ) given by / Z (cid:17) / T (Z ) / T (Z ) +Z 0Z η×η + γ γ f ×g +Z / / Z × Z [f , g ]Alg def= (cid:16)X × Y / Z (cid:17). Again, remaining details are left to the reader. Finally, to show that (iii) implies (i), consider the algebra morphisms: / T (Xi )(cid:17) T (κi ) / (cid:16)T 2 (X1 + X2 ) (cid:16)T 2 (Xi ) / T (X1 + X2 )(cid:17). The free functor C → Alg(T ) preserves coproducts, so these T (κi ) form a coproduct diagram in Alg(T ). As × is a coproduct in Alg(T ), by assumption, the cotuple [T (κ1), T (κ2 )] : T (X1) × T (X2 ) → T (X1 + X2 ) in Alg(T ) is an isomorphism. The copro jections ℓi : T (Xi) → T (X1) × T (X2 ) satisfy ℓ1 = hπ1 ◦ ℓ1 , π2 ◦ ℓ2 i = hid, 0i, and similarly, ℓ2 = h0, idi. Now we compute: µ µ bc ◦ [T (κ1), T (κ2 )] ◦ ℓ1 = hµ ◦ T (p1 ), µ ◦ T (p2 )i ◦ T (κ1) = hµ ◦ T (p1 ◦ κ1 ), µ ◦ T (p2 ◦ κ1 )i = hµ ◦ T (η), µ ◦ T (0)i = hid, 0i = ℓ1 . Similarly, bc ◦ [T (κ1 ), T (κ2)] ◦ ℓ2 = ℓ2 , so that bc ◦ [T (κ1 ), T (κ2)] = id, making bc an isomorphism. (cid:3) It is well-known (see for instance [15, 1]) that a category with finite biprod- ucts (⊕, 0) is enriched over commutative monoids: each homset carries a com- mutative monoid structure (+, 0), and this structure is preserved by pre- and 18 / / / / / / / / / / / post-composition. The addition operation + on homsets is obtained as f + g def= (cid:16)X / Y (cid:17). The zero map is neutral element for this addition. One can also describe a monoid structure on each ob ject X as / X ⊕ X / Y ⊕ Y (6) hid,idi f ⊕g [id,id] X ⊕ X [id,id] / X 0 0. (7) We have just seen that the Kleisli category of an additive monad has biprod- ucts, using the addition operation from Lemma 17. When we apply the sum description (7) to such a Kleisli category its biproducts, we obtain precisely the original addition from Lemma 17, since the codiagonal ∇ = [id, id] in the Kleisli category is given T (∇) ◦ bc−1 . 4.1 Additive commutative monads In the remainder of this section we focus on the category ACMnd(C) of monads that are both additive and commutative on a distributive category C. As usual, we simply write ACMnd for ACMnd(Sets). For T ∈ ACMnd(C), the Kleisli category Kℓ(T ) is both symmetric monoidal—with (×, 1) as monoidal structure, see Lemma 2—and has biproducts (+, 0). Moreover, it is not hard to see that this monoidal structure distributes over the biproducts via the canonical map (Z × X ) + (Z × Y ) → Z × (X + Y ) that can be lifted from C to Kℓ(T ). We shall write SMBLaw ֒→ SMLaw for the category of symmetric monoidal Lawvere theories in which (+, 0) form not only coproducts but biproducts. No- tice that a pro jection π1 : n + m → n is necessarily of the form π1 = [id, 0], where 0 : m → n is the zero map m → 0 → n. The tensor ⊗ distributes over (+, 0) in SMBLaw, as it already does so in SMLaw . Morphisms in SMBLaw are functors that strictly preserve all the structure. The following result extends Corollary 3. Lemma 20 The (finitary) Kleisli construction on a monad yields a functor KℓN : ACMnd → SMBLaw. Proof It follows from Theorem 19 that (+, 0) form biproducts in KℓN (T ), for T an additive commutative monad (on Sets). This structure is preserved by functors KℓN (σ), for σ : T → S in ACMnd. (cid:3) We have already seen in Lemma 12 that the functor T : Law → Mnd defined in Lemma 7 restricts to a functor between symmetric monoidal Lawvere theories and commutative monads. We now show that it also restricts to a functor be- tween symmetric monoidal Lawvere theories with biproducts and commutative additive monads. Again, this restriction is left adjoint to the finitary Kleisli construction. 19 / / / / o o Lemma 21 The functor T : SMLaw → CMnd from Lemma 12 restricts to SMBLaw → ACMnd. Further, this restriction is left adjoint to the finitary Kleisli construction KℓN : ACMnd → SMBLaw. Proof First note that TL (0) is final: TL (0) = `i L(1, i) × 0i ∼= L(1, 0) × 00 ∼= 1, where the last isomorphism follows from the fact that (+, 0) is a biproduct in L and hence 0 is terminal. The resulting zero map 0X,Y : X → T (Y ) is given by x 7→ [κ0 (! : 1 → 0, ! : 0 → Y )]. To prove that the bicartesian map bc : TL (X + Y ) → TL (X ) × TL (Y ) is an isomorphism, we introduce some notation. For [κi (g , v)] ∈ TL(X + Y ), where g : 1 → i and v : i → X + Y , we form the pullbacks (in Sets) iX _ vX i v iY _ vY κ1 X / X + Y κ2 Y By universality of coproducts we can write i = iX + iY and v = vX + vY : iX + iY → X + Y . Then we can also write g = hgX , gY i : 1 → iX + iY . Hence, for [ki (g , v)] ∈ TL (X + Y ), bc([κi (g , v)]) = (cid:0)[κiX (gX , vX )], [κiY (gY , vY )](cid:1). It then easily follows that the map TL(X ) × TL (Y ) → TL (X + Y ) defined by (8) ([κi (g , v)], [κj (h, w)]) 7→ [κi+j (hg , hi, v + w)] is the inverse of bc. Checking that the unit of the adjunction T : SMLaw ⇆ CMnd : KℓN pre- serves the product structure is left to the reader. This proves that also the restricted functors form an adjunction. (cid:3) In the next two sections we will see how additive commutative monads and symmetric monoidal Lawvere theories with biproducts relate to commutative semirings. 5 Semirings and monads This section starts with some clarification about semirings and modules. Then it shows how semirings give rise to certain “multiset” monads, which are both commutative and additive. It is shown that the “evaluate at unit 1”-functor yields a map in the reverse direction, giving rise to an adjunction, as before. 20 / / o o / o o A commutative semiring in Sets consists of a set S together with two com- mutative monoid structures, one additive (+, 0) and one multiplicative (·, 1), where the latter distributes over the former: s · 0 = 0 and s · (t + r) = s · t + s · r. For more information on semirings, see [8]. Here we only consider commuta- tive ones. Typical examples are the natural numbers N, or the non-negative rationals Q≥0 , or the reals R≥0 . One way to describe semirings categorically is by considering the additive monoid (S, +, 0) as an ob ject of the category CMon of commutative monoids, carrying a multiplicative monoid structure I 1→ S ·← S ⊗ S in this category CMon. The tensor guarantees that multiplication is a bihomomorphism, and thus distributes over additions. In the present context of categories with finite products we do not need to use these tensors and can give a direct categorical formulation of such semirings, as a pair of monoids 1 0→ S +← S × S and 1 1→ S ·← S × S making the following distributivity diagrams commute. id×0 S × 1 S × S (S × S ) × S dbl (S × S ) × (S × S ) ·×· / S × S ! 1 0 · / S +×id S × S · + / S where dbl = hπ1×id, π2×idi is the doubling map that was also used in Lemma 15. With the obvious notion of homomorphism between semirings this yields a cat- egory CSRng(C) of (commutative) semirings in a category C with finite prod- ucts. Associated with a semiring S there is a notion of module over S . It consists of a commutative monoid (M , 0, +) together with a (multiplicative) action ⋆ : S × M → M that is an additive bihomomorphism, that is, the action preserves the additive structure in each argument separately. We recall that the properties of an action are given categorically by ⋆ ◦ (· × id) = ⋆ ◦ (id × ⋆) ◦ α−1 : (S × S ) × M → M and ⋆ ◦ (1 × id) = π2 : 1 × M → M . The fact that ⋆ is an additive bihomomorphism is expressed by S × (M × M ) dbl ′ (S × M ) × (S × M ) dbl (S × S ) × M id×+ ⋆×⋆ M × M + +×id ∗ ∗ S × M S × M / M where dbl ′ is the obvious duplicator of S . Preservation of zeros is simply ⋆ ◦ (0 × id) = 0 ◦ π1 : 1 × M → M and ⋆ ◦ (id × 0) = 0 ◦ π2 : S × 1 → M . We shall assemble such semirings and modules in one category Mod(C) with triples (S, M , ⋆) as ob jects, where ⋆ : S × M → M is an action as above. A morphism (S1 , M1 , ⋆1 ) → (S2 , M2 , ⋆2) consists of a pair of morphisms f : S1 → S2 and g : M1 → M2 in C such that f is a map of semirings, f is a map of monoids, and the actions interact appropriately: ⋆2 ◦ (f × g ) = g ◦ ⋆1 . 21 / / / / / / / / / o o / o o 5.1 From semirings to monads To construct an adjunction between semirings and additive commutative mon- ads we start by defining, for each commutative semiring S , the so-called multiset monad on S and show that this monad is both commutative and additive. Definition 22 For a semiring S , define a “multiset” functor MS : Sets → Sets on a set X by MS (X ) = {ϕ : X → S supp(ϕ) is finite}, where supp(ϕ) = {x ∈ X ϕ(x) 6= 0} is cal led the support of ϕ. For a function f : X → Y one defines MS (f ) : MS (X ) → MS (Y ) by: MS (f )(ϕ)(y ) = Px∈f −1 (y) ϕ(x). Such a multiset ϕ ∈ MS (X ) may be written as formal sum s1x1 + · · · + sk xk , where supp(ϕ) = {x1 , . . . , xk } and si = ϕ(xi ) ∈ S describes the “multiplicity” of the element xi . In this notation one can write the application of MS on a map f as MS (f )(Pi sixi ) = Pi si f (xi ). Functoriality is then obvious. Lemma 23 For each semiring S , the multiset functor MS forms a commutative and additive monad, with unit and multiplication: η X x µ / 1x / MS (X ) / MS (X ) MS (MS (X )) Pi siϕi / λx ∈ X . Pi siϕi (x). Proof The verification that MS with these η and µ indeed forms a monad is left to the reader. We mention that for commutativity and additivity the relevant maps are given by: MS (X ) × MS (Y ) dst / MS (X × Y ) MS (X + Y ) bc / MS (X ) × MS (Y ) (ϕ, ψ) / λ(x, y ). ϕ(x) · ψ(y ) χ / (χ ◦ κ1 , χ ◦ κ2 ). Clearly, bc is an isomorphism, making MS additive. (cid:3) Lemma 24 The assignment S 7→ MS yields a functor M : CSRng → ACMnd. Proof Every semiring homomorphism f : S → R, gives rise to a monad mor- phism M(f ) : MS → MR with components defined by M(f )X (Pi sixi ) = Pi f (si )xi . It is left to the reader to check that M(f ) is indeed a monad morphism. (cid:3) For a semiring S , the category Alg(MS ) of algebras of the multiset monad MS is (equivalent to) the category ModS (C) ֒→ Mod(C) of modules over S . This is not used here, but just mentioned as background information. 22 / / / / / / / / 5.2 From monads to semirings A commutative additive monad T on a category C gives rise to two commutative monoid structures on T (1), namely the multiplication defined in Lemma 10 and the addition defined in Lemma 17 (considered for X = 1). In case the category C is distributive these two operations turn T (1) into a semiring. Lemma 25 Each commutative additive monad T on a distributive category C with terminal object 1 gives rise to a semiring E (T ) = T (1) in C. The mapping T 7→ E (T ) yields a functor ACMnd(C) → CSRng(C). Proof For a commutative additive monad T on C, addition on T (1) is given by T (∇) ◦ bc−1 : T (1) × T (1) → T (1) with unit 01,1 : 1 → T (1) as in Lemma 17, the multiplication is given by µ ◦ T (λ) ◦ st : T (1) × T (1) → T (1) with unit η1 : 1 → T (1) as in Lemma 10. It was shown in the lemmas just mentioned that both addition and multipli- cation define a commutative monoid structure on T (1). The following diagram proves distributivity of multiplication over addition. (T (1) × T (1)) × T (1) bc−1×id T (1 + 1) × T (1) T (∇)×id T (1) × T (1) dbl st st (T (1) × T (1)) × (T (1) × T (1)) T ((1 + 1) × T (1)) T (∇×id) T (1 × T (1)) st×st ∼= T (1 × T (1)) × T (1 × T (X )) bc−1 T (1 × T (1) + 1 × T (1)) T (λ)×T (λ) T (λ+λ T (λ) T 2(1) × T 2 (1) bc−1 µ×µ T ([T (κ1 ),T (κ2 )]) / T (T (1) + T (1)) )SSSSSSSSSSSSSSSS T (∇) T 2 (∇) T 2(1 + 1) T (1) × T (1) bc−1 µ / T (1 + 1) T (∇) T 2 (1) µ / T (1) Here we rely on Lemma 15 for the commutativity of the upper and lower square on the left. In a distributive category 0 ∼= 0 × X , for every ob ject X . In particular T (0 × T (1)) ∼= T (0) ∼= 1 is final. This is used to obtain commutativity of the 23 / / / / / / / / / ) / / / / upper-left square of the following diagram proving 0 · s = 0: T (1) ∼= / T (0) × T (1) T (!)×id T (1) × T (1) ! T (0) st st ∼= / T (0 × T (1)) %LLLLLLLLLL T (!) T (λ) T 2 (1) T (!×id) / T (1 × T (1)) T (λ) id / T 2(1) T (!) µ T (1) For a monad morphism σ : T → S , we define E (σ) = σ1 : T (1) → S (1). By Lemma 5, σ1 commutes with the multiplicative structure. As σ1 = T (id) ◦ σ1 = Ad((σ, id)), it follows from Lemma 17 that σ1 also commutes with the additive structure and is therefore a CSRng-homomorphism. (cid:3) 5.3 Adjunction between monads and semirings The functors defined in the Lemmas 24 and 25, considered on C = Sets, form an adjunction M : CSRng ⇆ ACMnd : E . To prove this adjunction we first show that each pair (T , X ), where T is a commutative additive monad on a category C and X an ob ject of C, gives rise to a module on C as defined at the beginning of this section. Lemma 26 Each pair (T , X ), where T is a commutative additive monad on a category C and X is an object of C, gives rise to a module Mod(T , X ) = (T (1), T (X ), ⋆). Here T (1) is the commutative semiring defined in Lemma 25 and T (X ) is the commutative monoid defined in Lemma 17. The action map is given by ⋆ = T (λ) ◦ dst : T (1) × T (X ) → T (X ). The mapping (T , X ) 7→ Mod(T , X ) yields a functor ACMnd(C) × C → Mod(C). Proof Checking that ⋆ defines an appropriate action requires some work but is essentially straightforward, using the properties from Lemma 15. For a pair of maps σ : T → S in ACMnd(C) and g : X → Y in C, we define a map Mod(σ, g ) by (T (1), T (X ), ⋆T ) (σ1 ,σY ◦T (g)) / (S (1), S (Y ), ⋆S ). Note that, by naturality of σ , one has σY ◦ T (g ) = S (g ) ◦ σX . It easily follows that this defines a Mod(C)-map and that the assignment is functorial. (cid:3) Lemma 27 The pair of functors M : CSRng ⇆ ACMnd : E forms an ad- junction, M ⊣ E . 24 / / / / / / % / / / Proof For a semiring S and a commutative additive monad T on Sets there are (natural) bijective correspondences: MS = M(S ) σ / T in CAMnd S f / E (T ) = T (1) in CSRng Given σ : MS → T , one defines a semiring map σ : S → T (1) by / T (1)(cid:17). σ = (cid:16)S Conversely, given a semiring map f : S → T (1), one gets a monad map f : MS → T with components: / MS (1) λs.(λx.s) σ1 f X / MS (X ) / Pif (si ) ⋆ η(xi ), / T (X ) given by Pisixi where the sum on the right hand side is the addition in T (X ) defined in Lemma 17 and ⋆ is the action of T (1) on T (X ) defined in Lemma 26. Showing that f is indeed a monad morphism requires some work. In doing so one may rely on the properties of the action and on Lemma 17. The details are left to the reader. (cid:3) Notice that the counit of the above adjunction EM(S ) = MS (1) → S is an isomorphism. Hence this adjunction is in fact a reflection. 6 Semirings and Lawvere theories In this section we will extend the adjunction between commutative monoids and symmetric monoidal Lawvere theories depicted in Figure 3 to an adjunction between commutative semirings and symmetrical monoidal Lawvere theories with biproducts, i.e. between the categories CSRng and SMBLaw. 6.1 From semirings to Lawvere theories Composing the multiset functor M : CSRng → ACMnd from the previous section with the finitary Kleisli construction KℓN yields a functor from CSRng to SMBLaw. This functor may be described in an alternative (isomorphic) way by assigning to every semiring S the Lawvere theory of matrices over S , which is defined as follows. Definition 28 For a semiring S , the Lawvere theory Mat(S ) of matrices over S has, for n, m ∈ N morphisms (in Sets) n × m → S , i.e. n × m matrices over S , as morphisms n → m. The identity idn : n → n is given by the identity matrix: idn (i, j ) = ( 1 0 if i 6= j. if i = j 25 / / / / / The composition of g : n → m and h : m → p is given by matrix multiplication: (h ◦ g )(i, k) = Pj g (i, j ) · h(j, k). The coprojections κ1 : n → n + m and κ2 : m → n + m are given by κ1 (i, j ) = ( 1 if i = j κ2 (i, j ) = ( 1 if j ≥ n and j − n = i 0 otherwise. 0 otherwise. Lemma 29 The assignment S 7→ Mat(S ) yields a functor CSRng → Law . The two functors MatE and KℓN : ACMnd → Law are natural ly isomorphic. Proof A map of semirings f : S → R gives rise to a functor Mat(f ) : Mat(S ) → Mat(R) which is the identity on ob jects and which acts on morphisms by post- composition: h : n × m → S in Mat(S ) is mapped to f ◦ h : n × m → T in Mat(T ). It is easily checked that Mat(f ) is a morphism of Lawvere theories and that the assigment is functorial. To prove the second claim we define two natural transformations. First we define ξ : MatE → KℓN with components ξT : Mat(T (1)) → KℓN (T ) that are the identity on ob jects and send a morphism h : n × m → T (1) in Mat(T (1)) to the morphism ξT (h) in KℓN (T ) given by hh( ,j)ij∈m bc−1 m / / T (1)m ξT (h)=(cid:16)n / T (m)(cid:17), where bc−1 m is the inverse of the generalised bicartesian map / T (1)m(cid:17). bcm = (cid:16)T (m) = T (`m 1) And secondly, in the reverse direction, we define θ : KℓN → MatE with com- ponents θT : KℓN (T ) → Mat(T (1)) that are the identity on ob jects and send a morphism g : n → T (m) in KℓN (T ) to the morphism θT (g ) : n × m → T (1) in Mat(T (1)) given by θT (g )(i, j ) = (πj ◦ bcm ◦ g )(i). (9) It requires some work, but is relatively straightforward to check that the com- ponents ξT and θT are Law-maps. To prove preservation of the composition by ξT and θT one uses the definition of addition and multiplication in T (1) and (generalisations of ) the properties of the map bc listed in Lemma 15. A short computation shows that the functors are each other’s inverses. The naturality of both ξ and θ follows from (a generalisation of ) point (i) of Lemma 15. (cid:3) The pair of functors M : CSRng ⇆ ACMnd : E forms a reflection, EM ∼= id (Lemma 27). Combining this with the previous proposition, it follows that also the functors Mat, KℓNM : CSRng → Law are naturally isomorphic. Hence, 26 / / the functor Mat : CSRng → Law may be viewed as a functor from commuta- tive semirings to symmetric monoidal Lawvere theories with biproducts. For a commutative semiring S the pro jection maps π1 : n+m → n and π2 : n+m → m in Mat(S ) are defined in a similar way as the copro jection maps from Def- inition 28. For a pair of maps g : m → p, h : n → q , the tensor product g ⊗ h : (m × n) → (p × q) is the map g ⊗ h : (m × n) × (p × q) → S defined as (g ⊗ h)((i0 , i1 ), (j0 , j1 )) = g (i0 , j0 ) · h(i1 , j1 ), where · is the multiplication from S . 6.2 From Lawvere theories to semirings In Section 3.1, just after Lemma 11, we have already seen that the homset L(1, 1) of a Lawvere theory L ∈ SMLaw is a commutative monoid, with multiplication given by composition of endomaps on 1. In case L also has biproducts we have, by (6), an addition on this homset, which is preserved by composition. Combining those two monoid structures yields a semiring structure on L(1, 1). This is standard, see e.g. [1, 15, 11]. The assignment of the semiring L(1, 1) to a Lawvere theory L ∈ SMBLaw is functorial and we denote this functor, as in Section 3.1, by H : SMBLaw → CSRng. 6.3 Adjunction between semirings and Lawvere theories Our main result is the adjunction on the right in the triangle of adjunctions for semirings, see Figure 4. Lemma 30 The pair of functors Mat : CSRng ⇄ SMBLaw : H, forms an adjunction Mat ⊣ H. Proof For S ∈ CSRng and L ∈ SMBLaw there are (natural) bijective corre- spondences: Mat(S ) F / / L in SMBLaw S f / H(L) in CSRng Given F one defines a semiring map F : S → H(L) = L(1, 1) by s 7→ F (1 × 1 λx. s−−−→ S ). Note that 1 × 1 λx. s−−−→ S is an endomap on 1 in Mat(S ) which is mapped by F to an element of L(1, 1). Conversely, given f one defines a SMBLaw-map f : Mat(S ) → L which sends a morphism h : n → m in Mat(S ), i.e. h : n × m → S in Sets, to the following morphism n → m in L, forming an n-cotuple of m-tuples f (h) = (cid:16)n (cid:2)(cid:10)f (h(i,j))(cid:11)j<m (cid:3)i<n / 27 / m(cid:17) . / CSRng M ⊣ Mat∼=KℓNM E ∼=HKℓN H ⊣ ACMnd T ⊥ KℓN SMBLaw Figure 4: Triangle of adjunctions starting from commutative semirings, with commutative additive monads, and symmetric monoidal Lawvere theories with biproducts. It is readily checked that F : S → L(1, 1) is a map of semirings. To show that f : Mat(S ) → L is a functor one has to use the definition of the semiring structure on L(1, 1) and the properties of the biproduct on L. One easily verifies that f preserves the biproduct. To show that it also preserves the monoidal structure one has to use that, for s, t ∈ L(1, 1), s ⊗ t = t ◦ s (= s ◦ t). (cid:3) The results of Section 5 and 6 are summarized in Figure 4. 7 Semirings with involutions In this final section we enrich our approach with involutions. Actually, such involutions could have been introduced for monoids already. We have not done so for practical reasons: involutions on semirings give the most powerful results, combining daggers on categories with both symmetric monoidal and biproduct structure. An involutive semiring (in Sets) is a semiring (S, +, 0, ·, 1) together with a unary operation ∗ that preserves the addition and multiplication, i.e. (s + t)∗ = s∗ + t∗ and 0∗ = 0, and (s · t)∗ = s∗ · t∗ and 1∗ = 1, and is involutive, i.e. (s∗ )∗ = s. The complex numbers with conjugation form an example. We denote the category of involutive semirings, with homomorphisms that preserve all structure, by ICSRng. The adjunction M : CSRng ⇆ ACMnd : E considered in Lemma 27 may be restricted to an adjunction between involutive semirings and so-called invo- lutive commutative additive monads (on Sets), which are commutative additive monads T together with a monad morphism ζ : T → T satisfying ζ ◦ ζ = id. We call ζ an involution on T , just as in the semiring setting. A morphism between such monads (T , ζ ) and (T ′ , ζ ′ ), is a monad morphism σ : T → T ′ preserving the involution, i.e. satisfying σ ◦ ζ = ζ ′ ◦ σ . We denote the category of involutive commutative additive monads by IACMnd. Lemma 31 The functors M : CSRng ⇆ ACMnd : E from Lemma 24 and Lemma 25 restrict to a pair of functors M : ICSRng ⇆ IACMnd : E . The restricted functors form an adjunction M ⊢ E . 28 } } * * 1 1 = = j j q q ζX : MS (X ) → MS (X ), Proof Given a semiring S with involution ∗, we may define an involution ζ on the multiset monad M(S ) = MS with components Pi sixi 7→ Pi s∗ i xi . Conversely, for an involutive monad (T , ζ ), the map ζ1 gives an involution on the semiring E (T ) = T (1). A simple computation shows that the unit and the counit of the adjunction M : CSRng ⇆ ACMnd : E from Lemma 27 preserve the involution (on semir- ings and on monads respectively). Hence the restricted functors again form an adjunction. (cid:3) The adjunction Mat : CSRng ⇄ SMBLaw : H from Lemma 30 may also be restricted to involutive semirings. To do so, we have to consider dagger categories. A dagger category is a category C with a functor † : Cop → C that is the identity on ob jects and satisfies, for all morphisms f : X → Y , (f † )† = f . The functor † is called a dagger on C. Combining this dagger with the categorical structure we studied in Section 6 yields a so-called dagger symmetric monoidal category with dagger biproducts, that is, a category C with a symmetric monoidal structure (⊗, I ), a biproduct structure (⊕, 0) and a dagger †, such that, for all morphisms f and g , (f ⊗ g )† = f † ⊗ g † , all the coherence isomorphisms α, ρ and γ are dagger isomorphisms and, with respect to the biproduct structure, κi = π† i , where a dagger isomorphism is an isomorphism f satisfying f −1 = f † . Further details may be found in [1, 2, 11]. We will denote the category of dagger symmetric monoidal Lawvere theories with dagger biproducts such that the monoidal structure distributes over the biproduct structure by DSMBLaw. Morphisms in DSMBLaw are maps in SMBLaw that (strictly) commute with the daggers. Lemma 32 The functors Mat : CSRng ⇄ SMBLaw : H defined in Section 6 restrict to a pair of functors Mat : ICSRng ⇄ DSMBLaw : H. The restricted functors form an adjunction, Mat ⊣ H. Proof For an involutive semiring S , we may define a dagger on the Lawvere theory Mat(S ) by assigning to a morphism f : n → m in Mat(S ) the morphism f † : m → n given by f † (i, j ) = f (j, i)∗ . (10) Some short and straightforward computations show that the functor † is in- deed a dagger on Mat(S ), which interacts appropriately with the monoidal and biproduct structure. For a dagger symmetric monoidal Lawvere theory L with dagger biproduct, it easily follows from the properties of the dagger that this functor induces an involution on the semiring H(L) = L(1, 1), namely via s 7→ s† . The unit and the counit of the adjunction Mat : CSRng ⇆ SMBLaw from Lemma 30 preserve the involution and the dagger respectively. Hence, also the restricted functors form an adjunction. (cid:3) 29 To complete our last triangle of adjunctions, recall that, for the Lawvere the- ory associated with a (involutive commutative additive) monad T , KℓN (T ) ∼= Mat(E (T )), see Proposition 29. Hence, using the previous two lemmas, the fini- tary Kleisli construction restricts to a functor KℓN : IACMnd → DSMBLaw. For the other direction we use the following result. Lemma 33 The functor T : SMBLaw → ACMnd from Lemma 12 restricts to DSMBLaw → IACMnd, and yields a left adjoint to KℓN : IACMnd → DSMBLaw. Proof To start, for a Lawvere theory L ∈ DSMBLaw with dagger † we have to define an involution ζ : TL → TL . For a set X this involves a map TL(X ) = (cid:0) `i∈N L(1, i) × X i(cid:17)/ ∼ / (cid:0) `i∈N L(1, i) × X i(cid:17)/ ∼ = TL (X ) / [κi (hg † 0 , . . . , g † [κi (g , v)] i−1 i, v)], ζX / where g : 1 → i is written as g = hg0 , . . . , gi−1 i using that i = 1 + · · · + 1 is not only a sum, but also a product. Clearly, ζ is natural, and satisfies ζ ◦ ζ = id. This ζ is also a map of monads; commutatution with multiplication µ requires commutativity of composition in the homset L(1, 1). The unit of the adjunction η : L → KℓN (TL ) ∼= Mat(TL(1)) commutes with daggers, since for f : n → m in L we get η(f )† = η(f † ) via the following argument in Mat(TL (1)). For i < n and j < m, η(f )† (i, j ) = η(f )(i, j )∗ by (10) = πj bcm (κm (f ◦ κi , idm ))∗ by (9) = κ1 (πj ◦ f ◦ κi , id1 )∗ by definition of bc, see (8) = κ1 (πj ◦ f ◦ κi , id)† since (−)∗ = (−)† on TL (1) = κ1 ((πj ◦ f ◦ κi )† , id) = κ1 (κ† i ◦ f † ◦ π† j , id) = κ1 (πi ◦ f † ◦ κj , id) = η(f † )(i, j ). (cid:3) In the definition of the involution ζ on the monad TL in this proof we have used that + is a (bi)product in the Lawvere theory L, namely when we decom- pose the map g : 1 → i into its components πa ◦ g : 1 → 1 for a < i. We could have avoided this biproduct structure by first taking the dagger g † : i → 1, and then precomposing with copro jections g † ◦ κa : 1 → 1. Again applying daggers, cotupling, and taking the dagger one gets the same result. This is relevant if one wishes to consider involutions/daggers in the context of monoids, where products in the corresponding Lawvere theories are lacking. By combining the previous three lemmas we obtain another triangle of ad- junctions in Figure 5. This concludes our survey of the interrelatedness of scalars, monads and categories. 30 / ICSRng M ⊣ Mat∼=KℓNM E ∼=HKℓN H ⊣ IACMnd T ⊥ KℓN DSMBLaw Figure 5: Triangle of adjunctions starting from involutive commutative semi- rings, with involutive commutative additive monads, and dagger symmetric monoidal Lawvere theories with dagger biproducts. References [1] S. Abramsky and B. Coecke. A categorical semantics of quantum protocols. In Logic in Computer Science, pages 415–425. IEEE, Computer Science Press, 2004. [2] S. Abramsky and B. Coecke. A categorical semantics of quantum protocols. In K. Engesser, Dov M. Gabbai, and D. Lehmann, editors, Handbook of Quantum Logic and Quantum Structures, pages 261–323. North Holland, Elsevier, Computer Science Press, 2009. [3] J. Ad´amek and J. Velebil. Analytic functors and weak pullbacks. Theory and Applications of Categories, 21(11):191–209, 2008. Available from URL: www.tac.mta.ca/tac/volumes/21/11/. [4] S. Awodey. Category Theory. Oxford Logic Guides. Oxford Univ. Press, 2006. [5] F. Borceux. Handbook of Categorical Algebra, volume 50, 51 and 52 of Encyclopedia of Mathematics. Cambridge Univ. Press, 1994. [6] P.-L. Curien. Operads, clones, and distributive laws. Available from URL: www.pps.jussieu.fr/~curien/Operads- Strasbourg.ps, 2008. [7] M. Fiore, G. Plotkin, and D. Turi. Abstract syntax and variable binding. In Logic in Computer Science, pages 193–202. IEEE, Computer Science Press, 1999. [8] J. S. Golan. Semirings and their Applications. Kluwer Academic Publishers, 1999. [9] S. Goncharov, L. Schroder, and T. Mossakowski. Kleene monads: Handling iteration in a framework of generic effects. In A. Kurz and A. Tarlecki, ed- itors, Conference on Algebra and Coalgebra in Computer Science (CALCO 2009), number 5728 in Lect. Notes Comp. Sci., pages 18–33. Springer, Berlin, 2009. 31 * * 1 1 < < j j q q [10] R. Hasegawa. Two applications of analytic functors. Theor. Comp. Sci., 272(1-2):113–175, 2002. [11] C. Heunen. Categorical Quantum Models and Logics. PhD thesis, Radboud Univ. Nijmegen, 2010. [12] M. Hyland and J. Power. The category theoretic understanding of univer- sal algebra: Lawvere theories and monads. In L. Cardelli, M. Fiore, and G. Winskel, editors, Computation, Meaning, and Logic: Articles dedicated to Gordon Plotkin, number 172 in Elect. Notes in Theor. Comp. Sci., pages 437–458. Elsevier, Amsterdam, 2007. [13] B. Jacobs. From coalgebraic to monoidal traces. In Coalgebraic Methods in Computer Science, Elect. Notes in Theor. Comp. Sci. Elsevier, Amsterdam, 2010, to appear. [14] A. Joyal. Foncteurs analytiques et esp`eces de structures. In G. Labelle and P. Leroux, editors, Combinatoire Enumerative, number 1234 in Lect. Notes Math., pages 126–159. Springer, Berlin, 1986. [15] G.M. Kelly and M. L. Laplaza. Coherence for compact closed categories. Journ. of Pure & Appl. Algebra, 19:193–213, 1980. [16] A. Kock. Closed categories generated by commutative monads. Journ. Austr. Math. Soc., XII:405–424, 1971. [17] S. Mac Lane. Categories for the Working Mathematician. Springer, Berlin, 1971. [18] F.W. Lawvere. Functorial Semantics of Algebraic Theories and Some Alge- braic Problems in the context of Functorial Semantics of Algebraic Theories. PhD thesis, Columbia Univ., 1963. Reprinted in Theory and Applications of Categories, 5:1–121, 2004. [19] H. Wolff. Monads and monoids on symmetric monoidal closed categories. Archiv der Mathematik, XXIV:113–120, 1973. 32
1807.05645	2	1807	2019-01-29T22:13:14	Stable noncommutative polynomials and their determinantal representations	[ "math.RA", "math.FA" ]	A noncommutative polynomial is stable if it is nonsingular on all tuples of matrices whose imaginary parts are positive definite. In this paper a characterization of stable polynomials is given in terms of strongly stable linear matrix pencils, i.e., pencils of the form $H+iP_0+P_1x_1+\cdots+P_dx_d$, where $H$ is hermitian and $P_j$ are positive semidefinite matrices. Namely, a noncommutative polynomial is stable if and only if it admits a determinantal representation with a strongly stable pencil. More generally, structure certificates for noncommutative stability are given for linear matrix pencils and noncommutative rational functions.	math.RA	math	STABLE NONCOMMUTATIVE POLYNOMIALS AND THEIR DETERMINANTAL REPRESENTATIONS JURIJ VOL CI C1 Abstract. A noncommutative polynomial is stable if it is nonsingular on all tuples of matrices whose imaginary parts are positive definite. In this paper a characterization of stable polynomials is given in terms of purely stable linear matrix pencils, i.e., pencils of the form H + iP0 + P1x1 + · · · + Pdxd, where H is hermitian and Pj are positive semidefinite matrices. Namely, a noncommutative polynomial is stable if and only if it admits a determinantal representation with a purely stable pencil. More generally, structure certificates for noncommutative stability are given for linear matrix pencils and noncommutative rational functions. 9 1 0 2 n a J 9 2 ] . A R h t a m [ 2 v 5 4 6 5 0 . 7 0 8 1 : v i X r a 1. Introduction A multivariate polynomial f ∈ C[x1, . . . , xd] is stable if f (α) 6= 0 whenever Im αj > 0 for all j = 1, . . . , d. Stable polynomials and their variations, such as Hurwitz and Schur polynomials, originated in control theory [FB87, Bos88, Kum89, KT-M99]. However, recent years saw a renewed interest in stable polynomials in a quite wide range of areas [Wag11]. A decade ago various problems in combinatorics, matrix theory and statistical mechanics were resolved using stable polynomials, such as the Johnson conjectures [BB08], new proofs of the Van der Waerden and the Schrijver-Valiant conjectures [Gur08], and Lee-Yang type theorems [BB09]. In real algebraic geometry [BCR98, BPT13], stable polynomials emerged through their connection to hyperbolic polynomials [KPV15, JT18], most prominently in the solutions of the Lax conjecture [HV07] and the Kadison-Singer conjecture [MSS15]. From a complex analysis perspective, stable polynomials are closely related to the Schur-Agler class of rational inner functions [Agl90, Kne11, GK-VVW16]. The common thread of these developments are determinantal representations of stable polynomials using linear matrix pencils with a distinguished structure [Bra11, NT12]. Namely, if S is a symmetric matrix and P1, . . . , Pd are positive semidefinite matrices, then (1.1) f = det(S + P1x1 + · · · + Pdxd) ∈ R[x1, . . . , xd] [BB08, Proposition 2.4]. Conversely, as a is either zero or a stable polynomial; see e.g. consequence of the celebrated Helton-Vinnikov theorem [HV07], every real stable poly- nomial f in two variables is of the form (1.1) by [BB08, Theorem 5.4]. However, the converse fails for polynomials in more than two variables [Bra11]. The existence of a spe- cial determinantal representation (1.1) is closely related to having a structural certificate Date: January 31, 2019. 2010 Mathematics Subject Classification. Primary 13J30, 15A22; Secondary 26C15, 93D05. Key words and phrases. Stable polynomial, linear matrix pencil, determinantal representation, Hurwitz stability, noncommutative rational function. 1Research supported by the Deutsche Forschungsgemeinschaft (DFG) Grant No. SCHW 1723/1-1. 1 2 J. VOL CI C for linear matrix pencils to be invertible on the positive orthant in Cd. Such problems have natural analogs in free analysis and free real algebraic geometry. Here, pencils are evaluated on matrices rather than on scalars, and these new noncommutative problems are often more tractable since matrix evaluations capture the structural properties more completely than just scalar evaluations. For example, Helton [Hel02, Theorem 1.1] showed that every positive noncommutative polynomial is a sum of hermitian squares; also see [HMV06, HKM12, BPT13, KPV17] for further results of this flavor. The aim of this paper is to introduce stable noncommutative polynomials and to prove that they admit "perfect" determinantal representations. This is achieved by proving a structural theorem for stable linear matrix pencils. Main results. Let x = (x1, . . . , xd) be freely noncommuting variables. In our noncom- mutative setting, the positive orthant in Cd is replaced by the set of all tuples of matrices whose imaginary part is positive definite, which we call the matricial positive orthant and denote Hd. Then we say that a linear matrix pencil L is stable if L(X) is invertible for every X ∈ Hd. For example, if H is a hermitian matrix and P0, . . . , Pd are positive semidefinite matrices such that ker H ∩Tj ker Pj = {0}, then L = H + iP0 + P1x1 + · · · Pdxd (1.2) is a stable pencil (Proposition 2.4). Due to their special structure we call pencils of the form (1.2) purely stable. Our first main result states that every stable pencil is built of purely stable pencils. Theorem A. A δ × δ linear pencil L is stable if and only if there exist D, E ∈ GLδ(C) such that L1 DLE =  ⋆ ⋆ . . . ⋆ Lℓ ,   where L1, . . . , Lℓ are purely stable pencils. Theorem A is a special case of Theorem 2.10 which deals more generally with rectan- gular pencils. Its proof also yields an algorithm relying on semidefinite programming for checking whether a pencil is stable (Subsection 2.3.1). Note that Theorem A is especially intriguing since it represents an algebraic certificate for invertibility on an open matricial set; usually such certificates are obtained for closed (convex) sets [HKM12, BMV18] or are less clean [KPV17]. We also obtain a strengthened version of Theorem A for hermitian pencils (Proposition 2.11), and a size bound for invertibility of linear matrix pencils on the matricial polydisk (Corollary 2.12). Next we characterize noncommutative rational functions that are regular on Hd (The- orem 3.2) by combining Theorem A and realization theory for noncommutative rational functions [BGM05, BR11], leading to determinantal representations of stable noncommu- tative polynomials. We say that f ∈ C<x> is stable if det f (X) 6= 0 for all X ∈ Hd. Theorem B. Let f ∈ C<x>. Then f is stable if and only if there exists a purely stable pencil L such that det f (X) = det L(X) for all matrix tuples X. STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 3 See Theorem 3.7 for the proof. Finally, we consider hermitian polynomials, which are In contrast to the commutative setting, noncommutative analogs of real polynomials. stable hermitian polynomials display surprisingly rigid behavior. Theorem C. Every stable irreducible hermitian polynomial is affine. See Theorem 3.9 for a more precise statement. The paper concludes with Section 4 on possible applications of the derived theory in multidimensional systems and circuits. Acknowledgments. The author thanks Andreas Thom for bringing this topic to his attention, and Igor Klep for fruitful suggestions. 2. Stable pencils In this section we completely characterize stable linear matrix pencils, i.e., rectangular pencils that have full rank on the matricial positive orthant. We prove that every such pencil is equivalent to a lower block triangular pencil whose diagonal blocks are stable for obvious reasons (and thus called purely stable pencils). This result is then strengthened for hermitian pencils. Lastly, the characterization is extended to other classical notions of stability. 2.1. Notation. We start by introducing the basic terminology used throughout the pa- per, including purely stable pencils. 2.1.1. Linear matrix pencils. For d ∈ N let x = (x1, . . . , xd) be a tuple of freely noncom- muting variables and let C<x> be the free C-algebra generated by x. If A0, . . . , Ad ∈ Cδ×ε, then L = A0 + A1x1 + · · · + Adxd ∈ Cδ×ε ⊗C C<x> = C<x>δ×ε is a linear matrix pencil of size δ × ε. If δ = ε, then we simply say that L is of size δ. If X ∈ Mn(C)d, then the evaluation of L at X is defined as L(X) = A0 ⊗ I +Xj Aj ⊗ Xj ∈ Cδn×εn, where ⊗ is the Kronecker product. Let us also denote Md =Sn∈N Mn(C)d. 2.1.2. Real and imaginary part of a matrix. Let Hn(C) ⊂ Mn(C) denote the R-subspace of hermitian matrices. For X ∈ Mn(C) let Re X = 1 2 (X + X ∗), Im X = 1 2i (X − X ∗). Then Re X, Im X ∈ Hn(C) and X = Re X + i Im X. Lemma 2.1. Let X ∈ Mn(C) and Im X (cid:23) 0. Then ker X = ker(Re X) ∩ ker(Im X). Proof. The inclusion ⊇ clearly holds. Conversely, let v ∈ ker X. Then h(Re X)v, vi + ih(Im X)v, vi = hXv, vi = 0. Now h(Re X)v, vi, h(Im X)v, vi ∈ R implies h(Re X)v, vi = h(Im X)v, vi = 0. 4 J. VOL CI C Since Im X (cid:23) 0, we have (Im X)v = 0, and therefore (Re X)v = Xv − i(Im X)v = 0. Hence v ∈ ker(Re X) ∩ ker(Im X). (cid:3) 2.2. Stable pencils. This subsection introduces stable pencils, which are the core objects of this paper. Then we single out two particular kinds of such pencils that are stable "for obvious reasons", purely stable and S-stable pencils. Let {(X1, . . . , Xd) ∈ Mn(C) : Im Xj ≻ 0 ∀j} ⊂ Md be the matricial positive orthant. The sets Hd ∩ Mn(C)d are closely related to Siegel upper half-spaces [vdG08, JT18]. Hd = [n∈N Definition 2.2. A linear matrix pencil L is stable if L(X) has full rank for all X ∈ Hd. The next property is the first step towards a structural characterization of stable pencils. Definition 2.3. A pencil L = H + iP0 +Pd H ∈ Hδ(C), Pj (cid:23) 0 ∀j = 0, . . . , d, j=1 Pjxj of size δ is purely stable if ker H ∩ d \j=0 ker Pj = {0}. The above terminology is justified by the following proposition. Proposition 2.4. Every purely stable pencil is stable. Proof. Let X ∈ Hd ∩ Mn(C)d and let L = H + iP0 +Pj>0 Pjxj be purely stable. Then Pj ⊗ Xj L(X) = H ⊗ I + iP0 ⊗ I +Xj>0 = H ⊗ I +Xj>0 Pj ⊗ Re Xj! + i P0 ⊗ I +Xj>0 Pj ⊗ Im Xj! . Note that Im L(X) (cid:23) 0. If v ∈ ker L(X), then by Lemma 2.1 and positive semidefiniteness we have v ∈ ker H ⊗ I +Xj>0 = ker(P0 ⊗ I) ∩ \j>0 Pj ⊗ Re Xj! ∩ ker P0 ⊗ I +Xj>0 ker(Pj ⊗ Im Xj)! ∩ ker H ⊗ I +Xj>0 Pj ⊗ Im Xj! Pj ⊗ Re Xj! . It is easy to see that ker(A ⊗ B) = ker A ⊗ Cn for every A ∈ Mδ(C) and B ∈ GLn(C). Since Im Xj ≻ 0, we have v ∈ ker Pj ⊗ Cn for all j, and consequently v ∈ ker H ⊗ Cn. (cid:3) Using purely stable pencils as building blocks, one can produce more stable pencils. Finally, ker H ∩Tj ker Pj = {0} implies v = 0. Definition 2.5. Let L = A0 +Pj>0 Ajxj with Aj ∈ Cδ×ε and δ ≥ ε. We temporarily say that L is S-stable if L1 DLE =  ⋆ ⋆ . . . ⋆ Lℓ   0 +Pj>0 At STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 5 for some D ∈ Cε×δ, E ∈ Cε×ε and purely stable pencils L1, . . . , Lℓ. If δ < ε, then we call L S-stable if Lt = At jxj is S-stable. Remark 2.6. The matrices D and E from Definition 2.5 necessarily have full rank and every S-stable pencil is stable by Proposition 2.4. Furthermore, if ℓ = 1 in Definition 2.5, then E is redundant: if DLE = L1 is purely stable, then ((E−1)∗D)L = (E−1)∗L1E−1 is also purely stable. Example 2.7. Let Then L is S-stable since −1(cid:19) . L =(cid:18)1 + 2x1 −x1 −x1 (cid:18)1 1 1 0(cid:19) L(cid:18)1 1 0 1(cid:19) =(cid:18)1 0 1 1(cid:19) +(cid:18)1 0 2 1(cid:19) x1. Suppose that DL is purely stable for some D ∈ GL2(C). From the R-linear system Im(DA1) = 0 in D we deduce that D =(cid:18)α1 + iβ α3 α2 + 2iβ 2α3 + α1 − iβ(cid:19) , αj, β ∈ R. Furthermore det Im(DA0) = − 1 4(α2 + α3)2, so Im(DA0) (cid:23) 0 implies α3 = −α2. Then an easy calculation shows that det Re(DA1) = − det D, so Re(DA1) (cid:23) 0 contradicts det D 6= 0. Therefore one cannot assume ℓ = 1 in Definition 2.5 in general. (cid:4) Let L = A0 +Pj>0 Ajxj be of size d and A0 ∈ GLδ(C). Then we say that L is an 0 generate Mδ(C) as a C-algebra (cf. [KV17, indecomposable pencil if A1A−1 Section 3.4] or [HKV18, Section 2.1])1. 0 , . . . , AdA−1 Lemma 2.8. Let L be an indecomposable pencil of size δ. If L is S-stable, then it has only one purely stable block; that is, DL is purely stable for some D ∈ GLδ(C). Proof. Let D, E ∈ GLδ(C) be such that (2.1) L1 DLE =  ⋆ ⋆ . . . ⋆ Lℓ ,   where L1, . . . , Lℓ are purely stable pencils. Then the coefficients of D(cid:0)LL(0)−1(cid:1)D−1 = DLE(cid:0)DL(0)E)−1 generate Mδ(C); however, they have a block lower triangular form as in (2.1), so ℓ = 1. By Remark 2.6 we thus have DL = L1 for some D ∈ GLδ(C) and a purely stable pencil L1. (cid:3) 1 Where such pencils were called irreducible. 6 J. VOL CI C 2.3. Main theorem. In this subsection we apply a truncated Gelfand-Naimark-Segal (GNS) construction to prove that every stable pencil is S-stable; see Theorem 2.10. We start with some preliminary notation. By x∗ = (x∗ 1, . . . , x∗ d) we denote the formal adjoints of variables xj and endow the free algebra C<x, x∗> with the corresponding involution. Let L = A0 +Pj Ajxj be a linear pencil of size δ × ε and δ ≥ ε. For ℓ = 0, 1, 2 let Vℓ denote the subspace of elements of degree at most ℓ in Mε(C) ⊗ C<x, x∗>. Furthermore define C1 =(Xj>0 C2 =(Xk Pj Im xj : Pj (cid:23) 0 ∀j) ⊂ V1, LkL∗ k : Lk ∈ V1) ⊂ V2, U = Cε×δL + L∗Cδ×ε ⊂ V1. The following lemma relies on a variant of an argument that was used to prove the one-sided real Nullstellensatz [CHMN13]. Lemma 2.9. Keep the notation from above. (1) C1 + C2 is a closed convex cone in V2. (2) Assume (2.2) Im(DA0) (cid:23) 0 and Im(DAj) = 0, Re(DAj) (cid:23) 0 for j > 0 =⇒ DL = 0 holds for all D ∈ Cε×δ. Then (2a) U ∩ (C1 + C2) = {0}, (2b) there exists X ∈ Hd ∩ Mε(C) such that ker L(X) 6= {0}. Proof. (1) It is clear that C1 and C2 are convex cones in V2, C1 is closed and C1 ∩C2 = {0}. Furthermore, using Caratheodory's theorem on convex hulls [Roc70, Theorem 17.1] it is easy to show that C2 is closed in V2; see e.g. [HKM12, Proposition 3.1]. Therefore C1 + C2 is closed by [Roc70, Corollary 9.1.3]. (2a) Let DL+L∗E ∈ U ∩(C1 +C2). Then DL+L∗E is hermitian and so 2(DL+L∗E) = (D + E)L + L∗(D + E)∗. Furthermore, DL + L∗E is of degree at most 1 and hence (D + E)L + L∗(D + E)∗ = P0 +Xj Pj Im xj for some Pj (cid:23) 0. Then Re((D +E)A0) (cid:23) 0 and Re((D +E)Aj) = 0, − Im((D +E)Aj) (cid:23) 0 for j > 0. For D = i(D + E) we thus have Im( DA0) (cid:23) 0 and Im( DAj) = 0, Re( DAj) (cid:23) 0 for j > 0. By the assumption (2.2) we have DL = 0, so (D + E)L = 0 and therefore DL + L∗E = 0. (2b) Let V ′ 2 ⊂ V2 be the R-subspace of hermitian elements. By (2a) and [Kle55, Theorem 2.5] there exists an R-linear functional λ0 : V ′ 2 → R satisfying λ0 ((C1 + C2) \ {0}) = R>0, λ0(U ∩ V ′ 2) = {0}. We extend λ0 to λ : V2 → C as λ(f ) = λ0(Re f ) + iλ0(Im f ). Then λ determines a scalar product hf1, f2i = λ(f ∗ 2 f1) on V1 because λ(C2 \ {0}) = R>0. Also note that λ(U) = {0}. STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 7 Let π : V1 → V0 be the orthogonal projection. Note that V0 = Mε(C). For every a, v ∈ Mε(C) and 1 ≤ j ≤ d we have hπ(axj), vi = haxj, vi = hxj, a∗vi = hπ(xj), a∗vi = haπ(xj), vi and thus (2.3) π(af ) = aπ(f ) ∀a ∈ Mε(C), f ∈ V1. For j = 1, . . . , d and a ∈ Mε(C) we define operators Yj : V0 → V0, f 7→ π(xjf ), ℓa : V0 → V0, f 7→ af. It is easy to see that λ(C1 \{0}) = R>0 implies Y = (Y1, . . . , Yg) ∈ Hd. By (2.3), operators ℓa and Yj commute. A straightforward argument shows that ℓ∗ a = ℓa∗, so ℓa also commute with Y ∗ j . Furthermore, the map Mε(C) → End(V0) ∼= Mε(C) = Mε(C) ⊗ Mε(C) given by a 7→ ℓa is a unital ∗-embedding of ∗-algebras. By a ∗-version of the Skolem- Noether theorem [Tak79, Theorem 11.9] there exists a unitary Q : V0 → Cε ⊗ Cε such that QℓaQ∗ = a ⊗ I for all a ∈ Mε(C). Since Yj and Y ∗ such that j commute with operators ℓa, there exist Xj ∈ Mε(C) QYjQ∗ = I ⊗ Xj. Since Q is unitary, we have QY ∗ j and consequently X = (X1, . . . , Xd) ∈ Hd. Let D ∈ Cε×δ be arbitrary and consider the pencil DL of size ε. By the previous j Q∗ = I ⊗ X ∗ paragraph, (DL)(X) can be viewed as an operator on V0 and (DL)(X) = ℓDA0 +Xj>0 If u ∈ V0 denotes the ε × ε identity matrix, then ℓDAj ◦ Yj. h(DL)(X)u, f i = hπ(DL), f i = hDL, f i = λ((f ∗D)L) = 0 for all f ∈ V0 by λ(U) = {0}. Hence (DL)(X)u = 0 for every D ∈ Cε×δ. Then it is easy to see that L(X)u = 0 and hence ker L(X) 6= {0}. (cid:3) Theorem 2.10. Let L be a linear pencil of size δ × ε. The following are equivalent: (1) L is stable; (2) L is S-stable; (3) L(X) has full rank for all X ∈ Hd ∩ Mmin{δ,ε}(C)d. Proof. (2) ⇒ (1) is already stated in Remark 2.6, and (1) ⇒ (3) is trivial. Hence we prove (3) ⇒ (2). Without loss of generality let δ ≥ ε. We prove the statement by induction on ε by looking at the solutions D ∈ Cε×δ of the system (2.4) Im(DA0) (cid:23) 0 and Im(DAj) = 0, Re(DAj) (cid:23) 0 for j > 0. 8 J. VOL CI C First let ε = 1. If L is not S-stable, then for every D ∈ C1×δ, DL is not purely stable. Hence every solution D of (2.4) satisfies DL = 0, so L(X) does not have full rank for some X ∈ Hd ∩ Cd by Lemma 2.9. Now assume the statement holds for all ε′ < ε and that L is not S-stable. By composing the coefficients of L on the left with the projection onto Pj ran Aj, we can without loss of generality assume that Pj ran Aj = Cd. Since L is not S-stable, DL is in particular not purely stable for any D ∈ Cε×δ, so every solution D of (2.4) satisfies K :=\j (cid:0) ker(DAj) ∩ ker(DAj)∗(cid:1) 6= {0} by Lemma 2.1. If every solution D of (2.4) satisfies DL = 0, then L(X) does not have full rank for some X ∈ Hd ∩ Mε(C)d by Lemma 2.9. Otherwise there exists a solution D of (2.4) such that DL 6= 0. Let Q be a ε × ε unitary matrix such that its columns form an orthonormal basis corresponding to the orthogonal decomposition Cε = K⊥ ⊕ K, and write D0(cid:19) , Q∗D =(cid:18)D1 By the definition of K we have LQ =(cid:0)L1 L0(cid:1) . Q∗DLQ =(cid:18)D1L1 0 0(cid:19) , 0 where L1 is a purely stable pencil. Therefore D1L0 = 0, and D0 = 0 sincePj ran Aj = Cd. If L0 were S-stable, then there would exist D2, E2 of appropriate sizes such that D2L0E2 would be a block lower triangular matrix with purely stable pencils on the diagonal. Then (cid:18)D1 D2(cid:19) LQ(I ⊕ E2) =(cid:18)D1L1 D2L1 D2L0E2(cid:19) 0 contradicts the assumption that L is not S-stable. Therefore L0 is not S-stable, so by the induction hypothesis there exists X ∈ Hd ∩ Mε(C)d such that L0(X) does not have full rank. Hence L(X) does not have full rank. (cid:3) 2.3.1. An algorithm. The proof of Theorem 2.10 can be used to devise an algorithm for testing whether a pencil is stable by solving a sequence of semidefinite programs (SDPs) [BPT13, WSV12]. Let L = A0 +Pj>0 Ajxj be of size δ × ε with δ ≥ ε. (1) Solve the following feasibility SDP for D ∈ Cε×δ: (2.5) tr Im(DA0) +Xj>0 Im(DA0) (cid:23) 0 Re(DAj) (cid:23) 0 for j > 0 Re(DAj)! = 1 Im(DAj) = 0 for j > 0. (2) If (2.5) is infeasible, then L is not stable. STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 9 (3) Otherwise let D be the output of (2.5) and let K = Tj≥0 ker(DAj). If K = {0}, then L is stable. If K 6= {0}, then let V be a matrix whose columns form a basis of K. By the proof of Theorem 2.10, L is stable if and only if LV is stable. Then we apply (1) to LV and continue. This procedure will eventually stop because LV is of smaller size than L. Similar algorithms exist for testing whether a pencil is of full rank on all hermitian tuples [KPV17] or on free spectrahedra given by monic hermitian pencils [HKMV]; the latter situation is especially interesting for the study of linear matrix inequalities [BEFB94]. However, in both preceding cases there is no clean structural analog of Theorem 2.10. 2.4. Hermitian coefficients. Classically, one is interested in symmetric or hermitian determinantal representations (1.1) of real polynomials. However, the constant term of a purely stable pencil is in general not hermitian. This can be amended for a particular for j = 0, . . . , d. class of pencils. We say that L = H0 +Pj>0 Hjxj is a hermitian pencil if Hj ∈ Hδ(C) Proposition 2.11. Let L = H0 +Pj>0 Hjxj be hermitian and D ∈ GLδ(C). (1) Assume Tj>0 ker Hj = {0}. If DL is purely stable, then DL is hermitian. (2) If L is indecomposable and stable, then L or −L is purely stable. Proof. (1) Note that DHj are hermitian and positive semidefinite for j > 0 because DL is purely stable. First we claim that the eigenvalues of D∗ are real. Let D∗v = λv for v 6= 0. Then for every j = 1, . . . , d we have (λ − ¯λ)hDHjv, vi = λhDHjv, vi − ¯λhHjD∗v, vi = hλHjv, D∗vi − hHjD∗v, λvi = 0. Suppose hDHjv, vi = 0 for all j > 0. Since DHj (cid:23) 0, we have DHjv = 0, so v = 0 by the assumption, contradicting v 6= 0. Therefore hDHjv, vi 6= 0 for some j > 0 and hence λ = ¯λ. Next we show that D∗ is diagonalizable. Let (D∗ − λI)2v = 0. Then h(DHj)(D∗ −λI)v, (D∗ −λI)vi = h(D −λI)DHj(D∗ −λI)v, vi = hDHj(D∗ −λI)2v, vi = 0 for all j > 0. Since DHj (cid:23) 0, it follows that (DHj)(D∗ − λI)v = 0 for all j > 0, so (D∗ − λI)v = 0. Therefore D∗ is diagonalizable. If D∗v = λv, then (2.6) 2ihIm(DH0)v, vi = hH0v, D∗vi − hH0D∗v, vi = 0 because λ ∈ R. Now if v1 and v2 are eigenvectors for D∗, then hIm(DH0)(v1 ± v2), (v1 ± v2)i ≥ 0 since Im(DH0) (cid:23) 0, which together with (2.6) implies (2.7) hIm(DH0)v1, v2i + hIm(DH0)v2, v1i = 0. Now let v ∈ Cd be arbitrary. Because D is diagonalizable, v can be written as a sum of eigenvectors of D. Therefore hIm(DH0)v, vi = 0 by (2.7), and so Im(DH0) = 0. 10 J. VOL CI C (2) Since L is indecomposable and stable, there exists D such that DL is purely stable by Theorem 2.10 and Lemma 2.8. Hence DL is hermitian by (1), i.e., DHj = HjD∗ for all j ≥ 0. Therefore HjH −1 0 D = HjD∗H −1 0 = DHjH −1 0 for all j > 0. Since H1H −1 matrix, so L or −L is purely stable. 0 , . . . , HgH −1 0 generate Mδ(C), it follows that D is a real scalar (cid:3) 2.5. Hurwitz and Schur stability. In control theory, there are also other stability notions, such as Hurwitz and Schur stability, that can be related to Definition 2.2. In this subsection we describe how to apply Theorem 2.10 and the algorithm from Subsection 2.3.1 to test other versions of noncommutative stability. We say that L is Hurwitz stable if L(X) has full rank for every X with Re(Xj) ≻ 0 for j = 1, . . . , d. Then L is Hurwitz stable if and only if L(−ix) is stable. Therefore one can directly derive the analogs of Theorem 2.10 and Subsection 2.3.1 for Hurwitz stable pencils. Let k · k denote the spectral norm of matrices, and let Dd = [n∈N(cid:8)(X1, . . . , Xd) ∈ Mn(C)d : kXjk < 1 ∀j(cid:9) be the noncommutative polydisk. We say that L is Schur stable if L(X) has full rank for every X ∈ Dd. Using the Cayley transform we see that L is Schur stable if and only if (2.8) has full rank for all X ∈ Hd. However, (2.8) is not a linear matrix pencil anymore. Let L(cid:0)(X1 − iI)(X1 + iI)−1, . . . , (Xg − iI)(Xg + iI)−1(cid:1) L = A0 +Pj Ajxj be of size δ × ε with δ ≥ ε and consider the pencil   −I ... I(xg + i) −I · · · Ag(xg − i) A0 L =  A1(x1 − i) I(x1 + i) . . . of size (dε + δ) × (dε + ε). Using Schur complements it is easy to check that (2.8) is invertible if and only if L(X) is invertible. If A0 does not have full rank, then L is not Schur stable. Now let A0 have full rank; i.e., after a left and a right basis change we can I ). Let L1 denote the Schur complement of L with respect to the ε × ε assume A0 = ( 0 block I in A0. Then L1 is a linear matrix pencil of size (dε + δ − ε) × (dε), and L1(X) is invertible if and only if (2.8) is invertible. Therefore L is Schur stable if and only if L1 is stable. In particular, we can test the Schur stability with a sequence of SDPs as in Subsection 2.3.1. Moreover, Theorem 2.10 implies the following size bound. Corollary 2.12. Let L be a pencil of size δ × ε. If L(X) has full rank for every X ∈ Dd of size d · min{δ, ε}, then L(X) has full rank for every X ∈ Dd. Remark 2.13. Via realization theory (see Subsection 3.1 below), Schur stable pencils are closely related to noncommutative rational functions that are regular on the noncommu- tative polydisk. A particularly interesting subset of such functions is the noncommutative STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 11 Schur-Agler class. One of its characteristic features is the existence of contractive repre- sentations; see [BMV18]. 3. Stability of noncommutative polynomials We are now ready to apply the preceding results to noncommutative polynomials and rational functions. First we characterize noncommutative rational functions whose do- mains contain the matricial positive orthant Hd (Theorem 3.2). Next we show that every stable noncommutative polynomial admits a determinantal representation with a purely stable pencil (Theorem 3.7). Finally, we somewhat surprisingly prove that every irre- ducible hermitian stable polynomial is affine (Theorem 3.9). 3.1. Noncommutative rational functions. After a short introduction of the free skew field and required realization theory, we describe noncommutative rational functions de- fined on the matricial positive orthant. 3.1.1. Free skew field. We give a condensed introduction of noncommutative rational func- tions using matrix evaluations of formal rational expressions following [K-VV12]. Origi- nally they were defined ring-theoretically [Ami66, Coh95]. Noncommutative rational expressions are syntactically valid combinations of complex numbers, variables x, arithmetic operations +, ·, −1 and parentheses (, ). Given a noncom- mutative rational expression r and X ∈ Mn(C)d, the evaluation r(X) ∈ Mn(C) is defined in the obvious way if all inverses appearing in r exist at X. The set of all X ∈ Md such that r is defined at X is is called the domain of r and denoted dom r. On the set of all expressions with nonempty domains we define an equivalence relation r1 ∼ r2 if and only if r1(X) = r2(X) for all X ∈ dom r1 ∩ dom r2. The equivalence classes with respect to this relation are called noncommutative rational functions. By [K-VV12, Proposition 2.1] they form a skew field denoted C (<x )>, which is the universal skew field of fractions of C<x> by [Coh95, Section 4.5]. We define the domain of a noncommutative rational function ∈ C (<x )> as the union of dom r over all representatives r of . 3.1.2. Realization theory. Let ∈ C (<x )> and assume that is regular at the origin, i.e., 0 ∈ dom . Then there exist δ ∈ N, b, c ∈ Cδ and a linear pencil L of size δ with L(0) = I, such that (3.1) = c∗L−1b. We say that (3.1) is a (descriptor) realization of of size δ; see [BGM05, Section 12] and [HMV06, Vol18]. In automata theory, such realizations are also called linear representations [BR11]. Remark 3.1. In general, admits various realizations. Those of the smallest size are called minimal, and possess distinguished properties that we now outline. Let c∗L−1b with L = I −Pj Ajxj be a minimal realization of of size δ. (1) It is easy to see that c∗v = 0 and Ajv = 0 ∀j =⇒ v = 0 and v∗b = 0 and v∗Aj = 0 ∀j =⇒ v = 0 12 J. VOL CI C holds for every v ∈ Cδ. This observation is also a consequence of a stronger result stating that minimal realizations are controllable and observable [BGM05, Theorem 9.1] (cf. reduced [BR11, Proposition 2.1]). (2) ∈ C<x> if and only if A1, . . . , Ad are jointly nilpotent by [CR99, Proposition 2.1]. (3) By [K-VV09, Theorem 3.1] and [Vol17, Theorem 3.10] we have (4) Assume that is hermitian, i.e., (X)∗ = (X ∗) for all X ∈ dom . Then admits dom =(cid:8)X ∈ Md : det L(X) 6= 0(cid:9). a minimal realization that is hermitian, = c∗ H0 +Xj Hjxj!−1 HjH −1 0 xj!−1 c, c = (H −1 0 c)∗ I +Xj where Hj ∈ Hδ(C); see [HMV06, Lemma 4.1] or [Vol18, Theorem 6.8]. 3.1.3. Rational functions on the matricial positive orthant. We can now apply Theorem 2.10 to noncommutative rational functions via realization theory. Theorem 3.2. Let ∈ C (<x )>. Then dom ⊃ Hd if and only if = c∗L−1b for some S-stable pencil L. Proof. (⇐) Let = c∗L−1b for a stable pencil L. The matrix L(β) is invertible for every β in the positive orthant of Cd by stability of L, so the complex polynomial det L(z) in d commuting variables is nonzero. Hence there exists α ∈ Rd such that det L(α) 6= 0. Then 0 ∈ dom (x + α) and (x + α) = c∗(cid:0)L(x + α)(cid:1)−1b = c∗(cid:0)I + L(α)−1(L − I)(cid:1)−1L(α)−1b is a realization of (x + α). By Remark 3.1(3) and stability of L we have and consequently dom ⊃ Hd since α ∈ Rd. dom (x + α) =(cid:8)X ∈ Md : det L(X + αI) 6= 0(cid:9) ⊇ Hd, (⇒) Let dom ⊃ Hd. Since dom ∩ Cd is a nonempty Zariski open set and Rd is a Zariski dense set in Cd, there exists α ∈ dom ∩ Rd. Note that (x + α) again satisfies If c∗L−1b is a minimal realization of (x + α), then L is stable dom (x + α) ⊃ Hd. by Remark 3.1(3). Hence L(x − α) is stable and thus S-stable by Theorem 2.10 and = c∗L(x − α)−1b. (cid:3) For later use we record two well-known determinantal identities. Lemma 3.3. Let P ∈ GLδ(C) and u, v ∈ Cδ×ε. Then det(P + uv∗) = det(I + v∗P −1u) det P, det(cid:18)0 u∗ u P(cid:19) = det(−u∗P −1u) det P. Let ∈ C (<x )> and 0 ∈ dom . We say that is indecomposable [KV17, Section 4.2] if the pencil appearing in its minimal realization is indecomposable. We record the following property of hermitian indecomposable functions. Proposition 3.4. Let ∈ C (<x )> be hermitian and indecomposable. If dom ⊃ Hd, then dom −1 ⊃ Hd. STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 13 Proof. By the assumption and Remark 3.1(4), admits a minimal realization c∗L−1c with L hermitian and indecomposable. Since dom ⊃ Hd, the proof of (⇒) in Theorem 3.2 shows that L is stable, so L or −L is purely stable by Proposition 2.11. By Lemma 3.3 we have (3.2) det(cid:0)−(c∗L−1c)(X)(cid:1) det L(X) = det(cid:18)(cid:18)0 c∗ c L(cid:19) (X)(cid:19) . Note that L and ( 0 c∗ X ∈ Hd. Therefore dom c L ) or their negatives are purely stable pencils, so det (X) 6= 0 for all (cid:3) −1 ⊃ Hd. Remark 3.5. The conclusion of Proposition 3.4(1) fails in general if is not hermitian and indecomposable; for example, consider = x1 − i and = 1 + x2 1. 3.2. Stable noncommutative polynomials. We say that f ∈ C<x> is stable if f (X) is invertible for every f ∈ Hd. That is, f is stable if and only if dom f −1 ⊃ Hd. In this sub- section we prove that every stable noncommutative polynomial admits a determinantal representation with a purely stable pencil, see Theorem 3.7. Then we turn our atten- tion to hermitian stable polynomials, which are noncommutative analogs of real stable polynomials. Quite contrary to the commutative setting, we show that every irreducible hermitian stable polynomial is affine (Theorem 3.9). Here f ∈ C<x> is irreducible if it cannot be written as f = f1f2 for some f1, f2 ∈ C<x> \C. The following lemma is a descriptor realization analog of [HKV18, Lemma 5.3] (which deals with Fornasini-Marchesini realizations). Lemma 3.6. Let f ∈ C<x> and f (0) = 1. If c∗L−1b is a minimal realization of f −1 with L = I −Pj Ajxj of size δ, then (1) f admits a minimal realization (3.3) (cid:0)−c∗ 1(cid:1) I −Pj Aj(I − bc∗)xj −Pj(Ajb)xj 0 1 of size δ + 1, !−1 1(cid:19) (cid:18)0 (2) Pj ran Aj = Cδ and Tj ker Aj = {0}, (3) det f (X) = det L(X) for all X ∈ Md, (4) L is indecomposable if f is irreducible. −1 of size δ + 1, see e.g. of of size δ, then (3.3) is a realization of 3.10]. In particular, if admits a minimal realization of size δ, then Proof. Let ∈ C (<x )> satisfy 0 ∈ dom and (0) = 1. If c∗(I −Pj Ajxj)−1b is a realization realization of size at least δ − 1. Now assume ∈ C<x> and let c∗(I −Pj Ajxj)−1b be Tj ker Aj 6= {0}, and by Remark 3.1(1) there exists v2 ∈Tj ker Aj such that c∗v2 = 1. Let its minimal realization. Then A1, . . . , Ad are jointly nilpotent matrices by Remark 3.1(3). Hence there exists v1 6= 0 such that v∗ 1Aj = 0 for all j. Moreover, joint nilpotency implies V ∈ C(δ+1)×(δ−1) be a matrix whose columns form an orthonormal basis of the orthogonal complement of {( v1 1 )} in Cδ+1. Combining [Vol17, Theorem −1 admits a minimal 0 ), ( v2 1 0(cid:1)(cid:18)0 (cid:0)v∗ 1(cid:19) = 0, 1 0(cid:1)(cid:18)Aj(I − bc∗) Ajb (cid:0)v∗ 0 (cid:19) = 0 0 14 and J. VOL CI C with (3.3) it is easy to see that (cid:0)−c∗ 1(cid:1)(cid:18)v2 1(cid:19) = 0, (cid:18)Aj(I − bc∗) Ajb 0 (cid:19)(cid:18)v2 (cid:0)(cid:0)−c∗ 1(cid:1) V(cid:1) V ∗ I −Pj Aj(I − bc∗)xj −Pj(Ajb)xj −1 admits a realization 1(cid:19) = 0 ! V!−1 1 0 0 (cid:18)V ∗(cid:18)0 1(cid:19)(cid:19) of size δ − 1. (1) Now fix f ∈ C<x> with f (0) = 1, and let c∗L−1b be a minimal realization of f −1 with L = I −Pj Ajxj. If f admitted a minimal realization of size at most δ, then f −1 would admit a minimal realization of size at most δ − 1 by the previous paragraph, which contradicts the assumption on size of c∗L−1b. Therefore f admits a minimal realization of size δ + 1 of the form (3.3). (2) We have just seen that (3.3) is a minimal realization of f . If v ∈ Cd is such that v∗Aj = 0 for all j, then (cid:0)v∗ 0(cid:1)(cid:18)0 1(cid:19) = 0, (cid:0)v∗ 0(cid:1)(cid:18)Aj(I − bc∗) Ajb 0 (cid:19) = 0, 0 so Remark 3.1(1) implies v = 0, and hence Pj ran Aj = Cδ. Similarly we obtain Tj ker Aj = {0}. (3) Matrices Aj(I −bc∗), which by (3.3) appear in a minimal realization of f , are jointly nilpotent by Remark 3.1(2). Next, c∗b = f (0)−1 = 1 implies f −1 = c∗L−1b = 1 + c∗L−1(I − L)b. By Lemma 3.3 we then have for every X ∈ dom f −1. But det L(X) · det f (X)−1 = det(cid:0)(L + (I − L)bc∗)(X)(cid:1) det(cid:0)(L + (I − L)bc∗)(X)(cid:1) = det I ⊗ I −Xj Aj(I − bc∗) ⊗ Xj! = 1 since Aj(I − bc∗) are jointly nilpotent, so det f (X) = det L(X) for all X ∈ Md. (4) Assume f is irreducible. For X (n) ∈ Mn(C)g one can view det f (X (n)) as a poly- nomial in gn2 variables, By [HKV18, Theorem 4.3], there exists n0 ∈ N such that det L(X (n0)) = det f (X (n0)) is an irreducible polynomial for all n ≥ n0. Using the results of [HKV18, Subsection 2.1] it is easy to see that L is indecomposable or Pj ran Aj 6= Cδ or Tj ker Aj 6= {0}. Therefore L is indecomposable by (2). Theorem 3.7. Let f ∈ C<x>. Then f is stable if and only if there exists a purely stable pencil L such that det f (X) = det L(X) for all X ∈ Md. (cid:3) Proof. The implication (⇐) trivially holds, so we consider (⇒). Since purely stable pen- cils are preserved under shifts along Rd and direct sums, it suffices to assume that f is irreducible and f (0) 6= 0. Let f −1 = c∗ L−1b be a minimal realization. The monic pencil L is indecomposable and det f (X) = det(cid:0)f (0) L(X)(cid:1) STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 15 If f is stable, then L is S-stable by Theorem for all X ∈ Md by Lemma 3.6(3)&(4). 3.2. By Lemma 2.8 there exists an invertible D such that f (0)D L is purely stable. If det D ∈ R>0, then take L := (det D)−1f (0)D L. Otherwise there is γ ∈ R + iR>0 such that det D = γ2. Then L := (γ−1I2) ⊕ (f (0)D L) is the desired purely stable pencil. (cid:3) Example 3.8. If β ∈ R, then a short calculation shows that (3.4) Im(cid:0)(T − βI)−1(cid:1) = −(T − βI)−1(Im T )(T ∗ − βI)−1 for all the matrices T with T − βI invertible. Now let r ∈ R(t) be an arbitrary univariate rational function of the form r = ℓ Xk=1 αk t − βk , αk ∈ R<0, βk ∈ R. If Im T ≻ 0, then (T − βkI) is invertible for all k, and thus (3.5) Im T ≻ 0 ⇒ Im(r(T )) ≻ 0 by (3.4). Write r = p/q for coprime p, q ∈ R[t] and let f = p(x1) + q(x1)x2 ∈ C<x>. Then f (X) = p(X1) + q(X1)X2 = q(X1)(cid:0)q(X1)−1p(X1) + X2(cid:1) is invertible for every X ∈ H2 because Im(q(X1)−1p(X1) + X2) ≻ 0 by (3.5). Therefore f is stable and irreducible. (cid:4) 3.2.1. Stable hermitian polynomials. As for noncommutative rational functions, we say that f ∈ C<x> hermitian if f (X)∗ = f (X ∗) for all X ∈ Md. Recall that there exist irreducible stable polynomials of arbitrary degree (Example 3.8). On the other hand, this is not true for hermitian polynomials. Theorem 3.9. Let f ∈ C<x> be hermitian and irreducible, and f (0) = 1. Then f is stable if and only if f = 1 ± (α1x1 + · · · + αdxd) for αj ∈ R≥0. Proof. Since (⇐) is clear, let us prove (⇒). By Remark 3.1(4), f −1 admits a hermitian minimal realization c∗L−1c of size δ with L = H0 +Pj>0 Hjxj. Note that c∗L−1c = (H −1 0 c)∗(LH −1 0 )−1c. Then L is indecomposable by Lemma 3.6(4). Furthermore L is stable because f is stable. Moreover, since L is hermitian, L or −L is purely stable by Proposition 2.11(2). By Lemma 3.6(1), (3.6) 0 c)∗ 1(cid:1) I +Pj HjH −1 (cid:0)−(H −1 is a minimal realization for f . Denote 0 (cid:0)I − c(H −1 0 c)∗(cid:1) xj Pj(HjH −1 0 1 0 c)xj !−1 1(cid:19) (cid:18)0 T = H −1 0 − (H −1 0 c)(H −1 0 c)∗. Therefore HjH −1 Observe that c∗H −1 of Cδ such that c is a multiple of e1 we have 0 (cid:0)I − c(H −1 0 c = f (0)−1 = 1 implies T c = 0. With respect to an orthonormal basis 0 c)∗(cid:1) = HjT are jointly nilpotent matrices by Remark 3.1(2). Hj =(cid:18)αj u∗ uj Pj(cid:19) , T =(cid:18)0 0 0 S(cid:19) j 16 J. VOL CI C for Pj, S ∈ Hδ−1(C), uj ∈ Cδ−1 and αj ∈ R. Since L or −L is purely stable, we can without loss of generality assume that Hj (cid:23) 0 for all j. Therefore Pj = VjV ∗ j , uj = Vjvj for some Vj ∈ Mδ−1(C) and vj ∈ Cδ−1. Furthermore, P1S, . . . , PgS are jointly nilpotent (δ − 1) × (δ − 1) matrices. Choose 1 ≤ j, k ≤ d and denote M = V ∗ j SVk. Then M ∗(MM ∗)δ−1 = V ∗ k S(PjSPkS)δ−1Vj = 0 and therefore M = 0. Consequently (3.7) HjT Hk = 0 ∀j, k. Combining (3.6) and (3.7) yields !−1 1 1 0 0 0 c)xj 1(cid:19) (cid:18)0 = 1 + c∗H −1 f =(cid:0)−(H −1 =(cid:0)−(H −1 0 c)∗ 1(cid:1) I +Pj HjT xj Pj(HjH −1 0 c)∗ 1(cid:1) I −Pj HjT xj −(cid:16)I −Pj HjT xj(cid:17)(cid:16)Pj Hjxj(cid:17) H −1 0 I −Xj 0 Xj = 1 +Xj (cid:0)c∗H −1 Hjxj! H −1 0 c(cid:1) xj. HjT xj! Xj Hjxj!!H −1 0 c = 1 + c∗H −1 0 HjH −1 0 c 0 c !(cid:18)0 1(cid:19) Finally, since L or −L is purely stable, the real numbers c∗H −1 sign. 0 HjH −1 0 c have the same (cid:3) 4. Applications Theoretical results of previous sections can be applied to multidimensional circuits and systems [Bos03, Bos17]. In engineering, one seeks a controlled system output, and is thus interested in stable systems. Given a d-dimensional linear time-invariant system, its sta- bility is related to the zero locus of the denominator f ∈ C[z1, . . . , zd] of its characteristic or transfer function. In the continuous case, stability corresponds to f not having zeros in the open poly-right-halfplane (f is a Hurwitz polynomial), while in the discrete case stability relates to f not having zeros in the open polydisk (f is a Schur polynomial). For example, consider the discrete Roesser state-space model   (4.1) xd(i1, . . . , id + 1) x1(i1 + 1, . . . , id) x1(i1, . . . , id) ...   = A  y1(i1, . . . , id) = C  ... ... xd(i1, . . . , id) x1(i1, . . . , id) xd(i1, . . . , id) + Bu(i1, . . . , id), + Du(i1, . . . , id)     STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 17 as in [Bas91], where xj, u, y are the state, input and output vectors, respectively, and A, B, C, D are constant matrices of appropriate sizes. Then the denominator of the trans- fer function for (4.1) equals (4.2) det I − A Mj Iδj zj!! , where δj is the dimension of xj. Hence one would like to test whether (4.2) is Schur. While there are certain procedures for checking the Schur or the Hurwitz property and their variations [FB87, RR89], they are computationally challenging since determining whether a polynomial has a zero in an open domain in Cd is a hard problem. Hence we propose the following relaxation. Returning to the model (4.1), one can first ask if the pencil (4.3) I − A Mj Iδj xj! is Schur stable as in Subsection 2.5. This can be done efficiently using the algorithm from Subsection 2.3.1. If (4.3) is a Schur stable pencil, then (4.2) is a Schur polynomial. While the converse fails in general, this relaxation is reasonable when it can be hypothesized that the stability of (4.1) strongly depends on a specific structure of the matrix A. Namely, the algorithm from Subsection 2.3.1 affirms or dismisses this hypothesis. Such relaxations are also applicable to other models of multidimensional linear systems S, whose stability can be translated into the Hurwitz/Schur property of the determinant of the pencil arising from matrices in S. Furthermore, the development of the noncommu- tative linear systems theory [BGM05, BGM06, BGM06'] might provide even more direct applications of the results in this paper. References [Agl90] J. Agler: On the representation of certain holomorphic functions defined on a polydisc, Topics in operator theory: Ernst D. Hellinger memorial volume, 47 -- 66, Oper. Theory Adv. Appl. 48, Birkhauser, Basel, 1990. 1 [Ami66] S.A. Amitsur: Rational identities and applications to algebra and geometry, J. Algebra 3 (1966) 304 -- 359. 11 [BGM05] J. A. Ball, G. Groenewald, T. Malakorn: Structured noncommutative multidimensional linear systems, SIAM J. Control Optim. 44 (2005) 1474 -- 1528. 2, 11, 12, 17 [BGM06] J. A. Ball, G. Groenewald, T. Malakorn: Conservative structured noncommutative multidi- mensional linear systems, The state space method generalizations and applications, 179 -- 223, Oper. Theory Adv. Appl. 161, Linear Oper. Linear Syst., Birkhauser, Basel, 2006. 17 [BGM06'] J. A. Ball, G. Groenewald, T. Malakorn: Bounded real lemma for structured noncommutative multidimensional linear systems and robust control, Multidimens. Syst. Signal Process. 17 (2006) 119 -- 150. 17 [BMV18] J. A. Ball, G. Marx, V. Vinnikov: Interpolation and transfer-function realization for the non- commutative Schur -- Agler class, in: Operator Theory in Different Settings and Related Applica- tions, 23 -- 116, Birkhauser, Cham, 2018. 2, 11 [Bas91] S. Basu: New results on stable multidimensional polynomials. III. State-space interpretations, IEEE Trans. Circuits Syst. 38 (1991) 755 -- 768. 17 [BR11] J. Berstel, C. Reutenauer: Noncommutative rational series with applications, Encyclopedia of Mathematics and its Applications 137, Cambridge University Press, Cambridge, 2011. 2, 11, 12 18 J. VOL CI C [BPT13] G. Blekherman, P. A. Parrilo, R. R. Thomas (eds.): Semidefinite optimization and convex algebraic geometry, MOS-SIAM Ser. Optim. 13, SIAM, Philadelphia, PA, 2013. 1, 2, 8 [BCR98] J. Bochnak, M. Coste, M.F. Roy: Real algebraic geometry, Results in Mathematics and Related Areas (3) 36, Springer-Verlag, Berlin, 1998. 1 [BB08] J. Borcea, P. Brand´en: Applications of stable polynomials to mixed determinants: Johnson's conjectures, unimodality, and symmetrized Fischer products, Duke Math. J. 143 (2008) 205 -- 223. 1 [BB09] J. Borcea, P. Brand´en: The Lee-Yang and P´olya-Schur programs. II. Theory of stable polynomials and applications, Comm. Pure Appl. Math. 62 (2009) 1595 -- 1631. 1 [Bos88] N. K. Bose: Robust multivariate scattering Hurwitz interval polynomials, Linear Algebra Appl. 98 (1988) 123 -- 136. 1 [Bos03] N. K. Bose: Multidimensional systems theory and applications, second edition, with contributions by B. Buchberger and J. P. Guiver, Kluwer Academic Publishers, Dordrecht, 2003. 16 [Bos17] N. K. Bose: Applied multidimensional systems theory, second edition, 2ith a preface by W. K. Jenkins, C. Lagoa and U. Srinivas, Springer, Cham, 2017. 16 [BEFB94] S. Boyd, L. El Ghaoui, E. Feron, V. Balakrishnan: Linear matrix inequalities in system and control theory, SIAM Studies in Applied Mathematics 15, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1994. 9 [Bra11] P. Brand´en: Obstructions to determinantal representability, Adv. Math. 226 (2011) 1202 -- 1212. 1 [CHMN13] J. Cimpric, J. W. Helton, S. McCullough, C. Nelson: A noncommutative real nullstellensatz corresponds to a noncommutative real ideal: algorithms, Proc. Lond. Math. Soc. 106 (2013) 1060 -- 1086. 6 [Coh95] P. M. Cohn: Skew fields. Theory of general division rings, Encyclopedia of Mathematics and its Applications 57, Cambridge University Press, Cambridge, 1995. 11 [CR99] P. M. Cohn, C. Reutenauer: On the construction of the free field, Internat. J. Algebra Comput. 9 (1999) 307 -- 323. 12 [FB87] A. Fettweis, S. Basu: New results on stable multidimensional polynomials. I. Continuous case, IEEE Trans. Circuits Syst. 34 (1987) 1221 -- 1232. 1, 17 [GK-VVW16] A. Grinshpan, D. S. Kaliuzhnyi-Verbovetskyi, V. Vinnikov, H. J. Woerdeman: Contractive determinantal representations of stable polynomials on a matrix polyball, Math. Z. 283 (2016) 25 -- 37. 1 [Gur08] L. Gurvits: Van der Waerden/Schrijver-Valiant like conjectures and stable (aka hyperbolic) ho- mogeneous polynomials: one theorem for all, Electron. J. Combin. 15 (2008) RP 66. 1 [Hel02] J. W. Helton: "Positive" noncommutative polynomials are sums of squares, Ann. of Math. (2) 156 (2002) 675 -- 694. 2 [HKM12] J. W. Helton, I. Klep, S. McCullough: The convex Positivstellensatz in a free algebra, Adv. Math. 231 (2012) 516 -- 534. 2, 6 [HKV18] J. W. Helton, I. Klep, J. Volcic: Geometry of free loci and factorization of noncommutative polynomials, Adv. Math. 331 (2018) 589 -- 626. 5, 13, 14 [HKMV] J. W. Helton, I. Klep, S. McCullough, J. Volcic: Noncommutative polynomials describing convex sets, preprint arXiv:1808.06669. 9 [HMV06] J. W. Helton, S. McCullough, V. Vinnikov: Noncommutative convexity arises from linear matrix inequalities, J. Funct. Anal. 240 (2006) 105 -- 191. 2, 11, 12 [HV07] J. W. Helton, V. Vinnikov: Linear matrix inequality representation of sets, Comm. Pure Appl. Math. 60 (2007) 654 -- 674. 1 [JT18] T. Jorgens, T. Theobald: Hyperbolicity cones and imaginary projections, Proc. Amer. Math. Soc. (2018). 1 [JT18] T. Jorgens, T. Theobald: Conic stability of polynomials Res. Math. Sci. 5 (2018) 5 -- 26. 4 [K-VV09] D. S. Kaliuzhnyi-Verbovetskyi, V. Vinnikov: Singularities of rational functions and minimal factorizations: the noncommutative and the commutative setting, Linear Algebra Appl. 430 (2009) 869 -- 889. 12 STABLE NC POLYNOMIALS AND DETERMINANTAL REPRESENTATIONS 19 [K-VV12] D. S. Kaliuzhnyi-Verbovetskyi, V. Vinnikov: Noncommutative rational their difference-differential calculus and realizations, Multidimens. Syst. Signal Process. 23 (2012) 49 -- 77. 11 functions, [KT-M99] V. L. Kharitonov, J. A. Torres-Munoz: Robust stability of multivariate polynomials. I. Small coefficient perturbations, Multidimens. Systems Signal Process. 10 (1999) 7 -- 20. 1 [Kle55] V. L. Klee: Separation properties of convex cones, Proc. Amer. Math. Soc. 6 (1955) 313 -- 318. 6 [KPV15] M. Kummer, D. Plaumann, C. Vinzant: Hyperbolic polynomials, interlacers, and sums of squares, Math. Program. 153 (2015) 223 -- 245. 1 [Kum89] A. Kummert: Synthesis of two-dimensional lossless m-ports with prescribed scattering matrix, Circuits Systems Signal Process. 8 (1989) 97 -- 119. 1 [KPV17] I. Klep, J. E. Pascoe, J. Volcic: Regular and positive noncommutative rational functions, J. Lond. Math. Soc. 95 (2017) 613 -- 632. 2, 9 [KV17] I. Klep, J. Volcic: Free loci of matrix pencils and domains of noncommutative rational functions, Comment. Math. Helv. 92 (2017) 105 -- 130. 5, 12 [Kne11] G. Knese: Rational inner functions in the Schur-Agler class of the polydisk, Publ. Mat. 55 (2011) 343 -- 357. 1 [MSS15] A. W. Marcus, D. A. Spielman, N. Srivastava: Interlacing families II: Mixed characteristic polynomials and the Kadison-Singer problem, Ann. of Math. 182 (2015) 327 -- 350. 1 [NT12] T. Netzer, A. Thom: Polynomials with and without determinantal representations, Linear Algebra Appl. 437 (2012) 1579 -- 1595. 1 [RR89] P.K. Rajan, H.C. Reddy: A Test Procedure for 2-D Discrete Scattering Hurwitz Polynomials, IEEE Trans. Acoust., Speech, Signal Process. 37 (1989) 118 -- 120. 17 [Roc70] R. T. Rockafellar: Convex analysis, Princeton Mathematical Series 28, Princeton University Press, Princeton, 1970. 6 [Tak79] M. Takesaki: Theory of operator algebras. I, Springer-Verlag, New York-Heidelberg, 1979. 7 [vdG08] G. van der Geer: Siegel modular forms and their applications, in: The 1-2-3 of modular forms, Universitext, 181 -- 245, Springer, Berlin, 2008. 4 [Vol17] J. Volcic: On domains of noncommutative rational functions, Linear Algebra Appl. 516 (2017) 69 -- 81. 12, 13 [Vol18] J. Volcic: Matrix coefficient realization theory of noncommutative rational functions, J. Algebra 499 (2018) 397 -- 437. 11, 12 [Wag11] D. G. Wagner: Multivariate stable polynomials: theory and applications, Bull. Amer. Math. Soc. (N.S.) 48 (2011) 53 -- 84. 1 [WSV12] H. Wolkowicz, R. Saigal, L. Vandenberghe (editors): Handbook of semidefinite programming: theory, algorithms, and applications, vol. 27, Springer Science & Business Media, 2012. 8 Jurij Volcic, Department of Mathematics, Texas A&M University, Texas E-mail address: [email protected] 20 J. VOL CI C NOT FOR PUBLICATION Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Stable pencils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1. Linear matrix pencils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.2. Real and imaginary part of a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Stable pencils. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. Main theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1. An algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4. Hermitian coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5. Hurwitz and Schur stability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. Stability of noncommutative polynomials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Noncommutative rational functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1. Free skew field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.2. Realization theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.3. Rational functions on the matricial positive orthant. . . . . . . . . . . . . . 3.2. Stable noncommutative polynomials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1. Stable hermitian polynomials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 3 3 3 3 3 4 6 8 9 10 11 11 11 11 12 13 15 16 17
1009.4892	2	1009	2010-10-17T15:56:46	Simplicity and maximal commutative subalgebras of twisted generalized Weyl algebras	[ "math.RA" ]	In this paper we show that each non-zero ideal of a twisted generalized Weyl algebra (TGWA) $A$ intersects the centralizer of the distinguished subalgebra $R$ in $A$ non-trivially. We also provide a necessary and sufficient condition for the centralizer of $R$ in $A$ to be commutative, and give examples of TGWAs associated to symmetric Cartan matrices satisfying this condition. By imposing a certain finiteness condition on $R$ (weaker than Noetherianity) we are able to make an Ore localization which turns out to be useful when investigating simplicity of the TGWA. Under this mild assumption we obtain necessary and sufficient conditions for the simplicity of TGWAs. We describe how this is related to maximal commutativity of $R$ in $A$ and the (non-) existence of non-trivial $\Z^n$-invariant ideals of $R$. Our result is a generalization of the rank one case, obtained by D. A. Jordan in 1993. We illustrate our theorems by considering some special classes of TGWAs and providing concrete examples.	math.RA	math	Simplicity and maximal commutative subalgebras of twisted generalized Weyl algebras Jonas T. Hartwig∗† Johan Öinert‡§ In this paper we show that each non-zero ideal of a twisted generalized Weyl algebra (TGWA) A intersects the centralizer of the distinguished subalgebra R in A non-trivially. We also provide a necessary and sufficient condition for the centralizer of R in A to be commutative, and give examples of TGWAs as- sociated to symmetric Cartan matrices satisfying this condition. By imposing a certain finiteness condition on R (weaker than Noetherianity) we are able to make an Ore localization which turns out to be useful when investigating simplicity of the TGWA. Under this mild assumption we obtain necessary and sufficient conditions for the simplicity of TGWAs. We describe how this is related to maximal commutativity of R in A and the (non-) existence of non-trivial Zn-invariant ideals of R. Our result is a generalization of the rank one case, obtained by D. A. Jordan in 1993. We illustrate our theorems by considering some special classes of TGWAs and providing concrete examples. Contents 1 Introduction 2 ∗E-mail: [email protected] †Partially supported by the Netherlands Organization for Scientific Research in the VIDI-project "Sym- metry and modularity in exactly solvable models" and by postdoctoral grants from the Swedish Research Council and from the São Paulo Research Foundation FAPESP (2008/10688-1). ‡Address: Department of Mathematical Sciences, University of Copenhagen, Universitetsparken 5, DK-2100 Copenhagen Ø, Denmark, E-mail: [email protected] §Partially supported by The Swedish Research Council, The Swedish Foundation for International Cooperation in Research and Higher Education (STINT), The Crafoord Foundation, The Royal Physiographic Society in Lund, The Swedish Royal Academy of Sciences and "LieGrits", a Marie Curie Research Training Network funded by the European Community as project MRTN-CT 2003- 505078. 1 1 Introduction 2 Preliminaries 2.1 Notation and definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Bilinear forms and radicals associated to group graded algebras . . . . . . 2.2.1 The gradation form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 The gradical 2.3 The TGWC and TGWA . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Essential subalgebras of TGWAs 4 Characterization of regularly graded TGWAs 5 5 6 6 8 9 10 12 5 Centralizers of R in TGWAs 13 5.1 Results for general TGWAs . . . . . . . . . . . . . . . . . . . . . . . . . . 13 5.2 Maximal commutative subalgebras of TGWAs associated to Cartan matrices 16 6 Ore localizations and a finiteness condition for TGWAs 18 7 Simplicity theorems for TGWAs 21 7.1 A weak simplicity result for general TGWAs . . . . . . . . . . . . . . . . . 21 7.2 On the structure of the localized algebra X −1A . . . . . . . . . . . . . . . 22 7.3 Simplicity of finitistic TGWAs . . . . . . . . . . . . . . . . . . . . . . . . . 24 7.4 Simplicity of TGWAs of Lie type A1 × · · · × A1 . . . . . . . . . . . . . . . 26 8 Examples 9 Comparison between TGWAs and strongly graded rings 28 30 1 Introduction Higher rank generalized Weyl algebras (GWAs) were introduced by Bavula in [1] and are defined as follows. Let R be a ring and σ : Zn → Aut(R), g 7→ σg, an action of Zn on R by ring automorphisms, t = (t1, . . . , tn) an n-tuple of non-zero elements of the center of R such that σi(tj) = tj for all i 6= j. The generalized Weyl algebra of rank (or degree) n ∈ Z>0, denoted R(σ, t), is the ring extension of R by X1, . . . , Xn, Y1, . . . , Yn subject to the relations Xir = σi(r)Xi, YiXi = ti, YiYj = YjYi, Yir = σ−1 i (r)Yi, ∀r ∈ R, i ∈ {1, . . . , n}, XiYi = σi(ti), XiXj = XjXi XiYj = YjXi, ∀i ∈ {1, . . . , n}, ∀i, j ∈ {1, . . . , n}, ∀i 6= j, (1a) (1b) (1c) (1d) := σei and ei = (0, . . . , where σi In [13] Mazorchuk and Turowska introduced a class of algebras called twisted generalized Weyl algebras (TGWAs), which is a generalization of the higher rank generalized Weyl algebras (see Section 2.3 for i 1, . . . , 0) ∈ Zn. 2 1 Introduction the precise definition). Basically relation (1c) is relaxed in order to also accomodate situations where the defining relations among the Xi's (and among the Yi's) are some q-commutation relations (say, XiXj = qXjXi, i < j) or some Serre-type relations (such as X 2 i Xj − 2XiXjXi + XjX 2 i = 0). Already GWAs of rank one include many interesting algebras such as U(cid:0)sl(2)(cid:1) and its quantization and deformations, as well as several quantized function spaces and so called generalized down-up algebras, see for example [2] and references therein. This unification of different algebras into one family has proved to be very fruitful. For example in [5] the authors classified all simple and indecomposable weight modules over a rank one GWA, which in particular gives in one stroke a description of such modules for characteristic zero, quantized, and modular U(cid:0)sl(2)(cid:1). One of the original motivations to introduce the TGWAs was that in higher rank sev- eral algebras which one would expect to be GWAs, such as the multiparameter quantized Weyl algebras defined in [12], are not GWAs (at least not in an obvious way). However they are TGWAs as was shown in [14]. In [15] some algebras of importance in the repre- and the extended orthogonal Gelfand-Tsetlin algebras (including certain localizations of sentation theory of gl(n), namely the Mickelsson-Zhelobenko algebra Z(cid:0)gl(n), gl(n − 1)(cid:1) U(cid:0)gl(n)(cid:1)), were shown to be examples of TGWAs. It is expected that the corresponding primitive quotients of U(cid:0)gl(3)(cid:1) are TGWAs and used this fact to produce multivariable quantized versions of these algebras are also TGWAs. Sergeev [23] proved that some When it comes to representation theory, several classes of simple weight (with respect to the commutative subring R) modules over TGWAs have been classified in [7, 13, 15], including simple graded modules [15]. Bounded and unbounded ∗-representations were described in [14]. analogs of Hahn polynomials. In this paper we investigate TGWAs from another point of view, namely that of graded algebras; every TGWA of rank n is Zn-graded in a natural way. It is known that the class of higher rank GWAs, hence the class of TGWAs, includes all skew group algebras over Zn (take ti = 1 for all i in the definition above). Another related fact, proved in [6], is that any TGWA where ti is regular for each i can be embedded in a crossed product algebra. Consider the following problem: Problem 1. Given a G-graded algebra A = Lg∈G where Ae is commutative, is it true that each non-zero ideal of A has non-zero intersection with the centralizer of Ae in A? This is known to be true for strongly group graded rings [21] and for crystalline graded rings [20]. However, TGWAs are in general not strongly graded nor crystalline graded. The first main result of this paper (see Section 3) shows that this also holds for TGWAs: Theorem A. Let A = A(R, σ, t, µ) be a twisted generalized Weyl algebra. Each non-zero ideal of A has non-zero intersection with the centralizer of R in A. To establish this, we first prove that the gradation form (see Section 2.2.1) of A is always non-degenerate (Corollary 3.3), an interesting and useful fact in itself. Graded 3 1 Introduction rings with this property have been studied before [4]. Then we can apply the general result from [18, Theorem 3]. We introduce the notion of regularly graded algebras, which is a generalization of crystalline graded algebras, and provide exact conditions for a TGWA to be regularly graded (Theorem 4.3). The second problem that we address is the following. Problem 2. Given a G-graded algebra A = Lg∈G Ag where Ae is commutative, when is Ae maximal commutative in A? When is the centralizer of Ae in A commutative? If one can prove that a commutative subalgebra R has a commutative centralizer in A, then it follows that the centralizer is the unique maximal commutative subalgebra in A which contains R. This situation is of great value in representation theory when studying weight modules over A with respect to the commutative subalgebra R. In [13] a sufficient condition for the degree zero subalgebra of a TGWA to have a commutative centralizer was given. We generalize this result by Theorem 5.5 and Remark 5.6 and provide necessary and sufficient conditions: Theorem B. Let A = A(R, σ, t, µ) be a twisted generalized Weyl algebra. Suppose that R is a domain or that R is Zn-simple with respect to the action σ : Zn → Autk(R). Then the centralizer of R in A is commutative if and only if ker(σ) has a Z-basis {k1, . . . , ks} such that [Amki , Alkj ] = 0 for all m, l ∈ Z and all i, j ∈ {1, . . . , s}, i 6= j, where [a, b] = ab − ba. As an application we prove, in Theorem 5.10, that the family of TGWAs, Tq(C), parametrized by a scalar q and a symmetric generalized Cartan matrix C, introduced in [8], satisfies this condition (but not the condition in [13]). We also prove that the number s in Theorem B in this case is equal to the number of connected components of the Coxeter graph of C (Theorem 5.9). Finally, the third problem we consider is the question of simplicity of TGWAs. Problem 3. Give necessary and sufficient conditions, expressed only in terms of the initial data (R, σ, t, µ), for a twisted generalized Weyl algebra A(R, σ, t, µ) to be simple. The phrase "expressed in terms of the initial data" is not meant to be precise, but rather to indicate the type of condition we aim for. Classical Weyl algebras are well- known to be simple, but generalized Weyl algebras are not always simple. A result of Jordan [9, Theorem 6.1] provides a criterion for the simplicity of a degree one generalized Weyl algebra R(σ, t), where R is a commutative Noetherian ring, σ ∈ Aut(R) and t ∈ R. Namely, R(σ, t) is simple if and only if (i) σ has infinite order; (ii) t is a regular element in R; (iii) R has no σ-invariant ideals except {0} and R; and (iv) for all positive integers m, Rt + Rσm(t) = R. One area where this type of results are of importance is the area of dynamical systems. In [24] skew group algebras associated to dynamical systems are studied and conditions for simplicity of certain Z-graded skew group algebras and their analytical analogues are obtained. The Z-graded skew group algebras are in fact examples of GWAs of rank one. In another paper by Jordan, [10], the simplicity of a certain localization of the multipa- rameter quantized Weyl algebra is proved. In [14] it was proved that the multiparameter 4 2 Preliminaries quantized Weyl algebra is an example of a TGWA, and in [6] that Jordan's simple local- ization is also a TGWA. The third main result of this paper (Theorem 7.12) unify and extend these two sim- plicity results by Jordan to a large family of twisted generalized Weyl algebras. Theorem C. Let A = A(R, σ, t, µ) be a twisted generalized Weyl algebra where R is Noetherian (or more generally, it is enough that A is so called R-finitistic), and where ti is regular in R for each i. Then A is simple if and only if the following three assertions hold: (i) AXi1Xi2 · · · XimA = A for all m ∈ Z≥0 and all i1, . . . , im ∈ {1, . . . , n}; (ii) R is Zn-simple with respect to the action given by σ; (iii) The center of A is contained in R. The method we use to prove this result is a localization technique inspired by [10]. To be able to carry out the localization one needs to know that certain subsets of the algebra are Ore sets. It turns out that this condition is in fact equivalent to the TGWA to be so called R-finitistic, a notion introduced in [8] (where it was called "locally finite over R"). Both conditions hold if R is Noetherian. In a special case, still covering all higher rank GWAs, condition (i) of Theorem C can be made completely explicit (Theorem 7.18). In Section 8 we provide several interesting examples of TGWAs and display some phenomena, which are not possible in other common classes of graded algebras. The paper is concluded with a discussion about our results for TGWAs in the context of graded algebras. We describe how our results for TGWAs are related to, and in several cases differ from, the corresponding results for other classes of graded algebras. Acknowledgements The first author is grateful to V. Futorny for interesting discussions. The second author is grateful to P. Lundström for interesting discussions. 2 Preliminaries 2.1 Notation and definitions Rings and ring homomorphisms are always assumed to be unital. Ideals are understood to be two-sided. Throughout this paper k will denote an arbitrary commutative ring, unless otherwise stated. Suppose that S is a ring. The group of units in S is denoted by U (S). An element a ∈ S is said to be regular in S if a is not a left nor a right zero-divisor in S. The set of regular elements of S is denoted by Sreg. If a, b ∈ S, then we let [a, b] denote the element ab − ba ∈ S. The centralizer (or commutant) of a subset T ⊆ S is denoted by CS(T ) and is defined as the set (cid:8)s ∈ S st = ts, ∀t ∈ T(cid:9) which is clearly a subring of S. If T 5 2 Preliminaries is commutative and T = CS(T ), then T is said to be maximal commutative in S. The center of S, CS(S), is denoted by Z(S). If X is a set, then the free S-bimodule on X is defined as SXS := Lx∈X SxS where S. Namely, FS(X) :=L∞ for x ∈ X each summand SxS is by definition isomorphic as an S-bimodule to S ⊗Z S via s1xs2 7→ s1⊗s2. The free S-ring FS(X) on X is defined as the tensor algebra of SXS over n=0(SXS)⊗S n, where (SXS)⊗S n = (SXS) ⊗S · · · ⊗S (SXS) (n factors) and (SXS)⊗S 0 = S by convention. Thus, a general element in FS (X) is a sum of monomials of the form s1x1s2x2 · · · skxksk+1 where si ∈ S and xi ∈ X, for i ∈ {1, . . . , k}. There is a natural ring homomorphism S → FS(X) given by inclusion into the degree zero component. Suppose that S is a k-algebra, i.e. S is a ring together with a ring homomorphism ηS : k → S such that ηS(λ)s = sηS(λ) for all s ∈ S, λ ∈ k. Suppose that S′ is another k-algebra. A k-algebra homomorphism ϕ : S → S′ is a ring homomorphism such that ϕ ◦ ηS = ηS′. If X and Y are nonempty subsets of S, then XY denotes the k-linear span of the set {xy x ∈ X, y ∈ Y }. If G is a group acting as k-algebra automorphisms of S, ρ : G ∋ g 7→ ρg ∈ Autk(S), then an ideal I ⊆ S is said to be G-invariant if ρg(I) = I for all g ∈ G. If S and {0} are the only G-invariant ideals of S, then S is said to be G-simple. Recall that, given a group G, a G-gradation on S is a set of k-submodules, {Sg}g∈G, of g, h ∈ G, then the gradation is a strong G-gradation. A (strongly) G-graded k-algebra is a k-algebra together with a (strong) G-gradation on it. Each element a of a G-graded k- S such that S =Lg∈G Sg and SgSh ⊆ Sgh, for g, h ∈ G. If in addition SgSh = Sgh, for all algebra S can be written as a =Pg∈G ag where ag ∈ Sg, for g ∈ G, and ag = 0 for all but finitely many g ∈ G. The support of a is defined as the set Supp(a) = {g ∈ G ag 6= 0}, the cardinality of which is denoted by Supp(a). For g ∈ G, the k-submodule Sg is referred to as the homogeneous component of degree g. An element s ∈ S is said to be homogeneous of degree g, written deg(s) = g, if s ∈ Sg for g ∈ G. The neutral element of G is denoted by e and the subring Se is refered to as the neutral component of S. A left where Ig := Sg ∩ I for g ∈ G. If I is a graded ideal, then the quotient ring S/I is a (right, two-sided) ideal of a G-graded k-algebra S is said to be graded if I = Pg∈G Ig, G-graded k-algebra Lg∈G(S/I)g with a gradation defined by (S/I)g =(cid:8)s + I s ∈ Sg(cid:9) for g ∈ G. 2.2 Bilinear forms and radicals associated to group graded algebras 2.2.1 The gradation form (By this it is understood that, for each g ∈ G, Ag is a k-submodule of A.) Denote by Let G be a group with neutral element e and A = Lg∈G Ag a G-graded k-algebra. pe : A → Ae the graded projection onto the neutral component, i.e pe(Pg∈G ag) = ae, where ag ∈ Ag for g ∈ G. To each G-graded k-algebra A we associate a map, which we shall refer to as the gradation form of A. It is defined in the following way: γA : A × A → Ae, (a, b) 7→ pe(ab) (2) 6 2 Preliminaries This form was introduced by Cohen and Rowen in [4] in their study of G-graded rings. Note that γA depends on the gradation on A. It is immediate that γA is k-bilinear and that γA(a, bc) = γA(ab, c) (3) for all a, b, c ∈ A. Furthermore, if a = Pg∈G ag and b = Pg∈G bg where ag, bg ∈ Ag, for g ∈ G, then The left respectively right radical of γA is defined as follows: γA(a, b) = Xg∈G agbg−1. (4) (5) (6) Radl(γA) :=(cid:8)b ∈ A γA(a, b) = 0, ∀a ∈ A(cid:9), Radr(γA) :=(cid:8)a ∈ A γA(a, b) = 0, ∀b ∈ A(cid:9). It is easy to check that Radl(γA) is a graded left ideal of A, while Radr(γA) is a graded right ideal of A. Definition 2.1. Let A be a G-graded k-algebra with gradation form γA. (i) If Radl(γA) = Radr(γA), then γA is said to be radical-symmetric and the left (and right) radical is simply denoted by Rad(γA). (ii) We say that γA is non-degenerate if it is radical-symmetric and Rad(γA) = {0}. (iii) If there exists an action of G by k-algebra automorphisms on Ae (denoted G ∋ g 7→ ρg ∈ Autk(Ae)) such that, for any g ∈ G, for all a ∈ Ag and all b ∈ Ag−1, then γA is said to be G-symmetric. γA(a, b) = ρg(cid:0)γA(b, a)(cid:1) (7) Remark 2.2. Note that if γA is G-symmetric, then it is radical-symmetric. Moreover, if γA is radical-symmetric, then Rad(γA) is a two-sided graded ideal of A. Remark 2.3. Also note that γA is non-degenerate in the sense of Definition 2.1 if and only if it is non-degenerate in the sense of Cohen and Rowen [4]. Furthermore, if this is the case, then the gradation on A is left (and right) non-degenerate in the sense of [18, Definition 2]. If A is a G-graded k-algebra with radical-symmetric gradation form γA, then we put ¯A := A/ Rad(γA). In that case Rad(γA) is a graded ideal and hence the quotient algebra ¯A has a natural G-gradation induced by the one on A; ¯A =Lg∈G ¯Ag, where ¯Ag =(cid:8)a + Rad(γA) ∈ ¯A a ∈ Ag(cid:9) for g ∈ G. For each g ∈ G we have that ¯Ag ≃ Ag/(Ag ∩ Rad(γA)) as k-modules. Since ¯A is a G-graded k-algebra, this allows us to define a gradation form on this algebra as well. The following lemma is proved under the hypothesis that γA is radical-symmetric. Lemma 2.4. The gradation form γ ¯A on ¯A is non-degenerate. 7 2 Preliminaries Proof. Since γA is radical-symmetric, so is γ ¯A. Take an arbitrary ¯a ∈ Rad(γ ¯A), i.e. such that γ ¯A(¯a, ¯b) = 0 for all ¯b ∈ ¯A. Choose a ∈ A such that ¯a = a + Rad(γA). By (4) we get γA(a, b) + Rad(γA) = Pg∈G agbg−1 + Rad(γA) = Pg∈G ¯ag¯bg−1 = γ ¯A(¯a, ¯b) = 0 in ¯A, for any b ∈ A. Thus γA(a, b) ∈ Rad(γA) ∩ Ae for all b ∈ A. However, 1 ∈ Ae and γA(r, 1) = r = γA(1, r) for each r ∈ Ae and hence Rad(γA) ∩ Ae = {0}. This shows that γA(a, b) = 0 for all b ∈ A, which means that a ∈ Rad(γA). Thus ¯a = 0 in ¯A, proving that Rad(γ ¯A) = {0}, i.e. γ ¯A is non-degenerate. 2.2.2 The gradical Let A be a G-graded k-algebra. The gradical of A (short for gradation radical), denoted by grRad(A), is defined as the sum of all graded ideals J of A which satisfy Ae ∩ J = {0}. The gradical grRad(A) is itself a graded ideal of A such that Ae ∩ grRad(A) = {0}, and by construction it is the unique maximal element in the set of all such graded ideals (ordered by inclusion). The gradical grRad(A) is related to the radicals of the gradation form γA of A, defined in Section 2.2.1, in the following way: Proposition 2.5. Let A be a G-graded k-algebra with gradation form γA. The following holds: (i) grRad(A) ⊆ Radl(γA) ∩ Radr(γA); (ii) If γA is G-symmetric, then Rad(γA) = grRad(A). Proof. (i) Let a ∈ grRad(A). Since grRad(A) is graded we can assume that a is homoge- neous, i.e. a ∈ Ag ∩grRad(A) for some g ∈ G. Let b ∈ A be arbitrary. Write b =Ph∈G bh where bh ∈ Ah for h ∈ G. Then, by (4), γA(a, b) = ab−g ∈ Ae ∩ grRad(A) = {0}. Also γA(b, a) = b−ga ∈ Ae ∩ grRad(A) = {0}. Since b was arbitrary, this proves that a ∈ Radl(γA) ∩ Radr(γA). (ii) If γA is G-symmetric, then it is radical-symmetric, so by part (i) it only remains to prove that Rad(γA) ⊆ grRad(A). Let a ∈ Rad(γA). Since Rad(γA) is graded we can assume that a ∈ Ag for some g ∈ G. Put J = AaA, the ideal in A generated by a. Since a is homogeneous, J is graded. If we show that Ae ∩ J = {0} then it follows that J ⊆ grRad(A), and in particular a ∈ grRad(A). Any element x of Ae ∩ J can be written as bhack x = Xh,k∈G hgk=e where bh, ch ∈ Ah for all h ∈ G. But then, for any h, k ∈ G with hgk = e we have, using (3) and that γA is G-symmetric, bhack = γA(bh, ack) = ρh(cid:0)γA(ack, bh)(cid:1) = ρh(cid:0)γA(a, ckbh)(cid:1) = 0 since a ∈ Rad(γA), proving that x = 0. Since x was arbitrary, Ae ∩ J = {0}. As shown above, this implies that a ∈ grRad(A). But a was arbitrary so we conclude that Rad(γA) ⊆ grRad(A). 8 2 Preliminaries 2.3 The TGWC and TGWA We shall now recall the definition of a twisted generalized Weyl algebra from [13], [14]. Fix a positive integer n and set n = {1, 2, . . . , n}. Let k be a commutative unital ring, and let R be a commutative unital k-algebra, σ : Zn → Autk(R), σ : g 7→ σg, be a group homomorphism from Zn to the group of k-algebra automorphisms of R, µ = (µij)i,j∈n be a matrix with invertible entries from k and t = (t1, . . . , tn) be an n-tuple of central elements of R. The quadruple (R, σ, t, µ) will be referred to as a TGW datum. For convenience we will denote σei, where ei = (0, . . . , given n commuting k-algebra automorphisms σ1, . . . , σn of R we put σg = σg1 for all g ∈ Zn and this defines a group homomorphism σ : Zn → Autk(R). i 1, . . . , 0) ∈ Zn by σi. Conversely, 1 ◦ · · · ◦ σgn n The twisted generalized Weyl construction (TGWC) A′ = A′(R, σ, t, µ) obtained from the TGW datum (R, σ, t, µ) is the free R-ring FR(Z) on a set Z = {Xi, Yi i ∈ n} of 2n symbols modulo the following relations: Xir = σi(r)Xi, YiXi = ti, XiYj = µijYjXi, Yir = σ−1 i (r)Yi, XiYi = σi(ti), for r ∈ R, i ∈ n, for i ∈ n, for i, j ∈ n, i 6= j. (8a) (8b) (8c) Due to relation (8b), it is not guaranteed that the k-algebra homomorphism ι : R → A′ given by composing the natural map R → FR(Z) with the canonical projection FR(Z) → A′ is injective. However, throughout this paper we will make the additional assumption that ι is injective. In [6, Corollary 2.17] it was proved that if t1, . . . , tn are regular elements of R, then ι is injective if and only if the following consistency conditions hold: σiσj(titj) = µijµjiσi(ti)σj(tj), tjσiσk(tj) = σi(tj)σk(tj), ∀i, j ∈ n, i 6= j, ∀i, j, k ∈ n, i 6= j 6= k 6= i. (9a) (9b) Condition (9a) appeared already in [13, 14]. Thus, when considering examples where the ti's are regular in R, it is enough to verify relations (9) to know that ι : R → A′ is injective. Relations (8) are homogeneous with respect to the Zn-gradation on the free R-ring FR(Z) uniquely defined by requiring deg Xi = ei, deg Yi = −ei, deg r = 0, for i ∈ n, r ∈ R, i 1, 0, . . . , n 0). Thus this Zn-gradation on FR(Z) descends to a Zn- 0 = ι(R). Thus, since we where ei = ( gradation {A′ always assume that ι is injective, we may identify R with A′ 1 0, . . . , 0, g}g∈Zn on the quotient A′. It is easy to see that A′ 0 via the isomorphism ι. The twisted generalized Weyl algebra (TGWA) A = A(R, σ, t, µ) of rank n is defined to be A′/I, where I = grRad(A′) is the gradical of A′ (i.e. the sum of all graded two- sided ideals of A′ intersecting R trivially). Since I is graded, A inherits a Zn-gradation {Ag}g∈Zn from A′. Note that R is isomorphic to its image in the quotient A = A′/I. We shall therefore identify these algebras and denote them with the same letter R. 9 3 Essential subalgebras of TGWAs By a monic monomial in a TGWC A′ or TGWA A we mean a product of the generators Xi, Yi, for i ∈ n. A product of an element of R and a monic monimial is referred to as a monomial. A monic monomial is called reduced if it has the form Yi1 · · · Yik Xj1 · · · Xjl, where ir, js ∈ n and {i1, . . . , ik} ∩ {j1, . . . , jl} = ∅. (10) The following lemma is useful and straightforward to prove. Lemma 2.6. A′ (respectively A) is generated as a left and as a right R-module by the reduced monomials in A′ (respectively A). 3 Essential subalgebras of TGWAs The ideal intersection property has been studied in e.g. [18]. In this paper, subalgebras having this property are said to be essential and they are defined as follows. Definition 3.1. A subalgebra S′ of a k-algebra S is said to be an essential subalgebra of S, if S′ ∩ I 6= {0} for each non-zero ideal I of S. One useful fact, is that injectivity of a k-algebra homomorphism ϕ : S → T follows from the injectivity of ϕS′: S′ → T (the restriction of ϕ to S′) if S′ is an essential subalgebra of S. Let A be a group graded k-algebra with commutative neutral component Ae. Consider the following question: Is the centralizer of Ae in A an essential subalgebra of A? The answer is known to be affirmative if A is strongly graded [21] or if A is a crystalline graded ring (CGR) [20], or more generally for any ring satisfying the condition in [18, Theorem 3]. We shall now prove that the answer is affirmative also for TGWAs. This is interesting since there are TGWAs which, with their natural Zn-gradation, are neither CGRs nor strongly graded. We begin with some useful preliminary resuts. The following interesting and useful commutation relation (proved under additional assumptions in [6], [8]) shows that the gradation form γA′ on an arbitrary TGWC A′ is always Zn-symmetric (in the sense of Definition 2.1). Lemma 3.2. Let A′ = A′(R, σ, t, µ) be a TGWC. For any g ∈ Zn, we have ab = σg(ba) (11) for all a ∈ A′ g and all b ∈ A′ −g. Hence the gradation form γA′ on A′ is Zn-symmetric. It suffices to prove the statement when a ∈ A′ Proof. Let g ∈ Zn be arbitrary. g is a monomial. Also, we may assume that a is monic because if the statement holds for a monic monomial then rab = rσg(ba) = σg(σ−g(r)ba) = σg(bra) for any r ∈ R, any b ∈ A′ If we first suppose that a = Xi, then b has degree −ei and hence b = rYi for some r ∈ R, by Lemma 2.6. We then get ab = XirYi = σi(r)σi(ti) = σi(rYiXi) = σi(ba). The case a = Yi is treated analogously. By iterating this argument we find that the statement holds for an arbitrary monic monomial a. −g and monic monomial a. 10 3 Essential subalgebras of TGWAs By Proposition 2.5 we obtain the following important result. Corollary 3.3. The gradical I = grRad(A′) of a TGWC A′ is equal to the radical of the gradation form γA′ on A′, and the gradation form γA on the corresponding TGWA A = A′/ grRad(A′) is non-degenerate. Remark 3.4. By Remark 2.3 this shows that for a general TGWA A, γA is non- degenerate in the sense of Cohen and Rowen [4] and furthermore, the grading on A is left (and right) non-degenerate in the sense of [18, Definition 2]. Remark 3.5. Let A′ = A′(R, σ, t, µ) be a TGWC. If we assume that µ is symmetric, then we can define an involution ∗ : A′ → A′. Recall the left and right Shapovalov-type forms from [15], F l : A′ × A′ → R and F r : A′ × A′ → R defined by F l(a, b) = p0(a∗b) and F r(a, b) = p0(ab∗) respectively. In [15] it was shown that, if R is a domain, then the radicals of F l and F r, i.e. Rad(F r) := {a ∈ A′ F r(a, b) = 0 for any b ∈ A} respectively Rad(F l) := {a ∈ A′ F l(b, a) = 0 for any b ∈ A}, coincide with eachother and with the ideal I. Corollary 3.3 generalizes this relation to the case of arbitrary TGWAs and arbitrary µ. Theorem 3.6. Let A = A(R, σ, t, µ) be a TGWA. Then CA(R) is an essential subalgebra of A. That is, each non-zero ideal of A intersects the centralizer of R in A non-trivially. Proof. This follows from Corollary 3.3 (Remark 3.4) and [18, Theorem 3]. For the con- venience of the reader we give a proof here. Let J be a non-zero ideal of A. Among all non-zero elements of J, choose one, a = Pg∈Zn ag, where ag ∈ Ag for g ∈ G, such that its support is minimal, i.e. such that Supp(a) is as small as possible. Pick some h ∈ Zn such that ah 6= 0. Then there are homogeneous elements b, c ∈ A such that deg(b) + deg(c) = −h and bahc 6= 0. Indeed, otherwise ah would generate a graded ideal in A with zero intersection with R. Put a′ = bac. Then a′ is a non-zero element of J such that Supp(a′) = Supp(a), but in addition a′ 0 = bahc 6= 0. We claim that a′ ∈ CA(R). Otherwise there is an r ∈ R with a′′ = [a′, r] 6= 0. But then a′′ is a non-zero element of J such that Supp(a′′) < Supp(a) because a′′ 0 6= 0. This contradicts the minimality of Supp(a). Hence a′ ∈ CA(R) ∩ J. This proves that CA(R) ∩ J 6= {0}. 0 ∈ R and a′ 0 = [a′ 0, r] = 0 since a′ Remark 3.7. Note that the statement in the above proof is concerned with ideals, not only graded ideals. It is immediate from the definition that any non-zero graded ideal of a TGWA intersects R non-trivially. The following example shows two things: (i) The corresponding statement is not true for all GWAs. (ii) Not all GWAs are isomorphic to a CGR with neutral component R. Example 3.8. Put R = C[H]/(H 2), and let t = H + (H). Let ε ∈ C be a non-root of unity and define σ ∈ AutC(R) by σ(t) = εt. Consider A = R(σ, t), the corresponding 11 4 Characterization of regularly graded TGWAs generalized Weyl algebra of rank one (equivalently, A is a TGWC of rank one). We claim that CA(R) = R. Suppose that a ∈ CA(R). Write a = Pk∈Z rkZ (k) where Z (k) equals X k if k ≥ 0 and Y −k otherwise. Then 0 = [a, t] = Pk∈Z rk(εk − 1)Z (k). Since A is Z-graded we get rk(εk − 1)Z (k) = 0 for each k ∈ Z. It is well-known that each Z (k) generates a free left R-module. Thus, since ε is not a root of unity, rk = 0 for all k ∈ Z \ {0}. Hence CA(R) = R. However, it is easy to see that X 2 generates an ideal in A which has zero intersection with R. 4 Characterization of regularly graded TGWAs Crystalline graded algebras were introduced by Nauwelaerts and Van Oystaeyen in [11] as a natural generalization of e.g. GWAs and G-crossed products. Not all TGWAs fit into this class. We shall now introduce the notion of regularly graded algebras, of which crystalline graded algebras are a special case. Definition 4.1. A G-graded k-algebra is said to be regularly graded if each homogeneous component contains a regular element. In this case the gradation is said to be regular. Note that the gradation form of a regularly graded algebra is necessarily non-degenerate. Moreover, a regular gradation is necessarily faithful (called component regular in [22]) in the sense of Cohen and Montgomery [3]. Remark 4.2. As already mentioned, G-crossed products are examples of crystalline graded algebras, and hence they are regularly graded. Moreover, each G-crossed product is strongly G-graded. However, not every strongly graded algebra is regularly graded. To see this, let A := M3(C) be the algebra of 3 × 3-matrices over C. It is possible to define a strong Z2-gradation on A = A¯0 ⊕ A¯1 (see e.g. [16, Example 6.11]) and one can show that A¯1 contains no regular element. Therefore A is strongly Z2-graded (hence faithfully graded and γA is non-degenerate) but not regularly graded. The following theorem gives a characterization of those TGWAs whose Zn-gradation is regular. Theorem 4.3. Let A = A(R, σ, t, µ) be a TGWA. The following assertions are equiva- lent: (i) A is regularly graded; (ii) For each i ∈ n, ti ∈ Rreg; (iii) Each monic monomial in A is non-zero and generates a free left (and right) R- module of rank one; (iv) If a ∈ A is a homogeneous element such that bac = 0 for some monic monomials b, c ∈ A with deg(a) + deg(b) + deg(c) = 0, then a = 0. 12 5 Centralizers of R in TGWAs Proof. (i) ⇒ (ii): If A is regularly graded, then for each i ∈ n, Aei contains a regular element. Since Aei = XiR it means there exists an r ∈ R such that Xir = σi(r)Xi is regular in A. But then Xi must be regular in A as well. Analogously Yi must be regular in A. Thus YiXi = ti is regular in R. 1 n · · · Z (gn) n i = X k i Z (−gn−1) n−1 · · · Z (−g1) , where Z (k) (ii) ⇒ (i): Assume that ti is regular in R for all i ∈ n. Let g ∈ Zn be arbitrary and let a = Z (g1) if k < 0. Put a∗ = Z (−gn) . Then aa∗ and a∗a are products of elements of the form σh(ti) (h ∈ Zn, i ∈ n), hence they are both regular in R. If b ∈ A with ab = 0, then a∗ab = 0 so a∗abh = 0 for each homogeneous component bh of b. But if bh 6= 0 for some h ∈ Zn then by Corollary 3.3 there is some c ∈ A−h such that bhc 6= 0. Then a∗abhc = 0 and bhc ∈ R \ {0} contradict the fact that a∗a is regular in R. So b = 0. Similarly ba = 0 for some b ∈ A implies b = 0. Thus a is regular in A. if k ≥ 0 and Z (−k) = Y −k i i 1 (ii) ⇒ (iii): Let a ∈ A be a monic monomial and let r ∈ R \ {0}. As in the previous step, aa∗ is a regular element of R, hence raa∗ 6= 0 and thus ra 6= 0. (iii) ⇒ (iv): By Lemma 3.2, we have bac = σh(acb) where h = deg(b). Thus acb = σ−1 h (bac) = 0. If a 6= 0, then by Corollary 3.3, da ∈ R \ {0} for some d ∈ A with deg(d) = − deg(a). But by (iii), da · cb = 0 implies that da = 0. This contradiction shows that a = 0. (iv) ⇒ (ii): Suppose that rti = 0 for some r ∈ R, i ∈ n. Then 0 = rti = rYiXi = Yiσi(r)Xi. By (iv) we get σi(r) = 0 and hence r = 0. This proves that ti is regular in R for each i ∈ n. Remark 4.4. Property (iv) in the above theorem is very convenient for proving relations in a regularly graded TGWA. 5 Centralizers of R in TGWAs 5.1 Results for general TGWAs Given a TGW datum (R, σ, t, µ), we shall denote the kernel of the group homomorphism σ : Zn → Autk(R) by K. Also we use the notation Z (k) i = (X k i , Y −k i k ≥ 0, , k < 0 where i ∈ n and k ∈ Z. Theorem 5.1. Let A = A(R, σ, t, µ) be a TGWA and let AK be the following subalgebra of A: AK := Mg∈K Ag. Then AK is contained in CA(R), the centralizer of R in A. Moreover, if R is a domain or if R is Zn-simple, then AK = CA(R). Proof. Take an arbitrary g = (g1, . . . , gn) ∈ K and a ∈ Ag. Suppose that a /∈ CA(R). Then there exists an r ∈ R such that ar 6= ra. On the other hand, ar = σg(r)a and 13 5 Centralizers of R in TGWAs hence σg(r) cannot be equal to r. This shows that σg 6= idR, which contradicts g ∈ K. Therefore a ∈ CA(R). Since a and g were arbitrary, this shows that AK ⊆ CA(R). For the converse, take an arbitrary non-zero a ∈ CA(R) (clearly 0 ∈ AK ∩ CA(R)) and let r ∈ R be arbitrary. Write a = Pg∈Zn ag, with ag ∈ Ag for g ∈ Zn. Then 0 = [a, r] = Pg∈Zn[ag, r]. Since deg r = 0 we have [ag, r] ∈ Ag for each g ∈ G and from the gradation on A = Lg∈Zn Ag we conclude that [ag, r] = 0 for each g ∈ G. Pick an arbitrary g ∈ G for which ag 6= 0. Then, since r was arbitrary, 0 = [ag, r] = (σg(r) − r)ag, ∀r ∈ R. (12) Since ag 6= 0 there is, by Corollary 3.3, an element c ∈ A−g such that agc 6= 0. Multiplying (12) from the right by c we get (σg(r) − r)agc = 0 for all r ∈ R, and 0 6= agc ∈ R. If R is a domain, then we get that σg(r) = r for all r ∈ R. If instead R is Zn-simple, then note that the set J = {s ∈ R (σg(r) − r)s = 0, ∀r ∈ R} is a non-zero (since agc ∈ J) Zn-invariant ideal of R and hence 1 ∈ J which means that σg = idR. In both cases we conclude that g ∈ K and since g ∈ Supp(a) was arbitrarily chosen we get a ∈ AK. This shows that CA(R) ⊆ AK. Corollary 5.2. Let A = A(R, σ, t, µ) be a TGWA. If R is maximal commutative in A, then σ : Zn → Autk(R) is injective. Conversely, if R is a domain or R is Zn-simple, then injectivity of σ implies that R is maximal commutative in A. It is an interesting question to ask if the centralizer CA(R) is commutative. If this is the case, then CA(R) is a maximal commutative subalgebra of A, more specifically the unique maximal commutative subalgebra of A containing R. The following result gives a description of commuting elements in the centralizer. Theorem 5.3. Let A = A(R, σ, t, µ) be a regularly graded TGWA. Let H be any subgroup of K = ker(σ) of rank one. Then the subalgebra Lg∈H Ag is commutative. Proof. Write H = Z · g, where g ∈ K. Let a, b be any homogeneous elements in A of degrees in H, say deg a = kg, deg b = mg where k, m ∈ Z. If we can show that ab − ba = 0, then we are done. By replacing g by −g we can assume that k + m ≥ 0. First we assume that k = 1. Let c be any monic monomial of degree −g. Then abcm+1 has degree zero, and thus abcm+1 = σg(bcm+1a) by Lemma 3.2, = bcm+1a = bacm+1 since g ∈ K, since ca = σ−g(ac) = ac by Lemma 3.2 and since g ∈ K. By Theorem 4.3(iv), we conclude that ab − ba = 0. Now let k be general. Let c be any monic monomial of degree −g. Then abck+m has degree zero, and thus, like before we have abck+m = σkg(bck+ma) = bck+ma = back+m 14 5 Centralizers of R in TGWAs where we have used the case k = 1, k + m times in the last step. This shows that (ab − ba)ck+m = 0 and by Theorem 4.3(iv), we conclude that ab − ba = 0. Combining Theorem 5.1 and Theorem 5.3 we immediately get the following sufficient condition for the centralizer of R in A to be commutative. Corollary 5.4. Let A = A(R, σ, t, µ) be a regularly graded TGWA, where R is either a domain or is Zn-simple. If K = ker(σ) has rank at most one, then the centralizer of R in A is commutative. The condition of K having rank at most one is not necessary. In the next section we give a large family of examples where the centralizer is commutative, which cover cases for which K has arbitrarily large rank. The following theorem, which is the main result of this section, gives a necessary and It is a sufficient condition for the subalgebra AK of a TGWA A to be commutative. generalization of [13, Lemma 5] (see Remark 5.7 below). Theorem 5.5. Let A = A(R, σ, t, µ) be a TGWA and let K = ker(σ). The subalgebra AK := Lg∈K Ag is commutative if and only if there exists a Z-basis {k1, . . . , ks} for K with the following property: For any two basis elements ki 6= kj, and m, l ∈ Z, we have [Amki, Alkj ] = 0. (13) Proof. Clearly (13) is necessary for AK to be commutative. Conversely, assume that {k1, . . . , ks} is a Z-basis for K such that (13) holds and let a, b ∈ AK be arbitrary homogeneous elements. Then there are unique integers αi, βi such that α := deg a = s Xi=1 αiki, β := deg b = βiki. s Xi=1 For i ∈ {1, . . . , s}, let a′ monomial of degree −βiki, and ci be a monic monomial of degree βiki. Put a′ = a′ and b′ = b′ i be a monic monomial of degree −αiki and b′ s and c = c1 · · · cs. Then using Lemma 3.2 and that α, β ∈ K we have i be a monic 1 · · · a′ s 1 · · · b′ (aba′b′)c = σα(ba′b′a)c = b(a′b′ac) = bσ−α−β(aca′b′) = baca′b′. (14) For any i ∈ {1, . . . , s}, the element ci commutes with a′ all three elements are contained in the subalgebra Lg∈Zki i by Theorem 5.3, because Ag and Zki is a subgroup of K of rank one. Also, if i 6= j, then ci commutes with a′ j due to assumption (13). Thus c commutes with both a′ and b′ which together with (14) entails that (ab−ba)a′b′c = 0. By Theorem 4.3(iv), this implies that ab − ba = 0. j and with b′ i and b′ Remark 5.6. If R is either a domain or Zn-simple, then by Theorem 5.1, AK coincides with the centralizer CA(R). 15 5 Centralizers of R in TGWAs Remark 5.7. Theorem 5.5 is a generalization of [13, Lemma 5] where it was proved that AK is commutative under the assumptions that µij = 1 for all i, j and that XiXj = XjXi j=1 kjej ∈ K with ki 6= 0. Even if we only assume that µij = 1 for all i 6= j, this assumption is still stronger than the one in Theorem 5.5 because there exist examples where AK is commutative without satisfying the condition of [13, Lemma 5] (see Example 5.11). for any i, j ∈ F (W ), where F (W ) is the set of i ∈ n such that there is a k =Pn The following is an example where the centralizer CA(R) of R in A is not commutative. Example 5.8. Let n = 2, R = k = C, σ1 = σ2 = idC, t1 = t2 = 1 and µ12 = 2, µ21 = 1 2. Then the only consistency relation from (9) is t1t2 = µ12µ21σ−1 2 (t1), which holds. Let A = A(R, σ, t, µ). Then X1r = rX1 for all r ∈ R so R is not maximal commutative in A. In fact A = CA(R), and A is not commutative since X1Y2 = 2Y2X1. 1 (t2)σ−1 5.2 Maximal commutative subalgebras of TGWAs associated to Cartan matrices In this section we show that an interesting and more explicit description of the centralizer CA(R) is possible for a class of TGWA introduced in [8]. Moreover, in all these cases the centralizer is commutative, hence maximal commutative in A = A(R, σ, t, µ). Recall that a matrix C = (aij)1≤i,j≤n with integer entries is called a generalized Cartan matrix if the following assertions hold: (i) aii = 2 for all i ∈ n; (ii) aij ≤ 0 for all i 6= j; (iii) For all i, j ∈ n, aij = 0 if and only if aji = 0. Generalized Cartan matrices are fundamental in the theory of Kac-Moody algebras. In this section we assume that k is a field. In [8] a family of TGWAs were constructed, denoted Tq,µ(C), where q ∈ k\{0}, µ = (µij)1≤i,j≤n and C = (aij)1≤i,j≤n, a symmetric generalized Cartan matrix. We will consider the special case when µij = 1 for all i, j ∈ n and denote these algebras by Tq(C). Their construction is as follows. Take R to be the following polynomial algebra over k: R = k[H (k) ij 1 ≤ i < j ≤ n, and k = aij, aij + 2, . . . , −aij]. Define σ1, . . . , σn ∈ Autk(R) by setting, for all i < j and k = aij, aij + 2, . . . , −aij: ij + H (k−2) ij , where H (aij −2) ij := 0, σj(H (k) σi(H (k) σr(H (k) ij ) = qkH (k) j (H (k) ij ) = σ−1 ij ), ij ) = H (k) ij , r 6= i, j 1 ◦ · · · ◦ σgn (15a) (15b) (15c) and define σ : Zn → Autk(R) by σg = σg1 put n for g ∈ Zn. For notational purposes, Hij = H (−aij ) ij , Hji = σ−1 j (Hij) for all i < j, and Hii = 1 for all i ∈ n 16 and define 5 Centralizers of R in TGWAs ti = Hi1Hi2 · · · Hin, for i ∈ n. One can verify that the σi's commute with eachother, and that consistency relations (9) hold (see [8] for details). The algebra Tq(C) is defined as the TGWA associated to the above data, Tq(C) = A(R, σ, t, µ), where µij = 1 for all i, j ∈ n. For q ∈ k\{0}, put [k]q = q−k+1 + q−k+3 + · · · + qk−1 if k ∈ Z≥0 and [−k]q = −[k]q if k ∈ Z<0. Recall that q is said to have quantum characteristic zero if, for any integer n, [n]q = 0 implies that n = 0. Let ΓC be the Coxeter graph associated to C; its vertex set is V (ΓC ) = {1, . . . , n} and i, j are connected if and only if aij < 0 (we do not need to label the edges here). Let Comp(ΓC ) be the set of connected components of the graph ΓC. Let K be the kernel of the group homomorphism σ : Zn → Autk(R). For each γ ∈ Comp(ΓC) define gγ ∈ Zn by gγ = Xi∈V (γ) ei where {e1, . . . , en} is the standard basis of Zn, and V (γ) is the vertex set of the subgraph γ. Let A = Tq(C). By Theorem 5.1, the centralizer CA(R) is equal to AK = Lg∈K Ag. We have the following description of the gradation group K. Theorem 5.9. Assume that q ∈ k\{0} has quantum characteristic zero. Let C be an n × n symmetric generalized Cartan matrix and let A = Tq(C) = A(R, σ, t, µ) be the twisted generalized Weyl algebra defined above. Then the set {gγ γ ∈ Comp(ΓC)} (16) forms a Z-basis for the kernel K of σ : Zn → Autk(R). In particular, the rank of the free abelian group K is equal to the number of connected components of the Coxeter graph ΓC of C. Proof. Clearly the gγ's are linearly independent over Z. Let g = (g1, . . . , gn) ∈ Zn and 1 ≤ i < j ≤ n. If aij = 0, then σg(H (0) ij . If aij < 0, then, by (15), ij ) = H (0) σg(H (2+aij ) ij gj −gi j (2+aij ) ij (H ) = σ = q(2+aij )(gj −gi)H (2+aij ) ) = ij + q(aij +1)(gj −gi−1)[gj − gi]qH (aij ) ij . So g ∈ K implies that gi = gj for all i, j for which aij < 0, i.e. which are connected in ΓC. This shows that any g ∈ K is a Z-linear combination of the elements gγ, for γ ∈ Comp(ΓC). Conversely, if g ∈ Comp(ΓC), then gi = gj for all i 6= j with aij < 0 and thus for all i < j and all k. So σg fixes all the generators of (H (k) ij ) = H (k) ij ij ) = σgj −gi σg(H (k) R and hence g ∈ K. j 17 6 Ore localizations and a finiteness condition for TGWAs A generalized Cartan matrix is said to be indecomposable if it cannot be rearranged, by applying simultaneous row and column permutations, into a block matrix with more than one block. An immediate corollary of Theorem 5.9 is that if the Coxeter graph ΓC of C is connected (which is equivalent to that C is indecomposable) then K has rank one, and hence by Corollary 5.4, CA(R) is commutative. The following theorem shows that this holds for any TGWA A = Tq(C) associated to a symmetric generalized Cartan matrix C, not necessarily indecomposable. Theorem 5.10. Let C be a symmetric generalized Cartan matrix, and q ∈ k\{0} of quantum characteristic zero. Let A = Tq(C). Then the centralizer CA(R) is commutative (hence maximal commutative in A). Proof. We will show that the Z-basis {gγ γ ∈ Comp(ΓC)} for K satisfies condition (13) of Theorem 5.5. Let γ, γ′ ∈ Comp(ΓC), γ 6= γ′, and assume that a ∈ Amgγ , b ∈ Algγ′ for some m, l ∈ Z. Then ab = ba due to the fact that XiYj = YjXi for all i 6= j, and, by [8, Theorem 5.2(c)], XiXj = XjXi and YiYj = YjYi for all i ∈ V (γ), j ∈ V (γ′). Thus, by Theorem 5.5, CA(R) is commutative. . Thus K = Z · (1, 1), so 1, 2 ∈ F (W ) in the notation of [13] (see Remark 5.7 ΓC is above). However, X1X2 6= X2X1 and hence [13, Lemma 5] cannot be applied to conclude Example 5.11. Let C = (cid:2) 2 −1 −1 2 (cid:3) and consider the algebra Tq(C). The Coxeter graph that the subalgebra AK := Lg∈Zn Ag is commutative. Nevertheless, by Theorem 5.10, AK is commutative. Furthermore, since R is a domain, AK coincides with the centralizer CR(A) by Theorem 5.1. 6 Ore localizations and a finiteness condition for TGWAs In this section we introduce a finiteness condition for TGW data, in order to have suitable Ore sets that allow us to define a well-behaved localization. The motivation for doing so is that this particular localization turns out to be useful in the subsequent section when deriving conditions for simplicity of TGWAs. Definition 6.1. Let (R, σ, t, µ) be a TGW datum and let S be a subalgebra of R. We say that (R, σ, t, µ) is left S-finitistic if for any i, j ∈ n, i 6= j, there exist some k ∈ Z≥0 and s1, . . . , sk ∈ S, such that i (tj) + s1σk−1 σk i (tj) + · · · + sk−1σi(tj) + sktj = 0. (17) Similarly (R, σ, t, µ) is called right S-finitistic if for any i, j ∈ n, i 6= j, there exist some m ∈ Z≥0 and s′ m ∈ S, such that 1, . . . , s′ tj + s′ i (tj) = 0. 1σi(tj) + . . . + s′ mσm (18) (R, σ, t, µ) is S-finitistic if it is both left and right S-finitistic. If (R, σ, t, µ) is S-finitistic, then A(R, σ, t, µ) is also said to be S-finitistic. 18 6 Ore localizations and a finiteness condition for TGWAs Remark 6.2. The cases S = R and S = k were considered in [8], with the slight difference that in [8] the existence of s1, . . . , sk and s′ m such that (17) and (18) hold was required also when i = j. 1, . . . , s′ Proposition 6.3. Let (R, σ, t, µ) be a TGW datum. (i) If S1 and S2 are subalgebras of R such that S1 ⊆ S2 and if (R, σ, t, µ) is left (right) S1-finitistic, then (R, σ, t, µ) is also left (right) S2-finitistic. (ii) If R is Noetherian, then (R, σ, t, µ) is R-finitistic. (iii) If S is a subalgebra of R and (R, σ, t, µ) is S-finitistic then for all i 6= j, min{k ∈ Z≥0 equation (17) holds for some s1, . . . , sk ∈ S} = = min{m ∈ Z≥0 equation (18) holds for some s′ 1, . . . , s′ m ∈ S}. (19) The common number in (19) is denoted by mij. Proof. Part (i) is trivial, and part (ii) was proved in [8]. For part (iii), fix i 6= j and assume that (17) and (18) hold with k and m minimal, but m 6= k. Suppose that m < k; the case m > k can be treated analogously. Then by multiplying (18) by −sk and adding to (17) we can assume that sk = 0. But then we can apply σ−1 to (17) and get a contradiction to the minimality of m. i To an S-finitistic TGW datum (R, σ, t, µ) of degree n we associate a matrix C = CS(R, σ, t, µ) = (aij)1≤i,j≤n given by aij =(1 − mij, 2, i 6= j i = j where mij was defined in (19). Proposition 6.4. If (R, σ, t, µ) is an S-finitistic TGW datum where S is a Zn-invariant subalgebra of R and ti ∈ Rreg for each i ∈ n, then CS(R, σ, t, µ) is a generalized Cartan matrix. Proof. Since ti ∈ Rreg for each i ∈ n we have mij ≥ 1 for all i 6= j. Hence aij ≤ 0 for all i 6= j. Suppose that aij = 0 for some i 6= j. This means that mij = 1 and hence σi(tj) = stj and tj = s′σi(tj) for some s, s′ ∈ S. We may combine these two relations with the regularity of tj to conclude that s′s = 1. By [6, Corollary 2.17] the regularity of the ti's implies that consistency relations (9) hold. The Zn-invariance of S in combination with relation (9a) and σi(tj) = stj and tj = s′σi(tj) yields mji = 1, i.e. aji = 0. Remark 6.5. The algebras Tq(C) associated to symmetric generalized Cartan matrices C (see Section 5.2) are k-finitistic and have the property that their respective generalized Cartan matrices, as defined above, is precisely C. This was the main point of their construction in [8]. 19 6 Ore localizations and a finiteness condition for TGWAs Recall that a regular left (right) Ore set S in an algebra A is a multiplicatively closed subset consisting of regular non-zero elements and containing 1 such that Sa ∩ As 6= ∅ (aS ∩sA 6= ∅) for all a ∈ A, s ∈ S. We now prove a theorem which connects the property of a TGWA A being R-finitistic to certain natural subsets of A being Ore sets. Theorem 6.6. Let A = A(R, σ, t, µ) be a regularly graded TGWA. For i ∈ n, define the following subsets of A: Xi := {X k i k ∈ Z≥0}, Yi := {Y k i k ∈ Z≥0}. (20) Then the following three assertions are equivalent: (i) (R, σ, t, µ) is left (right) R-finitistic; (ii) Xi is a regular left (right) Ore set in A for each i ∈ n; (iii) Yi is a regular left (right) Ore set in A for each i ∈ n. Proof. We consider the right-sided case and prove that (ii) is equivalent to (i). The other cases are analogous. Let i ∈ n be arbitrary. It suffices to prove the equivalence of the following two assertion: (a) Xi is a right regular Ore set in A; (b) For all j 6= i, there exist some s′ 1, . . . , s′ m ∈ R such that (18) holds. If Xi is invertible in A, then its inverse has to have degree −ei, hence be of the form rYi for some r ∈ R, which implies that ti is invertible. In this case assertions (a) and (b) are both easily seen to hold. Assume that Xi is not invertible. By definition, assertion (a) holds if and only if aXi ∩ sA 6= ∅, Since A is generated by R, Xj, Yj, for j ∈ n, and since rXi = Xiσ−1 YjXi = Xi · µ−1 if ij Yj for i 6= j and YiX 2 i = Xiσ−1 i ∀a ∈ A, ∀s ∈ Xi. (21) (r) for r ∈ R, and (ti), it follows that (21) holds if and only i (22) Since A is graded and Xi is not invertible, an equality XjX m i = Xia for some a ∈ A implies that m > 0 and a is homogeneous of degree (m − 1)ei + ej. For any m ∈ Z≥0 we ∀j 6= i, ∃m ∈ Z≥0 such that XjX m i ∈ XiA. have Amei+ej =Pm k=0 RX k i XjX m−k i and thus (22) is equivalent to m ∀j 6= i, ∃m ∈ Z>0, r1, . . . , rm ∈ R such that XjX m i = rkX k i XjX m−k i . (23) By Theorem 4.3(iv), the identity in (23) is equivalent to (putting r0 = −1): Xk=1 0 = rkX k i XjX m−k i YjY m i = m Xk=0 =(cid:16) Xk=0 m rkµm−k ij σk i σj(tj)(cid:17)X m i Y m i 20 7 Simplicity theorems for TGWAs Since X m to that assertion (b) holds. i = σm i Y m i (ti) · · · σi(ti) is a regular element of R it follows that (23) is equivalent Corollary 6.7. If A = A(R, σ, t, µ) is a regularly graded and R-finitistic TGWA, then the multiplicative monoid X in A generated by X1, . . . , Xn is a regular Ore set in A. Proof. Straightforward. 7 Simplicity theorems for TGWAs In this section we provide a description of when TGWAs are simple. 7.1 A weak simplicity result for general TGWAs Theorem 7.1. Let A = A(R, σ, t, µ) be a TGWA such that R is Zn-simple and maximal commutative in A. If J is a non-zero proper ideal of A, then any prime ideal of R containing J ∩ R, contains an element of the form σg(ti) for some i ∈ n and g ∈ Zn. Proof. Let J be a non-zero ideal of A. According to Theorem 3.6, J ∩ R is a nonzero ideal of R. Suppose that P is a prime ideal of R containing J ∩ R. If P would contain σg(J ∩ R) for all g ∈ Zn then P would also contain the sum J = Pg∈Zn σg(J ∩ R). Then J would be a non-zero proper Zn-invariant ideal of R, but this contradicts the Zn-simplicity of R. Thus there exists a g ∈ Zn such that σg(J ∩ R) is not contained in P . However, J ∩ R contains Ag(J ∩ R)A−g = σg(J ∩ R)AgA−g. Since P is a prime ideal we conclude that AgA−g ⊆ P . Choosing two monic monomials a ∈ Ag and b ∈ A−g, the product ab can be written as a product of elements of the form σh(ti), where h ∈ Zn and i ∈ n. Since P is prime it must contain at least one such factor, which proves the theorem. Corollary 7.2. Let A = A(R, σ, t, µ) be a TGWA. Consider the following countable union of Zariski-closed sets in Spec(R): S = [g∈Zn i∈n V(cid:0)σg(ti)(cid:1). If R is Zn-simple and maximal commutative in A, then S contains SJ ⊳A V (J ∩ R), i.e. the union of varieties of the ideals of the form J ∩ R where J ranges over the ideals of A. Remark 7.3. Corollary 7.2 can be interpreted as saying that there are "few" proper ideals of A, in other words A is "close" to being simple. In particular, if R is Zn-simple and maximal commutative in A and in addition ti is invertible for each i ∈ n, then A is simple. This will be made more precise in Corollary 7.14. 21 7 Simplicity theorems for TGWAs 7.2 On the structure of the localized algebra X −1A Let A = A(R, σ, t, µ) be a regularly graded TGWA which is R-finitistic. By Corollary 6.7, the multiplicative submonoid X of A, generated by X1, . . . , Xn, is an Ore set in A. Let B = X −1A. Since X consists of regular elements in A, the canonical map A → B is injective and we can, and will henceforth, regard A as a subalgebra of B. In this section we shall prove a key result about the algebra B (Theorem 7.8) which will later allow us to deduce the simplicity criterion for A in Section 7.3. Remark 7.4. The algebra B can be embedded into another localization of A which is a Zn-crossed product. Let T be the multiplicative submonoid of R generated by the set {σg(ti) g ∈ Zn, i ∈ n}. By [6, Theorem 2.15], T is an Ore set in A and the localization T −1A is a crossed product, isomorphic to the TGWA A(T −1R, σ, t, µ), where σ is uniquely extended to a Zn-action on T −1R. There is a k-algebra monomorphism τ : B → T −1A given by τ (Xi) = Xi, τ (X −1 i Yi and τ (r) = r for all r ∈ R and i ∈ n. ) = t−1 i Before we can continue we need to prove two lemmas, which will be of vital importance in the proof of our main results. Lemma 7.5. Let A = A(R, σ, t, µ) be a regularly graded and R-finitistic TGWA and put B = X −1A. The following assertions hold: (i) For any non-zero element b ∈ B, there exists an r ∈ Rreg such that rb ∈ A; (ii) Rreg ⊆ Breg; (iii) Z(A) ⊆ Z(B); (iv) If AxA = A for all x ∈ X , then Z(A) = Z(B). Proof. (i) Any element b ∈ B can be written as b = X −1 i1 YiXi = ti so tiX −1 element r = (σi1 ◦ · · · ◦ σik−1)−1(tik ) · · · σ−1 of A. a where a ∈ A. But i = Yi for i ∈ n. Therefore, multiplying b from the left by the regular i1 (ti2)ti1 we get Yi1 · · · Yik a, which is an element · · · X −1 ik (ii) Let r ∈ Rreg and b ∈ B\{0} be arbitrary. Seeking for a contradiction, assume that rb = 0. Without loss of generality we may assume that b is homogeneous. Since B = X −1A, there is an x ∈ X with 0 6= xb ∈ A. By Corollay 3.3 there is a homogeneous c ∈ A such that xbc ∈ R \ {0}. Since r ∈ Rreg we get σg(r)xbc 6= 0, where g = deg(x). However, σg(r)xbc = xrbc = 0, which is a contradiction. This shows that r ∈ Breg. (iii) Let a ∈ Z(A) and b ∈ B. Write b = x−1a1, where x ∈ X and a1 ∈ A. Then ax = xa and aa1 = a1a imply ab = ba. Thus a ∈ Z(B). (iv) Let b ∈ Z(B). Since A ⊆ B it is enough to show that b ∈ A. Since B = X −1A, we have b = x−1a for some x ∈ X , a ∈ A. By the assumption 1 = Pi cixdi for some ci, di ∈ A. Now note that, since b commutes with any element of A, A ∋Xi ciadi =Xi cixbdi =Xi cixdib = b. 22 7 Simplicity theorems for TGWAs This proves that b ∈ A. Hence b ∈ Z(A), so Z(B) ⊆ Z(A). The converse inclusion was shown in (iii). Remark 7.6. Under the same assumptions as in Lemma 7.5 one may also prove the following statements: (i) For any non-zero element b ∈ Z(B), there is a regular element r ∈ R such that rb ∈ Z(CA(R)); (ii) B0 is commutative; (iii) CA(B0) = CA(R); (iv) If R is Zn-simple, then Z(B) ∩ B0 ⊆ R. However, this paper makes no use of these facts, and therefore we omit the proof. The second lemma that we need is a technical step used in the proof of Theorem 7.8 below. Note that the Zn-gradation on A can be extended to a Zn-gradation on B by putting deg(X −1 ) = −ei for each i ∈ n. i Lemma 7.7. Let A = A(R, σ, t, µ) be a regularly graded and R-finitistic TGWA such that R is Zn-simple, and put B = X −1A. For any non-zero b ∈ B there exists an element b′ ∈ B with the following properties: (i) b′ ∈ BbB; (ii) (b′)0 = 1, where (b′)0 is the degree zero component of b′ with respect to the Zn- gradation on B; (iii) Supp(b′) ≤ Supp(b). Proof. Let b ∈ B be non-zero. Then b =Pg∈Zn bg where bg ∈ Bg, for g ∈ Zn and we may choose some h ∈ Zn such that bh 6= 0. By Theorem 4.3(iv), we know that bhc 6= 0 for any monic monomial c of degree −h. Thus, by replacing b by bc we can without loss of generality assume that b0 6= 0. By Lemma 7.5(i), there is an r ∈ Rreg such that rb ∈ A. By Lemma 7.5(ii) the element rb0 is non-zero. The set J =(cid:8)s ∈ R s + Xg∈Supp(b)\{0} cg ∈ BbB for some cg ∈ Bg(cid:9) contains the non-zero element rb0 (take cg = rbg for g ∈ Zn) and hence J is a non-zero ideal of R. We shall now show that J is Zn-invariant. If s ∈ J, then s+Pg∈Supp(b)\{0} cg ∈ BbB for some cg ∈ Bg. So for any i ∈ n, BbB ∋ Xi(s + Xg∈Supp(b)\{0} cg)X −1 i = σi(s) + Xg∈Supp(b)\{0} XicgX −1 i which shows that σi(s) ∈ J. Similarly σ−1 (s) ∈ J. Thus J is Zn-invariant. Since R is assumed to be Zn-simple we deduce that J = R. Hence 1 ∈ J, which means that there i are cg ∈ Bg such that b′ := 1 +Pg∈Supp(b)\{0} cg ∈ BbB. This b′ satisfies the required properties. Now we come to the main result about the algebra B which in particular implies that the center Z(B) of B is an essential subalgebra of B, in the sense of Defininition 3.1. 23 7 Simplicity theorems for TGWAs Theorem 7.8. Let A = A(R, σ, t, µ) be a regularly graded and R-finitistic TGWA such that R is Zn-simple, and put B = X −1A. Every non-zero ideal of B has non-empty in- tersection with Z(B)∩(cid:0)1+Pg∈Zn\{0} Bg(cid:1). In particular, Z(B) is an essential subalgebra of B. Proof. Let J ⊆ B be a non-zero ideal. Among all non-zero elements of J, let b ∈ J be one such that Supp(b) is as small as possible. By Lemma 7.7 we can construct an element b′ with b′ ∈ BbB ⊆ J, (b′)0 = 1, and Supp(b′) ≤ Supp(b). In fact, by minimality of Supp(b) among all non-zero elements of J, we have Supp(b′) = Supp(b). Let r ∈ R be arbitrary. Then (b′r − rb′)0 = r − r = 0 so Supp(b′r − rb′) < Supp(b′). By minimality of Supp(b′) and that b′r − rb′ ∈ J we must have b′r − rb′ = 0. Let i ∈ n. Then (Xib′X −1 i − b′ ∈ J so by minimality of Supp(b′) we conclude that Xib′X −1 i − b′ = 0. So Xib′ = b′Xi for all i ∈ n. Since B is generated as a ring by the elements of R and Xi, X −1 , for i ∈ n, we conclude that i − b′)0 = 1 − 1 = 0. Again Xib′X −1 b′ ∈ J ∩ Z(B) ∩(cid:0)1 +Pg∈Zn\{0} Bg(cid:1). 7.3 Simplicity of finitistic TGWAs i In this section, let A = A(R, σ, t, µ) be a regularly graded and R-finitistic TGWA. Recall that X is the multiplicative submonoid in A generated by X1, . . . , Xn, which by Theorem 6.6 is a regular Ore set in A, and put B = X −1A. Our first theorem in this section reduces the question of the simplicity of a TGWA to the simplicity of the localized algebra B. This method was inspired by the ideas of Jordan [9, Theorem 6.1], [10, Theorem 3.2]. Theorem 7.9. Let A = A(R, σ, t, µ) be a regularly graded and R-finitistic TGWA. Then A is simple if and only if the following two assertions hold: (i) B is simple; (ii) AxA = A for all x ∈ X . Proof. Suppose that A is simple. Let J be any non-zero ideal of B and let 0 6= b ∈ J. Since B = X −1A there is an x ∈ X such that xb ∈ A. Thus 0 6= xb ∈ A ∩ J so A ∩ J is a non-zero ideal of A, thus 1 ∈ J since A is simple. This shows that (i) holds. Since A is regularly graded, each element x ∈ X is non-zero by Theorem 4.3. So, since A is simple, (ii) must hold. For the converse, assume that condition (ii) holds. Then we prove something slightly stronger than that (i) implies A is simple. Namely, we show that whenever J is an ideal of A such that BJB = B, then J = A. So suppose J ⊆ A is an ideal with BJB = B. Using that X is an Ore set in A, it is straightforward to show that BJB =(cid:8)x−1a x ∈ X , a ∈ J(cid:9). Since BJB = B, we have 1 = x−1a for some x ∈ X , a ∈ J. Then x = a ∈ J. By condition (ii), A = AxA = AaA ⊆ J. Thus J = A. In what follows, it is useful to keep the following facts in mind. Note that the assump- tion of R-finitisticity is not required here. 24 7 Simplicity theorems for TGWAs Lemma 7.10. Let A = A(R, σ, t, µ) be a regularly graded TGWA. Consider the following assertions: (i) Z(A) is a field; (ii) Z(A) ⊆ R; (iii) Z(A) = RZn :=(cid:8)r ∈ R σg(r) = r, ∀g ∈ Zn(cid:9). Then (i)⇒(ii)⇒(iii). If R is Zn-simple, then all three assertions are equivalent. Proof. (i)⇒(ii): Suppose that Z(A)g 6= {0} for some g ∈ Zn \ {0} and let 0 6= a ∈ Z(A)g. Then 1 + a ∈ Z(A) and hence it is invertible. Using that Zn is an orderable group, fix an ordering < on Zn. Without loss of generality we may assume that g > 0. Let b be the inverse of 1 + a. Write b = bh1 + · · · + bhk where 0 6= bhi ∈ Bhi, hi ∈ Zn, for all i ∈ {1, . . . , k} and h1 < · · · < hk. In the product (1 + a)b, the term of lowest degree is 1bh1 and the term of highest degree is abhk which is non-zero since a is invertible. On the other hand, (1 + a)b = 1. Thus k = 1 and bh1 = abh1 = 1 which contradicts that g + h1 > h1. Hence Z(A) ⊆ R. (ii)⇒(iii): Let r ∈ Z(A). Then rXi = Xir for any i ∈ n. Equivalently, (r − σi(r))Xi = 0 for any i ∈ n. By Theorem 4.3(iii) we get σi(r) = r for all i ∈ n. Thus σg(r) = r for each g ∈ Zn. The converse inclusion is straightforward. Assume that R is Zn-simple. It is enough to prove that (iii)⇒(i). Let r ∈ Z(A) be non-zero. Then, by (iii), Rr is a non-zero Zn-invariant ideal of R, hence it contains 1 and thus r is invertible. The following result holds for an arbitrary TGWA. Lemma 7.11. Let A = A(R, σ, t, µ) be a TGWA. If A is simple, then R is Zn-simple. Proof. Suppose that J is a proper Zn-invariant ideal of R. Then, since R = A0, (AJA) ∩ R = Xg∈Zn AgJA−g = Xg∈Zn σg(J)AgA−g ⊆ JR ⊆ J. Hence AJA is a proper ideal of A. Since A is simple it follows that AJA = {0}, i.e. J = {0}. This proves that R is Zn-simple. The next result is the main theorem of this section and provides necessary and sufficient conditions for an R-finitistic TGWA to be simple. Theorem 7.12. Let A = A(R, σ, t, µ) be a regularly graded and R-finitistic TGWA. Then A is simple if and only if the following three assertions hold: (i) AxA = A for all x ∈ X ; (ii) R is Zn-simple; (iii) Z(A) ⊆ R. 25 7 Simplicity theorems for TGWAs Proof. Suppose that A is simple. By Theorem 7.9, condition (i) holds. Lemma 7.11 shows that (ii) holds. Simplicity of A also implies that Z(A) is a field. Hence (iii) follows from Lemma 7.10. For the converse, assume that (i) -- (iii) hold. By (i) and Theorem 7.9 it is enough to prove that B is simple. Let J be any non-zero ideal of B. Since (ii) holds, Theorem 7.8 applies to give J ∩ Z(B) ∩(cid:0)1 +Pg∈Zn\{0} Bg(cid:1) 6= ∅. By assumptions (i) and (iii), Lemma 7.5 (iv) implies that Z(B) ⊆ R. Hence Z(B) ∩(cid:0)1 +Pg∈Zn\{0} Bg(cid:1) = {1}. Thus 1 ∈ J and therefore J = B. Remark 7.13. In some cases assertion (i) can be made more explicit, see Section 7.4. Corollary 7.14. If A = A(R, σ, t, µ) is a regularly graded and R-finitistic TGWA where R is Zn-simple and maximal commutative in A, then A is simple if and only if AxA = A for all x ∈ X . Remark 7.15. Corollary 7.14 can be compared to the "generic" result that we proved earlier in Corollary 7.2, where it was shown that if R is Zn-simple and maximal commuta- tive in A, then in some sense A has "few" ideals. Corollary 7.14 explains this somewhat, in that it shows that the only ideals in A which obstruct the simplicity when R is Zn-simple and maximal commutative in A are those of the form AxA for x ∈ X . See Example 8.2 for a TGWA A = A(R, σ, t, µ) where R is Zn-simple and maximal commutative in A, but where AX1A is a proper non-zero ideal, and hence A is not simple. 7.4 Simplicity of TGWAs of Lie type A1 × · · · × A1 Definition 7.16. If A(R, σ, t, µ) is a regularly graded and R-finitistic TGWA, whose associated generalized Cartan matrix is of a certain type X then A(R, σ, t, µ) is said to be of Lie type X. Let A = A(R, σ, t, µ) be a regularly graded, R-finitistic TGWA. We assume in this section that A is of Lie type (A1)n = A1 × · · · × A1. This is equivalent to that for all i, j ∈ n there exist invertible rij, sij ∈ R such that XiXj = rijXjXi and YiYj = sijYiYj. This case covers all higher rank generalized Weyl algebras (such that ti ∈ Rreg for all i ∈ n). Indeed, they correspond to the case rij = sij = 1 for all i, j ∈ n. Furthermore, the TGWAs constructed in [15] (certain Mickelson-Zhelobenko algebras and Gelfand-Tsetlin algebras) are of Lie type (A1)n as well as some examples of so called crystalline graded rings. The following result shows how assertion (i) of Theorem 7.12 can be made more explicit in the case of TGWAs of Lie type (A1)n. Lemma 7.17. Let A = A(R, σ, t, µ) be a regularly graded and R-finitistic TGWA of Lie type A1 × · · · × A1. Then the following assertions are equivalent: (i) AxA = A for all x ∈ X ; (ii) Rti + Rσd i (ti) = R, for all i ∈ n and d ∈ Z>0. 26 7 Simplicity theorems for TGWAs Proof. First we show that AxA = A, ∀x ∈ X ⇐⇒ AX d i A = A, ∀i ∈ n, ∀d ∈ Z>0. The implication ⇒ is trivial since X d i A = A for all i, d and let x ∈ X . Since A is of Lie type (A1)n, we can reorder the factors in x (up to multiplicative invertible factors from R) and thus get AxA = Ax′A where x′ = X j1 i ∈ X for all i, d. Conversely, assume that AX d n for some jr ≥ 0. By assumption, AX j1 1 A = A. Thus 1 · · · X jn 1 ∈ (AX j1 1 A) ∩ R = Xg∈Zn AgX j1 1 A−g−j1e1. (24) · · · Z (gn) n . Also Since A is of Lie type (A1)n, for each g ∈ Zn we have Ag = RZ (g1) RZ (k) for all i 6= j and any k, l ∈ Z. Thus, (24) implies that i Z (l) j ⊆ RZ (l) j Z (k) 1 i rkZ (k) 1 X j1 1 Z (−k−j1) 1 1 =Xk∈Z for some rk ∈ R, k ∈ Z, (only finitely many non-zero). We can choose sk ∈ R, k ∈ Z, such that skZ (k) 1 x′Z (−k−j1) 1 Xk∈Z =Xk∈Z rkZ (k) 1 X j1 1 Z (−k−j1) 1 · X j2 2 · · · X jn n = X j2 2 · · · X jn n . This proves that X j2 this procedure for 2, 3, . . . , n yields 1 ∈ AxA, hence A = AxA. n ∈ Ax′A = AxA. Replacing x by X j2 2 · · · X jn 2 · · · X jn n and repeating It remains to show that AX d i (ti) = R, ∀i ∈ n, ∀d ∈ Z>0. Similar to the reasoning above we have for any d > 0 and any i ∈ n, AX d which implies the required equivalence. i A = A, ∀i ∈ n, ∀d ∈ Z>0 is equivalent to Rti + Rσd i A ∩ RX d−1 i Z (−k−1) i + RX d = RYiX d i X d i Yi = (cid:0)Rti + Rσd i (ti)(cid:1)X d−1 = Pk∈Z RZ (k) i i i Lemma 7.17 and Theorem 7.12 immediately imply the following result. Theorem 7.18. Let A = A(R, σ, t, µ) be a regularly graded TGWA which is R-finitistic of Lie type (A1)n. Then A is simple if and only if the following three assertions hold: (i) Rσd i (ti) + Rti = R, for all i ∈ n and d ∈ Z>0; (ii) R is Zn-simple; (iii) Z(A) ⊆ R. The next result deals with the case of higher rank generalized Weyl algebras. Taking n = 1 in the following theorem, we recover the case of [9, Theorem 6.1]. Theorem 7.19. If A = R(σ, t) is a generalized Weyl algebra of rank n, then A is simple if and only if the following four assertions hold: 27 8 Examples (i) ti ∈ Rreg for all i ∈ n; (ii) Rti + Rσd i (ti) = R for i ∈ n and all d ∈ Z>0; (iii) R is Zn-simple; (iv) σ : Zn → Autk(R) is injective. Proof. If A is simple, then all ti's must be regular. Indeed, if rti = 0 for some i ∈ n, where 0 6= r ∈ R, then one can check that the element rYi of A (which is non-zero in the present case of GWAs) generates an ideal which is proper: (ArYiA) ∩ R = {0}. Also, if all ti's are regular, it is well-known that A is isomorphic to the TGWA A(R, σ, t, µ), where µij = 1 for all i, j (see [13, Example 1.2]) and that it is R-finitistic and of Lie type (A1)n. Thus we can apply Theorem 7.18. It only remains to prove that Z(A) ⊆ R if and only if σ is injective. For this, the key point is that if g ∈ K = ker(σ), then Z(A)g = RZnZ (g1) (here it is crucial that A is a GWA). Then it is clear that K = {0} if and only if Z(A) ⊆ R. · · · Z (gn) 1 n Combining Theorem 7.19 and Corollary 5.2 we get the following. Corollary 7.20. If A = R(σ, t) is a simple generalized Weyl algebra of rank n, then R is a maximal commutative subalgebra of A. Remark 7.21. If A = R(σ, t) is a GWA of rank n where ti is non-invertible for each i ∈ n, then condition (iv) of Theorem 7.19 can be removed. Indeed, then σ : Zn → Autk(R) must be injective if Rti + Rσd i (ti) = R for all i ∈ n and all d ∈ Z>0, which follows from the fact that, for GWAs, σi(tj) = tj for i 6= j. 8 Examples The following example shows a simple TGWA where R is not maximal commutative. Example 8.1. Let A = A(R, σ, t, µ) be as in Example 5.8, i.e. n = 2, R = k = C, σ1 = σ2 = idC, t1 = t2 = 1 and µ12 = 2, µ21 = 1 2 . Note that in A we have YiXi = 1 = XiYi for i = 1, 2. We will show that Z(A) ⊆ R. Let a ∈ Z(A) be non-zero. Since Z(A) is a graded subalgebra of A we can assume that a is homogeneous. Let g = (g1, g2) ∈ Z2 1 X g2 be the degree of a. Since X2X1 = 2X1X2 (since Y2 = X −1 2 for some non-zero r ∈ R. Therefore 0 = [a, X1] = (2g2 − 1)rX g1+1 2 which implies g2 = 0. Similarly g1 = 0, which proves that Z(A) ⊆ R. Trivially R = C is Zn-simple and AxA = A for all x ∈ X (since X1 and X2 are invertible), hence Theorem 7.12 implies that A is simple. 2 ) we have a = rX g1 1 X g2 Condition (i) in Theorem 7.12 and condition (ii) in Theorem 7.19 are not superfluous, as the following example shows. Example 8.2. Let n = 1, k = C, R = C[u], σ1(u) = u + 1, t = u(u − 1) and A = R(σ, t). Then R is a maximal commutative subalgebra of A, and R is Z-simple. However, Rt + Rσ1(t) = Ru which is a proper ideal of R. Hence, by Theorem 7.19, A is not simple. 28 8 Examples The construction of the following TGWA is due to A. Sergeev [23, Section 1.5.2], but we consider sl(n + 1, C) instead of gl(n + 1, C). Example 8.3. Let h be the Cartan subalgebra consisting of all traceless diagonal ma- trices in the Lie algebra sl(n + 1, C). For i ∈ {1, . . . , n + 1}, let εi ∈ h∗ be given by εi(diag(λ1, . . . , λn+1)) = λi. For i ∈ n let αi := εi − εi+1 be the simple roots, and let hi ∈ h be such that αi(hj) = δij. Let f1, . . . , fn+1 ∈ C[u] be an arbitrary collection of polynomials in one indeterminate u. Take k = C and let R = SC(h), the symmetric algebra on h. For i ∈ n, define σi ∈ AutC(R) by requiring that σi(h) = h − αi(h), ∀h ∈ h. Let µij = 1 for all i, j and put t1 = f1(h1)f2(h2−h1), t2 = f2(h2−h1+1)f3(h3−h2), . . . , tn = fn(hn−hn−1+1)fn+1(hn). One can verify that the consistency relations (9) hold. Let S(f1, . . . , fn+1) = A(R, σ, t, µ) be the corresponding TGWA. One can show that it is R-finitistic. We will use Theorem 7.12 to prove the following result. Proposition 8.4. S(1, u, 1) is simple. Proof. Regard R as a polynomial algebra over C in variables h1, h2. We show that R is Z2-simple. Let J be any non-zero Z2-invariant ideal of R. Among all non-zero elements f of J, choose one whose h1-degree is minimal. If it is positive, then f − σ1(f ) has smaller degree, hence f − σ1(f ) = 0 (otherwise f − σ1(f ) would be a non-zero element of J since J is Z2-invariant, which would contradict the choice of f ). If the h1-degree of f is zero then we get σ1(f ) = f anyway. Next, among all non-zero elements f of J with σ1(f ) = f , choose one with minimal h2-degree. As above, σ2(f ) = f . That is, f ∈ RZ2, the subalgebra of invariants. However, RZ2 = C. This proves that J contains a non-zero constant, so J = R. Hence R is Z2-simple. Now, note that we have Y1X1 = t1 = h2 − h1 and X1Y1 = σ1(t1) = h2 − h1 + 1, hence [−Y1, X1] = 1. Similarly we get [Y2, X2] = 1. Since µij = 1 for all i, j, we have [Yj, Xi] = 0 if i 6= j. Thus, if x ∈ X has degree g = (g1, g2) ∈ Z>0 × Z>0 then we have 1 = 1 g1!g2! (ad −Y1)g1(cid:0)(ad Y2)g2(x)(cid:1) (25) where (ad a)(b) = [a, b] = ab − ba. Thus AxA = A for each x ∈ X , where A = S(1, u, 1). For any g = (g1, g2) ∈ Z2 we have σg(hi) = hi − gi for i = 1, 2. Thus it is clear that σ : Z2 → Autk(R) is injective. By Theorem 5.1, CA(R) = R. Since we always have Z(A) ⊆ CA(R), we get Z(A) ⊆ R. Hence, by Theorem 7.12, A is simple. Example 8.5. The construction of the following TGWA was first mentioned in [13, Example 1.3] and has been generalized to higher rank cases in [8, 23]. Let n = 2, R = k[H], σ1(H) = H + 1, σ2(H) = H − 1, t1 = H, t2 = H + 1, µ12 = µ21 = 1 and let A = A(R, σ, t, µ) be the associated TGWA. One may verify that the consistency relations 29 9 Comparison between TGWAs and strongly graded rings (9) hold and that A is k-finitistic of type A2. It was shown in [8, Example 6.3] that A is isomorphic to the k-algebra with generators X1, X2, Y1, Y2, H and defining relations X1H = (H + 1)X1, Y1H = (H − 1)Y1, Y1X1 = X2Y2 = H, X1Y2 = Y2X1, X2H = (H − 1)X2, Y2H = (H + 1)Y2, Y2X2 = X1Y1 = H + 1, X2Y1 = Y1X2, 1 = 0, 2 = 0, 1 = 0, 2 = 0. Using these relations it is easy to check that 0 6= X1X2 − X2X1 ∈ Z(A). Thus, by Theorem 7.12, A is not simple. 1 X2 − 2X1X2X1 + X2X 2 2 X1 − 2X2X1X2 + X1X 2 Y 2 1 Y2 − 2Y1Y2Y1 + Y2Y 2 Y 2 2 Y1 − 2Y2Y1Y2 + Y1Y 2 X 2 X 2 9 Comparison between TGWAs and strongly graded rings In [16, Theorem 3.5] it was proved that if S is a skew group ring with commutative neutral component Se, then Se is maximal commutative if and only if Se ∩I 6= {0} for all non-zero ideals I ⊆ S (i.e. Se is an essential subalgebra of S). For a TGWA A = A(R, σ, t, µ), maximal commutativity of R(= A0) in A implies that R is an essential subalgebra of A, by Theorem 3.6. However, the converse is not true because of Example 8.1 which is a simple TGWA (hence R trivially is essential) but R is not maximal commutative in A. Furthermore, in [16, Theorem 6.13] it was shown that if S is a G-graded skew group ring with commutative neutral component Se, then S is simple if and only if Se is G- simple and maximal commutative in S. For a TGWA A = A(R, σ, t, µ), simplicity of A implies that R is Zn-simple but does not imply that R is maximal commutative in A (Example 8.1 again). The converse need not hold either, by Example 8.2. Another result, [16, Theorem 6.6], states that if S is a strongly G-graded ring such that the neutral component Se is maximal commutative in S, then S is simple if and only if Se is G-simple. For a TGWA A = A(R, σ, t, µ), if R is Zn-simple and maximal commutative in A then still A need not be simple, see Example 8.2. However, by Lemma 7.11, if A(R, σ, t, µ) is a simple TGWA (even without R being maximal commutative) then R is Zn-simple. Finally, [16, Proposition 6.5] shows that if S is a strongly G-graded ring with commu- tative Se such that CS(Se) is G-simple, then S is simple. For TGWAs, the condition that CA(R) is Zn-simple could be interpreted as saying that CA(R) has no non-zero proper ideals J such that XiJ = JXi and YiJ = JYi for all i. In any case this need not hold for TGWAs because of Example 8.2. References [1] Bavula, V. V., Generalized Weyl algebras and their representations (Russian), Al- gebra i Analiz 4 (1992), no. 1, 75 -- 97; English transl. in St Petersburg Math. J. 4 (1993) 71 -- 92. [2] Cassidy, T., Shelton, B., Basic properties of generalized down-up algebras, J. Algebra 279 (2004), no. 1, 402 -- 421. 30 References [3] Cohen, M., Montgomery, S., Group-Graded Rings, Smash Products and Group Ac- tions, Trans. Amer. Math. Soc., 282, no. 1, (1984), 237 -- 258. [4] Cohen, M., Rowen, L. H., Group graded rings, Comm. Algebra 11 (1983), no. 11, 1253 -- 1270. [5] Drozd, Yu. A., Guzner, B. L., Ovsienko, S. A., Weight modules over generalized Weyl algebras, J. Algebra 184 (1996), 491 -- 504. [6] Futorny, V., Hartwig, J. T., Twisted Weyl algebras, crossed products, and represen- tations, arXiv:1005.0341v1 [math.RA], Preprint (2010). [7] Hartwig, J. T., Locally Finite Simple Weight Modules over Twisted Generalized Weyl Algebras, J. Algebra 303 (2006), no. 1, 42 -- 76. [8] Hartwig, J. T., Twisted Generalized Weyl Algebras, Polynomial Cartan Matrices and Serre-Type Relations. To appear in Comm. Algebra. Available as arXiv:0908.2054v1 [math.RA], (2009). [9] Jordan, D. A., Primitivity in Skew Laurent Polynomial Rings and Related Rings, Math. Z. 213 (1993), no. 3, 353 -- 371. [10] Jordan, D. A., A Simple Localization of The Quantized Weyl Algebra, J. Algebra 174 (1995), no. 1, 267 -- 281. [11] Nauwelaerts, E., Van Oystaeyen, F., Introducing Crystalline Graded Algebras, Al- gebr. Represent. Theory 11 (2008), no. 2, 133 -- 148. [12] Maltsiniotis, G., Groupes quantiques et structures différentielles, C. R. Acad. Sci. Paris Sér. I Math. 311 No. 12 (1990) 831 -- 834. [13] Mazorchuk, V., Turowska, L., Simple weight modules over twisted generalized Weyl algebras, Comm. Algebra 27 (1999), no. 6, 2613 -- 2625. [14] Mazorchuk, V., Turowska, L., ∗-representations of twisted generalized Weyl construc- tions, Algebr. Represent. Theory 5 (2002), no. 2, 163 -- 186. [15] Mazorchuk, V., Ponomarenko, M., Turowska, L, Some Associative Algebras Related to U (g) and Twisted Generalized Weyl Algebras, Math. Scand. 92 (2003), no. 1, 5 -- 30. [16] Öinert, J., Simple group graded rings and maximal commutativity, Operator Struc- tures and Dynamical Systems (Leiden, NL, 2008), 159 -- 175, Contemp. Math. 503, Amer. Math. Soc., Providence, RI, (2009). [17] Öinert, J., Lundström, P., Commutativity and Ideals in Category Crossed Products. To appear in the Proceedings of the Estonian Academy of Sciences. [18] Öinert, J., Lundström, P., The Ideal Intersection Property for Groupoid Graded Rings, arXiv:1001.0303v3 [math.RA], Preprint (2010). 31 References [19] Öinert, J., Silvestrov, S. D., Commutativity and Ideals in Algebraic Crossed Products, J. Gen. Lie T. Appl. 2 (2008), no. 4, 287 -- 302. [20] Öinert, J., Silvestrov, S. D., Commutativity and Ideals in Pre-Crystalline Graded Rings, Acta Appl. Math. 108 (2009), no. 3, 603 -- 615. [21] Öinert, J., Silvestrov, S. D., Theohari-Apostolidi, T., Vavatsoulas, H., Commuta- tivity and Ideals in Strongly Graded Rings, Acta Appl. Math. 108 (2009), no. 3, 585 -- 602. [22] Passman, D. S., Cancellative group-graded rings, Methods in ring theory (Antwerp, 1983), 403 -- 414, NATO Adv. Sci. Inst. Ser.C Math. Phys. Sci., 129, Reidel, Dor- drecht, 1984. [23] Sergeev, A., Enveloping Algebra U (gl(3)) and Orthogonal Polynomials in Sev- eral Discrete Indeterminates in "Noncommutative structures in mathematics and physics", Duplij S., Wess J. (eds.), Proc. NATO Advanced Reserch Workshop, Kiev, 2000. Kluwer, 2001, 113 -- 124. Also available at arXiv:math/0202182v1 [math.RT]. [24] Svensson, C., Crossed Product Structures Associated with Topological Dynamical Systems, PhD Thesis, Lund University, 2009. 32
1305.3906	1	1305	2013-05-16T19:57:03	Algebraic structures of tropical mathematics	[ "math.RA", "math.AC" ]	Tropical mathematics often is defined over an ordered cancellative monoid $\tM$, usually taken to be $(\RR, +)$ or $(\QQ, +)$. Although a rich theory has arisen from this viewpoint, cf. [L1], idempotent semirings possess a restricted algebraic structure theory, and also do not reflect certain valuation-theoretic properties, thereby forcing researchers to rely often on combinatoric techniques. In this paper we describe an alternative structure, more compatible with valuation theory, studied by the authors over the past few years, that permits fuller use of algebraic theory especially in understanding the underlying tropical geometry. The idempotent max-plus algebra $A$ of an ordered monoid $\tM$ is replaced by $R: = L\times \tM$, where $L$ is a given indexing semiring (not necessarily with 0). In this case we say $R$ layered by $L$. When $L$ is trivial, i.e, $L=\{1\}$, $R$ is the usual bipotent max-plus algebra. When $L=\{1,\infty\}$ we recover the "standard" supertropical structure with its "ghost" layer. When $L = \NN $ we can describe multiple roots of polynomials via a "layering function" $s: R \to L$. Likewise, one can define the layering $s: R^{(n)} \to L^{(n)}$ componentwise; vectors $v_1, \dots, v_m$ are called tropically dependent if each component of some nontrivial linear combination $\sum \a_i v_i$ is a ghost, for "tangible" $\a_i \in R$. Then an $n\times n$ matrix has tropically dependent rows iff its permanent is a ghost. We explain how supertropical algebras, and more generally layered algebras, provide a robust algebraic foundation for tropical linear algebra, in which many classical tools are available. In the process, we provide some new results concerning the rank of d-independent sets (such as the fact that they are semi-additive),put them in the context of supertropical bilinear forms, and lay the matrix theory in the framework of identities of semirings.	math.RA	math	ALGEBRAIC STRUCTURES OF TROPICAL MATHEMATICS ZUR IZHAKIAN, MANFRED KNEBUSCH, AND LOUIS ROWEN Abstract. Tropical mathematics often is defined over an ordered cancellative monoid M, usually taken to be (R, +) or (Q , +). Although a rich theory has arisen from this viewpoint, cf. [L1], idempotent semirings possess a restricted algebraic structure theory, and also do not reflect certain valuation-theoretic properties, thereby forcing researchers to rely often on combinatoric techniques. In this paper we describe an alternative structure, more compatible with valuation theory, studied by the authors over the past few years, that permits fuller use of algebraic theory especially in understanding the underlying tropical geometry. The idempotent max-plus algebra A of an ordered monoid M is replaced by R := L ×M, where L is a given indexing semiring (not necessarily with 0). In this case we say R layered by L. When L is trivial, i.e, L = {1}, R is the usual bipotent max-plus algebra. When L = {1, ∞} we recover the "standard" supertropical structure with its "ghost" layer. When L = N we can describe multiple roots of polynomials via a "layering function" s : R → L. Likewise, one can define the layering s : R(n) → L(n) componentwise; vectors v1, . . . , vm are called tropically dependent if each component of some nontrivial linear combination P αivi is a ghost, for "tangible" αi ∈ R. Then an n × n matrix has tropically dependent rows iff its permanent is a ghost. We explain how supertropical algebras, and more generally layered algebras, provide a robust algebraic foundation for tropical linear algebra, in which many classical tools are available. In the process, we provide some new results concerning the rank of d-independent sets (such as the fact that they are semi-additive), put them in the context of supertropical bilinear forms, and lay the matrix theory in the framework of identities of semirings. 3 1 0 2 y a M 6 1 ] . A R h t a m [ 1 v 6 0 9 3 . 5 0 3 1 : v i X r a 1. Introduction Tropical geometry, a rapidly growing area expounded for example in [Gat, ItMS, L1, MS, SS], has been based on two main approaches. The most direct passage to tropical mathematics is via logarithms. But valuation theory has richer algebraic applications (for example providing a quick proof of Kapranov's theo- rem), and much of tropical geometry is based on valuations on Puiseux series. The structures listed above are compatible with valuations, and in §2.4 we see how valuations fit in with this approach. In his overview, Litvinov [L2] describes tropicalization as a process of dequantization. Thus, one is motivated to develop the algebraic tools at the tropical level, in order to provide an intrinsic theory to support tropical geometry and linear algebra. The main mathematical structure of tropical geometry is the max-plus algebra, which is viewed algebraically as an ordered monoid. Considerable recent activity [CHWW, W] concerns geometry over monoids, but the ordering provides extra structure which enables us to draw on classical algebraic structure theory. The max-plus algebra is fine for answering many combinatoric questions, but it turns out that a more sophisticated structure is needed to understand the algebraic structure connected with valuations. Our overlying objective is to translate ordered monoids into an algebraic theory supporting tropical linear algebra and geometry, using the following approaches: • Algebraic geometry as espoused by Zariski and Grothendieck, using varieties and commutative al- gebra in the context of category theory. • Linear algebra via tropical dependence, the characteristic polynomial, and (generalized) eigenspaces. • Algebraic formulations for more sophisticated concepts such as resultants, discriminants, and Jaco- bians. Date: November 6, 2017. 2010 Mathematics Subject Classification. Primary 06F20, 11C08, 12K10, 14T05, 14T99, 16Y60; Secondary 06F25, 16D25. Key words and phrases. tropical algebra, layered supertropical domains, polynomial semiring, d-base, s-base, bilinear form. The research of the first and third authors was supported by the Israel Science Foundation (grant No. 448/09). The research of the first author also was conducted under the auspices of the Oberwolfach Leibniz Fellows Programme (OWLF), Mathematisches Forschungsinstitut Oberwolfach, Germany. 1 2 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN This approach leads to the use of polynomials and matrices, which requires two operations. Our task has been to pinpoint the appropriate category of semirings in which to work, or equivalently, how far do we dequantize in the process of tropicalization? In this survey we compare four structures, listed in increasing level of refinement: • The max-plus algebra, • Supertropical algebra, • Layered tropical algebras, • Exploded supertropical algebras. We review the layered algebra in §3, compare it to the max-plus algebra, and then in §4 survey its linear algebraic theory, especially in terms of different notions of bases, proving a new result (Proposition 4.25) about the semi-additivity of the rank of d-independent sets of a layered vector space. In §5 we see how these considerations lead naturally to a theory of identities. Due to lack of space, we often refer the reader to [IzKR4, IzKR5] for more details. 2. Algebraic Background We start by reviewing some notions which may be familiar, but are needed extensively in our exposition. The basic tropicalization, or dequantization, involves taking logarithms to (R, + ), which as explained in [L1] replaces conventional multiplication by addition, and conventional addition by the maximum. This is called the max-plus algebra of (R, + ). 2.1. Ordered groups and monoids. Recall that a monoid (M,· , 1) is a set with an associative operation · and a unit element 1. We usually work with Abelian monoids, in which the operation is commutative. The passage to the max-plus algebra in tropical mathematics can be viewed algebraically via ordered groups (such as (R, + )), and, more generally, ordered monoids. An Abelian monoid M := (M,· , 1) is cancellative if ab = ac implies b = c. There is a well-known localization procedure with respect to a submonoid S of a cancellative Abelian monoid M, obtained by taking M × S/ ∼ , where ∼ is the equivalence relation given by (a, s) ∼ (a′, s′) iff as′ = a′s. Localizing with respect to all of M yields its group of fractions, cf. [Bo, W]. We say that a monoid M is power- cancellative (called torsion-free by [W]) if an = bn for some n ∈ N implies a = b. A monoid M is called N-divisible (also called radicalizible in the tropical literature) if for each a ∈ M and m ∈ N there is b ∈ M such that bm = a. For example, (Q, + ) is N-divisible. Remark 2.1. The customary way of embedding an Abelian monoid M into an N-divisible monoid, is to adjoin m√a for each a ∈ M and m ∈ N, and define m√a n√b := mn√anbm. This will be power-cancellative if M is power-cancellative. An ordered Abelian monoid is an Abelian monoid endowed with a total order satisfying the property: a ≤ b implies ga ≤ gb, (2.1) for all elements a, b, g. Any ordered cancellative Abelian monoid is infinite. One advantage of working with ordered monoids and groups is that their elementary theory is well-known to model theorists. The theory of ordered N-divisible Abelian groups is model complete, cf. [M, p. 116] and [Sa, pp. 35, 36], which essentially means that every N-divisible ordered cancellative Abelian monoid has the same algebraic theory as the max-plus algebra (Q, + ), which is a much simpler structure than (R, + ). From this point of view, the algebraic essence of tropical mathematics boils down to (Q, + ). Sometimes we want to study its ordered submonoid (Z, + ), or even (N, + ), although they are not N-divisible. Nevertheless, just as one often wants to study the arithmetic of Q by viewing finite homomorphic images of Z, we want the option of studying finite homomorphic images of the ordered monoid (N, + ). Towards this end, we define the q-truncated monoid M = [1, q] := {1, 2, . . . , q}, given with the obvious ordering; the sum and product of two elements k, ℓ ∈ L are taken as usual, if not exceeding q − 1, and is q otherwise. In other words, q could be considered as the infinite element of the finite monoid M. ALGEBRAIC STRUCTURES 3 2.2. Semirings without zero. So far, dequantization has enabled us to pass from algebras to ordered Abelian monoids, which come equipped with a rich model theory ready to implement, and as noted above, there is a growing theory of algebraic geometry over monoids [CHWW]. But to utilize standard tools such as polynomials and matrices, we need two operations (addition and multiplication), and return to the language of semirings, using [Gol] as a general reference. We write † to indicate that we do not require the zero element. A semiring† (R, +,·, 1) is a set R equipped with binary operations + and · such that: • (R, +) is an Abelian semigroup; • (R,· , 1R) is a monoid with identity element 1R; • Multiplication distributes over addition. A semifield† is a semiring† in which every element is (multiplicatively) invertible. In particular, the max-plus algebras (Z, + ), (Q, + ), and (R, + ) are semifields†, since + now is the multiplication. A semiring is a semiring† with a zero element 0R satisfying a + 0R = a, a · 0R = 0R = 0R · a, ∀a ∈ R. We use semirings† instead of semirings since the zero element can be adjoined formally, and often is irrelevant. For example, the zero element of the max-plus algebra would be −∞, which requires special attention. A semifield is a semifield† with a zero element adjoined. Note that under this definition the customary field Q with the usual operations is not a semifield, since Q \ {0} is not closed under addition. and defined by: Any ordered Abelian monoid gives rise to a max-plus semiring†, where the operations are written ⊙ and ⊕ Associativity and distributivity (of ⊙ over ⊕) hold, but NOT negation, since a ⊕ b 6= −∞ unless a = b = −∞. Although the circle notation is standard in the tropical literature, we find it difficult to read when dealing with algebraic formulae. (Compare x4 + 7x3 + 4x + 1 with a ⊕ b := max{a, b}; a ⊙ b := a + b. x ⊙ x ⊙ x ⊙ x ⊕ 7 ⊙ x ⊙ x ⊙ x ⊕ 4 ⊙ x ⊕ 1.) The max-plus algebra satisfies the property that a + b ∈ {a, b}; we call this property bipotence. Thus, when appealing to the abstract theory of semirings we use the usual algebraic notation of · (often suppressed) and + respectively for multiplication and addition. In particular, the max-plus algebra, viewed as a semiring†, is idempotent in the sense that a + a = a for all a. Although idempotence pervades the theory, it turns out that what is really crucial for many applications is the following fact: Remark 2.2. In any idempotent semiring†, if a + b + c = a, then a + b = a. (Proof: a = a + b + c = (a + b + c) + b = a + b.) Let us call such a semiring† proper. Note that a proper semiring cannot have additive inverses other than 0, since if c + a = 0, then a = a + 0 = a + c + a, implying a = a + c = 0. Any proper semiring† R gives rise to a partial order, given by a ≤ b iff a + c = b for some c ∈ R. This is a total order when the semiring† R is bipotent. Thus, the categories of bipotent semirings† and ordered monoids are isomorphic, and each language has its particular advantages. 2.3. The function semiring†. Definition 2.3. The function semiring† Fun(S, R) is the set of functions from a set S to a semiring† R. Fun(S, R) becomes a semiring† under componentwise operations, and is proper when R is proper. Cus- tomarily one takes S = R(n), the Cartesian product of n copies of R. This definition enables us to work with proper subsets, but the geometric applications lie outside the scope of the present paper. 2.3.1. Polynomials and power series. Λ = {λ1, . . . , λn} always denotes a finite set of indeterminates com- muting with the semiring† R; often n = 1 and we have a single indeterminate λ. We have the polynomial semiring† R[Λ]. As in [IzR1], we view polynomials in R[Λ] as functions, but perhaps viewed over some extension R′ of R. More precisely, for any subset S ⊆ R(n), there is a natural semiring† homomorphism ψ : R[Λ] → Fun(S, R), (2.2) 4 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN obtained by viewing a polynomial as a function on S. When R is a semifield†, the same analysis is applicable to Laurent polynomials R[Λ, Λ−1], since the homomorphism λi 7→ ai then sends λ−1 . Likewise, when R is power-cancellative and divisible, we can also define the semiring† of rational polynomials R[Λ]rat, where the powers of the λi are taken to be arbitrary rational numbers. These can all be viewed as elementary formulas in the appropriate languages, so the model theory alluded to earlier is applicable to the appropriate polynomials and their (tropical) roots in each case. 7→ a−1 i i Other functions over the bipotent semiring† R of an ordered monoid M can be defined in the same way. For example, if M is an ordered submonoid of (R+,·), then we can define the formal exponential series exp(a) :=Xk ak k! (2.3) since a < m implies am+1 is defined in Fun(R, R). (m+1)! < am m! , and thus (2.3) becomes a finite sum. It follows at once that exp(λ) :=P λk k! 2.4. Puisuex series and valuations. Since logarithms often do not work well with algebraic structure , tropicalists have turned to the algebra of Puiseux series, denoted K, whose elements have the form p(t) = Xτ ∈Q≥0, cτ ∈K cτ tτ , where the powers of t are taken over well-ordered subsets of Q. Here K is any algebraically closed field of characteristic 0, customarily C. Intuitively, we view t as a "generic element." In the literature, the powers τ are often taken in R rather than Q, but it is enough to work with Q, for which it much easier to compute the powers of t. The algebra K is an algebraically closed field. Now recall that a valuation from an integral domain W to an ordered monoid (G, + ) is a multiplicative monoid homomorphism v : W \ {0} → G, i.e., with v(ab) = v(a) + v(b), and satisfying the property v(a + b) ≥ min{v(a), v(b)} for all a, b ∈ K. We formally put v(0) = ∞. For example, the field of Puiseux series has the order valuation v given by v(p(t)) := min{τ ∈ Q≥0 : cτ 6= 0}. The following basic observation in valuation theory shows why valuations are relevant to the tropical As t → 0, the dominant term in p(t) becomes cv(p(t))tv(p(t)). theory. Remark 2.4. If v(a) 6= v(b), then v(a + b) = min{v(a), v(b)}. Inductively, if v(a1), . . . , v(am) are distinct, then v(cid:18) mXi=1 ai(cid:19) = min{v(ai) : 1 ≤ i ≤ m} ∈ G. more deeply in [BiG]. Consequently, if P ai = 0, then at least two of the v(ai) are the same. These considerations are taken much When W is a field, the value monoid G is a group. Much information about a valuation v : W → G ∪{∞} can be garnered from the target v(W ), but valuation theory provides some extra structure: • The valuation ring Ov = {a ∈ W : v(a) ≥ 0}, • The valuation ideal Pv = {a ∈ W : v(a) > 0}, • The residue ring ¯W = Ov/Pv, a field if W is a field. For example, the valuation ring of the order valuation on the field K of Puiseux series is {p(t) ∈ K : cτ = 0 for τ < 0}, and the residue field is K. We replace v by −v to switch minimum to maximum, and ∞ by −∞. One can generalize the notion of valuation to permit W to be a semiring†; taking W = M, we see that the identity map is a valuation, which provides one of our main examples. ALGEBRAIC STRUCTURES 5 2.5. The standard supertropical semiring†. This construction, following [IzR1], refines the max-plus algebra and picks up the essence of the value monoid. From now on, in the spirit of max-plus, we write the operation of an ordered monoid M as multiplication. We start with an Abelian monoid M := (M,· ), an ordered group G := (G,· ), and an onto monoid homomorphism v : M → G. We write aν for v(a), for a ∈ M. Thus every element of G is some aν. We write a ∼=ν b if aν = bν. Our two main examples: • M = G is the ordered monoid of the max-plus algebra (the original example in Izhakian's disserta- tion); • M is the multiplicative group of a field F , and v : F × → G is a valuation. Note that we forget the original addition on the field F ! Our objective is to use the order on G to study M. Accordingly we want to define a structure on M ∪ G. The standard supertropical semiring† R is the disjoint union M∪G, made into a monoid by starting with the given multiplications on M and G, and defining a · bν and aν · b to be (ab)ν for a, b ∈ M. We extend v to the ghost map ν : R → G by taking νM = v and νG to be the identity on G. Thus, ν is a monoid projection. We make R into a semiring† by defining a + b =  a b aν for aν > bν; for aν < bν; for aν = bν. R is never additively cancellative (except for M = {1}). M is called the tangible submonoid of R. G is called the ghost ideal. R is called a supertropical domain† when the monoid M is (multiplicatively) cancellative. Strictly speaking, a supertropical domain† will not be a semifield† since the ghost elements are not invertible. Accordingly, we define a 1-semifield† to be a supertropical domain† for which M is a group. commutative algebra. Towards this end, we write Motivation: The ghost ideal G is to be treated much the same way that one treats the zero element in = b a gs if a = b or a = b + ghost. = 0 if a is a ghost.) Note that for a tangible, a gs (Accordingly, write a gs := G ∪ {0}. We may formally adjoin a zero element in a separate component; then the ghost ideal is G0 think of the ghost elements as uncertainties in classical algebra arising from adding two Puiseux series whose lowest order terms have the same degree. = b iff a = b. If needed, we could a + a = aν instead of a + a = a. R is a cover of the max-plus algebra of G, in which we "resolve" tangible idempotence, in the sense that This modification in the structure permits us to detect corner roots of tropical polynomials in terms of the algebraic structure, by means of ghosts. Namely, we say that a ∈ R(n) is a root of a polynomial f ∈ R[Λ] when f (a) ∈ G. This concise formulation enables us to apply directly many standard mathematical concepts from algebra, algebraic geometry, category theory, and model theory, as described in [IzKR1]–[IzKR5] and [IzR1]–[IzR6]. The standard supertropical semiring works well with linear algebra, as we shall see. 1 ··· λin 2.6. Kapranov's Theorem and the exploded supertropical structure. Given a polynomial f (Λ) = Pi pi(λi1 n ) ∈ K[Λ], where i = (i1, . . . , in), i.e., with each pi a Puiseux series, we define its tropicaliza- tion f to be the tropical polynomialPi v(pi)λi1 n . (In the tropical literature, this is customarily written in the circle notation.) By Remark 2.4, if a ∈ K(n) is a root of f in the classical sense, then v(a) is a tropical root of f . Kapranov showed, conversely, that any tropical root of f has the form v(a) for suitable a ∈ K(n), and valuation theory can be applied to give a rather quick proof of this fact, although we are not aware of an explicit reference. (See [R1, Proposition 12.58] for an analogous proof of a related valuation-theoretic result.) 1 ··· λin To prove Kapranov's theorem, one needs more than just the lowest powers of the Puiseux series appearing as coefficients of f , but also their coefficients; i.e., we also must take into account the residue field of the 6 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN order valuation on Puiseux series. Thus, we need to enrich the supertropical structure to include this extra information. This idea was first utilized by Parker [Par] in his "exploded" tropical mathematics. Likewise, Kapranov's Theorem has been extended by Payne [Pay1, Pay2], for which we need the following more refined supertropical structure, initiated by Sheiner [ShSh]: Definition 2.5. Given a valuation v : W → G, we define the exploded supertropical algebra R = W × G, viewed naturally as a monoid. (Thus we are mixing the "usual" world with the tropical world.) We make R into a semiring† by defining (c, a) + (d, b) =  (c, a) (d, b) (c + d, a) when a > b; when a < b; when a = b. Sheiner's theory parallels the standard supertropical theory, where now the ghost elements are taken to be the 0-layer {0} × G. 3. The layered structure The standard supertropical theory has several drawbacks. First, it fails to detect the multiplicity of a root of a polynomial. For example we would want 3 to have multiplicity 5 as a tropical root of the tropical polynomial (λ + 3)5; this is not indicated supertropically. Furthermore, serious difficulties are encountered when attempting to establish a useful intrinsic differential calculus on the supertropical structure. Also, some basic supertropical verifications require ad hoc arguments. These drawbacks are resolved by refining the ghost ideal into different "layers," following a construction of [WW, Example 3.4] and [AkGG, Proposition 5.1]. Rather than a single ghost layer, we take an indexing set L which itself is a partially ordered semiring†; often L = N under classical addition and multiplication. Ordered semirings† can be trickier than ordered groups, since, for example, a > b in (R,· ) does not imply −a > −b, but rather −a < −b. To circumvent this issue, we require all elements in the indexing semiring† to be non-negative. Construction 3.1 ( [IzKR4, Construction 3.2]). Suppose we are given a cancellative ordered monoid G, viewed as a semiring† as above. For any partially ordered semiring† L we define the semiring† R := R(L,G) to be set-theoretically L × G, where we denote the "layer" {ℓ} × G as Rℓ and the element (ℓ, a) as [ℓ]a ; we define multiplication componentwise, i.e., for k, ℓ ∈ L, a, b ∈ G, [k]a [ℓ]b = [kℓ](ab) , (3.1) and addition via the rules: [k]a + [ℓ]b =  The sort map s : R → L is given by s( [k]a ) = k. [k]a [ℓ]b [k+ℓ]a if a > b, if a < b, if a = b. (3.2) R is indeed a semiring†. We identify a ∈ G with [1]a ∈ R1. In most applications the "sorting" semiring† L is ordered, and its smallest nonzero element is 1. In this case, the monoid { [ℓ]a : 0 < ℓ ≤ 1} is called the tangible part of R. The ghosts are { [ℓ]a : ℓ > 1}, and correspond to the ghosts in the standard supertropical theory. The ghosts together with R0 comprise an ideal. If there is a zero element it would be [0]0 . One can view the various choices of the sorting semiring† L as different stages of degeneration of algebraic geometry, where the crudest (for L = {1}) is obtained by passing directly to the familiar max-plus algebra. The supertropical structure is obtained when L = {1,∞}, where R1 and R∞ are two copies of G, with R1 the tangible submonoid of R and R∞ being the ghost copy. Other useful choices of L include {1, 2,∞} (to distinguish between simple roots and multiple roots) and N, which enables us to work with the multiplicity of roots and with derivatives, as seen below. In order to deal with tropical integration as anti-differentiation, one should consider the sorting semirings† Q>0 and R>0, but this is outside our present scope. By convention, [ℓ]λ denotes [ℓ]1R λ. Thus, any monomial can be written in the form [ℓ]αi λi1 where i = (i1, . . . , in). We say a polynomial f is tangible if each of its coefficients is tangible. 1 ··· λin n ALGEBRAIC STRUCTURES 7 Note that the customary decomposition R = Lℓ∈L Rℓ in graded algebras has been strengthened to the partition R = Sℓ∈LRℓ. The ghost layers now indicate the number of monomials defining a corner root of a tangible polynomial. Thus, we can measure multiplicity of roots by means of layers. For example, (λ + 3)5 = [1]λ5 + [5]3 λ4 + [10]6 λ3 + [10]9 λ2 + [5]12 λ + [1]15 , and substituting 3 for λ gives [32]15 = [25]35 . 3.1. Layered derivatives. Formal derivatives are not very enlightening over the max-plus algebra. For example, if we take the polynomial f = λ2 + 5λ + 8, which has corner roots 3 and 5, we have f ′ = 2λ + 5, having corner root 3, but the common corner root 3 of f and f ′ could hardly be considered a multiple root of f . This difficulty arises from the fact that 1 + 1 6= 2 in the max-plus algebra. The layering permits us to define a more useful version of the derivative (where now R contains a zero element 0R): Definition 3.2. The layered derivative f ′ lay of f on R[λ] is given by: (cid:18) nXj=0 [ℓj ]αj λj(cid:19)′ lay := nXj=1 [jℓj ]αj λj−1. (3.3) In particular, for α = [1]α ∈ R1, (αλj )′ lay := [j]α λj−1 Thus, we have the familiar formulas: (1) (f + g)′ lay = f ′ lay + g′ layg + f g′ lay; lay. (j ≥ 2), (αλ)′ lay := α, and α′ lay := 0R. (2) (f g)′ This is far more informative in the layered setting (say for L = N) than in the standard supertropical lay = f ′ setting, in which (αλj )′ is ghost for all j ≥ 2. 3.2. The tropical Laplace transform. The classical technique of Laplace transforms has a tropical analog which enables us to compare the various notions of derivative. Suppose L is infinite, say L = N . Formally permitting infinite vectors (aℓ)ℓ∈L permits us to define a homomorphism R[[Λ]] → R(L, R) given by X akλk 7→(cid:0) [k]k! ak(cid:1). (Strictly speaking, we would want the image to be ( [ 1 require us to take L = Q+.) For example, explay(a) 7→ ( [k]ak ) where each ak = a. functions in the layered theory. Now we define ( [ℓ]aℓ )′ = ( [ℓ−1]aℓ ). Then exp′ lay = explay . This enables one to handle trigonometric k ]k! ak), but this would complicate the notation and 3.3. Layered domains† with symmetry, and patchworking. Akian, Gaubert, and Guterman [AkGG, Definition 4.1] introduced an involutory operation on semirings, which they call a symmetry, to unify the supertropical theory with classical ring theory. One can put their symmetry in the context of R(L,G). Definition 3.3. A negation map on a semiring† L is a function τ : L → L satisfying the properties: N1. τ (kℓ) = τ (k)ℓ = kτ (ℓ); N2. τ 2(k) = k; N3. τ (k + ℓ) = τ (k) + τ (ℓ). Suppose the semiring† L has a negation map τ of order ≤ 2. We say that R := R(L,G) has a symmetry σ when R is endowed with a map σ : R → R and a negation map τ on L, together with the extra axiom: S1. s(σ(a)) = τ (s(a)), ∀a ∈ R. 8 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN Example 3.4. Suppose L is an ordered semiring†. We mimic the well-known construction of Z from N. Define the doubled semiring† the direct product of two copies L1 and L−1 of L, where addition is defined componentwise, but multiplication is given by D(L) = L1 × L−1, (k, ℓ) · (k′, ℓ′) = (kk′ + ℓℓ′, kℓ′ + ℓk′). In other words, D(L) is multiplicatively graded by {±1}. D(L) is endowed with the product partial order, i.e., (k′, ℓ′) ≥ (k, ℓ) when k′ ≥ k and ℓ′ ≥ ℓ. Here is an example relating to "patchworking," [ItMS]. Example 3.5. Suppose G is an ordered Abelian monoid, viewed as a semiring† as in Construction 3.1. Define the doubled layered domain† but with addition and multiplication given by the following rules: R = R(D(L),G) = {((k, ℓ), a) : (k, ℓ) 6= (0, 0), a ∈ G}, ((k, ℓ), a) + ((k′, ℓ′), b) =   ((k, ℓ), a) ((k′, ℓ′), b) ((k + k′, ℓ + ℓ′), a) if a > b, if a < b, if a = b. ((k, ℓ), a) · ((k′, ℓ′), b) = ((kk′ + ℓℓ′, kℓ′ + k′ℓ), ab). Remark 3.6. In R = R(D(L),G), the symmetry σ : R → R given by σ : ((k, ℓ), a) 7→ ((ℓ, k), a) is analogous to the one described in [AkGG], and behaves much like negation. For example, when L = {1,∞}, we note that D(L) = {(1, 1), (1,∞), (∞, 1), (∞,∞)}, which is applicable to Viro's theory of patchworking, where the "tangible" part could be viewed as those elements of layer (1, 1), (1,∞), or (∞, 1). Explicitly, comparing with Viro's use of hyperfields in [Vi, § 3.5], we identify these three layers respectively with 0, 1, and −1 in his terminology, and the element (∞,∞) with the set {0, 1,−1}. 4. Matrices and linear algebra As an application, the supertropical and layered structures provide many of the analogs to the classical Hamilton-Cayley-Frobenius theory. Mn(R) denotes the semiring† of n× n matrices over a semiring R. (Note that the familiar matrix operations do not require negation.) Although one of the more popular and most applicable aspects of idempotent mathematics, idempotent matrix theory is handicapped by the lack of an element −1 with which to construct the determinant. Many ingenious methods have been devised to circumvent this difficulty, as surveyed in [AkBG]; also cf. [AkGG] and many interesting papers in this volume. Unfortunately these give rise to many different notions of rank of matrix, and often are difficult to understand. The layered (and more specifically, supertropical) theories give a unified and relatively straightforward notion of rank of a matrix, eigenvalue, adjoint, etc. 4.1. The supertropical determinant. This discussion summarizes [IzR3]. We define the supertropical determinant A of a matrix A = (ai,j ) to be the permanent: (ai,j) = Xπ∈Sn a1,π(1) ··· an,π(n). (4.1) Defining the transpose matrix (ai,j)t to be (aj,i), we have (cid:12)(cid:12)(ai,j)t(cid:12)(cid:12) = (ai,j ) . A = 0R iff "enough" entries are 0R to force each summand in Formula 4.1 to be 0R. This property, which in classical matrix theory provides a description of singular subspaces, is too strong for our purposes. We now take the natural supertropical version. Write T for the tangible elements of our supertropical semiring R, and T0 = T ∪ {0}. Definition 4.1. A matrix A is nonsingular if A ∈ T ; A is singular when A ∈ G0 . ALGEBRAIC STRUCTURES 9 The standard supertropical structure often is sufficient for matrices, since it enables us to distinguish between nonsingular matrices (in which the tropical n × n determinant is computed as the unique maximal product of n elements in one track) and singular matrices. The tropical determinant is not multiplicative, as seen by taking the nonsingular matrix A = (cid:18)0 2(cid:19). Then A2 =(cid:18)1 2 Theorem 4.2. For any n × n matrices over a supertropical semiring R, we have 3 4(cid:19) is singular and (cid:12)(cid:12)A2(cid:12)(cid:12) = 5ν 6= 2 · 2. But we do have: 1 0 AB gs = AB . In particular, AB = A B whenever AB is tangible. We say a permutation σ ∈ Sn attains A if A ∼=ν aσ(1),1 ··· aσ(n),n. • By definition, some permutation always attains A. • If there is a unique permutation σ which attains A, then A = a1,σ(1) ··· an,σ(n). • If at least two permutations attain A, then A must be singular. Note in this case that if we replaced all nonzero entries of A by tangible entries of the same ν-value, then A would still be singular. 4.2. Quasi-identities and the adjoint. Definition 4.3. A quasi-identity matrix IG is a nonsingular, multiplicatively idempotent matrix equal to I + ZG, where ZG is 0R on the diagonal, and whose off-diagonal entries are ghosts or 0R. IG = 1R by the nonsingularity of IG. Also, for any matrix A and any quasi-identity, IG, we have There is another notion to help us out. AIG = A + AG, where AG = AZG ∈ Mn(G0 ). Definition 4.4. The (i, j)-minor A′ of A. The adjoint matrix adj(A) of A is defined as the transpose of the matrix (a′ i,j of a matrix A = (ai,j) is obtained by deleting the i row and j column i,j), where a′ i,j =(cid:12)(cid:12)A′ i,j(cid:12)(cid:12). (4.2) Remark 4.5. (i) Suppose A = (ai,j). An easy calculation using Formula (4.1) yields Consequently, ai,j a′ nXj=1 i,j ≤ν A for each i, j. A = ai,j a′ i,j, ∀i. (ii) If we take k 6= i, then replacing the i row by the k row in A yields a matrix with two identical rows; thus, its determinant is a ghost, and we thereby obtain Likewise nXj=1 nXj=1 ai,j a′ k,j ∈ G0 , ∀k 6= i; aj,i a′ j,k ∈ G0 , ∀k 6= i. (4.3) One easily checks that adj(B) adj(A) = adj(AB) for any 2 × 2 matrices A and B. However, this fails for larger n, cf. [IzR3, Example 4.7]. We do have the following fact, which illustrates the subtleties of the supertropical structure, cf. [IzR3, Proposition 5.6]: Proposition 4.6. adj(AB) = adj(B) adj(A) + ghost. Definition 4.7. For A invertible, define IA = A , adj(A) A I ′ A = adj(A) A A. 10 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN The matrices IA and I ′ A are quasi-identities, as seen in [IzR3, Theorem 4.13]. The main technique of proof is to define a string (from the matrix A) to be a product ai1,j1 ··· aik,jk of entries from A and, given such a string, to define its digraph to be the graph whose edges are (i1, j1), . . . , (ik, jk), counting multiplicities. A k-multicycle in a digraph is the union of disjoint simple cycles, the sum of whose lengths is k; thus every vertex in an n-multicycle appears exactly once. A careful examination of the digraph in conjunction with Hall's Marriage Theorem yields the following major results from [IzR3, Theorem 4.9 and Theorem 4.12]: Theorem 4.8. (i) A adj(A) = An . (ii) adj(A) = An−1 . In case A is a nonsingular, we define A∇ = adj(A) A . Thus AA∇ = IA, and A∇A = I ′ A. Note that I ′ A and IA may differ off the diagonal, although This result is refined in [IzR4, Theorem 2.18]. One might hope that A adj(A)A = A A, but this is false in general! The difficulty is that one might not be able to extract an n-multicycle from IAA = AA∇A = AI ′ A. k,j ak,ℓ. For example, when n = 3, the term a1,1(a1,3a3,2)a2,2 = a1,1a′ have the following positive result from [IzR4, Theorem 4.18]: Theorem 4.9. adj(A) adj( adj(A)) adj(A) ∼=ν An−1 adj(A) for any n × n matrix A. 4.3. The supertropical Hamilton-Cayley theorem. ai,ja′ 2,1a2,2 does not contain an n-multicycle. We do (4.4) Definition 4.10. Define the characteristic polynomial fA of the matrix A to be fA = λI + A , and the tangible characteristic polynomial to be a tangible polynomial cfA = λn +Pn are tangible and bαi ∼=ν αi, such that fA = λn +Pn i=1bαiλn−i, where αi Under this notation, we see that αk ∈ R arises from the dominant k-multicycles in the digraph of A. We i=1 αiλn−i. say that a matrix A satisfies a polynomial f ∈ R[λ] if f (A) ∈ Mn(G0 ). Theorem 4.11. (Supertropical Hamilton-Cayley, [IzR3, Theorem 5.2]) Any matrix A satisfies both its characteristic polynomial fA and its tangible characteristic polynomial cfA. 4.4. Tropical dependence. Now we apply supertropical matrix theory to vectors. As in classical math- ematics, one defines a module (often called semi-module in the literature) analogously to module in classical algebra, noting again that negation does not appear in the definition. It is convenient to stipulate that the module V has a zero element 0V , and then we need the axiom: a0V = 0V for all a ∈ R. Also, if 0 ∈ R then we require that 0v = 0V for all v ∈ V . In what follows, F always denotes a 1-semifield. In this case, a module over F is called a (supertropical) vector space. The natural example is F (n), with componentwise operations. As in the classical theory, there is the usual familiar correspondence between the semiring Mn(F ) and the linear transformations of F (n). For v = (v1, . . . , vn), w = (w1, . . . , wn) ∈ F (n), we write v gs Here is an application of the adjoint matrix, used to solve equations. = w when vi gs = wi for all 1 ≤ i ≤ n. Remark 4.12. Suppose A is nonsingular, and v ∈ F (n). Then the equation Aw = v + ghost has the solution w = A∇v. Indeed, writing IA = I + ZG for a ghost matrix ZG, we have Aw = AA∇v = IAv = (I + ZG)v gs = v. This leads to the supertropical analog of Cramer's rule [IzR4, Theorem 3.5]: ALGEBRAIC STRUCTURES 11 Theorem 4.13. If A is a nonsingular matrix and v is a tangible vector, then the equation Ax gs solution over F which is the tangible vector having value A∇v. = v has a Our next task is to characterize singularity of a matrix A in terms of "tropical dependence" of its rows. In some ways the standard supertropical theory works well with matrices, since we are interested mainly in whether or not this matrix is nonsingular, i.e., if its determinant is tangible; at the outset, at least, we are not concerned with the precise ghost layer of the determinant. Definition 4.14. A subset W ⊂ F (n) is tropically dependent if there is a finite sum P αiwi ∈ G(n) , with each αi ∈ T0 , but not all of them 0R; otherwise W ⊂ F (n) is called tropically independent. A vector v ∈ F (n) is tropically dependent on W if W ∪ {v} is tropically dependent. 0 By [IzKR2, Proposition 4.5], we have: Proposition 4.15. Any n + 1 vectors in F (n) are tropically dependent. Theorem 4.16. ([IzR3, Theorem 6.5]) Vectors v1, . . . , vn ∈ F (n) are tropically dependent, iff the matrix whose rows are v1, . . . , vn is singular. Corollary 4.17. The matrix A ∈ Mn(F ) over a supertropical domain F is nonsingular iff the rows of A are tropically independent, iff the columns of A are tropically independent. Proof. Apply the theorem to A and At, which are the same. (cid:3) There are two competing supertropical notions of base of a vector space, that of a maximal independent set of vectors, and that of a minimal spanning set, but this is unavoidable since, unlike the classical theory, these two definitions need not coincide. 4.5. Tropical bases and rank. The customary definition of tropical base, which we call s-base (for spanning base), is a minimal spanning set (when it exists). However, this definition is rather restrictive, and a competing notion provides a richer theory. Definition 4.18. A d-base (for dependence base) of a vector space V is a maximal set of tropically independent elements of V . A d,s-base is a d-base which is also an s-base. The rank of a set B ⊆ V , denoted rank(B), is the maximal number of d-independent vectors of B. Our d-base corresponds to the "basis" in [MS, Definition 5.2.4]. In view of Proposition 4.15, all d-bases of F (n) have precisely n elements. This leads us to the following definition. Definition 4.19. The rank of a vector space V is defined as: rank(V ) := max(cid:8) rank(B) : B is a d-base of V(cid:9). We have just seen that rank(F (n)) = n. Thus, if V ⊂ F (n), then rank(V ) ≤ n. We might have liked rank(V ) to be independent of the choice of d-base of V , for any vector space V . This is proved in the classical theory of vector spaces by showing that dependence is transitive. However, transitivity of dependence fails in the supertropical theory, and, in fact, different d-bases may contain different numbers of elements, even when tangible. An example is given in [MS, Example 5.4.20], and reproduced in [IzKR2, Example 4.9] as being a subspace of F (4) having d-bases both of ranks 2 and 3. Example 4.20. The matrix A =   4 4 4 4 0 4 1 4 2   has rank 2, but is "ghost annihilated" by the tropically independent vectors v1 = (1, 1, 0)t and v2 = (1, 1, 1)t; i.e., Av1 = Av2 = (5ν, 5ν, 5ν)t, although 2 + 2 > 3. We do have some consolations. Proposition 4.21 ([IzKR2, Proposition 4.11]). For any tropical subspace V of F (n) and any tangible v ∈ V, there is a tangible d-base of V containing v whose rank is that of V . Proposition 4.22 ([IzKR2, Proposition 4.13]). Any n × n matrix of rank m has ghost annihilator of rank ≥ n − m. 12 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN 4.5.1. Semi-additivity of rank. Definition 4.23. A function rankS : S → N is monotone if for all S2 ⊆ S1 ⊆ S we have rankS(S2 ∪ {s}) − rankS(S2) ≥ rankS(S1 ∪ {s}) − rankS(S1) (4.5) for all s ∈ S. Note that (4.5) says that rankS(S1)− rankS(S2) ≥ rankS(S1 ∪{s})− rankS(S2 ∪{s}). Also, taking S2 = ∅ yields rankS(S1 ∪ {s}) − rankS(S1) ≤ 1. Lemma 4.24. If rankS : S → N is monotone, then rankS(S1) + rankS(S2) ≥ rankS(S1 ∪ S2) + rankS(S1 ∩ S2) for all S1, S2 ⊂ S. Proof. Induction on m = rankS(S2 \ S1). If m = 0, i.e., S2 ⊆ S1, then the left side of (4.6) equals the right side. Thus we may assume that m ≥ 1. Pick s in a d-base of S2 \ S1. Let S′ 2 = S2 \ {s}. Noting that rankS(S′ 2 \ S1) = m − 1, we see by induction that rankS(S1) + rankS(S′ 2) ≥ rankS(S1 ∪ S′ 2) + rankS(S1 ∩ S′ 2), (4.6) (4.7) or (taking S1 ∪ S′ 2 instead of S2 in (4.5)), rankS(S1) − rankS(S1 ∩ S2) = rankS(S1) − rankS(S1 ∩ S′ 2) ≥ rankS(S1 ∪ S′ 2) − rankS(S′ 2) ≥ rankS(S1 ∪ S2) − rankS(S2), yielding (4.6). Proposition 4.25. rank(S1) + rank(S2) ≥ rank(S1 ∪ S2) + rank(S1 ∩ S2) for all S1, S2 ⊂ S. Proof. rank is a monotone function, since each side of (4.5) is 0 or 1, depending on whether or not s is independent of Si, and only decreases as we enlarge the set. (cid:3) (cid:3) 4.6. Supertropical eigenvectors. The standard definition of an eigenvector of a matrix A is a vector v, with eigenvalue β, satisfying Av = βv. It is well known [BrR] that any (tangible) matrix has an eigenvector. Example 4.26. The characteristic polynomial fA of A =(cid:18) 4 0 0 1 (cid:19) is (λ + 4)(λ + 1) + 0 = (λ + 4)(λ + 1), and the vector (4, 0) is a eigenvector of A, with eigenvalue 4. However, there is no eigenvector having eigenvalue 1. In general, the lesser roots of the characteristic polynomial are "lost" as eigenvalues. We rectify this deficiency by weakening the standard definition. Definition 4.27. A tangible vector v is a generalized supertropical eigenvector of a (not necessarily = βmv for some m; the tangible) matrix A, with generalized supertropical eigenvalue β ∈ T0 , if Amv gs minimal such m is called the multiplicity of the eigenvalue (and also of the eigenvector). A supertropical eigenvector is a generalized supertropical eigenvector of multiplicity 1. Example 4.28. The matrix A =(cid:18) 4 0 0 1 (cid:19) of Example 4.26 also has the tangible supertropical eigenvector v = (0, 4), corresponding to the supertropical eigenvalue 1, since Av = (4ν, 5) = 1v + (4ν,−∞). Proposition 4.29. If v is a tangible supertropical eigenvector of A with supertropical eigenvalue β, the matrix A + βI is singular (and thus β must be a (tropical) root of the characteristic polynomial fA of A). Conversely, we have: Theorem 4.30 ([IzR3, Theorem 7.10]). Assume that νT : T → G is 1:1. For any matrix A, the dominant tangible root of the characteristic polynomial of A is an eigenvalue of A, and has a tangible eigenvector. The other tangible roots are precisely the supertropical eigenvalues of A. Let us return to our example A =(cid:18)0 1 0 2(cid:19) . Its characteristic polynomial is λ2 + 2λ + 2 = (λ + 0)(λ + 2), ALGEBRAIC STRUCTURES 13 whose roots are 2 and 0. The eigenvalue 2 has tangible eigenvector v = (0, 2) since Av = (2, 4) = 2v, but there are no other tangible eigenvalues. A does have the tangible supertropical eigenvalue 0, with tangible supertropical eigenvector w = (2, 1), since Aw = (2, 3ν) = 0w + (−∞, 3ν). Note that A + 0I = (cid:18)0ν singular, because A + 0I = 2ν. Furthermore, A2 =(cid:18)1 2 2(cid:19) is 3 4(cid:19) is a root of λ2 + 4A, and thus A is a root of g = λ4 + 4λ2 = (λ(λ + 2))2, but 0 0 1 is not a root of g although it is a root of fA. This shows that the naive formulation of Frobenius' theorem fails in the supertropical theory, and is explained in the work of Adi Niv [N]. 4.7. Bilinear forms and orthogonality. One can refine the study of bases by introducing angles, i.e., orthogonality, in terms of bilinear forms. Let us quote some results from [IzKR2]. Definition 4.31. A (supertropical) bilinear form B on a (supertropical) vector space V is a function B : V × V → F satisfying B(v1 + v2, w1 + w2) gs = B(v1, w1) + B(v1, w2) + B(v2, w1) + B(v2, w2), for all α ∈ F and vi ∈ V, and wj ∈ V ′. B(αv1, w1) = αB(v1, w1) = B(v1, αw1), We work with a fixed bilinear form B = h , i on a (supertropical) vector space V ⊆ F (n). The Gram matrix of vectors v1, . . . , vk ∈ F (n) is defined as the k × k matrix ··· ··· . . . ··· eG(v1, . . . , vk) = hv1, v1i hv2, v1i hv1, v2i hv2, v2i   ... hvk, v1i hvk, v2i ... hv1, vki hv2, vki ... hvk, vki   . (4.8) The set {v1, . . . , vk} is nonsingular (with respect to B) when its Gram matrix is nonsingular. In particular, given a vector space V with s-base {b1, . . . , bk}, we have the matrix eG = eG(b1, . . . , bk), which can be written as (gi,j) where gi,j = hbi, bji. The singularity of eG does not depend on the choice of s-base. Definition 4.32. For vectors v, w in V , we write v⊥⊥w when hv, wi ∈ G0 , that is hv, wi gs that v is left ghost orthogonal to w. We write W ⊥⊥ for {v ∈ V : v⊥⊥w for all w ∈ W.} Definition 4.33. A subspace W of V is called nondegenerate (with respect to B), if W ⊥⊥ ∩ W is ghost. The bilinear form B is nondegenerate if the space V is nondegenerate. Lemma 4.34. Suppose {w1, . . . , wm} tropically spans a subspace W of V , and v ∈ V. IfPm for all βi ∈ T , then v ∈ W ⊥⊥. Theorem 4.35. ([IzKR2, Theorem 6.7]) Assume that vectors w1, . . . , wk ∈ V span a nondegenerate sub- space W of V . If eG(w1, . . . , wk) ∈ G0 , then w1, . . . , wk are tropically dependent. Corollary 4.36. If the bilinear form B is nondegenerate on a vector space V , then the Gram matrix (with respect to any given supertropical d,s-base of V ) is nonsingular. Definition 4.37. The bilinear form B is supertropically alternate if hv, vi ∈ G0 supertropically symmetric if hv, wi + hw, vi ∈ G0 for all v, w ∈ V . for all v ∈ V. B is i=1 βihv, wii ∈ G0 = 0F , and say We aim for the supertropical version ([IzKR2, Theorem 6.19]) of a classical theorem of Artin, that any bilinear form in which ghost-orthogonality is symmetric must be a supertropically symmetric bilinear form. Definition 4.38. The (supertropical) bilinear form B is orthogonal-symmetric if it satisfies the following property for any finite sum, with vi, w ∈ V : 14 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN Xi hvi, wi ∈ G0 iff Xi hw, vii ∈ G0 , (4.9) B is supertropically orthogonal-symmetric if B is orthogonal-symmetric and satisfies the additional property that hv, wi ∼=ν hw, vi for all v, w ∈ V satisfying hv, wi ∈ T . The symmetry condition extends to sums, and after some easy lemmas we obtain ([IzKR2, Theorem 6.19]): Theorem 4.39. Every orthogonal-symmetric bilinear form B on a vector space V is supertropically sym- metric. 5. Identities of semirings, especially matrices The word "identity" has several interpretations, according to its context. First of all, there are well- known matrix identities such as the Hamilton-Cayley identity which says that any matrix is a root of its characteristic polynomial. Since the classical theory of polynomial identities is tied in with invariant theory, we also introduce layered polynomial identities (PIs), to enrich our knowledge of layered matrices. 5.1. Polynomial identities of semirings†. We draw on basic concepts of polynomial identities, i.e., PI's, say from [R2, Chapter 23]. Since semirings† do not involve negatives, we modify the definition a bit. Definition 5.1. The free N-semiring† N{x1, x2, . . .} is the monoid semiring† of the free (word) monoid {x1, x2, . . .} over the commutative semiring† N. Definition 5.2. A (semiring†) polynomial identity (PI) of a semiring† R is a pair (f, g) of (noncom- mutative) polynomials f (x1, . . . , xm), g(x1, . . . , xm) ∈ N{x1, . . . , xm} for which ∀r1, . . . , rm ∈ R. f (r1, . . . , rm) = g(r1, . . . , rm), We write (f, g) ∈ id(R) when (f, g) is a PI of R. Remark 5.3. A semigroup identity of a semigroup S is a pair (f, g) of (noncommutative) monomials f (x1, . . . , xm), g(x1, . . . , xm) ∈ N{x1, . . . , xm} for which f (s1, . . . , sm) = g(s1, . . . , sm), ∀s1, . . . , sm ∈ S. If S is contained in the multiplicative semigroup of a semiring† R, the semigroup identities of S are precisely the semiring† PIs (f, g) where f and g are monomials. Akian, Gaubert and Guterman [AkGG, Theorem 4.21] proved their strong transfer principle, which immediately implies the following easy but important observation: Theorem 5.4. If f, g ∈ N{x1, . . . , xn} have disjoint supports and f − g is a PI of Mn(Z), then f = g is also a semiring† PI of Mn(R) for any commutative semiring† R. Proof. Since Z is an infinite integral domain, f − g is also a PI of Mn(C), where C = Z[ξ1, ξ2, . . . ] denotes the free commutative ring in countably many indeterminates, implying (f, g) is a semiring† PI of Mn(N[ξ1, ξ2, . . . ]). But the semiring† Mn(R) is a homomorphic image of Mn(N[ξ1, ξ2, . . . ]), implying (f, g) ∈ id(Mn(R)). (cid:3) Corollary 5.5. Any PI of Mn(Z) yields a corresponding semiring† PI of Mn(R) for all commutative semirings† R. Proof. Take f to be the sum of the terms having positive coefficient, and g to be the sum of the terms having negative coefficient, and apply the theorem. (cid:3) Many (but not all) matrix PIs can be viewed in terms of Theorem 5.4, although semiring versions of basic results such as the Amitsur-Levitzki Theorem and Newton's Formulas often are more transparent here. We say that polynomials f (x1, . . . , xm) and g(x1, . . . , xm) are a t-alternating pair if f and g are inter- changed whenever we interchange a pair xi and xj for some 1 ≤ i < j ≤ t. For example, x1x2 and x2x1 are a 2-alternating pair. Sometimes we write the non-alternating variables as y1, y2, . . . ; we write y as shorthand for all the yj. ALGEBRAIC STRUCTURES 15 Definition 5.6. We partition the symmetric group St of permutations in t letters into the even permu- tations S+ t . Given a t-linear polynomial h(x1, . . . , xt; y), we define the t- alternating pair t and the odd permutations S− and The standard pair is Stnt := (h+ h(xσ(1), . . . , xσ(t); y) h(xσ(1), . . . , xσ(t); y). t h− h+ alt(x1, . . . , xt; y) := Xσ∈S+ alt(x1, . . . , xt; y) := Xσ∈S− Stnt :=(cid:18) Xσ∈S+ alt, h− t alt), where h = x1 ··· xt. Explicitly, xσ(1) ··· xσ(t)(cid:19). xσ(1) ··· xσ(t), Xσ∈S− alt), where h = x1y1x2y2 ··· xtyt. Explicitly, t t The Capelli pair is Capt := (h+ alt, h− Capt :=(cid:18) Xσ∈S+ t xσ(1)y1xσ(2)y2 ··· yt−1xσ(t)yt, Xσ∈S− t xσ(1)y1xσ(2)y2 ··· yt−1xσ(t)yt(cid:19). Proposition 5.7. Any t-alternating pair (f, g) is a PI for every semiring† R spanned by fewer than t elements over its center. Proof. Suppose R is spanned by {b1, b2, . . . , bt−1}. We need to verify f(cid:18)X αi,1bi1 , . . . ,X αi,tbit, . . .(cid:19) = g(cid:18)X αi,1bi1, . . . ,X αi,tbit , . . .(cid:19). Since f and g are linear in these entries, it suffices to verify f (bi1 , . . . , bit , . . . ) = g(bi1 , . . . , bit, . . . ) (5.1) for all i1, . . . , it. But by hypothesis, two of these must be equal, say ik and ik′ , so switching these two yields (5.1) by the alternating hypothesis. (cid:3) Let ei,j denote the matrix units. The semiring† version of the Amitsur-Levitzki theorem [AmL], that Stn2n ∈ id(Mn(N)), is an immediate consequence of Theorem 5.4, and its minimality follows from: Lemma 5.8. Any pair of multilinear polynomials f (x1, . . . , xm) and g(x1, . . . , xm) having no common mono- mials do not comprise a PI of Mn(R) unless m ≥ 2n. Proof. Rewriting indices we may assume that x1 ··· xm appears as a monomial of f, but not of g, and we note (for ℓ =(cid:2) m 2(cid:3) + 1) that f (e1,1, e1,2, e2,2, e2,3, . . . , ek−1,k, ek,k, . . . ) = e1,ℓ 6= 0, g(e1,1, e1,2, e2,2, e2,3, . . . , ek−1,k, ek,k, . . . ) = 0. but (cid:3) Likewise, the identical proof of [R2, Remark 23.14] shows that the Capelli pair Capn2 is not a PI of Mn(C), and in fact (e1,1, 0) ∈ Capn2(Mn(R)) for any semiring† R. 5.2. Surpassing identities. The surpassing identity f gs for all a1, . . . , am ∈ R. Example 5.9. Take the general 2 × 2 matrix A = (cid:18)a b A2 =(cid:18) a2 + bc bc + d2(cid:19) , so c(a + d) b(a + d) c = g(a1, . . . , am) = g holds when f (a1, . . . , am) gs d(cid:19) . Then tr(A) = a + d and A = ad + bc. A2 + adI =(cid:18)a(a + d) + bc c(a + d) b(a + d) bc + d(a + d)(cid:19) = tr(A)A + bcI, implying A2 + A I = tr(A)A + bcνI, 16 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN yielding the surpassing identity A2 + A I gs = tr(A)A for 2 × 2 matrices. We might hope for a surpassing identity involving alternating terms in the Hamilton-Cayley polynomial, but a cursory examination of matrix cycles dashes our hopes. Example 5.10. Let A =   . Then A2 = − d a   c − − − b − cd ab − − cd ac bc − − in this case, where α denotes the other coefficient in fA. But for A =  A3 = αA + A   , implying   and A3 =  cd2 acd abc c2d abc − − bcd abc  a − −  we have − b − − − c A3 + αA + 2 − − −  − abc − − − abc   = tr(A)A2 + A , so neither A3 + αA nor tr(A)A2 + A necessarily surpasses the other. 5.3. Layered surpassing identities. Since we want to deal with general layers, we write 2a (instead of aν) for a + a, but note that s(2a) = 2s(a). When working with the layered structure, we can extend the notion of PI from Definition 5.2 by making use of the following relations that arise naturally in the theory. Definition 5.11. The L-surpassing relation L = is given by = b a L iff either   a = b + c a = b, a ∼=ν b with c s(b)-ghost, with a s(b)-ghost. (5.2) It follows that if a L = b, then a + b is s(b)-ghost. When a 6= b, this means a ≥ν b and a is s(b)-ghost. Definition 5.12. The surpassing (L, ν)-relation L≡ν is given by a L≡ν b iff = b a L and a ∼=ν b. (5.3) The surpassing L-identity f L = g holds for f, g ∈ Fun(R(n), R) if f (a1, . . . , an) L = g(a1, . . . , an) for all a1, . . . , an ∈ R. The surpassing (L, ν)-identity f L≡ν g holds for f, g ∈ Fun(R(n), R) if f (a1, . . . , an) L≡ν g(a1, . . . , an) for all a1, . . . , an ∈ R. 5.3.1. Layered surpassing identities of commutative layered semirings. Just as the Boolean algebra satisfies the PI x2 = x, we have some surpassing identities for commutative layered domains†. Proposition 5.13. (Frobenius identity) (x1 + x2)m L≡ν xm Proof. This is just a restatement of [IzKR5, Remark 5.2]. 1 + xm 2 . (cid:3) Proposition 5.14. (x1 + x2 + x3)(x1x3 + x2x3 + x1x2) L≡ν (x1 + x2)(x1 + x3)(x2 + x3). More generally, let g1 =Pi xi, g2 =Pi<j xixj, . . . , and gm−1 =PiQj6=i xj . Then g1 ··· gm−1 L≡ν Yi<j (xi + xj). Proof. This is just a restatement of [IzR1, Theorem 8.51]. (5.4) (cid:3)   or a = b. 5.4. Layered surpassing identities of matrices. ALGEBRAIC STRUCTURES 17 We applied the strong transfer principle of Akian, Gaubert, and Guterman [AkGG, Theorem 4.21] to the (standard) supertropical matrix semiring in [IzR4]. We would like to make a similar argument in the layered case, but must avoid the following kind of counterexamples, pointed out by Adi Niv: Example 5.15. Suppose A = (cid:18) [1]10 A2 = [8]28 , which does not N-surpass A2 (and does not even N-surpass A). [10]0(cid:19) . Then A2 = (cid:18) [1]20 [2]14 [2]4 [2]4 [2]14 [4]8 (cid:19) , so A = [10]10 whereas The difficulty in the example was that some ν-small entry of A has a high layer which provides A a high layer but does not affect the powers of A. There is a version of surpassing which is useful in this context. Definition 5.16. An element c ∈ R is a strong ℓ-ghost (for ℓ ∈ L+) if s(c) ≥ 2ℓ. = b holds in an L-layered domain† R, if either The strong ℓ-surpassing relation a Sℓ a = b + c with c a strong ℓ-ghost (5.5) Sℓ = bi,j for We often take ℓ = s(b). In this case b + b Sℓ = b (as well as b + b ℓ = b). The strong ℓ-surpassing relation (ai,j) Sℓ = (bi,j) holds for matrices (ai,j) and (bi,j), if ai,j each i, j. We say that a matrix A is ℓ- layered if each entry has layer ≥ ℓ. We are ready for our other two versions of layered identities. Definition 5.17. The strong (ℓ, d)-surpassing identity f = g holds for f, g ∈ Fun(Mn(R)(m), Mn(R)) Sℓ;d if f (A1, . . . , Am) S ℓ = g(A1, . . . , Am) with ℓ = ℓd, for all ℓ-layered matrices A1, . . . , Am ∈ Mn(R). In the standard supertropical theory we take ℓ = ℓ = 1, but in the general layered theory we may need to consider other ℓ. Formally set P (x1, . . . , xℓ) = P + − P − and Q(x1, . . . , xℓ) = Q+ − Q−. We say Q is admissible if the monomials of Q+ and Q− are distinct, for each pair (i, j). We then obtain the following metatheorem, along the lines of [AkGG] (just as in [IzR4, Theorem 2.4]): Theorem 5.18. Suppose P = Q is a homogeneous matrix identity of Mn(Z) of degree d, with Q admissible. Then the matrix semiring† Mn(R) satisfies the strong (ℓ, d)-surpassing identity P + + P − Sℓ;d = Q+ + Q−. Here are some applications. Corollary 5.19. AB Sℓ;d = AB for L-layered n × n matrices A and B, where d = 2n. Given an L-layered matrix A and the polynomial we define the polynomial ffA to be where s(bαi) = ℓn−i and bαi ∼=ν αi. Theorem 5.20. ffA(A) fA := λI + A = αnλn + ··· + α1λ + α0, cfA = bαnλn−1 + ··· +bα2λ +bα1, Sℓ;d = adj(A), where d = n − 1, for any ℓ-layered matrix A. Proof. This is an identity for Mn(Z), using the usual determinant. (cid:3) Proposition 5.21. adj( adj(A)) = An−2 A, where d = n − 1, for any ℓ-layered matrix A. Sℓ;d Questions for further thought: 18 Z. IZHAKIAN, M. KNEBUSCH, AND L. ROWEN Q1. What are all the semiring† PIs of Mn(R)? Specifically, we have the Specht-like question: Q2. Are all semiring† PIs of Mn(R) a consequence of a given finite set? Example 5.22. It is shown in [IzM] that the semiring of 2 × 2 matrices over the max-plus algebra satisfies the semigroup identity (5.6) AB2A AB AB2A = AB2A BA AB2A. The way of proving this identity is essentially based on showing that pairs of polynomials corresponding to compatible entries in the right and the left product above define the same function. This identification is performed by using the machinery of Newton polytopes, and thus is valid also for supertropical polynomials. From the results of [IzM], we also conclude that this identity is minimal. [AkBG] M. Akian, R. Bapat, and S. Gaubert. Max-plus algebra, In: Hogben, L., Brualdi, R., Greenbaum, A., Mathias, R. (eds.) Handbook of Linear Algebra. Chapman and Hall, London, 2006. [AkGG] M. Akian, S. Gaubert, and A. Guterman. Linear independence over tropical semirings and beyond. In Tropical and References [AmL] [BiG] [ItMS] Idempotent Mathematics, G.L. Litvinov and S.N. Sergeev, (eds.), Contemp. Math. 495:1–38, 2009. S.A¿ Amitsur and J. Levitzki. Minimal identities for algebras. Proc. American Mathematical Society 1, 449–463 1950. R. Bieri and R. Groves, The geometry of the set of characters induced by valuations. J. fur die Reine und angevandte Mathematik, 374:168–195, 1984. N. Bourbaki, Alg. Comm. VI, §3, No.1. R. A. Brualdi and H. J. Ryser. Combinatorial matrix theory. Cambridge University Press, 1991. [Bo] [BrR] [CHWW] G. Cortinas, C. Haesemeyer, M. Walker, and C. Weibel, Toric varieties, monoid schemes, and descent. Preprint, 2010. [DeS] [Gat] [Gol] M. Devlin and B. Sturmfels, Tropical convexity. Documenta Mathematica 9 (2004), 1–27, Erratum 205–6. A. Gathmann, Tropical algebraic geometry. Jahresbericht der DMV 108:3–32, 2006. J. Golan, The theory of semirings with applications in mathematics and theoretical computer science, Vol. 54, Longman Sci & Tech., 1992. I. Itenberg, G. Mikhalkin and E. Shustin, Tropical Algebraic Geometry, Oberwolfach Seminars, 35, Birkhauser Verlag, Basel, 2007. Z. Izhakian. Tropical arithmetic and matrix algebra. Commun. in Algebra 37(4),1445–1468, 2009. [Iz] [IzKR1] Z. Izhakian, M. Knebusch, and L. Rowen. A Glimpse at Supertropical Valuation theory An. St. Univ. Ovidius Constanta 19, no. 2, 131–142, 2011. [IzKR2] Z. Izhakian, M. Knebusch, and L. Rowen, Supertropical linear algebra, Pacific J. of Math., to appear. (Preprint at arXiv:1008.0025.) [IzKR3] Z. Izhakian, M. Knebusch, and L. Rowen, Dual spaces and bilinear forms in supertropical linear algebra, Linear and Mult. Algebra, to appear. (Preprint at arXiv:1201.6481.) [IzKR4] Z. Izhakian, M. Knebusch, and L. Rowen. Layered tropical mathematics, preprint at arXiv:0912.1398. (Submitted May 2012) [IzKR5] Z. Izhakian, M. Knebusch, and L. Rowen. Categorical layered mathematics. Contemporary Mathematics, proceedings [IzM] [IzR1] [IzR2] [IzR3] [IzR4] [IzR5] [IzR6] [J] [Ko] [L1] [L2] [L3] [MS] [M] [N] [Par] [Pay1] of the CIEM Workshop on Tropical Geometry, to appear. (Preprint at arXiv:1207.3487.) Z. Izhakian and S. W. Margolis. Semigroup identities in the monoid of 2-by-2 tropical matrices. Semigroup Furom 80(2), 191–218, 2010. Z. Izhakian and L. Rowen, Supertropical algebra. Adv. in Math 225(4), 2222–2286, 2010. Z. Izhakian and L. Rowen, The tropical rank of a tropical matrix. Commun. in Algebra, 37(11), 3912–3927, 2009. Z. Izhakian and L. Rowen, Supertropical matrix algebra, Israel J. Math., 182(1), 383–424, 2011. Z. Izhakian and L. Rowen, Supertropical matrix algebra II: Solving tropical equations. Israel J. Math. 186(1), 69–97, 2011. Z. Izhakian and L. Rowen, Supertropical matrix algebra III: Powers of matrices and generalized eigenspaces. J. of Algebra, 341(1), 125–149, 2011. Z. Izhakian and L. Rowen, Supertropical polynomials and resultants, J. Algebra 324, 1860–1886, 2010. Jacobson, N., Basic Algebra, Freeman, 1980. V.M. Kopytov, Lattice ordered groups (Russian), Nauka, Moscow (1984) G. Litvinov, The Maslov dequantization, idempotent and tropical mathematics: a very brief introduction. J. of Math. Sciences, 140(3),426–444, 2007. G. Litvinov, Dequantization of mathematical structures and tropical idempotent mathematics: An introductory lecture. (See next reference), 2012 G. Litvinov and V.P. Maslov, Tropical and idempotent mathematics: International Workshop. Pncelet Laboratory and Moscow Center for Continuous Mathematical Education, Independent University of Moscow, 2012. D. Maclagan and B. Sturmfels, Tropical Geometry, preprint, 2009. D. Marker, Model theory: An introduction, Springer Graduate texts in mathematics; 217, 2002. A. Niv, Characteric polynomials of supertropical matrices, Commun. in Algebra, to appear, 2012. B. Parker. Exploded fibrations, preprint at arXiv: 0705.2408v1, 2007. S. Payne. Fibers of tropicalizations, Arch. Math., 2010 Correction: preprint at arXiv: ?? [math.AG], 2012. ALGEBRAIC STRUCTURES 19 [Pay2] [R1] [R2] [Sa] [ShSh] [SS] [Vi] [W] [WW] S. Payne. Analytification is the limit of all tropicalizations, preprint at arXiv: 0806.1916v3 [math.AG], 2009. L.H. Rowen. Graduate Algebra: Commutative View. Pure and Applied Mathematics 73, Amer. Math. Soc., 2006. L.H. Rowen. Graduate algebra: A noncommutative view Amer. Math. Soc., 2008. G.E. Sacks, Saturated Model Theory, Mathematical Lecture Noets 80 Benjamin, 1972. E. Sheiner and S. Shnider, An exploded-layered version of Payne's generalization of Kapranov's theorem, preprint, 2012. D. E. Speyer and B. Sturmfels, Tropical mathematics, Math. Mag., 82 (2009), 1631/2-173. O. Viro, Hyperfields for tropical geometry I. Hyperfields and dequantization Preprint at arXiv:math.AG/1006.3034v2 C. Weibel, EGA for monoids. Preprint, 2010. H.J. Weinert and R. Wiegandt, On the structure of semifields and lattice-ordered groups, Periodica Mathematica Hungaria 32 (1-2) (1996), 129–147. Department of Mathematics, Bar-Ilan University, Ramat-Gan 52900, Israel E-mail address: [email protected] Department of Mathematics, NWF-I Mathematik, Universitat Regensburg 93040 Regensburg, Germany E-mail address: [email protected] Department of Mathematics, Bar-Ilan University, 52900 Ramat-Gan, Israel E-mail address: [email protected]
1503.04346	2	1503	2015-09-03T20:46:59	Archimedean classes of matrices over ordered fields	[ "math.RA" ]	Let $(F,\le)$ be an ordered field and let $A,B$ be square matrices over $F$ of the same size. We say that $A$ and $B$ belong to the same archimedean class if there exists an integer $r$ such that the matrices $r A^T A-B^T B$ and $r B^T B-A^T A$ are positive semidefinite with respect to $\le$. We show that this is true if and only if $A=CB$ for some invertible matrix $C$ such that all entries of $C$ and $C^{-1}$ are bounded by some integer. We also show that every archimedean class contains a row echelon form and that its shape and archimedean classes (in $F$) of its pivots are uniquely determined. For matrices over fields of formal Laurent series we construct a canonical representative in each archimedean class. The set of all archimedean classes is shown to have a natural lattice structure while the semigroup structure does not come from matrix multiplication. Our motivation comes from noncommutative real algebraic geometry and noncommutative valuation theory.	math.RA	math	ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS J. CIMPRI C Abstract. Let (F, ≤) be an ordered field and let A, B be square matrices over F of the same size. We say that A and B belong to the same archimedean class if there exists an integer r such that the matrices rAT A − BT B and rBT B − AT A are positive semidefinite with respect to ≤. We show that this is true if and only if A = CB for some invertible matrix C such that all entries of C and C−1 are bounded by some integer. We also show that every archimedean class contains a row echelon form and that its shape and archimedean classes (in F ) of its pivots are uniquely determined. For matrices over fields of formal Laurent series we construct a canonical representative in each archimedean class. The set of all archimedean classes is shown to have a natural lattice structure while the semigroup structure does not come from matrix multiplication. Our motivation comes from noncommutative real algebraic geometry and noncommutative valuation theory. 1. Introduction The notion of the natural valuation of an ordered field was introduced by Baer in [3]. Through [2] it motivated Krull to introduce valuations with non-archimedean value groups in [13]. Krull's valuation theory was extended to skew-fields by Schilling [18] and natural valuations of ordered skew-fields were studied by Con- rad [7] and Holland [11]. For matrices over skew-fields the valuation theory was developed in [8], [17] and [19] but it can be shown that orderings exist only in the 1 × 1 case. On the other hand, partial orderings also exist in other cases but their natural valuations have only been studied in the commutative 1 × 1 case; see [4]. It would be interesting to study other cases, too. We will concentrate on the simplest case, namely matrices over ordered fields with transpose as involution and positive semidefinitness as partial ordering, be- cause the theory is already nontrivial and the results may be of interest in linear algebra. We define a relation (cid:23)n on Mn(F ) by A (cid:23)n B if and only if there exist an integer r such that rBT B − AT A is positive semidefinite with respect to the order- ing of F . This relation is reflexive and transitive, but it need not be antisymmetric. The corresponding equivalence relation ∼n on Mn(F ) defined by A ∼n B if and only if A (cid:23)n B and B (cid:23)n A is called archimedean equivalence and the elements of the factor set Mn(F )/∼n are called archimedean classes. The canonical projection vn : Mn(F ) → Mn(F )/∼n is called the natural valuation. Date: October 21, 2018. 2010 Mathematics Subject Classification. 12J15, 13A18, 13J30, 15B33. Key words and phrases. ordered fields, positive semidefinite matrices, valuations and their generalizations. Supported by grant P1-0222 of the Slovenian Research Agency. 1 2 J. CIMPRI C The paper is organized as follows. In section 2 we recall the construction of the natural valuation of an ordered field and generalize it to matrices over ordered fields. Our construction does not produce a value function in the sense of Morandi [17], so we carefully explain the relationship between the two. In section 3 we show that each archimedean class in Mn(F ) contains a row echelon form and that its shape and the natural valuations of its pivots are uniquely determined. In section 4 we show that the relation (cid:23)n induces a lattice structure on Mn(F )/∼n. In section 5 we characterize relations (cid:23)n and ∼n by divisibility. In section 6 we try to find a canonical representative in each archimedean class. This works for matrices over formal Laurent series fields but not in general. In section 7 we discuss different ways of introducing a semigroup structure on Mn(F )/∼n. 2. Preliminaries on natural valuations We will recall the construction of the natural valuations of an ordered field and generalize it to an ordered ring with involution. We will illustrate our construction on the matrix ring Mn(F ) with transpose involution ordered in two different ways. The natural valuation of the first ordering is a value function in the sense of Morandi while the natural valuation of the second ordering does not have an analogue in noncommutative valuation theory. 2.1. Natural valuations of ordered fields. Let (F,≤) be an ordered field. Clearly, char F = 0, and so F contains Q. For every a, b ∈ F write a (cid:23) b iff there exists r ∈ N such that a ≤ rb (or equivalently, a2 ≤ r2b2). Write a ∼ b iff a (cid:23) b and b (cid:23) a and note that this relation is a congruence on the multiplica- tive semigroup of F . The congruence classes are also called archimedean classes of (F,≤). The factor semigroup F/ ∼ is linearly ordered by [a] (cid:23) [b] iff a (cid:23) b. The canonical projection from F to F/ ∼ will be denoted by v and called the natural valuation of ≤. We will write F/∼ = Γ ∪ {∞} where ∞ := {0} is the congruence class of zero and Γ is the set of all other congruence classes. Clearly, Γ is an abelian group with [a] + [b] := [ab] as operation and 0 := [1] as neutral element. We will call it the value group of v. a Let us now briefly sketch an alternative construction. We say that an element a ∈ F is bounded if there exists r ∈ N such that −r ≤ a ≤ r. An element a ∈ F is infinitesimal if −r ≤ a ≤ r for every r ∈ N. The set V of all bounded elements is a valuation subring of F and the set m of all infinitesimal elements is the only maximal ideal of V . The set U := V \ m is a subgroup of the multiplicative group F × = F \ {0} and the factor group Γ := F ×/U is linearly ordered by aU (cid:23) bU iff b ∈ V . The natural valuation of (F,≤) is then the canonical projection v : F × → Γ extended to F by v(0) = ∞ where ∞ 6∈ Γ is larger from all elements from Γ. For every ordered group G, the field R(G)) of formal Laurent series can be ordered by the sign of the lowest nonzero coefficient. The corresponding value group is G and the natural valuation assigns to each element the least element of G with nonzero coefficient. Hahn's embedding theorem for ordered fields says that every ordered field with value group Γ has an order-preserving embedding into R((Γ)). See [10] for the origins; a complete proof appeared much later. In [14] it is shown that the real closure of an ordered field has a truncation closed embedding into R((¯Γ)) (where ¯Γ is the division hull of Γ) which maps the field into R((Γ)). An ordered field is archimedean iff it has no unbounded elements iff it has no nonzero infinitesimal elements iff Γ has only one element iff it is a subfield of R. ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS 3 2.2. Ordered rings with involution. The construction from the previous section can be extended to ordered rings with involution. Let R be a ring with involution ∗ and let ≤ be a relation on the subset S := {a ∈ R a∗ = a} which is reflexive, transitive and satisfies 0 ≤ a∗a for every a ∈ R. We define two relations on R by a (cid:23) b iff there is r ∈ N such that a∗a ≤ rb∗b, a ∼ b iff a (cid:23) b and b (cid:23) a. Clearly, (cid:23) is reflexive and transitive and ∼ is an equivalence relation. As above, we say that the canonical projection v : R → R/ ∼ is the natural valuation of ≤. Its complement is denoted by Γ and Consider the equivalence class ∞ := v(0). called the value set of v. Whenever we work with several rings or several orderings, we will add a suitable index to ≤, (cid:23), ∼, v or Γ. Note that v(a) = ∞ iff a∗a = 0, so Γ can be empty. Lemma 2.2 will show that the factor set R/∼ = Γ ∪ {∞} need not have any algebraic structure because ∼ need not be a congruence relation on the multiplicative semigroup of R and it need not be compatible with the involution. Moreover, the partial ordering of R/∼ induced by (cid:23) need not be a linear ordering. Example 2.1. If R is a partially ordered commutative ring with positive squares then Γ ∪ {∞} has the structure of a partially ordered commutative semigroup. In particular, if R is a partially ordered field with positive squares then Γ has the structure of a partially ordered abelian group. See [4] for additional information. Let R be the ring Mn(F ) of all n × n matrices with transpose as involution. Then S is the vector space Sn(F ) of all symmetric n× n matrices. We will consider two orderings of S: C ≤ D iff tr C ≤ tr D in subsection 2.3 and C ≤ D iff D − C is positive semidefinite in subsection 2.4. The first ordering is uncommon but its natural valuation is the most common valuation on R. The second ordering is the most common ordering of R but its natural valuation is very uncommon. The rest of the paper will then give more details about this uncommon valuation. 2.3. Noncommutative valuation theory. Let (F,≤F ) be an ordered field with natural valuation vF : F → ΓF ∪ {∞} and let n be a natural number. We assume that Sn(F ) is ordered by C ≤ D iff tr C ≤F tr D. The corresponding ordering (cid:23) of Mn(F ) is then A (cid:23) B iff tr AT A ≤F r tr BT B for some r ∈ N. We will denote the natural valuation of ≤ by w. The mapping c → cIn identifies F with a subset of Mn(F ). Moreover, we have that cIn (cid:23) dIn iff c (cid:23)F d. It follows that ΓF can be identified with a subset of Γ. To show that Γ = ΓF , it suffices to observe that for every matrix A = [aij] ∈ Mn(F ), (1) where kAk∞ := maxi,j aij is the max norm of A and In is the identity matrix. Furthermore, relation (1) implies that for every A ∈ Mn(F ), (2) A ∼ kAk∞ · In, w(A) = vF (kAk∞). Since vF is the natural valuation, vF (max{x, y}) = vF (x+y) = min{vF (x), vF (y)} for every x, y ∈ F ≥0. It follows that for every matrix A = [aij ] we have that (3) vF (aij ). vF (kAk∞) = min i,j Relations (2) and (3) imply that the natural valuation w : Mn(F ) → Γ ∪ {∞} satisfies [19, Definition 1.2], i.e. w is a vF -value function on Mn(F ) which for every 4 J. CIMPRI C A, B ∈ Mn(F ) satisfies w(A)+w(B) ≤ w(AB) and w(AT A) = 2w(A). By the proof of [19, Theorem 2.2], w is the only such function. Therefore w can be considered as the canonical extension of vF from F to Mn(F ). 2.4. The main example. A matrix A ∈ Sn(F ) is positive semidefinite iff vT Av ≥ 0 for every v ∈ F n. The set of all positive semidefinite matrices will be de- noted by S + It defines a partial ordering ≤n of Sn(F ) by A ≤n B iff B − A ∈ S + n (F ). The corresponding relations on Mn(F ) are defined by A (cid:23)n B iff AT A ≤n rBT B for some r ∈ N and A ∼n B iff A (cid:23)n B and B (cid:23)n A. We will call ∼n archimedean equivalence. The natural valuation of ≤n will be denoted by vn : Mn(F ) → Mn(F )/∼n where we decompose Mn(F )/∼n into Γn and ∞ = {0n}. Note that AT A ≤ rBT B implies tr AT A ≤F r tr BT B, thus vn(A) (cid:23)n vn(B) implies w(A) (cid:23) w(B). Therefore, we have a mapping φ : Γn ∪ {∞} → ΓF ∪ {∞} which is order-preserving and makes the following diagram commutative: n (F ). Mn(F ) vn ✲ Γn k · k∞ w φ ❄ vF F ✲ ❄ ✲ ΓF As above, the mapping vF (c) 7→ vn(cIn) identifies ΓF with a subset of Γn. In other words, we can consider vn as a refinement of w. Let us show that the properties of vn are much worse than the properties of w. Lemma 2.2. Notation from above. If n ≥ 2, then (Namely, A ∼n B always implies AC ∼n BC but not CA ∼n CB.) (1) ∼n is not a congruence relation on the multiplicative semigroup of Mn(F ). (2) (cid:23)n is not a linear ordering of Γn ∪ {∞}. (3) There exists A ∈ Mn(F ) such that vn(A) 6= vn(AT ). Proof. To prove (1) for n = 2, note that 1 0 (cid:21) ∼2 (cid:20) 1 (cid:20) 0 1 0 0 1 (cid:21) but (cid:20) 1 0 0 0 (cid:21)(cid:20) 0 1 1 0 (cid:21) 6∼2 (cid:20) 1 0 0 0 (cid:21)(cid:20) 1 0 0 1 (cid:21) . 0 1 To prove (2) and (3) for n = 2, note that A = (cid:20) 0 0 (cid:21) satisfies A 6(cid:23)2 AT and AT 6(cid:23)2 A. To get examples for larger n just add some zero rows and columns. (cid:3) Lemma 2.3 characterizes relations (cid:23)n and ∼n for archimedean fields. Note that A ≤n (tr A)In (4) for every A ∈ S + Lemma 2.3. Notation from above. Pick any A, B ∈ Mn(F ). If A (cid:23)n B then ker A ⊇ ker B. If F is an archimedean field then the converse is also true. If F is a non-archimedean field then the converse fails already for n = 1. n (F ) since (tr A)In − A = P1≤i<j≤n(Eij − Eji)T A(Eij − Eji). ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS 5 Proof. The first claim is clear. To prove the second claim note that ker A ⊇ ker B implies that there is some C of appropriate size such that A = CB. The trace inequality (4) implies that AT A = BT C T CB ≤n tr(C T C)BT B. If F is an archimedean field, then tr C T C is bounded by some natural number, so A (cid:23)n B. If F is a non-archimedean field, then it contains some unbounded element t. Note that ker[t] = ker[1] = 0 but [t] 6(cid:23)1 [1]. (cid:3) 3. Row echelon forms We will show that each archimedean class contains a row echelon form whose shape and natural valuations of its pivots are uniquely determined. 3.1. QR decomposition. Let M∗,n(F ) := S∞ m=1 Mm,n(F ) be the set of all matri- ces over F with n columns and arbitrary many rows. For technical reasons we will also study extensions of the relations (cid:23)n and ∼n from Mn(F ) to M∗,n(F ). For every A, B ∈ M∗,n(F ) we write, as above, A (cid:23)n B iff AT A ≤n rBT B for some r ∈ N and A ∼n B iff A (cid:23)n B and B (cid:23)n A. By the Gram-Schmidt algorithm, every subspace of F n has an orthogonal basis with respect to the standard inner product. However, we do not always have an orthonormal basis unless F is pythagorean. Instead, we will use normalization in the ℓ∞-norm kvk∞ := maxi vi. We say that a matrix C is a row echelon form if all zero rows of C are at the bottom of C and if for each i the first nonzero element in the (i + 1)-th nonzero row is on the right-hand side of the first nonzero element in the i-th nonzero row. The first nonzero element in the i-th nonzero row is also called the i-th pivot. Lemma 3.1 is a variant of QR decomposition. Lemma 3.1. For every A ∈ Mm,n(F ) with rank r ≥ 1 there exists matrices Q ∈ Mm,r(F ) and R ∈ Mr,n(F ) such that (1) A = QR. (2) The columns of Q are orthogonal and ℓ∞-normalized. (3) R is a row echelon form with positive pivots and no zero rows. i = vki − Pi−1 j=1 hvki ,w′ ii ii w′ hw′ i,w′ Moreover, Q and R are uniquely determined by A. We also have that Q ∼r Ir. Proof. Let vi be the i-th column of A for each i = 1, . . . , n and let r be the rank of A. Let k1 be the first index such that vk1 6= 0, k2 the first index such that vk2 6∈ span{vk1}, k3 the first index such that vk3 6∈ span{vk1 , vk2}, etc. Now set 1 = vk1 and w′ w′ ik∞ for i = 1, . . . , r and note that Q := [w1 . . . wr] satisfies (2). Now pick cij ∈ F such that vj = Pr i=1 cij wi for j = 1, . . . , n and note that R := [cij] satisfies (1) and (3). The matrix D := QT Q is diagonal since wi are orthogonal and it satisfies Ir ≤n D ≤n mIn since wi are ℓ∞-normalized. It follows that Q ∼n Ir. Suppose that QR = Q′R′, where Q = [w1 . . . wr] and Q′ = [z1 . . . zr] satisfy (1) and R = [cij] and S = [dij ] satisfy (2). Let ci,ki and dj,lj be the pivots of R and R'. The t-th column of QR = Q′R′ is c1tw1 + . . .+ cttwt = d1tz1 + . . .+ dttzt. For t = min{k1, l1} one gets that w1 = z1; for t = min{k2, l2} one gets that w2 = z2; and so on. i for i = 2, . . . , r. Write wi = w′ i/kw′ (cid:3) The natural embedding j : Mn(F ) → M∗,n(F ) satisfies the property A (cid:23)n B iff j(A) (cid:23)n j(B) for every A, B ∈ Mn(F ). In other words, j is an o-embedding. It 6 J. CIMPRI C follows that j induces a mapping j′ : Mn(F )/ ∼n→ M∗,n(F )/ ∼n which is also an o-embedding. Corollary 3.2 will show that every element of M∗,n(F )/∼n contains a square matrix which implies that the mapping j′ is onto. Therefore, j′ is an isomorphism of partially ordered sets. Corollary 3.2. Every archimedean class of M∗,n(F ) contains a square matrix which is a row echelon form. Proof. Pick any A ∈ M∗,n(F ) of rank r. The claim is clear if r = 0. Otherwise we can decompose A = QR where Q ∼r Ir and R is a row echelon form. It follows that QR ∼n R, so R belongs to the archimedean class of A. Since R ∼n (cid:20) R 0 (cid:21) for any number of zero rows, the claim follows. (cid:3) 3.2. Shape. For every row echelon form C ∈ Mk,n(F ) we define its shape shape(C) = {(i, j) ∈ Nk × Nn ∃j0 ≤ j : cij0 6= 0}. where Nn := {1, . . . , n}. It consists of all positions that lie above the "staircase". Proposition 3.3. Let A, B ∈ M∗,n(F ) be row echelon forms. If A (cid:23) B then shape(A) ⊆ shape(B). Moreover, for every pivot position (i, ki) of shape(B) we have that v(ai,ki ) (cid:23) v(bi,ki ). Proof. Suppose that for some row echelon forms A, B ∈ M∗,n(F ) we have that vn(A) (cid:23) vn(B) but shape(A) 6⊆ shape(B). Let i be the smallest integer such that the i-th row of shape(A) is not contained in shape(B) (which can have less than i rows) and let j be the smallest integer such that (i, j) ∈ shape(A). It follows that (i, j) is a pivot position of shape(A) (so aij 6= 0) and (i, j) 6∈ shape(B) and the j-th column of shape(B) does not contain any pivot position of shape(B) (otherwise shape(A) would have a step of size ≥ 2.) Recall that for every elementary matrix E, C → CE is an elementary column transformation of the matrix C. Let us perform the standard Gauss algorithm on the columns of B. We use the first pivot to kill all other elements in the first row of B, then we use the second pivot to kill all other elements in the second row of B and so on. Let Q be the product of all elementary matrices that we used in this procedure. Note that B = BQ has the same shape and the same pivots as B but all non-pivot elements are zero. In particular, the j-th column of B is zero. We claim that the j-th column of A := AQ is also zero. The assumption vn(A) (cid:23)n vn(B) implies that there exists r ∈ N such that AT A ≤ rBT B. It follows that AT A ≤ r BT B which implies the claim. Finally, note that the (i, j)-th element of A is aij because the i-th pivot of B (if it exists) lies on the right-hand side of aij , and so the elementary column transformations from the product Q did not act on aij. Since the j-th column of A is zero, it follows that aij = 0 which is not possible by the choice of (i, j). This contradiction finishes the proof of the first part. To prove the second part, let A and B be as above. Write u for the ki-th column of A and v for the ki-th column of B. As above, AT A ≤ r BT B implies that uT u ≤ rvT v for some r ∈ N. Since u = (u1, . . . , ui−1, ai,ki , 0, . . . , 0)T for some u1, . . . , ui−1 ∈ F and v = (0, . . . , 0, bi,ki, 0, . . . , 0), it follows that a2 , i.e. v(ai,ki ) (cid:23) v(bi,ki ). (cid:3) i,ki ≤ rb2 i,ki and A ⊒n B iff A ≤ rB for some r ∈ N A ≈n B iff A ⊒n B and B ⊒n A. ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS 7 Corollary 3.4. Let A, B ∈ M∗,n(F ) be row echelon forms. If vn(A) = vn(B) then shape(A) = shape(B) and for each i, the natural valuation of the i-th pivot of A is equal to the natural valuation of the i-th pivot of B. Corollary 3.4 implies that we can define the shape of an archimedean class. 4. Γn is a lattice We will show that the partial ordering induced by (cid:23)n on the value set Γn is in fact a lattice ordering, i.e. each finite subset of Γn has supremum and infimum. 4.1. Lattice structure of positive semidefinite matrices. Let (F,≥) be an ordered field. For every A, B ∈ S + n (F ) write n (F ), A 7→ AT A satisfies A (cid:23)n B iff i(A) ⊒n i(B), The mapping i : M∗,n(F ) → S + i is an o-embedding. Therefore i induces a mapping i′ : M∗,n(F )/ ∼n→ i.e. n (F )/ ≈n which is also an o-embedding. The plan is to show that S + S + n (F )/ ≈n is a lattice and then pull this result back to M∗,n(F )/ ∼n which is isomorphic to Γn ∪ {∞}. One can also deduce Lemma 2.3 from Lemma 4.1 this way. Lemma 4.1. Suppose that A, B ∈ S + n (F ). If A ⊒n B then also ker A ⊇ ker B. If F is an archimedean field, then we also have the converse. For non-archimedean fields the converse already fails for n = 1. Proof. If v ∈ ker B and A ⊒n B then clearly vT Av = 0. By [15, Corollary 2.4], there exists an invertible P ∈ Mn(F ) and a diagonal D ∈ Sn(F ) with entries di such that A = P T DP . It follows that Pi di(P v)2 i = 0. Since 0 ≤n A, also 0 ≤n D, which implies that di(P v)i = 0 for each i. Therefore DP v = 0, and so v ∈ ker A. Suppose now that F is archimedean and ker A ⊇ ker B. Pick an invertible Q ∈ Mn(F ) such that E := QT BQ is a diagonal matrix with nonzero entries e1, . . . , ek. Write C := QT AQ and note that ker C ⊇ ker E implies that cij = 0 if either i > k or j > k. It follows that C = M ⊕ 0n−k ≤n (tr M )Ik ⊕ 0n−k ≤ uD where u ∈ N is such that min{e1,...,ek} ≤ u. Therefore A ≤n uB. tr M (cid:3) By the Gram-Schmidt algorithm (or [15, Proposition 1.3]), every subspace U of F n satisfies U ⊕ U ⊥ = F n. It follows that for every C ∈ Sn(F ) we have that F n = im C⊕ ker C. We define its Moore-Penrose inverse C + = (Cim C )−1⊕ 0ker C . For every A, B ∈ S + n (F ) we define their parallel sum A : B = A(A + B)+B. The basic properties of A : B from [1] carry over from R to general ordered fields. In particular, we have the following generalization of [1, Corollary 21]. Lemma 4.2. If A, B, C ∈ S + Proof. Note that ker(A + C) ⊆ ker C. Namely, (A + C)x = 0 implies that xT Ax = xT Cx = 0 which implies that Cx = 0 as in the proof of Lemma 4.1. It follows that im C ⊆ im(A+C), which implies that C(A+C)+(A+C) = C = (A+C)(A+C)+C. If A ≤n B, these identities imply that 0 ≤n C(B + C)+(B− A)(B + C)+C + C((B + C)+ − (A + C)+)(A + C)((B + C)+ − (A + C)+)C = C(A + C)+C − C(B + C)+C = C − C(B + C)+C − (C − C(A + C)+C) = B : C − A : C. n (F ) and A ≤n B then A : C ≤n B : C. (cid:3) 8 J. CIMPRI C Proposition 4.3. For every [A], [B] ∈ S + their least upper bound are given by the formulas n (F )/≈n, their greatest lower bound and [A] ⊓n [B] = [A + B] and [A] ⊔n [B] = [A : B]. Proof. Pick A, B, C ∈ S + n . Since A ≤n A + B and B ≤n A + B we have that A ⊒n A + B and B ⊒n A + B. If A ⊒n C and B ⊒n C, then A ≤ rC and B ≤ sC for some r, s ∈ N. If follows that A + B ≤ (r + s)C, thus A + B ⊒n C. This proves the first part. To prove the second part, note that A : B ≤n A and A : B ≤n B by [1, Lemma 18], which implies that A : B ⊒n A and A : B ⊒n B. If C ⊒n A and C ⊒n B then C ≤ tA and C ≤ tB for some t ∈ N. By Lemma 4.2, we have that 1 2 C = C : C ≤n (tA) : C ≤n (tA) : (tB) = t(A : B) which implies that C ⊒n A : B. (cid:3) For every A, B ∈ S + n (F ), we clearly have that ker(A + B) = ker A∩ ker B and by [1, Lemma 3] we also have that ker(A : B) = ker A + ker B. By Lemma 4.1, we can define the kernel of an equivalence class by ker[A] := ker A. By Proposition 4.3, ker is a lattice homomorphism from (S + n (F )/ ≈n,⊔n,⊓n) to (Subspaces(F n), +,∩). 4.2. Lattice structure of rectangular matrices. Proposition 4.4 shows that Γn ∪ {∞} is a lattice. Proposition 4.4. For every two elements vn(A), vn(B) ∈ Γn ∪ {∞} their greatest lower bound and their least upper bound are given by the formulas B (cid:21)) and vn(A)∨nvn(B) = vn((cid:20) A(AT A + BT B)+BT B vn(A)∧nvn(B) = vn((cid:20) A B(AT A + BT B)+AT A (cid:21)). Proof. Let i′ be as above. By Lemma 4.3, we have that i′(vn((cid:20) A B (cid:21))) = [AT A + BT B]n = [AT A] ⊓n [BT B] = i′(vn(A)) ⊓n i′(vn(B)) and i′(vn((cid:20) A(AT A + BT B)+BT B B(AT A + BT B)+AT A (cid:21))) = = [AT A(AT A + BT B)+BT B(AT A + BT B)+AT A+ +BT B(AT A + BT B)+AT A(AT A + BT B)+BT B]n = = [(AT A + BT B)(AT A + BT B)+(AT A : BT B)]n = = [AT A : BT B]n = [AT A] ⊔n [BT B]n = i′(vn(A)) ⊔n i′(vn(B)). Since i′ is an o-imbedding, this implies the formulas. (cid:3) By Propositions 4.3 and 4.4, the mapping ker◦i′ is a lattice homomorphism from (M∗,n(F )/∼n,∨n,∧n) to (Subspaces(F n), +,∩). Corollary 4.5. For every A, B ∈ M∗,n(F ) of the same size we have that vn(A + B) (cid:23) vn(A) ∧n vn(B). Proof. This follows from Lemma 4.4 and 0 ≤n (A − B)T (A − B) = 2(cid:20) A B (cid:21)T (cid:20) A B (cid:21) − (A + B)T (A + B) (cid:3) Corollary 4.6. For every A.B ∈ M∗,n(F ), we have shape(vn(A))∪shape(vn(B)) ⊆ shape(vn(A) ∧2 vn(B)). The inclusion can be strict. ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS 9 11E11 + ET Proof. The first part follows from Proposition 3.3. To prove the second part, note that the relation ET 2 I2 implies that v2(E11)∧2 v2(E12) = v2(I2). Therefore, shape(v2(E11) ∧2 v2(E12)) = shape(v2(I2)) = {(1, 1), (1, 2), (2, 2)}. On the other hand, shape(vn(A)) ∪ shape(vn(B)) = {(1, 1), (1, 2)}. 5. Bounded and bibounded elements 12E12 = I T (cid:3) Let (F,≤) be an ordered field and n an integer. We say that a matrix A ∈ M∗,n(F ) is bounded (w.r.t. ≤n) if A (cid:23)n In where In is the identity matrix. We say that A ∈ M∗,n(F ) is bibounded if A ∼n In. We will also use this terminology for scalars and vectors which can be identified with elements of M∗,1(F ). A scalar a is bounded iff v(a) ≥ 0. It is bibounded iff v(a) = 0. Every vector v clearly satisfies v ∼1 kvk∞ where kvk∞ is the ℓ∞ norm of v. It follows that a vector is bounded iff all its components are bounded and that a vector is bibounded iff its ℓ∞ norm is bibounded. We can make every nonzero vector bibounded by dividing it with its ℓ∞ norm. 5.1. A characterization of (cid:23)n and ∼n. The aim if this section is to characterize the relations (cid:23)n and ∼n in terms of divisibility. Proposition 5.1. For every A ∈ Mk,n(F ) and B ∈ Ml,n(F ) we have that A (cid:23)n B iff A = CB for some bounded C ∈ Mk,l(F ). Proof. If A = CB for some bounded C ∈ Mk,l(F ) then A = CB (cid:23)n IlB = B. Conversely, suppose that A (cid:23)n B. Since F l = im B ⊕ (im B)⊥ w.r.t. to the standard inner product, we can decompose every v ∈ F l as v = Bx + y, yT Bx = 0. Let us define Cv = Ax. To show that C is well-defined note that Bx + y = Bx′ + y′ implies Bx = Bx′ and y = y′. Thus, x−x′ ∈ ker B ⊆ ker A which implies Ax = Ax′. To show that C is bounded, pick any v ∈ F l and decompose it as v = Bx + y with yT Bx = 0. We have that vT C T Cv = xT AT Ax ≤ rxT BT Bx ≤ rxT BT Bx+ryT y = r(xT BT Bx + yT Bx + xT BT y + yT y) = rvT v. It follows that C T C ≤l rIl. Proposition 5.2. For every A ∈ Mk,n(F ) and B ∈ Ml,n(F ) where k ≥ l, we have that A ∼n B iff A = CB for some bibounded C ∈ Mk,l(F ). Proof. Suppose that A = CB for some bibounded C ∈ Mk,l(F ). Since C ∼l Il, it follows that A = CB ∼n IlB = B. Since sIl ≤l C T C ≤l rIl for some r, s ∈ N, it follows that sBT Bn ≤ BT C T CB ≤n rBT B which implies that A ∼n B. Conversely, if A ∼n B, then ker A = ker B. It follows that t := dim(im B)⊥ = l − dim im B = l − n + dim ker B ≤ k − n + dim ker A = dim(im A)⊥ =: t′. Pick an orthogonal basis u1, . . . , ut of (im B)⊥ and an orthogonal basis v1, . . . , vt′ of (im A)⊥. Now make all ui and vj bibounded by dividing them with suitable scalars. i=1 αiui. We define Cz := Every element z ∈ F l can be written as z = Bx + Pt Ax +Pt Let us show that C is bibounded. Since A ∼n B and ui ∼1 1 ∼1 vi for every i = 1, . . . , t, we can pick r, s ∈ N such that sBT B ≤n AT A ≤n rBT B and suT i ui ≤ vT i vi ≤ ruT i ui for every i = 1, . . . , t. We claim that szT z ≤ zT C T Cz ≤ rzT z for every z ∈ F l which implies that sIl ≤ C T C ≤ rIl. Pick any z = Bx + y ∈ F l where y = P αiui ∈ (im B)⊥. We have that zT z = xT BT Bx + yT Bx + xT BT y + yT y = xT BT Bx + yT y = xT BT Bx + Pt i ui. On the other hand, zT C T Cz = i=1 αivi. Since ker A = ker B, C is well-defined and one-to-one. i=1 α2 (cid:3) i uT 10 J. CIMPRI C xT AT Ax + (Cy)T Ax + (Ax)T Cy + yT C T Cy = xT AT Ax + yT C T Cy = xT AT Ax + (cid:3) i vT i vi. The claim follows. Pt i=1 α2 5.2. A characterization of bounded and bibounded matrices. We will char- acterize bounded and bibounded matrices in terms of their entries; see Proposition 5.5. We start with some preparation. The following is well-known. n (F ) and A ≤n B and A is invertible, then B is Lemma 5.3. If A, B belong to S + also invertible, A−1, B−1 ∈ S + Proof. Note that B is invertible since A ≤n B implies ker A ⊇ ker B by Lemma 4.1. Now use A−1 − B−1 = (A−1 − B−1)A(A−1 − B−1) + B−1(B − A)B−1. Let us characterize the analogues of bounded and bibounded elements in S + n (F ) and B−1 ≤n A−1. (cid:3) n (F ). Lemma 5.4. For every matrix A ∈ S + n (F ) we have the following. (1) A ⊒n In iff all diagonal entries of A are bounded. (2) A ≈n In iff A ⊒n In and det A is bibounded. Proof. One direction of claim (1) is clear and the other follows from A ≤ (tr A)In. Suppose that A ⊒n In and det A is bibounded. By (1) all diagonal entries of A are bounded. Since a2 ij ≤ aiiajj for all i, j, it follows that nondiagonal entries of A are also bounded. Therefore all minors of A are bounded. The assumption that det A is bibounded implies that A−1 = (det A)−1Cof(A) exists and all its entries are bounded. By (1) it follows that A−1 ⊒n In which implies that In ⊒n A. To prove the other direction of claim (2), suppose that In ⊒n A. For k = 1, . . . , n write Ak for the upper left k × k corner of A. By assumption, there exists s ∈ N such that sIn ≤n A, and so sIk ≤n Ak for all k. Lemma 5.3 implies that Ak is invertible and A−1 k )kk ≤ s−1 for all k. Now, (det A)−1 is bounded since (det A)−1 = (det A1)−1Qn det Ak ≤ s−n. (cid:3) k ≤ s−1Ik for every k. Thus, det Ak−1 = (A−1 det Ak−1 det Ak k=2 Proposition 5.5 is a corollary of Lemma 5.4. Proposition 5.5. For every element A ∈ Mm,n(F ) we have the following. (1) A is bounded iff all entries of A are bounded. (2) A is bibounded iff A is bounded, m ≥ n, and at least one of the n×n minors of A is bibounded. A square matrix A is bibounded iff it is invertible and both A and A−1 are bounded. j=1 a2 Proof. The diagonal entries of AT A are Pn ij where i = 1, . . . , n. They are bounded iff all aij are bounded. Claim (1) now follows from Lemma 5.4. To prove claim (2) note that, by Lemma 5.4, a matrix A ∈ Mm,n(F ) is bibounded iff it is bounded and det AT A is bibounded. If m < n, then det AT A = 0, so it is not bibounded. The Binet-Cauchy theorem (see [9], for example) implies that det AT A is equal to the sum of squares of all n × n minors of A. It follows that det AT A is bibounded iff the n × n minor of A of the largest absolute value is bibounded iff at least one of the n × n minors of A is bibounded. (cid:3) Proposition 5.5 implies that a matrix A ∈ Mn(F ) is bounded iff w(A) ≥ 0. However, condition w(A) = 0 is not equivalent to biboundedness. ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS 11 5.3. Bibounded elementary matrices. Recall that the Gauss algorithm con- sists of a series of elementary row tranformations that can be represented as left multiplications by elementary matrices Eij(α), Ei(α) and Pij where α ∈ F . Lemma 5.6. For every α ∈ F , we have the following. (1) A matrix of the form Eij (α) is bibounded iff α is bounded. (2) A matrix of the form Ei(α) is bibounded if α is bibounded. (3) A matrix of the form Pij is always bibounded. Proof. (1) By Lemma 5.5, Eij (α) is bounded iff α is bounded. Since Eij (α)−1 = Eij(−α) and α is bounded iff −α is bounded, the claim follows. the claim follows from the definition of a bibounded element of F . (2) By Lemma 5.5, Ei(α) is bounded iff α is bounded. Since Ei(α)−1 = Ei(1/α), (3) Follows from Lemma 5.5. (cid:3) We already know that for every A ∈ Mk,n(F ) there exists a bibounded Q ∈ Mk(F ) such that QA is a row echelon form. We can also prove this result by a bibounded version of Gauss algorithm. If the first column of A contains only zeros then move to the next column. Otherwise pick in the first column an element of the largest absolute value (the first pivot) and move it to the first row by an appropriate P1j . Now kill all elements below the pivot by Ej1(−aj1/a11) where j = 2, . . . , n. By the choice of the pivot, the elements −aj1/a11 are bounded. Thus the matrices Ej1(−aj1/a11) are bibounded. Finally move to the next column. If all elements from a22 to an2 are zero, then move to the next column. Otherwise pick an element of the largest absolute value (the second pivot) and move it to the second row by an appropriate P2j. Now kill all elements below the pivot with Ej2(−aj2/a22), j = 3, . . . , n, which are bibounded by the choice of the pivot. Finally, move to the next column. Continue until you run out of columns. The elements above a pivot cannot be killed unless their valuation is greater or equal to the valuation of the pivot. This is already clear on the matrix E12(a). However, if A is bibounded and square, then the corresponding row echelon form is also bibounded and square. By Lemma 5.5 this can only happen if it has bounded entries and bibounded determinant. It follows that its diagonal entries are also bibounded. Therefore we can use a bibounded version of Gauss algorithm to kill all elements above the diagonal. Finally, a bibounded diagonal matrix is clearly a product of bibounded Ei(α). This proves Proposition 5.7. Proposition 5.7. A square matrix over F is bibounded iff it is a product of bi- bounded elementary matrices. Remark 5.8. For every A ∈ M∗,n(F ) and every row echelon form B ∈ M∗,n(F ), A (cid:23)n B iff (cid:20) B A (cid:21) ∼n (cid:20) B 0 (cid:21) iff bibounded Gauss on (cid:20) B A (cid:21) gives (cid:20) B 0 (cid:21). 6. Archimedean canonical forms We want to choose a canonical representative in each archimedean class of M∗,n(F ). This works for Laurent series fields but it does not work in general. Let Γ be a linearly ordered Abelian group and let F = R((Γ)) be the field of formal Laurent series with real coefficients and with exponents in Γ. A row echelon 12 J. CIMPRI C form C ∈ M∗,n(F ) with pivots ci,ki , i = 1, . . . , r, is called an archimedean canonical form if it has no zero rows and there exist m1, . . . , mr ∈ Γ such that (i) ci,ki = tmi and (ii) for every j < ki, cj,ki has no terms with exponents ≥ mi. Explicitly, C = 0 0 0 ...   . . . . . . . . . tm1 0 0 ... c1,k1+1 0 0 ... . . . . . . . . . c1,k2 tm2 0 ... c1,k2+1 c2,k2+1 0 ... . . . . . . . . . c1,k3 c2,k3 tm3 ... c1,k3+1 c2,k3+1 c3,k3+1 ... . . . . . . . . . . . .   Note that an archimedean canonical form contains more information than just the information about the archimedean classes of its entries. Proposition 6.1. For every nonzero A ∈ M∗,n(F ) there exists a unique archimedean canonical form C ∈ M∗,n(F ) such that A ∼n C. Proof. By Proposition 3.1 we may assume that A is a row echelon form with no zero rows. Each pivot ai,ki of A can be decomposed uniquely as ai,ki = uitmi where ui is bibounded. If we divide each nonzero row of A with ui we get a matrix B such that B ∼n A and bi,ki = tmi. Now we perform bibounded row transformations of B which use the pivot tmi to kill all terms with exponents ≥ mi in all entries above the pivot. The result is an archimedean canonical form C such that C ∼n B. This proves the existence part. To prove uniqueness pick another archimedean canonical form D ∼n A. By Lemma 2.3, C and D have the same rank and thus the same number of rows, say r. By Corollary 3.4, C ∼n D implies that C and D have the same shape and the same pivots in each row. By Proposition 5.2, there exists a bibounded Q ∈ Mr(F ) such that D = QC. If q1, . . . , qr are the columns of Q and e1, . . . , er are the columns of Ir, then D = QC implies that (5) (6) (7) tm1 q1 = tm1e1 c1,k2q1 + tm2 q2 = d1,k2 e1 + tm2e2 c1,k3q1 + c2,k3q2 + tm3 q3 = d1,k3 e1 + d2,k3 e2 + tm3 e3 ... Equation (5) implies that q1 = e1. Equation (6) implies that (c1,k2 − d1,k2 )e1 = tm2(e2 − q2). Since all powers that appear in the components of the left-hand side are < m2 and all powers that appear in the components of the right-hand side are ≥ m2, it follows that c1,k2 = d1,k2 and q2 = e2. Similarly, equation (7) implies that (c1,k3 − d1,k3 )e1 + (c2,k3 − d2,k3)e2 = tm3 (e3− q3). Since all powers that appear in the components of the left-hand side are < m3 and all powers that appear in the components of the right-hand side are ≥ m3, it follows that c1,k3 = d1,k3 , c2,k3 = d2,k3 and q3 = e3. This process stops after r steps and gives that Q = Ir. (cid:3) Remark 6.2. Since every ordered field F with value group Γ can be o-embedded into R((Γ)) by Hahn's embedding theorem, it is tempting to assume that archimedean canonical forms exist for matrices over general fields. However, the problem is that to construct the archimedean canonical form we do not need just ring operations but also the operation of truncation which can take us out of the image of F in ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS 13 R((Γ)). If we want the image of F to be truncation closed we must pass to real closures. Let ¯F be the real closure of (F,≤) and let ¯Γ be he division hull of Γ. In [14] it is shown that there exists an o-embedding ψ : ¯F → R((¯Γ)) such that ψ( ¯F ) is truncation closed in R((¯Γ)) and ψ(F ) is contained in R((Γ)). Pick now any A ∈ Mn(F ) and compute the archimedean canonical form C of ψn(A) := [ψ(aij )] in Mn(R((Γ))). The properties of ψ imply that there exists B ∈ Mn( ¯F ) such that ψn(B) = C. Since ψn is an o-embedding, it follows that B ∼n A in Mn( ¯F ). So C is a representative of the archimedean class of A in Mn( ¯F ). This is not very useful because in Mn( ¯F ) we have better representatives of archimedean classes. Namely, if Pi σiPi is the spectral decomposition of AT A and τi are representatives of archimedean classes of √σi, then A ∼ Pi τiPi. We already know that the relation ∼n is not a congruence relation on the multiplicative semigroup on Mn(F ). More precisely, A ∼n B always implies AC ∼n BC but it does not always imply CA ∼n CB. Therefore, the factor set Mn(F )/ ∼n= Γn ∪ {∞} is not a semigroup in the usual way. We can address this issue in at least three different ways: 7. Can we multiply archimedean classes? (1) We give up on the multiplicative structure of Mn(F )/∼n and consider the following construction instead. For every C ∈ Mn(F ), the mapping φC : Mn(F )/∼n→ Mn(F )/∼n, φC (vn(A)) := vn(AC) is order-preserving. The mapping φ : C 7→ φC from the multiplicative semi- group of Mn(F ) to the semigroup of order-preserving maps from Mn(F )/∼n to Mn(F )/∼n is a homomorphism of semigroups. (2) We define some unnatural multiplication on Mn(F )/∼n; see section 7.1. (3) We modify the definition of the relation ∼n; see section 7.2. 7.1. An unnatural multiplication. We will define a multiplication on Mn(F )/ ∼n. Although the definition is unappealing its properties are very good. Proposition 7.1. The operation (cid:3) on Mn(F ) defined by has the following properties: A(cid:3)B := kAk∞kBk∞In (1) It is associative and commutative. (2) If A (cid:23)n B for some A, B ∈ Mn(F ) then A(cid:3)C (cid:23)n B(cid:3)C for every C ∈ (3) vn is surmultiplicative, i.e. vn(AB) (cid:23)n vn(A)(cid:3)vn(B) := vn(A(cid:3)B) for (4) φ is multiplicative, i.e. φ(vn(A)(cid:3)vn(B)) = φ(vn(A))φ(vn(B)) for every M∗,n(F ). In particular ∼n is a congruence relation on (Mn(F ), (cid:3)). every A ∈ Mn(F ) and every B ∈ Mn(F ). A, B ∈ Mn(F ). (vn(B)(cid:3)vn(C)) for every A, B, C ∈ Mn(F ). (5) ∧n is distributive over (cid:3), i.e. (vn(A)∧nvn(B))(cid:3)vn(C) = (vn(A)(cid:3)vn(C))∧n (6) ∨n is not distributive over (cid:3). Proof. Claim (1) is clear since (A(cid:3)B)(cid:3)C = kAk∞kBk∞kCk∞In = A(cid:3)(B(cid:3)C). Claim (2) follows from the fact that A (cid:23)n B implies kAk∞ (cid:23) kBk∞. For every A ∈ M∗,n(F ) we have that AT A ≤n (tr AT A)In ≤ n2kAk2 ∞In, so (8) A (cid:23)n kAk∞In. 14 J. CIMPRI C Claim (3) follows from inequality (8). Namely, AB (cid:23)n (kAk∞In)B = B(kAk∞In) (cid:23)n (kBk∞In)(kAk∞In). Claim (4) follows from φ(vn(A)(cid:3)vn(B)) = φ(vn(A(cid:3)B)) = v(kA(cid:3)Bk∞) = v(kAk∞kBk∞) = v(kAk∞)v(kBk∞) = φ(vn(A))φ(vn(B)). If we multiply max{kAk∞,kBk∞}2 ≤ kAk2 ∞ and we use k(cid:20) A with kCk2 ∞ + kBk2 ∞ ≤ 2 max{kAk∞,kBk∞}2 B (cid:21) k∞ = max{kAk∞,kBk∞} we get that B (cid:21) (cid:3)C ∼n (cid:20) A(cid:3)C (cid:20) A B(cid:3)C (cid:21) . This implies claim (5). To prove claim (6), take A = (cid:20) 1 0 and any C with kCk∞ = 1. Then 0 0 (cid:21) , B = (cid:20) 0 0 0 1 (cid:21) and (v2(A) ∨2 v2(B))(cid:3)v2(C) = 0(cid:3)C = 0 (v2(A)(cid:3)v2(C)) ∨2 (v2(B)(cid:3)v2(C)) = I2 ∨2 I2 = I2. (cid:3) It would be interesting to know if there exists a multiplication on Γn such that we also have distributivity for ∨n. 7.2. A variant of relations (cid:23)n and ∼n. Let F be an ordered field and n an integer. For every A, B ∈ Mn(F ) we write A ≫n B iff CA (cid:23)n CB for every C ∈ Mn(F ). A ≡n B iff A ≫n B and B ≫n A. It is clear that ≡n is a congruence relation on the multiplicative semigroup of Mn(F ) and that the congruence class of zero is a singleton. Proposition 7.2. For every A, B ∈ Mn(F ), the following are equivalent. (1) A ≫n B. (2) For every y ∈ F n there exists r ∈ N such that AT yyT A ≤n rBT yyT B. (3) For every y ∈ F n there exists a bounded αy ∈ F such that yT A = αyyT B. (4) There exists a bounded α ∈ F such that A = αB. Proof. (1) implies (2). Just replace C with a matrix whose first row is yT and other rows are zero. (2) implies (3). Use that for every vectors w, z ∈ F n which satisfy wwT ≤n rzzT for some r ∈ F there exists α ∈ F such that w = αz and α2 ≤ r. (3) implies (4). By assumption AT and BT are locally linearly dependent. By Theorem 2.3 in [5] either AT and BT are linearly dependent or their exists a nonzero v ∈ F n such that im AT and im BT are subsets of the span of v. If either rank A > 1 or rank B > 1 then we are in the first case. Consequently A = αB for some α. If rank A ≤ 1 and rank B ≤ 1 then there exist a, b ∈ F n such that AT = vaT and BT = vbT . By (3), for every y ∈ F n there exists a bounded αy ∈ F such that (yT a)vT = αy(yT b)vT . Since v 6= 0, it follows that yT a = αyyT b. Therefore ARCHIMEDEAN CLASSES OF MATRICES OVER ORDERED FIELDS 15 yT b = 0 implies yT a = 0 for every y ∈ F n. A = αB in this case, too. By (3), α must be bounded. In particular, a = αb. Therefore (4) implies (1). This is clear. The following is now clear: (cid:3) Corollary 7.3. For every matrices A, B ∈ Mn(F ) we have that A ≡n B iff either B and vF (kAk∞) = A = B = 0n or if A 6= 0n and B 6= 0n and vF (kBk∞). In other words, the factor set Mn(F )6=0n / ≡n is isomorphic to Kn×ΓF where Kn = {A ∈ Mn(F ) kAk∞ = 1}. A = 1 kBk∞ 1 kAk∞ References [1] Anderson, W. N., Jr.; Duffin, R. J. Series and parallel addition of matrices. J. Math. Anal. Appl. 26 (1969) 576-594. [2] Artin, E.; Schreier, O. Algebraische Konstruktion reeller Korper. Abh. Math. Sem. Univ. Hamburg 5 (1927), no. 1, 85-99. [3] Baer, R. Uber nicht-archimedisch geordnete Korper. Sitzungsber. Heidelb. Akad. Wiss. Math.- Naturw. Kl. 8 (1927), 3 -- 13. [4] Becker, E. Partial orders on a field and valuation rings. Comm. Algebra 7 (1979), no. 18, 1933-1976. [5] Bresar, M.; Semrl, P. On locally linearly dependent operators and derivations. Trans. Amer. Math. Soc. 351 (1999), no. 3, 1257-1275. [6] Cimpric, J. A representation theorem for Archimedean quadratic modules on -rings. Canad. Math. Bull. 52 (2009), no. 1, 39-52. [7] Conrad, P. On ordered division rings. Proc. Amer. Math. Soc. 5 (1954), no. 2, 323-328. [8] Dubrovin, N. I. Noncommutative valuation rings. (Russian) Trudy Moskov. Mat. Obshch. 45 (1982), 265-280. [9] Gantmacher, F. R. The theory of matrices. Vols. 1, 2. Translated by K. A. Hirsch. Chelsea Publishing Co., New York 1959 Vol. 1, x+374 pp. Vol. 2, ix+276 pp. [10] Hahn, H. Uber die nichtarchimedischen Grossensysteme. Sitzungsber. Akad. Wiss. Wien, Math.-Naturw. Kl. Abt. IIa 116 (1907), 601 -- 655. [11] Holland, S. S., Jr. -valuations and ordered *-fields. Trans. Amer. Math. Soc. 262 (1980), no. 1, 219-243. Erratum. Trans. Amer. Math. Soc. 267 (1981), no. 1, 333. [12] Knus, M.-A.; Merkurjev, A.; Rost, M.; Tignol, J.-P. The book of involutions. With a preface in French by J. Tits. American Mathematical Society Colloquium Publications, 44. American Mathematical Society, Providence, RI, 1998. xxii+593 pp. [13] Krull, W. Allgemeine Bewertungstheorie. J. Reine Angew. Math. 167 (1932) 160 -- 196. [14] Kuhlmann, F.-V.; Kuhlmann, S.; Marshall, M.; Zekavat, M. Embedding ordered fields in formal power series fields. J. Pure Appl. Algebra 169 (2002), no. 1, 71 -- 90. [15] Lam, T. Y. Introduction to quadratic forms over fields. Graduate Studies in Mathematics, 67. American Mathematical Society, Providence, RI, 2005. xxii+550 pp. ISBN: 0-8218-1095-2 [16] Lang, S. The theory of real places. Ann. of Math. 57 (1953), 378-391. [17] Morandi, P. J. Value functions on central simple algebras. Trans. Amer. Math. Soc. 315 (1989), no. 2, 605-622. [18] Schilling, O. F. G. Noncommutative valuations. Bull. Amer. Math. Soc. 51 (1945), 297-304. [19] Tignol, J.-P.; Wadsworth, A. R. Valuations on algebras with involution. Math. Ann. 351 (2011), no. 1, 109-148. University of Ljubljana, Faculty of Mathematics and Physics, Department of Math- ematics, Jadranska 21, SI-1000 Ljubljana, Slovenia E-mail address: [email protected]
0710.5565	2	0710	2010-05-23T03:27:34	A Note on $\aleph_{0}$-injective Rings	[ "math.RA" ]	A ring $R$ is called right $\aleph_{0}$-injective if every homomorphism from a countably generated right ideal of $R$ to $R_{R}$ can be extended to a homomorphism from $R_{R}$ to $R_{R}$. In this note, some characterizations of $\aleph_{0}$-injective rings are given. It is proved that if $R$ is semilocal, then $R$ is right $\aleph_{0}$-injective if and only if every homomorphism from a countably generated small right ideal of $R$ to $R_{R}$ can be extended to one from $R_{R}$ to $R_{R}$. It is also shown that if $R$ is right noetherian and left $\aleph_{0}$-injective, then $R$ is \emph{QF}. This result can be considered as an approach to the Faith-Menal conjecture.	math.RA	math	A Note on ℵ0-injective Rings Liang Shen Department of Mathematics, Southeast University Nanjing 210096, P.R. China E-mail: [email protected] Abstract: A ring R is called right ℵ0-injective if every right homomorphism from a countably generated right ideal of R to RR can be extended to a homo- morphism from RR to RR. In this note, some characterizations of ℵ0-injective rings are given. It is proved that if R is semiperfect, then R is right ℵ0- injective if and only if every homomorphism from a countably generated small right ideal of R to RR can be extended to one from RR to RR. It is also shown that if R is right noetherian and left ℵ0-injective, then R is QF. This result can be looked as an approach to the Faith-Menal conjecture. Key Words: ℵ0-injective rings; Faith-Menal conjecture; Quasi-Frobenius rings. (2000) Mathematics Subject Classification: 16D50; 16L60. 1. INTRODUCTION Throughout this paper rings are associative with identity. Write J and Sl for the Jacobson radical and the left socle of a ring R respectively. Use N ⊆ess M to mean that N is an essential submodule of M. For a subset X of a ring R, the left annihilator of X in R is l(X) = {r ∈ R : rx = 0 for all x ∈ X}. Right annihilators are defined analogously. f = c· means that f is a homomorphism multiplied by an element c on the left side. It is mentioned in [13] that a ring R is called right ℵ0-(or countably) injective if every right homomorphism from a countably generated right ideal of R to RR can be extended to a homomorphism from RR to RR. Recall that a ring R is called right F-injective if every right homomorphism from a finitely generated 1 2 right ideal of R to RR can be extended to one from RR to RR. And a right FP-injective ring R satisfies that every right homomorphism from a finitely generated submodule of a free right R-module FR to RR can be extended to one from FR to RR. The left side of the above injectivities can be defined similarly. It is obvious that right self-injective rings are right ℵ0-injective and right ℵ0-injective rings are right F -injective. But neither of the converses is true (see [13, Example 10.46]). The example also shows that a right FP- injective ring may not be right ℵ0-injective. But it is still unknown whether a right F -injective ring is right FP-injective. We have the following arrow diagrams on injectivities of rings: right self-injectivity ⇒ : right ℵ0-injectivity ⇒ : right F -injectivity ?⇒ ⇐ right FP -injectivity, right self-injectivity ⇒ : right FP -injectivity ; ⇐? right ℵ0-injectivity. Recall that a ring R is quasi-Frobenius (QF ) if R is one-sided noetherian and one-sided self-injective. There are three unresolved Faith conjectures on QF rings (see [10]). One of them is the Faith-Menal conjecture, which was raised by Faith and Menal in [3]. The conjecture says that every strongly right Johns ring is QF. A ring R is called right Johns if R is right noetherian and every right ideal of R is a right annihilator. R is called strongly right Johns if the matrix ring Mn(R) is right Johns for all n ≥ 1. In [4], Johns used a false result of Kurshan [6, Theorem 3.3] to show that right Johns rings are right artinian. Later in [2], Faith and Menal gave a counter example to show that right Johns rings may not be right artinian. They characterized strongly right Johns rings as right noetherian and left FP-injective rings (see [3, Theorem 1.1]). So the Faith-Menal conjecture is equivalent to say that every right noetherian and left FP-injective ring is QF. In this short article, some characterizations of ℵ0- injective rings are explored. It is proved in Theorem 9 that if R is semiperfect, then R is right ℵ0-injective if and only if every homomorphism from a countably generated small right ideal of R to RR can be extended to one from RR to RR. Since FP-injectivity implies F -injectivity, it is unknown whether a right noetherian and left F -injective ring is QF. It is proved in Theorem 14 that a right noetherian and left ℵ0-injective ring is QF. This result can be looked as an approach to the Faith-Menal conjecture. 2. RESULTS 3 First we explore some basic characterizations of ℵ0-injective rings. Proposition 1. A direct product of rings R = Qi∈I Ri is right ℵ0-injective if and only if Ri is right ℵ0-injective, ∀i ∈ I. Proof. For i ∈ I, let πi and ιi be the ith projection map and the ith inclusion map canonically. If R is right ℵ0-injective, for each i, suppose fi : Ti → Ri is Ri-linear where Ti is a countably generated right ideal of Ri. Then the map 0 × · · · × Ti × · · · × 0 → 0 × · · · × Ri × · · · × 0 given by (0, · · · , ti, · · · , 0) 7−→ (0, · · · , fi(ti), · · · , 0) is R-linear with 0×· · ·×Ti ×· · ·×0 a countably generated right ideal of R. So it has the form c· where c ∈ R. Thus fi = πi(c)·. Conversely, let γ : T → R be R-linear, where T is a countably generated right ideal of R. For each i ∈ I, let Ti = {x ∈ Ri ιi(x) ∈ T }. Since T is countably k=1 akR, where ak ∈ R, k = 1, 2, · · · . Then it is easy to prove k=1 πi(ak)Ri is a countably generated right ideal of Ri, ∀i ∈ I. Now define γi : Ti → Ri by γi(x) = πiγ(ιi(x)), x ∈ Ti. Since Ri is right ℵ0-injective, there exists ci ∈ Ri such that γi = ci·. For each ¯t = htii ∈ T , write γ(¯t) = ¯s = hsii. Since T is a right ideal of R, ti ∈ Ti, ∀i ∈ I. Thus si = πi(¯s · ιi(1i)) = πi(γ(¯t) · ιi(1i)) = πiγ(¯t · ιi(1i)) = πiγ(ιi(ti)) = γi(ti) = citi, whence ¯s = hcii · ¯t. So γ = hcii·. This shows that R is right ℵ0-injective. generated, T =P∞ that Ti = P∞ (cid:3) Proposition 2. If R is right ℵ0-injective, then l(I ∩ K)=l(I)+l(K), where I and K are countably generated right ideals of R. Proof. It is only to be shown that l(I ∩ K)⊆l(I)+l(K). Let x ∈ l(I ∩ K). Define a right R-homomorphism f from I + K to RR such that f (i + k) = xi, where i ∈ I and k ∈ K. Then it is clear that f is well-defined. Since I and K are both countably generated right ideals of R, I + K is also a countably generated right ideal of R. As R is right ℵ0-injective, f can be extended to a homomorphism from RR to RR. Hence there exist an element c ∈ R such that f = c·. Thus, by the definition of f , c ∈ l(K) and x − c ∈ l(I). So x = (x − c) + c ∈ l(I) + l(K). (cid:3) 4 Recall that a ring R is called right Kasch if each simple right R-module can embed into RR. Or equivalently, every maximal right ideal of R is a right annihilator. Left Kasch rings can be defined similarly. Proposition 3. If R is right Kasch and right ℵ0-injective, then every countably generated right ideals of R is a right annihilator. Proof. Let I be a countably generated right ideal of R. If I is not a right annihilator, then there exists a nonzero element x ∈ R such that x ∈ rl(I)\I. Now let K = I + xR. Then K = K/I is finitely generated. Hence K has a maximal submodule M . Since R is right Kasch, K/M can embed into RR. Thus there exists a homomorphism f from K to RR with f (I) = 0 and f (x) 6= 0. Since R is right ℵ0-injective, f = c· for some c ∈ R. So c ∈ l(I). Since x ∈ rl(I), f (x) = cx = 0. This is a contradiction. (cid:3) Theorem 4. Let R be a right ℵ0-injective ring. For any idempotent e ∈ R with ReR=R, the corner ring eRe is also right ℵ0-injective. Proof. Let S = eRe and θ : T → S be a right S-homomorphism from a countably generated right ideal T of S to SS. Define ¯θ : T R → RR by ¯θ(P tiri) = P θ(ti)ri, ti ∈ T , ri ∈ R. Assume P tiri = 0. For any r ∈ R, 0 = P tirire = P ti(erire). So 0 = P θ(ti)(erire) = [P θ(ti)ri]re. Since ReR = R, it is clear that P θ(ti)ri = 0. Hence ¯θ is a well-defined right R- homomorphism. Since T is a countably generated right ideal of S, T R is also a countably generated right ideal of R. As R is right ℵ0-injective, ¯θ = c· for some c ∈ R. Then for each t ∈ T , θ(t) = eθ(t) = e¯θ(t) = ect = (ec)et = (ece)t. Hence θ = (ece)·, as required. (cid:3) Remark 5. The condition that ReR = R in the above theorem is necessary. For example (see [5, Example 9]), let R be the algebra of matrices over a field a x 0 0 0 0 0 b 0 0 0 0 0 0 c y 0 0 0 0 0 a 0 0 0 0 0 0 b z 0 0 0 0 0 c     K of the form R = , a, b, c, x, y, z ∈ K. Set e = e11 + e22 + e44 + e55, which is a sum of canonical matrix units. It is 5 clear that e is an idempotent of R such that ReR 6= R. R is right ℵ0-injective, but eRe is not right ℵ0-injective. Proof. [5, Example 9] shows that R is a QF ring and eRe is not a QF ring. Since R is QF , R is right self-injective and eRe is left noetherian. So R is right ℵ0-injective. If eRe is right ℵ0-injective, then eRe is QF by Theorem 14. This is a contradiction. (cid:3) It is natural to ask whether right ℵ0-injectivity is a Morita invariant. Question 6. If R is right ℵ0-injective, is Mn×n(R) (n ≥ 2) right ℵ0-injective? The method in the proof of the following theorem is owing to [8, Theorem 1] Theorem 7. The following are equivalent for a ring R and an integer n ≥ 1: (1) Mn(R) is right ℵ0-injective. (2) For each countably generated right R-submodule T of Rn, every R-linear map γ: T→ R can be extended to Rn→ R. (3) For each countably generated right R-submodule T of Rn, every R-linear map γ: T→ Rn can be extended to Rn→ Rn. Proof. We prove for the case n=2. The others are analogous. (1)⇒(2). Given γ: T → R where T is a countably generated right R-submodule of R2, consider the countably generated right ideal T =[T T ] = {[α β]α, β ∈ T } of M2(R). The map γ : T →M2(R) defined by γ([α β])=" γ(α) γ(β) 0 # , α, β ∈ T 0 is M2(R)-linear. By (1), there exists C ∈ M2(R) such that γ = C·. So γ = α·, where α is the first row of C. Hence γ can be extended to a homomorphism from R2 to R. (2)⇒(3). Given (2), consider γ: T → R2 where T is a countably generated right R- submodule of R2. Let πi : R2 → R be the ith projection, i = 1, 2. Then (2) provides an R-linear map γi : R2 → R extending πi ◦ γ, i = 1, 2. Thus γ : R2 → R2 extends γ where γ(x) = [γ1(x) γ2(x)]T , x ∈ R2. 6 (3)⇒(1). Write S = M2(R), consider γ: T → SS where T is a countably generated right ideal of S. Then it is easy to prove that T =[T0 T0] where T0 = {x ∈ R2 [x 0] ∈ T } is a right countably generated right R-submodule of R2. For x ∈ T0, the S- linearity of γ shows that γ[x 0] = γ([x 0]" 1 0 0 0 #) = γ([x 0])" 1 0 0 0 # = [y 0] for some y ∈ R2. Writing y = γ0(x) yields an R-linear map γ0: T0 → R2 such that γ[x 0] = [γ0(x) 0], x ∈ T0. Then γ0 extends to an R-linear map γ : R2 → R2 by (3). Hence γ0 = C· for some C ∈ S. If [x y] ∈ T it follows that γ([x y]) = γ([x 0] + [y 0]" 0 1 0 0 #) = [γ0(x) 0] + [γ0(y) 0]" 0 1 0 0 # = [Cx Cy] = C[x y]. This shows γ = C·. (cid:3) Recall that a right ideal L of a ring R is called small if, for any proper right 6= RR. Let I be a right ideal of R. I is said to lie ideal L′ of R, L + L′ over a direct summand of RR if there exists an idempotent e ∈ R such that I = eR ⊕ (I ∩ (1 − e)R), where I ∩ (1 − e)R is a small right ideal of R. Lemma 8. [7, Corollary 2.10] A ring R is semiperfect if and only if every countably generated right ideal of R lies over a direct summand of RR. Theorem 9. Let R be semiperfect. If every homomorphism from a countably generated small right ideal of R to RR can be extended to one from RR to RR, then R is right ℵ0-injective. Proof. Let I be a countably generated right ideal of R and f be a homo- morphism from I to RR. By the above lemma, I = eR ⊕ K, where e is an idempotent of RR and K = I ∩ (1 − e)R is a small right ideal of R. Since I is countably generated, K is also countably generated. By hypothesis, there exists a homomorphism g from (1 − e)R to RR such that gK = f K. For each x ∈ R, define F (x) = f (x1) + g(x2) where x1 = ex and x2 = (1 − e)x. It is clear that F I = f . (cid:3) Now we turn to the main theorem of this note. First look at some lemmas. Lemma 10. If R is a left ℵ0-injective ring with ACC on right annihilators, then R is left finite dimensional. 7 Proof. Assume R is not left finite dimensional. Then there are nonzero el- ements ai ∈ R, i = 1, 2, . . . , such that {Rai}∞ i=1 is an independent family of proper left ideals of R. Let Ik = ⊕∞ i=kRai, k = 1, 2, . . .. Then r(I1) ⊆r(I2)⊆ · · · . Since R satisfies ACC on right annihilators, there exists n ∈ N such that r(In) = r(In+1). As In = In+1 ⊕ Ran, we have r(In)=r(In+1)∩r(an). So r(In) ⊆r(an). Since R is left ℵ0-injective, by the symmetry of Proposition 2, R=r(0)=r(In+1 ∩ Ran)= r(In+1)+r(an)=r(In)+r(an)=r(an). Thus an = 0. This is a contradiction. (cid:3) Recall that a ring R is called left P-injective (2-injective) if every homomor- phism from a principal (2-generated) left ideal of R to RR can be extended to one from RR to RR. Lemma 11. [9, Theorem 3.3]If R is left P -injective and left finite dimensional, then R is semilocal. Lemma 12. [1, Theorem 2.7]If R is right noetherian and left P-injective, then J is nilpotent. Lemma 13. [11, Corollary 3]If R is a left 2-injective ring with ACC on left annihilators, then R is QF. Now we obtain the main theorem. Theorem 14. If R is right noetherian and left ℵ0-injective, then R is QF. Proof. Since R is right noetherian, R satisfies ACC on right annihilators. By Lemma 10, R is left finite dimensional. Since R is left ℵ0-injective, R is left P -injective. So R is semilocal and J is nilpotent by Lemma 11 and Lemma 12. Thus R is semiprimary. Hence R is right artinian. So R satisfies ACC on left annihilators. Then R is QF by Lemma 13. (cid:3) By Lemma 13, we see that if R is a left ℵ0-injective ring with ACC on left annihilators, then R is QF. It is natural to ask the following question: Question 15. Can right noetherian condition in the above theorem be weak- ened to the condition satisfying ACC on right annihilators? Remark 16. The answer is "yes" if we can show that J is right T -nilpotent. By Lemma 10 and Lemma 11, R is semilocal. If J is right T -nilpotent, then 8 R is right perfect. So R is left GPF (i.e., R is left P -injective, semiperfect and Sl ⊆ess RR). Thus R is left Kasch by [10, Theorem 5.31]. By [9, Lemma 2.2], R is right P -injective. So R is left and right mininjective. Recall that a ring R is called right mininjective if every homomorphism from a minimal right ideal of R to RR can be extended to one from RR to RR. Left mininjective rings can be defined similarly. Then by [12, Theorem 2.5], R is QF. ACKNOWLEDGEMENTS The research is supported by the Foundation of Southeast University (No.4007011034). References [1] J. L. G´omez Pardo, and P. A. Guil Asensio, (1998) Torsionless modules and rings with finite essential socle. Abelian groups Module Theory, and Topology, Padua, 1997. Lecture Notes in Pure and Appl. Math., Vol. 201. New York: Marcel Dekker, Inc., pp. 261-278. [2] C. Faith and P. Menal, A counter example to a conjecture of Johns, Proc. Amer. Math. Soc., 116 (1992), 21-26. [3] C. Faith and P. Menal, The structure of Johns rings, Proc. Amer. Math. Soc., 120 (1994), 1071-1081. [4] B. Johns, Annihilator conditions in noetherian rings, J. Algetra, 49 (1977), 222-224. [5] K. Koike, Dual rings and cogenerator rings, Math. J. Okayama Univ., 37 (1995), 99-103. [6] R.P. Kurshan, Rings whose cyclic modules have finitely generated socle, J. Algetra, 15 (1970), 376-386. [7] W.K. Nicholson, Semiregular modules and rings, Canadian J. Math 28 (1976), 1105- 1120. [8] W.K. Nicholson, J.K. Park and M.F. Yousif, Extensions of simple-injective rings, Comm. Algebra, 28(10) (2000), 4665-4675. [9] W.K. Nicholson and M.F. Yousif, Principally injective rings, J. Algebra 174 (1995), 77-93. [10] W.K. Nicholson and M.F. Yousif, Quasi-Frobenius Rings, Cambridge University Press, 2003. [11] E.A. Rutter, Rings with the principal extension property, Comm. Algebra 3(3) (1975), 203-212. [12] L. Shen and J.L. Chen, New characterizations of Quasi-Frobenius rings, Comm. Algebra 34 (2006), 2157-2165. [13] A.A. Tuganbaev, Semidistributive modules and rings, Kluwer Academic Publishers, 1998. 9
1312.5037	1	1312	2013-12-18T04:28:37	Schr\"odinger representations from the viewpoint of monoidal categories	[ "math.RA" ]	The Drinfel'd double D(A) of a finite-dimensional Hopf algebra A is a Hopf algebraic counterpart of the monoidal center construction. Majid introduced an important representation of the Drinfel'd double, which he called the Schr\"odinger representation. We study this representation from the viewpoint of the theory of monoidal categories. One of our main results is as follows: If two finite-dimensional Hopf algebras A and B over a field k are monoidally Morita equivalent, i.e., there exists an equivalence F from the module category over A to the module category over B of k-linear monoidal categories, then the equivalence between the module categories over D(A) and D(B) induced by F preserves the Schr\"odinger representation. As an application, we construct a family of invariants of finite-dimensional Hopf algebras under the monoidal Morita equivalence. This family is parameterized by braids. The invariant associated to a braid b is, roughly speaking, defined by "coloring'' the closure of b by the Schr\"odinger representation. We investigate what algebraic properties this family have and, in particular, show that the invariant associated to a certain braid closely relates to the number of irreducible representations.	math.RA	math	Schrodinger representations from the viewpoint of monoidal categories Kenichi Shimizu and Michihisa Wakui Abstract The Drinfel'd double D(A) of a finite-dimensional Hopf algebra A is a Hopf algebraic counterpart of the monoidal center construction. Majid introduced an important representa- tion of D(A), which he called the Schrodinger representation. We study this representation from the viewpoint of the theory of monoidal categories. One of our main results is as follows: If two finite-dimensional Hopf algebras A and B over a field k are monoidally Morita equiv- alent, i.e., there exists an equivalence F : AM → BM of k-linear monoidal categories, then the equivalence D(A) M induced by F preserves the Schrodinger representation. Here, AM for an algebra A means the category of left A-modules. M ≈ D(B) As an application, we construct a family of invariants of finite-dimensional Hopf alge- bras under the monoidal Morita equivalence. This family is parameterized by braids. The invariant associated to a braid b is, roughly speaking, defined by "coloring" the closure of b by the Schrodinger representation. We investigate what algebraic properties this family have and, in particular, show that the invariant associated to a certain braid closely relates to the number of irreducible representations. Key words: Hopf algebra, monoidal category, Schrodinger representation, Drinfel'd double Mathematics Subject Classifications (2010): 16T05, 16D90, 18D10 1 Introduction Drinfel'd doubles of Hopf algebras [5] are one of the most important objects in not only Hopf algebra theory, but also in other areas including category theory and low-dimensional topology. Let A be a finite-dimensional Hopf algebra over a field k, and (D(A), R) be its Drinfel'd double. Due to Majid [25], it is known that there is a canonical representation of D(A) on A, which is called the Schrodinger representation (or the Schrodinger module). This representation is an extension of the adjoint representation of A, and originates from quantum mechanics as explained in Majid's book [25, Examples 6.1.4 & 7.1.8] (see Section 2 for the precise definition of the Schrodinger representation). The Schrodinger module is also addressed by Fang [10] as an algebra in the Yetter-Drinfel'd category A AYD via the Miyashita-Ulbrich action. In this paper, we study the Schrodinger module over the Drinfel'd double from the viewpoint of the theory of monoidal categories. We say that two finite-dimensional Hopf algebras A and B over the same field k are monoidally Morita equivalent if AM and BM are equivalent as k-linear monoidal categories, where H M for an algebra H is the category of H-modules. One of our main results is that the Schrodinger module is an invariant under the monoidal Morita equivalence 1 in the following sense: If F : AM → BM is an equivalence of k-linear monoidal categories, then the equivalence D(A)M ≈ D(B)M induced by F preserves the Schrodinger modules. To prove this result, we introduce the notion of the Schrodinger object for a monoidal category by using the monoidal center construction. It turns out that the Schrodinger module over D(A) is characterized as the Schrodinger object for AM. Once such a characterization is established, the above result easily follows from general arguments. As an application of the above category-theoretical understanding of the Schrodinger module, we construct a new family of monoidal Morita invariants, i.e., invariants of finite-dimensional Hopf algebras under the monoidal Morita equivalence. Some monoidal Morita invariants have been discovered and studied; see, e.g., [7, 8, 9, 17, 28, 29, 37, 39]. Our family of invariants is parametrized by braids. Roughly speaking, the invariant associated with a braid b is defined by "coloring" the closure of b by the Schrodinger module. Since the quantum dimension (in the sense of Majid [25]) of the Schrodinger module is a special case of our invariants, we call the invariant associated with b the braided dimension of the Schrodinger module associated with b. We investigate what algebraic properties this family of invariants have and, in particular, show that the invariant associated to a certain braid closely relates to the number of irreducible representations. This paper organizes as follows: In Section 2, we recall the definition of the Schrodinger mod- ule over the Drinfel'd double D(A) of a finite-dimensional Hopf algebra A. We also describe the definition of another Schrodinger representation of D(A) on A∗cop, which is introduced by Fang [10]. We refer the corresponding left D(A)-module as the dual Schrodinger module. Following [13] we also describe the definition of Radford's induction functors and its properties. It is shown that the Schrodinger module and the dual Schodinger module are isomorphic to the images of the trivial left A-module and the trivial right A-comodule under Radford's induction functors, respectively. Furthermore, we examine the relationship between the Schrodinger module over D(A∗) and the dual Schrodinger module over D(A). In Section 3, we study the categorical aspects of the Schrodinger module and the dual Schrodinger module. We introduce a Schrodinger object for a monoidal category C as the object of Z(C) representing the functor HomC(Π(−), I), where Z(C) is the monoidal center of C, Π : Z(C) → C is the forgetful functor, and I is the unit object of C. By using the properties of Radford's induction functor, we show that the Schrodinger module over D(A) is a Schrodinger object for AM under the identification Z(AM) ≈ D(A)M. Once this characterization is obtained, the invariance of the Schrodinger module (stated above) is easily proved. A similar result for the dual Schrodinger module is also proved. In Section 4, based on our category-theoretical understanding of the Schrodinger module, we introduce a family of monoidal Morita invariants parameterized by braids. We give formulas for the invariants associated with a certain series of braids, and give some applications. Note that some monoidal Morita invariants, such as ones introduced in [7] and [37], factor through the 2 Drinfel'd double construction. Our invariants have an advantage that they do not factor through that. On the other hand, our invariants have a disadvantage in the non-semisimple situation: For any braid b, the braided dimension of the Schrodinger module of D(A) associated with b is zero unless A is cosemisimple. From this result, we could say that our invariants are not interesting as monoidal Morita invariants for non-cosemisimple Hopf algebras. However, endomorphisms on the Schodinger module induced by braids are not generally zero, and thus may have some information about A. To demonstrate, we give an example of a morphism induced by a braid, which turns out to be closely related to the unimodularlity of A. Throughout this paper, k is an arbitrary field, and all tensor products ⊗ are taken over k. For k-vector spaces X and Y , denoted by TX,Y is the k-linear isomorphism from X ⊗Y to Y ⊗X defined by TX,Y (x ⊗ y) = y ⊗ x for all x ∈ X and y ∈ Y . For a coalgebra (C, ∆, ε) and c ∈ C we use the Sweedler's notation ∆(c) =P c(1) ⊗ c(2). If (M, ρ) is a right C-comodule and m ∈ M , then we also use the notation ρ(m) =P m(0) ⊗ m(1). For a Hopf algebra A, denoted by ∆A, εA and SA are the coproduct, the counit, and the antipode of A, respectively. If the antipode of A is bijective, then one has two Hopf algebras Acop and Aop, which are defined from A by replacing ∆A by the opposite coproduct ∆cop A = TA,A ◦ ∆A and replacing the product by the opposite product, respectively. A Hopf algebra map α : A −→ B induces a Hopf algebra map form Acop to Bcop. The map is denoted by αcop. If A is a finite-dimensional, then the dual vector space A∗ is also a Hopf algebra so that hpq, ai =P p(a(1))q(a(2)), 1A∗ = εA, h∆A∗(p), a ⊗ bi = p(ab), εA∗(p) = p(1A), SA∗(p) = p ◦ SA for p, q ∈ A∗ and a, b ∈ A, where h , i stands for the natural pairing between A and A∗, or A ⊗ A and A∗ ⊗ A∗. Denoted by AM is the k-linear monoidal category whose objects are left A-modules and morphisms are A-module maps, and MA is the k-linear monoidal category whose objects are right A-comodules and morphisms are A-comodule maps. For general facts on Hopf algebras, refer to Abe's book [1], Montgomery's book [27] and Sweedler's book [38], and for general facts on monoidal categories, refer to MacLane's book [21], Kassel's book [19] and Joyal and Street's paper [15]. 2 Schrodinger modules of the Drinfel'd double 2.1 Preliminaries: the Drinfel'd double Let A be a finite-dimensional Hopf algebra A over the field k. The Drinfel'd double D(A) of A [5] is the Hopf algebra such that as a coalgebra D(A) = A∗cop ⊗ A, and the multiplication is given by (p ⊗ a) · (p′ ⊗ a′) =Xhp′ (1), S−1 A (a(3))ihp′ (3), a(1)ipp′ (2) ⊗ a(2)a′ 3 for all p, p′ ∈ A∗ and a, a′ ∈ A, where ((∆A ⊗ id) ◦ ∆A)(a) =P a(1) ⊗ a(2) ⊗ a(3), ((∆A∗ ⊗ id) ◦ ∆A∗)(p) =P p(1) ⊗ p(2) ⊗ p(3). The unit element is εA ⊗ 1A, and the antipode is given by A∗ (p(3)), a(1)iS−1 A∗ (p(2)) ⊗ SA(a(2)). SD(A)(p ⊗ a) =Xhp(1), a(3)ihS−1 For all p ∈ A∗ and a ∈ A, the element p ⊗ a in D(A) is frequently written in the form p ⊲⊳ a, since the Drinfel'd double can be viewed as a bicrossed product A∗cop ⊲⊳ A due to Majid [22]. The Drinfel'd double D(A) has a canonical quasitriangular structure as described below [5]. Let {ei}d 1A) ∈ D(A) ⊗ D(A) satisfies the following conditions. i=1 be its dual basis for A∗. Then R =Pd i=1(εA ⊲⊳ ei) ⊗ (e∗ i=1 be a basis for A, and {e∗ i }d i ⊲⊳ • ∆cop(x) = R · ∆(x) · R−1 for all x ∈ D(A), • (∆ ⊗ id)(R) = R13R23, • (id ⊗ ∆)(R) = R13R12. Here Rij denotes the element in D(A) ⊗ D(A) ⊗ D(A) obtained by substituting the first and the second components in R to the i-th and j-th components, respectively, and substituting 1 to elsewhere. Thus, the pair (D(A), R) is a quasitriangular Hopf algebra [5]. Let A and B be two finite-dimensional Hopf algebras over k. If they are isomorphic, then their Drinfel'd doubles are. More precisely, an isomorphism f : A −→ B of Hopf algebras yields an isomorphism of quasitriangular Hopf algebras D(f ) := (f −1)∗ ⊗ f : D(A) −→ D(B). Let A be a finite-dimensional Hopf algebra over k. As shown in [33, Theorem 3] and [37, Propositions 2.10 & 3.5], for the Drinfel'd doubles (D(A), R), (D(Aop cop∗), R) and (D(A∗), R′), there are isomorphisms: (D(A), R) ∼= (D(Aop cop∗)op, R21) ∼= (D(A∗)op, R′ 21) ∼= (D(A∗)cop, R′ 21). (2.1) Here, the first isomorphism f1 : (D(A), R) −→ (D(Aop cop∗)op, R21) is given by f1(p ⊲⊳ a) = a ⊲⊳ p (p ∈ A∗, a ∈ A) under the identification A∗∗ = A. The second isomorphism is given by f2 := S ⊗ tS−1 : (D(Aop cop∗)op, R21) −→ (D(A∗)op, R′ 21) under the identification A∗∗ = A, where we regard S as a Hopf algebra isomorphism from Aop cop to A. Finally, the third isomorphism is given by the antipode SD(A∗) [37, Proposition 2.10(1)]. Composing f1, f2 and SD(A∗) we have an isomorphism of quasitriangular Hopf algebras φA : (D(A), R) −→ (D(A∗)cop, R′ 21) such that φA(p ⊲⊳ a) = (ιA(1A) ⊲⊳ p) · (ιA(a) ⊲⊳ εA) (2.2) for all p ∈ A∗ and a ∈ A, where ιA : A −→ A∗∗ is the usual isomorphism of vector spaces. We note the following property of the isomorphism φA: 4 Lemma 2.1. φA∗ ◦ φA = D(ιA). Proof. As is well-known, ιA∗ = (ι−1 A )∗. Hence, for all a ∈ A and p ∈ A∗, (φA∗ ◦ φA)(p ⊲⊳ a) = φA∗(ιA(1A) ⊲⊳ p) · φA∗(ιA(a) ⊲⊳ εA) = (ιA∗ (p) ⊲⊳ ιA(1A)) · (ιA∗(εA) ⊲⊳ ιA(a)) = D(ιA)(p ⊲⊳ a). 2.2 Schrodinger modules The Drinfel'd double D(A) has a canonical representation, which is called the Schrodinger representation as described in Majid's book [25, Examples 6.1.4 & 7.1.8]. This representation is obtained by unifying the left adjoint action of A and the right A∗-action ↼ (See below (2.3) and (2.4) for precise definition). It is generalized to quasi-Hopf case by Bulacu and Torrecillas [2, Section 3]. Another generalization is given by Fang [10, Section 2]. He introduced two Schrodinger representations for the Drinfel'd double defined by a generalized Hopf pairing. One corresponds to the original Schrodinger representation, and the other corresponds to a dual version of it. Specializing in our setting, we will describe these representations below. There are four actions defined as follows. a ◮ c =X a(1)cS(a(2)) a ↼ p =Xhp, a(1)ia(2) q ◭ p =X S(p(1))qp(2) a ⇀ q =X q(1)hq(2), ai (a, c ∈ A), (p ∈ A∗, a ∈ A), (p, q ∈ A∗), (q ∈ A∗, a ∈ A). By using these actions two left actions • of D(A) on A and A∗ can be defined by (p ⊲⊳ a) • b = (a ◮ b) ↼ S−1(p), (p ⊲⊳ a) • q = (a ⇀ q) ◭ S−1(p) (2.3) (2.4) (2.5) (2.6) (2.7) (2.8) for all a, b ∈ A∗ and p, q ∈ A∗. We call the left D(A)-modules (A, •) and (A∗, •) the Schrodinger module and the dual Schrodinger module, and denote them by D(A)A and D(A)A∗cop, respectively. Proposition 2.2. Regard the dual Schrodinger module D(A∗)(A∗)∗cop of D(A∗) as a left D(A)- module by pull-back along the isomorphism φA defined by (2.2). Then D(A∗)(A∗)∗cop is isomor- phic to the Schrodinger module D(A)A as a left D(A)-module. Proof. Let ¯• be the left action on D(A∗)(A∗)∗cop = A as a left D(A)-module by pull-back along 5 φA. Then, for p ∈ A∗ and a, b ∈ A, we have (p ⊲⊳ a) ¯• b = φA(p ⊲⊳ a) • b = (1A ⊲⊳ p) •(cid:0)(a ⊲⊳ εA) • b(cid:1) = (1A ⊲⊳ p) •(cid:0)X a(2)bS−1(a(1))(cid:1) =Xha(4)b(2)S−1(a(1)), pia(3)b(1)S−1(a(2)) =Xha(1)S(b(2))S(a(4)), S−1(p)iS−1(cid:0)a(2)S(b(1))S(a(3))(cid:1) = S−1(cid:0)(p ⊲⊳ a) • S(b)(cid:1). This implies that S(cid:0)(p ⊲⊳ a) ¯• b(cid:1) = (p ⊲⊳ a) • S(b). Thus, the composition S ◦ ι−1 gives an isomorphism D(A∗)(A∗)∗cop −→ D(A)A of left D(A)-modules. A : A∗∗ −→ A Corollary 2.3. Regard the dual Schrodinger module D(A)(A)∗cop as a left D(A∗)-module by pull-back along the isomorphism (φ−1 A )cop : (D(A∗), R′) −→ (D(A)cop, R21). Then D(A)(A)∗cop is isomorphic to the Schrodinger module D(A∗)A∗ as a left D(A∗)-module. Proof. Let F1 : D(A)M −→ D(A∗∗)M and F2 : D(A)M −→ D(A∗)M be the isomorphisms of categories induced by isomorphisms D(ι−1 A ) and φA∗ , respectively. By Lemma 2.1, F = F2 ◦ F1 is the isomorphism of categories induced by φ−1 A . Applying Proposition 2.2 to A∗, we have F (D(A)(A)∗cop) = (F2 ◦ F1)(D(A)(A)∗cop) = F2(D(A∗∗)(A∗∗)∗cop) = D(A∗)A∗. 2.3 Radford's induction functors and Schrodinger modules In this subsection we show that the Schrodinger module D(A)A is isomorphic to the image of the trivial A-module under Radford's induction functor. This implies that the Schrodinger modules are remarkable objects from the viewpoint of category theory. Radford's induction functors are described by using Yetter-Drinfel'd modules, which were introduced in [40] and called crossed bimodules. Let us recall the definition of Yetter-Drinfel'd modules [35]. Let A be a bialgebra over k. Suppose that M is a vector space over k equipped with a left A-module action · and a right A-comodule coaction ρ on it. The triple M = (M, ·, ρ) is called a Yetter-Drinfel'd A-module, if the following compatibility condition, that is called the Yetter-Drinfel'd condition, is satisfied for all a ∈ A and m ∈ M : X(a(1) · m(0)) ⊗ (a(2)m(1)) =X(a(2) · m)(0) ⊗ (a(2) · m)(1)a(1). (2.9) A k-linear map f : M −→ N between two Yetter-Drinfel'd A-modules is called a Yetter-Drinfel'd homomorphism if f is an A-module map and an A-comodule map. For two Yetter-Drinfel'd 6 A-modules M and N , the tensor product M ⊗ N becomes a Yetter-Drinfel'd A-module with the action and coaction: a · (m ⊗ n) =X(a(1) · m) ⊗ (a(2) · n), m ⊗ n 7−→X m(0) ⊗ n(0) ⊗ n(1)m(1) for all a ∈ A, m ∈ M, n ∈ N . Then, the Yetter-Drinfel'd A-modules and the Yetter-Drinfel'd homomorphisms make a monoidal category together with the above tensor products. This category is called the Yetter-Drinfel'd category, and denoted by AYDA. By Yetter [40] it is proved that the Yetter-Drinfel'd category AYDA has a prebraiding c, which is a collection of Yetter-Drinfel'd homomorphisms cM,N : M ⊗ N −→ N ⊗ M defined by cM,N (m ⊗ n) =X n(0) ⊗ (n(1) · m) for all m ∈ M, n ∈ N . Furthermore, he also proved that if A is a Hopf algebra, then the prebraiding c is a braiding for AYDA. In [34] Radford constructed a functor from AM to AYDA, that is a right adjoint of the forgetful functor RA : AYDA −→ AM as shown in [13]. Lemma 2.4 (Radford[34, Proposition 2], Hu and Zhang[13, Lemma 2.1]). Let A be a bialgebra over k, and suppose that Aop has antipode S. Let L ∈ AM. Then L ⊗ A ∈ AYDA, where the left A action · and the right A-coaction ρ are given by h · (l ⊗ a) =X(h(2) · l) ⊗ h(3)aS(h(1)), ρ(l ⊗ a) =X(l ⊗ a(1)) ⊗ a(2) for all h, a ∈ A and l ∈ L. The correspondence L 7→ L ⊗ A is extended to a functor IA : AM −→ AYDA that is a right adjoint of the forgetful functor RA. The functor IA will be referred to as Radford's induction functor derived from left A-modules. From the above lemma there is a natural k-linear isomorphism ϕ : HomAM(RA(M ), V ) −→ Hom AY DA(M, IA(V )) for all M ∈ AYDA and V ∈ AM. This isomorphism is given by f ∈ HomAM(RA(M ), V ) 7−→ ϕ(f ) ∈ Hom AY DA(M, IA(V )), (cid:0)ϕ(f )(cid:1)(m) =X f (m(0)) ⊗ m(1) (m ∈ M ). (2.10) Now we recall the following easy lemma from the category theory: Lemma 2.5. Let P : C′ −→ C and Q : D′ −→ D be functors, where C, C′, D and D′ are arbitrary categories, and suppose that there are equivalences F : C −→ D and F ′ : C′ −→ D′ of categories such that the diagram F ′ −−−−→ D′ yQ D C′ Py C −−−−→ F 7 commutes up to isomorphism. If P has a right (left) adjoint I, then J = F ′ ◦ I ◦ F is a right (left) adjoint to Q, where F is a quasi-inverse functor of F . Proof. We only prove the case where P has a right adjoint. Let F ′ be a quasi-inverse of F ′. Then there are isomorphisms HomD′(cid:16)X, J(Y )(cid:17) ∼= HomC′(cid:16)F ′ (X), (I ◦ F )(Y )(cid:17) )(X), F (Y )(cid:17) ′ ∼= HomC(cid:16)(P ◦ F ∼= HomC(cid:16)(F ◦ Q)(X), F (Y )(cid:17) ∼= HomD(cid:16)Q(X), Y(cid:17), which are natural in X ∈ D′ and Y ∈ D. Hence J is a right adjoint to Q. If A is a finite-dimensional Hopf algebra over k, then a Yetter-Drinfel'd A-module M becomes a left D(A)-module by the action given by (p ⊲⊳ a) · m =Php, (am)(1)i(am)(0) for p ∈ A∗, a ∈ A and m ∈ M . This construction establishes an isomorphism AYDA ∼= D(A) monoidal categories (see, e.g., [23]). M of k-linear braided For a while, we denote by F ′ : AYDA −→ D(A) M and R′ A : D(A) M −→ AM the above isomorphism and the restriction functor, respectively. Since R′ A ◦ F ′ = RA (= id ◦ RA), the functor I ′ A by Lemma 2.5. In what follows, based on this observation, we identify D(A)M with AYDA via the isomorphism F ′ and, abusing notation, write R′ A = F ′ ◦ IA is a right adjoint to R′ A as RA and IA, respectively. A and I ′ Proposition 2.6. Let A be a finite-dimensional Hopf algebra over k. Then, the k-linear map Φ : D(A)A −→ IA(k) defined by Φ(a) = 1 ⊗ S−1(a) for all a ∈ A is an isomorphism of left D(A)-modules. Here, k in IA(k) means the trivial left A-module. Proof. Let V be a left A-module. The left A∗ action corresponding to the right coaction of IA(V ) ∈ AYDA is given by p · (v ⊗ a) =Xhp, (v ⊗ a)(1)i(v ⊗ a)(0) =Xhp, a(2)i v ⊗ a(1) for all p ∈ A∗, v ∈ V, a ∈ A. Thus, the left D(A)-action on IA(V ) is given as follows. (p ⊲⊳ h) · (v ⊗ a) = p · (h · (v ⊗ a)) =X p · ((h(2) · v) ⊗ h(3)aS−1(h(1))) =Xhp, (h(3)aS−1(h(1)))(2)i (h(2) · v) ⊗ (h(3)aS−1(h(1)))(1) =Xhp, h(5)a(2)S−1(h(1))i (h(3) · v) ⊗ (h(4)a(1)S−1(h(2))) for all p ∈ A∗, h, a ∈ A, v ∈ V . In particular, when V is the trivial A-module k, (p ⊲⊳ h) · (1 ⊗ a) =Xhp, h(4)a(2)S−1(h(1))i 1 ⊗ (h(3)a(1)S−1(h(2))) =XhS−1(p), h(1)S(a(2))S(h(4))i 1 ⊗ S−1(h(2)S(a(1))S(h(3))). 8 So, identifying k ⊗ H = H, we have This is equivalent to (p ⊲⊳ h) · a =XhS−1(p), h(1)S(a(2))S(h(4))i S−1(h(2)S(a(1))S(h(3))). S((p ⊲⊳ h) · S−1(a)) =XhS−1(p), h(1)a(1)S(h(4))i h(2)a(2)S(h(3)). (2.11) The right-hand side of the above equation coincides with the left action of the Schrodinger module D(A)A. As in a similar to Lemma 2.4, the following holds. Lemma 2.7 (Radford[34, Proposition 1], Hu and Zhang[13, Remark 2.2]). Let A be a bialgebra over k, and suppose that Aop has antipode S. Let N ∈ MA. Then A ⊗ N ∈ AYDA, where the left A action · and the right A-coaction ρ are given by h · (a ⊗ n) = ha ⊗ n, ρ(h ⊗ n) =X(h(2) ⊗ n(0)) ⊗ h(3)n(1)S(h(1)) for all h, a ∈ A and n ∈ N . The correspondence N 7→ A ⊗ N is extended to a functor I A : MA −→ AYDA that is a left adjoint of the forgetful functor RA. The functor I A will be referred to as Radford's induction functor derived from right A-comodules. Let A be a finite-dimensional Hopf algebra over k. The Hopf algebra embedding  : A∗cop ֒→ D(A) defined by (p) = p ⊲⊳ 1A induces a monoidal functor D(A)M −→ A∗cop M ∼= (MA)rev. Here, (MA)rev means the reversed monoidal category of MA. Under the identification AYDA = D(A)M, the above functor D(A)M −→ (MA)rev coincides with the forgetful functor RA given in Lemma 2.7. So, we denote this functor by RA, again. By Lemmas 2.5 and 2.7, we see that the Radford's induction functor I A : (MA)rev −→ AYDA = D(A) M is a left adjoint of RA : D(A) M −→ (MA)rev. Let A be a finite-dimensional Hopf algebra over k. Then the bialgebra Aop has an antipode since the antipode S of A is bijective [31, Theorem 1]. Let M = (M, ·, ρM ) be a finite-dimensional Yetter-Drinfel'd A-module. Then the dual vector space M ∗ becomes a Yetter-Drinfel'd A- module with respect to the following action · and coaction ρM ∗: (a · α)(m) = α(S(a) · m) (a ∈ A, α ∈ M ∗, m ∈ M ), ρM ∗(e∗ j ) = dXi=1 i ⊗ S−1(aji) e∗ (j = 1, . . . , d), where {ei}d j = 1, . . . , d. i=1 is a basis for M and {e∗ i }d i=1 is its dual basis, and ρM (ej) = Pd i=1 ei ⊗ aij for 9 Proposition 2.8. Let A be a finite-dimensional Hopf algebra over k. Then, the k-linear map defined by Φ(q) = S−1(q) ⊗ 1 for all q ∈ A∗ is an isomorphism of left D(A)-modules. Here, k in I A(k) means the trivial right A-comodule. Φ : D(A)A∗cop −→(cid:0)I A(k)(cid:1)∗ Proof. Let W be a finite-dimensional right A-comodule. Applying Radford's induction functor to the dual A-comodule W ∗, we have a Yetter-Drinfel'd A-module I A(W ∗) = A ⊗ W ∗. As is also a Yetter-Drinfel'd A-module whose mentioned before Proposition 2.8, the dual(cid:0)I A(W ∗)(cid:1)∗ left action is given by (h · f )(a ⊗ α) = f (S(h) · (a ⊗ α)) = f (S(h)a ⊗ α) (2.12) for all f ∈ (A ⊗ W ∗)∗, h, a ∈ H, α ∈ W ∗. The coaction ψ of (cid:0)I A(W ∗)(cid:1)∗ i=1 be bases for A and W , respectively. Let {fsi}1≤s≤r 1≤i≤d j=1 and {ei}d Let {xs}r is given as follows. be the dual basis for {xs ⊗ e∗ , that is, fsi ∈ (A ⊗ W ∗)∗ and fsi(xt ⊗ e∗ j ) = δs,tδi,j for all i, j = 1, . . . , d and i }1≤s≤r 1≤i≤d s, t = 1, . . . , r. Then, ψ(fsi) = rXt=1 coaction of W , and write ρW (ej) =Pd where a(t,j),(s,i) ∈ A is defined by ρ(xs ⊗ e∗ dXj=1 i ) =Pd ρ(xs ⊗ e∗ t=1(xt ⊗ e∗ i=1 ei ⊗ aij (aij ∈ A). Since j=1Pr ftj ⊗ S−1(a(s,i),(t,j)), i ) =X((xs)(2) ⊗ e∗ =Xh(xs)(2), x∗ j ) ⊗ (xs)(3)S−1(aij)S−1((xs)(1)) j ) ⊗ (xs)(3)S−1((xs)(1)aij) t i(xt ⊗ e∗ j ) ⊗ a(t,j),(s,i). Let ρW be the by Lemma 2.7(1), we see that a(t,j),(s,i) =Ph(xs)(2), x∗ t i(xs)(3)S−1((xs)(1)aij), and whence ψ(fsi) = h(xt)(2), x∗ rXt=1 dXj=1 si ftj ⊗ S−1(cid:0)(xt)(3)S−1((xt)(1)aji)(cid:1). Under the canonical identification (A ⊗ W ∗)∗ ∼= W ∗∗ ⊗ A∗ ∼= W ⊗ A∗ as vector spaces, we regard W ⊗ A∗ as a Yetter-Drinfel'd A-module. Then, from the above observation it follows that the action and coaction on W ⊗ A∗ are given by a · (w ⊗ p) = w ⊗ (a · p), ψ(w ⊗ p) = hp, (xt)(2)i(w(0) ⊗ x∗ rXt=1 t ) ⊗ S−1(cid:0)(xt)(3)S−1((xt)(1)w(1))(cid:1) for all a ∈ A, w ∈ W, p ∈ A∗, where a · p is the action of A on A∗ as the dual A-module of the left regular A-module. 10 Considering the case where W is the trivial right A-comodule k, we have a Yetter-Drinfel'd A-module structure on A∗ whose action and coaction are given by a · q = ψ(p) = rXt=1 rXt=1 t =Xhq(1), S(a)iq(2), hq, S(a)xtix∗ rXt=1 t ⊗ S−1(cid:0)(xt)(3)S−1((xt)(1))(cid:1). ha · q, xtix∗ t = hp, (xt)(2)ix∗ So, the corresponding left D(A)-module structure on A∗ is given by (p ⊲⊳ h) · q = p · (h · q) hh · q, (xt)(2)ihS−1(p(2)), (xt)(3)ihS−2(p(1)), (xt)(1)ix∗ t t hS−2(p(1))(h · q)S−1(p(2)), xtix∗ hh · q, (xt)(2)ihp, S−1(cid:0)(xt)(3)S−1((xt)(1))(cid:1)ix∗ =X rXt=1 =X rXt=1 =X rXt=1 =Xhq(1), S(h)iS−2(p(1))q(2)S−1(p(2)) S((p ⊲⊳ h) · S−1(q)) =Xhq(2), hip(2)q(1)S−1(p(1)). t for all p, q ∈ A∗ and a ∈ A. Therefore, the left D(A) action on A∗ ∼= (cid:0)I A(k∗)(cid:1)∗ equation satisfies the (2.13) The right-hand side coincides with the left D(A) action on the dual Schrodinger module D(A)A∗cop. Since k∗ ∼= k as right A-comodules, the equation (2.13) gives rise to an isomorphism(cid:0)I A(k)(cid:1)∗ ∼= D(A)A∗cop. 2.4 Tensor products of Schrodinger modules In this subsection, we compute the tensor product of Schrodinger modules by using Proposi- tions 2.6 and 2.8. First, we provide the following lemma, which is proved straightforwardly. Lemma 2.9. Let A be a Hopf algebra over k with bijective antipode. Then: (1) There is a natural isomorphism of Yetter-Drinfel'd modules Φ : IA(V ) ⊗ M −→ IA(V ⊗ RA(M )) (V ∈ AM, M ∈ AYDA) given by Φ(v ⊗ a ⊗ m) =P v ⊗ m(0) ⊗ m(1)a for v ∈ V , a ∈ A and m ∈ M . (2) There is a natural isomorphism of Yetter-Drinfel'd modules Ψ : I A(V ⊗ RA(M )) −→ I A(V ) ⊗ M (V ∈ MA, M ∈ AYDA) given by Ψ(a ⊗ v ⊗ m) =P a(1) ⊗ v ⊗ a(2)m for v ∈ V , a ∈ A and m ∈ M . 11 For a Hopf algebra A, we denote by Aad the adjoint representation of A, i.e., the vector space A endowed with the left A-module structure given by (2.3). Proposition 2.10. Let A be a finite-dimensional Hopf algebra over k. Then Proof. By Proposition 2.6 and Lemma 2.9, (D(A)A)⊗n ∼= IA(A ⊗(n−1) ad ). (D(A)A)⊗n ∼= IA(k) ⊗ (D(A)A)⊗(n−1) ∼= (IA ◦ RA)((D(A)A)⊗(n−1)) ∼= IA(A ⊗(n−1) ad ). The following result is a non-semisimple generalization of a part of [3, Proposition 4]. Proposition 2.11. (D(A)A) ⊗ (D(A)A∗cop) ∼= D(A) as left D(A)-modules. Proof. Let, in general, H be a finite-dimensional Hopf algebra. As is well-known, there are natural isomorphisms of vector spaces HomH(X ⊗ H, Y ) ∼= Homk(X, Y ) ∼= HomH(X, Y ⊗ H) (X, Y ∈ H M). (2.14) Since (RA ◦ I A)(k) ∼= A as left A-modules, there are natural isomorphisms of vector spaces HomD(A)(X, (D(A)A) ⊗ (D(A)A∗cop)) ∼= HomD(A)(X, IA(k) ⊗ I A(k)∗) ∼= HomD(A)(X ⊗ I A(k), IA(k)) ∼= HomA(RA(X ⊗ I A(k)), k) ∼= HomA(RA(X) ⊗ A, k) ∼= RA(X)∗ = X ∗ (by (2.14)) for X ∈ D(A)M. On the other hand, HomD(A)(X, D(A)) ∼= X ∗ by (2.14). Hence the result follows from Yoneda's lemma. 3 Categorical aspects of Schrodinger modules 3.1 The center of a monoidal category We first recall the center construction for monoidal categories, which was introduced by Drinfel'd (see Joyal and Street [16] and Majid [24]). Let C = (C, ⊗, I, a, r, l) be a monoidal category, and let V ∈ C be an object. A half-braiding for V is a natural isomorphism c−,V : (−) ⊗ V −→ V ⊗ (−) such that the diagram (X ⊗ Y ) ⊗ V cX⊗Y,V −−−−−→ V ⊗ (X ⊗ Y ) a−1 V,X,Y←−−−− (V ⊗ X) ⊗ Y aX,Y,Vy X ⊗ (Y ⊗ V ) −−−−−−→ idX ⊗cY,V xcX,V ⊗idY (X ⊗ V ) ⊗ Y X ⊗ (V ⊗ Y ) −−−−→ a−1 X,V,Y 12 aX,Y,Vy xidW ⊗cX,W commutes for all X, Y ∈ C. The center of C, denoted by Z(C), is the category defined as follows: An object of Z(C) is a pair (V, c−,V ) consisting of an object V ∈ C and a half braiding c−,V for V . If (V, c−,V ) and (W, c−,W ) are objects of Z(C), then a morphism from (V, c−,V ) to (W, c−,W ) is a morphism f : V −→ W in C such that (f ⊗ idX) ◦ cX,V = cX,W ◦ (idY ⊗ f ) for all X ∈ C. The composition of morphisms in Z(C) is given in an obvious way. Forgetting the half-braiding defines a faithful functor ΠC : Z(C) −→ C, which will be referred to as the forgetful functor from Z(C) to C. The category Z(C) is in fact a braided monoidal category: First, the tensor product of objects of Z(C) is defined by (V, c−,V ) ⊗ (W, c−,W ) = (V ⊗ W, c−,V ⊗W ), where the half-braiding c−,V ⊗W for V ⊗ W is a unique natural isomorphism making the diagram X ⊗ (V ⊗ W ) cX,V ⊗W −−−−−→ (V ⊗ W ) ⊗ X aX,V,W←−−−−− (X ⊗ V ) ⊗ W (X ⊗ V ) ⊗ W −−−−−−→ cX,V ⊗idX (V ⊗ X) ⊗ W −−−−−→ aV,X,W V ⊗ (X ⊗ W ) commute for all X ∈ C. The unit object is I Z(C) := (I, l−1 ◦ r). The associativity and the unit constraints for Z(C) are defined so that the forgetful functor ΠC is strict monoidal. Finally, the braiding of Z(C) is given by c =ncV,W : (V, c−,V ) ⊗ (W, c−,W ) −→ (W, c−,W ) ⊗ (V, c−,V )o(V,c−,V ),(W,c−,W )∈Z(C) . Since the center construction is described purely in terms of monoidal categories, it is natural to expect that equivalent monoidal categories have equivalent centers. We omit to give a proof of the following well-known fact, since the proof is easy but quite long. Lemma 3.1. For each monoidal equivalence F : C −→ D between monoidal categories C and D, there exists a unique braided monoidal equivalence Z(F ) : Z(C) −→ Z(D) such that ΠD ◦ Z(F ) = ΠC ◦ F as monoidal functors. The assignment F 7→ Z(F ) enjoys the following properties: (1) If C F−→ D G−→ E is a sequence of monoidal equivalences between monoidal categories C, D and E, then Z(G ◦ F ) = Z(G) ◦ Z(F ). (2) Let F1, F2 : C −→ D be monoidal equivalences. For each monoidal natural transformation α : F1 −→ F2, a monoidal natural transformation Z(α) : Z(F1) −→ Z(F2) 13 is defined by Z(α)X = αX for X = (X, σX ) ∈ Z(C). Moreover, α 7→ Z(α) preserves the horizontal and the vertical compositions of natural transformations. By a k-linear monoidal category, we mean a monoidal category C such that HomC(X, Y ) is a vector space over k for all objects X, Y ∈ C, and the composition and the tensor product of morphisms in C are linear in each variables. We note that if C is a k-linear monoidal category, then so is Z(C) in such a way that the functor ΠC is k-linear. If F : C −→ D is a k-linear monoidal equivalence between k-linear monoidal categories, then Z(F ) is k-linear. 3.2 Schrodinger modules in monoidal categories Let C be a monoidal category. We say that an object A ∈ Z(C) is a Schrodinger object if Z(C)) ∼= HomZ(C)(X, A), which is natural in the variable X ∈ Z(C). Note that such an object is unique up to isomorphism by Yoneda's lemma (if it there exists a bijection HomC(ΠC(X), I exists). Lemma 3.2. Let C and D be monoidal categories. (1) Suppose that ΠC : Z(C) −→ C has a right adjoint functor IC : C −→ Z(C). Then the object IC(I) is a Schrodinger object for C. (2) Suppose that AC is a Schrodinger object for C, and there exists an equivalence F : C −→ D of monoidal categories. Then Z(F )(AC ) is a Schrodinger object for D. Proof. Part (1) follows immediately from the definition of the adjoint functor. Part (2) is shown as follows: Let F be a quasi-inverse of F . Then Z(F ) is a quasi-inverse of Z(F ). Hence, HomZ(D)(cid:0)X, Z(F )(AC )(cid:1) ∼= HomZ(C)(cid:0)Z(F )(X), AC(cid:1) ∼= HomC(cid:0)(ΠC ◦ Z(F ))(X), I(cid:1) ∼= HomD(cid:0)(F ◦ ΠD)(X), I(cid:1) ∼= HomD(cid:0)ΠD(X), F (I)(cid:1) ∼= HomD(cid:0)ΠD(X), I(cid:1) for all X ∈ Z(C). Therefore, Z(F )(AC) is the Schrodinger object for D. Now, let A be a Hopf algebra over k with bijective antipode. Then ΦA : Z(AM) −→ AYDA, (V, c−,V ) 7→ (V, ρ), (3.1) where ρ : V → V ⊗ A is the linear map given by ρ(v) = cA,V (1A ⊗ v) (v ∈ V ), is an isomorphism of k-linear braided monoidal categories [16, 24]. Theorem 3.3. Let A and B be Hopf algebras over k with bijective antipode. Suppose that there is an equivalence F : AM −→ BM of k-linear monoidal categories. (1) SA := (Φ−1 A ◦ IA)(k) ∈ Z(AM) is a Schrodinger object for AM. 14 Proof. To prove Part (1), note that the diagram (2) There uniquely exists an equivalence eF : AYDA −→ BYDB of k-linear braided monoidal categories such that RB ◦ eF = F ◦ RA as monoidal functors. (3) The functor eF in Part (2) satisfies IB ◦ F ∼= eF ◦ IA. yRA ΠAMy ΦA−−−−→ AYDA AM −−−−→ Z(AM) (3.2) AM id commutes. Applying Lemma 2.5 to this diagram, we see that Φ−1 to ΠAM. Hence SA is a Schrodinger object for AM by Part (1) of Lemma 3.2. A ◦ IA is a right adjoint functor B ◦ Z(F ) ◦ ΦA satisfies the required conditions. To show the uniqueness, note that (3.2) is in fact a commutative diagram of monoidal Now we show Part (2). It is easy to see that eF = Φ−1 functors. Thus, if eF satisfies the required conditions, then, by (3.2), ΠB M ◦ (ΦB ◦ eF ◦ Φ−1 as monoidal functors. Hence eF = Φ−1 By Part (2) and Lemma 2.5, I ′ := eF ◦ IA ◦ F is a right adjoint to RA. By the uniqueness of the right adjoint functor, we have I ′ ∼= IB. Hence, IB ◦ F ∼= eF ◦ IA. To prove Part (3), we recall that IA is a right adjoint to RA. Let F be a quasi-inverse of F . A ) = RB ◦ eF ◦ Φ−1 Suppose that A is finite-dimensional. Then we obtain an isomorphism B ◦ Z(F ) ◦ ΦA by the uniqueness part of Lemma 3.1. A = F ◦ RA ◦ Φ−1 A = F ◦ ΠAM ΨA : Z(AM) ∼=−→ D(A)M (3.3) of k-linear braided monoidal categories by composing (3.1) and the isomorphism AYDA ∼= D(A) (which we have recalled in Section 2). Note that M ΠAM = RA ◦ ΨA as monoidal functors, where RA is the restriction functor D(A) M −→ AM (cf. (3.1) and (3.2)). Let IA be the right adjoint functor of RA given in Section 2. Recall from Proposition 2.8 that the Schrodinger module D(A)A is isomorphic to IA(k). By the same way as the previous theorem, we prove: Theorem 3.4. Let A and B be finite-dimensional Hopf algebras over the same field k. Suppose that there is an equivalence F : AM −→ BM of k-linear monoidal categories. (1) SA := Ψ−1 A (D(A)A) ∈ Z(AM) is a Schrodinger object for AM. 15 An important corollary is the following invariance of the Schrodinger module: M −→ D(B) (2) There uniquely exists an equivalence eF : D(A) categories such that RB ◦ eF = F ◦ RA as monoidal functors. (3) The functor eF in Part (2) satisfies IB ◦ F ∼= eF ◦ IA. Corollary 3.5. The equivalence eF : D(A) module, i.e., eF (D(A)A) ∼= D(B)B. Remark 3.6. An equivalence G : D(A) M −→ D(B) does not preserve the Schrodinger module in general. M −→ D(B) M of k-linear braided monoidal M of Theorem 3.4 preserves the Schrodinger M of k-linear braided monoidal categories Remark 3.7. As Masuoka pointed out to us, Corollaries 3.5 and 3.9 below can be also derived from the point of view of cocycle deformations by using the action given in [26, Proposition 5.1]. 3.3 The dual Schrodinger module as a Schrodinger object Let A be a finite-dimensional Hopf algebra over k. Recall from Section 2 that there exists an isomorphism D(A) ∼= D(A∗)cop of quasitriangular Hopf algebras. Since MA is isomorphic to A∗M as k-linear monoidal categories, we have isomorphisms Z(MA)rev ∼= Z(A∗M)rev ∼= (D(A∗) M)rev ∼= D(A∗)cop M ∼= D(A) M (3.4) of k-linear braided monoidal categories. Now, let ΞA : Z(MA)rev −→ D(A)M be the composition of the above isomorphisms. One can check that the diagram Z(MA)rev ΞA−−−−→ D(A) ΠMAy M yRA (MA)rev −−−−→ (MA)rev id (3.5) commutes, where RA is the functor used in Subsection 2.1. By Lemma 2.7 the functor I A : MA −→ D(A) M is a left adjoint to RA. The following theorem is proved by the same way as Theorem 3.4. Theorem 3.8. Let A and B be finite-dimensional Hopf algebras over the same field k. Suppose that there is an equivalence F : MA −→ MB of k-linear monoidal categories. (1) There uniquely exists an equivalence eF : D(A) categories such that RB ◦ eF = F ◦ RA as monoidal functors. (2) The functor eF in Part (1) satisfies I B ◦ F ∼= eF ◦ I A. M −→ D(B) M of k-linear braided monoidal 16 In Proposition 2.8 it is shown that the dual Schrodinger module D(A)A∗cop is isomorphic to the left dual of I A(k). The following corollary is obtained by Part (2) of the above theorem, and the fact that a monoidal equivalence preserves left duals. Corollary 3.9. The equivalence eF : D(A)M −→ D(B)M of Theorem 3.8 preserves the dual Schrodinger module, i.e., eF (D(A)A∗cop) ∼= D(B)B∗cop. We have introduced the notion of a Schrodinger object to explain categorical nature of the Schrodinger module. The following theorem claims that also the dual Schrodinger object can be interpreted in terms of a Schrodinger object: Theorem 3.10. SA := Ξ−1 A (D(A)A∗cop) ∈ Z(MA) is a Schrodinger object for MA. Proof. For simplicity, write HomMA(X, Y ) as HomA(X, Y ). By Proposition 2.8 and Lemma 2.9, we obtain natural isomorphisms HomD(A)(X, D(A)A∗cop) ∼= HomD(A)(I A(k) ⊗ X, k) ∼= HomD(A)(I A(k ⊗ RA(X)), k) ∼= HomA(RA(X), k) for X ∈ D(A)M. Hence there are natural isomorphisms HomZ(MA)(X, SA) ∼= HomD(A)(ΞA(X), D(A)A∗cop) for X ∈ Z(MA). This shows that SA ∈ Z(MA) is a Schrodinger object for MA. ∼= HomA(cid:0)(RA ◦ ΞA)(X), k(cid:1) ∼= HomA(cid:0)ΠMA(X), k(cid:1) The following theorem is a slight weak version of Proposition 2.2 and its corollary. Empha- sizing the role of the Schrodinger object, we now reexamine the proof of Proposition 2.2 and its corollary. Theorem 3.11. For a finite-dimensional Hopf algebra A over k, we denote by FA : (D(A∗) M)rev = D(A∗)cop M −→ D(A) M (3.6) the isomorphism of k-linear braided monoidal categories induced by the isomorphism φA given by (2.2). Then there are isomorphisms FA(cid:0) D(A∗)A∗(cid:1) ∼= D(A)A∗cop and FA(cid:0) D(A∗)(A∗)∗cop(cid:1) ∼= D(A)A. Proof. To establish the first isomorphism, we first note that the category isomorphism FA is expressed as follows (cf. (3.4)): FA : D(A∗)cop M ∼= (D(A∗) M)rev ΨA∗ −−−−→ Z(A∗ M)rev ∼=−−−−→ Z(MA)rev ΞA−−−−→ D(A) M, 17 where the arrow Z(A∗ M)rev −→ Z(MA)rev is the isomorphism induced by A∗M ∼= MA. Theo- rem 3.4 tells us that D(A∗)A∗ is an object corresponding to a Schrodinger object for A∗M. On the other hand, Theorem 3.10 tells us that D(A)A∗ is an object corresponding to a Schrodinger object for MA. Hence, by Lemma 3.2, we have FA(D(A∗)A∗) ∼= D(A)A∗cop. Once the first isomorphism is obtained, the second isomorphism can be obtained by replacing A with A∗ (cf. the proof of Corollary 2.3). 4 Applications Motivated by the construction of quantum representations of the n-strand braid group Bn due to Reshetikhin and Turaev [36], a family of monoidal Morita invariants of a finite-dimensional Hopf algebra, which is indexed by braids, can be obtained from the Schrodinger module. Let A be a finite-dimensional Hopf algebra. It turns out that the invariant associated with the identity element 1 ∈ B1 is equal to the quantum dimension the Schrodinger module D(A)A in the sense of Majid [25], and thus equal to Tr(S2) by [25, Example 9.3.8], where S is the antipode of A (see Bulacu-Torrecillas [2] for the case of quasi-Hopf algebras). As is well-known, Tr(S2) has the following representation-theoretic meaning: Tr(S2) 6= 0 if and only if A is semisimple and cosemisimple [30]. In this section, we show that the invariants derived from other braids, like , involve further interesting results connecting with representation theory. The invariant associated with a braid b is, roughly speaking, defined by "coloring" the closure of b by the Schrodinger module as if we were computing the quantum invariant of a (framed) link. Such an operation is not allowed in general since D(A) we will use the (partial) braided trace, introduced below, to define invariants. M may not be a ribbon category. So 4.1 Partial traces in braided monoidal categories We briefly recall the concept of duals in a monoidal category C = (C, ⊗, I, a, r, l). For an object X ∈ C the triple (X ∗, eX , nX) consisting of an object X ∗ ∈ C and morphisms eX : X ∗ ⊗ X −→ I, nX : I −→ X ⊗ X ∗ in C is said to be a left dual if two compositions l−1 −−−→ I ⊗ X X nX ⊗id −−−−→ (X ⊗ X ∗) ⊗ X a−−→ X ⊗ (X ∗ ⊗ X) id⊗eX−−−−→ X ⊗ I r−−→ X, X ∗ r−1 −−−→ X ∗ ⊗ I id⊗nX−−−−→ X ∗ ⊗ (X ⊗ X ∗) a−1 −−→ (X ∗ ⊗ X) ⊗ X ∗ eX ⊗id −−−−→ I ⊗ X l−→ X are equal to idX and idX ∗, respectively. If all objects in C have left duals, then the monoidal category is called left rigid. From now on, all monoidal categories are assumed to be strict although almost all definitions and results are not needed this assumption. 18 Let (C, c) is a left rigid braided monoidal category. We choose a left dual (X ∗, eX , nX ) for each object X ∈ C. Let f : X ⊗ Y −→ X ⊗ Z be an morphism in C. Then the following composition Trl,X (f ) : Y −→ Z can be defined: c l−1 Y−−−→ I ⊗ Y Y nX ⊗idY −−−−−−→ X ⊗ X ∗ ⊗ Y id⊗f −−−−→ X ∗ ⊗ X ⊗ Z c−1 X ∗,X −−−−−−−−→ X ∗ ⊗ X ⊗ Y ⊗idY eX ⊗idZ −−−−−−→ I ⊗ Z lZ−−→ Z. The morphism Trl,X c (f ) : Y −→ Z is said to be the left partial braided trace of f on X. Similarly, for a morphism f : Y ⊗ X −→ Z ⊗ X, the right partial braided trace of f on X is defined by the composition r−1 Y−−−→ Y ⊗ I Y idY ⊗nX −−−−−−→ Y ⊗ X ⊗ X ∗ id⊗cX,X ∗ −−−−−−−→ Z ⊗ X ∗ ⊗ X f ⊗id −−−−→ Z ⊗ X ⊗ X ∗ idZ ⊗eX −−−−−−→ Z ⊗ I rZ−−−→ Z. Trl,X c (f ) = X* X f X X* Z Y Trr,X c (f ) = Z Y X* X f X X* Figure 1: the left and right partial traces The left and right partial braided traces on X are does not depend on the choice of left duals of X. For morphisms f : X ⊗ Y −→ X ⊗ Z and g : Y ′ −→ Y, h : Z −→ Z ′, we have and for morphisms f ′ : Y ⊗ X −→ Y ⊗ X and g : Y ′ −→ Y, h : Z −→ Z ′, we have Trl,X c (cid:0)(idX ⊗ h) ◦ f ◦ (idX ⊗ g)(cid:1) = h ◦ Trl,X (cid:0)(h ⊗ idX) ◦ f ′ ◦ (g ⊗ idX)(cid:1) = h ◦ Trr,X c c Trr,X c (f ) ◦ g : Y −→ Z, (4.1) (f ′) ◦ g : Y −→ Z. (4.2) For an endomorphism f : X −→ X in (C, c), the left braided trace Trl,X (f ) and the right c braided trace Trr,X They coincide with the following compositions, respectively. (f ) are defined by Trl,X (f ) := Trl,X c c c (f ⊗ idI) and Trr c(f ) := Trr,X c (idI ⊗ f ). Trl c(f ) : I Trr c(f ) : I nX−−−→ X ⊗ X ∗ nX−−−→ X ⊗ X ∗ f ⊗id c−1 X ∗,X−−−−→ X ∗ ⊗ X −−−→ X ⊗ X ∗ cX,X ∗ id⊗f −−−→ X ∗ ⊗ X −−−−→ X ∗ ⊗ X eX−−−→ I, eX−−−→ I. 19 Lemma 4.1. Let (C, c) and (D, c′) be left rigid braided monoidal categories, and (F, φ, ω) : C −→ D be a braided monoidal functor. Then (1) For any morphism f : X ⊗ Y −→ X ⊗ Z in C (2) For any morphism f : Y ⊗ X −→ Z ⊗ X in C c F(cid:0)Trl,X F(cid:0)Trr,X c c′ (f )(cid:1) = Trl,F (X) (f )(cid:1) = Trr,F (X) (cid:0)φ−1 X,Z ◦ F (f ) ◦ φX,Y(cid:1). Z,X ◦ F (f ) ◦ φY,X(cid:1). (cid:0)φ−1 c′ (4.3) (4.4) Proof. For each X ∈ C we choose a left dual (X ∗, eX , nX ). Then (F (X ∗), e′ left dual of F (X), where F (X), n′ F (X)) is a F (X) := ω−1 ◦ F (eX ) ◦ φX ∗,X : F (X ∗) ⊗ F (X) −→ I′, e′ F (X) := φ−1 n′ X,X ∗ ◦ F (nX) ◦ ω : I′ −→ F (X) ⊗ F (X ∗). By using this left dual of F (X) and computing the partial braided trace Trl,F (X) φX,Y(cid:1), we have the desired equation (4.3). Part (2) is also proved as in the same manner with Part (1). X,Z ◦ F (f ) ◦ c (cid:0)φ−1 Let us describe a relationship between left and right partial traces. Let c be a braiding Y,X : for a monoidal category C. Then, the collection ¯c consisting of isomorphisms ¯cX,Y := c−1 X ⊗ Y −→ Y ⊗ X over all pairs (X, Y ) of objects in C is also a braiding for C. Let (C, c) be a left rigid braided monoidal category chosen left duals (X ∗, eX , nX) for all objects X in C. Then, for two objects X, Y in C there is a natural isomorphism jX,Y : Y ∗ ⊗ X ∗ −→ (X ⊗ Y )∗ such that eX⊗Y ◦ (jX,Y ◦ idX⊗Y ) = eY ◦ (idY ∗ ⊗ eX ⊗ idY ) [12]. For any morphism f : X −→ Y in (C, c), there is a unique morphism tf : Y ∗ −→ X ∗ in C, which is characterized by eX ◦ (tf ⊗ idX) = eY ◦ (idY ∗ ⊗ f ). The morphism tf is called the left transpose of f . The left and right partial traces are related as follows. Lemma 4.2. For any morphism f : X ⊗ Y −→ X ⊗ Z in a left rigid braided monoidal category (C, c), Trr,X ∗ c (j−1 X,Y ◦ tf ◦ jX,Z ) = t(cid:0)Trl,X ¯c (f )(cid:1). Proof. The equation of the lemma is obtained from a graphical calculus depicted as in Figure 2. Let M be an object in a strict left rigid braided monoidal category (C, c). For each endomor- phism f ∈ End(M ⊗n) and each positive integer k (1 ≤ k ≤ n), we set Trl,k Trr,k c (f ) := Trr,M ⊗k (f ), and c c (f ) := Trl,M ⊗k c (f ), (Trl,1 c ◦ · · · ◦ Trl,1 r c(f ) := (Trr,1 c ◦ · · · ◦ Trr,1 c )(f ). (4.5) z l c(f ) := fTr n } { c )(f ), fTr 20 z n } { * Y (cid:1) * Y (cid:1) * Y (cid:1) * Y (cid:1) * Y (cid:1) X* X* X* X* X* X Z jX,Y(cid:160)-1 (X⊗Y(cid:160))* tf(cid:160) (X⊗Z(cid:160))* jX,Z(cid:160) = X jX,Y(cid:160)-1 f(cid:160) jX,Z(cid:160) jX,Z(cid:160) = f(cid:160) = f(cid:160) = X jX,Y(cid:160)-1 X** X X* X* X** f Y X X* * Z(cid:160) * Z(cid:160) * Z(cid:160) * Z(cid:160) * Z(cid:160) Figure 2: a graphical calculus for the proof of Lemma 4.2 The modified traces (4.5) are preserved by a braided monoidal functor. More precisely: Proposition 4.3. Let C = (C, c) and D = (D, c′) be strict left rigid braided monoidal categories, and (F, φ, ω) : C −→ D be a braided monoidal functor. Let M be an object in C, and k be a positive integer, and define the isomorphism φ(k) : F (M )⊗k −→ F (M ⊗k) in D by φ(1) := idF (M ), φ(k) := φM,M ⊗(k−1) ◦ (idF (M ) ⊗ φ(k−1)) (k ≥ 2). Then for an endomorphism f on M ⊗n in C, the following equations hold. l r c′(cid:0)(φ(n))−1 ◦ F (f ) ◦ φ(n)(cid:1) = ω−1 ◦(cid:0)F(cid:0)fTr fTr c′(cid:0)(φ(n))−1 ◦ F (f ) ◦ φ(n)(cid:1) = ω−1 ◦(cid:0)F(cid:0)fTr fTr Therefore, if C, D and (F, φ, ω) are k-linear, then l r c(f )(cid:1)(cid:1) ◦ ω, c(f )(cid:1)(cid:1) ◦ ω. l c′(cid:0)(φ(n))−1 ◦ F (f ) ◦ φ(n)(cid:1) =fTr fTr as elements in k. l c(f ), fTr r c′(cid:0)(φ(n))−1 ◦ F (f ) ◦ φ(n)(cid:1) =fTr Proof. We set g := (φ(n))−1 ◦ F (f ) ◦ φ(n). By (4.1) and Lemma 4.1(1) we have (4.6) (4.7) r c(f ) The same arguments for f1 := Trl,1 Trl,1 M,M ⊗(n−1) ◦ F (f ) ◦ φM,M ⊗(n−1)(cid:1)(cid:17) ◦ φ(n−1) c′ (g) = (φ(n−1))−1 ◦(cid:16)Trl,1 c′(cid:0)φ−1 = (φ(n−1))−1 ◦ F(cid:0)Trl,1 c (f )(cid:1) ◦ φ(n−1). c′ (g)(cid:1) = (φ(n−2))−1 ◦ F(cid:0)Trl,1 c′(cid:0)Trl,1 c (f )(cid:1)(cid:1) ◦ φ(n−2). c′ (g) provide the equation c (f ) and g1 := Trl,1 c (cid:0)Trl,1 Trl,1 By repeating the same arguments, the equation (Trl,1 c′ ◦ · · · ◦ Trl,1 (Trl,1 c ◦ · · · ◦ Trl,1 (4.8) z n−1 } { z c′ )(g) = F(cid:0) 21 n−1 } { c )(f )(cid:1) c )(f ) and applying Trl,1 c′ to the equation (4.8), we is obtained. Setting fn−1 := have the desired equation n−1 z (Trl,1 } c ◦ · · · ◦ Trl,1 { c′(cid:16)F(cid:0)fn−1(cid:1)(cid:17) = ω−1 ◦ F(cid:0)Trl,1 l fTr c′(g) = Trl,1 As in a similar way, the equation (4.7) can be shown by using φ(k) = φM ⊗(k−1),M ◦ (φ(k−1) ⊗ l c(f ) = c(f )(cid:1) ◦ ω. idF (M )). The last two equations are immediately derived from (4.6) and (4.7) by settingfTr λidk andfTr c (fn−1)(cid:1) ◦ ω = ω−1 ◦ F(cid:0)fTr As the same manner of the proof of the above proposition with help from Lemma 4.2 we r c(f ) = λ′idk for some λ, λ′ ∈ k. l have: Proposition 4.4. Let C = (C, c) be a k-linear strict left rigid braided monoidal category. Let M be an object in C, and k be a positive integer, and define the isomorphism j(k) : (M ∗)⊗k −→ (M ⊗k)∗ in C by j(1) := idM ∗, j(k) := jM ⊗(k−1),M ◦ (idM ∗ ⊗ j(k−1)) (k ≥ 2). Then for an endomorphism f on M ⊗n in C, the following equation holds as elements in k: (4.9) Given a quasitriangular Hopf algebra (A, R), a braiding cR = { cR X,Y : X ⊗ Y −→ Y ⊗ X }X,Y ∈AM is defined by r l ¯c(f ). c(cid:0)(j(n))−1 ◦ tf ◦ j(n)(cid:1) =fTr fTr (cR)X,Y (x ⊗ y) = TX,Y(cid:0)R · (x ⊗ y)(cid:1) for all x ∈ X and y ∈ Y , where R · (x ⊗ y) is the diagonal action of R to X ⊗ Y . Denoted by (A,R)M is the braided monoidal category (AM, cR). We use the notation TrR instead of TrcR. Example 4.5. Let (A, R) a quasitriangular Hopf algebra over k, and u be the Drinfel'd element of it. It is well-known that the Drinfel'd element u is invertible, and when R is written as R =Pj αj ⊗βj, the inverse is given by R−1 = (S ⊗id)(R) =Pj S(αj)⊗βj, and u =Pj S(βj)αj and u−1 =Pj βj S2(αj) [6, 32]. Let M be a finite-dimensional left A-module. For any a ∈ A the action of a on M is denoted by aM . Then for any A-module endomorphism f on M ⊗n the following formulas hold: l R(f ) = Tr((u−1 M ⊗ · · · ⊗ u−1 M ) ◦ f ), r R(f ) = Tr((uM ⊗ · · · ⊗ uM ) ◦ f ), (4.10) (4.11) fTr fTr where Tr in the right-hand side stands for the usual trace on linear transformations. 22 Proof. Here, we only prove the first equation since the second equation can be proved by the same argument. The equation (4.10) can be shown by induction on n as follows. Let {ei}d i=1 be a basis for M . For any a ∈ A, a· ei is expressed as a· ei =Pd R (f ) =P Mi′,i(βj)Mk,i′(f )Mi,k(S2(αj)) =P Mi′,i′(u−1 i′=1 Mi′,i(a) ei′ for M ◦ R(f ) = Trl,1 l M ) ◦ g) holds for any A-module endomorphism g on M ⊗(n−1). Let f be an A-module endomorphism on M ⊗n. Then g := Trl,1 R (f ) is an A-module endomorphism on M ⊗(n−1). Applying the induction hypothesis, we R(g) = Tr((u−1 M ⊗ · · · ⊗ u−1 M ◦ f ). f ) = Tr(u−1 some Mi′,i(a) ∈ k. ThenfTr Next, assume that the equationfTr R(f ) =fTr havefTr R(g) = Tr((u−1 l l l (n−1) {z M ⊗ · · · ⊗ u−1 M ) ◦g) = Tr((u−1 M ⊗ · · · ⊗ u−1 M ) ◦f ). } n {z } 4.2 Construction of monoidal Morita invariants In this subsection we introduce a family of monoidal Morita invariants of a finite-dimensional Hopf algebra by using partial braided traces. Let C = (C, c) be a strict left rigid braided monoidal category, and M be an object in C. Then there is a representation ρM : Bn −→ GL(M ⊗n) of the n-strand braid group Bn such that each positive crossing and negative crossing correspond to cM,M and c−1 M,M , respectively [36]. For each b ∈ Bn we set b - diml c(M ) :=fTr l c(cid:0)ρM (b)(cid:1), b - dimr c(M ) :=fTr r c(cid:0)ρM (b)(cid:1). By Example 4.5, 1 - dimr c(M ) = (the quantum dimension of M ) (4.12) in the sense of [25], where 1 is the identity element of B1. Lemma 4.6. Let M and N be two objects in C. (1) If M and N are isomorphic, then b - diml (2) b - dimr c(M ∗) = b - diml ¯c(M ). c(M ) = b - diml c(N ), b - dimr c(M ) = b - dimr c(N ). Proof. (1) Let ϕ : M −→ N be an isomorphism. The map ϕ⊗n : M ⊗n −→ N ⊗n is also an isomorphism. Let ρM : Bn −→ GL(M ⊗n) and ρN : Bn −→ GL(N ⊗n) be the representations induced from the braiding c. Since cN,N ◦ (ϕ ⊗ ϕ) = (ϕ ⊗ ϕ) ◦ cM,M from naturality of c, the endomorphisms f := ρM (b) and g := ρN (b) satisfy g ◦ ϕ⊗n = ϕ⊗n ◦ f . Thus, g is expressed l as g = (ϕ⊗n) ◦ f ◦ (ϕ⊗n)−1, and it follows from Proposition 4.3 that b - diml c(g) = c(N ) is also shown by the same c(M ). The equation b - dimr l c(f ) = b - diml c(M ) = b - dimr c(N ) = fTr (2) By the definition of the natural isomorphism jM,N : N ∗ ⊗ M ∗ −→ (M ⊗ N )∗, it is easy to see that jM,N ◦ cM ∗,N ∗ = t(cM,N ) ◦ jN,M . It follows that the representation ρM ∗ : Bn −→ fTr argument. 23 is the isomorphism defined in Proposition 4.4. Thus we have b - dimr GL((M ∗)⊗n) induced from the braiding c satisfies j(n) ◦ ρM ∗(b) = t(cid:0)ρM (b)(cid:1) ◦ j(n), where j(n) c(cid:0)(j(n))−1 ◦ t(cid:0)ρM (b)(cid:1) ◦ j(n)(cid:1) =fTr Let (A, R) be a quasitriangular Hopf algebra over k, and M be a finite-dimensional left ¯c(cid:0)ρM (b)(cid:1) = b - diml c(M ∗) = fTr ¯c(M ). r l A-module. For each b ∈ Bn we set b - diml R(M ) := b - diml cR(M ), b - dimr R(M ) := b - diml cR(M ). In particular, in the case where (A, R) = (D(H), R) for some finite-dimensional Hopf algebra H over k, we denote b - diml(M ) := b - diml R(M ), b - dimr(M ) := b - dimr R(M ). Now we will show that b - diml(D(H)H) and b - dimr(D(H)H) are preserved under k-linear monoidal equivalences of the module categories. Theorem 4.7. Let A and B be finite-dimensional Hopf algebras over k. If AM and BM are equivalent as k-linear monoidal categories, then b - dimr(D(A)A) = b - dimr(D(B)B) for all b ∈ Bn. The same statement holds for b - diml. c′ F (M ) = b - dimr Proof. Let (F, φ, ω) : (AM, c) −→ (BM, c′) be a k-linear braided monoidal functor. Then b - dimr cM for all finite-dimensional left A-module M and b ∈ Bn. This equation can be shown as follows. Since φM,M ◦ c′ F (M ),F (M ) = F (cM,M ) ◦ φM,M , and f := ρM (b) : M ⊗n −→ M ⊗n is expressed by compositions and tensor products of c±1 M,M and idM , the isomorphism φ(n) : F (M )⊗n −→ F (M ⊗n) defined in Proposition 4.2 satisfies φ(n) ◦ ρF (M )(b) = F (f ) ◦ φ(n). Therefore, by Proposition 4.2 we have b - dimr c′(cid:0)F (M )(cid:1) =fTr r c′(cid:0)(φ(n))−1 ◦ F (f ) ◦ φ(n)(cid:1) =fTr r c(f ) = b - dimr c(M ). Suppose that (F, φ, ω) : AM −→ BM is an equivalence of k-linear monoidal categories. By Corollary 3.5, F (D(A)A) ∼= D(B)B as left D(B)-modules. It follows from Lemma 4.6 that b - dimr R′(D(B)B), where R and R′ are the canonical universal R-matrices of D(A) and D(B), respectively. R′ F (D(A)A) = b - dimr R(D(A)A) = b - dimr By using Theorem 3.11 the monoidal Morita invariants b - diml and b - dimr of the dual Schrodinger modules D(A)A∗cop and D(A∗)(A∗)∗cop are computable from the monoidal Morita invariants of the Schrodinger modules D(A∗)(A∗) and D(A)A, respectively. Proposition 4.8. Let A be a finite-dimensional Hopf algebra over k, and R, R′ be the canonical universal R-matrices of D(A), D(A∗), respectively. For any b ∈ Bn, the following equations hold. (1) b - diml(D(A∗)(A∗)∗cop) = b - diml(D(A)A), b - dimr(D(A∗)(A∗)∗cop) = b - dimr(D(A)A). (2) b - diml(D(A∗)A∗) = b - diml(D(A)A∗cop), b - dimr(D(A∗)A∗) = b - dimr(D(A)A∗cop). 24 M, cR′ )rev −→ (D(A) M, cR) be the equivalence of braided monoidal R(FA(M )) = R′(M ) from the proof of Theorem 4.7. Since FA(M ) and D(A)A are isomorphic as R(FA(M )) = Proof. (1) Let FA : (D(A∗) categories defined in Theorem 3.11. Setting M := D(A∗)(A∗)∗cop, we have b - dimr b - dimr left D(A)-modules by Proposition 2.2, it follows from Lemma 4.6(1) that b - dimr b - dimr b - diml(D(A∗)(A∗)∗cop) = b - diml(D(A)A) can be proved. R(D(A)A). Thus, the second equation is obtained. Similarly, the equation Part (2) can be proved as in the proof of (1) by using Corollary 2.3 instead of Proposition 2.2. Let A be a finite-dimensional Hopf algebra. In view of Example 4.5, it is important to know the action of the Drinfel'd element u ∈ D(A) on a given D(A)-module M to compute the braided dimension of M . Below we give formulas for the actions of u and S(u) on the Schrodinger module D(A)A. Recall that a left integral in A is an element Λ ∈ A such that aΛ = ε(a)Λ for all a ∈ A. A right integral in A is a left integral in Aop. It is known that a non-zero left integral Λ ∈ A always exists (under our assumption that A is finite-dimensional), and is unique up to a scalar multiple. Hence one can define α ∈ A∗ by Λa = hα, aiΛ for a ∈ A. The map α is in fact an algebra map, and does not depend on the choice of Λ. We call α the distinguished grouplike element of A∗. The Hopf algebra A is said to be unimodular if the distinguished grouplike element α ∈ A∗ is the counit of A, or, equivalently, Λ ∈ A is central. Lemma 4.9. With the above notations, we have u ⇀ a =X S2(a(1))(cid:10)α−1, a(2)(cid:11) for all a ∈ D(A)A, where α−1 = α ◦ S. and S(u) ⇀ a = S−2(a) Proof. Recall that u =Pi S−1 basis to {ei}. We first compute the action of S(u). Since S2(u) = u, A∗ (e∗ i ) ⊲⊳ ei, where {ei} is a fixed basis of A, and {e∗ i } is the dual S(u) = S−1(u) =Xi (ε ⊲⊳ S−1 A (ei)) · (e∗ i ⊲⊳ 1). Hence, for all a ∈ D(A)A, we have S(u) ⇀ a =Xi (cid:10)e∗ =Xi (cid:10)e∗ =X S−1(cid:16)S−1(a(1))(2)(cid:17)a(2)S−1(a(1))(1) = S−2(a). i , S−1(a(1))(cid:11)(cid:0)S−1(ei) ⇀ a(2)(cid:1) i , S−1(a(1))(cid:11) S−1(ei(2))a(2)ei(1) Next, we compute the action of u. Fix a non-zero right integral λ ∈ A∗, and define g ∈ A to be the unique element such that pλ = hp, giλ for all p ∈ A∗ (i.e., the distinguished grouplike 25 element of (Acop)∗∗ = (A∗ op)∗ regarded as an element of A). Radford showed in [30] that D(A) is unimodular, and g ⊲⊳ α is the distinguished grouplike element of D(A)cop. Hence, by [32, Theorem 2], we have u ⇀ a = S(u) ⇀ (g ⊲⊳ α) ⇀ a for all a ∈ D(A)A. Using the formula of the fourth power of the antipode [31], we obtain u ⇀ a =Xhα, S−1(a(1))i S−2(ga(2)g−1) =X S2(a(1))(cid:10)α−1, a(2)(cid:11). Combining Example 4.5 and Lemma 4.9, we obtain the following proposition: Proposition 4.10. Let A be a finite-dimensional Hopf algebra over k. If A is involutory (i.e. the square of the antipode is the identity) and unimodular, then we have b - diml(D(A)A) = b - dimr(D(A)A) = Tr(ρ(b)) for all b ∈ Bn, where ρ : Bn −→ GL((D(A)A)⊗n) is the braid group action. 4.3 Examples We denote by σi ∈ Bn (i = 1, . . . , n − 1) the braid of n strands with only one positive crossing between the i-th and the (i + 1)-st strands. For integers p and q with p ≥ 2, the braid tp,q := (σ1σ2 · · · σp−1)q ∈ Bp is called the (p, q)-torus braid, as its closure is the (p, q)-torus link. The below is an example of the computation of the braided dimension associated with b = t2,q. Lemma 4.11. Let (A, R) be a quasitriangular Hopf algebra over k, and u be the Drinfel'd element of it. For each non-negative integer m and finite-dimensional left A-module X, ) if q = 2m, X ◦ TX,X(cid:1) ) X if q = 2m + 1, if q = 2m, ◦ TX,X ) if q = 2m + 1. t2,q - diml R t2,q - dimr R X = X = X )Tr(um−1(u−m)(2) P Tr(cid:0)um−1(u−m)(1) P Tr(cid:0)(cid:0)um−1 ⊗ um−1(cid:1)∆(u−m)R21 P Tr(um+1(u−m)(1) P Tr((um+1 ⊗ um+1)∆(u−m)R21X⊗X )Tr(um+1(u−m)(2) X⊗X X Here, for elements a, b ∈ A the notation a ⊗ bX⊗X stands for the left action on X ⊗ X defined by x ⊗ y 7−→ (a · x) ⊗ (b · y) for all x, y ∈ X. Proof. The formula for t2,q - diml R X can be obtained as follows. Let {es}d s=1 be a basis for X, and {e∗ s}d s=1 be its dual basis. Let R(q) be the element in A ⊗ A defined by R(q) =((R21R)m (R21R)mR21 if q = 2m, if q = 2m + 1. 26 By Example 4.5, we see that t2,q - diml R X =(Tr((u−1 ⊗ u−1)R(q) Tr((u−1 ⊗ u−1)R(q) ) X⊗X if q is even, ◦ TX,X ) if q is odd. X⊗X (4.13) Since R21R = ∆(u−1)(u ⊗ u) = (u ⊗ u)∆(u−1) [6], it follows that (R21R)m = (um ⊗ um)∆(u−m). Substituting this equation to (4.13) we obtain the formula for t2,q - diml X in the lemma. By a R similar consideration, the formula for t2,q - dimr R X can be obtained. In the case where A is semisimple, the braided dimension of the Schrodinger module associ- ated with t2,2 has the following representation-theoretic meaning: Theorem 4.12. Suppose that k is an algebraically closed field of characteristic zero. If A is a finite-dimensional semisimple Hopf algebra over k, then t2,2- diml(D(A)A) = t2,2- dimr(D(A)A) = dim(A) ♯Irr(A), where ♯Irr(A) is the number of isomorphism classes of irreducible A-modules. Proof. It is sufficient to show t2,2- diml(D(A)A) = dim(A) ♯Irr(A) in view of Proposition 4.10. By the assumption, Radford induction functor IA : AM −→ D(A) M is isomorphic to the functor D(A) ⊗A (−) by [14, Lemma 2.3]. Combining this fact with Proposition 2.9, we have (D(A)A) ⊗ (D(A)A) ∼= I A(Aad) ∼= D(A) ⊗A Aad. Hence, by Lemma 4.11, t2,2- diml(D(A)A) =X Tr((u−1)(1) )Tr((u−1)(2) D(A)A = Tr(u−1 (D(A)A)⊗(D(A)A)) = Tr(u−1 ) D(A)A D(A)⊗AAad ). (4.14) We use some results on the Frobenius-Schur indicator [20]. Let V be a finite-dimensional left A-module. The "third formula" [18, §6.4] of the n-th Frobenius-Schur indicator νn(V ) (n = 1, 2, . . .) expresses νn(V ) by using the Drinfel'd element, as νn(V ) = 1 dim(A) Tr(un D(A)⊗AV ). Since u is of finite order [7], dim(A) νn(V ) ∈ Z[ξ] (⊂ k), where ξ ∈ k is a root of unity of the same order as u. Hence, if we denote by z 7→ z the ring automorphism of Z[ξ] defined by ξ 7→ ξ−1, then we have Tr(u−n D(A)⊗AV ) = dim(A) νn(V ). On the other hand, the "first formula" [18, §2.3] yields ν1(V ) = dim(cid:0)HomA(k, V )(cid:1). Considering the case where V is the adjoint representation Aad, we obtain Tr(u−1 D(A)⊗AAad) = dim(A) ν1(Aad) = dim(cid:0)Hom(k, Aad)(cid:1) = dim(A) ♯Irr(A). Now the result follows from (4.14). 27 As this theorem suggests, the Schrodinger module D(A)A has much information about the category of A-modules, at least, in the semisimple case. However, the computation of the braided dimension is not easy in general. Fortunately, if A is a group algebra, then the braided dimension of D(A)A closely relates to the link group of the closure of the braid, and can be computed in the following way: Theorem 4.13. Let b ∈ Bn. If A = k[G] is the group algebra of a finite group G, then b - diml(D(A)A) = b - dimr(D(A)A) = ♯Hom(π1(R3 \bb), G) in k, wherebb is the link obtained by closing the braid b, and π1 means the fundamental group. Proof. Set X = D(A)A for simplicity. Then the braiding cX,X is given by cX,X(g ⊗ h) = h ⊗ (h−1 ◮ g) = h ⊗ h−1gh (g, h ∈ G). Let Bn act on Gn by (σi)(g1, . . . , gn) = (g1, . . . , gi−1, gi+1, g−1 i+1gigi+1, gi+2, . . . , gn) (g1, . . . , gn ∈ G). By Proposition 4.10, b - diml(D(A)A) and b - dimr(D(A)A) are equal to the number of fixed points of (b) regarded as an element of k. On the other hand, the number of fixed points of (b) has been studied by Freyd and Yetter [11] in relation with link invariants arising from crossed G-sets. The claim of this theorem follows from [11, Proposition 4.2.5]. Example 4.14. We consider the case where A = k[G] is the group algebra of a finite group G. If k is an algebraically closed field of characteristic zero, then we obtain t2,2 - diml(D(A)A) = t2,2 - dimr(D(A)A) = G · ♯Conj(G), t2,2 - diml(D(A∗)A∗) = t2,2 - dimr(D(A∗)A∗) = G2 (4.15) by Theorem 4.12, where Conj(G) is the set of conjugacy classes of G. In particular, t2,2 - diml(D(A)A) 6= t2,2 - diml(D(A∗)A∗) whenever G is non-abelian. This result is interesting from the viewpoint that some other monoidal Morita invariants, such as ones introduced in [7] and [37], are in fact invariants of the braided monoidal category of the representations of the Drinfel'd double. In topology, the link H =bt2,2 is known as the Hopf link. Since π1(R \ H) is the free abelian group of rank two, we have t2,2 - diml(D(A)A) = t2,2 - dimr(D(A)A) = ♯Comm(G) (4.16) by Theorem 4.13, where Comm(G) = {(x, y) ∈ G × G xy = yx}. Comparing (4.15) with (4.16), we get G · ♯Conj(G) = ♯Comm(G). Although this formula itself is well-known in finite 28 group theory, we expect that some non-trivial formulas for finite groups (or, more generally, for finite-dimensional semisimple Hopf algebras) would be obtained via the investigation of the braided dimension. By (4.12) and [25, Example 9.3.8], we have 1 - dimr(D(A)A) = Tr(S2 A) (see [2] for the quasi- Hopf case). In particular, 1 - dimr(D(A)A) = 0 whenever A is not cosemisimple. More strongly, we have the following theorem: Theorem 4.15. Let A be a finite-dimensional Hopf algebra. If A is not cosemisimple, then we have b - diml(D(A)A) = b - dimr(D(A)A) = 0 for all braids b. Proof. Let, in general, X be a finite-dimensional Hopf algebra, let Λ ∈ X be a left integral, and let λ ∈ X ∗ be the right integral such that hλ, Λi = 1. By [30, Proposition 2 (a)], Tr(cid:16)X −→ X; x 7→ S2(x(2))hp, x(1)i(cid:17) = hλ, 1ihp, Λi for all p ∈ X ∗. Applying this formula to X = Aop cop, we have Tr(cid:16)A −→ A; a 7→ S2(a(1))hp, a(2)i(cid:17) = 0 for all p ∈ A∗, since the Hopf algebra A is not cosemisimple and thus hλ, 1i = 0. (4.17) Now, let b ∈ Bn be a braid. By (4.8), b - diml(D(A)A) = Trl c(ef ), where Trl,1 c ◦ · · · ◦ Trl,1 c )(ρ(b)). n−1 } { z ef = ( EndD(A)(D(A)A) Let f : Aad −→ k be the A-linear map corresponding to ef under the isomorphism given by (2.10) and Proposition 2.6. Then we have ef (a) =Phf, a(2)i a(1) for all a ∈ D(A)A, and therefore b - diml(D(A)A) is equal to the trace of the linear map ∼=−−−−→ HomD(A)(D(A)A, IA(k)) ∼=−−−−→ HomA(Aad, k) A −→ A; a 7→ u ⇀ ef (a) =X S2(a(1))hα−1, a(2)ihf, a(3)i (a ∈ A). Hence, b - diml(D(A)A) = 0 by (4.17). The equation b - dimr(D(A)A) = 0 is proved in a similar way. By this theorem, we could say that the braided dimension of the Schrodinger module is not interesting as a monoidal Morita invariant for non-cosemisimple Hopf algebras. However, the endomorphism of D(A)A induced by a braid, such as ef in the above proof, is not generally zero, and thus may have some information about A. For example, let us consider the map zM := Trr,1 c (ρM (σ1)) : M −→ M (4.18) 29 for finite-dimensional M ∈ D(A) check that zM is given by the action of z := uS(u) on M . Hence, if M = D(A)A, then M, where ρM : B2 −→ GL(M ⊗2) is the action of B2. One can zM (a) = z • a = a(1)hα−1, a(2)i (a ∈ D(A)A) (4.19) by Lemma 4.9, where α ∈ A∗ is the distinguished grouplike element. Therefore this map has the following information: zM for M = D(A)A is the identity if and only if A is unimodular. Acknowledgements The authors would like to thank Professor Akira Masuoka for showing a direction to solve our problem and for continuous encouragement. For this research the first author (K.S.) is supported by Grant-in-Aid for JSPS Fellows (24·3606), and the second author (M.W.) is partially supported by Grant-in-Aid for Scientific Research (No. 22540058), JSPS. References [1] E. Abe, Hopf Algebras, Cambridge Univ. Press, Cambridge, 1980 (original Japanese version published by Iwanami Shoten, Tokyo, 1977). [2] D. Bulacu and B. Torrecillas, The representation-theoretic rank of double of quasi-Hopf algebras, J. Pure and Appl. Algebra 212 (2008), 919 -- 940. [3] S. Burciu, On some representations of the Drinfel'd double, J. Algebra 296 (2006), 480 -- 504. [4] R. H. Crowell and R. H. Fox, Introduction to knot theory, Ginn and Co., or: G.T.M. 57, Springer-Verlag, 1963. [5] V. G. Drinfel'd, Quantum groups, In: Proceedings of the International Congress of Mathe- matics, Berkeley, CA., 1987, 798 -- 820. [6] V. G. Drinfel'd, On almost cocommutative Hopf algebras, Leningrad Math. J. 1 (1990), 321 -- 342. [7] P. Etingof and S. Gelaki, On the exponent of finite-dimensional Hopf algebras, Math. Res. Lett. 6 (1999), 131 -- 140. [8] P. Etingof and S. Gelaki, Isocategorical groups, International Mathematical Research No- tices no. 2 (2001), 59 -- 76. [9] P. Etingof and S. Gelaki, On the quasi-exponent of finite-dimensional Hopf algebras, Math. Res. Lett. 9 (2002), 277 -- 287. [10] X. Fang, Quantum groups, q-Boson algebras and quantized Weyl algebras, International. J. Math. 22 (2011), 675 -- 694. 30 [11] P. Freyd and D. N. Yetter, Braided compact closed categories with applications to low di- mensional topology, Adv. in Math. 77 (1989), 156 -- 182. [12] P. Freyd and D. N. Yetter, Coherence theorems via knot theory, J Pure Appl. Algebra 78 (1992), 49 -- 76. [13] J. Hu and Y. Zhang, The β-character algebra and a commuting pair in Hopf algebras, Algebra Represent. Theory 10 (2007), 497 -- 516. [14] J. Hu and Y. Zhang, Induced Modules of Semisimple Hopf Algebras, Algebra Colloquium 4 (2007), 571 -- 584. [15] A. Joyal, R. Street, Braided tensor categories, Adv. in Math. 102 (1993) 20 -- 78. [16] A. Joyal, R. Street, Tortile Yang-Baxter operators in tensor categories, J. Pure Appl. Alge- bra 71 (1991), 43 -- 51. [17] Y. Kashina, S. Montgomery and S.-H. Ng, On the trace of the antipode and higher indica- tors, Israel J. Math. 188 (2012), 57 -- 89. [18] Y. Kashina, Y. Sommerhauser, and Y. Zhu, On higher Frobenius-Schur indicators, Mem. Amer. Math. Soc., 181, 2006. [19] C. Kassel, Quantum Groups, G.T.M. 155, Springer-Verlag, New York, 1995. [20] V. Linchenko and S. Montgomery, A Frobenius-Schur theorem for Hopf algebras, Algebr. Represent. Theory 3 (2000), 347 -- 355. [21] S. MacLane, Categories for the working mathematician, G.T.M. 5, Springer-Verlag, New York, 1971. [22] S. Majid, Physics for algebraists: Non-commutative and non-cocommutative Hopf algebras by a bicrossproduct construction, J. Algebra 130 (1990), 17 -- 64. [23] S. Majid, Double quasitriangular Hopf algebras, Commun. Algebra 19 (1991), 3061 -- 3073. [24] S. Majid, Representations, duals and quantum doubles of monoidal categories, Rend. Circ. Math. Palermo (2) Supp. 26 (1991), 197 -- 206. [25] S. Majid, Foundations of quantum group theory, Cambridge University Press, 1995. [26] A. Masuoka, Construction of quantized enveloping algebras by cocycle deformation, The Arabian Journal of Science and Engineering -- Theme Issues, Vol.33, No.2C (2008), 387 -- 406. [27] S. Montgomery, Hopf algebras and their action on rings, C.B.M.S. 82, American Mathe- matical Society, 1993. 31 [28] S.-H. Ng and P. Schauenburg, Higher Frobenius-Schur indicators for pivotal categories, Con- temp. Math. 441, Amer. Math. Soc., Providence, RI, 2007, 63 -- 90. [29] S.-H. Ng and P. Schauenburg, Frobenius-Schur indicators and exponents of spherical cate- gories, Adv. in Math. 211 (2007), 34 -- 71. [30] D. E. Radford, The Trace Function and Hopf Algebras, J. Algebra 163 (1994), 583 -- 622. [31] D. E. Radford, The order of the antipode of a finite dimensional Hopf algebra is finite, Amer. J. Math. 98 (1976), 333 -- 355. [32] D. E. Radford, On the antipode of a quasitriangular Hopf algebra, J. Algebra 151 (1992), 1 -- 11. [33] D. E. Radford, Minimal quasitriangular Hopf algebras, J. Algebra 157 (1993), 285 -- 315. [34] D. E. Radford, On oriented quantum algebras derived from representations of the quantum double of a finite-dimensional Hopf algebras, J. Algebra 270 (2003), 670 -- 695. [35] D. E. Radford and J. Towber, Yetter-Drinfel'd categories associated to an arbitrary bialge- bra, J. Pure Appl. Algebra 87 (1993), 259 -- 279. [36] N. Y. Reshetikhin and V. G. Turaev, Ribbon graphs and their invariants derived from quan- tum groups, Comm. Math. Phys. 127 (1990), 1 -- 26. [37] K. Shimizu, Monoidal Morita invariants for finite group algebras, J. Algebra 323 (2010), 397 -- 418. [38] M. E. Sweedler, Hopf algebras, Benjamin, New York, 1969. [39] M. Wakui, Polynomial invariants for a semisimple and cosemisimple Hopf algebra of finite dimension, J. Pure and Appl. Algebra 214 (2010), 701 -- 728. [40] D. N. Yetter, Quantum groups and representations of monoidal categories, Math. Proc. Cambridge Phil. Soc. 108 (1900), 261 -- 290. Graduate School of Mathematics, Nagoya University, Furo-cho, Chikusa-ku, Nagoya 464-8602, Japan E-mail address: [email protected] Department of Mathematics, Faculty of Engineering Science, Kansai University, Suita-shi, Osaka 564-8680, Japan E-mail address: [email protected] 32
1310.6648	1	1310	2013-10-24T15:44:58	Leavitt path algebras of Cayley graphs arising from cyclic groups	[ "math.RA" ]	For any positive integer $n$ we describe the Leavitt path algebra of the Cayley graph $C_n$ corresponding to the cyclic group $\Z/n\Z$. Using a Kirchberg-Phillips-type realization result, we show that there are exactly four isomorphism classes of such Leavitt path algebras, arising as the algebras corresponding to the graphs $C_i$ ($3\leq i \leq 6$).	math.RA	math	Leavitt path algebras of Cayley graphs arising from cyclic groups Gene Abrams and Benjamin Schoonmaker Abstract. For any positive integer n we describe the Leavitt path algebra of the Cayley graph Cn corresponding to the cyclic group Z/nZ. Using a Kirchberg-Phillips-type realization result, we show that there are exactly four isomorphism classes of such Leavitt path algebras, arising as the algebras cor- responding to the graphs Ci (3 ≤ i ≤ 6). For each finite group H the Cayley graph CH of H is a directed graph which en- codes information about the relationships between elements of H and a set of gener- ators of H. In the particular case where Hn = Z/nZ (and n ≥ 3), the Cayley graph CHn (which we denote simply by Cn) consists of n vertices {v1, v2, . . . , vn} and 2n edges {e1, e2, . . . , en, f1, f2, . . . , fn} for which s(ei) = vi, r(ei) = vi+1, s(fi) = vi, r(fi) = vi−1, where indices are interpreted mod n, and where s(e) (resp., r(e)) denotes the source (resp., range) vertex of the edge e. (More precisely, the graph Cn described here is the Cayley graph for the group Z/nZ with respect to the subset {1, n − 1}.) So, for instance, C3 is the graph C3 = •v1 =③③③③③③③③③ !❉❉❉❉❉❉❉❉❉ •v2 •v3 For the cases n = 1 and n = 2, we define the graphs C1 and C2 in a manner consistent with the above description (i.e., as a graph with n vertices and 2n edges 1991 Mathematics Subject Classification. Primary 16S99 Secondary 05C25. Key words and phrases. Leavitt path algebra, Cayley graph. The first author is partially supported by a Simons Foundation Collaboration Grants for Mathematicians Award #208941. Part of this work formed the basis of the second author's Master of Science presentation at the University of Colorado Colorado Springs, April 2013. The authors are grateful to Attila Egri-Nagy, who brought to the attention of the first author the potential connection between Leavitt path algebras and Cayley graphs during the conference "Graph C*- algebras, Leavitt path algebras and symbolic dynamics", held at University of Western Sydney, February 2013. 1 ! = 8 8 o o k k ACn =   0 1 0 0 1 1 0 0 1 1 0 ... 0 0 · · · · · · . . . · · · 0 1 0 0 0 0 ... 0 1 1 0   . 2 GENE ABRAMS AND BENJAMIN SCHOONMAKER with appropriate source and range relations), as follows: C1 = •v1 C2 = •v1 •v2 Less formally, Cn is the graph with n vertices and 2n edges, where each vertex emits two edges, one to both of its neighboring vertices. We denote by ACn the incidence matrix of Cn, which for n ≥ 3 is easily seen to be the matrix For the cases n = 1 and 2 we have AC1 = (2), and AC2 =(cid:18)0 2 2 0(cid:19). For any field K and directed graph E the Leavitt path algebra LK(E) has been the focus of sustained investigation since 2004. We give here a basic description of LK(E); for additional information, see e.g. [AAP1] or [AAS]. Definition of Leavitt path algebra. Let K be a field, and let E = (E0, E1, r, s) be a directed graph with vertex set E0 and edge set E1. The Leavitt path K-algebra LK(E) of E with coefficients in K is the K-algebra generated by a set {v v ∈ E0}, together with a set of variables {e, e∗ e ∈ E1}, which satisfy the following relations: vw = δv,wv for all v, w ∈ E0, s(e)e = er(e) = e for all e ∈ E1, r(e)e∗ = e∗s(e) = e∗ for all e ∈ E1, (V) (E1) (E2) (CK1) e∗e′ = δe,e′ r(e) for all e, e′ ∈ E1, An alternate description of LK(E) may be given as follows. For any graph (CK2) v =P{e∈E1s(e)=v} ee∗ for every v ∈ E0 for which 0 < s−1(v) < ∞. E let bE denote the "double graph" of E, gotten by adding to E an edge e∗ for each edge e ∈ E1. Then LK(E) is the usual path algebra KE∗, modulo the ideal generated by the relations (CK1) and (CK2). ✷ It is easy to show that LK(E) is unital if and only if E0 is finite. This is of course the case when E = Cn. Let E be a directed graph with vertices v1, v2, . . . , vn and adjacency matrix AE = (ai,j). We let Fn denote the free abelian monoid on the generators v1, v2, . . . , vn (so Fn ∼= ⊕n Z+ as monoids). We denote the identity element of this monoid by z. We let Rn(E) denote the submonoid of Fn generated by the relations i=1 vi = nXj=1 ai,jvj for each non-sink vi. The graph monoid ME of E is defined as the quotient monoid The elements of ME are typically denoted using brackets. ME = Fn/Rn(E). X X $ $ b b W W LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS ARISING FROM CYCLIC GROUPS 3 As a representative example, we explicitly describe the graph monoid MC3 associated to the Cayley graph C3. This is the free abelian monoid on the generators v1, v2, v3, modulo the submonoid generated by the relations v1 = v2 + v3, v2 = v1 + v3, and v3 = v1+v2. Note that, for instance, v1+(v1+v2+v3) = (v1+v2)+(v1+v3) = v3 + v2 = v1, so that [v1] = [v1] + [v1 + v2 + v3] in MC3. Let x denote the element [v1 + v2 + v3] of MC3. Then a similar computation yields that [v2] = [v2] + x and [v3] = [v3] + x in MC3. Moreover, [v1] + [v1] = [v1] + [v2 + v3] = x, and in a similar fashion we also have 2[v2] = 2[v3] = x in MC3. Thus we see that MC3 = {[z], [v1], [v2], [v3], [v1] + [v2] + [v3]}. (We have not justified why these five elements are distinct in MC3, but this can be done easily; see e.g. [AS, page 171].) More generally, for any n ≥ 1, the monoid MCn is generated by [v1], [v2], . . . , [vn], subject to the relations [vi] = [vi−1] + [vi+1] (for all 1 ≤ i ≤ n), where subscripts are interpreted mod n. (This description also covers the cases n = 1 and n = 2.) We present now the background information required to achieve our main result (Theorem 8), which yields a description of the Leavitt path algebras corresponding to the Cayley graphs {Cn n ≥ 1}. The cornerstone of the result is a utilization of the Algebraic Kirchberg Phillips Theorem. To motivate and explain how this theorem is used, in the following three paragraphs we make a streamlined visit to three elegant, fundamental results in the theory of Leavitt path algebras and purely infinite simple algebras. For a unital K-algebra A, the set of isomorphism classes of finitely generated projective left A-modules is denoted by V(A). We denote the elements of V(A) using brackets; for example, [A] ∈ V(A) represents the isomorphism class of the left regular module AA. V(A) is a monoid, with operation ⊕, and zero element [{0}]. The monoid (V(A), ⊕) is conical; this means that the sum of any two nonzero elements of V(A) is nonzero, or, rephrased, that V(A)∗ = V(A) \ {0} is a semigroup under ⊕. A striking property of Leavitt path algebras was established in [AMP, Theorem 3.5], to wit: (∗) Moreover, V(LK(E)) ∼= ME as monoids. [LK(E)] ↔ Xv∈E0 [v] under this isomorphism. A unital K-algebra A is called purely infinite simple in case A is not a division ring, and A has the property that for every nonzero element x of A there exists b, c ∈ A for which bxc = 1A. It is shown in [AGP, Corollary 2.2] that if A is a unital purely infinite simple K-algebra, then the semigroup (V(A)∗, ⊕) is in fact a group, and, moreover, that V(A)∗ ∼= K0(A), the Grothendieck group of A. (Indeed, if V(LK(E))∗ is a for unital Leavitt path algebras, the converse is true as well: group, then LK(E) is purely infinite simple. This converse is not true for general K-algebras.) Summarizing: when LK(E) is unital purely infinite simple we have the following isomorphism of groups: (∗∗) K0(LK(E)) ∼= V(LK(E))∗ ∼= M ∗ E. 4 GENE ABRAMS AND BENJAMIN SCHOONMAKER The finite graphs E for which the Leavitt path algebra LK(E) is purely infinite simple have been explicitly described in [AAP2], to wit: (∗ ∗ ∗) LK(E) is purely infinite simple ⇐⇒ E is cofinal, sink-free, and satisfies Condition (L). Somewhat more fully, these are the graphs E for which: every vertex in E connects (via some directed path) to every cycle of E; every cycle in E has an exit (i.e., in each cycle of E there is a vertex which emits at least two edges); and E contains at least one cycle. (The structure of the field K plays no role in determining the purely infinite simplicity of LK(E).) We now have the necessary background information in hand which allows us to present the powerful tool which will yield our main result. The proof of the following theorem utilizes deep results and ideas in the theory of symbolic dynamics. The letters K and P in its name derive from E. Kirchberg and N.C. Phillips, who (independently in 2000) proved an analogous result for graph C∗-algebras. The Algebraic KP Theorem. [ALPS, Corollary 2.7] Suppose E and F are finite graphs for which the Leavitt path algebras LK(E) and LK(F ) are purely in- finite simple. Suppose that there is an isomorphism ϕ : K0(LK(E)) → K0(LK(F )) for which ϕ([LK(E)]) = [LK(F )], and suppose also that the two integers det(IE0 − At and det (IF 0 − At nonpositive). Then LK(E) ∼= LK(F ) as K-algebras. F ) have the same sign (i.e., are either both nonnegative, or both E) We note that, as of Fall 2013, it is not known whether the hypothesis regarding the germane determinants can be eliminated from the statement of The Algebraic KP Theorem. On the other hand, is the case that the determinant hypothesis can be eliminated in the analogous graph C∗-algebra result established by Kirchberg and Phillips. Thus [ALPS, Corollary 2.7] is sometimes called the 'Restricted' Algebraic Kirchberg Phillips Theorem. Our goal for the remainder of this short note is to analyze the data required to invoke The Algebraic KP Theorem in the context of the collection of algebras {LK(Cn) n ∈ N}. We start by noting that displays (∗ ∗ ∗) and (∗∗) immediately give Proposition 1. For each n ≥ 1 the K-algebra LK(Cn) is unital purely infinite simple. In particular, M ∗ Cn = (MCn \ {[z]}, +) is a group. Referring to the explicit description of MC3 given above, it is easy to see that M ∗ C3 = {[v1], [v2], [v3], [v1] + [v2] + [v3]} ∼= Z/2Z × Z/2Z. Using the previous computations, we see that x = [v1] + [v2] + [v3] is the identity element of the group M ∗ Indeed, we will see below that for any n ∈ N, the . (However, for an arbitrary graph elementPn E, the analogous element Pv∈E0 [v] need not be the identity of M ∗ i=1[vi] is the identity element in M ∗ Cn C3. Let E be a finite directed graph for which E0 = n, and let AE denote the usual incidence matrix of E. Let BE denote the matrix In − At E. We view BE both as a matrix, and as a linear transformation BE : Zn → Zn, via left multiplication (viewing elements of Zn as column vectors). In the situation where LK(E) is E.) LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS ARISING FROM CYCLIC GROUPS 5 purely infinite simple, so that in particular M ∗ to K0(LK(E))), we have that E is a group (necessarily isomorphic K0(LK(E)) ∼= M ∗ E ∼= Zn/Im(BE) = Coker(BE). (See [AALP, Section 3] for a complete discussion.) Under this isomorphism we have [vi] 7→ ~bi + Im(BE), where ~bi is the element of Zn which is 1 in the ith coordinate and 0 elsewhere. For any n×n matrix T ∈ Mn(Z), we may view T as a linear transformation from Zn to Zn in the usual way. Then the finitely generated abelian group Zn/Im(T ) may be described by analyzing the Smith normal form of T (see e.g. [S]). Specifically, if the Smith normal form of T is the diagonal matrix diag(α1, α2, ..., αn), then Zn/Im(T ) ∼= Z/α1Z⊕ Z/α2Z⊕· · ·⊕ Z/αnZ, where Z/1Z is interpreted as the trivial group {0}. (Most computer software packages, e.g. Mathematica and Scientific Notebook, contain a built-in Smith normal form function.) Using this method, where we let T be the matrix BE = In − At E, we present here a description of the groups M ∗ C12. (Of course the description of some of these groups can be achieved using a more straightforward approach than the utilization of Smith normal form, as was done above for the group M ∗ C1 through M ∗ C3.) ∼= Z/3Z, M ∗ C5 ∼= Z/3Z, M ∗ C11 M ∗ C1 M ∗ C7 ∼= {0}, M ∗ ∼= Z×Z, C2 ∼= {0}, M ∗ ∼= Z×Z. C8 The displayed isomorphisms suggest a pattern, first noticed by A. Egri-Nagy ∼= Z/2Z×Z/2Z, M ∗ C4 ∼= Z/2Z×Z/2Z, M ∗ ∼= Z/3Z, M ∗ C3 ∼= Z/3Z, M ∗ C9 ∼= {0}, M ∗ C6 ∼= {0}, M ∗ C10 C12 and shared with the first author in a private communication. Although for an arbitrary graph E there is no appropriate notion of "the inverse of an element" in the semigroup M ∗ E (since this need not be a group), we use the standard minus sign notation to denote inverses in M ∗ E whenever this semigroup is actually a group. In particular, it is appropriate to consider an element of the form −[x] in M ∗ Cn for any [x] ∈ M ∗ Cn . Lemma 2. For each [vi] in the group M ∗ Cn we have [vi] = −[vi+3]. Conse- . quently, [vi] = [vi+6] for all 1 ≤ i ≤ n in M ∗ Cn In M ∗ Proof. Substituting yields [vi+1] = [vi] + [vi+1] + [vi+3]. Since M ∗ Cn cancel [vi+1], and the result follows immediately. Cn we have [vi+1] = [vi] + [vi+2] and [vi+2] = [vi+1] + [vi+3]. is a group, we can ✷ Proposition 3. For n, m ≥ 1, if n ≡ m mod 6, then M ∗ Cn ∼= M ∗ Cm . Proof. By the previously displayed isomorphisms, it suffices to show that if n ≥ 6 and n ≡ N mod 6 with 6 ≤ N ≤ 11, then M ∗ . (We choose N ≥ 6 in Cn order to avoid some notational issues involving the interpretation of integers mod6.) For any graph E having E0 = t we denote by πt the canonical homomorphism of monoids Ft → Ft/Rt(E). We note that since none of the generating relations i=1 mivi = 0 for mi ∈ N, πt restricts to a which produce Rt(E) are of the form Pn t → (Ft/Rt(E))∗. homomorphism π∗ ∼= M ∗ CN t : F ∗ Define the semigroup homomorphism ϕ : F ∗ N by setting ϕ(vi) = vimod6 n → M ∗ for each generator vi (1 ≤ i ≤ n) of F ∗ CN be the composition π∗ N ◦ ϕ, so that, in particular, ψ(vi) = [vimod6] for 1 ≤ i ≤ n. To show that ψ factors to a homomorphism from M ∗ , we need only Cn show that ψ takes each of the relations vi = vi−1 + vi+1 (1 ≤ i ≤ n, interpreted modn) in Rn(E) to a relation valid in M ∗ CN ; in other words, it suffices to show that n , and extending linearly. Let ψ : F ∗ n → F ∗ to M ∗ CN 6 GENE ABRAMS AND BENJAMIN SCHOONMAKER ψ(vi) = ψ(vi−1) + ψ(vi+1) in M ∗ We consider five cases. The point here is that CN we must understand the given relations with two things in mind: the interpretation of the integers kmod6 which arise in the definition of ϕ as integers between 1 and 6 (inclusive), as well as the interpretation of the subscripts in the expressions in M ∗ CN as integers modN . Case 1a: 1 ≤ i − 1 and i + 1 ≤ n and i − 1 ≡ 1, 2, 3, or 4, mod 6. Then ψ(vi) = [vimod6] = [vi−1mod6] + [vi+1mod6] = ψ(vi−1) + ψ(vi+1). Case 1b: 1 ≤ i − 1 and i + 1 ≤ n and i − 1 ≡ 5mod6; so i ≡ 6mod6 and i + 1 ≡ 1mod6. Then using Lemma 2 we have ψ(vi) = [vimod6] = [v6] = [v5] + [v7] = [v5] + [v1] = [vi−1mod6] + [vi+1mod6] = ψ(vi−1) + ψ(vi+1). Case 1c: 1 ≤ i − 1 and i + 1 ≤ n and i − 1 ≡ 6mod6; so i ≡ 1mod6 and i + 1 ≡ 2mod6. Then using Lemma 2 we have ψ(vi) = [vimod6] = [v1] = [v7] = [v6] + [v8] = [v6] + [v2] = [vi−1mod6] + [vi+1mod6] = ψ(vi−1) + ψ(vi+1). Case 2: i = 1. So vi−1 = vn in Fn by definition. Then using that N mod6 ≡ nmod6 and Lemma 2, we get ψ(v1) = [v1] = [vN ] + [v2] = [vN mod6] + [v2] = [vnmod6] + [v2] = [vi−1mod6] + [vi+1mod6] = ψ(vi−1) + ψ(vi+1). Case 3: i = n. So vi+1 = v1 in Fn by definition. Then using that N mod6 ≡ nmod6 and Lemma 2, we get ψ(vn) = [vnmod6] = [vN mod6] = [vN ] = [vN −1] + [v1] = [vN −1mod6] + [v1] = [vn−1mod6] + [v1] = [vi−1mod6] + [vi+1mod6] = ψ(vi−1) + ψ(vi+1). Thus ψ preserves the relations which generate Rn(E), and so ψ extends to a group homomorphism ψ : M ∗ Cn → M ∗ CN . In a completely analogous manner, for n ≡ N mod6 with 6 ≤ N ≤ 11 there for which τ ([vi]) = [vimod6]. It is exists a group homomorphism τ : M ∗ then clear that τ and ψ are inverses, thus establishing the result for n, m ≥ 6. → M ∗ Cn CN The observation made prior to the Proposition shows that the cases n = ✷ 1, 2, 3, 4, 5 satisfy the statement as well. As a consequence of Proposition 3, and using the previously displayed compu- tations, we see that there are, up to isomorphism, only four groups represented by the collection {M ∗ Cn n ∈ N}, as follows. Corollary 4. The following is a complete description of the isomorphism classes of the groups M ∗ Cn for n ∈ N. (1) M ∗ Cn (2) M ∗ Cn (3) M ∗ Cn (4) M ∗ Cn ∼= {0} in case n ≡ 1mod6 or n ≡ 5mod6. ∼= Z/3Z in case n ≡ 2mod6 or n ≡ 4mod6. ∼= Z/2Z × Z/2Z in case n ≡ 3mod6. ∼= Z × Z in case n ≡ 6mod6. LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS ARISING FROM CYCLIC GROUPS 7 With the first piece of the analysis now in place, we turn our attention to describing the element [LK(Cn)] of the group K0(LK(Cn)); by (∗), this amounts i=1[vi] in the group M ∗ Cn . to describing the element Pn Lemma 5. For each n ∈ N, Pn Proof: Let x denote the elementPn Rn(Cn), we have i=1[vi] is the identity element of the group M ∗ Cn . i=1[vi] of M ∗ Cn . Using the defining relations x = nXi=1 [vi] = nXi=1 ([vi−1] + [vi+1]) = nXi=1 [vi−1] + nXi=1 [vi+1] = x + x, as we interpret the indices of the generating elements modulo n in M ∗ . But the Cn equation x + x = x in a group yields immediately that x is the identity element. ✷ Suppose that m, n are integers for which M ∗ Cn trivially such an isomorphism must send the elementPn Pm (see Corollary 4). Then to the element Cm, as Lemma 5 shows that each of these expressions is the identity element in the respective group. Cm i=1[vi] of M ∗ Cn i=1[vi] of M ∗ ∼= M ∗ The final piece of the hypotheses in the Algebraic KP Theorem involves deter- minants of appropriate matrices, which we analyze in the next result. Proposition 6. For each n ∈ N, det(In − At Cn ) ≤ 0. Proof: An n × n matrix B = (bi,j) is circulant in case we have bi+1,j+1 = bi,j for 1 ≤ i ≤ n − 1, 1 ≤ j ≤ n − 1; b1,j+1 = bn,j for j + 1 ≤ n; and bi+1,1 = bi,n for i + 1 ≤ n. Less formally, B is circulant in case each subsequent row of B is obtained from the previous row by moving each entry of the previous row one place to the right, and moving the last entry of the previous row to the first position of the subsequent row. (The last row gets moved to the first row in this way as well.) If B is circulant, then there is a formula (derived from an analysis of the eigenvectors of B) which expresses det(B) as the following product: det(B) = n−1Yj=0 (b1 + b2ωj + b3ω2 j + · · · + bnωn−1 j ) where (b1 b2 b3 · · · bn) is the first row of B, and ωj = e is an nth root of unity in C. (See e.g. [KS] for a description of some of the many places in which circulant matrices arise.) n 2πij In the case of the Cayley graph Cn (for n ≥ 3), the matrix B = In − At has b1 = 1, b2 = bn = −1, and bi = 0 for i = 3, 4, . . . n−1. Using that eiθ = cos θ+i sin θ together with the displayed equation, we get: Cn (1 − ωj − ωn−1 j ) = n−1Yj=0 (1 − e 2πij n − e 2πij(n−1) n ) (1 − cos 2πj n − i sin 2πj n − cos 2πj(n − 1) n − i sin 2πj(n − 1) n ) det(In − At Cn) = = = n−1Yj=0 n−1Yj=0 n−1Yj=0 (1 − 2 cos 2π n j), 8 GENE ABRAMS AND BENJAMIN SCHOONMAKER with the final equality coming as a direct result of the basic trigonometry facts that, for any integer j, cos 2πj(n−1) = cos 2πj n and sin 2πj(n−1) = − sin 2πj n . n n When j = 0 we have 1 − 2 cos( 2π 2 we have 1 − 2 cos( 2π even, when j = n 1−2 cos 2π This yields the result. n j = 1−2 cos 2π n j) = 1 − 2 · cos 0 = −1 < 0. In case n is n j) = 1 − 2 cos π = 3 > 0. Furthermore, since n (n−j)) ≥ 0. ✷ n j)(1−2 cos 2π n (n−j), we see that (1−2 cos 2π We note as a consequence of the previous analysis that det(In − At ) = 0 precisely when one of the factors 1 − 2 cos 2π n j (0 ≤ j ≤ n − 1) equals 0. This can easily be shown to happen precisely when n is a multiple of 6. This information is consistent with the observation that the only values of n for which the group M ∗ Cn is infinite are multiples of 6. Cn Of the four groups which arise up to isomorphism as a group of the form M ∗ Cn (for n ≥ 1), we see that two are cyclic: {0} and Z/3Z. Purely infinite simple unital Leavitt path algebras LK(E) whose corresponding K0 groups are cyclic and for which det(IE0 − At E) ≤ 0 are relatively well-understood, and arise from the classical Leavitt algebras LK(1, n), as follows. For any integer n ≥ 2, LK(1, n) is the free associative K-algebra in 2n generators x1, x2, ..., xn, y1, y2, ..., yn, subject to the relations yixj = δi,j1K and xiyi = 1K. nXi=1 These algebras were first defined and investigated in [L], and formed the motivating examples for the more general notion of Leavitt path algebra. It is easy to see that for n ≥ 2, if Rn is the graph having one vertex and n loops (the "rose with n petals" graph), then LK(Rn) ∼= LK(1, n). It is clear from (∗ ∗ ∗) that each LK(Rn) is purely infinite simple; it is straightforward from (∗∗) that K0(LK(Rn)) ∼= M ∗ Rn is the cyclic group Z/(n − 1)Z of order n − 1, where the regular module [LK(Rn)] in K0(LK(Rn)) corresponds to 1 in Z/(n − 1)Z. Now let d ≥ 2, and consider the graph Rd n having two vertices v1, v2; d − 1 edges from v1 to v2; and n loops at v2: Rd n = •v1 (d−1) / •v2 (n) n). By standard Morita equivalence theory, K0(Md(LK(1, n))) ∼= K0(LK(1, n)). It is shown in [AALP] that the matrix algebra Md(LK(1, n)) is isomorphic to LK(Rd Moreover, the element [Md(LK(1, n))] of K0(Md(LK(1, n))) corresponds to the ele- ment d in Z/(n−1)Z. In particular, the element [Mn−1(LK(1, n))] of K0(Mn−1(LK(1, n))) corresponds to n − 1 ≡ 0 in Z/(n − 1)Z. Finally, an easy computation yields that det(I2 − At ) = −(n − 1) < 0 for all n, d. Therefore, by invoking the Algebraic KP Theorem, the previous discussion immediately yields the following. Rd n simple. Suppose that M ∗ Proposition 7. Suppose E is a graph for which LK(E) is unital purely infinite E is isomorphic to the cyclic group Z/(n − 1)Z, via an iso- E to the element d of Z/(n − 1)Z. morphism which takes the elementPv∈E0 [v] of M ∗ E) is negative. Then LK(E) ∼= Md(LK(1, n)). Finally, suppose that det(IE0 − At We now have all the ingredients in place to achieve our main result. Theorem 8. For each n ≥ 1 let Cn denote the Cayley graph corresponding to the cyclic group Z/nZ (with respect to the subset {1, n − 1}) as described previously. / f f LEAVITT PATH ALGEBRAS OF CAYLEY GRAPHS ARISING FROM CYCLIC GROUPS 9 Then up to isomorphism the collection of Leavitt path algebras {LK(Cn) n ∈ N} is completely described by the following four pairwise non-isomorphic classes of K-algebras. (1) LK(Cn) ∼= LK(Cm) in case m ≡ 1 or 5 mod6 and n ≡ 1 or 5 mod6. In this case, these algebras are isomorphic to LK(1, 2). (2) LK(Cn) ∼= LK(Cm) in case m ≡ 2 or 4 mod6 and n ≡ 2 or 4 mod6. In this case, these algebras are isomorphic to M3(LK(1, 4)). (3) LK(Cn) ∼= LK(Cm) in case m, n ≡ 3 mod6. (4) LK(Cn) ∼= LK(Cm) in case m, n ≡ 6 mod6. Proof: We seek to invoke the Algebraic KP Theorem. By Proposition 1, LK(Cn) is purely infinite simple for each n ∈ N. For any of the four indicated cases, we choose a pair of integers m, n. By Corollary 4 and (∗∗) we have K0(LK(Cn)) ∼= K0(LK(Cm)), and, by the observation made subsequent to Lemma 5 together with (∗), this isomorphism necessarily takes [LK(Cn)] to [LK(Cm)]. By Proposition 6, det(In −At ) are both nonpositive. The Algebraic KP Theorem now gives the bulk of the result. The two extra statements in parts (i) and (ii) follow directly from statements (i) and (ii) of Corollary 4 together with Proposition 7. ✷ ) and det(Im−At Cn Cm In [AAP3] a description is given of the Leavitt path algebras associated to additional collections of Cayley-type graphs. . References [AALP] G. Abrams, P.N. ´Anh, A. Louly, E. Pardo, The classification question for Leavitt path algebras, J. Algebra 320(5) (2008), 1983 -- 2026. [AAS] G. Abrams, P. Ara, M. Siles Molina, Leavitt path algebras: a primer and handbook, Lecture Notes Series in Mathematics, Springer Verlag, to appear. [AAP1] G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293(2) (2005), 319 -- 334. [AAP2] G. Abrams, G. Aranda Pino, Purely infinite simple Leavitt path algebras, J. Pure App. Alg. 207(3) (2006), 553 -- 563. [AAP3] G. Abrams, G. Aranda Pino, Leavitt path algebras for generalized Cayley graphs of cyclic groups, in preparation. [ALPS] G. Abrams, A. Louly, E. Pardo, C. Smith, Flow invariants in the classification of Leavitt path algebras, J. Algebra 333 (2011), 202 -- 231. [AS] G. Abrams, J. Sklar, The graph menagerie: Abstract algebra and the Mad Veterinarian, Math. Magazine 83(3) (2010), 168 -- 179. [AGP] P. Ara, K. Goodearl, E. Pardo, K0 of purely infinite simple regular rings, K-Theory 26 (2002), 69 -- 100. [AMP] P. Ara, M.A. Moreno, E. Pardo, Non-stable K-theory for graph algebras, Algebr. Repre- sent. Theor. 10(2) (2007), 157 -- 178. [KS] I. Kra, S. Simanca, On circulant matrices, Notices Amer. Math. Soc. 59(3) (2012), 368 -- 377. [L] W. Leavitt, The module type of a ring, Trans. Amer. Math. Soc. 103(1) (1962), 113 -- 130. [S] C. Sims, Computation with finitely presented groups, Encyclopedia of Mathematics and its Applications, Cambridge University Press, 1994. Department of Mathematics, University of Colorado, Colorado Springs, CO 80918 U.S.A. E-mail address: [email protected] Department of Mathematics, Brigham Young University, Provo, UT 84602 U.S.A. E-mail address: [email protected]
1701.03750	1	1701	2017-01-13T17:43:41	On Lie algebras associated with modules over polynomial rings	[ "math.RA" ]	Let $\mathbb K$ be an algebraically closed field of characteristic zero. Let $V$ be a module over the polynomial ring $\mathbb K[x,y]$. The actions of $x$ and $y$ determine linear operators $P$ and $Q$ on $V$ as a vector space over $\mathbb K$. Define the Lie algebra $L_V=\mathbb K\langle P,Q\rangle \rightthreetimes V$ as the semidirect product of two abelian Lie algebras with the natural action of $\mathbb K\langle P,Q\rangle$ on $V$. We show that if $\mathbb K[x,y]$-modules $V$ and $W$ are isomorphic or weakly isomorphic, then the corresponding associated Lie algebras $L_V$ and $L_W$ are isomorphic. The converse is not true: we construct two $\mathbb K[x,y]$-modules $V$ and $W$ of dimension $4$ that are not weakly isomorphic but their associated Lie algebras are isomorphic. We characterize such pairs of $\mathbb K[x, y]$-modules of arbitrary dimension. We prove that indecomposable modules $V$ and $W$ with $\dim V=\dim W\geq 7$ are weakly isomorphic if and only if their associated Lie algebras $L_V$ and $L_W$ are isomorphic.	math.RA	math	ON LIE ALGEBRAS ASSOCIATED WITH MODULES OVER POLYNOMIAL RINGS A.P. PETRAVCHUK, K.YA. SYSAK Abstract. Let K be an algebraically closed field of characteristic zero. Let V be a mod- ule over the polynomial ring Krx, ys. The actions of x and y determine linear operators P and Q on V as a vector space over K. Define the Lie algebra LV " KxP, Qy%V as the semidirect product of two abelian Lie algebras with the natural action of KxP, Qy on V . We show that if Krx, ys-modules V and W are isomorphic or weakly isomorphic, then the corresponding associated Lie algebras LV and LW are isomorphic. The converse is not true: we construct two Krx, ys-modules V and W of dimension 4 that are not weakly isomorphic but their associated Lie algebras are isomorphic. We characterize such pairs of Krx, ys-modules of arbitrary dimension. We prove that indecomposable modules V and W with dim V " dim W ě 7 are weakly isomorphic if and only if their associated Lie algebras LV and LW are isomorphic. 7 1 0 2 n a J 3 1 ] . A R h t a m [ 1 v 0 5 7 3 0 . 1 0 7 1 : v i X r a 1. Introduction Let K be an algebraically closed field of characteristic zero. Let V be a module over the polynomial ring Krx, ys. Define the commuting linear operators P and Q on V by setting P pvq " x v and Qpvq " y v for all v P V . Conversely, if P and Q are commuting linear operators on V , then the vector space V can be considered as the Krx, ys-module with multiplication f px, yq v " f pP, Qqpvq for all v P V and f px, yq P Krx, ys. For each Krx, ys-module V , the metabelian Lie algebra LV " KxP, Qy%V , which is the external semidirect product of the abelian Lie algebra KxP, Qy of dimension 2 and the abelian Lie algebra V with the natural action of KxP, Qy on V . We say that the Lie algebra LV is associated with the Krx, ys-module V . construct let us Modules over polynomial rings were studied by many authors: Gelfand and Ponomarev [3] proved that the problem of classifying finite dimensional modules over Krx, ys contains the problem of classifying matrix pairs up to similarity. Quillen [4] and Suslin [5] studied projective modules over polynomial rings in connection with Serre's problem. Our goal is to study relations between finite dimensional Krx, ys-modules V and the corresponding associated Lie algebras LV . For each automorphism θ of the abelian Lie algebra KxP, Qy with θpP q " α11P ` α12Q, θpQq " α21P ` α22Q, αij P K, Date: July 8, 2021. 2000 Mathematics Subject Classification. Primary 17B30, 17B60; Secondary 13C13. Key words and phrases. Lie algebra, polynomial ring, metabelian algebra, module, weak isomorphism, linear operator. 1 2 A.P. PETRAVCHUK, K.YA. SYSAK the semidirect product KxθpP q, θpQqy%V is isomorphic to LV . The corresponding trans- formation of the ring Krx, ys is the automorphism of Krx, ys defined by θpxq " α11x`α12y and θpyq " α21x ` α22y. This automorphism defines the "twisted" module Vθ with the fol- lowing multiplication on the vector space V : x v " θpxq v and y v " θpyq v for all v P V. The modules V and Vθ are called weakly isomorphic; this notion in matrix form was studied by Belitskii, Lipyanski, and Sergeichuk [2]. We show in Proposition 1 that if Krx, ys-modules W and V are isomorphic (or even weakly isomorphic), then the associated Lie algebras LV and LW are isomorphic. The converse statement is not true: Lemma 3 gives an example of Krx, ys-modules V and W that are not weakly isomorphic, but their Lie algebras LV and LW are isomorphic. Nevertheless, if V and W are indecomposable modules with dimK V " dimK W ě 7, then by Theorem 2 V and W are weakly isomorphic if and only if the associated Lie algebras LV and LW are isomorphic. Theorem 1 gives a characterization of pairs pV, W q of Krx, ys-modules that are not weakly isomorphic but their associated Lie algebras are isomorphic. It shows that the problem of classifying finite dimensional Lie algebras of the form L " B%A with an abelian ideal A and a two-dimensional abelian subalgebra B is equivalent to the problem of classifying finite dimensional Krx, ys-modules up to weak isomorphism. We think that the latter problem is wild. The wildness of some classes of metabelian Lie algebras was established by Belitskii, Bondarenko, Lipyanski, Plachotnik, and Sergeichuk [1, 2]. From now on, K is an algebraically closed field of characteristic zero. All Lie algebras and modules over Krx, ys that we consider are finite dimensional over K. Let L1 and L2 be Lie algebras over a field K. Let ϕ : L1 Ñ DerL2 be a homomorphism of Lie algebras, in which DerL2 is the Lie algebra of all K-derivations of L2. The external semidirect product of L1 and L2 (denoted by L1%ϕL2 or L1%L2) is the vector space L1'L2 with the Lie bracket rpa1, b1q, pa2, b2qs " pra1, a2s, D1pb2q ´ D2pb1qq, where D1 :" ϕpa1q and D2 :" ϕpa2q. If a Lie algebra L contains an ideal N and a subalgebra B such that L " N ` B and N X B " 0, then L is the internal semidirect product of Lie algebras B and N (we write L " B%N) with the natural homomorphism of the Lie algebra B into the Lie algebra DerKN. Let W be a subspace of a vector space V over K. Elements v1, . . . , vn P V are called linearly independent over W if v1 ` W, . . . , vn ` W are linearly independent in the quotient space V {W. We write V » W if V and W are isomorphic Krx, ys-modules. 2. Krx, ys-modules with isomorphic associated Lie algebras Belitskii, Lipyanski, and Sergeichuk [2] considered a notion that is weaker than simi- larity of matrix pairs. Their notion in the case of commuting matrices can be formulated ON LIE ALGEBRAS ASSOCIATED WITH MODULES OVER POLYNOMIAL RINGS 3 in terms of Krx, ys-modules as follows. Let θ be a linear automorphism of the polynomial ring Krx, ys, defined by linear homogeneous polynomials. Let Vθ be the Krx, ys-module defined above. Definition 1. Krx, ys-modules V and W are called weakly isomorphic if there exists a series of Krx, ys-modules V1 :" V, V2, . . . , Vk :" W such that for each i " 1, 2, . . . , k either Vi » Vi`1, or Vi`1 " pViqθi for some linear automorphism θi P AutpKrx, ysq. Each weak isomorphism is an equivalence relation on the class of all Krx, ys-modules. Clearly, isomorphic Krx, ys-modules are weakly isomorphic. The converse is false by the following example. Example 1. Let V be an n-dimensional vector space over a field K. Choose a basis of V over K and take any nonzero n n matrix A with entries in K. Denote by V the Krx, ys- module with the underlying vector space V and the action of x and y on V determined by the matrix pair pA, 0nq. Let θ P AutpKrx, ysq be the automorphism such that θpxq " y and θpyq " x. Define the action of Krx, ys on Vθ by the matrix pair p0n, Aq. Then the modules V and Vθ are not isomorphic since the pairs pA, 0nq and p0n, Aq are not similar. Remark 1. Krx, ys-modules V and W are weakly isomorphic if and only if there exists a Krx, ys-module U such that U » V and W " Uθ for some linear automorphism θ of the ring Krx, ys. Proposition 1. If V and W are weakly isomorphic modules over the ring Krx, ys, then the corresponding associated Lie algebras LV and LW are isomorphic. Proof. An isomorphism ϕ of Krx, ys-modules V and W can be extend to the isomorphism ϕ of the Lie algebras LV " KxP, Qy%V and LW " KxS, T y%W via ϕpP q " S and ϕpQq " T , and further by linearity. Assume that V and W are weakly isomorphic but not isomorphic. Then there exists a Krx, ys-module U such that U » V and Uθ " W for some linear automorphism θ P AutpKrx, ysq given by a nonsingular 2 2 matrix raijs over K: θpxq " α11x ` α12y, θpyq " α21x ` α22y. It suffices to show that LU and LUθ are isomorphic Lie algebras. Write LU " KxT1, T2y%U, where T1, T2 : U Ñ U are commuting linear operators that determine the actions of x and y on U. Then LUθ " KxθpT1q, θpT2qy%U, where θpT1q " α11T1`α12T2 and θpT2q " α21T1` α22T2. It is easy to see that KxθpT1q, θpT2qy " KxT1, T2y. Thus, LU and LUθ is the same Lie algebra. (cid:3) Remark 2. Let LV " KxP, Qy%V and LW " KxS, T y%W be the Lie algebras associated with Krx, ys-modules V and W . If ϕ is an isomorphism of Lie algebras LV and LW such that ϕpV q " W , then the restriction of ϕ on V is a weak isomorphism of modules V and W . 4 A.P. PETRAVCHUK, K.YA. SYSAK Let us prove Remark 2. Let ϕpP q " α1S ` α2T ` w1 and ϕpQq " β1S ` β2T ` w2, where αi, βj P K and w1, w2 P W. Consider the weakly isomorphic module Wθ with the automorphism θ defined by θpxq " α1x ` α2y and θpyq " β1x ` β2y. We can assume that ϕpP q " S ` w1 and ϕpQq " T ` w2. Then the restriction ϕ of ϕ on V is an isomorphism of Krx, ys-modules V and Wθ. Lemma 1. Let V and W be Krx, ys-modules that are finite dimensional over K. Let the associated Lie algebras LV " KxP, Qy%V and LW " KxS, T y%W be isomorphic. Let ϕ : LV Ñ LW be an isomorphism of Lie algebras. Write W1 :" ϕpV q X W and V1 :" ϕ´1pW1q. Then V1 is a submodule of V and W1 is a submodule of W. Moreover, (i) if ϕpV q ` W " LW , then dimK V {V1 " dimK W {W1 " 2; (ii) if dimK ϕpV q ` W {W " 1, then dimK V {V1 " dimK W {W1 " 1. Proof. We have the canonical isomorphism for Krx, ys-modules (1) ϕpV q ` W {W » ϕpV q{pϕpV q X W q " ϕpV q{W1. Let ϕpV q ` W " LW . By (1) and dimK LW {W " 2, dimK ϕpV q{W1 " 2. Using the isomorphism ϕ´1, we get dimK V {V1 " 2. Since dimK LV {V1 " 4, dimK LW {W1 " 4. Then dimK W {W1 " 2. The statement (ii) is proved analogously. (cid:3) Lemma 2. Let V and W be Krx, ys-modules that are finite dimensional over K. Let LV " KxP, Qy%V and LW " KxS, T y%W be the corresponding associated Lie algebras. Let there exist an isomorphism of Lie algebras ϕ : LV Ñ LW such that ϕpKxP, Qyq ` W " LW . Then the Krx, ys-modules V and W are weakly isomorphic. Proof. Since ϕ is injective and dimK V " dimK W ă 8, it is easy to see that ϕpV q Ď W implies ϕpV q " W. By Remark 2, we can assume that ϕpV q Ę W . Since ϕpKxP, Qyq ` W " LW , we can also assume that (2) ϕpP q " S ` w1, ϕpQq " T ` w2 for some elements w1, w2 P W (passing to a weakly isomorphic module Wθ, if it is needed). We consider two cases. Case 1: ϕpV q ` W " LW . Write W1 :" ϕpV q X W and V1 :" ϕ´1pW1q. By Lemma 1, V1 is a submodule of the Krx, ys-module V with dimK V {V1 " 2 and W1 is a submodule of W with dimK W {W1 " 2. Therefore, V " Kxv1, v2y ` V1 for some v1, v2 P V zV1. Let ϕpv1q " α11S ` α12T ` u1 and ϕpv2q " α21S ` α22T ` u2 for some αij P K and u1, u2 P W . Since ϕpV q ` W " LW , the matrix rαijs is nonsingular. Therefore, we can choose v1 and v2 such that (3) ϕpv1q " S ` u1, ϕpv2q " T ` u2. Since rP, Qs " 0 in the Lie algebra LV , and ϕ is an isomorphism of Lie algebras, we get 0 " ϕprP, Qsq " rϕpP q, ϕpQqs " rS ` w1, T ` w2s " Spw2q ´ T pw1q, ON LIE ALGEBRAS ASSOCIATED WITH MODULES OVER POLYNOMIAL RINGS 5 and so (4) T pw1q " Spw2q. Analogously, the equality rv1, v2s " 0 on LV implies (5) Spu2q " T pu1q. The images of products of elements from LV in the Lie algebra LW are (6) (7) ϕprP, v1sq " rS ` w1, S ` u1s " Spu1q ´ Spw1q " Spu1 ´ w1q. ϕprQ, v2sq " rT ` w2, T ` u2s " T pu2q ´ T pw2q " T pu2 ´ w2q. Using (5) and (4), we obtain (8) (9) ϕprP, v2sq " rS ` w1, T ` u2s " Spu2q ´ T pw1q " Spu2 ´ w2q. ϕprQ, v1sq " rT ` w2, S ` u1s " T pu1q ´ Spw2q " T pu1 ´ w1q. The equality rV, V1s " 0 in the Lie algebra LV implies ϕprV, V1sq " 0 " rϕpV q, ϕpV1qs " rϕpV q, W1s. Since rW, W1s " 0 and LW " ϕpV q`W, we obtain rLW , W1s " 0, which implies rLV , V1s " 0. The last two equalities ensure (10) rP, V1s " rQ, V1s " 0, rS, W1s " rT, W1s " 0. Let us show that the linear operators P and Q act trivially on the quotient module V {V1. Let us consider, for example, P pv1q " rP, v1s. Since P pv1q P V and ϕpP pv1qq P W (by (6)), we have ϕpP pv1qq P W1. Using V1 " ϕ´1pW1q, we find that P pv1q P V1. Using (8), (9), and (7), we analogously prove that P pv2q, Qpv1q, Qpv2q P V1. Define the linear map pϕ : V Ñ W as follows: pϕpvq " ϕpvq for v P V1, pϕpv1q " u1 ´ w1, and pϕpv2q " u2 ´ w2. It is easy to check for all v P V1 that pϕpx vq " pϕpP pvqq " 0 " Spϕpvqq " x pϕpvq, pϕpx v1q " pϕpP pv1qq " ϕpP pv1qq " Spu1 ´ w1q " x pu1 ´ w1q " x pϕpv1q. Analogous relations hold for x v2, y v1, and y v2. Therefore, the linear mapping pϕ is a homomorphism of Krx, ys-modules V and W. Let us show that pϕ is a surjective homomorphism. It is sufficient to verify that u1 ´ w1 and u2 ´ w2 are linearly independent over W1 in the vector space W . Conversely, suppose that αpu1 ´ w1q ` βpu2 ´ w2q P W1 for some α, β P K such that at least one of them is nonzero. Then ϕpαpv1 ´ P q ` βpv2 ´ Qqq " αpu1 ´ w1q ` βpu2 ´ w2q P W1, αpv1 ´ P q ` βpv2 ´ Qq P V1, and therefore αP ` βQ P V, which is impossible because KxP, Qy X V " 0. We have proved that pϕ is surjective. Since the modules V and W have 6 A.P. PETRAVCHUK, K.YA. SYSAK the same dimension over K, the mapping pϕ is an isomorphism between Krx, ys-modules V and W. Case 2: ϕpV q`W ‰ LW . Then dimKpϕpV q`W q{W " 1. By Lemma 1, W1 " ϕpV qXW is a submodule of W with dimK W {W1 " 1, and V1 " ϕ´1pW1q is a submodule of V with dimK V {V1 " 1. Take any v0 P V zV1. Then V " Kxv0y ` V1 and ϕpv0q " αS ` βT ` w0 for some α, β P K and w0 P W . As earlier, we assume that ϕpP q " S ` w1, ϕpQq " T ` w2. Then ϕprP, vosq " rϕpP q, ϕpv0qs " rS ` w1, αS ` βT ` w0s " Spw0q ´ αSpw1q ´ βT pw1q, ϕprQ, vosq " rϕpQq, ϕpv0qs " rT ` w2, αS ` βT ` w0s " T pw0q ´ αSpw2q ´ βT pw2q. Using (4), we get (11) ϕprP, vosq " Spw0 ´ αw1 ´ βw2q, ϕprQ, vosq " Spw0 ´ αw1 ´ βw2q. These equalities imply ϕpP pv0qq, ϕpQpv0qq P W . As in Case 1, P pv0q, Qpv0q P V1. Define the linear map rϕ : V Ñ W by rϕpvq " ϕpvq for v P V1 and rϕpv0q " w0 ´ αw1 ´ βw2. By (11), rϕ is a homomorphism of Krx, ys-modules (note that rP, v1s, rQ, v1s P V1 for all v1 P V1). Let us show that w0 ´ αw1 ´ βw2 R W1. Conversely, let there exist v P V1 such that w0 ´ αw1 ´ βw2 " ϕpvq. Then ϕpv0 ´ vq " αS ` βT ` w0 ´ w0 ` αw1 ` βw2 " αpS ` w1q ` βpT ` w2q " ϕpαP ` βQq. Since ϕ is an isomorphism of Lie algebras, these equalities imply v0 ´ v " αP ` βQ. Then v0 ´ v P V X KxP, Qy " 0. Hence v " v0 and ϕpv0q P W1, which contradicts our choice of v0. Thus, w0 ´ αw1 ´ βw2 R W1, rϕ is a surjective homomorphism, and so it is an isomorphism of Krx, ys-modules. (cid:3) Lemma 3. Let V " Kxv1, v2, a1, a2y and W " Kxw1, w2, b1, b2y be modules over the ring Krx, ys with the following actions of x and y: (i) xv1 " a1, xv2 " a2, and the other products of x, y and the basis elements of V are zero, (ii) xw1 " b1, yw1 " b2, and the other products of x, y and the basis elements of W are zero. Then the Krx, ys-modules V and W are not weakly isomorphic, but the corresponding associated Lie algebras LV and LW are isomorphic. Proof. The annihilator AnnW pKrx, ysq :" tw P W Krx, ys w " 0u of Krx, ys in W is a submodule of W. Then AnnW pKrx, ysq " Kxw2, b1, b2y and AnnWθpKrx, ysq " AnnW pKrx, ysq " Kxw2, b1, b2y for any linear automorphism θ P AutpKrx, ysq. Let us show that AnnV pKrx, ysq " Kxa1, a2y. Take any v P V and write v " γ1v1 ` γ2v2 ` z, where γ1, γ2 P K and z P Kxa1, a2y. If v P AnnV pKrx, ysq, then x v " 0 and therefore γ1a1 ` γ2a2 " 0 because x z " 0. Since a1 and a2 are linearly independent ON LIE ALGEBRAS ASSOCIATED WITH MODULES OVER POLYNOMIAL RINGS 7 over K, we get γ1 " γ2 " 0. Thus v " z P Kxa1, a2y. Clearly, Kxa1, a2y Ď AnnV pKrx, ysq. Therefore, Kxa1, a2y " AnnV pKrx, ysq. Now suppose that the Krx, ys-modules V and W are weakly isomorphic. Then there exists Krx, ys-module U such that V » U and W " Uθ for some linear automorphism θ of the ring Krx, ys. Since V » U, dimK AnnU pKrx, ysq " 2. It is easy to see that AnnU pKrx, ysq " AnnUθpKrx, ysq. Then AnnW pKrx, ysq is of dimension 2 over K, which is impossible. The obtained contradiction shows that V and W are not weakly isomorphic. Let LV and LW be the Lie algebras associated with Krx, ys-modules V and W . Define the linear map ϕ : LV Ñ LW by ϕpP q " ´w1, ϕpQq " ´w2, ϕpv1q " S, ϕpv2q " T, ϕpa1q " b1, ϕpa2q " b2. Then ϕprP, v1sq " ϕpP pv1qq " ϕpa1q " b1 " Spw1q " rS, w1s " r´w1, Ss " rϕpP q, ϕpv1qs, ϕprQ, v1sq " ϕpQpv1qq " ϕp0q " 0 " Spw2q " r´w2, Ss " rϕpQq, ϕpv1qs. We analogously check that ϕprP, v2sq " rϕpP q, ϕpv2qs and ϕprQ, v2sq " rϕpQq, ϕpv2qs. Since the other products of basic elements are zero, we find that ϕ is an isomorphism of the Lie algebras LV and LW . (cid:3) Lemma 4. Let V and W be Krx, ys-modules that are finite dimensional over K and that are not weakly isomorphic. Let LV " KxP, Qy%V and LW " KxS, T y%W be their associated Lie algebras. Let there exist an isomorphism ϕ : LV Ñ LW of Lie algebras such that ϕpKxP, Qyq Ď W . Then V " V0 ' V2 and W " W0 ' W2, where V2 and W2 are isomorphic Krx, ys-modules with the trivial action of Krx, ys on them, and where V0 and W0 are not weakly isomorphic Krx, ys-modules with dimK V0 " dimK W0 ď 6. Proof. We have ϕpV q ` W " LW since otherwise ϕpLV q Ď ϕpKxP, Qyq ` ϕpV q Ď ϕpV q ` W Ă LW , which is impossible by ϕpLV q " LW . Write W1 :" ϕpV q X W and V1 :" ϕ´1pW1q. By Lemma 1, W1 is a submodule of codimension 2 in W , and V1 is a submodule of codimension 2 in V. Take any v1, v2 P V zV1 such that V " Kxv1, v2y ` V1. By conditions of the lemma, ϕpKxP, Qyq Ď W, and so ϕpP q " w1 and ϕpQq " w2 for some w1, w2 P W. Since ϕpV q ` W " LW , there are v1, v2 P V such that ϕpv1q " S ` u1 and ϕpv2q " T ` u2, where u1, u2 P W. Since rv1, v2s " 0, we get ϕprv1, v2sq " rϕpv1q, ϕpv2qs " rS ` u1, T ` u2s " Spu2q ´ T pu1q " 0, which implies Spu2q " T pu1q. Similarly, (12) (13) (14) ϕprP, v1sq " rϕpP q, ϕpv1qs " rw1, S ` u1s " ´Spw1q, ϕprP, v2sq " rϕpP q, ϕpv2qs " rw1, T ` u2s " ´T pw1q, ϕprQ, v1sq " rϕpQq, ϕpv1qs " rw2, S ` u1s " ´Spw2q, 8 (15) A.P. PETRAVCHUK, K.YA. SYSAK ϕprQ, v2sq " rϕpQq, ϕpv2qs " rw2, T ` u2s " ´T pw2q. It follows from rv1, V1s " rv2, V1s " 0 that ϕprv1, V1sq " rS ` u1, W1s " rS, W1s " 0, ϕprv2, V1sq " rT ` u2, W1s " rT, W1s " 0. Hence W1 Ď ZpLW q and therefore V1 " ϕ´1pW1q Ď ZpLV q, where ZpLV q and ZpLW q are the centers of the Lie algebras LV and LW . Let us show that the derived subalgebra L1 V of the Lie algebra LV is of dimension ď 4 over K. All g1, g2 P L are represented in the form g1 " α1P ` β1Q ` γ1v1 ` δ1v2 ` u3, g2 " α2P ` β2Q ` γ2v1 ` δ2v2 ` u4, where αi, βi, γi, δi P K, i " 1, 2, and u3, u4 P V1. We have rg1, g2s " α11rP, v1s ` α12rP, v2s ` α21rQ, v1s ` α22rQ, v2s for some αij P K. Thus, L1 rQ, v2s, and so dimK L1 V ď 4. V is a K-linear hull of the elements rP, v1s, rP, v2s, rQ, v1s, and Consider the subalgebra V0 of the Lie algebra LV generated by v1, v2, and their images under the action of the operators P and Q. It is easy to show that V0 is a submodule of Krx, ys-module V , and V0 is the linear hull of elements v1, v2, P pv1q, P pv2q, Qpv1q, and Qpv2q, and so dimK V0 ď 6. Take any subspace V2 of the vector space V1 such that V1 " pV0 X V1q ' V2. It is easy to see that V " V0 ' V2 and that Krx, ys acts trivially on V2 since V2 Ă V1 and rP, V1s " rQ, V1s " 0. Thus, V " V0 ' V2 is a direct sum of Krx, ys-submodules. Analogously, we consider the subalgebra W0 of the Lie algebra LW that is generated W coincides by w1, w2, and their images under the action of S and T . It is clear that L1 with the K-linear hull of elements rS, w1s, rS, w2s, rT, w1s, and rT, w2s. Since LV and LW W ď 4. Furthermore, W0 is a submodule are isomorphic Lie algebras, dimK L1 of W and dimK V0 " dimK W0 ď 6. Take any direct complement W2 of W0 X W1 in the Krx, ys-module W1. Then W " W0 ' W2 is a direct sum of submodules, and Krx, ys acts on W2 trivially since W2 Ă W1. Using dimK V2 " dimK W2, we get that V2 and W2 are isomorphic Krx, ys-modules. By the conditions of lemma, the Krx, ys-modules V and W are not weakly isomorphic, and so V0 and W0 are not weakly isomorphic too. (cid:3) V " dimK L1 Lemma 5. Let V and W be Krx, ys-modules that are finite dimensional over K. Let LV " KxP, Qy%V and LW " KxS, T y%W be the corresponding associated Lie algebras. Let there exist an isomorphism ϕ : LV Ñ LW of Lie algebras such that dimKpϕpV q ` W q{W " 1 and dimKpϕpKxP, Qyq ` W q{W " 1. Then the Krx, ys-modules V and W are weakly isomorphic. Proof. Write W1 :" ϕpV q X W and V1 :" ϕ´1pW1q. By Lemma 1, W1 is a submodule of codimension 1 in W and V1 is a submodule of codimension 1 in V . Take any v1 P V zV1. Then Kxv1y ` V1 " V. Passing to a weakly isomorphic module Wθ, if it is needed, we can ON LIE ALGEBRAS ASSOCIATED WITH MODULES OVER POLYNOMIAL RINGS 9 assume that ϕpv1q " S ` u1 for some u1 P W. The condition dimKpKxP, Qy ` W q{W " 1 implies that ϕpKxP, Qyq X W ‰ 0, and therefore ϕpαP ` βQq " w1 P W for some α, β P K and nonzero w1 P W. Passing to a weakly isomorphic module Vσ, if it is needed, we can assume that ϕpP q " w1 and get (16) ϕpv1q " S ` u1, ϕpP q " w1. Since V is an abelian subalgebra of LV , rv1, V1s " 0. Using ϕpV1q " W1, we obtain ϕprv1, V1sq " rϕpv1q, ϕpV1qs " rS ` u1, W1s " rS, W1s " 0. Similarly, rw1, W1s " 0 implies ϕ´1prw1, W1sq " rP, V1s " 0. Since P R V1, ϕpP q " w1 R W1. Therefore, W " Kxw1y ` W1. Without loss of generality, we can assume that u1 P Kxw1y, where u1 is defined in (16). Indeed, u1 " α1w1 ` w2 for some α1 P K and w2 P W1. If w2 ‰ 0, then we can take v1 ´ v2 instead of v1 for some v2 P V1 such that ϕpv2q " w2 (it is possible because ϕ : V1 Ñ W1 is a bijection). Therefore, u1 " α1w1. Further, ϕpQq " γS ` δT ` u2 for some γ, δ P K, u2 P W. Note that δ ‰ 0 since otherwise ϕpKxP, Q, v1, V1yq Ď KxSy ` W -" LW , which is impossible. We can assume that ϕpQq " T ` u3 for some u3 P W (passing to a module Wπ if it is needed, where πpxq " x, πpyq " y{δ ´ γx{δ). Moreover, replacing Q by Q1 " Q ´ µP for some µ P K and using ϕpP q " w1, we can assume that u3 P W1. This means that we go to the module Vρ with the automorphism ρ defined by ρpxq " x and ρpyq " y ´ µx. The last two automorphisms save the relations (16), so we can use them in the sequel. By 0 " ϕprP, Qsq " rϕpP q, ϕpQqs " rw1, T ` u3s " ´T pw1q, we have (17) We get (18) T pw1q " 0. ϕprP, v1sq " rϕpP q, ϕpv1qs " rw1, S ` u1s " ´Spw1q, (19) ϕprQ, v1sq " rϕpQq, ϕpv1qs " rT ` u3, S ` u1s " T pu1q ´ Spu3q " 0. The last equation in (19) holds because T pu1q " 0 (by (17)) and Spu3q " 0, since u3 P W1 and rS, W1s " 0. By ϕprQ, v1sq " 0, rQ, v1s " 0. Let us show that P pv1q P V1. Indeed, by (18) ϕpP pv1qq " ´Spw1q P W . Moreover, P pv1q P V and thus ϕpP pv1qq P ϕpV q. Then ϕpP pv1qq P W1 and P pv1q P ϕ´1pW1q " V1. Define the linear map ψ : V Ñ W by ψpvq " ϕpvq for v P V1 and ψpv1q " ´w1. Then the restriction of ψ on V1 is an isomorphism of Krx, ys modules V1 and W1. Using (18) and (19), we find that ψ is an isomorphism of Krx, ys-modules V and W. (cid:3) 10 A.P. PETRAVCHUK, K.YA. SYSAK 3. The main theorems Theorem 1. Let V and W be Krx, ys-modules that are finite dimensional over K. Let LV " KxP, Qy%V and LW " KxS, T y%W be their associated Lie algebras. Let LV and LW be isomorphic. Then one of the following conditions holds: (i) V and W are weakly isomorphic Krx, ys-modules; (ii) V " V0 ' V2 and W " W0 ' W2, where V0, W0 are not weakly isomorphic sub- modules such that dimK V0 " dimK W0 ď 6, and where V2 and W2 are submodules of equal dimension such that Krx, ys acts trivially on them. Proof. Let ϕ : LV Ñ LW be an isomorphism of Lie algebras. If ϕpV q " W or ϕpKxP, Qyq` W " LW , then V and W are weakly isomorphic modules by Remark 2 and Lemma 2. If dimK ϕpV q ` W {W " 1 and dimK ϕpKxP, Qyq ` W {W " 1, then the modules V and W are weakly isomorphic by Lemma 5. If ϕpKxP, Qyq Ď W and the modules V and W are not weakly isomorphic, then V and W are of type (ii) of the theorem by Lemma (4). Thus, we can assume that dimKpϕpKxP, Qyq ` W q{W " 1 and ϕpV q ` W " LW . By Lemma 1, W1 " ϕpV q X W is a submodule of codimension 2 in W and V1 " ϕ´1pW1q is a submodule of codimension 2 in V. Choose v1, v2 P V zV1 such that V " Kxv1, v2y ` V1 and (20) ϕpv1q " S ` u1, ϕpv2q " T ` u2 for some u1, u2 P W (which is possible since v1 and v2 are linearly independent over V1, and ϕpV q ` W " LW ). As in the proof of Lemma 4, we find that rP, V1s " rQ, V1s " 0 and rS, W1s " rT, W1s " 0. Since dimK ϕpKxP, Qyq ` W {W " 1, we get ϕpKxP, Qyq X W ‰ 0. Take a nonzero w3 P ϕpKxP, Qyq X W and write ϕpαP ` βQq " w3 for some α, β P K. We can assume that ϕpQq " w3 (passing to a weakly isomorphic module Vθ if it is needed). Moreover, ϕpKxP, Qyq Ę W , so we assume that ϕpP q " S ` u3 for some u3 P W (passing to a weakly isomorphic module Wσ if it is needed). Since rP, Qs " 0 and rv1, v2s " 0, we get rϕpP q, ϕpQqs " rS ` u3, w3s " Spw3q " 0 and rϕpv1q, ϕpv2qs " rS ` u1, T ` u2s " Spu2q ´ T pu1q " 0. As a consequence, (21) Similarly, (22) (23) (24) Spw3q " 0, Spu2q " T pu1q. ϕprP, v1sq " rϕpP q, ϕpv1qs " rS ` u3, S ` u1s " Spu1 ´ u3q, ϕprP, v2sq " rS ` u3, T ` u2s " Spu2q ´ T pu3q " ´T pu1 ´ u3q, ϕprQ, v1sq " rϕpQq, ϕpv1qs " rw3, S ` u1s " ´Spw3q " 0, ON LIE ALGEBRAS ASSOCIATED WITH MODULES OVER POLYNOMIAL RINGS 11 (25) ϕprQ, v2sq " rϕpQq, ϕpv2qs " rw3, T ` u2s " ´T pw3q (we use (21) in (23) and (24)). Since ϕ is an isomorphism of the Lie algebras LV and LW , and ϕprQ, v1sq " 0 by (24), we get rQ, v1s " 0. Observe that Kxu1 ´ u3, w3y ` W1 " W . Indeed, it is easy to see that u1 ´ u3 " ϕpv1 ´ P q. Since v1 ´ P and Q are linear independent in LV over V1, their images u1 ´ u3 and w3 under ϕ are linearly independent over W1 in W . Write V0 :" Kxv1, v2, P pv1q, P pv2q, Qpv2qy. It is easy to show that V0 is a submodule of the Krx, ys-module V and dimK V0 ď 5. Take any K-submodule V2 Ă V1 such that V1 " pV0 X V1q ' V2 (this submodule exists because P and Q act trivially on V1). Then V " V0'V2 is a direct sum of Krx, ys-submodules. Similarly, let us consider the submodule W0 :" Kxu1 ´ u3, w3, Spu1 ´ u3q, Spw3q, T pu1 ´ u3qy of W and take a submodule W2 of W1 such that W1 " pW0 X W1q ' W2 (it exists since T and S act trivially on W1). Then W " W0 ' W2 is a direct sum of submodules, and Krx, ys acts trivially on W2. Furthermore, dimK V0 " dimK W0 ď 5 and dimK V2 " dimK W2. Thus, V2 and W2 are isomorphic Krx, ys-modules. If V and W are not weakly isomorphic, then the Krx, ys- modules V0 and W0 are also not weakly isomorphic. We see that V and W are of type (ii) of the theorem. (cid:3) The following theorem is a direct consequence of the previous results. Theorem 2. Let V and W be indecomposable modules over the ring Krx, ys. Let LV " KxP, Qy%V and LW " KxS, T y%W be their associated Lie algebras. Let dimK V " dimK W ě 7. Then the Krx, ys-modules V and W are weakly isomorphic if and only if the Lie algebras LV and LW are isomorphic. Corollary 1. The problem of classifying finite dimensional Lie algebras of the form L " B%A with an abelian ideal A and a two-dimensional abelian subalgebra B is equivalent to the problem of classifying finite dimensional Krx, ys-modules up to weak isomorphism. The authors are grateful to V.V.Sergeichuk for useful discussions and advice. References [1] G. Belitskii, V.M. Bondarenko, R. Lipyanski, V.V. Plachotnik, V.V. Sergeichuk, The problems of classifying pairs of forms and local algebras with zero cube radical are wild, Linear Algebra Appl. 402 (2005) 135(cid:22)142. [2] G. Belitskii, R. Lipyanski, V. V. Sergeichuk, Problems of classifying associative or Lie algebras and triples of symmetric or skew-symmetric matrices are wild, Linear Algebra Appl. 407 (2005) 249 -- 262. [3] I. M. Gelfand, V. A. Ponomarev, Remarks on the classification of a pair of commuting linear trans- formations in a finite dimensional space, Functional Anal. Appl. 3 (1969) 325(cid:22)326. [4] D. Quillen, Projective modules over polynomial rings, Invent. Math. 36 (1976) 167 -- 171. [5] A. Suslin, Projective modules over polynomial rings are free, Soviet Math. Doklady 17 (1976) 1160(cid:22) 1164. 12 A.P. PETRAVCHUK, K.YA. SYSAK A.P. Petravchuk: Department of Algebra and Mathematical Logic, Faculty of Mechanics and Mathematics, Taras Shevchenko National University of Kyiv, 64, Volodymyrska street, 01033 Kyiv, Ukraine E-mail address: [email protected] , [email protected] K.Ya. Sysak: Department of Algebra and Mathematical Logic, Faculty of Mechanics and Mathematics, Taras Shevchenko National University of Kyiv, 64, Volodymyrska street, 01033 Kyiv, Ukraine E-mail address: [email protected]
1510.05098	2	1510	2017-02-28T14:01:55	Cosilting Modules	[ "math.RA" ]	We study the class of modules, called cosilting modules, which are defined as the categorical duals of silting module. Several characterizations of these modules and connections with silting modules are presented. We prove that Bazzoni theorem about the pure-injectivity of cotilting modules is also valid for cosilting modules.	math.RA	math	COSILTING MODULES SIMION BREAZ AND FLAVIU POP Abstract. We study the class of modules, called cosilting modules, which are defined as the categorical duals of silting modules. Several characterizations of these modules and connections with silting modules are presented. We prove that Bazzoni's theorem about the pure-injectivity of cotilting modules is also valid for cosilting modules. 1. Introduction Silting objects in triangulated categories, introduced by Keller and Vossieck in [15] for bounded derived categories, are important tools in the study of homotopy or derived categories since they are in correspondence with other important concepts such as (co-)t-structures or simple-minded collections of objects (see [17]). In [4] the authors introduced the notion of (partial) silting R-modules (where R is a unital associative ring) in order to study, in the category of all right R-modules, the class of cokernels of those homomorphisms between projective modules which represent, as 2-term complexes, silting objects in the derived category D(R). In this paper we study the modules which are defined as categorical duals (in the category of all right R-modules) of silting modules. Therefore (partial) cosilting modules are kernels of homomorphisms ζ between injective modules such that the class Bζ, consisting of all modules X with the property that HomR(X, ζ) is an epimorphism, has some closure properties which are similar to those properties satisfied by the Ext-orthogonal classes associated to (partial) cotilting modules. In fact, the class of all (partial) cosilting modules contains all (partial) cotilting modules. The main characterization of cosilting modules presented in this paper is Theorem 3.7, where cosilting modules are described as those partial cosilting modules such that Bζ has a special precovering property. Using this theorem we prove that every partial cosilting module is a direct summand of a cosilting module (Theorem 3.12). We also describe some connections of (partial) cosilting modules with (partial) silting modules (Proposition 3.6 and Corollary 3.8). Then we show in Theorem 4.7 that all (partial) cosilting modules are pure-injective via the techniques used by Bazzoni [7] (where she proved that all cotilting modules are pure-injective). Throughout this paper, by a ring R we will understand a unital associative ring, an R-module is a right R-module and we will denote by Mod-R (respectively, by R-Mod) the category of all right (respectively, left) R-modules. For an R-module T , we consider the perpendicular class ⊥T defined as follows ⊥T = {X ∈ Mod-R Ext1 R(X, T ) = 0}. 2010 Mathematics Subject Classification. 16E30, 18G15. Key words and phrases. Cosilting module, silting module, T -cogenerated, pure-injective module, torsion-free class. S. Breaz is supported by the Babe¸s-Bolyai University grant GSCE-30254. 1 2 SIMION BREAZ AND FLAVIU POP An R-module is said to be T -generated if it is an epimorphic image of a direct sum of copies of T and an R-module is called T -cogenerated if it can be embedded into a direct product of copies of T . By Gen(T ) (respectively, by Cogen(T )) we will denote the class of all T -generated (respectively, T -cogenerated) R-modules. We also denote by Add(T ) (respectively, by Prod(T )) the class of all modules which are isomorphic to direct summands of direct sums (respectively, of direct products) of copies of T . We refer to [13] for other notions and notations used in this paper. 2. The class Bζ In [4], the authors introduce the notion of silting module by defining the class Dσ = {X ∈ Mod-R HomR(σ, X) is an epimorphism} where σ : P1 → P0 is an R-homomorphism of (projective) R-modules. In order to define the dual notion, of cosilting module, we define the class Bζ as follows. If ζ : Q0 → Q1 is an R-homomorphism, then the class Bζ is defined as Bζ = {X ∈ Mod-R HomR(X, ζ) is an epimorphism}. We mention that the results presented in this section, also work in the dual case. Dualizing the notion of defect functor considered in [9] and [16], for an R- homomorphism ζ : Q0 → Q1 we can consider the codefect functor which assigns to every module X the abelian group CokerHomR(X, ζ). In this context the class Bζ is in fact the kernel of this functor. The proof of the following lemma is a simple exercise. Lemma 2.1. Let ζ : Q0 → Q1 be an R-homomorphism and assume that ζ = τζ ◦ πζ is the canonical decomposition of ζ. The following statements are equivalent for an R-module X: (1) X ∈ Bζ; (2) HomR(X, τζ) is an isomorphism and HomR(X, πζ) is an epimorphism. Corollary 2.2. Let ζ : Q0 → Q1 be an R-homomorphism with T = Ker(ζ) and let ζ = τζ ◦ πζ be the canonical decomposition. The following statements are equivalent for an R-module X which belongs to ⊥T : (1) X ∈ Bζ; (2) HomR(X, τζ) is an isomorphism. In the following results we establish some closure properties of the class Bζ. Lemma 2.3. Let ζ : Q0 → Q1 be an R-homomorphism. (1) The class Bζ is closed under direct sums. (2) If Q1 is injective then the class Bζ is closed under submodules. (3) If Q0 is injective then the class Bζ is closed under extensions. (4) If Q0 is injective and T = Ker(ζ) then Bζ ⊆ ⊥T . (5) Assume that Q1 is injective. If T = Ker(ζ) and 0 → A f −→ B g −→ X → 0 is an exact sequence such that A and B belong to the class Bζ and X ∈ ⊥T then X ∈ Bζ. COSILTING MODULES 3 Proof. (1) This follows from the fact that every contravariant Hom functor sends direct sums into direct products. (2) Let Y be in Bζ and let f : X → Y be a monomorphism. If h ∈ HomR(X, Q1) then, by the injectivity of Q1, there is an homomorphism u : Y → Q1 such that h = uf . Since HomR(Y, ζ) is an epimorphism, there exists v ∈ HomR(Y, Q0) with u = ζv, so we have the following commutative diagram Q1 h / X 0 ζ u a❇❇❇❇❇❇❇❇ f Q0 v Y such that the bottom row is exact. Then HomR(X, ζ)(vf ) = h. It follows that HomR(X, ζ) is an epimorphism, hence X ∈ Bζ. f g −→ Y (3) Let 0 → X −→ Z → 0 be an exact sequence with X, Z ∈ Bζ. If h ∈ HomR(Y, Q1) there is k ∈ HomR(X, Q0) such that ζk = hf (since HomR(X, ζ) is epic). Since Q0 is injective, there is u : Y → Q0 with k = uf . Because (ζu−h)f = 0, there exists v : Z → Q1 such that ζu − h = vg. But Z ∈ Bζ, hence there is r ∈ HomR(Z, Q0) with ζr = v. All these homomorphisms are represented in the following diagram: 0 / X f Y k ❆❆❆❆❆❆❆❆ u 0 Z v ④ r !❈❈❈❈❈❈❈❈ h ④ }④ ④ g ζ Q0 / Q1. Therefore HomR(Y, ζ)(u − rg) = h, and we obtain that HomR(Y, ζ) is an epimor- phism since h was arbitrarily. Hence Y ∈ Bζ, and the proof is complete. (4) Let X ∈ Bζ. Applying the covariant HomR(X, −) functor to the short exact sequence 0 → T i −→ Q0 πζ−→ Im(ζ) → 0, where ζ = τζ ◦ πζ is the canonical decomposition, we obtain the exact sequence 0 → HomR(X, T ) −→ HomR(X, Q0) −→ HomR(X, Im(ζ)) −→ Ext1 R(X, T ) → 0. By Lemma 2.1, we have that the homomorphism HomR(X, πζ) is an epimorphism, hence Ext1 R(X, T ) = 0. It follows that X ∈ ⊥T . (5) Suppose that ζ = τζ ◦ πζ is the canonical decomposition. Applying the contravariant functor HomR(−, Im(ζ)), we have the exact sequence 0 → HomR(X, Im(ζ)) −→ HomR(B, Im(ζ)) −→ HomR(A, Im(ζ)). We also apply the contravariant functor HomR(−, Q1), and we obtain the exact sequence 0 → HomR(X, Q1) −→ HomR(B, Q1) −→ HomR(A, Q1) → 0. Since A and B lie in Bζ, it follows by Lemma 2.1, that both HomR(A, τζ ) and HomR(B, τζ ) are isomorphisms. It follows that HomR(f, Im(ζ)) is an epimorphism, o o / / / O O a O O / / / / / ! / / ✤ ✤ ✤ } / 4 SIMION BREAZ AND FLAVIU POP hence we have the following commutative diagram with exact rows 0 0 / HomR(X, Im(ζ)) HomR(B, Im(ζ)) HomR(A, Im(ζ)) / HomR(X, Q1) / HomR(B, Q1) / HomR(A, Q1) 0 / 0 Applying the Snake Lemma we get that HomR(X, τζ ) is an isomorphism. The (cid:3) conclusion follows from Corollary 2.2. 3. Cosilting Modules Definition 3.1. We say that an R-module T is: (1) partial cosilting (with respect to ζ), if there exists an injective copresentation of T 0 → T f −→ Q0 ζ −→ Q1 such that: (a) T ∈ Bζ, and (b) the class Bζ is closed under direct products; (2) cosilting (with respect to ζ), if there exists an injective copresentation of T such that Cogen(T ) = Bζ. 0 → T f −→ Q0 ζ −→ Q1 Remark 3.2. If ζ : Q0 → Q1 is a homomorphism such that Q0 and Q1 are injective R-modules then the class Bζ is closed under submodules and extensions (Lemma 2.3). Hence the condition (b) in the definition of partial cosilting module is equiva- lent to the fact that the class Bζ is a torsion-free class. Moreover, since Cogen(T ) is closed under direct products, it is easy to see that every cosilting module is partial cosilting. In the following example we will see that every (partial) cotilting module is (partial) cosilting and, for every ring R, the trivial module 0 is cosilting. Moreover, even for some hereditary rings there exists non-zero cosilting modules which are not cotilting. Example 3.3. (a) Let us recall, from [11], that a right R-module T is partial cotilting if and only if Cogen(T ) ⊆ ⊥T and the class ⊥T is a torsion-free class. Moreover, T is cotilting if and only if Cogen(T ) = ⊥T . We note that the class ⊥T is closed under submodules if and only if id(T ) ≤ 1 (id(T ) denotes the injective dimension of T ), so every partial cotilting module is of injective dimension at most 1. Let T be an R-module of injective dimension at most 1, and consider an exact sequence 0 → T f −→ Q0 ζ −→ Q1 → 0 with Q0 and Q1 injective R-modules. Then T is (partial) cotilting if and only if T is (partial) cosilting with respect to ζ. The direct implications are obvious since Bζ = ⊥T . If we assume that T is partial cosilting with respect to ζ, we have Cogen(T ) ⊆ Bζ = ⊥T . Since Bζ is closed under direct products it follows that T is partial cotilting module. Moreover, if T is cosilting with respect to ζ, we obtain Cogen(T ) = Bζ = ⊥T , hence T is cotilting. / / / / / / / / / / / COSILTING MODULES 5 (b) For every ring R we consider an injective cogenerator E for Mod-R. If ζ : 0 → E is the trivial homomorphism then Bζ contains only the trivial module, hence 0 = Ker(ζ) is a cosilting module. (c) Let κ be a field, and let R be the ring of all upper 2×2 triangular matrices over κ. In the category Mod-R there are two simple modules S = (0, κ) and T = Q/S, where Q = (κ, κ). If ζ : Q → T ⊕ T is a homomorphism such that Ker(ζ) = S, then X ∈ Bζ if and only if Ext1 R(X, S) = 0 and HomR(X, T ) = 0. Let X ∈ Bζ. Since Q ⊕ T is an injective cogenerator for Mod-R, there exists a set I such that X can be embedded in QI ⊕ T I. Moreover, we have HomR(X, T ) = 0, and it follows that X can be embedded in QI . Therefore, we can suppose that X is a submodule of the direct product QI . Since the direct products are exact in module categories, we can view SI as a submodule of QI such that QI/SI ∼= T I. Therefore, X can be embedded in T I. Using again HomR(X, T ) = 0, it follows that X ⊆ SI , hence X ∈ Cogen(S). Then Bζ ⊆ Cogen(S). ∼= X+SI SI X∩SI Conversely, if X ∈ Cogen(S), it is easy to see that there exists a set I such that R(X, S) = 0 and HomR(X, T ) = 0, It follows that Ext1 X is isomorphic to S(I). hence X ∈ Bζ Therefore Bζ = Cogen(S), hence S is cosilting. On the other hand, S is not a cotilting module, since Q ∈ ⊥S \ Cogen(S). Applying Lemma 2.3 we obtain: Lemma 3.4. If the R-module T is partial cosilting with respect to the injective copresentation ζ : Q0 → Q1, then Cogen(T ) ⊆ Bζ ⊆ ⊥T . Some of the properties of (partial) cotilting modules are still valid for (partial) cosilting modules. We recall that the class Copres(T ) is the class of all modules X which are kernels of homomorphisms between direct products of copies of T . In the following we will denote by ◦T = {X ∈ Mod-R HomR(X, T ) = 0}, the torsion class induced by T . Corollary 3.5. Let ζ : Q0 → Q1 be a homomorphism between injective modules. If T = Ker(ζ), then the following are true: (1) If T is partial cosilting then the pair (◦T , Cogen(T )) is a torsion pair. (2) T is cosilting with respect to ζ if and only if (◦T , Bζ) is a torsion pair. (3) If T is a cosilting R-module then Cogen(T ) = Copres(T ). Proof. (1) This follows from the inclusion Cogen(T ) ⊆ ⊥T and [10, Proposition 1.4.2] by a standard proof. For reader's convenience, we present the details of this proof. Suppose that T is partial cosilting R-module with respect to the injective cop- resentation ζ : Q0 → Q1. We will prove that the class Cogen(T ) is a torsion-free class. Since Cogen(T ) is closed under both submodules and direct products, it is enough to prove that Cogen(T ) is closed under extensions. If S is the endo- morphism ring of T , let us denote by ∆ both contravariant functors HomR(−, T ) and HomS(−, T ). We recall that a module X is T -cogenerated if and only if the canonical homomorphism δX : X → ∆2(X) is a monomorphism. Let 0 → X → Y → Z → 0 be an exact sequence such that X, Z ∈ Cogen(T ). R(Z, T ) = 0, hence the induced sequence Since Cogen(T ) ⊆ ⊥T , we have that Ext1 0 → ∆(Z) → ∆(Y ) → ∆(X) → 0 6 SIMION BREAZ AND FLAVIU POP is also exact. Applying Snake Lemma to the following commutative diagram with exact rows 0 −−−−→ X −−−−→ Y −−−−→ Z yδX yδY yδZ −−−−→ 0 , 0 −−−−→ ∆2(X) −−−−→ ∆2(Y ) −−−−→ ∆2(Z) we obtain that δY is a monomorphism, hence Y ∈ Cogen(T ). Thus the class Cogen(T ) is closed under extensions. By [10, Proposition 1.4.2], we have that (T , Cogen(T )) is a torsion pair, where T = {X ∈ Mod-R HomR(X, F ) = 0, for all F ∈ Cogen(T )}. It is easy to show that T = ◦T and the proof is complete. (2) The direct implication is a consequence of (1). Conversely, if (◦T , Bζ) is a torsion pair, then Bζ is closed under direct products and T ∈ Bζ. It follows that T is partial cosilting, and we apply (1) to obtain that (◦T , Cogen(T )) is a torsion pair. Then Bζ = Cogen(T ), and the proof is complete. (3) Assume that T is cosilting R-module. It is obvious that Copres(T ) ⊆ Cogen(T ). For the reverse inclusion, let X ∈ Cogen(T ). By [10, Lemma 4.2.1], f −→ T I such that HomR(f, T ) is an there exists a monomorphism 0 → X/RejT (X) epimorphism, where RejT (X) = Tf ∈HomR(X,T ) Ker(f ). But RejT (X) = 0 since X is a T -cogenerated R-module, hence we have an exact sequence Since HomR(f, T ) is an epimorphism, the sequence 0 → X f −→ T I g −→ Y → 0. 0 → Ext1 R(Y, T ) Ext1 R(g,T ) −→ Ext1 R(T I, T ) Ext1 R(f,T ) −→ Ext1 R(X, T ) is exact. But Ext1 R(Y, T ) = 0, i.e. Y ∈ ⊥T . By Lemma 2.3(5), it follows that Y ∈ Bζ, hence Y ∈ Cogen(T ). Therefore, X ∈ Copres(T ). (cid:3) R(T I , T ) = 0 since Cogen(T ) ⊆ ⊥T . It follows that Ext1 In order to construct partial cosilting modules, we can use the same technique as in [11, Proposition 2.8] . Proposition 3.6. Let S be a commutative ring, and let R be an S-algebra. If E is an injective cogenerator for the category Mod-S, we denote by (−)d = HomS(−, E) : R-Mod ⇄ Mod-R : HomS(−, E) = (−)d the Hom-contravariant functors induced by E. Suppose that P2 → P1 ζ → P0 → M → 0 is a projective presentation of the left R-module M . The following statements are true: (1) If X ∈ Bζ d then X d ∈ Dζ . (2) If all projective modules Pi are finitely presented and Y ∈ Dζ then Y d ∈ Bζ d. (3) Suppose that all projective modules Pi are finitely presented. Then the left R- module M is partial silting with respect to ζ if and only if the dual M d is a partial cosilting right R-module with respect to ζd. COSILTING MODULES 7 Proof. Let U be the image of ζ and let ζ = τζ ◦ πζ be the canonical decomposition of ζ (i.e. πζ : P1 → U and τζ : U → P0). By the dual result of Lemma 2.1 we obtain that a left R-module Y is in the class Dζ if and only if Ext1 R(M, Y ) = 0 and HomR(πζ , Y ) is an isomorphism. Moreover, the exact sequence 0 → M d → P d 0 ζ d → P d 1 ζ ◦τ d is an injective copresentation of M d and ζd = πd ζ is the canonical decomposition of ζd. Therefore, a right R-module X belongs to the class Bζ d if and only if Ext1 ζ ) is an isomorphism. R(X, M d) = 0 and HomR(X, πd (1) Let X ∈ Bζ d . Then we have Ext1 R(X, M d) = 0. By [13, Lemma 2.16(b)] we observe that Ext1 R(X, M d) = 0 if and only if TorR R(M, X d) = 0 (by the left- side version of [13, Lemma 2.16(b)]). Now we can use [13, Lemma 2.16(a)] and its left-side version to observe that there are natural isomorphisms of functors 1 (X, M )d = 0, and this is equivalent to Ext1 1 (X, M )d, so that Ext1 R(X, M d) ∼= TorR HomR(X, (−)d) ∼=→ (X ⊗R −)d ∼=→ HomR(−, X d), and it follows that HomR(πζ , X d) is an isomorphism. Therefore, X d ∈ Dζ. (2) Suppose that Y ∈ Dζ. Then Ext1 R(M, Y )d = 0, 1 (Y d, M ) = 1 (Y d, M )d = 0, and it follows by [13, Lemma 2.16(b)] that and we can use the left version of [13, Lemma 2.16(d)] to obtain TorR 0. Therefore, TorR Ext1 R(Y d, M d) = 0. Moreover, HomR(πζ , Y ) is an isomorphism, hence HomR(πζ, Y )d is an isomor- R(M, Y ) = 0, hence Ext1 phism. Now we can use the left version of [13, Lemma 2.16(c)] to observe that Y d ⊗R πζ : Y d ⊗R P1 → Y d ⊗R U is an isomorphism, hence (Y d ⊗R πζ )d is an isomorphism. By [13, Lemma 2.16(a)] it follows that HomR(Y d, πd ζ ) is an isomorphism, and the proof is complete. (3) Suppose that M is partial silting with respect to ζ. We will prove that M d is partial cosilting with respect to ζd. Since M ∈ Dζ , it follows, by (2), that M d ∈ Bζ d . Applying the same techniques as before, we observe that we have a natural isomorphism Ext1 R(−, M d) ∼= TorR 1 (−, M )d. But E is an injective cogenerator, so for every right R-module X we have Ext1 R(X, M d) = 0 if and only if TorR 1 (X, M ) = 0. Since the left modules M and U are finitely presented, it follows by [19, Theorem A] that the functor TorR 1 (−, M ) commutes with direct products, hence the class ⊥M d is closed under direct products. In the same way, since E is an injective cogenerator we have that HomR(X, πd ζ ) is an isomorphism if and only if the homomorphism X ⊗R πζ is an isomorphism, and it follows that the class of all right R-modules X such that HomR(X, πd ζ ) is an isomorphism is closed under direct products (we note that P1 and U are finitely presented, hence the tensor functor induced by them commutes with respect direct products). Therefore the class Bζ d is closed under direct products. Hence M d is partial cosilting with respect to ζd. 8 SIMION BREAZ AND FLAVIU POP Now, suppose that M d is partial cosilting with respect to ζd. A left R-module X is in Dζ if and only if Ext1 R(M, X) = 0 and HomR(πζ , X) is an isomorphism. Since M , P1 and U are finitely presented, it follows that the functors Ext1 R(M, −), HomR(P1, −) and HomR(U, −) commute with direct sums. Therefore Dζ is closed under direct sums. In order to prove that M ∈ Dζ , let us observe that Ext1 R(M, M ) = 0 if and only R(M, M )d = 0. Since M has a projective resolution in which the first three 1 (M d, M ) = 0. if Ext1 projective modules are finitely presented, this is equivalent to TorR Applying again [13, Lemma 2.16(b)], this is equivalent to Ext1 R(M d, M d) ∼= TorR 1 (M d, M )d = 0. Since M d ∈ ⊥M d, it follows that Ext1 R(M, M ) = 0. We have to prove that HomR(πζ , M ) is an isomorphism. Since E is an injective cogenerator, this is equivalent to HomR(πζ , M )d is an isomorphism. Applying the Hom-Tensor relations, we obtain that HomR(πζ , M )d is an isomorphism if and only if M d ⊗R πζ is an isomorphism. Again, this is equivalent to (M d ⊗R πζ)d is an isomorphism, and this is true if and only if HomR(M d, πd ζ ) is an isomorphism. (cid:3) Let us recall from [20] that if M is an R-module then σ[M ] denotes the full sub- category of Mod-R whose objects are submodules of M -generated modules. Recall that σ[M ] is a Grothendieck category, and if E is an injective cogenerator in Mod-R then TrM (E) = Pf ∈HomR(M,E) Im(f ), the maximal M -generated submodule of E, is an injective cogenerator for the category σ[M ]. If 0 → T → Q0 Theorem 3.7. Let T be an R-module and let E be an injective cogenerator in ζ → Q1 is an injective copresentation for T then the Mod-R. following are equivalent: (1) T is cosilting with respect to ζ; (2) T has the following properties: (a) T is partial cosilting with respect to ζ, and (b) there exists an exact sequence 0 → T1 → T0 γ → E such that T0, T1 ∈ Prod(T ) and for every T ′ ∈ Bζ the homomorphism HomR(T ′, γ) is epic. Proof. (1)⇒(2) Suppose that T is cosilting with respect to ζ. Let I be the tor- sion part of R with respect to the torsion pair (◦T , Cogen(T )). Then for every X ∈ Cogen(T ) the canonical homomorphism HomR(R/I, X) → HomR(R, X) is an isomorphism, hence X is R/I-generated. In particular Cogen(T ) ⊆ σ[R/I]. Let α : R/I (Λ) → TrR/I (E) be an epimorphism. Since R/I ∈ Cogen(T ), it follows that R/I (Λ) ∈ Cogen(T ), hence we have a monomorphism 0 → R/I (Λ) → T Ω. We can view this monomorphism in the category σ[R/I], and we obtain that α can be extended to an epimorphism β : T Ω → TrR/I (E), by the injectivity of TrR/I (E) in the category σ[R/I]. Now the kernel K = Ker(β) is T -cogenerated, hence, taking into account Corollary 3.5, it is T -copresented. It follows that we have an exact sequence 0 → K → T Z → L → 0 such that L ∈ Cogen(T ). COSILTING MODULES 9 Using all these data we construct the pushout diagram 0 0 0 0 / K / T Ω / TrR/I (E) / T Z / X / TrR/I (E) L 0 L 0 / 0 / 0 . Since Cogen(T ) is a torsion-free class, it is closed under extensions, and it follows, by the second vertical exact sequence, that X ∈ Cogen(T ). Therefore, the exact sequence 0 → T Z → X → TrR/I (E) → 0 is in σ[R/I]. Let V ∈ Cogen(T ). If we apply the covariant extension functor Extσ[R/I](V, −) on the above exact sequence, we obtain the following exact sequence of abelian groups Extσ[R/I](V, T Z) → Extσ[R/I](V, X) → Extσ[R/I](V, TrR/I (E)). In this exact sequence the first group is zero since Extσ[R/I](V, T Z) ≤ Ext1 R(V, T Z) = 0, and the third group is also trivial since TrR/I (E) is injective in σ[R/I]. Therefore, every short exact sequence 0 → X → U → V → 0 with U ∈ σ[R/I] and V ∈ Cogen(T ) splits. By Corollary 3.5, there exists a short exact sequence 0 → X → T Γ → W → 0 with W ∈ Cogen(T ). Since this is an exact sequence in σ[R/I] and Cogen(T ) ⊆ σ[R/I], it splits, hence X ∈ Prod(T ). Therefore, in the short exact sequence 0 → T Z → X → TrR/I (E) → 0 the first two terms are in Prod(T ) and TrR/I (E) is a submodule of E, so it can be viewed as an exact sequence (♯) as in (b). 0 → T Z → X γ → E Moreover, if U belongs to Bζ = Cogen(T ) and α : U → E is a homomorphism then Im(α) ⊆ TrR/I (E), since U is an R/I-generated module by what has been / / / / / / / / 10 SIMION BREAZ AND FLAVIU POP shown so far. Then HomR(U, E) = HomR(U, TrR/I (E)), and we obtain that the natural homomorphism HomR(U, γ) : HomR(U, X) → HomR(U, E) is epic since Ext1 tions stated in (b). R(U, T Z) = 0. Therefore, the exact sequence (♯) satisfies all condi- (2)⇒(1) It is enough to prove that Bζ ⊆ Cogen(T ). Let X ∈ Bζ. Let E be an injective cogenerator for Mod-R. Then, for every non-zero element x ∈ X, there exists a homomorphism α : X → E such that α(x) 6= 0. By (b) we can lift α to a homomorphism α : X → T0. From T0 ∈ Prod(T ), it follows that, for every 0 6= x ∈ X, there exists a homomorphism α′ : X → T such that α′(x) 6= 0. It follows that X is T -cogenerated, and the proof is complete. (cid:3) We obtain the cosilting version of [10, Proposition 5.2.7]. Corollary 3.8. Suppose that we are in the same hypotheses as in Proposition 3.6(3). (1) If M is a silting module with respect to ζ then M d is a cosilting module with respect to ζd. (2) Suppose that R is an Artin algebra, S is the center of R and (−)d is the standard duality between finitely presented left and right modules induced by the injective envelope of S/J(S). Then M is a silting module with respect to ζ if and only if M d is a cosilting module with respect to ζd. Proof. (1) Suppose that M is silting with respect to ζ. By [4, Proposition 3.11] there exists an exact sequence R α→ M0 → M1 → 0 such that M0, M1 ∈ Add(M ), and for every left R-module Y ∈ Dζ the homomor- phism HomR(α, Y ) is an epimorphism. Applying the functor (−)d we obtain an exact sequence 0 → M d 1 → M d 0 αd → Rd 0 , M d 1 ∈ Prod(M d) and Rd is an injective cogenerator for Mod-R. By such that M d Theorem 3.7 and Proposition 3.6(3), it is enough to prove that for every X ∈ Bζ d the homomorphism HomR(X, αd) is epic. Let X ∈ Bζ d . By Proposition 3.6(1) we know that X d ∈ Dζ, so the homomor- phism HomR(α, X d) is epic. Moreover, for every Y ∈ R-Mod we have the natural isomorphisms HomR(X, Y d) ∼= (X ⊗R Y )d ∼= HomR(Y, X d), and if we take Y = M0, respectively Y = R, we obtain that HomR(X, αd) is an epimorphism. (2) Suppose that T = M d is a cosilting right R-module with respect to ζd. Then T is partial cosilting with respect to ζd. We observe that in the proof of Theorem 3.7 we can choose E = Rd as an injective cogenerator. Since E is finitely generated we can suppose that the set Λ is finite. Therefore (R/I)(Λ) is finitely generated. Moreover, using [10, Corollary 1.3.3], it follows that T Ω is a direct summand of a direct sum T (Γ) of copies of T . Since (R/I)(Λ) is finitely generated, it follows that we can embed it in a finite direct sum of copies of T . Therefore we can suppose that Ω is finite. Therefore K is finitely generated. Let 0 → K → T Z → L → 0 COSILTING MODULES 11 be an exact sequence such that L ∈ Cogen(T ). Since T Z is a direct summand of a direct sum T (Z ′) it follows that we can embed K in a short exact sequence f → T (Z ′) → L′ → 0 such that L′ ∈ Cogen(T ). But K is finitely generated, 0 → K hence there exists a finite subset Z ′′ of Z ′ such that α(K) ⊆ T (Z ′′), and it is not hard to see that T (Z ′′)/f (K) ∈ Cogen(T ) since it can be embedded in L′. Therefore, in the proof of Theorem 3.7 we can suppose that Z is finite. Then the module X used in the proof of this theorem is finitely generated, and we use one more time [10, Corollary 1.3.3] to conclude that X ∈ add(T ), where add(T ) denotes the class of all direct summands of finite direct sums of copies of T = M d. Therefore, we can construct an exact sequence 0 → T1 → T0 γ → E such that E = Rd and T0, T1 ∈ add(M d). Since the functors (−)d give a duality between finitely presented left and right R-modules we can write this exact sequence as γ dd → Edd, and we observe that for every Y ∈ Bζ d we have that 0 → T1 → T dd 0 HomR(Y, γdd) : HomR(Y, T dd 0 ) → HomR(Y, Edd) is an epimorphism. This implies that (Y ⊗R γd)d : (Y ⊗R T d 0 )d → (Y ⊗R Ed)d is an epimorphism, so Y ⊗R γd is a monomorphism. Let X ∈ Dζ. By Proposition 3.6(2) we obtain that X d ∈ Bζ d, so (X d ⊗R γd) : X d ⊗R Ed → X d ⊗R T d 0 is a monomorphism. This implies that HomR(γd, X)d : HomR(Ed, X)d → HomR(T d is a monomorphism, hence HomR(γd, X) is an epimorphism. 0 , X)d Therefore, since R = Ed, we have an exact sequence γ d → T d 0 → T d 1 → 0 R 0 , T d such that T d 1 ∈ add(M ), and for every X ∈ Dζ the homomorphism HomR(γd, X) is epic. By [4, Proposition 3.11] we obtain that M is silting with respect to ζ, and the proof is complete. (cid:3) Remark 3.9. (added in proof) It was proved by Angeleri and Hrbek in [3, Propo- sition 3.4] that the statement (1) in Corollary 3.8 is valid for all silting modules. The main ingredient in this general proof is the fact that the classes of all modules which are (co)generated by (co)silting modules are definable (we refer to Corollary 4.8 for the cosilting case). For reader's convenience we will present an example which shows that the con- verse proved in Corollary 3.8(2) is not valid for the general case. Example 3.10. Let P be the set of all primes. contravariant functor (−)d = Hom(−,Qp∈P Z(p∞)) : Mod-Z → Mod-Z induced by the direct product Qp∈P Z(p∞) of all injective envelopes for all simple Z-modules. If M = h 1 p p ∈ Pi then for every prime p we consider a short exact sequence 0 → If R = S = Z, we consider the 12 SIMION BREAZ AND FLAVIU POP L → M → ⊕q6=pZ(q) → 0 such that L ∼= Z, and it follows that Hom(M, Z(p∞)) ∼= Hom(L, Z(p∞)) ∼= Z(p∞). Therefore, M d ∼= Qp∈P Z(p∞). If 0 → F−1 ζ → F0 → M → 0 is a projective (free) resolution for M then we obtain a split exact sequence of injective abelian groups 0 → M d → F d −1 → 0. Since M d is an injective 0 cogenerator for Mod-Z, it follows that M d is cosilting with respect to ζd. But M is not (partial) silting since Ext1 Z(M, M ) 6= 0. ζ d → F d We mention that if we start with a cosilting module N then its dual N d is not necessarily a (partial) silting module. In order to see this we will use the examples constructed in [1, Section 3]. Example 3.11. We will use the same setting as in Example 3.10. Then N = Q⊕Q/Z is a cotilting Z-module. The dual module N d = Hom(N,Qp∈P Z(p∞)) contains a direct summand isomorphic to Q and a direct summand isomorphic to the group bZp of all p-adic integers. If we suppose that N d is partial silting then we obtain, using [4, Remark 3.8(1)], that Ext1 is a cotorsion group. But this is not true, by [8, Proposition 1.7]. Therefore, N d is not (partial) silting. By a similar argument it follows that the example constructed in [1, Section 3.4] can be used to see that over tame hereditary artin algebras of infinite representation type there exists a cotilting module W such that its dual W d is not (partial) silting. p ) = 0, hence bZ (ω) p (ω) Z(Q, bZ We thank Lidia Angeleri-Hugel for indicating us the following result: the dual of [4, Theorem 3.12] is also valid. Theorem 3.12. Let T be a partial cosilting R-module with respect to an injective ζ → Q1. Then there exists an R-module M and an copresentation 0 → T → Q0 injective copresentation 0 → T ⊕ M → Q′ 1 such that T ⊕ M is cosilting with 0 respect to ζ ′ and Bζ = Bζ ′. ζ ′ → Q′ Proof. Let E be an injective cogenerator for Mod-R. If Λ = HomR(E, Q1), and ψ : E → QΛ 1 is the canonical homomorphism, then we denote by M the pullback of 0 ⊕ E → QΛ the diagram QΛ 1 0 induced by ζΛ and ψ. We consider the exact sequence ψ ← E. Then M is the kernel of the map γ : QΛ ζ Λ → QΛ 1 0 → M ⊕ T → QΛ 0 ⊕ E ⊕ Q0 γ⊕ζ → QΛ 1 ⊕ Q1, and we will prove that M ⊕ T is cosilting with respect to γ ⊕ ζ. Note that Bγ⊕ζ = Bγ ∩ Bζ. Let X be a module in Bζ. Then X ∈ Bζ Λ. Since every homomorphism α : X → QΛ 0 , it follows that α = γδ, where δ : X → E ⊕ QΛ 0 is the homomorphism induced by β and 0. Therefore, Bζ ⊆ Bγ, hence Bγ⊕ζ = Bζ, hence Bγ⊕ζ is closed under direct products. 1 factorizes as α = ζΛβ, with β : X → QΛ COSILTING MODULES 13 Now, we will prove that M ∈ Bζ. Let α : M → Q1 be a homomorphism. We embed M in the solid rectangle of the following commutative diagram 0 0 0 / T Λ υ M ǫ ξ ρ µ α E ψ / T Λ QΛ 0 ζ Λ QΛ 1 λ πλ πλ πλ / T / Q0 ζ / Q1, and we observe that αυ : T Λ → Q1 factorizes through ζ (since T Λ ∈ Bζ), hence αυ = ζξ with ξ : T Λ → Q0. Since Q0 is injective there exists a homomorphism ρ : M → Q0 such that ξ = ρυ. Note that ζρ − α factorize through ǫ, hence there exists λ : E → Q1 such that λǫ = ζρ − α. In the above diagram πλ denote the canonical projections onto the λ-th corresponding components. Since πλψ = λ we obtain ζρ − α = λǫ = πλψǫ = ζπλµ, hence α factorizes through ζ. Then M ∈ Bζ = Bγ⊕ζ. It follows that M ⊕ T is partial cosilting with respect to γ ⊕ ζ. By Theorem 3.7, in order to complete the proof, it is enough to prove that every homomorphism β : M ′ → E, with M ′ ∈ Bζ, factorize through ǫ. This is true since ψβ factorizes through ζΛ, and M is constructed as a pullback of ψ and ζΛ. (cid:3) The following Corollary is already known, and it can be deduced from [2, p.93] and the dual of Bongarz Lemma, [13, Proposition 6.43]. Corollary 3.13. Every partial cotilting module is a direct summand of a cotilting module. Proof. If in the proof of Theorem 3.12 we assume that ζ is an epimorphism, then we obtain that γ ⊕ ζ is an epimorphism, so M ⊕ T is cotilting by Example 3.3. (cid:3) 4. Cosilting Modules are Pure Injective We recall that a short exact sequence in Mod-R is said to be pure-exact if the covariant functor HomR(F, −) preserves its exactness for every finitely pre- sented module F . An R-module U is pure-injective if the contravariant functor HomR(−, U ) preserves the exactness of every pure-exact sequence. In order to prove the main result of this section, i.e. all cosilting modules are pure-injective, we need the following characterization of pure-injective modules. Proposition 4.1. [18, Theorem 7.1] An R-module U is pure-injective if and only if the contravariant functor HomR(−, U ) preserves the exactness of the canonical short exact sequence 0 → U (λ) −→ U λ −→ U λ/U (λ) → 0 for every cardinal λ. We also need some useful lemmas related to cardinals. Let I be a set and let I be a family of subsets of I. We say that I is almost disjoint if the intersection of any two distinct elements of I is finite. Lemma 4.2. [7, Lemma 2.3] Let λ be an infinite cardinal. Then there is a family of λℵ0 countable almost disjoint subsets of λ. / / / / / / / / ✤ ✤ ✤ / / ✤ ✤ ✤ ✤ ✤ ✤ / / / 14 SIMION BREAZ AND FLAVIU POP Lemma 4.3. [14, Lemma 3.1] For any cardinal γ, there is an infinite cardinal λ ≥ γ such that λℵ0 = 2λ. Consequently, for every cardinal γ, there is an infinite cardinal λ such that γ ≤ λ < λℵ0 . We have the following construction given by Bazzoni in the proof of [7, Propo- sition 2.5]. We present the details for reader's convenience. Proposition 4.4. Let T be an R-module and let λ be an infinite cardinal. Then there is a submodule T (λ) ≤ V ≤ T λ such that V /T (λ) ∼= X (λℵ0 ), where X = T ℵ0/T (ℵ0). Proof. By Lemma 4.2, there is a family I = {Iα}α≤λℵ0 of λℵ0 countable almost disjoint subsets of λ. We view T Iα as embedded in T λ and we consider pα : T Iα → T λ/T (λ) be the restriction to T Iα of the canonical projection p : T λ → T λ/T (λ), for all α ≤ λℵ0 . We note that Ker(pα) = T (Iα) and Im(pα) ∼= X, for all α ≤ λℵ0 . Moreover, the sum Pα≤λℵ0 Im(pα) in T λ/T (λ) is actually a direct sum. Since all Im(pα) are submodules of T λ/T (λ), it follows that Lα≤λℵ0 Im(pα) ∼= X (λℵ0 ) is also a submodule of T λ/T (λ), hence there is a submodule T (λ) ≤ V ≤ T λ such that V /T (λ) ∼= X (λℵ0 ). (cid:3) Proposition 4.5. Let T be a partial cosilting R-module with respect to the injective copresentation ζ : Q0 → Q1. Then T ℵ0/T (ℵ0) ∈ Bζ. Proof. Suppose that T 6= 0 is partial cosilting with respect to the injective copre- sentation 0 → T f −→ Q0 ζ −→ Q1. By definition, T belongs to Bζ, hence for every cardinal γ we have T (γ) ∈ Bζ (by Lemma 2.3) and T γ ∈ Bζ. By Lemma 3.4, we obtain Ext1 R(T γ, T ) = 0 = Ext1 R(T (γ), T ). By Lemma 2.3 it is enough to prove that T ℵ0/T (ℵ0) ∈ ⊥T . We denote by X the R-module T ℵ0/T (ℵ0). Using Lemma 4.3 we can fix an infinite cardinal λ ≥ HomR(T, T ) such that λℵ0 = 2λ. We note that from HomR(T, T ) 6= 0 we have λ ≥ HomR(T, T ) ≥ 2, and it follows that HomR(T, T )λ = 2λ, since 2λ ≤ HomR(T, T )λ ≤ λλ = λℵ0λ = 2λλ = 2λ. Applying HomR(−, T ) to the canonical short exact sequence 0 → T (λ) i−→ T λ p −→ T λ/T (λ) → 0, we obtain the inequality HomR(T, T )λ ≥ (cid:12)(cid:12)(cid:12)Ext1 R(T λ/T (λ), T )(cid:12)(cid:12)(cid:12) . By Proposition 4.4, there is a submodule T (λ) ≤ V ≤ T λ such that V /T (λ) ∼= X (λℵ0 ). Since Bζ is closed under submodules (cf. Lemma 2.3), we observe that V belongs to Bζ hence Ext1 R(V, T ) = 0. Applying the HomR(−, T ) functor to the COSILTING MODULES 15 following pushout commutative diagram 0 y ry sy y 0 0 y yi yp y 0 0 x T (λ) T (λ) 0 −−−−→ V u−−−−→ T λ v−−−−→ T λ/V −−−−→ 0 0 −−−−→ V /T (λ) −−−−→ T λ/T (λ) −−−−→ T λ/V −−−−→ 0 j q (cid:13)(cid:13)(cid:13) we obtain the commutative diagram −−−−→ Ext2 R(T λ/V, T ) (cid:13)(cid:13)(cid:13) Ext1 R(T λ/T (λ), T ) −−−−→ Ext1 R(V /T (λ), T ) −−−−→ Ext2 R(T λ/V, T ). It follows that the natural homomorphism Ext1 R(T λ/T (λ), T ) −→ Ext1 R(V /T (λ), T ) is an epimorphism, hence we have the inequality between cardinals Now, if we assume that Ext1 R(V /T (λ), T )(cid:12)(cid:12)(cid:12) = (cid:12)(cid:12)Ext1 (cid:12)(cid:12)(cid:12)Ext1 R(T λ/T (λ), T )(cid:12)(cid:12)(cid:12) ≥ (cid:12)(cid:12)(cid:12)Ext1 2λ = HomR(T, T )λ ≥ (cid:12)(cid:12)Ext1 R(X, T ) 6= 0, then (cid:12)(cid:12)Ext1 R(X, T )(cid:12)(cid:12)λℵ0 R(X, T )(cid:12)(cid:12)λℵ0 . R(X, T )(cid:12)(cid:12) ≥ 2, and we obtain ≥ 2λℵ0 = 22λ , which is a contradiction. It follows that Ext1 and the proof is complete. R(X, T ) = 0, hence T ℵ0/T (ℵ0) ∈ ⊥T , (cid:3) We have the following result which holds in general. Proposition 4.6. [7, Lemma 2.7] Let X and U be two R-modules. If X ℵ0/X (ℵ0) ∈ Cogen(U ) then X λ/X (λ) ∈ Cogen(U ), for every cardinal λ. Theorem 4.7. Let T be an R-module. If T is partial cosilting then T is pure- injective. Proof. By Theorem 3.12 it is enough to prove that every cosilting module is pure- injective. In order to obtain this we will prove that for every cardinal λ then functor HomR(−, T ) preserves the exactness of the canonical short exact sequence 0 → T (λ) −→ T λ −→ T λ/T (λ) → 0. Suppose that T is cosilting with respect to ζ. By Proposition 4.5, we have T ℵ0/T (ℵ0) ∈ Bζ. Moreover, T ℵ0/T (ℵ0) ∈ Bζ = Cogen(T ), and applying Proposition 4.6 we obtain T λ/T (λ) ∈ Cogen(T ) for every cardinal λ. It follows that T λ/T (λ) ∈ ⊥T for every cardinal λ, and it follows that T is pure-injective. (cid:3) 16 SIMION BREAZ AND FLAVIU POP We recall that a full subcategory (or a subclass) of Mod-R is a definable sub- category if it is closed in Mod-R under direct products, direct limits and pure submodules. Corollary 4.8. If T is a cosilting R-module with respect to ζ, then the class Cogen(T ) = Bζ is definable. Proof. It is enough to prove that Bζ is closed under direct limits. If B = lim−→i∈I Bi such that all Bi are modules from Bζ, then there exists a pure exact sequence 0 → L → M Bi → B → 0 by [20, 33.9(2)]. Since T is pure injective, it follows that B ∈ ⊥T . By Lemma 2.3(5) we obtain B ∈ Bζ, and the proof is complete. (cid:3) i∈I Acknowledgment We would like to thank the referee for his/her suggestions and comments to improve this article. References [1] L. Angeleri-Hugel. "Infinite dimensional tilting theory", in Advances in representation theory of algebras. Selected papers of the 15th international conference on representations of algebras and workshop (ICRA XV), Bielefeld 2012, edited by D. J. Benson, H. Krause, A. Skowronski, EMS Series of Congress Reports 2014: 1 -- 37. [2] L. Angeleri-Hugel, and F. Coelho. "Infinitely generated complements to partial tilting mod- ules." Math. Proc. Camb. Phil. Soc. 132, (2002): 89 -- 96. [3] L. Angeleri-Hugel, and M. Hrbek. "Silting modules over commutative rings." Int. Math. Res. Not., to appear. [4] L. Angeleri-Hugel, F. Marks, and J. Vit´oria. "Silting Modules." Int. Math. Res. Not., Vol. 2016, No. 4, (2016), 1251 -- 1284. [5] L. Angeleri Hugel, A.Tonolo, and J. Trlifaj, "Tilting preenvelopes and cotilting precovers", Algebr. Represent. Theory 4 (2001): 155 -- 170. [6] M. Auslander, and I. Reiten. "Applications of contravariantly finite subcategories." Adv. in Math. 86 (1991): 111 -- 152. [7] S. Bazzoni. "Cotilting modules are pure-injective." Proc. Amer. Math. Soc. 131 (2003): 3665 -- 3672. [8] S. Bazzoni and J. Stov´ıcek. "Sigma-cotorsion modules over valuation domains", Forum Math. 21 (2009): 893 -- 920. [9] S. Breaz, and J. Zemlicka. "The defect functor of a homomorphism and direct unions." Algebr. Represent. Theory 19 (2016), 181 -- 208. [10] R. R. Colby, and K. R. Fuller. Equivalence and duality for module categories. With tilting and cotilting for rings, Cambridge University Press, Cambridge, 2004. [11] R. Colpi, A. Tonolo, and J. Trlifaj. "Partial cotilting modules and the lattices induced by them." Comm. Algebra 25, (1997): 3225 -- 3237. [12] X. H. Fu, P. A. Guil Asensio, I. Herzog, and B. Torrecillas. "Ideal Approximation Theory", Adv. in Math. 244, (2013): 750 -- 790. [13] R. Gobel, J. Trlifaj. Approximations and Endomorphism Algebras of Modules, Walter de Gruyter, Berlin, 2012. [14] P. Griffith. "On a subfunctor of Ext." Arch. Math. 21, (1970): 17 -- 22. [15] B. Keller, and D. Vossieck. "Aisles in derived categories.", Bull. Soc. Math. Belg. Sr. A 40, (1988): 239 -- 253. [16] H. Krause. "A short proof for Auslanders defect formula.", Lin. Alg. Appl. 365, (2003): 267 -- 270. [17] S. Koenig, and D. Yang. "Silting objects, simple-minded collections, t-structures and co-t- structures for finite-dimensional algebras." Doc. Math. 19, (2014): 403 -- 438. COSILTING MODULES 17 [18] C. U. Jensen, and H. Lenzing. Model Theoretic Algebra, Gordon and Breach Science Pub- lishers, New York, 1989. [19] R. Strebel. "A homological finiteness criterion." Math. Z. 151, (1976): 263 -- 275. [20] R. Wisbauer. Foundations of module and ring theory, Algebra, Logic and Applications. 3., Gordon and Breach Science Publishers, 1991. Simion Breaz: "Babes¸-Bolyai" University, Faculty of Mathematics and Computer Science, Str. Mihail Kogalniceanu 1, 400084, Cluj-Napoca, Romania E-mail address: [email protected] Flaviu Pop: "Babes¸-Bolyai" University, Faculty of Economics and Business Adminis- tration, str. T. Mihali, nr. 58-60, 400591, Cluj-Napoca, Romania E-mail address: [email protected]; [email protected]
1906.04969	1	1906	2019-06-12T07:00:54	Structure and classification of Hom-associative algebras	[ "math.RA" ]	The purpose of this paper is to study the structure and the algebraic varieties of Hom-associative algebras. We give characterize multiplicative simple Hom-associative algebras and show some examples deforming the $2\times 2$-matrix algebra to simple Hom-associative algebras. We provide a classification of $n$-dimensional Hom-associative algebras for $n\leq3$. Then study their derivations and compute small Hom-Type Hochschild cohomology groups. Furthermore, we discuss their irreducible components.	math.RA	math	STRUCTURE AND CLASSIFICATION OF HOM-ASSOCIATIVE ALGEBRAS. AHMED ZAHARI AND ABDENACER MAKHLOUF Abstract. The purpose of this paper is to study the structure and the algebraic varieties of Hom-associative algebras. We give characterize multiplicative simple Hom-associative algebras and show some examples deforming the 2 × 2-matrix algebra to simple Hom-associative algebras. We provide a classification of n-dimensional Hom-associative algebras for n ≤ 3. Then study their derivations and compute small Hom-Type Hochschild cohomology groups. Furthermore, we discuss their irreducible components. Introduction The first motivation to study nonassociative Hom-algebras came from quasi-deformations of Lie algebras of vector fields, in particular q-deformations of Witt and Virasoro algebras. The deformed algebras arising when replacing usual derivation by a σ-derivations are no longer Lie algebras. It was observed in the pioneering works, mainly by physicists, that in these examples a twisted Jacobi identity holds. Motivated by these examples and their generalization on the one hand, and the desire to be able to treat within the same framework such well-known generalizations of Lie algebras as the color and Lie superalgebras on the other hand, quasi-Lie algebras and subclasses of quasi-hom-Lie algebras and hom-Lie algebras were introduced by Hartwig, Larsson and Silvestrov in [6, 7]. The Hom-associative algebras play the role of associative algebras in the Hom-Lie setting. They were introduced by the second author and Silvestrov in [11]. Usual functors between the categories of Lie algebras and associative algebras were extended to Hom-setting, see [15] for the construction of the enveloping algebra of a Hom-Lie algebra. A Hom-associative algebra (A, µ, α) is consisting of a vector space, a multiplication and a linear self map; It may be viewed as a deformation of an associative algebra, in which the associativity condition is twisted by a linear map α and such that when α = id, the Hom-associative alge- bra degenerates to exactly an associative algebra. We aim in this paper to study the structure of Hom-associative algebras. We give a characterization of multiplicative simple Hom-associative algebras and show some examples deforming the 2 × 2-matrix algebra to simple Hom-associative algebras. Moreover we compute some invariants and discuss irreducible components of the corre- sponding algebraic varieties. Let A be an n-dimensional K-linear space and {e1, e2,··· , en} be a Key words and phrases. Hom-associative algebra, simple Hom-associative algebra, classification, cohomology, irreducible component. 1 2 AHMED ZAHARI AND ABDENACER MAKHLOUF k=1 Ck ij, were µ(ei, ej) =(cid:80)n α(ei) =(cid:80)n basis of A. A Hom-algebra structure on A with product µ is determined by n3 structure constants Ck ijek and by α which is identified by n2 structure constants aij, where j=1 ajiej. Requiring the algebra structure to be Hom-associative and unital gives rise to sub-variety HAssn (resp. UHAssn) of kn3+n2. Base changes in A result in the natural transport of structure action of GLn(K) on HAssn. Thus isomorphism classes of n-dimensional Hom-algebras are one-to-one correspondence with the orbits of the action of GLn(K) on HAssn. The decomposi- tion of HAssn into irreducible components with respect to Zariski topology is called the geometric classification of n-dimensional algebras. The paper is organized as follows. In the first section we give the basics about Hom-associative algebras and provide some new properties. Moreover, we discuss unital Hom-associative algebras. Section 2 deals with simple multiplicative Hom-associative algebras. We present one of the main results of this paper, that is a characterization of simple multiplicative Hom-associative algebras. Indeed, we show that they are all obtained by twistings of simple associative algebras. Moreover, we give all simple Hom-associative algebras, which are related to 2 × 2 matrix algebra. Section 3 is dedicated to describe algebraic varieties of Hom-associative algebras and provide classification, up to isomorphism, of 2-dimensional and 3-dimensional Hom-associative algebras. In Section 4, we study their derivations and twisted derivations, whereas in Section 5, we compute their Hom-type Hochschild cohomology. In the last section, we consider the geometric classification problem, using one-parameter formel deformations, and describe the irreducible components. 1. Structure of Hom-associative algebras Let K be an algebraically closed field of characteristic 0, A be a linear space over K. We refer to a Hom-algebra by a triple (A, µ, α), where µ : A × A → A is a bilinear map (multiplication) and α is a homomorphism of A (twist map). 1.1. Definitions. Definition 1.1. [11]. A Hom-associative algebra is a triple (A, µ, α) consisting of a linear space A, a bilinear map µ : A × A → A and a linear space homomorphism α : A → A satisfying (1.1) (1.2) µ(α(x), µ(y, z)) = µ(µ(x, y), α(z)). α(µ(x, y)) = µ(α(x), α(y)). Usually such a Hom-associative algebras are called multiplicative. Since we are dealing only with multiplicative Hom-associative algebras, we shall call them Hom-associative algebras for simplicity. We denote the set of all Hom-associative algebras by HAss. In the language of Hopf algebras, the multiplication of a Hom-associative algebra over A consists of a linear map µ : A ⊗ A → A and Condition (1.1) writes µ(α(x) ⊗ µ(y ⊗ z)) = µ(µ(x ⊗ y) ⊗ α(z)). Definition 1.2. A unital Hom-associative algebra is given by a quadruple (A, µ, α, u), where u ∈ A, such that 3 • (A, µ, α) is a Hom-associative algebra, • µ(x, u) = µ(u, x) = α(x) ∀x ∈ A, • α(u) = u. Definition 1.3. Let (A1, µ1, α1) and (A2, µ2, α2) be two Hom-associative algebras (resp. unital Hom-associative algebras with u1, u2 the units). A linear map ϕ : A1 → A2 is called a Hom- associative algebras morphism if (1.3) ϕ(µ1(x, y)) = µ2(ϕ(x), ϕ(y)) and α2 ◦ ϕ(x) = ϕ ◦ α1(x), ∀x, y ∈ A. and ϕ(u1) = u2 for unital algebras. In particular, Hom-associative algebras (A1, µ1, α1) and (A2, µ2, α2) are isomorphic if ϕ is also bijective. 1.2. Structure of Hom-associative algebras. We state in this section some properties on the structure of Hom-associative algebras which are not necessarily multiplicative. Proposition 1.4 ([14]). Let (A, µ, α) be a Hom-associative algebra and β : A → A be a Hom- associative algebra morphism. Then (A, βµ, βα) is a Hom-associative algebra. In particular, if (A, µ) is an associative algebra and β is an algebra morphism, then (A, βµ, β) is a Hom-associative algebra. Definition 1.5. Let (A, µ, α) be a Hom-associative algebra. If there is an associative algebra (A, µ(cid:48)) such that µ(x, y) = αµ(cid:48)(x, y), ∀x, y ∈ A, we say that (A, µ, α) is of associative type and (A, µ(cid:48)) is its compatible associative algebra or the untwist of (A, µ, α). Corollary 1.6. Let (A, µ, α) be a multiplicative Hom-associative algebra where α is invertible then (A, µ(cid:48) = α−1 ◦ µ) is an associative algebra and α is an automorphism with respect to µ(cid:48). Hence, (A, µ, α) is of associative type and (A, µ(cid:48) = α−1 ◦ µ) is its compatible associative algebra. Proof. We prove that (A, α−1 ◦ µ) is an associative algebra. Indeed, µ(cid:48)(µ(cid:48)(x, y), z) = α−1 ◦ µ(α−1µ(x, y), z) = α−1 ◦ µ(α−1µ(x, y), α−1 ◦ α(z)) = α−2 ◦ µ(µ(x, y), α(z)) = α−2 ◦ µ(α(x), µ(y, z)) = α−1 ◦ µ(x, α−1 ◦ µ(x, y)) = µ(cid:48)(x, µ(cid:48)(y, z)). Moreover, α is an automorphism with respect to µ(cid:48). Indeed, µ(cid:48)(α(x), α(y)) = α−1 ◦ µ(α(x), α(y)) = α ◦ α−1 ◦ µ(x, y) = α ◦ µ(cid:48)(x, y). (cid:3) 4 AHMED ZAHARI AND ABDENACER MAKHLOUF Remark 1.7. Notice that if α is not invertible, assuming µ = αµ leads to µ(α(x), µ(y, z)) = µ(µ(x, y), α(z)) αµ(α(x), αµ(y, z)) = αµ(αµ(x, y), α(z)) α2(µ(x, µ(y, z))) = α2(µ(µ(x, y), z)), which means that µ is associative up to α2. Proposition 1.8. Let (A1, µ1, α1) and (A2, µ2, α2) be two Hom-associative algebras and φ : A1 → A2 be an invertible Hom-associative algebra morphism. If (A1, µ1, α1) is of associative type and (A1, µ(cid:48) 1) is its compatible associative algebra then (A2, µ2, α2) is of associative type with compatible associative algebra (A2, µ(cid:48) 2) is an algebra morphism. 2 = φ ◦ µ1 ◦ (φ−1 ⊗ φ−1)) such that φ : (A1, µ(cid:48) 1) → (A2, µ(cid:48) Proof. Because φ is a homomorphism from (A1, µ1, α1) to (A2, µ2, α2), then α2φ = φα1, ∀x, y ∈ A, φ defines µ2 by µ2(φ(x), φ(y)) = φµ1(x, y). It is easy to check that (A2, µ2) is an associative algebra. Furthermore µ2(φ(x), φ(y)) = φ ◦ µ1(x, y) = φ ◦ α1 ◦ µ(cid:48) 1(x, y)) = α2 ◦ φµ(cid:48) 1(x, y) = α2µ(cid:48) 2(φ(x), φ(y)). We show that µ2 is an associative algebra such that µ2(u, v) = φ ◦ µ1(φ−1(u), φ−1(v)) with x = φ−1(u), y = φ−1(v) and z = φ−1(w) for all x, y, z ∈ A. µ2(µ2(u, v), w) = φ ◦ µ1(φ−1 ⊗ φ−1)(φ ◦ µ1(φ−1 ⊗ φ−1)(u, v), w) = φ ◦ µ1(φ−1 ⊗ φ−1)(φ ◦ µ1(φ−1(u), φ−1(v)), w) = φ ◦ µ1(µ1(φ−1(u), φ−1(v)), φ−1(w)) = φ ◦ µ1(φ−1(u), µ1(φ−1(v), φ−1(w))) = φ ◦ µ1(φ−1 ⊗ φ−1)(φ ⊗ φ)(φ−1(u), µ1(φ−1(v), φ−1(w)) = φ ◦ µ1(φ−1 ⊗ φ−1)(u, φµ1(φ−1(v), φ−1(w))) = µ2(u, µ2(v, w)). Hence, (A2, µ2) is an associative algebra. (cid:3) Proposition 1.9. Let (A, µ, α) be a n-dimensional Hom-associative algebra and φ : A → A be an invertible linear map. Then there is an isomorphism with a n-dimensional Hom-associative algebra (A, µ(cid:48), φαφ−1) where (cid:9) are the structure constants of µ with respect to the ba- sis {e1, . . . , en}, then µ(cid:48) has the same structure constants with respect to the basis {φ(e1), . . . , φ(en)} µ(cid:48) = φ◦µ◦(φ−1⊗φ−1). Furthermore, if(cid:8)C k when φ(ep) =(cid:80)n k=1 akpek. ij 5 Proof. We prove for any invertible linear map φ : A → A, (A, µ(cid:48), φαφ−1) is a Hom-associative algebra. µ(cid:48)(µ(cid:48)(x, y), φαφ−1(z)) = φµ(φ−1 ⊗ φ−1)(φµ(φ−1 ⊗ φ−1)(x, y), φαφ−1(z)) = φµ(µ(φ−1(x), φ−1(y)), αφ−1(z)) = φµ(αφ−1(x), µ(φ−1(y), φ−1(z))) = φµ(φ−1 ⊗ φ−1)(φ ⊗ φ)(αφ−1(x), µ(φ−1 ⊗ φ−1)(y, z))) = φµ(φ−1 ⊗ φ−1)(φαφ1(x), φµ(φ−1 ⊗ φ−1)(y, z)) = µ(cid:48)(φαφ−1(x), µ(cid:48)(y, z)). So (A, µ(cid:48), φαφ−1) is a Hom-associative algebra. It is also multiplicative. Indeed, φαφ−1µ(cid:48)(x, y) = φαφ−1φµ(φ−1 ⊗ φ−1)(x, y) = φαµ(φ−1 ⊗ φ−1)(x, y) = φµ(αφ−1(x), αφ−1(y)) = φµ(φ−1 ⊗ φ−1)(φ ⊗ φ)(αφ−1(x), αφ−1(y)) = µ(cid:48)(φαφ−1(x), φαφ−1(y)). Therefore φ : (A, µ, α) → (A, µ(cid:48), φαφ−1) is a Hom-associative algebras morphism, since φ ◦ µ = φ ◦ µ ◦ (φ−1 ⊗ φ−1) ◦ (φ ⊗ φ) = µ(cid:48) ◦ (φ ⊗ φ) and (φαφ−1) ◦ φ = φ ◦ α. It is easy to see that {φ(ei),··· , φ(en)} is a basis of A. For i, j = 1,··· , n, we have µ2(φ(ei), φ(ej)) = φµ1(φ−1(ei), φ−1(ej)) = φµ(ei, ej) =(cid:80)n k=1 Ck ijφ(ek). (cid:3) Remark 1.10. A Hom-associative algebra (A, µ, α) is isomorphic to an associative algebra if and only if α = id. Indeed, φ ◦ αφ−1 = id is equivalent to α = id. Remark 1.11. Proposition 1.9 is useful to make a classification of Hom-associative algebras. Indeed, we have to consider the class of morphisms which are conjugate. Representations of these classes are given by Jordan forms of the matrices corresponding to the morphisms. Any n × n matrix over K is equivalent, up to basis change, to a Jordan canonical form, then we choose φ such that the matrix of φαφ−1 = γ, where γ is a Jordan canonical form. Hence, to obtain the classification, we consider only Jordan forms for the structure map of Hom- associative algebras. Proposition 1.12. Let (A, µ, α) be a Hom-associative algebra. Let (A, µ(cid:48), φαφ−1) be its isomorphic Hom-associative algebra described in Proposition 1.9. If ψ is an automorphism of (A, µ, α), then φψφ−1 is an automorphism of (A, µ, φαφ−1). Proof. Note that γ = φαφ−1. We have φψφ−1γ = φψφ−1φαφ−1 = φψαφ−1 = φαψφ−1 = φαφ−1φψφ−1 = γφψφ−1. For any x, y ∈ A, φψφ−1µ(cid:48)(φ(x), φ(y)) = φψφ−1φµ(x, y) = φψµ(x, y) = φµ(ψ(x), ψ(y)) = µ(cid:48)(φψ(x), φψ(y)) = µ(cid:48)(φψφ−1(φ(x)), φψφ−1(φ(y))). By Definition, φψφ−1 is an automorphism of (A, µ(cid:48), φαφ−1). (cid:3) 6 AHMED ZAHARI AND ABDENACER MAKHLOUF The following characterization was given for Hom-Lie algebras in [13]. Proposition 1.13. Given two Hom-associative algebras (A, µA, α) and (B, µB, β), there is a Hom- associative algebra (A ⊕ B, µA⊕B, α + β), where the bilinear map µA⊕B(., .) : (A ⊕ B) × (A ⊕ B) → (A ⊕ B) is given by µA⊕B(a1 + b1, a2 + b2) = (µA(a1, a2), µB(b1, b2)), ∀ a1, a2 ∈ A, ∀ b1, b2 ∈ B, and the linear map (α + β) : A ⊕ B → A ⊕ B is given by (α + β)(a, b) = (α(a), β(b))∀a ∈ A, b ∈ B. Proof. For any ai ∈ A, bi ∈ B, by direct computation, we get µA⊕B((α + β)(a1, b1), µA⊕B(a2 + b2, a3 + b3)) == µA⊕B((α + β)(a1, b1), (µA(a2, a3), µB(b2, b3))) = µA⊕B((α(a1), β(b1)), (µA(a2, a3), µB(b2, b3))) = (µA(α(a1), µA(a2, a3)), µB(β(b1), µB(b2, b3))) = (µA(µA(a1, a2), α(a3)), µB(µB(b1, b2), β(b3))) = µA⊕B(µA⊕B(a1 + b1, a2 + b2), (α + β)(a3, b3))). (cid:3) This ends the proof. A Hom-associative algebra morphism φ : (A, µA, α) → (B, µB, β) is a linear map φ : A → B such that φ ◦ µA(a, b) = µB ◦ (φ(a), φ(b)),∀a, b ∈ A, φ ◦ α = β ◦ φ. Denote by ξφ ⊂ A ⊕ B, the graph of linear map φ : A → B. Proposition 1.14. A linear map φ : (A, µA, α) → (B, µB, β) is a Hom-associative algebra mor- phism if and only if the graph ξφ ⊂ A⊕ B is a Hom-associative subalgebra of (A⊕ B, µA⊕B, α + β). Proof. Let φ : (A, µA, α) → (B, µB, β) be a Hom-associative algebra morphism. Then for any a, b ∈ A, we have µA⊕B((a, φ(a)), (b, φ(b)) = (µA(a, b), µB(φ(a), φ(b))) = (µA(a, b), φµA(a, b)). Thus the graph ξφ is closed under the product µA⊕B. Furthermore, since φ ◦ α = β ◦ φ we have (α + β)(a, φ(a)) = (α(a), β ◦ φ(a)) = (α(a), φ ◦ α(a)), which implies that (α + β) ⊂ ξφ. Thus ξφ is a Hom-associative subalgebra of (A ⊕ B, µA⊕B, α + β). Conversely, if the graph ξφ ⊂ A ⊕ B is a Hom-associative subalgebra of (A ⊕ B, µA⊕B, α + β), then we have µA⊕B((a, φ(a)), (b, φ(b))) = (µA(a, b), µB(φ(a), φ(b)) ∈ ξφ, which implies that µB(φ(a), φ(b)) = φ ◦ µA(a, b). Furthermore, (α + β)(ξφ) ⊂ ξφ yields that (α + β)(a, φ(a)) = (α(a), β ◦ φ(a)) ∈ ξφ, which is equivalent to the condition β ◦ φ(a) = φ ◦ α(a). Therefore, φ is a Hom-associative algebra (cid:3) morphism. 7 1.3. Unital Hom-associative algebras. In this section we discuss unital Hom-associative alge- bras. We denote by UHAssn the set of n-dimensional unital Hom-associative algebras. Proposition 1.15. Let (A, µ, α) be a Hom-associative algebra. We set A = span(A, u) the vector space generated by elements of A and u. Assume µ(x, u) = µ(u, x) = α(x), ∀x ∈ A and α(u) = u. Then ( A, µ, α, u) is a unital Hom-associative algebra. Proof. It is straightforward to check the Hom-associativity. For example µ(µ(x, y), α(u)) = µ(µ(x, y), u) = α(µ(x, y)) = µ(α(x), α(y)) = µ(α(x), µ(y, u)). (cid:3) Remark 1.16. Some unital Hom-associative algebras cannot be obtained as an extension of a non- unital Hom-associative algebra. Remark 1.17. Let (A, µ, α, u) be a n-dimensional unital Hom-associative algebra and φ : A → A be an invertible linear map such that φ(u) = u. Then it is isomorphic to a n-dimensional Hom- associative algebra (A, µ(cid:48), φαφ−1, u) where µ(cid:48) = φ ◦ µ ◦ (φ−1 ⊗ φ−1). Moreover, if (cid:8)C k (cid:9) are the structure constants of µ with respect to the basis {e1, . . . , en} with e1 = u being the unit, then µ(cid:48) has the same structure constants with respect to the basis {φ(e1), . . . , φ(en)} with u the unit element. ij Indeed, we use Proposition 1.9 and Definition 1.2. The unit is conserved since µ(cid:48)(x, e1) = φ ◦ µ(φ−1(x), φ−1(e1)) = φ ◦ α ◦ φ−1(x). Proposition 1.18. Let (A1, µ1, α1, u1) and (A2, µ2, α2, u2) be two unital Hom-associative algebras. Suppose there exists a Hom-associative algebra morphism φ : A1 → A2 with φ(u1) = u2. If (A1, µ(cid:48) 1) is an untwist of (A1, µ1, α1, u1) then there exists an untwist of (A2, µ2, α2, u2) such that φ : (A1, µ(cid:48) 2) is an algebra morphism. 1) → (A2, µ(cid:48) 1, u(cid:48) 1, u(cid:48) 2, u(cid:48) Proof. Since φ is a homomorphism from (A1, µ1, α1, u1) to (A2, µ2, α2, u2), then α2φ = φα1, and for all x ∈ A we have µ2(φ(x), φ(u1)) = µ2(φ(x), u2) = α2 ◦ φ(x) and φ ◦ µ1(x, u1) = φ ◦ α1(x). By Proposition 1.8, we can see that (A2, µ2, u2) is also an associative algebra. Furthermore µ(cid:48) 2(φ(x), φ(u1)) = µ(cid:48) 2(φ(x), u2) = φ ◦ α(cid:48) 1 ◦ φ(x) = φ ◦ α1 ◦ µ1(x, u1) = α2 ◦ φ ◦ µ1(x, u1) = α2 ◦ µ2(φ(x), u2). (cid:3) 2. Simple Hom-associative algebras In this section, we study and characterize simple multiplicative Hom-associative algebras. Then we provide exemples by considering 2 × 2 matrix algebra. This study is inspired by the study of simple Hom-Lie algebras in [16]. Definition 2.1. Let (A, µ, α) be a Hom-associative algebra. A subspace H of A is called a Hom- associative subalgebra of (A, µ, α) if α(H) ⊆ H and µ(H, H) ⊆ H. In particular, a Hom-associative subalgebra H is said to be a two-sided ideal of (A, µ, α) if µ(H, A) ⊆ H and µ(A, H) ⊆ H. 8 AHMED ZAHARI AND ABDENACER MAKHLOUF Definition 2.2. The set C(A) = {x ∈ Aµ(x, y) = µ(y, x), µ(α(x), y) = µ(y, α(x)), ∀y ∈ A} is called the center of (A, µ, α). Clearly, C(A) is a two-sided ideal. Lemma 2.3. Let (A, µ, α) be a multiplicative Hom-associative algebra, then (Ker(α), µ, α) is a two-sided ideal. Proof. Obviously, α(x) = 0 ∈ Ker(α) for any x ∈ Ker(α). Since αµ(x, y) = µ(α(x), α(y) = µ(0, y) = 0 for any x ∈ Ker(α) and y ∈ A, we get µ(x, y) ∈ Ker(α). On the other hand, we have α(y) = 0 ∈ Ker(α) for any y ∈ Ker(α). Since αµ(x, y) = µ(α(x), α(y)) = µ(x, 0) = 0 for any x ∈ Ker(α) and y ∈ A, we get µ(x, y) ∈ Ker(α). Therefore, (Ker(α), µ, α) is a (cid:3) two-sided ideal of (A, µ, α). Definition 2.4. Let (A, µ, α) (α (cid:54)= 0) be a non trivial Hom-associative algebra. It is said to be a simple Hom-associative algebra if it has no proper two-sided ideal. Theorem 2.5. Let (A, µ, α) be a finite dimensional simple Hom-associative algebra. Then α is an automorphism, the Hom-associative algebra is of associative type with a simple compatible associa- tive algebra. Proof. According to Lemma 2.3, Ker(α) is a two-sided ideal. Since the Hom-associative algebra is simple, either Ker(α) = {0} or Ker(α) = A. The Hom-associative algebra is nontrivial, therefore Ker(α) (cid:54)= A. Thus, A is of associative type. Let (A, µ(cid:48) = α−1µ) be the induced associative algebra of the multiplicative simple Hom-associative algebra (A, µ, α). Clearly, α is both an automorphism of (A, µ, α) and (A, µ(cid:48)). Indeed αµ(cid:48)(x, y) = αα−1µ(x, y) = α−1µ(α(x), α(y)) = µ(cid:48)(α(x), α(y)). Suppose that A1 (cid:54)= 0, is the maximal two-sided ideal of (A, µ(cid:48)). Because α(A1) is also a two-sided ideal of (A, µ(cid:48)), then α(A1) ⊆ A1. Moreover, and µ(A1, A) = αµ(cid:48)(A1, A) ⊆ α(A1) ⊆ A1 µ(A, A1) = αµ(cid:48)(A, A1) ⊆ α(A1) ⊆ A1. So A1 is a two-sided ideal of (A, µ, α). Then A1 = A, and we have µ(A, A) = µ(A1, A) = αµ(cid:48)(A1, A) (cid:38) α(A1) ⊆ A1 = A 9 and µ(A, A) = µ(A, A1) = αµ(cid:48)(A, A1) (cid:38) α(A1) ⊆ A1 = A. Furthermore, since (A, µ, α) is a multiplicative simple Hom-associative algebra, we clearly have (cid:3) µ(A, A) = A. It is contradiction. Hence A1 = 0. By the above theorem, there exists an induced associative algebra for any multiplicative simple Hom-associative algebra (A, µ, α) and α is an automorphism of the induced associative algebra, in addition to this their products are mutually determined. Theorem 2.6. Two simple Hom-associative algebras (A1, µ1, α) and (A2, µ2, β) are isomorphic if and only if there exists an associative algebra isomorphism ϕ : A1 → A2 (between their induced as- sociative algebras) satisfying ϕ◦α = β◦ϕ. In other words, the two associative algebra automorphisms α and β are conjugate. Proof. Let (A1, µ1) and (A2, µ2) be the induced associative algebras of (A1, µ1, α) and (A2, µ2, β), respectively. Suppose ϕ : (A1, µ1, α) → (A2, µ2, β) is an isomorphism of Hom-associative algebras, then ϕ ◦ α = β ◦ ϕ, thus ϕ ◦ α−1 = β−1 ◦ ϕ. Moreover, ϕµ1(x, y) = ϕ◦α−1◦αµ1(x, y) = ϕ◦α−1µ1(x, y) = β−1◦ϕµ1(x, y) = β−1(µ2(ϕ(x), ϕ(y))) = µ2(ϕ(x), ϕ(y)). So, ϕ is an isomorphism between the two induced associative algebras. On the other hand, if there exists an isomorphism ϕ between the induced associative algebras (A1, µ1) and (A2, µ2) such that ϕ ◦ α = β ◦ ϕ, then ϕµ1(x, y) = ϕ ◦ αµ1(x, y) = β ◦ µ2(ϕ(x), ϕ(y)) = β(µ2(ϕ(x), ϕ(y)) = µ2(ϕ(x), ϕ(y)). (cid:3) 2.1. Examples of simple Hom-associative algebras. We consider the simple associative alge- bra defined by 2 × 2 matrices, which we denote by M2. Let B = {Eij} i=1,2 be the canonical basis given by elementary matrices. We seek first for algebra morphisms ϕ of M2, that is linear maps such that j=1,2 ϕ(Eij).ϕ(Ekl) = ϕ(Eij.Ekl) = δjkϕ(Eil). Then we apply the previous theorem to construct families of 4-dimensional simple Hom-associative algebras. We obtain by straightforward calculation the following algebra morphisms where δij is the Kronecker symbol. Morphism 1 ϕ(E11) = E11 − i Morphism 2 ϕ(E11) = E11 + i ϕ(E21) = E21 β1 ϕ(E21) = E21 β1 √ β2√ β1 E21 ϕ(E12) = i ϕ(E22) = i √ β1 √ β2√ β1 E21 + E22 √ β2E11 + β1E12 + β2E21 − i √ √ β1 β2E22 √ β2√ β1 √ E21 ϕ(E12) = −i β1 √ ϕ(E22) = −i β2√ β1 E21 + E22 √ √ β2E11 + β1E12 + β2E21 + i β1 √ β2E22 ϕ(E12) = β2E21 E22 ϕ(E22) = E11 − i √ β2 √ γ1E21 , E12 + β2E21 − β4E22 , , , , , λ1 E12 + β2E21 + β2 λ1 E22 1E21 − β1E22 − β3γ1E21 + β3E22 γ1 10 AHMED ZAHARI AND ABDENACER MAKHLOUF E21 1 β2 ϕ(E21) = γ1E21 ϕ(E21) = E21 β1 ϕ(E22) = λ1E21 + E22 ϕ(E21) = − λ2 ϕ(E22) = −β3γ1E21 + E22 E11 − β2 γ2 2 ϕ(E22) = λ1E21 + E22 Morphism 3 ϕ(E11) = E11 − λ1E21 ϕ(E12) = − β2 Morphism 4 ϕ(E11) = E11 − λ1E21 ϕ(E12) = β1λ1E11 + β1E12 − β1λ2 Morphism 5 ϕ(E11) = E11 + β3γ1E21 ϕ(E12) = −β3E11 + E12 Morphism 6 ϕ(E11) = i Morphism 7  ϕ(E11) = β4 Morphism 8 ϕ(E11) = E11 + γ2E12 Morphism 9 ϕ(E11) = −γ2E21 + E22 E12 + E22 ϕ(E12) = β4E11 − β2 4 β2 E12 ϕ(E21) = −γ2E11 + γ4E12 − γ2 ϕ(E12) = E12 γ1 E12 + γ1E21 + γ2E22 ϕ(E22) = − β2 ϕ(E21) = −γ2E11 − γ2 β2 ϕ(E21) = E12 β2 ϕ(E22) = E11 − β4 + γ1E21 − i √ β2 √ γ1√ β2 E11 + E12 β2 γ1E21 + E22 ϕ(E21) = i √ γ1√ β2 ϕ(E12) = E21 γ4 √ 2 γ1 β2 E12 + E22 β1 where β1, β2, β3, β4, λ1, λ2, λ3, λ4, γ1, γ2, γ3, γ4 ∈ C are parameters. They lead to simple Hom-associative algebras (M2,∗, ϕ) where Eij ∗ Epq = ϕ(EijEpq). 2 γ4 E21 + γ2E22 ϕ(E22) = E11 + γ2 γ4 E21 Therefore, the multiplication tables are given as follows : Algebra 1 Algebra 2 E11 ∗ E11 = E11 − i E12 ∗ E21 = E11 − i E21 ∗ E11 = E21 E22 ∗ E21 = E11 β1 β1 √ β2√ √ β1 β2√ β1 √ E21 E11 ∗ E12 = i √ E21 E12 ∗ E22 = i E21 ∗ E12 = −i E22 ∗ E22 = −i β1 β1 √ β2√ √ β1 β2√ β1 √ √ √ β2E11 + β1E12 + β2E21 − i √ β2E11 + β1E12 + β2E21 − i E21 + E22 √ √ β1 β1 β2E22 β2E22 E21 + E22 E11 ∗ E11 = E11 + i E12 ∗ E21 = E11 + i E21 ∗ E11 = E21 E22 ∗ E21 = E21 β1 β1 √ β2√ √ β1 β1√ β1 β1 √ E21 E11 ∗ E12 = −i √ E21 E12 ∗ E22 = −i β1 √ E21 ∗ E12 = −i β2√ √ β1 E22 ∗ E22 = −i β2√ β1 E21 + E22 E21 + E22 √ √ β1E11 + β1E12 + β2E21 + i β1 √ √ β1 β2E11 + β2E12 + β2E21 + i √ β1E22 √ β2E22 E11 ∗ E11 = E11 − λ1E21 E11 ∗ E12 = − β2 E12 ∗ E21 = E11 − λ1E21 E12 ∗ E22 = − β2 E21 ∗ E11 = − λ2 E22 ∗ E21 = − λ2 E11 − β2 E11 − β2 E21 ∗ E12 = λ1E21 + E22 E22 ∗ E22 = λ1E21 + E22. E21 E21 1 β2 λ1 λ1 γ2 γ2 1 β2 2 E12 + β2E21 + β2 λ1 2 E12 + β2E21 + β2 λ1 11 E22 E22 E11 ∗ E11 = E11 − λ1E21 E11 ∗ E12 = β1λ1E11 + β1E12 − β1λ2 E12 ∗ E21 = E11 − λ1E21 E12 ∗ E22 = β1λ1E11 + β1E12 − β1λ2 E21 ∗ E11 = E21 E22 ∗ E21 = E21 E21 ∗ E12 = λ1E21 + E22 E22 ∗ E22 = λ1E21 + E22 β1 β1 1E21 − β1E22 1E21 − β1E22 E11 ∗ E11 = E11 + β3γ1E21 E11 ∗ E12 = −β3E11 + E12 E12 ∗ E21 = E11 + β3γ1E21 E12 ∗ E22 = −β3E11 + E12 E21 ∗ E11 = γ1E21 E22 ∗ E21 = γ1E21 E21 ∗ E12 = −β3γ1E21 + E22 E22 ∗ E22 = −β3γ1E21 + E22 γ1 γ1 − β3γ1E21 + β3E22 − β3γ1E21 + β3E22 √ √ β2 √ E11 ∗ E11 = i √ E12 ∗ E21 = i β2 √ E21 ∗ E11 = i γ1√ √ β2 E22 ∗ E21 = i γ1√ β2 E11 ∗ E12 = β2E21 E12 ∗ E22 = β2E21 √ E22 E21 ∗ E12 = E11 − i β2 √ E22 E22 ∗ E22 = E11 − i β2 √ γ1E21 √ γ1E21 √ γ1√ √ β2 γ1√ β2 Algebra 3 Algebra 4 Algebra 5 Algebra 6 Algebra 7  Algebra 8 Algebra 9 γ1E21 + E22 γ1E21 + E22 E11 + E12 β2 E11 + E12 β2 + γ1E21 − i + γ1E21 − i β2 E11 ∗ E11 = β4 E12 ∗ E21 = β4 E21 ∗ E11 = E12 E22 ∗ E21 = E12 E12 + E22 E11 ∗ E12 = β4E11 − β2 E12 + E22 E12 ∗ E22 = β4E11 − β2 4 β2 E12 E21 ∗ E12 = E11 − β4 E22 ∗ E22 = E11 − β4 E12 4 β2 β2 β2 β2 β2 β2 E12 + β2E21 − β4E22 E12 + β2E21 − β4E22 E11 ∗ E11 = E11 + γ2E12 E12 ∗ E21 = E11 + γ2E12 E21 ∗ E11 = −γ2E11 − γ2 E22 ∗ E21 = −γ2E11 − γ2 2 γ1 2 γ1 E11 ∗ E12 = E12 E12 ∗ E22 = E12 E12 + γ1E21 + γ2E22 E21 ∗ E12 = − β2 E12 + γ1E21 + γ2E22 E22 ∗ E22 = − β2 β1 γ1 γ1 β1 E12 + E22 E12 + E22 E11 ∗ E11 = −γ2E21 + E22 E12 ∗ E21 = −γ2E21 + E22 E21 ∗ E11 = −γ2E11 + γ4E12 − γ2 E22 ∗ E21 = −γ2E11 + γ4E12 − γ2 2 γ4 2 γ4 E11 ∗ E12 = E21 E12 ∗ E22 = E21 γ4 γ4 E21 + γ2E22 E21 ∗ E12 = E11 + γ2 E21 + γ2E22 E22 ∗ E22 = E11 + γ2 γ4 E21 E21 γ4 12 AHMED ZAHARI AND ABDENACER MAKHLOUF 3. Algebraic varieties of Hom-associative algebras and Classification In this section, we deal with algebraic varieties of Hom-associative algebras with a fixed dimen- sion. A Hom-associative algebra is identified with its structure constants with respect to a fixed basis. Their set corresponds to an algebraic variety where the ideal is generated by polynomials corresponding to the Hom-associativity condition. 3.1. Algebraic varieties HAssn. Let A be a n-dimensional K-linear space and {e1,··· , en} be a basis of A. A Hom-algebra structure on A with product µ determined by n3 structure constants Ck ijek and a structure map α determined by n2 structure constants aji, k=1 Ck j=1 ajiej. ij where µ(ei, ej) =(cid:80)n where α(ei) =(cid:80)n  (3.1) If we require this algebra structure to be Hom-associative, then this limits the set of structure constants (Ck ij, aij) to a cubic sub-variety of the affine algebraic variety Kn3+n2 defined by the (cid:80)n (cid:80)n following polynomial equations system : (cid:80)n m=1 ailCm jkCs p=1 aspCp i, j, k, s = 1,··· , n. i, j, s = 1,··· , n. ij −(cid:80)n lm = 0, pq = 0, p=1 l=1 Moreover if µ is commutative, we have Ck i, j, k = 1,··· , n. (cid:80)n ijCs lm − amkCl q=1 apiaqjCs ij = Ck ji The first set of equation correspond to the Hom-associative condition µ(α(ei), µ(ej, ek)) = µ(µ(ei, ej), α(ek)) and the second set to multiplicativity condition α ◦ µ(ei, ej) = µ(α(ei), α(ej)). We denote by HAssn the set of all n-dimensional multiplicative Hom-associative algebras. Assume that e1 = u, the unit, in the basis B. It turns out that in addition to the system (3.1), 1iek = we have the following condition with respect to unitality : u1.ei = ei.u1 = α(ei) ⇒(cid:80)n (cid:80)n k=1 Ck i1ek =(cid:80)n k=1 akiek, that is k=1 Ck Ck i1 = Ck 1i = aki ∀i, k. (3.2) We denote by UHAssn the algebraic varieties of n-dimensional unital Hom-associative algebras. 3.2. Action of linear group on the algebraic varieties HAssn. The group GLn(K) acts on the algebraic varieties of Hom-structures by the so-called transport of structure action defined as follows. Let A = (A, µ, α) be a n-dimensional Hom-associative algebra defined by multiplication µ and a linear map α. Given f ∈ GLn(K), the action f · A transports the structure, Θ : GLn(K) × HAssn −→ HAssn (cid:55)−→ (A, f−1 ◦ µ ◦ (f ⊗ f ), f ◦ α ◦ f−1) defined for x, y ∈ A, by (f, (A, µ, α)) (3.3) The conjugate class is given by Θ(f, (A, µ, α)) = (A, f−1 ◦ µ◦ (f ⊗ f ), f ◦ α◦ f−1)) for f ∈ GLn(K). f · µ(x, y) = f−1µ(f (x), f (y)), f · α(x) = f−1α(f (x)). The orbit of a Hom-associative algebra A of HAssn is given by 13 ϑ(A) = {A(cid:48) = f · A, f ∈ GLn(K)} . The orbits are in 1-1 correspondence with the isomorphism classes of n-dimensional Hom- associative algebras. The stabilizer is Stab((A, µ, α)) =(cid:8)f ∈ GLn(K)(f−1 ◦ µ ◦ (f ⊗ f ) = µ and f ◦ α = α ◦ f(cid:9) . We characterize in terms of structure constants the fact that two Hom-associative algebras are in the same orbit (or isomorphic). Let (A, µ1, α1) and (A, µ2, α2) be two n-dimensional Hom-associative algebras. They are isomorphic if there exists ϕ ∈ GLn(K) such that (3.4) ϕ ◦ µ1 = µ2(ϕ ⊗ ϕ) and ϕ ◦ α1 = α2 ◦ ϕ. Remark 3.1. Conditions (3.4) are equivalent to µ1 = ϕ−1 ◦ µ2 ◦ ϕ ⊗ ϕ and α1 = ϕ−1 ◦ α2 ◦ ϕ. We set with respect to a basis {ei}i=1,··· ,n: p=1 apiep, α1(ei) =(cid:80)n ϕ(ei) =(cid:80)n µ1(ei, ej) =(cid:80)n ijek ijaqk−(cid:80)n (cid:80)n Conditions (3.4) translate to the following system : k=1 Ck ijek, µ2(ei, ej) =(cid:80)n k=1 Ck (cid:80)n p=1 Dq k=1 j=1 αjiej, α2(ei) =(cid:80)n k=1 αjiaqk−(cid:80)n pkapiakj = 0, and,(cid:80)n k=1 Dk j=1 βjiej i, j = 1,··· , n. i = 1,··· , n k=1 akiβqk = 0, i, j, q = 1,··· , n. 3.3. Algebraic Variety HAss2. A Hom-associative algebra is identified to its structure constants (C k i,j) and (aij) with respect to a given basis. They satisfy the first family of system (3.1), for which the solutions belong to the 14 AHMED ZAHARI AND ABDENACER MAKHLOUF algebraic variety defined by the following Groebner basis. (cid:104)a21c1 a12(c1 a12(c2 −a11c1 a12c1 11c1 −a22c1 a12c1 11c2 a12c1 11c2 −a11c2 11c1 a12c2 a12c1 11c1 −a12c1 a12c2 11c1 −a12c2 21 − a11c2 12 − a21c1 11c1 11c1 11c1 12 − a22c1 11c1 11c1 11)2 + a11c1 12 − a11c1 11c1 11)2 + a12c2 11c1 12 − a21(c1 12)2 + a11c1 11c1 12 − a12c1 12 + a12c1 12c2 11c1 22 + a22c1 12c2 21 + a12c1 12c1 12 − a22c1 11c1 11 + a11c2 11c2 12 − a11c2 11 + a12c2 11c2 11c1 12 − a21c1 12 + a11c2 12c2 11c1 12)2 − a12c2 11c1 12 + a12(c2 21 + a22(c1 21)2 + a12c1 12c2 12)2 − a12c2 12 − a22(c1 11c1 21 + a22c1 21 + a12c2 12c2 21c2 12 − a22c1 12 − a12c2 11c1 12c2 12 + a11c2 11c2 21, a21c1 11c1 12 − a12c2 12 + a11c1 11c1 12c2 21 − a21c1 21 + a22c1 12c1 11c1 21 − a11c2 21 + a21(c1 12c1 11c1 21 − a12c1 21c2 21 + a22c2 12c1 22 − a12c1 12c1 21c2 21c1 22, a22c2 12)2 − a12c2 11c2 12 + a11(c2 21 − a21c2 12c1 21 + a22c1 11c2 21 − a21c2 21c2 21 + a21c1 11c1 21 − a22c2 12c2 21 + a22c1 12c1 21 − a11c1 22 − a21c1 11c1 12c1 22 − a22c2 11c1 21 + a11c1 12c1 22 − a21c2 21 − a11c2 11c1 12c1 22 + a21c2 21 + a11c2 12c2 11c1 21 − a11c2 12 − a21c1 11c2 11c2 21 − a22c2 11c1 12c1 21 + a21c1 21 − a11c1 21 + a22c2 11c1 21c2 21 − a21c2 12c2 21)2 + a11c1 22 − a22c2 21c1 22, 22 − a22c2 22 − a12c2 12c2 21 − a22c2 12c2 21 + a21c1 21 − a11(c2 21)2 + a22c2 21c2 12c2 22 + a21c2 21 − a12(c2 12c2 21)2 + a22c2 22 − a11c1 21c2 22 + a22c2 21c1 22 + a11c1 21c1 22 + a21c1 12c1 22 − a11c2 22 + a22c2 21c2 21c2 22 − a22c2 22 + a11c2 21c1 12c2 12c2 11c2 11c2 22, 11c2 12 + a11c2 21, 12c1 22 + a21c2 22 − a21c2 21c1 22 + a21c2 12c1 22, 21c1 22, 22, 22, 22 + a22c2 22 + a21c2 22 − a21c2 21c2 12c2 21c2 22, 22, 22, 22 − a22c2 21c2 22 − a21c1 22c2 12c2 22 + a21c1 22 − a21(c2 22 + a21(c2 12c2 22, 22c2 22)2, 22)2(cid:105) 22, If the Hom-associative algebra is multiplicative, it should satisfy further the second family of (3.1), that is, it belongs to the intersection with the the algebraic variety defined by the following Groebner basis. (cid:104)a11c1 a11a12c1 a11a12c1 12c1 a2 a21c1 a11a12c2 a11a12c2 a2 12c2 11 − a2 11c1 11 + a12c2 11 − a11a21c1 12 − a11a21c1 21 − a2 21c1 22, 11 + a11c1 11 − a12a21c1 12 − a11a2,2c1 12 + a11c1 12 − a12a22c1 12 + a12c2 21 − a11a22c1 21 + a11c1 11 + a22c2 11 − a11a21c2 11 − a12a22c1 11 − a2 11c2 22 − a2 22c1 12 − a11a21c2 22 + a12c2 21 − a2 21c2 22, 22, 12 − a12a21c1 21 + a12c2 21 − a21a22c1 21 − a21a22c1 22, 22 12 + a22c2 11 + a21c1 11 − a12a21c2 12 − a11a22c2 21 + a22c2 21 + a21c1 12 − a12a21c2 21 − a11a22c2 22 + a22c2 12 + a21c1 12 − a12a22c2 11 − a12a22c2 21 − a21a22c2 22, 21 − a21a22c2 22, 22 − a2 22(cid:105). 22c2 Describing the algebraic varieties by solving such systems lead to the 2-dimensional and 3- dimensional Hom-associative algebras classifications. 3.4. Classification of 2-dimensional Hom-associative algebras. We have to consider two classes of morphisms which are given by Jordan forms, namely they . We check whether the previous are  a 0 0 b  and  a 1 0 a are represented by the matrices isomorphic. We provide all 2-dimensional Hom-associative algebras, corresponding to solutions of the system (3.1). To this end, we use a computer algebra system. 15 Lemma 3.2. Let α be a diagonal morphism such that α(e1) = pe1, α(e2) = qe2, p (cid:54)= q with respect to basis {e1, e2}. Then any ϕ : A → A such that ϕ ◦ α = α ◦ ϕ is of the form ϕ(e1) = λe1 and ϕ(e2) = ρe2 with respect to the same basis. Proof. Let ϕ(e1) = λ1e1 +λ2e2 and ϕ(e2) = ρ1e1 +ρ2e2. On the one hand, ϕ◦α(e1) = λ1pe1 +λ2pe2 and α(cid:48) ◦ ϕ(e1) = λ1p(cid:48)e1 + λ2q(cid:48)e2. So we have λ1p = λ1p(cid:48) and λ2p = λ2q(cid:48). On the other hand, ϕ ◦ α(e2) = qρ1e1 + qρ2e2 and α(cid:48) ◦ ϕ(e2) = ρ1p(cid:48)e1 + ρ2q(cid:48)e2. We have ρ1q = ρ1p(cid:48) and ρ2q = ρ2q(cid:48). Then we have λ1(p − p(cid:48)) = 0, λ2(p − q(cid:48)) = 0, q = q(cid:48), we have λ2(p − q(cid:48)) = 0 and ρ2(q − q(cid:48)) = 0. If p (cid:54)= q, so λ2 = ρ1 = 0. Hence the lemma with λ = λ1 and ρ = ρ2. ρ2(q − q(cid:48)) = 0. If p = p(cid:48) and ρ1(q − p(cid:48)) = 0, (cid:3) Theorem 3.3. Every 2-dimensional multiplicative Hom-associative algebra is isomorphic to one of the following pairwise non-isomorphic Hom-associative algebra (A,∗, α), where ∗ is the multipli- cation and α the structure map. We set {e1, e2} to be a basis of K2. e2 ∗ e1 = e2, e2 ∗ e2 = e1, α(e1) = e1, α(e2) = −e2; e2 ∗ e1 = 0, e2 ∗ e1 = 0, e2 ∗ e1 = e2, e2 ∗ e1 = 0, e2 ∗ e1 = 0, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = 0; e2 ∗ e2 = 0, α(e1) = e1, α(e2) = 0; e2 ∗ e2 = 0, α(e1) = e1, α(e2) = e2; e2 ∗ e2 = 0, α(e1) = 0, α(e2) = ke2; e2 ∗ e2 = 0, α(e1) = e1, α(e2) = e2; e2 ∗ e1 = be1, e2 ∗ e2 = ce1, α(e1) = 0, α(e2) = e1, e1 ∗ e2 = e2, e1 ∗ e2 = 0, e1 ∗ e2 = 0, e1 ∗ e2 = e2, e1 ∗ e2 = 0, e1 ∗ e2 = 0, e1 ∗ e2 = ae1, 1 : e1 ∗ e1 = −e1, A2 2 : e1 ∗ e1 = e1, A2 3 : e1 ∗ e1 = e1, A2 4 : e1 ∗ e1 = e1, A2 5 : e1 ∗ e1 = e1, A2 6 : e1 ∗ e1 = e2, A2 7 : e1 ∗ e1 = 0, A2 where a, b, c, k ∈ C; 8 : e1∗ e1 = 0, A2 9 : e1∗ e1 = 0, A2 e2∗ e2 = e1 + e2, α(e1) = e1, α(e2) = e1 + e2; e2∗ e2 = e1 + e2, α(e1) = e1, α(e2) = e1 + e2. Proof. The proof follows from straightforward calculation using Definition 1.3 and Lemma 3.2. (cid:3) e2∗ e1 = 0, e2∗ e1 = e1, e1∗ e2 = e1, e1∗ e2 = 0, Proposition 3.4. The Hom-associative algebras A1, A4, A6, A8, A9 are of associative type. Proof. Indeed, we set in the following corresponding associative algebras : e2 · e1 = −e2, e2 · e2 = e1. 1 : e1 · e1 = −e1, A2 4 : e1 · e1 = e1, A2 6 : e1 · e1 = e2, A2 8 : e1 · e1 = 0, A2 9 : e1 · e1 = 0, A2 e1 · e2 = −e2, e1 · e2 = e2, e1 · e2 = 0, e1 · e2 = e1, e1 · e2 = e1, e2 · e1 = e2, e2 · e1 = 0, e2 · e1 = 0, e2 · e1 = 0, e2 · e2 = 0. e2 · e2 = 0. e2 · e2 = e2. e2 · e2 = e2. Remark 3.5. It turns out that A2 algebra. 2, A2 3, A2 5, A2 7 cannot be obtained by twisting of an associative (cid:3) 16 AHMED ZAHARI AND ABDENACER MAKHLOUF 3.5. Classification of 3-dimensional Hom-associative algebras. We seek for all 3-dimensional Hom-associative algebras. We consider two classes of morphism which are given by Jordan form, namely they are represented by the matrices  a 0 0 0 0 b  ,  a 1  , 0  a 1  . 0 0 a 0 0 a 1 Using similar calculation as in previous section, we obtain the following classification. 0 0 c 0 0 b 0 0 a Theorem 3.6. Every 3-dimensional multiplicative Hom-associative algebra is isomorphic to one of the following pairwise non-isomorphic Hom-associative algebra (A,∗, α), where ∗ is the multipli- cation and α the structure map. We set {e1, e2, e3} to be a basis of K3 : (the non written products and images of α are equal to zero) e2 ∗ e3 = e2 + e3, e3 ∗ e2 = e2 + e3, e3 ∗ e3 = e3 ∗ e3 = p3e3, α(e1) = e1, α(e2) = e2; e3 ∗ e3 = p3e3, α(e1) = e1, α(e2) = e2, α(e3) = e3; e3∗e2 = p4e1, e2∗e3 = p4e1, e3∗e1 = p5e1, e3 ∗ e3 = p2e3, α(e1) = e1, α(e2) = e1 + e2; e3 ∗ e2 = e1, α(e2) = e1, α(e3) = e3; e3 ∗ e3 = e1, α(e1) = e1, α(e2) = e1 + e2 ∗ e2 = e2 + e3 e2 ∗ e2 = p2e2, e2 ∗ e2 = p2e2, 1 : e1 ∗ e1 = e1, A3 e2 + e3, α(e1) = e1; 2 : e1 ∗ e1 = p1e1, A3 3 : e1 ∗ e1 = p1e1, A3 4 : e1∗e2 = p1e1, e1∗e3 = p2e1, e2∗e2 = p3e1, A3 e3 ∗ e3 = p6e1, α(e2) = e1; 5 : e2 ∗ e2 = p1e1, A3 6 : e1 ∗ e2 = e1, A3 7 : e2 ∗ e2 = e1, A3 e2, α(e3) = e3; 8 : e1 ∗ e2 = −e3, A3 9 : e2 ∗ e3 = p1e1, A3 10 : e2 ∗ e2 = p1e1, A3 11 : e1 ∗ e3 = p1e1, A3 12 : e2∗e3 = −p1e1, A3 e2 ∗ e2 = e1, e2 ∗ e3 = e1, e2 ∗ e3 = e1, e3 ∗ e2 = e1, e2 ∗ e2 = e3, α(e1) = e1, α(e2) = e1 + e2, α(e3) = e3; e2 ∗ e1 = e3, e3 ∗ e2 = p2e1, α(e1) = ae1, α(e2) = e1 + ae2, α(e3) = e3; e3 ∗ e3 = p2e1, α(e1) = e1, α(e2) = e1 + e2, α(e3) = −e3; e2 ∗ e3 = p2e1, e3∗e2 = p1e1, e3 ∗ e3 = p3e1, α(e2) = e1, α(e3) = e2; e3∗e3 = p2e1, α(e1) = e1, α(e2) = e1+e2, α(e3) = e2 + e3. Proposition 3.7. The Hom-associative algebras A3 3, A3 7, A3 8, A3 9, A3 10, A3 12 are of associative type. Proof. Indeed, we set in the following the corresponding associative algebras : 3 :e1 · e1 = p1e1, A3 7 :e2 · e1 = e1 A3 8 :e1 · e2 = −e3, A3 9 :e2 · e3 = p1 A3 a e1, e2 · e2 = p2e2, e3 · e3 = p3e3; e2 · e2 = e1, e2 · e1 = e3 e3 · e2 = p2 a e1; e2 · e3 = e1, e2 · e2 = e3; e3 · e2 = e1, e3 · e3 = e1; 10 :e2 · e2 = p1e1, A3 12 :e2 · e1 = e3 A3 e3 · e3 = p2e1; e2 · e3 = −p1e1, where pi are parameters. e3 · e2 = p1e1, e3 · e3 = p2e1; 17 (cid:3) Remark 3.8. It turns out that A3 ciative algebra. 1, A3 2, A3 4, A3 5, A3 6, A3 11 cannot be obtained by twisting of an asso- Theorem 3.9. Every 2-dimensional unital multiplicative Hom-associative algebra is isomorphic to one of the following pairwise non-isomorphic Hom-associative algebra (A,∗, α), where ∗ is the multiplication and α the structure map. We set {e1, e2} to be a basis of K2 where e1 is the unit : e2 ∗ e2 = e1 + e2, α(e1) = e1, α(e2) = e2; e2 ∗ e2 = e1, α(e1) = e1, α(e2) = −e2; e2 ∗ e1 = e2, e2 ∗ e1 = −e2, 1 : e1 ∗ e1 = e1, A(cid:48)2 2 : e1 ∗ e1 = e1, A(cid:48)2 3 : e1 ∗ e1 = e1, A(cid:48)2 3 : e1 ∗ e1 = e1, A(cid:48)2 e1 ∗ e2 = e2, e1 ∗ e2 = −e2, e1 ∗ e2 = 0, e1 ∗ e2 = e2, e2 ∗ e1 = 0, e2 ∗ e1 = e2, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = 0; e2 ∗ e2 = 0, α(e1) = e1, α(e2) = e2. Proposition 3.10. The unital Hom-associative algebras A(cid:48)2 1, A(cid:48)2 2, A(cid:48)2 4 are of associative type. Proof. Indeed, we set in the following the corresponding associative algebras : A(cid:48)2 A(cid:48)2 A(cid:48)2 1 : e1 · e1 = e1, 2 : e1 · e1 = e1, 4 : e1 · e1 = e1, e1 · e2 = e2, e1 · e2 = e2, e1 · e2 = e2, e2 · e1 = e2, e2 · e1 = e2, e2 · e1 = e2, e2 · e2 = e1 + e2. e2 · e2 = e1. e2 · e2 = 0. (cid:3) Remark 3.11. It turns out that A(cid:48)2 3 cannot be obtained by twisting of an associative algebra. Theorem 3.12. Every 3-dimensional unital multiplicative Hom-associative algebra is isomorphic to one of the following pairwise non-isomorphic Hom-associative algebras (A,∗, α), where ∗ is the multiplication and α the structure map. We set {e1, e2, e3} to be a basis of K3 where e1 is the unit (the no written products and images of α are equal to zero) : 1 : e1 ∗ e1 = e1, A(cid:48)3 e2 ∗ e2 = e2 + e3, e2 ∗ e3 = e2 + e3, e3 ∗ e2 = e2 + e3, e3 ∗ e3 = e2 + e3, α(e1) = e1; 2 : e1 ∗ e1 = e1, A(cid:48)3 3 : e1 ∗ e1 = e1, A(cid:48)3 e2 ∗ e2 = e2, e2 ∗ e2 = e2, e3 ∗ e1 = e3, e1 ∗ e3 = −e3, e3 ∗ e3 = e1 + e3, α(e1) = e1, α(e3) = e3; e3 ∗ e3 = e1, α(e1) = e3 ∗ e1 = −e3, e1, α(e3) = −e3; 4 : e1 ∗ e1 = e1, A(cid:48)3 e1 ∗ e2 = e2, e1, α(e2) = e2; 5 : e1 ∗ e1 = e1, A(cid:48)3 e1, α(e2) = −e2; e1 ∗ e2 = −e2, e2 ∗ e1 = e2, e2 ∗ e2 = e1 + e2, e3 ∗ e3 = e3, α(e1) = e2 ∗ e1 = −e2, e2 ∗ e2 = e1, e3 ∗ e3 = e3, α(e1) = 18 AHMED ZAHARI AND ABDENACER MAKHLOUF 6 : e1 ∗ e1 = e1, A(cid:48)3 e1 ∗ e2 = e2, e1 ∗ e3 = e3, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e2 ∗ e3 = e3, e3 ∗ e1 = e3, e3 ∗ e2 = e3, e3 ∗ e3 = e2 + e3, α(e1) = e1, α(e2) = e2, α(e3) = e3; e1 ∗ e2 = e2, e1 ∗ e3 = −e3, 7 : e1 ∗ e1 = e1, A(cid:48)3 e3, e3 ∗ e1 = −e3, e3 ∗ e2 = e3, e3 ∗ e3 = e2, α(e1) = e1, α(e2) = e2, α(e3) = −e3; e2 ∗ e1 = e2, e2 ∗ e2 = −e2, e2 ∗ e3 = 8 : e1 ∗ e1 = e1, e1 ∗ e2 = −e2, e1 ∗ e3 = e3, e2 ∗ e1 = −e2, e2 ∗ e2 = e3, e2 ∗ e3 = e2, e3 ∗ e1 = A(cid:48)3 e3, e3 ∗ e2 = e2, e3 ∗ e3 = −e3, α(e1) = e1, α(e2) = −e2, α(e3) = e3; e1 ∗ e2 = ae2, e1 ∗ e3 = e3, e2 ∗ e1 = ae2, e3 ∗ e1 = e3, e3 ∗ e3 = 9 : e1 ∗ e1 = e1, A(cid:48)3 e3, α(e1) = e1, α(e2) = ae2, α(e3) = e3; ae2, α(e3) = −e3; e1 ∗ e2 = ae2, 11 : e1 ∗ e1 = e1, A(cid:48)3 a2e3, α(e1) = e1, α(e2) = ae2, α(e3) = a2e3; 10 : e1 ∗ e1 = e1, e1 ∗ e2 = ae2, e1 ∗ e3 = −e3, e2 ∗ e1 = ae2, e3 ∗ e1 = −e3, α(e1) = e1, α(e2) = A(cid:48)3 e1 ∗ e3 = a2e3, e2 ∗ e1 = ae2, e2 ∗ e2 = e3, e3 ∗ e1 = 12 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e1 ∗ e3 = be3, e2 ∗ e1 = e2, e2 ∗ e2 = 1 A(cid:48)3 b e2, e2 ∗ e3 = e3, e3 ∗ e1 = be3, α(e1) = e1, α(e2) = e2, α(e3) = be3; 13 : e1 ∗ e1 = e1, e1 ∗ e2 = −e2, e1 ∗ e3 = be3, e2 ∗ e1 = −e2, e3 ∗ e1 = be3, α(e1) = e1, α(e2) = A(cid:48)3 −e2, α(e3) = be3; 14 : e1∗e1 = e1, e1∗e2 = b2e2, e1∗e3 = be3, A(cid:48)3 e2∗e1 = b2e2, e3∗e1 = be3, e3∗e3 = e2, α(e1) = e1, α(e2) = b2e2, α(e3) = be3; 15 : e1 ∗ e1 = e1, e1 ∗ e2 = ae2, e1 ∗ e3 = be3, e2 ∗ e1 = ae2, e3 ∗ e1 = be3, α(e1) = e1, α(e2) = A(cid:48)3 ae2, α(e3) = be3. Property 3.13. The unital Hom-associative algebras A(cid:48)3 are of associative type. 6, A(cid:48)3 7, A(cid:48)3 8, A(cid:48)3 9, A(cid:48)3 10, A(cid:48)3 11, A(cid:48)3 12, A(cid:48)3 13, A(cid:48)3 14, A(cid:48)3 15 Proof. Indeed, we set in the following, the corresponding associative algebras : A(cid:48)3 A(cid:48)3 6 :e1· e1 = e1, e1· e2 = e2, e1· e3 = e3, e2· e1 = e2 e2· e2 = e2, e2· e3 = e3, e3· e1 = e3, e3· e2 = e3, e3 · e3 = e2 + e3; 7 :e1 · e1 = e1, e1 · e2 = e2, e1 · e3 = e3, e3 · e2 = −e3, e2 · e1 = e2 e2 · e2 = −e2, e2 · e3 = −e3, e3 · e1 = e3, e3 · e3 = e2; 19 A(cid:48)3 8 :e1· e1 = e1, e1· e2 = e2, e1· e3 = e3, e2· e1 = e2 e2· e2 = e3, e2· e3 = e2, e3· e1 = e3, e3· e2 = −e2, e3 · e3 = −e3; 9 :e1 · e1 = e1, A(cid:48)3 10 :e1 · e1 = e1, A(cid:48)3 11 :e1 · e1 = e1, A(cid:48)3 12 :e1·e1 = e1, A(cid:48)3 14 :e1 · e1 = e1, A(cid:48)3 e3 · e1 = e3, e3 · e3 = e3; e3 · e1 = e3; e2 · e2 = e3, e1 · e3 = e3, e1 · e3 = e3, e1 · e3 = e3, e1·e3 = e3, e1 · e3 = e3, e1 · e2 = e2, e1 · e2 = e2, e1 · e2 = e2, e1·e2 = e2, e1 · e2 = e2, e2 · e1 = e2, e2 · e1 = e2, e2 · e1 = e2 e3 · e1 = e3; e2·e1 = e2 e2·e2 = e2, e2·e3 = e3, e2 · e1 = e2, e3 · e1 = e3, e3 · e3 = e2. e3·e1 = e3; (cid:3) Remark 3.14. It turns out that A(cid:48)3 algebra. 1, A(cid:48)3 2, A(cid:48)3 3, A(cid:48)3 4, A(cid:48)3 5 cannot be obtained by twisting an associative 4. Derivations of Hom-associative algebras Let (A, µ, α) be a multiplicative Hom-associative algebra. For any nonnegative integer k, we denote by αk the k-times composition of α, i.e αk = α ◦ ··· ◦ α (k-times). In particular, α0 = id and α1 = α. Definition 4.1. For any non-negative integer k, a linear map D : A −→ A is called an αk-derivation of a Hom-associative (A, µ, α), if (4.1) and (4.2) D ◦ α = α ◦ D D ◦ µ(f, g) = µ(D(f ), αk(g)) + µ(αk(f ), D(g)). Denote by Derαk (A) the set of αk-derivations of a multiplicative Hom-associative algebra (A, µ, α). For any f ∈ A satisfying α(f ) = f, we define Dk(f ) : A → A by Dk(f )(g) = µ(αk(g), f ), ∀g ∈ A. Then D(f ) is an αk+1-derivation, which we will call an Inner αk+1-derivation. In fact, we have Dk(f )(α(g)) = µ(αk+1(g), f ) = α(µ(αk(g), f ) = α ◦ Dk(f )(g), which implies that identity (4.1) in Definition 4.1 is satisfied. On the other hand, we have Dk(f )µ(g, h) = µ(αk(µ(g, h), f ) = µ(µ(αk(g), αk(h)), α(f )) = µ(αk+1(g), µ(αk(h), f )) + µ(µ(αk(g), f ), αk+1(h)) = µ(αk+1(g), Dk(f )(h)) + µ(Dk(f )(g), αk+1(h)). Therefore, Dk(f ) is an αk+1-derivation. The set of αk-derivations is denoted by Innerαk (A) , i.e. Innerαk (A) =(cid:8)µ(αk−1(•), ff ∈ A, α(f ) = f(cid:9) . (4.3) 20 AHMED ZAHARI AND ABDENACER MAKHLOUF For any D ∈ Derαk (A) and D(cid:48) ∈ Derαs (A), we define their commutator [D, D(cid:48)] as usual : [D, D(cid:48)] = D ◦ D(cid:48) − D(cid:48) ◦ D. Proposition 4.2. For any D ∈ Derαk (A) and D(cid:48) ∈ Derαs(A), we have [D, D(cid:48)] ∈ Derαk+s(A). Proof. For any f, g ∈ A, we have [D, D(cid:48)] µ(f, g) = D ◦ D(cid:48)µ(f, g) − D(cid:48) ◦ Dµ(f, g) = D(µ(D(cid:48)(f ), αs(g)) + µ(αs(f ), D(cid:48)(g))) − D(cid:48)(µ(D(f ), αk(g)) + µ(αk(f ), D(g))) = µ(D ◦ D(cid:48)(f ), αk+s(g)) + µ(αk ◦ D(cid:48)(f ), D ◦ αs(g)) + µ(D ◦ αs(f ), αk ◦ D(cid:48)(g)) + µ(αk+s(f ), D ◦ D(cid:48)(g)) −µ(D(cid:48) ◦ D(f ), αk+s(g)) − µ(αs ◦ D(f ), D(cid:48) ◦ αk(g)) − µ(D(cid:48) ◦ αk(f ), αs ◦ D(g)) − µ(αk+s(f ), D(cid:48) ◦ D(g)). Since D and D(cid:48) satisfy D◦α = α◦D, D(cid:48)◦α = α◦D(cid:48), we obtain αk◦D(cid:48) = D(cid:48)◦αk, D◦αs = αs◦D. Therefore, we have [D, D(cid:48)] µ(f, g) = µ(αk+s(f ), [D, D(cid:48)] (g)) + µ([D, D(cid:48)] (f ), αk+s(g)). Furthermore, it is straightforward to see that [D, D(cid:48)] ◦ α = D ◦ D(cid:48) ◦ α − D(cid:48) ◦ D ◦ α = α ◦ D ◦ D(cid:48) − α ◦ D(cid:48) ◦ D = α ◦ [D, D(cid:48)] , which yields that [D, D(cid:48)] ∈ Derαk+s (A). (cid:3) 5. Cohomology of Hom-associative algebras. In this section, we deal with a cochain complex that defines a cohomology of multiplicative Hom-associative algebras and then compute the cohomology groups of Hom-associative algebras obtained in the 2-dimensional and 3-dimensional classifications. Let (A, µ, α) be a Hom-associative algebra, for n ≥ 1 we define a K-vector space C n n-cochains as follows : ϕ ∈ C n Hom(A, A) is a n-linear map ϕ : An → A satisfying Hom(A, A) of α ◦ ϕ(x0,··· , xn−1) = ϕ(α(x0), α(x1),··· α(xn−1)) for all x0, x1,··· , xn−1 ∈ A. +(cid:80)n Definition 5.1. We call, for n ≥ 1, n-coboundary operator of a Hom-associative algebra (A, µ, α) the linear map δn (5.1) Homϕ(x0, x1, . . . , xn) = µ(αn−1(x0), ϕ(x1, x2, . . . , xn)) δn Hom(A, A) → C n+1 Hom(A, A) defined by Hom : C n k=1(−1)kϕ(α(x0), α(x1), . . . , α(xk−2), µ(xk−1, xk), α(xk+1), . . . , α(xn)) + (−1)n+1µ(ϕ(x0, x1, . . . , xn−1), αn−1(xn)). The space of n-cocycles is defined by Z n space of n-coboundaries is defined by Bn nth cohomology group of the Hom-associative algebra A the quotient H n Hom(A, A) = {ϕ ∈ C n Hom(A, A) =(cid:8)φ ∈ δn−1 Homϕ : ϕ ∈ C n−1(A, A)(cid:9) . We call the Homϕ = 0} , and the Hom(A, A) : δn Hom(A, A) = Zn Hom(A,A) Hom(A,A) . Bn In particular, a 2-coboundary operator of Hom-associative algebra A is given by the map 21 δ2 Hom : C 2 Hom(A, A) → C 3 Hom(A, A), ϕ (cid:55)→ δ2 Homϕ defined by Homϕ(x, y, z) = ϕ(α(x), µ(y, z)) − ϕ(µ(x, y), α(z)) + µ(α(x), ϕ(y, z)) − µ(ϕ(x, y), α(z)). δ2 The conditions δ2 In order to compute the second cohomology group, we set for a 2-cochain ϕ, ϕ(ei, ej, ek) = f k ijek. Homϕ(ei, ej, ek) = 0 and α ◦ ϕ(ei, ej) = ϕ(α(ei), α(ej)) translate to the following (cid:80)n q=1(apiCq pq) = 0, i, j, k, r = 1, . . . , n. ijaqkCr ijaqkf r jkCr jkf r p=1 ij −(cid:80)n (cid:80)n pq − Cp p=1 aspf p p=1 q=1 apiaqjf s pq + apif q pq = 0, pq − f p i, j, s = 1, . . . , n. system (cid:80)n (cid:80)n Recall that δ1f (ei, ej) = f (ei).ej − f (ei.ej) + ei.f (ej). Remark 5.2. The following groups correspond in Deformation theory to the space of obstructions to extend a deformation of order p to a deformation of order p + 1. A 3-coboundary operator of Hom-associative algebra A is given by a map Hom : C 3(A, A) → C 4(A, A), ψ (cid:55)→ δ3 δ3 Homψ defined as Homψ(x, y, z, w) = µ(α2(x), ψ(y, z, w)) − ψ(µ(x, y), α(z), α(w)) δ3 +ψ(α(x), µ(y, z), α(w)) − ψ(α(x), α(y), µ(z, w)) + µ(ψ(x, y, z)), α2(w)). For the computations, we set, for a 3-cochain ψ, ψ(ei, ej, ek, es) = ϕs 0 and α ◦ ψ(ei, ej, ek) = ψ(α(ei), α(ej), α(ek)) translate to the following system : ijkes. Conditions δ3 Homψ(ei, ej, ek, es) =  (cid:80)n (cid:80)n (cid:80)n (cid:80)n ijkals −(cid:80)n ijkaqlarqCs s=1 ϕs +ϕp p=1 q=1 r=1(apiarpϕq pr) = 0, ijaqkarlϕs rq − Cp jklCs (cid:80)n i, j, k, r, s = 1,··· , n. s=1 apiaqjaskϕl (cid:80)n q=1 p=1 pqr + apiCq jkϕs pqr − apiaqjCr klϕs pqr pqs = 0, i, j, k, l = 1, = ··· , n. The cohomology class is given by solving the equation ψ = δ2 5.1. Cohomology and Obstructions in HAss2. Cohomology in HAss2. In the following, we compute the 2-cocycles Z 2 and the 2-cohomology group H 2 and then Z 3 and H 3 for the 2-dimensional and 3-dimensional Hom-associative algebras provided in the classification. We do write only non-trivial images of basis elements. Homϕ, where ϕ is a 2-cochain. (1) For A2 (2) For A2 4, A2 8 and A2 1, A2 2, the Z 2 is 1-dimensional generated by the 2-cocycle defined as : ϕ2(e2, e2) = e2. 9, the Z 2 is 0-dimensional. Thus, we have H 2 = (cid:104)0(cid:105). Thus, we have H 2 = (cid:104)0(cid:105). 22 AHMED ZAHARI AND ABDENACER MAKHLOUF (3) For A2 3, the Z 2 is 3-dimensional generated by the 2-cocycle generators : ϕ1(e1, e2) = e2, ϕ2(e2, e1) = e2, ϕ3(e2, e2) = e2. Thus, we have H 2 = (cid:104)ϕ2, ϕ3(cid:105). (4) For A2 5, the Z 2 is 1-dimensional generated by the 2-cocycle generators : ϕ1(e1, e1) = e1. Thus, we have H 2 = (cid:104)0(cid:105). (5) For A2 6 the Z 2 is 2-dimensional generated by the 2-cocycle generators : ϕ1(e1, e1) = e2, ϕ2(e1, e2) = e2, ϕ2(e2, e1) = e2. Thus, we have H 2 = (cid:104)ϕ2(cid:105). (6) For A2 7, the Z 2 is 2-dimensional generated by the 2-cocycle generators : ϕ1(e1, e2) = e1, ϕ2(e2, e1) = e1. Thus, we have H 2 = (cid:104)ϕ1, ϕ2(cid:105). Obstructions spaces of HAss2. (1) For A2 (2) For A2 8, A2 4, A2 9, the Z 3 is 0-dimensional. Thus, we have H 3 = (cid:104)0(cid:105) . 1, A2 2, the Z 3 is 4-dimensional generated by the 3-cocycle generators : ψ1(e1, e2, e1) = e2, ψ1(e1, e2, e2) = −e2, ψ2(e2, e1, e2) = e2, ψ3(e2, e2, e1) = e2, ψ4(e2, e2, e2) = e2. Thus, we have H 3 = (cid:104)ψ1, ψ2, ψ3, ψ4(cid:105) . (3) For A2 3, the Z 3 is 6-dimensional generated by the 3-cocycle generators : ψ1(e1, e2, e1) = e2, ψ3(e2, e1, e1) = e2, ψ5(e2, e2, e1) = e2, ψ2(e1, e2, e2) = e2, ψ4(e2, e1, e2) = e2, ψ6(e2, e2, e2) = e2. Thus, we have H 3 = (cid:104)ψ1,··· , ψ6(cid:105) . (4) For A2 5, the Z 3 is 4-dimensional generated by the 3-cocycle generators : ψ1(e1, e1, e1) = e1, ψ2(e1, e1, e2) = e1, ψ3(e1, e2, e1) = e1, ψ4(e2, e1, e1) = e1, ψ4(e2, e1, e2) = −e1. Thus, we have H 3 = (cid:104)ψ1, ψ3, ψ4(cid:105) . (5) For A2 6, the Z 3 is 3-dimensional generated by the 3-cocycle generators : ψ1(e1, e1, e1) = e1, ψ1(e2, e1, e1) = 2e2, ψ2(e1, e1, e1) = e2, (cid:104)ψ1, ϕ2, ϕ3(cid:105) . ψ3(e1, e1, e2) = e2. ψ3(e2, e1, e1) = −e2, Thus, we have H 3 = (6) For A2 7, the Z 3 is 2-dimensional generated by the 3-cocycle generators : ψ1(e1, e2, e2) = e1, ψ1(e2, e2, e2) = 1 2 e2, (cid:104)ψ1, ψ2(cid:105) . ψ2(e2, e2, e1) = e1, ψ2(e2, e2, e2) = 1 2 e2. Thus, we have H 3 = 5.2. Cohomology and Obstructions in HAss3. Cohomology in HAss3. (1) For A3 (2) For A3 1, the Z 2 is 12-dimensional.Thus, we have H 2 = (cid:104)ϕ1, . . . , ϕ12(cid:105) . 2, the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ1(e3, e3) = e3. Thus, we have H 2 = (cid:104)ϕ1(cid:105) . (3) For A3 (4) For A3 9, the Z 2 is 0-dimensional. Thus, we have H 2 = (cid:104){0}(cid:105) . 3,A3 4, the Z 2 is 7-dimensional. Thus, we have H 2 = (cid:104)ϕ1, . . . , ϕ7(cid:105) . 23 (5) For A3 5, the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ1(e3, e3) = e3. Thus, we have H 2 = (cid:104)0(cid:105) . (6) For A3 6, the Z 2 is 3-dimensional generated by the 2-cocycle generators ϕ1(e1, e2) = e1, ϕ2(e2, e1) = e1, ϕ3(e3, e3) = e3. Thus, we have H 2 = (cid:104)ϕ3(cid:105) . (7) For A3 7 , the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ1(e1, e2) = e1. Thus, we have H 2 = (cid:104)ϕ1(cid:105) . (8) For A3 8, the Z 2 is 2-dimensional generated by the 2-cocycle generators ϕ1(e1, e1) = e3, ϕ2(e1, e2) = e3, ϕ2(e2, e1) = −e3. Thus, we have H 2 = (cid:104)ϕ1, ϕ2(cid:105) . (9) For A3 10, the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ1(e3, e3) = e1. Thus, we have H 2 = (cid:104)0(cid:105) . (10) For A3 11, the Z 2 is 2-dimensional generated by the 2-cocycle generators ϕ1(e1, e3) = e1, ϕ2(e3, e1) = e1. Thus, we have H 2 = (cid:104)ϕ2(cid:105) . (11) For A3 12 , the Z 2 is 0-dimensional. Thus, we have H 2 = (cid:104)0(cid:105) . Obstructions spaces of HAss3. (1) For A3 (2) For A3 (3) For A3 1, the Z 3 is 42-dimensional and we have H 3 = (cid:104)ψ1,··· , ψ42(cid:105)\(cid:104)ψ31, ψ32, ψ37, ψ38(cid:105). 2, the Z 3 is 9-dimensional and we have H 3 = (cid:104)ψ1,··· , ψ9(cid:105) . 3, the Z 3 is 0-dimensional generated by the 3-cocycle generators. Thus, we have H 3 = (cid:104){0}(cid:105) . (4) For A3 (5) For A3 (6) For A3 (7) For A3 (8) For A3 (9) For A3 4, the Z 3 is 2-dimensional and we have H 3 = (cid:104)ψ1, . . . , 2(cid:105) 5, the Z 3 is 11-dimensional and we have H 3 = (cid:104)ψ1,··· , ψ11(cid:105). 6, the Z 3 is 18-dimensional and we have H 3 = (cid:104)ψ1, ψ4, ψ5, ψ6, ψ7, ψ8, ψ9, ψ10, ψ14, ψ18(cid:105). 7, the Z 3 is 14-dimensional and we have H 3 = (cid:104)ψ1,··· , ψ14(cid:105). 8, the Z 3 is 7-dimensional and we have H 3 = (cid:104)ψ1,··· , ψ7(cid:105) . 9, the Z 3 is 5-dimensional generated by the following 3-cocycles generators : ψ1(e1, e3, e3) = e1, ψ1(e2, e3, e3) = e2, ψ2(e2, e3, e3) = e1, ψ3(e3, e2, e3) = e1, ψ4(e3, e3, e2) = e2, ψ4(e3, e3, e1) = e1 ψ5(e3, e3, e2) = e1. Thus, we have H 3 = (cid:104)ψ1,··· , ψ5(cid:105). (10) For A3 10, the Z 3 is 7-dimensional generated by the 3-cocycle generators : ψ1(e1, e2, e2) = e1, ψ1(e2, e2, e1) = −e1, ψ1(e2, e2, e3) = −e3, ψ1(e3, e2, e2) = e3, (cid:104)ψ1,··· , ψ4(cid:105). ψ2(e1, e3, e3) = e1, ψ3(e2, e2, e2) = e1, ψ2(e2, e3, e3) = e2, ψ2(e3, e3, e1) = −e1, ψ2(e3, e3, e2) = −e2, ψ4(e2, e3, e3) = e1, ψ5(e3, e2, e3) = e1, ψ6(e3, e3, e2) = e1. Thus, we have H 3 = (11) For A3 (12) For A3 11, the Z 3 is 19-dimensional and we have H 3 = (cid:104)ψ1, . . . , ψ19(cid:105)\(cid:104)ψ4, ψ10, ψ13(cid:105) . 12, the Z 3 is 7-dimensional and we have H 3 = (cid:104)ψ1,··· , ψ5(cid:105). 24 AHMED ZAHARI AND ABDENACER MAKHLOUF 5.3. Cohomology of associative type algebras in HAssn. We compute the third cohomology of the associative algebras corresponding to Hom-associative algebras of associative type. The coboundary operator may be obtained from the coboundary operator of Hom-associative algebras by taking α equals to the identity map. We have the following observation. Theorem 5.3. Let (A, µ, α) be a Hom-associative algebra of associative type where µ = αµ and (A, µ(cid:48)) is an associative algebra. Let ϕ(cid:48) be a n-cocycle with respect to Hochschild cohomology of If ϕ(cid:48) satisfies αϕ(cid:48) = ϕ(cid:48) ◦ (α ⊗ α) then αϕ(cid:48) is a n-cocycle of (A, µ, α) with respect to (A, µ(cid:48)). Hom-type Hochschild cohomology. Proof. Let δn Ass ϕ(x0, . . . , xn) = µ(x0, ϕ(x1, x2, . . . , xn)) + n(cid:88) (−1)k ϕ(x0, . . . , xk−2, µ(xk−1, xk), xk+1, . . . , xn) +(−1)n+1 µ( ϕ(x0, . . . , xn−1), xn). k=1 If ϕ satisfies α ◦ ϕ(x0, . . . , xn−1) = ϕ(α(x0), . . . , α(xn−1)) (−1)kϕ(α(x0), . . . , α(xk−2), µ(xk−1, xk), α(xk+1), . . . , α(xn)) + (−1)n+1µ(ϕ(x0, . . . , xn−1), αn−1(xn)) = 0. + for x0, . . . , xn−1 ∈ A, by equation (5.1), we have µ(αn−1(x0), ϕ(x1, x2, . . . , xn)) n(cid:88) By multiplication α = (α1, . . . , αn), we obtain α ◦ ϕ is a n-cocycle for (A, αµ, α). 5.4. Cohomology and Obstructions in Ass2. Computation of cohomology in Ass2. k=1 (1) For A2 1,the Z 2 is 4-dimensional generated by the 2-cocycle generators : ϕ1(e1, e1) = e1, ϕ1(e2, e1) = e2, (cid:3) Thus, we have ϕ2(e1, e2) = −e1, ϕ2(e2, e1) = −e2, ϕ3(e2, e2) = e1, ϕ4(e2, e2) = e2. ϕ2(e1, e1) = e2, ϕ1(e1, e2) = e2, H 2 = (cid:104) ϕ1(cid:105). 4, the Z 2 is 4-dimensional generated by the 2-cocycle generators : ϕ1(e1, e1) = e1, ϕ1(e2, e1) = e2, (2) A2 ϕ2(e1, e1) = e2, ϕ1(e1, e2) = e2, (cid:104) ϕ3(cid:105). 6, the Z 2 is 4-dimensional generated by the 2-cocycle generators : ϕ1(e1, e1) = e1, ϕ3(e1, e2) = e1, ϕ4(e2, e2) = e2, (3) A2 ϕ2(e1, e1) = e2, (cid:104) ϕ1, ϕ3, ϕ4(cid:105). ϕ3(e2, e1) = e1, ϕ3(e2, e2) = e2, ϕ4(e2, e1) = e2 ϕ3(e2, e2) = e1, ϕ4(e2, e2) = e2. Thus, we have H 2 = Thus, we have H 2 = (4) A2 8, the Z 2 is 4-dimensional, the Z 2 is 4-dimensional generated by the 2-cocycle generators 25 : ϕ1(e1, e1) = e1, ϕ2(e1, e2) = e1, ϕ1(e2, e1) = e2, ϕ2(e2, e2) = e2. Thus, we have H 2 = (cid:104) ϕ1, ϕ2(cid:105). (5) For A2 9, the Z 2 is 4-dimensional generated by the 2-cocycle generators : ϕ1(e1, e1) = e1, ϕ2(e1, e2) = e1, ϕ1(e2, e1) = e2, ϕ2(e2, e2) = e2. Thus, we have H 2 = (cid:104) ϕ1, ϕ2(cid:105). Obstructions spaces of Ass2. (1) For A2 1 , the Z 3 is 4-dimensional generated by the 3-cocycle generators : ψ1(e1, e1, e2) = e1, ψ1(e1, e2, e2) = −e2, ψ1(e2, e1, e2) = e2, ψ1(e2, e2, e2) = e1, Thus, we have H 3 = ψ2(e1, e1, e2) = e2, ψ3(e2, e1, e1) = e1, ψ2(e1, e2, e2) = e1, ψ2(e2, e1, e2) = −e1, ψ2(e2, e2, e2) = e2, (cid:68) ψ1, . . . , ψ4 (cid:69) . ψ3(e2, e1, e2) = e2, ψ3(e2, e2, e2) = −e2 ψ3(e2, e2, e2) = e1, ψ4(e2, e1, e1) = e2, ψ4(e2, e1, e2) = −e1, ψ4(e2, e2, e1) = e1 ψ4(e2, e2, e2) = e2. (2) For A2 4, the Z 3 is 5-dimensional generated by the 3-cocycle generators : ψ2(e1, e1, e2) = e2, ψ3(e2, e1, e1) = e1, ψ3(e2, e1, e2) = −e2, ψ4(e2, e1, e2) = e2, ψ3(e2, e1, e2) = e2, ψ5(e2, e2, e2) = e2. Thus, we have H 3 = ψ1(e1, e1, e2) = e1, ψ1(e1, e2, e2) = −e2, ψ1(e2, e1, e2) = e2, (cid:68) ψ1,··· , ψ5 (cid:68) ψ1,··· , ψ5 (cid:69) . (cid:69) . (3) For A2 6, the Z 3 is 5-dimensional generated by the 3-cocycle generators : ψ1(e1, e1, e2) = e2, ψ2(e1, e2, e1) = e1, ψ3(e1, e2, e2) = e1, ψ3(e2, e2, e1) = −e1, ψ4(e2, e1, e2) = e2, ψ5(e2, e2, e2) = e2. Thus, we have H 3 = (4) For A2 8, the Z 3 is 5-dimensional generated by the 3-cocycle generators : ψ1(e1, e1, e1) = e1, ψ1(e1, e2, e1) = −e2, ψ1(e2, e1, e1) = e2, (cid:68) ψ1,··· , ψ5 (cid:69) . ψ2(e1, e1, e2) = e1, ψ2(e1, e2, e1) = e1, ψ2(e1, e2, e2) = −e2, ψ2(e2, e1, e2) = e2, ψ3(e1, e2, e2) = e1, ψ4(e2, e2, e1) = e1, ψ5(e2, e2, e2) = e1. Thus, we have H 3 = (5) For A2 9, the Z 3 is 5-dimensional generated by the 3-cocycle generators : ψ1(e1, e1, e1) = e1, ψ1(e1, e2, e1) = −e2, ψ1(e2, e1, e1) = e2, (cid:68) ψ1,··· , ψ5 (cid:69) . ψ2(e1, e1, e2) = e1, ψ2(e1, e2, e1) = e1, ψ2(e1, e2, e2) = −e2, ψ2(e2, e1, e2) = e2, ψ3(e1, e2, e2) = e1, ψ4(e2, e2, e1) = e1, ψ5(e2, e2, e2) = e1. Thus, we have H 3 = 26 AHMED ZAHARI AND ABDENACER MAKHLOUF 5.5. Cohomology and Obstructions in Ass3. Computation of cohomology in Ass3. (1) For A3 (2) For A3 (3) For A3 (4) For A3 (5) For A3 (6) For A3 3 the Z 2 is 9-dimensional and we have H 2 = (cid:104) ϕ1,··· , ϕ9(cid:105). 7 the Z 2 is 4-dimensional and we have H 2 = (cid:104) ϕ1,··· , ϕ4(cid:105). 8, the Z 2 is 8-dimensional and we have H 2 = (cid:104) ϕ1,··· , ϕ8(cid:105). 9, the Z 2 is 10-dimensional and we have H 2 = (cid:104) ϕ1,··· , ϕ10(cid:105)\(cid:104) ϕ3, ϕ6, ϕ8, ϕ9(cid:105). 10, the Z 2 is 10-dimensional and we have H 2 = (cid:104) ϕ1,··· , ϕ10(cid:105)\(cid:104) ϕ3, ϕ6, ϕ8, ϕ9(cid:105). 12, the Z 2 is 2-dimensional generated by the 2-cocycle generators : ϕ2(e3, e2) = −e1, ϕ2(e3, e3) = −e1. ϕ1(e2, e2) = 2e2, ϕ2(e2, e1) = e3, ϕ1(e3, e2) = e3, ϕ2(e2, e3) = e1 ϕ1(e1, e2) = e1, ϕ1(e2, e1) = e1, (cid:104) ϕ1, ϕ2(cid:105). We have H 2 = Obstructions spaces of Ass3. (1) For A3 3, the Z 3 is 18-dimensional generated by the 3-cocycle generators. (2) For A3 7, the Z 3 is 5-dimensional generated by the 3-cocycle generators. Thus, we have H 3 = Thus, we have H 3 = Thus, we have H 3 = Thus, we have H 3 = Thus, we have H 3 = . . (cid:68) ψ1, . . . , ψ18 (cid:69) (cid:69) (cid:68) ψ1,··· , ψ5 (cid:68) (cid:68) (cid:68) ψ1,··· , ψ24 ψ1,··· , ψ27 ψ1,··· , ψ23 (cid:69) (cid:69) (cid:69) . . . (3) For A3 8, the Z 3 is 24-dimensional generated by the 3-cocycle generators. (4) For A3 9, the Z 3 is 27-dimensional generated by the 3-cocycle generators. (5) For A3 10, the Z 3 is 23-dimensional generated by the 3-cocycle generators. (6) For A3 12, the Z 3 is 2-dimensional generated by the 3-cocycle generators : ψ1(e1, e2, e2) = e1, ψ1(e2, e2, e1) = −e1, ψ1(e3, e2, e2) = e3, (cid:68) (cid:69) . ψ1, ψ2 ψ2(e2, e1, e2) = e1, ψ2(e2, e1, e3) = e1, ψ2(e2, e2, e1) = e1, ψ2(e2, e2, e3) = e2, ψ2(e2, e3, e2) = −e3, ψ2(e2, e3, e3) = −e3, ψ2(e3, e2, e1) = e1. Thus, we have H 3 = 5.6. Cohomology and Obstructions in UHAss2. Cohomology in UHAss2. (1) For A(cid:48)2 (2) For A(cid:48)2 2 and A(cid:48)2 1 , A(cid:48)2 3 , the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ(e2, e2) = e2. We 4 the Z 2 is 0-dimensional and we have H 2 = (cid:104)0(cid:105). have H 2 = (cid:104)0(cid:105). Obstructions in UHAss2. (1) For A(cid:48)2 (2) For A2 2 and A(cid:48)2 1 , A(cid:48)2 3, the Z 3 is 4-dimensional generated by the 3-cocycle generators : 4 the Z 3 is 0-dimensional and we have H 2 = (cid:104)0(cid:105). ψ(cid:48) 1(e1, e2, e1) = e2, 1(e1, e2, e2) = −e2, ψ(cid:48) ψ(cid:48) 2(e2, e1, e2) = e2, ψ(cid:48) 3(e2, e2, e1) = e2, ψ(cid:48) 4(e2, e2, e2) = e2. Thus, we have H 3 = 27 (cid:68) ψ1, ψ2 (cid:69) . 5.7. Cohomology and Obstructions in UHAss3. Computation of cohomology in UHAss3. 1 , the Z 2 is 12-dimensional generated by the 2-cocycle generators : ϕ(cid:48) 3(e2, e1) = e2, 3(e3, e1) = −e2, ϕ(cid:48) ϕ(cid:48) 4(e2, e1) = e3, ϕ(cid:48) 4(e3, e1) = e3, ϕ(cid:48) 5(e2, e2) = e2, ϕ(cid:48) 6(e2, e2) = e3, ϕ(cid:48) 7(e2, e3) = e2, ϕ(cid:48) 8(e2, e3) = e3, ϕ(cid:48) 9(e3, e2) = e2, ϕ(cid:48) 10(e3, e2) = e2, ϕ(cid:48) 11(e3, e3) = e2, 12(e3, e3) = −e3. ϕ(cid:48) We have 2 , the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ(cid:48)(e2, e2) = e2. (1) For A(cid:48)3 ϕ(cid:48) 1(e1, e2) = e2, 1(e1, e3) = −e2, ϕ(cid:48) ϕ(cid:48) 2(e1, e2) = e3, 2(e1, e3) = −e3, ϕ(cid:48) H 2 = (cid:104)ϕ(cid:48) 12(cid:105). 1, . . . , ϕ(cid:48) (2) For A(cid:48)3 Thus, we have H 2 = (cid:104)ϕ(cid:48) 1(cid:105). Thus, we have H 2 = (cid:104)0(cid:105). Thus, we have H 2 = (cid:104)ϕ(cid:48) 1(cid:105). (6) For A(cid:48)3 Thus, we have H 2 = (cid:104)ϕ(cid:48) 1(cid:105). 10, A(cid:48)3 9 , A(cid:48)3 the 2-cocycle generators ϕ(cid:48) = 0. 6 , A(cid:48)3 7 , A(cid:48)3 8 , A(cid:48)3 (3) For A(cid:48)3 3 , the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ(cid:48)(e2, e2) = e2. (4) For A(cid:48)3 4 , the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ(cid:48)(e3, e3) = e3. (5) For A(cid:48)3 5 , the Z 2 is 1-dimensional generated by the 2-cocycle generators ϕ(cid:48)(e3, e3) = e2. 11, A(cid:48)3 12, A(cid:48)3 13, A(cid:48)3 14, A(cid:48)3 15 the Z 2 is 0-dimensional generated by Obstructions Spaces in UHAss3. (1) For A(cid:48)3 (2) For A(cid:48)3 (3) For A(cid:48)3 (4) For A(cid:48)3 (5) For A(cid:48)3 (6) For A(cid:48)3 1 , the Z 3 is 42-dimensional and we have H 3 = (cid:104)ψ1,··· , ψ42(cid:105)\(cid:104)ψ31, ψ32, ψ37, ψ38(cid:105). 2 , the Z 3 is 11-dimensional. Thus, we have H 3 = (cid:104)ψ(cid:48) 3 , the Z 3 is 11-dimensional. Thus, we have H 3 = (cid:104)ψ(cid:48) 4 , the Z 3 is 11-dimensional. Thus, we have H 3 = (cid:104)ψ(cid:48) 5 , the Z 3 is 11-dimensional. Thus, we have H 3 = (cid:104)ψ(cid:48) 6 ,A(cid:48)3 H 3 = (cid:104)0(cid:105). 1, . . . , ψ(cid:48) 1, . . . , ψ(cid:48) 1, . . . , ψ(cid:48) 1, . . . , ψ(cid:48) 11(cid:105)\(cid:104)ψ(cid:48) 11(cid:105)\(cid:104)ψ(cid:48) 11(cid:105)\(cid:104)ψ(cid:48) 11(cid:105)\(cid:104)ψ(cid:48) 5(cid:105). 4, ψ(cid:48) 7, ψ(cid:48) 2, ψ(cid:48) 8, ψ(cid:48) 3, ψ(cid:48) 10(cid:105). 9, ψ(cid:48) 15, the Z 3 is 0-dimensional. Thus, we have 8, ψ(cid:48) 10(cid:105). 9 ,A(cid:48)3 10,A(cid:48)3 12,A(cid:48)3 13,A(cid:48)3 7 ,A(cid:48)3 8 , A(cid:48)3 14, A(cid:48)3 11,A(cid:48)3 10(cid:105). 5.8. Cohomology and Obstructions in UAss2. 28 AHMED ZAHARI AND ABDENACER MAKHLOUF Computation of cohomology in UAss2. (1) For A(cid:48)2 1 , the Z 2 is 5-dimensional generated by the 2-cocycle generators : 1(e1, e1) = e1, 1(e1, e2) = e1, 1(e2, e1) = e2, ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) (cid:69) . (cid:68) ϕ(cid:48) (cid:68) ϕ(cid:48) 1, ϕ(cid:48) 2 (cid:69) . ϕ(cid:48) ϕ(cid:48) 2(e1, e1) = e1 + e2, 2(e2, e1) = e1 + e2, ϕ(cid:48) ϕ(cid:48) 3(e2, e2) = e1, 4(e2, e2) = e2. Thus, we have H 2 = 1, ϕ(cid:48) 2, ϕ(cid:48) 3 (2) For A(cid:48)2 2 , the Z 2 is 4-dimensional generated by the 2-cocycle generators : 1(e1, e1) = e1, 1(e1, e2) = e2, 1(e2, e1) = e2, ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) 2(e1, e1) = e2, 2(e1, e2) = e1, 2(e2, e1) = e2, ϕ(cid:48) ϕ(cid:48) 3(e2, e2) = e2, 4(e2, e2) = e2. Thus, we have H 2 = (3) For A(cid:48)2 4 , the Z 2 is 4-dimensional generated by the 2-cocycle generators : ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) Obstructions spaces in UAss2. 1(e1, e1) = e1, 1(e1, e2) = e2, 1(e2, e1) = e2, 2(e1, e1) = e2, 3(e2, e2) = e1, 4(e2, e2) = e2. Thus, we have H 2 = (cid:104)ϕ(cid:48) 3(cid:105). 2, ϕ(cid:48) ϕ(cid:48) ϕ(cid:48) (1) For A(cid:48)2 (cid:69) ψ(cid:48) (cid:68) ψ(cid:48) 1 , the Z 3 is 1-dimensional generated by the 3-cocycle generators : 1(e1, e2, e2) = e1, ψ(cid:48) 1(e2, e1, e2) = −e1, ψ(cid:48) 1(e2, e2, e2) = −e1 + e2. Thus, we have H 3 = . (2) For A(cid:48)2 2 , the Z 3 is 2-dimensional generated by the 3-cocycle generators : 1 1(e1, e1, e2) = e1 − e2, ψ(cid:48) 1(e1, e2, e2) = e1 − e2, ψ(cid:48) 1(e2, e1, e1) = −e1 − e2, ψ(cid:48) 1(e2, e2, e1) = −e1 + e2, ψ(cid:48) ψ(cid:48) 2(e2, e1, e2) = e1 + e2, 2(e2, e2, e2) = e1 − e2. ψ(cid:48) (3) For A(cid:48)2 4 , the Z 3 is 1-dimensional generated by the 3-cocycle generators : ψ(cid:48) 1(e1, e2, e2) = e1, 1(e2, e1, e2) = −e2. Thus, we have H 3 = ψ(cid:48) (cid:68) ψ(cid:48) (cid:69) . 1 Thus, we have H 3 = (cid:68) ψ(cid:48) (cid:69) . 1, ψ(cid:48) 2 (cid:68) ϕ(cid:48) (cid:68) ϕ(cid:48) (cid:68) ϕ(cid:48) (cid:68) ϕ(cid:48) 5.9. Cohomology and Obstructions in UAss3. Computation of cohomology in UAss3. 13 (1) For A(cid:48)3 (2) For A(cid:48)3 (3) For A(cid:48)3 (4) For A(cid:48)3 (5) For A(cid:48)3 (6) For A(cid:48)3 (7) For A(cid:48)3 (8) For A(cid:48)3 6 , the Z 2 is 13-dimensional. Thus, we have H 2 = 7 , the Z 2 is 12-dimensional. Thus, we have H 2 = . 8 , the Z 2 is 12-dimensional. Thus, we have H 2 = 5, ϕ(cid:48) 4, ϕ(cid:48) 9 , the Z 2 is 11-dimensional. Thus, we have H 2 = 10, the Z 2 is 12-dimensional. Thus, we have H 2 = (cid:104) ϕ4, . . . , ϕ12(cid:105). 11, the Z 2 is 10-dimensional. Thus, we have H 2 = (cid:104) ϕ4, . . . , ϕ10(cid:105). 12, the Z 2 is 10-dimensional generated. Thus, we have H 2 = (cid:104) ϕ1, . . . , ϕ10(cid:105)\(cid:104) ϕ3, ϕ5(cid:105). 14, the Z 2 is 11-dimensional. Thus, we have H 2 = (cid:104) ϕ3, ϕ4, ϕ5, ϕ7, ϕ8(cid:105). 1, . . . , ϕ(cid:48) 1, . . . , ϕ(cid:48) 1, . . . , ϕ(cid:48) 3, ϕ(cid:48) 1, ϕ(cid:48) 6, ϕ(cid:48) 7, ϕ(cid:48) (cid:69) 12 12 . . . 8 Obstructions spaces in UAss3. (1) For A(cid:48)3 6 , the Z 3 is 18-dimensional. Thus, we have H 3 = (cid:68) ψ(cid:48) 1, . . . , ψ(cid:48) 18 (cid:69) (cid:69) (cid:69) (cid:69) . (cid:68) ψ(cid:48) 1,··· , ψ(cid:48) 18 (cid:69) . 29 . (2) For A(cid:48)3 (3) For A(cid:48)3 (cid:69) (cid:68) ψ(cid:48) 1(e2, e1, e1) = −e2, ψ(cid:48) ψ(cid:48) 1(e1, e1, e2) = e2, 1(e3, e1, e1) = −e3, ψ(cid:48) ψ(cid:48) 1(e1, e1, e3) = e3, Thus, we have H 3 = 7 , the Z 3 is 18-dimensional. Thus, we have H 3 = . 8 , the Z 3 is 2-dimensional generated by the 3-cocycle generators : 2(e2, e3, e2) = −e3, ψ(cid:48) 2(e2, e3, e3) = −e2. ψ(cid:48) (cid:69) (cid:69) (cid:69) ψ(cid:48) 2(e2, e2, e2) = e2, ψ(cid:48) 2(e2, e2, e3) = e3. 9 , the Z 3 is 16-dimensional. Thus, we have H 3 = 10, the Z 3 is 19-dimensional. Thus, we have H 3 = 11, the Z 3 is 20-dimensional. Thus, we have H 3 = 12, the Z 3 is 20-dimensional. Thus, we have H 3 = 14, the Z 3 is 20-dimensional. Thus, we have H 3 = 1, . . . , ψ(cid:48) 1, . . . , ψ(cid:48) 1, . . . , ψ(cid:48) 1, . . . , ψ(cid:48) 1, . . . , ψ(cid:48) (cid:68) ψ(cid:48) (cid:68) ψ(cid:48) (cid:68) ψ(cid:48) (cid:68) ψ(cid:48) (cid:68) ψ(cid:48) (4) For A(cid:48)3 (5) For A(cid:48)3 (6) For A(cid:48)3 (7) For A(cid:48)3 (8) For A(cid:48)3 1, ψ(cid:48) 2 . . (cid:69) (cid:69) 20 18 19 20 . 16 . . . 6. Deformations and irreducible components of Hom-associative algebras In this section, we aim to discuss the geometric classification of HAssn and UHAssn for n = 2, 3. We use to this end one parameter formal deformation theory introduced first by Gerstenhaber for associative algebras and extended to Hom-associative algebras in [1, 10]. Definition 6.1. Let (A, µ, α) be a Hom-associative algebra. A formal deformation of the Hom- associative algebra A is given by a K[[t]]-bilinear map µt : A[[t]] × A[[t]] −→ A[[t]] of the form i≥0 tiµi where each µi is a K-bilinear-map µi : A × A → A (extended to be K[[t]]-bilinear) µt =(cid:80) and µ0 = µ such that hold for x, y, z ∈ A the following condition (6.1) µt(µt(x, y), α(z)) = µt(α(x), µt(y, z)) 1,t, α(cid:48) Suppose that (A [[t]] , µ1,t, α1,t) and (A [[t]] , µ(cid:48) 1,t) are Hom-associative deformations of the Hom-associative algebras (A, µ, α). They are said equivalent if there exists a formal isomorphism between them, i.e. a K [[t]]-linear map ϕt, compatible with both the deformed multiplications and tiϕi, where the ϕi are linear maps ϕi : A → A the deformed twisting maps, of the form ϕt = and ϕ0 = idA. Compatibility with the deformed multiplications means that ϕt ◦ µt = µ(cid:48) ◦ (ϕt ⊗ ϕt), compatibility to the twisting maps means ϕt ◦ αt = α(cid:48) ◦ ϕt. (cid:88) i≥0 Proposition 6.2. Let µ1,t = φ−1 ◦ µ2 ◦ (φ ⊗ φ) and α1,t = φ−1 ◦ α2 ◦ φ. Then if (A, µ2, α2) is Hom-associative then (A [[t]] , µ1,t, α1,t) is Hom-associative. 30 AHMED ZAHARI AND ABDENACER MAKHLOUF Proof. By straightforward computation, we have µ1,t(α1,t(x), µ1,t(y, z)) = φ−1µ2(φ(φ−1 ◦ α2 ◦ φ(x)), φφ−1 ◦ µ2(φ(y), φ(z))) = φ−1µ2(α2 ◦ φ(x), µ2(φ(y), φ(z))) = φ−1µ2(µ2(φ(x), φ(y)), α2 ◦ φ(z)) = φ−1µ2(φ ◦ φ−1(µ2(φ(x), φ(y))), φ ◦ φ−1α2φ(z))) = µ1,t(µ1,t(x, y), α1,t(z)). (cid:3) Definition 6.3. A Hom-associative algebra A is called formally rigid, if every formal deformation of A is trivial. It is called geometrically rigid, if its orbid ϑ(µ) is open in HAssn. Then ϑ(µ) is an irreducible component of HAssn. Remark 6.4. Any irreducible component C of HAssn containing A also contains all degenerations of A. Indeed, we have ϑ(µ) ⊂ C so that ϑ(µ) is contained in C, since C is closed. Proposition 6.5. The irreducible components of HAss2 are the Zariski closure of orbits of Hom- associative algebras Ω =(cid:8)A2 (cid:9). 3, A2 5 Irreducible Components HAss2 Proposition 6.6. The irreducible components of HAss3 are the Zariski closure of orbits of Hom- associative algebras Ω =(cid:8)A3 (cid:9). 2, A3 5, A3 9 Irreducible components of Ass3 Proposition 6.7. The irreducible components of UHAss2 are the Zariski closure of orbits of Hom- associative algebras Ω =(cid:8)A(cid:48)2 3 , A(cid:48)2 4 (cid:9). A23A22A42A25A3233AA35A37A3931 Irreducible components of UHAss2 Proposition 6.8. The irreducible components of UHAss3 are the Zariski closure of orbits of Hom- (cid:9). associative algebras Ω =(cid:8)A(cid:48)3 9 , A(cid:48)3 10, A(cid:48)3 11, A(cid:48)3 12, A(cid:48)3 13, A(cid:48)3 14, A(cid:48)3 15 Irreducible components of UHAss3 Proposition 6.9. The irreducible components of UAss3 are the Zariski closure of orbits of Hom- associative algebras Ω = . 10, A3 14 (cid:110) A3 (cid:111) A3'2A4'2A2'2A1'2'3A7'3A6'3A8'3A11'3A13'3A10'3A14'3A12'3A15'3A932 AHMED ZAHARI AND ABDENACER MAKHLOUF Irreducible components of Ass3 References [1] F. Ammar, Z. Ejbehi, A. Makhlouf, Cohomology and deformations of Hom-algebras, J. Lie Theory 21 (4) (2011) 813 -- 836. [2] A. Armour, H. Chen and Y. Zhang, Classification of 4-dimensional superalgebras, Comm. in Algebra 37 (2009), 3697 -- 3728. [3] M. Goze and A. Makhlouf, Classification and rigid associative algebras in low dimensions, Lois d'algèbres et variétés algébriques (Colmar, 1991), 5-22, Travaux en cours, 50 Hermann, Paris, (1996). [4] M. Goze, and A. Makhlouf, On the rigid complex associative algebra, Comm. Algebra 18 (12) (1990) 4031 -- 4046. [5] Y. Fregier, A. Gohr and S. Silvestrov, Unital algebras of Hom-associative type and surjective or injective twistings, J. Gen. Lie Theory Appl. Vol. 3 (4), (2009) 285 -- 295. [6] J.T. Hartwig, D. Larsson and S. Silvestrov, Deformations of Lie algebras using σ- derivations, J. Algebra 295 (2006), 314 -- 361. [7] D. Larsson and S. Silvestrov, Quasi-hom-Lie algebras, central extensions and 2-cocycle- like identities, J. Algebra 288 (2005), 321 -- 344. [8] X.X. Li, Structures of multiplicative Hom-Lie algebras, Advances in Mathematics (China), 43(6)(2014)817 -- 823. [9] A. Makhlouf, Algèbres associatives et calcul formel, Theoret. Comput. Sci. 187(1997), no. 1-2, 123 -- 145. [10] A. Makhlouf and S. Silvestrov, Notes on formal deformations of Hom-associative and Hom-Lie algebras, Forum Mathematicum, vol. 22 (4) (2010) 715 -- 759. [11] A. Makhlouf and S. Silvestrov, Hom-algebra structures, J. Gen. Lie Theory Appl. Vol.2 (2008), No.2, 51 -- 64. [12] A. Makhlouf, Degeneration, rigidity and irreducible components of Hopf algebras, Algebra Colloq. 12(2005), no.2, 241 -- 254 [13] Y. Sheng, Representations of hom-Lie algebras, Algebras and Representation Theory, 15 (6) (2012), 1081 -- 1098. [14] D. Yau, Hom-algebras and homology, J. Lie Theory 19 (2009), 409 -- 421. [15] D. Yau, Enveloping algebra of Hom-Lie algebras, J. Gen. Lie Theory Appl. 2 (2008) 95 -- 108. [16] X. Chen and W. Han, Classification of multiplicative simple Hom-Lie algebras, J. Lie Theory 21(4)(2015) '3A6~'3A7~'3A12~'3A14~'3A10~'3A9~'3A11~'3A8~Université de Haute Alsace, IRIMAS-département de Mathématiques, 6, rue des Frères Lumière F-68093 Mulhouse, France E-mail address: [email protected] Université de Haute Alsace, IRIMAS-département de Mathématiques, 6, rue des Frères Lumière F-68093 Mulhouse, France E-mail address: [email protected] 33
1506.06574	2	1506	2015-12-10T07:37:37	DG Poisson algebra and its universal enveloping algebra	[ "math.RA" ]	In this paper, we introduce the notions of differential graded (DG) Poisson algebra and DG Poisson module. Let $A$ be any DG Poisson algebra. We construct the universal enveloping algebra of $A$ explicitly, which is denoted by $A^{ue}$. We show that $A^{ue}$ has a natural DG algebra structure and it satisfies certain universal property. As a consequence of the universal property, it is proved that the category of DG Poisson modules over $A$ is isomorphic to the category of DG modules over $A^{ue}$. Furthermore, we prove that the notion of universal enveloping algebra $A^{ue}$ is well-behaved under opposite algebra and tensor product of DG Poisson algebras. Practical examples of DG Poisson algebras are given throughout the paper including those arising from differential geometry and homological algebra.	math.RA	math	DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG Abstract. In this paper, we introduce the notions of differential graded (DG) Poisson alge- bra and DG Poisson module. Let A be any DG Poisson algebra. We construct the universal enveloping algebra of A explicitly, which is denoted by Aue. We show that Aue has a nat- ural DG algebra structure and it satisfies certain universal property. As a consequence of the universal property, it is proved that the category of DG Poisson modules over A is isomorphic to the category of DG modules over Aue. Furthermore, we prove that the no- tion of universal enveloping algebra Aue is well-behaved under opposite algebra and tensor product of DG Poisson algebras. Practical examples of DG Poisson algebras are given throughout the paper including those arising from differential geometry and homological algebra. Introduction The Poisson bracket was originally introduced by French Mathematician Sim´eon Denis Poisson in search for integrals of motion in Hamiltonian mechanics. For Poisson algebras, many important generalizations have been obtained in both commutative and noncommu- tative settings recently: Poisson orders [1], graded Poisson algebras [6, 9], noncommuta- tive Leibniz-Poisson algebras [7], left-right noncommutative Poisson algebras [8], Poisson PI algebras [19], double Poisson algebras [26], noncommutative Poisson algebras [27], Novikov-Poisson algebras [28] and Quiver Poisson algebras [30]. One of the interesting aspects of Poisson algebras is the notion of Poisson universal enveloping algebra, which was first introduced by Oh in order to describe the category of Poisson modules [22]. Since then, many progresses have been made in various directions of studying Poisson universal enveloping algebras, such as universal derivations and automorphism groups [25], Poisson- Ore extensions [15], Poisson Hopf algebras [16], and deformation theory [29]. In this paper, our aim is to study Poisson algebras and their universal enveloping algebras in the differential graded setting. Roughly speaking, a DG Poisson algebra is a differential graded commutative algebra together with a Poisson bracket satisfying certain compatible conditions. The Poisson bracket in our definition is assumed to have arbitrary degree rather 2010 Mathematics Subject Classification. 16E45, 16S10, 17B35, 17B63. Key words and phrases. differential graded algebras, differential graded Hopf algebras, differential graded Lie algebras, differential graded Poisson algebras, universal enveloping algebras, monoidal category. 1 2 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG than zero, which seems to be more natural as it appears in various examples like Schouten- Nijenhuis bracket in differential geometry and Gerstenhaber-Lie bracket in homological algebra. There are many equivalent ways to define the universal enveloping algebra of a DG Poisson algebra, among which we choose to use the explicit generators and relations; see Definition 3.1 and Remark 3.4. Regarding the universal property, we succeed to gener- alize it to the DG setting in Proposition 3.3, which induces an equivalence of two module categories as Theorem 4.2. Moreover, by considering the opposite algebra and the tensor product of DG Poisson algebras, we can view the universal enveloping algebra as a tensor functor from the category of DG Poisson algebras to the category of DG-algebras, which is summarized in Theorem 4.5. Moreover, any commutative Poisson algebra gives rise to a Lie-Rinehart pair associated to its Kahler differential [15, §5.2], which might provide an alternative approach to the results in our paper once we add the DG structure. Differential graded algebras arise naturally in algebra, representation theory and alge- braic topology. For instance, Koszul complexes, endomorphism algebras, fibers of ring ho- momorphisms, singular chain and cochain algebras of topological spaces, and bar resolu- tions all admit natural differential graded algebra structures. Moreover, differential graded commutative algebra provides powerful techniques for proving theorems about modules over commutative rings [2]. On the other side, using deformation quantization theory [10, 12, 13], Poisson brackets can be used as a tool to study noncommutative algebras. For example, by considering the Poisson structure on the center of Weyl algebras, Belov and Kontsevich proved that the Jacobian Conjecture is stably equivalent to the Dixmier Conjecture [4]. Therefore, as a generalization of both structures, we expect more study of DG Poisson algebras for its own interests, and further applications in a wide range of mathematical fields. The paper is organized as follows. Some background materials are reviewed in Section 1. The definition of DG Poisson algebra is given in Section 2, and the universal enveloping algebra of a DG Poisson algebra is defined in Section 3. Basic properties of universal enveloping algebras are discussed in Section 4. Several further examples of DG Poisson algebras are provided in Section 5. In last Section 6, future projects related to this paper are discussed. Acknowledgments. The authors first want to give their sincere gratitudes to James Zhang for introducing them this project. They also want to thank Yanhong Bao, Jiwei He, Xue- feng Mao, Cris Negron and James Zhang for many valuable discussions and suggestions on this paper. Additional thanks are given to the anonymous referees for several sugges- tions that improved the exposition of the paper. The first author is partially supported by National Natural Science Foundation of China [11001245, 11271335 and 11101288]. DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA 3 1. Preliminaries Throughout we work over a base field k, all vector spaces and linear maps are over k. In what follows, an unadorned ⊗ means ⊗k and Hom means Homk. All gradings are referred to Z-gradings with index occurs in upper superscript regarding cohomology. All elements and homomorphisms will be homogeneous in the graded setting. We refer the reader to [11, 23] for the following well-known definitions. A graded (associative) algebra means a Z-graded algebra A = Li∈Z Ai with k ⊆ A0 and AiA j ⊆ Ai+ j for all i, j ∈ Z. Moreover, A is said to be graded commutative if ab = (−1)abba for all elements a, b ∈ A, where a, b denote the degree of a, b in A. A differential graded algebra (or DG-algebra, for short) is a graded algebra A together with a map dA : A → A of degree 1 such that (1) (differential) d2 A (2) (graded Leibniz rule) dA(ab) = dA(a)b + (−1)aadA(b) for all elements a, b ∈ A. = 0; and Note that the cohomology of a DG-algebra is always a graded algebra. Let A be a DG-algebra. A left differential graded (DG) module over A is a graded left A-module M = Li∈Z Mi together with a linear map dM : M → M of degree 1 such that = 0; and (1) d2 M (2) dM(a · m) = dA(a) · m + (−1)aa · dM(m) for all elements a ∈ A and m ∈ M, where · denotes the A-module action on M. Right DG A-modules and DG A-bimodules can be defined in the similar manner. The category of left DG A-modules (resp. right DG A-modules) will be denoted by DGrMod- A (resp. A-DGrMod). For any M, N ∈ DGrMod-A, let HomA(M, N) be the set of all A-module morphisms f : M → N of degree 0 such that f dM = dN f . A differential graded (DG) Lie algebra is a graded vector space L = Li∈Z Li together with a bilinear Lie bracket [−, −] : L ⊗ L → L of degree p and a differential d : L → L of degree 1 (d2 = 0) such that (1) (antisymmetry) [a, b] = −(−1)(a+p)(b+p)[b, a]; (2) (Jacobi identity) [a, [b, c]] = [[a, b], c] + (−1)(a+p)(b+p)[b, [a, c]]; and (3) d[a, b] = [d(a), b] + (−1)(a+p)[a, d(b)] for all elements a, b, c ∈ L. It is worth pointing out that in above definition, we assume that the Lie bracket has arbitrary degree p rather than p = 0 as usual. It applies to many cases, for example, the Gerstenhaber Lie bracket defined for the Hochschild cohomology for any associative algebra has degree 4 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG −1 (p = −1 and d = 0). Note that Jacobi identity of the Lie bracket may also be expressed in a more symmetrical form: for any a, b, c ∈ L, we have (−1)(a+p)(c+p)[a, [b, c]] + (−1)(b+p)(a+p)[b, [c, a]] + (−1)(c+p)(b+p)[c, [a, b]] = 0. Let L be a DG Lie algebra with Lie bracket of degree p. A left differential graded (DG) Lie module over L is a graded vector space M = Li∈Z Mi together with a differential dM : M → M of degree 1 (d2 = 0) and a bilinear map [−, −]M : L ⊗ M → M of degree p M such that (1) [[a, b]L, m]M = [a, [b, m]M]M + (−1)(a+p)(b+p)[b, [a, m]M]M; and (2) dM([a, m]M) = [dL(a), m]M + (−1)(a+p)[a, dM(m)]M for all elements a ∈ L and m ∈ M. Right DG Lie L-modules and DG Lie L-bimodules can be defined in the similar manner. The category of left DG Lie L-modules (resp. right DG Lie L-modules) will be denoted by DGrLie-L (resp. L-DGrLie). For any M, N ∈ DGrLie-A, let HomL(M, N) be the set of all Lie L-module morphisms f : M → N of degree 0 such that f dM = dN f . 2. DG Poisson algebras and modules In literature, a Poisson algebra usually means a commutative algebra together with a Lie bracket which is a biderivation [15, Definition 2.1]. The notion of DG Poisson algebra can be considered as a combination of commutative algebras and Lie algebras in the differential graded setting. In the following definition, we allow the Poisson bracket to have arbitrary degree p. Definition 2.1. A differential graded Poisson algebra (or DG Poisson algebra, for short) is a quadruple (A, ·, {−.−}, d), where A = Li∈Z Ai is a graded vector space such that (i) (A, ·, d) is a graded commutative DG-algebra with differential d : A → A of degree 1. (ii) (A, {−, −}, d) is a DG Lie algebra with Poisson bracket {−, −} : A ⊗A → A of degree p. (iii) (Poisson identity): {a, b·c} = {a, b}·c+(−1)(a+p)bb·{a, c} for all elements a, b, c ∈ A. Remark 2.2. A noncommutative version of a DG Poisson algebra can be made without "graded commutative" in Definition 2.1 (i). In the following, we will consider Poisson modules over a DG Poisson algebra A with Poisson bracket of degree p. In order to make compatible with the degree of the Poisson bracket of A, we additionally require that the bracket A ⊗ M → M in the Poisson module DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA 5 structure for any Poisson module M shares the same degree p with the degree of the Poisson bracket of A. Definition 2.3. A left differential graded (DG) Poisson module over A is a quadruple (M, ∗, {−, −}M, dM), where M = Li∈Z Mi is a graded vector space with differential dM : M → M of degree 1 such that (i) (M, ∗, dM) ∈ DGrMod −A via A-module action ∗ : A ⊗ M → M of degree 0. (ii) (M, {−.−}M, dM) ∈ DGrLie −A via Lie A-module action {−.−}M : A ⊗ M → M of degree p. (iii) {a, b ∗ m}M = {a, b}A ∗ m + (−1)(a+p)bb ∗ {a, m}M, and (iv) {a · b, m}M = a ∗ {b, m}M + (−1)abb ∗ {a, m}M for all elements a, b ∈ A and m ∈ M. Right DG Poisson A-modules and DG Poisson A-bimodules are defined in the similar manner. The category of left DG Poisson A-modules (resp. right DG Poisson A-modules) will be denoted by DGrP-A (resp. A-DGrP). For any M, N ∈ DGrP-A, let HomA(M, N) be the set of all morphisms f : M → N of degree 0, which are both A-module and Lie A-module morphisms satisfying f dM = dN f . Example 2.4. The following are some trivial examples of DG Poisson algebras. (i) Let (A, ·, {−, −}, d) be a DG Poisson algebra with Poisson bracket of degree 0. Then the degree 0 piece (A0, ·, {−, −}) is an ordinary Poisson algebra. Conversely, any Poisson algebra can be considered as a DG Poisson algebra concentrated in degree 0 with trivial differential. (ii) Graded commutative DG-algebras are DG Poisson algebras with trivial Poisson bracket. (iii) Graded Poisson algebras are DG Poisson algebras with trivial differential. (iv) DG Lie algebras are DG Poisson algebras with trivial product added by a copy of the base field k. Example 2.5. Let (V, dV) be a differential graded vector space with differential dV : V → V of degree 1. There are two natural ways to construct DG Poisson algebras from V. One is to take the Hom complex of V, denoted by Hom(V, V). It is easy to check that Hom(V, V) is a noncommutative DG-algebra with the product given by the composition, and the differential given by d( f ) = dV f − (−1) f f dV for any f ∈ Hom(V, V). Moreover, one sees that Hom(V, V) becomes a DG Poisson algebra, where the Poisson bracket is given by the graded commutator defined by [ f , g] := f g − (−1) f gg f for any f , g ∈ Hom(V, V). The other way is to consider the graded symmetric algebra S (V) over V such that S (V) := T (V)/( f ⊗ g − (−1) f gg ⊗ f ), 6 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG where T (V) is the tensor algebra over V, and f , g ∈ V. Therefore, S (V) becomes a DG Poisson algebra via (1) (differential) d( f ) = dV f − (−1) f f dV, and (2) (Poisson bracket) { f , g} := [ f , g] for any f , g ∈ V. 3. Universal enveloping algebras of DG Poisson algebras The universal enveloping algebra of an ordinary Poisson algebra is given in [22]. Our aim is to generalize the definition to the differential graded setting. Let A be a DG Poisson algebra with Poisson bracket {−, −} of degree p. Definition 3.1. Let MA = {Ma : a ∈ A} and HA = {Ha : a ∈ A} be two copies of the graded vector space A endowed with two linear isomorphisms M : A → MA sending a to Ma and H : A → HA sending a to Ha. The universal enveloping algebra Aue of A is defined to be the quotient algebra of the free algebra generated by MA and HA, subject to the following relations: (i) Mab = MaMb, (ii) H{a,b} = HaHb − (−1)(a+p)(b+p)HbHa, (iii) Hab = MaHb + (−)abMbHa, (iv) M{a,b} = HaMb − (−1)(a+p)b MbHa, (v) M1 = 1 for all elements a, b ∈ A. Lemma 3.2. The universal enveloping algebra Aue has a natural DG-algebra structure induced by A such that Ma = a, Ha = a + p and d(Ma) = Md(a), d(Ha) = Hd(a) for any a ∈ A. Proof. It is clear that all the relations (i)-(v) in Definition 3.1 are homogeneous. Then, Aue is a graded algebra. It remains to show that d preserves all the relations, which is routine to check, e.g., for (ii) we have d(H{a,b}) = Hd{a,b} = H{d(a),b} + (−1)(a+p)H{a,d(b)} = Hd(a)Hb − (−1)(d(a)+p)(b+p)HbHd(a) + (−1)(a+p)(HaHd(b) − (−1)(a+p)(d(b)+p))Hd(b)Ha) = (Hd(a)Hb + (−1)(a+p)HaHd(b)) − (−1)(a+p)(b+p)(Hd(b)Ha + (−1)(b+p)HbHd(a)) = (d(Ha)Hb + (−1)HaHad(Hb)) − (−1)(a+p)(b+p)(d(Hb)Ha + (−1)HbHbd(Ha)) = d(HaHb − (−1)(a+p)(b+p)HbHa). DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA This completes the proof. 7 (cid:3) The universal enveloping algebra of an ordinary Poisson algebra can be described by certain universal property; see [16, §1.2]. The same thing happens to any DG Poisson algebra A, which is determined by the similar universal property. We say a triple (B, f , g) has property P with respect to A (or P-triple, for short) if (P1) B is a DG-algebra and f : A → B is a DG-algebra map of degree 0; (P2) g : (A, {−, −}) → B is a DG Lie algebra map of degree p, where the Lie bracket on B is given by the graded commutator; (P3) f ({a, b}) = g(a) f (b) − (−1)(a+p)b f (b)g(a), and (P4) g(ab) = f (a)g(b) + (−1)ab f (b)g(a), for all a, b ∈ A. Note that in Definition 3.1, MA, HA induces two linear maps from A to Aue, where we abuse to keep the same notations as M, H : A → Aue. It is clear that the triple (Aue, M, H) has property P, which is universal in the following sense. Proposition 3.3. For any P-triple (B, f , g), there exists a unique DG-algebra map φ : Aue → B of degree 0 such that the following diagram A f ❄❄❄❄❄❄❄❄ g Aue ⑤ ⑤ ∃!φ ⑤ ~⑤ M H B bi-commutes, i.e., f = φM and g = φH. Proof. Suppose the map φ exists. We get φ(Ma) = f (a) and φ(Ha) = g(a) for all a ∈ A by using bi-commutativity f = φM and g = φH. In order to show that φ is well-defined on Aue, we need to show that φ preserves all the relations (i)-(v) in Definition 3.1. (P1) implies that φ(Mab) = f (ab) = f (a) f (b) = φ(Ma)φ(Mb) for any a, b ∈ A and φ(M1) = f (1) = 1 = φ(1). So φ preserves (i) and (v). Similarly, φ preserves (ii) because of (P2); (iii) is preserved by φ by (P4); and (P3) implies that (iv) is preserved by φ. Moreover, it is clear that φ commutes with the differentials of Aue and B for f and g are DG maps. Finally, the uniqueness of φ is obvious. This completes the proof. (cid:3) Remark 3.4. The other way to define the universal enveloping algebra of an ordinary Pois- son algebra is though smash product [20, Definition 4.1.3] modulo certain relations; see [30, §2]. We can carry the same idea for any DG Poisson algebra A. Consider (A, {−, −}) as a DG Lie algebra, and denote by U(A) its universal enveloping algebra. It is important to point out that U(A) is a differential graded Hopf algebra with respect to the total grading coming from the grading of A, where elements in A ⊂ U(A) are all primitive. It is routine / / ~ 8 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG to check that A becomes a differential graded U(A)-module algebra via h ·a := {h, a} for all elements h, a ∈ A. As a consequence, the universal enveloping algebra Aue is the quotient algebra of the smash product A#U(A), subject to the relations 1#(ab) − a#b − (−1)abb#a for all elements a, b ∈ A. 4. Some basic properties Recall in Example 2.5, let V be a graded vector space with differential dV of degree 1. Denote by Hom(V, V) all the homogenous maps from V to itself. which has a natural differential given by d( f ) = dV f − (−1) f f dV for any f ∈ Hom(V, V). Moreover, one sees that Hom(V, V) becomes a DG-algebra via composition of maps such that d( f g) = d( f )g + (−1) f f d(g) for any f , g ∈ Hom(V, V). Now, let A be a DG Poisson algebra with Poisson bracket {−, −}A of degree p. By Definition 2.3, any DG Poisson A-module structure on V requires two linear maps from A to Hom(V, V), i.e., one A-module action ∗ : A ⊗ V → V of degree 0 and another Lie A- module action {−.−}V : A⊗V → V of degree p satisfying conditions described in Definition 2.3 (i)-(iv). The following lemma is straightforward. Lemma 4.1. The DG vector space (V, dV) belongs to DGrP-A if and only if there exits a P-triple (Hom(V, V), f , g) such that the A-module and Lie A-module actions on V are given by f and g correspondingly. We will see that the notion of universal enveloping algebra for DG Poisson algebra is proper in its usual sense as there exists an equivalence of two module categories. Theorem 4.2. The category of left (resp. right) DG Poisson modules over A is equivalent to the category of left (resp. right) DG modules over Aue, i.e., DGrP −A ≡ DGrMod −Aue (resp. A − DGrP ≡ Aue − DGrMod). Proof. Let M ∈ DGrP-A. By Lemma 4.1, it gives us a P-triple (Hom(M, M), f , g). Then by Proposition 3.3, we get a unique DG-algebra map φ : Aue → Hom(M, M) of degree 0 making certain diagram bi-commute. Hence, we can view M as a DG Aue-module via φ. Moreover, it is direct to show that φ respects homomorphisms between any two DG Poisson A-modules. Hence, we have a functor Φ : DGrP-A → DGrMod-Aue induced by φ. Finally, using the canonical triple (Aue, M, H), we see that any M ∈ DGrMod-Aue can be viewed as a DG Poisson A-module by Lemma 4.1 again, which yields an inverse functor of Φ. (cid:3) In the following, we will use DGA to denote the category of all DG-algebras, and use DGPA[p] to denote the category of all DG Poison algebras whose Poisson brackets are DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA 9 of degree p along with standard morphisms. It is well-known that how to take opposite algebra and tensor product in DGA with application of Koszul signs. In DGPA[p], the opposite DG Poisson algebra of A is denoted by Aop, where Aop = A as DG-algebras and the Poisson bracket of Aop is given by {a, b}Aop := (−1)(a+p)(b+p){b, a}A = −{a.b}A, for all a, b ∈ Aop. Now, let B be another DG Poisson algebra with Poisson bracket {−, −}B of same degree p. The tensor product A ⊗ B in DGA is again a graded commutative DG-algebra, where the product of homogenous elements of degree d1 and d2 is twisted by (−1)d1d2. Moreover, it is straightforward to check that A ⊗ B becomes a DG Poisson algebra with Poisson bracket of degree p by setting (4.1) {a1 ⊗ b1, a2 ⊗ b2}A⊗B := (−1)(a2 +p)b1{a1, a2}A ⊗ b1b2 + (−1)(b1+p)a2a1a2 ⊗ {b1, b2}B for any ai ∈ A and bi ∈ B, i = 1, 2. Example 4.3. Let A be a DG Poisson algebra with Poisson bracket of degree p, and (M, ∗, {−, −}M, dM) a left DG Poisson module over A. Then (Mop, ∗op, {−, −}Mop, dMop) is a right DG Poisson module over the opposite DG Poisson algebra Aop, where (Mop, dMop) = (M, dM) as differential graded vector spaces, and m ∗op a := (−1)ama ∗ m, {m, a}Mop := (−1)(a+p)(m+p){a, m}M, for any a ∈ A, m ∈ M. Moreover, let B be another DG Poisson algebra with Poisson bracket of same degree p, and (N, ∗, {−, −}N, dN) a left DG Poisson module over B. Then, the tensor product M ⊗ N can be viewed as a left DG Poisson module over the DG Poisson tensor product A ⊗ B, where we use the obvious DG A ⊗ B-module structure on M ⊗ N, and set up {a ⊗ b, m ⊗ n}M⊗N := (−1)(m+p)b{a, m}M ⊗ (b ∗ n) + (−)m(b+p)(a ∗ m) ⊗ {b, n}N, for any a ∈ A, b ∈ B, m ∈ M, n ∈ N. Lemma 4.4. Let A, B be two DG Poisson algebras with Poisson brackets of degree p. Suppose there are two P-triples (C, f , g) and (D, j, k) with respect to A, B correspondingly. Then, there is a tensor P-triple T := (C ⊗ D, f ⊗ j, f ⊗ k + (−1)pBg ⊗ j) with respect to A ⊗ B. We use (−1)pBg ⊗ j to mean that ((−1)pBg ⊗ j)(a ⊗b) = (−1)pbg(a) ⊗ j(b) for any a ∈ A, b ∈ B. Proof. It is tedious to check that (P1)-(P4) all hold for the triple T defined above. We only check (P1) and (P3) here, and leave the rest to the readers. (P1): first of all, we show that 10 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG f ⊗ j : A ⊗ B → C ⊗ D is a graded algebra map of degree 0. For any ai ∈ A, bi ∈ B, i = 1, 2, we have ( f ⊗ j)[(a1 ⊗ b1)(a2 ⊗ b2)] = (−1)a2b1( f ⊗ j)(a1a2 ⊗ b1b2) = (−1)a2b1 f (a1a2) ⊗ j(b1b2) = (−1) f (a2) j(b1) f (a1) f (a2) ⊗ j(b1) j(b2) = ( f (a1) ⊗ j(b1))( f (a2) ⊗ j(b2)) = ( f ⊗ j)(a1 ⊗ b1)( f ⊗ j)(a2 ⊗ b2). Then, we have to show that f ⊗ j commutes the differentials. We have, for any a ∈ A, b ∈ B, ( f ⊗ j)dA⊗B(a ⊗ b) = ( f ⊗ j)(dA(a) ⊗ b + (−1)aa ⊗ dB(b)) = f dA(a) ⊗ j(b) + (−1)a f (a) ⊗ jdB(b) = dC f (a) ⊗ j(b) + (−1)a f (a) ⊗ dD j(b) = dC⊗D( f ⊗ j)(a ⊗ b). (P3): for the left side of the identity, by Equation (4.1), we have ( f ⊗ j)({a1 ⊗ b1, a2 ⊗ b2}A⊗B) =(−1)(a2 +p)b1 f ({a1, a2}A) ⊗ j(b1b2) + (−1)a2(b1+p) f (a1a2) ⊗ j({b1, b2}B) =(−1)(a2 +p)b1(g(a1) f (a2) − (−1)(a1 +p)a2 f (a2)g(a1)) ⊗ j(b1) j(b2)+ (−1)a2 (b1+p) f (a1) f (a2) ⊗ (k(b1) j(b2) − (−1)(b1+p)b2 j(b2)k(b1)) =(−1)a2 b1+pb1 g(a1) f (a2) ⊗ j(b1) j(b2) − (−1)a1a2+pb1+pa2+a2b1 f (a2)g(a1) ⊗ j(b1) j(b2) + (−1)a2b1+pa2 f (a1) f (a2) ⊗ k(b1) j(b2) − (−1)b1b2+pb2+pa2+a2 b1 f (a1) f (a2) ⊗ j(b2)k(b1). For the right side of the identity, let χ = f ⊗ k + (−1)pBg ⊗ j, we have χ(a1 ⊗ b1)( f ⊗ j)(a2 ⊗ b2) − (−1)(a1+b1+p)(a2 +b2)( f ⊗ j)(a2 ⊗ b2)χ(a1 ⊗ b1) =( f (a1) ⊗ k(b1) + (−1)pb1g(a1) ⊗ j(b1))( f (a2) ⊗ j(b2)) − (−1)(a1+b1+p)(a2+b2)( f (a2) ⊗ j(b2)) ( f (a1) ⊗ k(b1) + (−1)pb1g(a1) ⊗ j(b1)) =(−1) f (a2)k(b1) f (a1) f (a2) ⊗ k(b1) j(b2) + (−1)pb1 + f (a2) j(b1)g(a1) f (a2) ⊗ j(b1) j(b2) − (−1)(a1+b1+p)(a2 +b2)+ f (a1) j(b2) f (a2) f (a1) ⊗ j(b2)k(b1) − (−1)(a1+b1+p)(a2 +b2)+pb1+g(a1) j(b2) f (a2)g(a1) ⊗ j(b2) j(b1). Note that f (a2) f (a1) = (−1) f (a1) f (a2) f (a1) f (a2), j(b2) j(b1) = (−1) j(b1) j(b2) j(b1) j(b2), and f , j have degree 0, g, k have degree p. Replacing these things in the last equality, we will see that (P3) holds. (cid:3) DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA 11 Theorem 4.5. Let A, B be two DG Poisson algebras with Poisson brackets of the same degree p. Then we have (i) (Aop)ue (cid:27) (Aue)op. (ii) (A ⊗ B)ue (cid:27) Aue ⊗ Bue. (iii) (A ⊗ Aop)ue (cid:27) Aue ⊗ (Aue)op. Moreover, ue : DGPA[p] → DGA is a tensor functor. Proof. We only prove (ii) here. We can get (i) from the same fashion, and (iii) is a corollary of (i) and (ii). First of all, note that we have two canonical P-triples (Aue, MA, HA) and (Bue, MB, HB). By Lemma 4.4, there is a tensor P-triple: T := (Aue ⊗ Bue, MA ⊗ MB, MA ⊗ HB + (−1)pBHA ⊗ MB) with respect to A ⊗ B. It suffices to show that T has the universal property stated in Propo- sition 3.3. Then the uniqueness argument will imply (ii). Now, suppose (D, f , g) is any P-triple. We will proceed by completing the following diagram: A ⊗ B iA ②②②②②②②②② b❊❊❊❊❊❊❊❊❊ iB A f,g B MA,HA fA,gA ❄❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧⑧⑧ fB,gB MB,HB / D ∃!φ ~⑤ o❴ ❴ ❴ ❴ ❴ ❴ ❴ ❴ a❇ Aue ⊗ Bue / Ae ⑤ ⑤ ∃!φA ⑤ ❇ ∃!φB ❇ ❇ / Bue iAue %❏❏❏❏❏❏❏❏❏❏ 9tttttttttt iBue By the natural inclusion iA : A → A ⊗ B, it is clear that the triple (D, fA, gA) has property P with respect to A. Then there exists a unique DG-algebra map φA : Aue → D of degree 0 such that fA = φAMA and gA = φAHA. Similarly, there exists a unique DG-algebra map φB : Bue → D of degree 0 such that fB = φBMB and gB = φBHB. Once we establish the following lemma, by the universal property of the tensor product, we will have a unique DG-algebra map φ : Aue ⊗Bue → D of degree 0 making the above diagram bi-commutative. Finally, the uniqueness of φ is clear by tracking the diagram. (cid:3) Lemma 4.6. In the above diagram, for any x ∈ Aue, y ∈ Bue, we have φA(x)φB(y) = (−1)xyφB(y)φA(x). Proof. By abuse of language, we use the same notations M, H to denote the two linear maps from A and B to Aue and Bue correspondingly. Note that it suffices to consider x = Ma, Ha and y = Mb, Hb for any a ∈ A, b ∈ B. We only check for x = Ma and y = Mb here, and / ~ % / o ? b / a 9 12 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG leave the rest to the readers. We have φA(Ma)φB(Mb) = fA(a) fB(b) = f (a ⊗ 1) f (1 ⊗ b) = f (a ⊗ b) = (−1)ab f (1 ⊗ b) f (a ⊗ 1) = (−1)abφB(Mb)φA(Ma). (cid:3) 5. Further examples of DG Poisson algebras In this section, examples of DG Poisson algebras are provided arising from Lie theory, differential geometry, homological algebra and deformation theory. See more examples in [5, 9]. 5.1. Graded symmetric algebras of DG Lie algebras. Let (L, dL, [−, −]L) be a DG Lie algebra. As in Example 2.5, we use S (L) to denote the graded symmetric algebra over L, i.e., S (L) := T (L)/(a ⊗ b − (−1)abb ⊗ a), for any a, b ∈ L. The differential dL of L can be extended to the graded symmetric algebra S (L) such that S (L) becomes a graded commutative DG-algebra. Here, we consider the total grading on S (L) coming from the grading of L. Moreover, the Lie bracket on L also can be extended to a Poisson bracket on S (L) by Jacobi identity such that {a, b}S (L) := [a, b]L, for any a, b ∈ L. Hence, the graded symmetric algebra S (L) over any DG Lie algebra L has a natural DG Poisson algebra structure. Next, note that we can view L as a left DG Lie L-module via the Lie bracket [−, −]L. The graded semi-product L ⋊ L is constructed in the following way: (1) L ⋊ L = LL L as graded vector spaces, (2) Lie bracket [−, −]L⋊L is 0 on the Lie subalgebra L + 0 and equals [−, −]L from the pair of components 0 + L and L + 0 to L + 0, (3) the differential on L ⋊ L is given by dL component-wisely. We denote by U(L ⋊ L) the graded universal enveloping algebra of L ⋊ L, which is a DG-algebra. Therefore, we have the following isomorphism: S (L)ue (cid:27) U(L ⋊ L). The isomorphism Φ : S (L)ue → U(L ⋊ L) can be constructed explicitly by Definition 3.1 such that Φ(ML) = L + 0 and Φ(HL) = 0 + L in L ⋊ L. Moreover, S (L) has a differential graded Hopf structure which is compatible with the Poisson bracket. See [15, Proposition 6.3] for the trivial case when L is only a Lie algebra without DG structure. DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA 13 5.2. Gerstenhaber algebras. Consider DG Poisson algebras with trivial differential. In general, most of such examples appear to be in the context of Gerstenhaber algebra, which by definition is a graded commutative algebra G = Li∈Z Gi together with a Lie bracket [−, −] : G⊗G → G of degree −1 satisfying: (1) (antisymmetry) [a, b] = −(−1)(a−1)(b−1)[b, a], (2) (Poisson identity) [a, bc] = [a, b]c + (−1)(a−1)bb[a, c], (3) (Jacobi identity) [a, [b, c]] = [[a, b], c] + (−1)(a−1)(b−1)[b, [a, c]] for any a, b, c ∈ G. For instance, in homological algebra, Hochschild cohomology of any associated algebra together with the Gerstenhaber Lie bracket [18]; in differential geometry, the alternating multivector fields on a smooth manifold together with the Schouten-Nijenhuis bracket ex- tending the Lie bracket of vector fields [21], are all Gertenhaber algebras. Moreover, given an arbitrary Lie-Rinehart algebra (R, L) [24], consider the graded exterior R-algebra Λ∗ RL RL ex- over L. The Lie bracket [−.−] of L induces a Gerstenhaber algebra structure on Λ∗ plicitly given by, for any u = α1 ∧ · · · ∧ αl ∈ Λl R L, where α1, · · · , αn ∈ L, then RL and v = αl+1 ∧ · · · ∧ αn ∈ Λn−l [u, v] = (−1)u X (−1)( j+k)[α j, αk] ∧ α1 ∧ · · · bα j · · · bαk · · · ∧ αn. j≤l<k Suppose A is a DG Poisson algebra with Poisson bracket of degree −1. Then, it is clear that the cohomology ring H A is a Gerstenhaber algebra whose Lie bracket is induced by the Poisson bracket on the cohomology. In the other hand, let G = Li∈Z Gi be a Gerstenhaber algebra with a Lie bracket [−, −] of degree −1. Suppose there is some element α ∈ G2 satisfying [α, α] = 0. Hence we can define a map d : G → G of degree 1 such that d := [α, −]. If char k , 2, one sees that d2 = 0, hence we have a differential d on G. Moreover, it is straightforward to check that, for any a, b ∈ G, we have (1) d(ab) = d(a)b + (−1)aad(b), and (2) d[a, b] = [d(a), b] + (−1)a−1[a, d(b)]. Therefore, G together with d = [α, −] is a DG Poisson algebra. 5.3. Deformations of graded commutative DG-algebras. Let A = Li≥Z Ai be a graded commutative DG-algebra together with differential d of degree 1. Consider a graded de- formation of A over the ring k[[]], which contains a bimodule map ⋆ : A[[]] ⊗ A[[]] → A[[]] making A[[]] into a DG-algebra where deg = 0. We can write the product of two elements a, b ∈ A as a ⋆ b = ab + B1(a, b) + B2(a, b)2 + · · · + Bi(a, b)i + · · · , for bilinear maps Bi : A ⊗ A → A of degree 0. Suppose A[[]] is not graded commutative. Then there is some smallest integer m ≥ 1 such that Bm(a, b) , (−1)abBm(b, a) for some 14 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG a, b ∈ A. As a consequence, we can define a Poisson bracket associated to that deformation: for a, b ∈ A, we have {a, b} := {−m(a ⋆ b − (−1)abb ⋆ a)}=0 = Bm(a, b) − (−1)abBm(b, a). It is routine to check that this degree 0 bilinear map {−, −} : A ⊗ A → A makes A into a DG Poisson algebra. 6. Closing discussions We close this paper by proposing some questions to be considered in future projects: (1) In Xu's paper [27], a noncommutative Poisson structure on an associative algebra A was defined to be any cohomology class π ∈ H2(A, A) in the second Hochschild cohomology group of A satisfying [π, π] = 0, where [−, −] is the Gerstenhaber bracket on H•(A, A). Can we define noncommutative DG Poisson algebras using the same idea? (2) Let A be a DG Poisson algebra. One sees easily that there exists a natural graded Poisson algebra map fA : H(A)ue → H(Aue), where H means taking the cohomology ring. We ask when is fA an isomorphism? Moreover, we ask if f : A → B is a quasi- isomorphism of DG Poisson algebras, does f always induce a quasi-isomorphism between Aue and Bue? (3) The classical Beilinson-Ginzburg-Soergel Koszul duality states a derived equiva- lence between a Koszul algebra and its Koszul dual. Later, it is generalized to the DG setting by Lu-Palmieri-Wu-Zhang in [14] using A∞-algebras . Can we establish Koszul duality for certain DG Poisson algebras in terms of the derived equivalence of their DG Poisson modules? References [1] K.A. Brown and I. Gordon, Poisson orders, symplectic reflection algebras and representation theory, J. Reine Angew. Math., 559 (2003), 193-216. [2] K.A. Beck and S. Sather-Wagstaff, A somewhat gentle introduction to differential graded commutative algebra, preprint, arXiv:1307.0369v1. [3] A. A. Beilinson, V. A. Ginsburg and V. V. Schechtman, Koszul duality patterns in representation theory, J. Amer. Math. Soc., 9 (1996), no. 2, 473-527. [4] A. Belov and M. Kontsevich, The Jacobian conjecture is stably equivalent to the Dixmier conjecture, Mosc. Math. J., 7(2) (2007), 209-218. [5] J. V. Beltr´an and J. Monterde, Graded Poisson structures on the algebra of differential forms, Com- ment. Math. Helvetici, 70 (1995), 383-402. [6] A.J. Calder´on, On extended graded Poisson algebras, Linear Alg. Appl., 439 (2013), 879-892. DG POISSON ALGEBRA AND ITS UNIVERSAL ENVELOPING ALGEBRA 15 [7] J.M. Casas and T. Datuashvili, Noncommutative Leibniz-Poisson algebras, Comm. Algebra, 34(7) (2006), 2507-2530. [8] J.M. Casas, T. Datuashvili and M. Ladra, Left-right noncommutative Poisson algebras, Cent. Eur. J. Math., 12(1) (2014), 57-78. [9] A.S. Cattaneo, D. Fiorenza and R. Longoni, Graded Poisson algebras, Encyclopedia Math. Phy., (2006), 560-567. [10] P. Etingof and D. Kazhdan, Quantization of Poisson algebraic groups and Poisson homogeneous spaces, Sym´etries quantiques (Les Houches, 1995), 935-946, North-Holland, Amsterdam, 1998. [11] Y. F´elix, S Halperin and J.-C. Thomas, Rational Homotopy Theory, Springer-Verlag New York (2001). [12] J. Huebschmann, Poisson cohomology and quantization, J. Reine Angew. Math., 408 (1990) 57-113. [13] M. Kontsevich, Deformation quantization of Poisson manifolds, Lett. Math. Phys., 66(3) (2003), 157- 216. [14] D. M. Lu, J. H.Palmieri, Q. Wu, and J. J. Zhang, Koszul equivalences in A∞-algebras, New York J. Math., 14 (2008), 325-378. [15] J.-F. Lu, X. Wang and G. Zhuang, Universal enveloping algebras of Poisson Hopf algebras, J. Algebra, 426 (2015), 92-136. [16] J.-F. Lu, X. Wang and G. Zhuang, Universal enveloping algebras of Poisson Ore-extensions, Proc. Amer. Math. Soc., 143 (2015), no. 11, 4633-4645. [17] J.-F. Lu, X. Wang, G. Zhuang. A note on the duality between Poisson homology and cohomology, preprint, arXiv:1404.4819. [18] M. Gerstenhaber, The cohomology structure of an associative ring, Ann. Math., 2 (1963), 78:267-288. [19] S.P. Mishchenko, V.M. Petrogradsky and A. Regev, Poisson PI algebras, Trans. Amer. Math. Soc., 359(10) (2007), 4669-4694. [20] S. Montgomery, Hopf Algebras and Their Actions on Rings, CBMS Reg. Conf. Ser. Math. vol. 82 (1993), Amer. Math. Soc., Providence, RI. [21] A. Nijenhuis, Jacobi-type identities for bilinear differential concomitants of certain tensor fields I, Indagationes Math., 17 (1955), 390-403. [22] S.-Q. Oh, Poisson enveloping algebras, Comm. Algebra, 27 (1999), 2181-2186. [23] D. Quillen, Rational homotopy theory, Ann. Math., 2 (1969) 90:205-295. [24] G. S. Rinehart, Differential forms on general commutative algebras, Trans. Amer. Math. Soc., 108 (1963), 195-222. [25] U. Umirbaev, Universal enveloping algebras and universal derivations of Poisson algebras, J. Algebra, 354 (2012), 77-94. [26] M. Van den Berger, Double Poisson algebras, Trans. Amer. Math. Soc., 360(11) (2008), 5711-5769. [27] P. Xu, Noncommutative Poisson algebras, Amer. J. Math., 116(1) (1994), 101-125. [28] X.-P. Xu, Novikov-Poisson algebras, J. Algebra, 190(2) (1997), 253-279. [29] Y.-H. Yang, Y. Yao and Y. Ye, (Quasi-)Poisson enveloping algebras, Acta Math. Sin. (Engl. Ser.), 29(1) (2013), 105-118. [30] Y. Yao, Y. Ye and P. Zhang, Quiver Poisson algebras, J. Algebra, 312(2) (2007), 570-589. 16 JIAFENG L U, XINGTING WANG, AND GUANGBIN ZHUANG Lu: Department of Mathematics, Zhejiang Normal University, Jinhua, Zhejiang 321004, P.R. China E-mail address: [email protected], [email protected] Wang: Department of Mathematics, Temple University, Philadelphia 19122, USA E-mail address: [email protected] Zhuang: Department of Mathematics, University of Southern California, Los Angeles 90089-2532, USA E-mail address: [email protected]
1805.08965	1	1805	2018-05-23T05:24:09	Group rings and the RS-property	[ "math.RA" ]	The object of this paper is to study (infinite) groups whose integral group rings have only trivial central units. This property is closely related to a property, here called the RS-property (\cite{DMS05}, \cite{RS90}), involving conjugacy in the group.	math.RA	math	Group rings and the RS-property Gurmeet K. Bakshi Centre for Advanced Study in Mathematics, Panjab University, Chandigarh 160014, India. email: [email protected] Sugandha Maheshwary ∗† Indian Institute of Science Education and Research, Mohali, Sector 81, Mohali (Punjab)-140306, India. email: [email protected] Inder Bir S. Passi Centre for Advanced Study in Mathematics, Panjab University, Chandigarh-160014, India & Indian Institute of Science Education and Research, Mohali, Sector 81, Mohali (Punjab)-140306, India. email: [email protected] Abstract The object of this paper is to study (infinite) groups whose integral group rings have only trivial central units. This property is closely related to a property, here called the RS-property ([6], [19]), involving conjugacy in the group. Keywords : integral group rings, unit group, trivial central units, FC-subgroup. MSC2000 : 16U60; 16S34; 20C07; 20E06; 20E45; 20F14 ∗Research supported by DST, India (INSPIRE/04/2017/000897). †Corresponding author 1 1 Introduction Given a group G, let U(Z[G]) be the group of units of the integral group ring Z[G] and let Z(U(Z[G])) be its center. Trivially, Z(U(Z[G])) contains ±Z(G), where Z(G) denotes the center of G. In case Z(U(Z[G])) = ±Z(G), i.e., all central units of Z[G] are trivial, following [3], we call G a cut-group or a group with the cut-property. The question of classifying cut-groups was explicitly posed, for the first time, by Goodaire and Parmenter [8]. As an answer, Ritter and Sehgal [19] gave a characterization for finite cut-groups which was later generalized by Dokuchaev, Polcino Milies and Sehgal [6] to arbitrary groups. Let us say that an element x ∈ G of finite order has the RS-property (or is an RS-element) in G if xj ∼G x±1 for all j ∈ U(o(x)), (1) where o(x) denotes the order of x, U(n) := {j : 1 ≤ j ≤ n, gcd(j, n) = 1}, and y ∼G z denotes y is conjugate to z in G. Let Φ(G) denote the FC-subgroup of G, i.e., the subgroup consisting of those elements of G which have only finitely many conjugates in G, and Φ+(G) its torsion subgroup. Then the characterization of cut-groups given in [6] can be stated to say that G is a cut-group if, and only if, every element of Φ+(G) has the RS-property in G. Recently, finite cut-groups and their properties have been explored further ([2]-[4], [14]; see also [15]). Our purpose in the present work is to examine the class of infinite cut-groups. This class is neither subgroup-closed nor is it quotient-closed. In Section 2, we show (Lemma 3) that if A is an RS-subgroup of a group G, then A/N is RS-subgroup of G/N for every finite normal subgroup N of G contained in A and thus deduce (Theorem 4) that the class of cut-groups is closed under quotients by finite normal subgroups. We examine the class of cut- groups under extensions (Theorem 5) including amalgams and HNN extensions and thus provide several interesting examples of infinite cut-groups. Extending the result on characterization of finite metacyclic cut-groups ([3], Theorem 5), a classification (Theorem 6) of infinite metacyclic cut-groups has been given. We also classify p-groups (Theorem 9) and nilpotent groups (Theorem 10) which are cut-groups. For a finite group G and a normal subgroup A of G, it has been shown (Theorem 11) that Z(U(Z[G])) ∩ (1 + ∆(G)∆(A)) is trivial, if and only if, the rank of Z(U(Z[G])) equals that of Z(U(Z[G/A])), where ∆(G) denotes the augmentation ideal of Z[G]. Given a finite normal subgroup A of a solvable 2 group G, G/CG(A), where CG(A) denotes the centralizer of A in G, is a finite solvable group. Thus if G/CG(A) is a cut-group, then, by ([2], Theorem 1.2), no prime other that 2, 3, 5 and 7 can divide the order of G/CG(A). If, in addition, Z(U(Z[H])) ∩ (1 + ∆(H)∆(A)) is trivial, then we show (Corollary 12) that the order of A is also not divisible by any prime different from 2, 3, 5 and 7, where H := A ⋊ G/CG(A). In Section 3, we consider symmetric central units; a unit P ugg ∈ Z[G] being symmetric if P ugg = P ugg−1. We prove (Theorem 13) that, modulo the triv- ial units, the group of central units is isomorphic to a torsion free subgroup of Z(U(Z[G])) consisting of symmetric central units. Consequently, it follows that all central units of Z[G] are trivial, if so are all the symmetric central units. Finally, in Section 4, we observe (Theorem 15) that Zi(U)/Zi−1(U) is of finite exponent for all i ≥ 2, provided Z(G) is of finite exponent or G is generated by torsion elements of bounded exponent, where Zi(U) denotes the ith term of the upper central series of U(Z[G]). 2 cut-groups Let G be an arbitrary group. Let us say that a subgroup A of G is an RS-subgroup of G, if every torsion element of A is an RS-element in G. Such subgroups are relevant for the study of cut-groups because of the following characterization: If A is a normal subgroup of a finite group G, then A is an RS-subgroup of G if, and only if, Z(U(Z[G]))∩ Z[A] consists of trivial units ([6], Theorems 8 & 9). Furthermore, a key step in the reduction of the investigation of infinite cut-groups to that of finite groups is provided by the following: Lemma 1. ([6], Lemma 10) If G is an arbitrary group and A a finite normal subgroup of G, then there exists a finite extension H of A such that Z(U(Z[G])) ∩ Z[A] = Z(U(Z[H])) ∩ Z[A]. The various known criteria, available in [6], [13] and [19] for a group to be a cut-group can be put together as follows: 3 Theorem 2. For every group G, the following statements are equivalent: (i) G is a cut-group. (ii) Φ+(G) is an RS-subgroup of G. (iii) Z(U(Z[G])) ∩ Z[Φ+(G)] = ±T (Z(G)), where T (Z(G)) denotes the torsion subgroup of Z(G). (iv) ±G = NU (G), the normalizer of G in U := U(Z[G]). Clearly, the property of being an RS-subgroup is subgroup-closed. This property also turns out to be closed under taking quotients by finite normal subgroups. To be precise, we have Lemma 3. Let G be a group and A a subgroup of G. Let N be a finite normal subgroup of G contained in A. If A is an RS-subgroup of G, then A/N is an RS-subgroup of G/N. Proof. Let a := aN ∈ A/N be a torsion element. We need to check that for all j ∈ U(o(a)), aj ∼G/N a±1. Since N is finite, every element of the coset aN is a torsion element. Let a0 be an element of minimal order in the coset aN. Let i be such that ij ≡ 1 mod o(a), so that aij 0 = a0, and consequently, by the minimality of the order of a0, o(aij 0 ) = o(a0). This gives gcd(ij, o(a0)) = 1 and hence gcd(j, o(a0)) = 1. Since by assumption A is an RS-subgroup of G, we have aj 0 ∼G a±1 0 , which yields (a)j ∼G/N a±1. If G is a finite cut-group, then so is G/N for every N EG [19]. As a generalization of this fact, we have the following: Theorem 4. If G is a cut-group, then G/N is a cut-group for every finite normal subgroup N of G. Proof. Let A/N be a finite normal subgroup of G/N. Since N is finite, A is a finite normal subgroup of G and thus it is an RS-subgroup of G. By Lemma 3, A/N is an RS-subgroup of G/N. Consequently, Φ+(G/N) is an RS -subgroup of G/N. Hence Theorem 2 yields that G/N is a cut-group. We next consider the behaviour of cut-groups under extensions. Note that, in view of Theorem 2, every group G with Φ+(G) = {1} is a cut-group; in particular, a torsion-free group is a cut-group. 4 Theorem 5. (i) Let G be a normal subgroup of the group Π and Q = Π/G. (a) If Q is a cut-group, and G ∩ Φ+(Π) = {1} (in particular, if Φ+(G) = {1}), then Π is a cut-group. (b) If Φ+(Q) = {1}, and Φ+(G) is an RS-subgroup of Π (in particular, if G is a cut-group), then Π is a cut-group. (ii) If Π = G ∗A G′ is an amalgam of arbitrary groups G and G′ with the amal- gamated subgroup A an RS-subgroup of G or G′, then G is a cut-group, provided A 6= G and A 6= G′. In particular, the free product of arbitrary non-trivial groups is a cut-group. (iii) If Π is an HNN extension of a group G over isomorphic subgroups A and B such that one of A or B is an RS-subgroup of G, then Π is a cut-group. Proof. (i) Let f : Π −→ Q be the canonical epimorphism and let Q be a cut- group. Let H be a finite normal subgroup of Π, so that f (H) is a finite normal subgroup of the cut-group Q. Let h ∈ H and j a positive integer coprime to o(h). Clearly then j is relatively prime to o(f (h)), and hence f (h)j ∼Q f (h)±1, i.e., there exists q ∈ Q such that q−1f (h)jq = f (h)±1. (2) Let y ∈ Π be such that f (y) = q. In view of Eq. (2), we have f (y−1hjyh∓1) = 1. Observe that y−1hjyh∓1 ∈ G ∩ Φ+(Π). Now, if G ∩ Φ+(Π) = {1}, then y−1hjyh∓1 = 1, i.e., hj ∼Π h±1. Hence H is an RS-subgroup of Π and conse- quently, Π is a cut-group. This proves (a). Next consider Φ+(Q) = {1}. Then every finite normal subgroup of Π is necessarily contained in G. Therefore, Φ+(Π) is contained in G, and hence in Φ+(G). Now if, Φ+(G) is an RS-subgroup of Π, then it follows that Φ+(Π) is an RS-subgroup of Π and therefore Π is a cut-group. Finally, we can see that if G is a cut-group, then Φ+(G) is an RS-subgroup of G and therefore also of Π. This proves (b). (ii) Let Π = G ∗A G′ be an amalgam of arbitrary groups G and G′ with the amalgamated subgroup A an RS-subgroup of G or G′. Suppose that A 6= G and A 6= G′. 5 If the index of A in both G and G′ is two, then Π/A is isomorphic to the free product C2 ∗ C2 and hence Φ+(Π/A) = {1}. Also A an RS-subgroup of one of G or G′ implies that A is an RS-subgroup of Π. Consequently Φ+(A) is an RS-subgroup of Π. Thus (i)(b) yields that Π is a cut-group. If the index of A is atleast three in either G or G′, then, by ([5], Proposition 1), Φ+(Π) is a subgroup of A. Since A is an RS-subgroup of either G or G′, A is also an RS-subgroup of Π. Consequently, Φ+(Π) is an RS-subgroup of Π, and hence Π is a cut-group. (iii) Let Π be an HNN extension of a group G over isomorphic subgroups A and B of G. If either A or B is a proper subgroup of G, then, by ([5], Proposition 3), Φ+(Π) is a subgroup of both A and B. Since one of A or B is an RS-subgroup of G, it follows that Φ+(Π) is an RS-subgroup of G and, consequently, that of Π, as desired. If A = B = G, then G is a cut-group and Π/G is infinite cyclic. Hence, by (i)(a), Π is a cut-group. The above theorem enables us to construct interesting examples of cut-groups. Examples 1. While not every finite simple group is a cut-group ([1], [7], also see [15], Theorem 2), observe that if G is an infinite simple group, then clearly Φ+(G) = {1} and therefore it is a cut-group. Furthermore, as an immediate consequence of Theorem 5(i)(a), it follows that An extension of an infinite simple group by a cut-group is a cut- group. 2. Recall that P SL(n, k) is simple if k is a field of characteristic 0 and n ≥ 2. If the roots of unity in k are of exponent dividing 4 or 6 (e.g., if k = Q or R), then Theorem 5(i)(b) yields that SL(n, k) is a cut-group for n ≥ 2. 3. It is known that the modular group P SL(2, Z) is the free product of cyclic groups C2 and C3. Thus, Theorem 5(ii) yields that The modular group P SL(2, Z) is a cut-group. 6 4. Observe that SL(2, Z) is isomorphic to C4 ∗C2 C6 and thus, by Theorem 5(ii), SL(2, Z) is a cut-group. 5. The Baumslang Solitar group BS(m, n) := ha, t t−1amt = ani, where m and n are non-zero integers, is an HNN extension. Thus, by Theorem 5(iii) The Baumslag Solitar groups BS(m, n) are cut-groups. We now proceed to show that Theorem 5 enables us to classify infinite metacyclic cut-groups. It may be mentioned that a complete list (up to isomorphism) of finite metacyclic cut-groups has been computed in [3]. For every group G, Φ+(Z(G)) ⊆ Φ+(G). If the equality holds, then we can see that G is cut-group if, and only if, Z(G) is a cut-group, i.e., each central torsion element must have order dividing 4 or 6. This observation is helpful for the study of infinite metacyclic cut-groups. Theorem 6. An infinite non-abelian metacyclic group G is a cut-group if, and only if, it is isomorphic to one of the following groups: (i) ha, b bn = 1, ba = a−1bi, n ∈ {0, 2, 4, 6, 8, 12}; (ii) ha, b am = 1, ba = arbi, m ≥ 3, 1 6= r ∈ U(m) and U(m) = h−1, ri. Proof. Let G be an infinite metacyclic group and N = hai a cyclic normal sub- group of G with G/N = hbNi cyclic. Let m = o(a) and n = o(b). As G is infinite, one of m or n must be 0. Case I: m = 0. In this case, G ∼= ha, b bn = 1, ba = a−1bi. If n = 0, then, by Theorem 5(i), G is a cut-group. We thus assume that n 6= 0. Observe that Z(G) = hb2i. We assert that Φ+(G) = Φ+(Z(G)) = hb2i. Let A be a finite normal subgroup of G and let g ∈ A, so that g = aαbβ for some α, β ∈ Z. Now, gb = a−αbβ = a−2αg ∈ A implies α = 0, as A is finite. Furthermore, ga = a(−1)+(−1)β bβg ∈ A implies 2 β and hence A ⊆ hb2i. Consequently, Φ+(G) ⊆ hb2i; however, the reverse inclusion is obvious. Thus the assertion follows. Consequently, by the foregoing observation, G is a cut-group if, and only if, n ∈ {2, 4, 6, 8, 12}. 7 Case II: m 6= 0. In this case, n = 0 and so G ∼= ha, b am = 1, ba = arbi, r ∈ U(m). Since G is non-abelian, m ≥ 3. By Theorem 5(i)(b), G is a cut-group if, and only if, a is an RS-element of G. It is easy to see that a is an RS-element of G if, and only if, U(m) = h−1, ri. We next consider p-groups. Theorem 7. A non-trivial normal subgroup A of a finite p-group G is an RS- subgroup of G if, and only if, one of the following holds: (i) p = 2 and a3 ∼G a±1 for all a ∈ G; (ii) p = 3 and a2 ∼G a−1 for all a ∈ G . Proof. A non-trivial normal subgroup A of G must intersect Z(G) non-trivially, as G is a p-group. However, a central element of G is an RS-element if, and only if, its order divides 4 or 6. This gives that p = 2 or 3. If p = 2, then a3 ∼G a±1, as 3 ∈ U(o(a)) and for the same Let 1 6= a ∈ A. reason, if p = 3, then a2 ∼G a±1. In the latter case, we just need to check that a2 6∼G a. Consider the lower central series {γi(G)}≥0 of G. Since G is a p-group, there exists n ≥ 1 such that γn(G) = {1}. Hence we can find 1 ≤ i < n such that a ∈ γi(G) \ γi+1(G). If a2 ∼G a, then a2 = g−1ag for some g ∈ G, which implies a = [a, g] ∈ [γi(G), G] = γi+1(G), contradicting the choice of i. Conversely, let G be a finite p-group, p ∈ {2, 3}, and A a normal subgroup of G satisfying (i) or (ii) according as p = 2 or 3. Let a ∈ A. If p = 2, then U(o(a)) = h3i×h−1i and hence a3 ∼G a±1 which implies aj ∼G a±1 for all j ∈ U(o(a)). If p = 3, then 2 is a primitive root modulo o(a) and hence a2 ∼G a−1 implies that aj ∼G a−1 for all j ∈ U(o(a)). Therefore, in either case, a is an RS-element of G. 8 The above result yields information about the RS-subgroups of a finite nilpotent group. Given a group G, let π(G) denote the set of primes p for which G contains an element of order p. Corollary 8. Let A be a normal RS-subgroup of a finite nilpotent group G, then π(A) ⊆ {2, 3}. Proof. Observe that for every prime p ∈ π(G), the Sylow p-subgroup of A is an RS-subgroup of the Sylow p-subgroup of G. Thus the assertion follows from Theorem 7. In view of Lemma 1, Theorems 2 & 7, we obtain the following: Theorem 9. A p-group G is a cut-group if, and only if, one of the following holds: (i) p = 2 and a3 ∼G a±1 for all a ∈ Φ+(G) ; (ii) p = 3 and a2 ∼G a−1 for all a ∈ Φ+(G) ; (iii) Φ+(G) = {1}. It may be noted that, while a finite p-group, which is a cut-group, must necessarily be a 2-group or a 3-group, this is not the case for an infinite p-group to be a cut- group. For example, for any prime p, the wreath product Cp ≀ A, where A is a direct sum of infinitely many copies of Cp, is a cut-group, since Φ(Cp ≀ A) = {1}. For more general examples of groups with Φ+(G) = {1}, see [17]-[18]. Let Gp denote the subset of G consisting of p-elements in G. For nilpotent cut- groups, we have Theorem 10. A nilpotent group G is a cut-group if, and only if, either Φ+(G) = {1} or Φ+(G) is a {2, 3}-group and the following conditions hold: (i) for all a ∈ Φ+(G)2, a3 ∼G a±1; (ii) for all a ∈ Φ+(G)3, a2 ∼G a−1. 9 Proof. Let G be a nilpotent group with Φ+(G) 6= {1}. Let x ∈ Φ+(G) be an element of prime order (say p) and A a finite normal subgroup of G containing x. Since G is a cut-group, A is an RS-subgroup of G. Consider H := A ⋊ G/CG(A). Observe that A is an RS-subgroup of the finite nilpotent group H. Thus, by Corollary 8, π(A) ⊆ {2, 3}, which gives that p = 2 or 3. Furthermore, as in Theorem 7, it turns out that a3 ∼G a±1 if p = 2, and a2 ∼G a−1 if p = 3. The converse can be seen easily as U(2i) = h3i × h−1i, U(3j) = h2i and U(2i3j) = U(2i) × U(3j) for all i, j ≥ 1. Theorem 11. Given a normal subgroup A of a finite group G, the following statements are equivalent: (i) Z(U(Z[G])) ∩ (1 + ∆(G)∆(A)) is trivial. (ii) ρ(G) = ρ(G/A), where ρ(G) denotes the rank of Z(U(Z[G])). Proof. Suppose (i) holds. Let Q[G] ∼= ⊕1≤i≤mMni(Di) be the Wedderburn decomposition of Q[G]. By reordering, if necessary, we can assume that and Q[G](1 − A) ∼= ⊕1≤i<rMni(Di) Q[G/A] ∼= Q[G] A ∼= ⊕r≤i≤mMni(Di), where 1 ≤ r < m. The center Z(Di) of each division ring Di is a finite extension of Q. Let Oi be the ring of integers of Z(Di). We then have and ρ(G) = X ρ(U(Oi)) 1≤i≤m ρ(G/A) = X ρ(U(Oi)). r≤i≤m Consequently, ρ(G/A) ≤ ρ(G). Thus, to establish (ii), it suffices to prove that under the natural map π : Z[G] → Z[G/A], any set of linearly independent central units in Z[G] are mapped to linearly independent central units of Z[G/A]. 10 Let u1, u2, · · · , ut be linearly independent central units in Z[G]. Denote π(ui) by ui. Suppose uk1 t = 1 for some integers k1, k2, · · · , kt. Then 2 · · · ukt 1 uk2 1 uk2 uk1 2 · · · ukt t − 1 ∈ ∆(G, A). 1 uk2 2 · · · ukt So u = uk1 t ∈ 1 + ∆(G, A). Consequently, u ≡ a mod ∆(G)∆(A) for some a ∈ A. This gives ua−1 ≡ 1 mod ∆(G)∆(A). Since G is finite and u is central, it follows that um belongs to 1 + ∆(G)∆(A) for some m ≥ 1. Therefore, um and hence u is trivial, i.e., u = uk1 t ∈ ±Z(G). Raising it to a suitable power and using the fact that u1, u2, · · · , ut are linearly independent, it follows that ki = 0 for all i, thus proving the validity of (ii). 2 · · · ukt 1 uk2 Conversely, if (ii) holds, then it follows that under the natural map π : Z[G] → Z[G/A], any set of linearly independent central units in Z[G] are mapped to linearly independent central units of Z[G/A]. Suppose u = X a∈A uaa ∈ Z(U(Z[G])) ∩ (1 + ∆(G)∆(A)). Then π(u) = 1 and hence {u} is mapped to a linearly dependent set. Thus {u} is linearly dependent and so u can't have infinite order. Consequently u is a trivial central unit and (i) holds. Corollary 12. Let A be a finite normal subgroup of a solvable group G and let H := A ⋊ G/CG(A). Suppose that (i) Z(U(Z[H])) ∩ (1 + ∆(H)∆(A)) is trivial; (ii) G/CG(A) is a cut-group. Then, π(A) ⊆ {2, 3, 5, 7}. Proof. Observe that H is finite. Since H/A ∼= G/CG(A) and G/CG(A) is a cut- group, ρ(H/A) = ρ(G/CG(A)) = 0. In view of (i), Theorem 11 yields that ρ(H) = ρ(H/A). Thus ρ(H) = 0, i.e., H is a finite solvable cut-group. Consequently, by ([2], Theorem 1.2) , π(H) ⊆ {2, 3, 5, 7}, which gives the desired result. 3 Symmetric central units Given u = P ugg ∈ Z[G], let u∗ = P ugg−1. An element u ∈ Z[G] is called symmetric, if u = u∗. 11 It is known (see [12], Corollary 7.1.9) that if G is a finite abelian group, then U(Z[G]) is the direct product of ±G and a torsion free group of symmetric units. As a generalization of this result, we have the following: Theorem 13. For every group G, the following statements hold: (i) There is an exact sequence 1 → ±Z(G) → Z(U(Z[G])) → S → 1, where S is a torsion free subgroup of Z(U(Z[G])) consisting of symmetric central units. (ii) If u ∈ Z(U(Z[G])), then there exists an element g ∈ Z(G) such that u = gu∗. Furthermore, u2 ∈ Z(G)S. (iii) If the symmetric central units of Z[G] are trivial, then so are all central units, i.e., G is a cut-group. Proof. (i) Observe that the map θ : Z(U(Z[G])) → Z(U(Z[G])), u 7→ uu∗, is a group homomorphism and ker θ = ±Z(G). Let S be the image of θ. We will show that S is torsion free. Let u ∈ Z(U(Z[G])). If uu∗ is of finite order, then it belongs to ±Z(G). Suppose uu∗ = ±h, If h 6= 1, then from the above equation it follows that where h ∈ Z(G). Pg∈G u2 g = 0, which is not so as u 6= 0. Hence, h = 1 and uu∗ = ±1. How- ever, uu∗ can't be −1 as the augmentation of uu∗ is 1. Thus uu∗ = 1 and consequently, S is torsion free. (ii) As u is central in Z[G], it can be readily verified that θ(u2) = θ(uu∗). Hence, by (i), u2 = ±guu∗, where g ∈ Z(G). Since both u2 and uu∗ are of augmentation 1, we have u2 = guu∗. Hence, (ii) follows. (iii) If symmetric central units are trivial, then from (ii) it follows that u2 ∈ Z(G) for all Z(U(Z[G])). However, by (i), Z(U(Z[G]))/ ± Z(G) is torsion free. There- fore, (iii) follows. In the above theorem, (ii) may be campared with ([16], Theorem 1). An imme- diate consequence of the above theorem is the following: 12 Corollary 14. For every group G with periodic center, in particular if G is a finite group, Z(U(Z[G]))/ZS (U(Z[G])) is a torsion group, where ZS(U(Z[G])) is the subgroup of Z(U(Z[G])) consisting of symmetric central units. 4 Hypercentral units Let Z0(U) ≤ Z1(U) ≤ · · · ≤ Zn(U) ≤ · · · be the upper central series of U = U(Z[G]). Let Z∞(U) = ∪i≥1Zi(U) be the subgroup of U(Z[G]) consisting of the hypercentral units. Theorem 15. For all i ≥ 2, Zi(U)/Zi−1(U) is of finite exponent, provided Z(G) is of finite exponent or G is generated by torsion elements of bounded exponent. Proof. In view of ([9], Lemma 4.2, p. 432), it suffices to show that Z2(U)/Z1(U) is of finite exponent. Let u ∈ Z2(U). Suppose Z(G) is of finite exponent, say m. Consider an arbitrary g ∈ G. By ([10], Proposition 4.1), [u, g] ∈ Z(G). Therefore, [um, g] = [u, g]m = 1, i.e., um ∈ Z1(U), as desired. Next, suppose that G is generated by X and the elements of X have bounded exponent, say m. Then, for any x ∈ X, [u, x] ∈ Z1(U) implies that [um, x] = [u, x]m = [u, xm] = 1. Consequently um is a central unit, as desired. This proves the result. Corollary 16. If G is such that all units in Z[G] are hypercentral and one of the following conditions hold: (i) Z(G) is of finite exponent; (ii) G is generated by torsion elements of bounded exponent, then U(Z[G]) can't contain a free subgroup of rank ≥ 2. [For the classification of groups G with all units hypercentral, see [11].] Proof. Suppose F is a free subgroup of rank ≥ 2 contained in U(ZG). Consider any 1 6= u ∈ F . As u is hypercentral, by Theorem 15, it follows that a power of u belongs to Z1(U); let this power be m. So um commutes with all elements of F , i.e., it belongs to Z(F ). But F being free of rank ≥ 2, has trivial center. Consequently um = 1. This is not possible, as F is free. 13 Acknowledgement Inder Bir S. Passi is thankful to the Indian National Science Academy, New Delhi (India) for their support and to Ashoka University, Sonipat (India), for making available their facilities. References [1] R. Zh. Aleev, A. V. Kargapolov, and V. V. Sokolov, The ranks of central unit groups of integral group rings of alternating groups, Fundam. Prikl. Mat. 14 (2008), no. 7, 15 -- 21. [2] A. Bachle, Integral group rings of solvable groups with trivial central units, Forum Math., https://doi.org/10.1515/forum-2017-0021. [3] G. K. Bakshi, S. Maheshwary, and I. B. S. Passi, Integral group rings with all central units trivial, J. Pure Appl. Algebra 221 (2017), no. 8, 1955 -- 1965. [4] D. Chillag and S. Dolfi, Semi-rational solvable groups, J. Group Theory 13 (2010), no. 4, 535 -- 548. [5] Y. de Cornulier, Infinite conjugacy classes in groups acting on trees, Groups Geom. Dyn. 3 (2009), no. 2, 267 -- 277. [6] M. Dokuchaev, C. Polcino Milies, and S. K. Sehgal, Integral group rings with trivial central units II, Comm. Algebra 33 (2005), no. 1, 37 -- 42. [7] R. A. Ferraz, Simple components and central units in group algebras, J. Algebra 279 (2004), no. 1, 191 -- 203. [8] E. G. Goodaire and M. M. Parmenter, Units in alternative loop rings, Israel J. Math. 53 (1986), no. 2, 209 -- 216. [9] P. Hall, The collected works of Philip Hall. Compiled by K. W. Gruenberg and J. E. Roseblade., Oxford (UK): Clarendon Press, 1988 (English). [10] M. Hertweck, E. Iwaki, E. Jespers, and S.O. Juriaans, On hypercentral units in integral group rings, J. Group Theory 10 (2007), no. 4, 477 -- 504. 14 [11] E. Iwaki and S. O. Juriaans, Hypercentral unit groups and the hyperbolicity of a modular group algebra, Comm. Algebra 36 (2008), no. 4, 1336 -- 1345. [12] E. Jespers and ´A. del R´ıo, Group Ring Groups, Volume 1: Orders and Generic Constructions of Units, De Gruyter, Berlin-Boston, 2015. [13] E. Jespers, S. O. Juriaans, J. M. de Miranda, and J. R. Rogerio, On the normalizer problem, J. Algebra 247 (2002), no. 1, 24 -- 36. [14] S. Maheshwary, Integral Group Rings With All Central Units Trivial: Solv- able Groups, Indian J. Pure Appl. Math. 49 (2018), no. 1, 169 -- 175. [15] S. Maheshwary and I. B. S. Passi, The upper central series of the unit groups of integral group rings: a survey (to appear in Indian Statistical Institute Series, Springer). [16] C. Polcino Milies and S. K. Sehgal, Central units of integral group rings, Commun. Algebra 27 (1999), no. 12, 6233 -- 6241. [17] J. P. Pr´eaux, Group extensions with infinite conjugacy classes, Confluentes Math. 5 (2013), no. 1, 73 -- 92. [18] J.P. Pr´eaux, Wreath product of groups with infinite conjugacy classes, preprint(2006), 2 pages, arXiv:math/0612685 [math.GR]. [19] J. Ritter and S. K. Sehgal, Integral group rings with trivial central units, Proc. Amer. Math. Soc. 108 (1990), no. 2, 327 -- 329. 15
1001.2653	1	1001	2010-01-15T14:33:41	Two examples about zero torsion linear maps on Lie algebras	[ "math.RA" ]	The question of whether or not any zero torsion linear map on a non abelian real Lie algebra g is necessarily an extension of some CR-structure is considered and answered in the negative. Two examples are provided, one in the negative and one in the positive.In both cases, the computation up to equivalence of all zero torsion linear maps on g is used for an explicit description of the equivalence classes of integrable complex structures on the direct product g x g.	math.RA	math	Two examples about zero torsion linear maps on Lie algebras ∗ L. Magnin Institut Math´ematique de Bourgogne † [email protected] November 14, 2018 Abstract The question of whether or not any zero torsion linear map on a non abelian real Lie algebra g is necessarily an extension of some CR- structure is considered and answered in the negative. Two examples are provided, one in the negative and one in the positive. In both cases, the computation up to equivalence of all zero torsion linear maps on g is used for an explicit description of the equivalence classes of integrable complex structures on g × g. 1 Introduction. Given a real Lie algebra g, the determination up to equivalence of zero torsion linear maps from g to g plays an important role in the computation of complex structures on direct products involving g ([2]). In the present note, we consider the question of whether or not any such zero torsion linear map for non abelian g is necessarily an extension of some CR-structure. We answer the question in the negative by computing (up to equivalence) all zero torsion linear maps from the real 3-dimensional Heisenberg Lie algebra n into itself. The result is then used to exhibit a complete set of representatives of equivalence classes of complex structures on n × n. We also compute all zero torsion linear maps ∗Math. Subj. Class. [2000] : 17B30. Key words : Complex structures, CR-structures, zero torsion, Heisenberg Lie algebra, sl(2, R). †UMR CNRS 5584, Universit´e de Bourgogne, BP 47870, 21078 Dijon Cedex, France. 1 on sl(2, R). In that case they are extensions of CR-structures. We deduce a complete set of representatives of equivalence classes of complex structures on sl(2, R) × sl(2, R). 2 Preliminaries. Let G0 be a connected finite dimensional real Lie group, with Lie algebra g. A linear map J : g → g is said to have zero torsion if it satisfies the condition [JX, JY ] − [X, Y ] − J[JX, Y ] − J[X, JY ] = 0 ∀X, Y ∈ g. (1) If J has zero torsion and satisfies in addition J 2 = −1, J is an (integrable) com- plex structure on g. That means that G0 can be given the structure of a complex manifold with the same underlying real structure and such that the canoni- cal complex structure on G0 is the left invariant almost complex structure J associated to J (For more details, see [3]). To any (integrable) complex struc- ture J is associated the complex subalgebra m = n X := X − iJX ; X ∈ go of the complexification gC of g. In that way, (integrable) complex structures can be identified with complex subalgebras m of gC such that gC = m ⊕ ¯m, bar denoting conjugation. J is said to be abelian if m is. When computing the matrices of the zero torsion maps in some fixed basis (xj)16j6n of g, we will denote by ijk (1 6 i, j, k 6 n) the torsion equation obtained by projecting on xk the equation (1) with X = xi, Y = xj. The automorphism group Aut g of g acts on the set of all zero torsion linear maps and on the set of all complex structures on g by J 7→ Φ ◦ J ◦ Φ−1 ∀Φ ∈ Aut g. Two J, J ′ on g are said to be equivalent (notation: J ≡ J ′) if they are on the same Aut g orbit. For complex structures and simply connected G0, this amounts to the existence of an f ∈ Aut G0 such that f : (G0, J) → (G0, J ′) is biholomorphic. 3 Case of sl(2, R). Let G = SL(2, R) denote the Lie group of real 2 ×2 matrices with determinant 1 σ = (cid:18)a b c d(cid:19) , ad − bc = 1. (2) Its Lie algebra g = sl(2, R) consists of the zero trace real 2 × 2 matrices X = (cid:18)x z −x(cid:19) = xH + yX+ + zX− y 2 with basis H = ( 1 0 −1 0 ) and commutation relations 0 −1 ), X+ = ( 0 1 0 0 ), X− = ( 0 0 [H, X+] = 2X+, [H, X−] = −2X−, [X+, X−] = H. (3) Beside the basis (H, X+, X−), we shall also make use of the basis (Y1, Y2, Y3) where Y1 = 1 2 (X+ + X−), with commutation relations 2(X+ − X−), Y3 = 1 2 H, Y2 = 1 [Y1, Y2] = Y3, [Y1, Y3] = Y2, [Y2, Y3] = Y1. (4) The adjoint representation of G on g is given by Ad(σ)X = σXσ−1. The matrix Φ of Ad(σ) (σ as in (2)) in the basis (H, X+, X−) is Φ =   1 + 2bc −ac −2ab 2cd bd a2 −b2 −c2 d2   . (5) The adjoint group Ad(G) is the identity component of Aut g and one has Aut g = Ad(G) ∪ Ψ0Ad(G) , Ψ0 = diag(1, −1, −1). (6) The adjoint action of G on g preserves the form x2 + yz. The orbits are : (i) the trivial orbit {0}; (ii) the upper sheet z > 0 of the cone x2 + yz = 0 (orbit of X−); (iii) the lower sheet z < 0 of the cone x2 + yz = 0 (orbit of −X−); (iv) for all s > 0 the one-sheet hyperboloid x2 + yz = s2 (orbit of sH); (v) for all s > 0 the upper sheet z > 0 of the hyperboloid x2 + yz = −s2 (orbit of s(−X+ + X−)); (vi) for all s > 0 the lower sheet z < 0 of the hyperboloid x2 + yz = −s2 (orbit of s(X+ − X−)). The orbits of g under the whole Aut g are, beside {0}: (I) the cone x2 + yz = 0 (orbit of X−); (II) the one-sheet hyperboloid x2 + yz = s2 (orbit of sH) (s > 0); (III) the two-sheet hyperboloid x2 + yz = −s2 (orbit of s(X+ − X−)) (s > 0). Lemma 1. Let g = sl(2, R), and J : g → g any linear map. J has zero torsion if and only if it is equivalent to the endomorphism defined in the basis (Y1, Y2, Y3) (resp. (H, X+, X−)) by J∗(λ) =   0 0 −1 0 0 λ 1 0 0   , λ ∈ R , (7) J∗(λ) 6≡ J∗(µ) for λ 6= µ 3 (resp. J(α) =   0 − 1 2 − 1 2 1 α −α 1 −α α   , α ∈ R , (8) J(α) 6≡ J(β) for α 6= β). Proof. Let J = (ξi are in the basis (H, X+, X−): j)16i,j63 in the basis (H, X+, X−). The 9 torsion equations 121 122 123 131 132 133 231 232 233 2 = 0, 3 = 0, 2(ξ2 (ξ2 3ξ1 2 + ξ3 2 − ξ1 2 + (ξ2 1 − 2(ξ2 1)ξ1 2 + 1 + (ξ2 2 + ξ1 1)ξ3 1 − (ξ2 2(ξ2 1ξ1 2)2) − ξ3 1ξ2 1 + 2ξ1 (ξ3 2)ξ3 2 + 2ξ1 1)ξ3 3)ξ1 1 − 2ξ1 1 + 2ξ2 1ξ2 3 − (ξ2 2(ξ2 2 − 2ξ1 1)ξ2 3 + (ξ2 1 + 2ξ1 3)ξ2 1 − 2ξ3 3 − 2(ξ3 2ξ2 3 − 2 + 2ξ3 1ξ2 ξ3 1ξ1 1 − 2ξ3 2 − ξ1 3 + (ξ2 2ξ2 1 − ξ3 2ξ1 2 − 1 − ξ2 3ξ1 4ξ1 3)ξ2 2 + ξ3 2 − (ξ2 3ξ1 1 = 0, 2ξ1 3 − (ξ2 2 + ξ3 3)ξ3 1 = 0. 1 + 2ξ1 3)ξ3 1 − 2ξ3 2ξ2 3 = 0, 3ξ3 2 + 2ξ3 2 = 0, 3)ξ3 1 + 2ξ1 3ξ2 3 = 0, 3)2 = 0, 1)ξ3 4ξ2 4ξ3 3 = 0, 3 = ξ2 J has at least one real eigenvalue λ. Let v ∈ g, v 6= 0, an eigenvector associated to λ. From the classification of the Aut g orbits of g, we then get 3 cases according to whether v is on the orbit (I),(II),(III) (in the cases (II), (III) one may choose v so that s = 1). Case 1. There exists ϕ ∈ Aut g such that v = ϕ(X−). Then, replacing J by ϕ−1Jϕ, we may suppose ξ1 3 = 0. That case is impossible from 132 and 133. Case 2. There exists ϕ ∈ Aut g such that v = ϕ(H). Then we may suppose ξ2 3ξ3 1 = ξ3 3 = 0. Then 123 and 132 successively give ξ3 1 = 0. Now 122 and 231 read resp. −ξ2 2)2 + 1 = 0. Hence that case is impossible. Case 3. There exists ϕ ∈ Aut g such that v = ϕ(X+ − X−). Then we may suppose that v = X+ − X−. Now instead of the basis (H, X+, X−), we consider the basis (Y1, Y2, Y3). The matrix of J in the basis (Y1, Y2, Y3) has the form 2 6= 0, and 232, 233 yield ξ1 3 = ξ2 2)2 + 1 = 0, and ξ2 1 = 0. Then from 122, ξ2 2 + 2ξ1 3ξ3 1 and ξ1 2 − (ξ2 2 + (ξ2 2 = ξ1 3ξ3 J∗ =   η1 1 0 η1 3 η2 1 λ η2 3 η3 1 0 η2 3   . 4 Then the 9 torsion equations ∗ijk (the star is to underline that the new basis is in use) for J in that basis are: 3)λ − (η3 (η3 1 + η1 (η1 1 + λ)η2 1λ − 1 + (η3 η1 3 + η2 η2 3η1 1η1 1)2 + (η2 1λ + 1 + (η2 η1 η2 3η1 1 − η2 1(η1 η1 1λ + 1 − (η1 ∗121 ∗122 ∗123 ∗131 ∗132 ∗133 ∗231 ∗232 ∗233 1 = 0, 3)η1 1 − η1 1η3 1 = 0, 1 + λ)η3 1 − η2 1η3 3 − (η1 1η1 3 − η2 1)2 − (η1 1 + η2 3η3 3)2 + η3 3 + η3 3)2 + (η1 3η3 1 − λ)η3 1 = 0, 3)η3 1) − η2 3 = 0, 3 = 0, η2 3η1 1 + η1 3 − (η3 3)λ + (η3 3 + λ)η2 1 − η1 (η3 3 = 0. 3 = 0, 3 = 0, 1 − λ)η3 3 = 0, From ∗121 and ∗233, η1 1(η3 1 − η1 3) = −η3 3(η3 1 − η1 3). (9) 3η1 3 − η1 1 = η2 1 = η1 3 = η1 1)2 + (η2 3 − η1 1)η2 1 − λ)η3 3 6= η1 1 6= 0. Then λ = 0. From ∗132, η3 1 is not possible either since then ∗231 would read (η1 3. Then λη3 1)η2 3 = 0, and ∗132 gives η1 1 = 0. 1.1) Consider the subcase η3 1) Suppose first that η3 1 = 0. 3 6= η1 3 = 0. Suppose η3 1 = 0, (η3 ∗131 and ∗133 read resp. (η3 1. 3, which implies η3 1λ + 1 = (η1 Then η2 3 = 0 by ∗231. As ∗123 then reads 1 = 0, this case η3 1 is not possible. Now, the case η3 1)2 + 1 = 0. We conclude that the subcase 1.1) is not possible. Hence we are in the subcase 1 )2 3 = −1+(η3 1.2) η3 3)2 + 2 = 0. This subcase 1.2) is not possible either. and ∗132 reads (η2 Hence case 1) is not possible, and we are necessarily in the case 2) η3 1 6= η1 3. From (9), η3 1η1 3 = 0 hence 1 )2+(η2 η3 1 6= 0 and η1 . Then ∗232 η3 1 reads η2 1)2 + (η1 3 = 0, which implies 1 = 0. Now ∗121 reads λ(1 + (η1 η2 1)2). The 1 )2−1 1 )2+(η3 subcase η1 2η1 1 1)2 + (η3 and ∗121 would read ((η1 1 − 1)2) = 0. Hence η1 1 = 0. Then ∗123 reads (ξ3 1)2 = 1 now implies the vanishing of all the torsion equations. In that case 1)2 + (η3 1 6= 0 is not possible since then ∗123 would yield λ = − (η1 1 = η2 3)2) = 0, i.e. η2 1)2) = −η1 1)2 + (η3 1 + 1)2)((η1 1)2 = 1. The condition (ξ3 1 )2+(η2 3 = − (η1 3)2 + λ2 + 1)(η3 1 6= 0. Then ∗231 yields η3 3)2 + 1 + η3 3(η1 1 +λ) η3 1 1 + λ)2(η2 1)2 − (η3 . From ∗122, η2 1(1 + (η1 3 = −η1 1. Then ∗132 reads (η1 1)2 + (η2 1)2 + (η2 3(((η2 3)2+1 η1 1 J∗ =   0 0 −ε 0 λ 0 0 ε 0   , 5 ε = ±1. Then in the basis (H, X+, X−) J =   2 − ε 2 − λ 2 2 0 − ε λ ε ε − λ 2 λ 2   The cases ε = ±1 are equivalent under Ψ0. Remark 1. Recall that a rank r (r > 1) CR-structure on a real Lie algebra g can be defined ([4]) as (p, Jp) where p is some 2r-dimensional vector subspace of g and Jp : p → p is a linear map such that (a): J 2 [X, Y ] − [JpX, JpY ] ∈ p ∀X, Y ∈ p, (c): (1) holds for Jp for all X, Y ∈ p. Then clearly J∗(λ) is an extension of a CR-structure. p = −1, (b): 4 Case of sl(2, R) × sl(2, R). 1 , Y (1) , Y (1) We consider the basis (Y (1) ) of sl(2, R) × sl(2, R), with the upper index referring to the first or second factor. The automor- phisms of sl(2, R) × sl(2, R) fall into 2 kinds: the first kind is comprised of the diag(Φ1, Φ2), Φ1, Φ2 ∈ Aut sl(2, R), and the second kind is comprised of the the Γ◦diag(Φ1, Φ2), with Γ the switch between the two factors of sl(2, R)×sl(2, R). , Y (2) , Y (2) , Y (2) 2 3 1 2 3 Lemma 2. Any integrable complex structure J on sl(2, R) × sl(2, R) is equiv- alent under some first kind automorphism to the endomorphism given in the basis (Y (1) ) by the matrix , Y (2) , Y (2) , Y (1) , Y (2) 1 2 1 2 3 J∗(ξ2 2, ξ2 5) = 3 , Y (1)  0 0 ξ2 0 2 0 1 0 0 0 − (ξ2 2 )2+1 ξ2 5 0 0  −1 0 0 0 0 0 0 0 0 −ξ2 0 2 0 0 1 0 0 ξ2 0 5 0 0 0 −1 0 0   , ξ2 2, ξ2 5 ∈ R , ξ2 5 6= 0. (10) 2, ξ2 5) is equivalent to J∗(ξ ′2 J∗(ξ2 tomorphism if and only if ξ ′2 2, ξ ′2 2 = ξ2 5) under some first (resp. second) kind au- 2, ξ ′2 ). J3 J4(cid:19) , (J1, J2, J3, J4 3 × 3 blocks), an inte- j)16i,j66 = (cid:18)J1 J2 5 (resp. ξ ′2 5 = − (ξ2 2 = −ξ2 2 )2+1 ξ2 5 5 = ξ2 2, ξ ′2 Proof. Let J = (ξi grable complex structure in the basis (Y (k) ℓ ). From lemma 1, with some first 6 kind automorphism, one may suppose J1 =   As T r(J) = 0, ξ5 [1], CSsl22.red and its output). 5 = −ξ2 0 0 ξ2 2 1 0 0 −1 0 0   , J4 =   0 0 ξ5 5 1 0 0 −1 0 0   . 2. Then one is led to (10) and the result follows (see Remark 2. The complex subalgebra m associated to J∗(ξ2 5) has basis Y (1) 2)Y (2) 1 = Y (1) . The complexification sl(2) × sl(2) of sl(2, R) × sl(2, R) has weight spaces de- composition with respect to the Cartan subalgeba h = CY (1) 2 + (1 + iξ2 1 = Y (2) 1 − iY (1) 1 − iY (2) 2 = −iξ2 5Y (1) , Y (2) , Y (2) 2, ξ2 2 ⊕ CY (2) : 3 2 3 2 ) ⊕ C(Y (2) h ⊕ C(Y (1) 1 + iY (1) 1 ⊕ C Y (2) Then m = (h ∩ m) ⊕ C Y (1) , which is a special case of the general fact proved in [5] that any complex (integrable) structure on a reductive Lie group of class I is regular. 1 with h ∩ m = C Y (2) 1 ⊕ C Y (2) 1 + iY (2) ) ⊕ C Y (1) . 3 3 1 2 5 Case of n. Let n the real 3-dimensional Heisenberg Lie algebra with basis (x1, x2, x3) and commutation relations [x1, x2] = x3. Lemma 3. Let J : n → n any linear map. J has zero torsion if and only if it is equivalent to one of the endomorphisms defined in the basis (x1, x2, x3) by: (i) (ii) (iii) 0 −1 0 1 0 0 ξ1 1 0 0 0 ξ1 1 0 S(ξ3 D(ξ1 3) =   1) =   T (a, b) =   0 −ab 1 0 b 0 0 0 ξ3 3 (ξ1 0 0 1)2−1 2ξ1 1   ,     , 0 0 ab−1 b , ξ3 3 ∈ R (11) ξ1 1 ∈ R, ξ1 1 6= 0 (12) a, b ∈ R, b 6= 0 (13) Any two distinct endomorphisms in the preceding list are non equivalent. T (a, b) is equivalent to T ′(a, b) =   b −b 0 a 0 0 0 0 ab−1 b   7 (14) Proof. Let J = (ξi ξ3 3(ξ2 j)16i,j63 in the basis (x1, x2, x3). The 9 torsion equations are: 121 122 123 131 132 133 231 232 233 2 + ξ1 ξ1 3(ξ2 1) = 0, 2 + ξ1 ξ2 3(ξ2 1) = 0, 2ξ1 1) − ξ2 1ξ1 1 + ξ2 2 + ξ1 3ξ1 ξ2 3 = 0, (ξ2 3)2 = 0, 1) + ξ1 3 − ξ1 2ξ1 3)2 = 0, (ξ1 ξ2 3ξ1 3 = 0, 3) − ξ2 2 − ξ3 2 + 1 = 0, 3 = 0, 2 = 0. ξ2 3(ξ3 ξ1 3(ξ2 3ξ1 Hence ξ1 3 = ξ2 3 = 0 , and we are left only with equation 123 which reads ξ3 3 T r(A) = det (A) − 1 (15) 1 ξ1 2 ξ2 1 ξ2 1 0 0 ∗ ∗ ξ3 where A = (cid:16) ξ1 similar over C, hence over R, to ( 0 −1 does not belong to the spectrum of ( 0 −1 of n for suitable α, β ∈ R, one gets J ≡ S(ξ3 2(cid:17) . Suppose first T r(A) = 0. Then A2 = −I, so that A is 1 0 ) . Hence J ≡ (cid:16) 0 −1 0 3(cid:17). Now, since ξ3 1 0 ) , taking the automorphism (cid:16) 1 0 0 α β 1(cid:17)   ≡ D(ξ1 b (cid:1) for some a, b ∈ R, and b 6= 0 from the trace. Then J ≡ T (a, b). Finally, T ′(a, b) ≡ T (a, b) since a 0 ) are similar for they have the same spectrum 1). If A is not a scalar matrix, then A is similar to (cid:0) 0 −ab 1I, then J =   3). Suppose now T r(A) 6= 0. Then . If A is a scalar matrix, i.e. A = ξ1 3 = det (A)−1 ξ3 0 0 1 )2−1 2ξ1 1 ξ1 1 0 0 ξ1 1 T r(A) 0 1 0 3 (ξ1 ∗ ∗ 1 1 b (cid:1) and ( b −b the matrices (cid:0) 0 −ab and are no scalar matrices. Remark 3. S(ξ3 D(ξ1 1), T (a, b) are not. 3) is an extension of a rank 1 CR-structure, however 6 CR-structures on n. Lemma 4. (i) Any linear map J : n → n which has zero torsion and is an extension of a rank 1 CR-structure on n such that p is nonabelian is equivalent to a unique 0 −1 1 0 0 0   0 0 ξ3 3   , ξ3 3 ∈ R. 8 (16) 0 −1 ξ1 3 ξ2 1 3 ξ3 0 3 0 0 J =   3 = ξ2   (18) (ii) Any linear map J : n → n which is an extension of a rank 1 CR-structure on n such that p is abelian is equivalent to a unique 0 0 ξ1 0 1 0 1 0 −1 0     , ξ1 1 ∈ R. (17) J has nonzero torsion. Proof. For any nonzero X ∈ p, (X, JpX) is a basis of p. In case (i), [X, JpX] 6= 0, since p is non abelian. Then [X, JpX] = µx3 , µ 6= 0, and x3 6∈ p since oth- erwise p would be abelian. One may extend Jp to n in the basis (X, JpX, µx3) as and J has zero torsion only if ξ1 3 = 0. In case (ii), necessarily x3 ∈ p since p is abelian. Hence (x3, Jpx3) is a basis for p. Take any linear extension J of Jp to n. There exists some eigenvector y1 6= 0 of J associated to some eigenvalue ξ1 1 ∈ R. Then y1 6∈ p, which implies [y1, Jx3] 6= 0, for otherwise y1 would commute to the whole of n and then be some multiple of x3 ∈ p. Hence [y1, Jx3] = λx3, λ 6= 0, and dividing y1 by λ one may suppose λ = 1. In the basis y1, y2 = Jx3, y3 = x3 one has J =   and (ii) follows. 0 0 ξ1 0 1 0 1 0 −1 0   (19) 7 Complex structures on n × n. We will use for commutation relations [x1, x2] = x5, [x3, x4] = x6. The au- tomorphisms of n × n fall into 2 kinds. The first kind is comprised of the 9 matrices Φ = b1 1 b2 1 0 0 b5 1 b6 1   b1 2 b2 2 0 0 b5 2 b6 2 0 0 b3 3 b4 3 b5 3 b6 3 0 0 b3 4 b4 4 b5 4 b6 4 0 0 0 0 b1 1b2 2 − b1 2b2 1 0 0 0 0 0 0   , b3 3b4 4 − b3 4b4 3 2 − b1 1b2 (b1 2b2 1)(b3 3b4 4 − b3 4b4 3) 6= 0. The second kind ones are Ψ = ΘΦ where Φ is first kind and Θ =   0 0 1 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 .   (20) (21) Lemma 5. Any integrable complex structure J on n × n is equivalent under some first kind automorphism to one of the following: (i) Sε(ξ5 5) = 0 −1 0 0 1 0 0 0 −1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0   0 0 0 0 0 0 0 0 5 −ε((ξ5 ξ5 5)2 + 1) −ξ5 ε 5   , ε = ±1 , ξ5 5 ∈ R. (22) 5) is equivalent to Sε(ξ5 Sε′(ξ ′5 second) kind automorphism if and only if ε′ = ε, ξ ′5 −ξ5 5) (ε, ε′ = ±1; ξ ′5 5, ξ5 5 ). 5 ∈ R) under some first (resp. 5 = 5 (resp. ε′ = −ε, ξ ′5 5 = ξ5 (ii) D(ξ1 1) = ξ1 1 0 1 0 0   0 −((ξ1 ξ1 1 0 1 0 1)2 + 1) 0 −ξ1 1 0 0 −((ξ1 0 1)2 + 1) 0 −ξ1 1 0 0 0 0 0 10 0 0 0 0 1 )2−1 2ξ1 1 1 2ξ1 1 (ξ1 0 0 0 0 1 )2+1)2 − ((ξ1 2ξ1 1 1)2 1−(ξ1 2ξ1 1 ξ1 1 ∈ R \ {0}. ,   (23) 1) is equivalent to D(ξ1 D(ξ ′1 1) (ξ ′1 automorphism if and only if ξ ′1 1, ξ1 1 = ξ1 1 ∈ R) under some first (resp. second) kind 1 (resp. ξ ′1 1 = −ξ1 1 ). 3ξ3 3 −ξ4 3 )2+1−ξ4 3 ξ3 3 3) = T (ξ3 3, ξ4 3ξ3 0 −ξ4 3 3 − (ξ3 1 −ξ3 ξ3 0 3 0 1 0 0 ξ3 3 ξ3 3 ξ4 3 0 3ξ3 ξ4 3 − 1 ξ3 3 −ξ3 3 0 0 (iii)   0 0 0 0 3 ξ3 − ξ4 3 ξ3 3 1 −1 0 0 0 0 3 ξ3 −2)ξ4 3+(ξ3 (ξ3 3)2 ξ4 3 ξ3 −1 3 ξ3 3 3 )2+1 − (ξ4 3 ξ3 3   ξ3 3 ∈ R \ {0}, ξ4 3 ∈ R. , (24) 0 0 0 0 3, ξ ′4 3) is equivalent to T (ξ3 T (ξ ′3 3 ∈ R \ {0}, ξ ′4 first (resp. second) kind automorphism if and only if ξ ′3 ξ ′3 3) (ξ ′3 3, ξ4 3, ξ3 3 = −ξ4 3). 3 = −ξ3 3, ξ ′4 3, ξ4 3 = ξ3 3 ∈ R.) under some 3 (resp. 3 = ξ4 3, ξ ′4 Finally, the cases (i),(ii), (iii) are mutually non equivalent, either under first or second kind automorphism. Proof. Let J = (ξi Denote J1 = (cid:16) ξ1 1 = (cid:18) ξ1 1 ξ1 ξ2 1 ξ2 ξ5 1 ξ5 2 ξ1 5 2 ξ2 5 2 ξ5 j)16i,j66 an integrable complex structure in the basis (xk)16k66. 1 ξ1 2 ξ2 1 ξ2 4(cid:17). Then 4(cid:17), J3 = (cid:16) ξ3 6(cid:19) are zero torsion linear maps from n to n, 2(cid:17), J4 = (cid:16) ξ3 2(cid:17), J2 = (cid:16) ξ1 3 = (cid:18) ξ3 3 ξ1 4 ξ2 3 ξ2 4 ξ3 6 4 ξ4 6 4 ξ6 1 ξ3 ξ4 3 ξ4 ξ6 3 ξ6 1 ξ3 2 ξ4 1 ξ4 3 ξ3 4 ξ4 3 ξ4 5(cid:19) and J ∗ J ∗ hence equivalent to type (11), (12) or (13) in lemma 3. It can be checked that their being of different types would contradict with J 2 = −1. Hence, mod- ulo equivalence under some first kind automorphism, we get 3 cases: case 1: 3 6= 0); 1 = D(ξ1 3 = D(ξ3 3), (ξ1 1), J ∗ 1, ξ3 J ∗ 1 = (cid:16) 0 −1 0 1 0 0 0 0 ξ5 5(cid:17), J ∗ 1 = (cid:18) 0 ξ1 1 0 0 0 0 ξ6 3 = (cid:16) 0 −1 0 5(cid:19), J ∗ 6(cid:17) ; case 2: J ∗ 3 = (cid:18) ξ3 3 0 0 ξ6 2 0 1 ξ2 2 0 0 0 ξ5 3 ξ4 3 0 −ξ3 0 0 6(cid:19), (ξ2 case 3: J ∗ 2, ξ3 3 6= 0). Case 1 (resp. 2, 3) leads to (22) (resp. (23), (24)) (see [1], programs "n2case1.red", "n2case2.red", "n2case3.red", and their outputs.) The assertion about equivalence in cases 1,2 are readily proved, as is equivalence under some first kind automorphism in case 3 and the nonequivalence of the 3 types. Consider now Θ T (ξ3 3)Θ−1. 3, ξ4 It is equivalent under some first kind automorphism to some T (η3 3, η4 3). That 3 (cid:17) are similar, which amounts to implies that the matrices (cid:16) ξ3 their having same trace and same determinant, i.e. η3 3) is equivalent to T (ξ3 T (ξ ′3 3. As 3) under some second kind automorphism if 0 (cid:17), (cid:16) 0 −η4 3, ξ4 3 = −ξ3 3 = −ξ4 3 η3 3 1 −ξ3 3, ξ ′4 3, η4 −ξ3 3 3 ξ4 3 11 and only if it is equivalent to Θ T (ξ3 phism, the assertion about second kind equivalence in case 3 follows. 3)Θ−1 under some first kind automor- 3, ξ4 Remark 4. In case 3, had we used J ∗ separate further into 2 subcases: subcase ξ1 2 6= 0: 3 = (cid:18) 0 ξ3 4 0 1 ξ4 4 0 0 0 ξ6 6(cid:19), then we would have to T (ξ1 2, ξ2 2) = 0 2 − ξ2 ξ1 2 ξ1 2 ξ1 2 +1 ξ2 1 2 ξ1 2 0 −ξ1 0 2 ξ2 1 1 2 0 0 0 −(ξ1 2 + 1) −ξ2 2 ξ1 2 −ξ2 2 0 0 0 0 0   0 0 0 0 − ξ1 2 +1 ξ2 2 ξ1 2 ξ2 2 0 0 0 0 − (ξ2 2 )2+(ξ1 2+1)2 ξ2 2 ξ1 2 ξ1 2 +1 ξ2 2   , ξ1 2ξ2 2 6= 0 ; subcase ξ1 2 = 0: T (ξ2 2) =   0 1 1 −ξ2 0 0 ξ2 2 0 2)2 + 1) 0 0 −1 0 0 0 1 0 0 0 1 −ξ2 0 2 − 1 0 0 ξ2 2 2 )2+1 0 − (ξ2 ξ2 2 0 0 0 0 0 1 ξ2 2 1 ξ2 2   , ξ2 2 6= 0 . 2 −((ξ2 0 0 Remark 5. Sε(ξ5 5) is abelian. Remark 6. If one looks for zero torsion linear maps instead of complex struc- tures, then J ∗ 3 may be of different types. 1 and J ∗ References [1] http://www.u-bourgogne.fr/monge/l.magnin/2examples/ 2examplesindex.html or http://www.u-bourgogne.fr/IMB/magnin/public html/2examples/ 2examplesindex.html [2] L. Magnin, Left invariant complex structures on U(2) and SU(2) × SU(2) revisited, preprint arXiv:0809.1182v2, 2008. 12 [3] L. Magnin, Complex structures on indecomposable 6-dimensional nilpo- tent real Lie algebras, Internat. J. Alg. Comput., 17, #1, 2007, 77-113. [4] G. Gigante, G. Tomassini, CR-structures on a real Lie algebra, Adv. Math., 94, 1992, 67-81. [5] D. M. Snow, Invariant complex structures on Lie groups, J. Reine Angew. Math., 371, 1986, 191-215. 13
1905.09343	1	1905	2019-05-22T19:34:17	Sectionally pseudocomplemented posets	[ "math.RA" ]	The concept of a sectionally pseudocomplemented lattice was introduced by I. Chajda as an extension of relative pseudocomplementation for not necessarily distributive lattices. The typical example of such a lattice is the non-modular lattice N5. The aim of this paper is to extend the concept of sectional pseudocomplementation from lattices to posets. At first we show that the class of sectionally pseudocompelemented lattices forms a variety of lattices which can be described by two simple identities. This variety has nice congruence properties. We summarize properties of sectionally pseudocomplemented posets and show differences to relative pseudocomplementation. We prove that every sectionally pseudocomplemented poset is completely L-semidistributive. We introduce the concept of congruence on these posets and show when the quotient structure becomes a poset again. Finally, we study the Dedekind-MacNeille completion of sectionally pseudocomplemented posets. We show that contrary to the case of relatively pseudocomplemented posets, this completion need not be sectionally pseudocomplemented but we present the construction of a so-called generalized ordinal sum which enables us to construct the Dedekind-MacNeille completion provided the completions of the summands are known.	math.RA	math	Sectionally pseudocomplemented posets Ivan Chajda, Helmut Langer and Jan Paseka Abstract The concept of a sectionally pseudocomplemented lattice was introduced in [3] as an extension of relative pseudocomplementation for not necessarily distribu- tive lattices. The typical example of such a lattice is the non-modular lattice N5. The aim of this paper is to extend the concept of sectional pseudocomple- mentation from lattices to posets. At first we show that the class of sectionally pseudocompelemented lattices forms a variety of lattices which can be described by two simple identities. This variety has nice congruence properties. We summarize properties of sectionally pseudocomplemented posets and show differences to rela- tive pseudocomplementation. We prove that every sectionally pseudocomplemented poset is completely L-semidistributive. We introduce the concept of congruence on these posets and show when the quotient structure becomes a poset again. Finally, we study the Dedekind-MacNeille completion of sectionally pseudocomplemented posets. We show that contrary to the case of relatively pseudocomplemented posets, this completion need not be sectionally pseudocomplemented but we present the construction of a so-called generalized ordinal sum which enables us to construct the Dedekind-MacNeille completion provided the completions of the summands are known. AMS Subject Classification: 06A11, 06D15, 06B23 Keywords: Sectional pseudocomplementation, poset, congruence, Dedekind-MacNeille completion, generalized ordinal sum. 1 Introduction The concept of relative pseudocomplemented lattices was introduced by R. P. Dilworth in [7]. The usefulness of this concept was shown in numerous papers and books, see e.g. the famous paper [1] by R. Balbes and the monograph [2] by G. Birkhoff. This concept was extended to posets recently by the first and second author in [6]. Relatively pseudocom- plemented lattices turn out to be distributive, a property which also holds for relatively pseudocomplemented posets (see [6]). In order to extend relative pseudocomplementation in lattices to the non-distributive case, sectional pseudocomplementation was introduced in [3]. The aim of the present paper is to extend sectional pseudocomplementation to posets which, of course, need not be distributive. The concept of a sectionally pseudocomplemented lattice was introduced by the first author in [3]. Recall that a lattice (L, ∨, ∧) is sectionally pseudocomplemented if for all 1Support of the research by OAD, project CZ 02/2019, and support of the research of the first author by IGA, project PrF 2019 015, is gratefully acknowledged. 1 a, b ∈ L there exists the pseudocomplement of a ∨ b with respect to b in [b, 1], in other words, there exists a greatest element c of L satisfying (a ∨ b) ∧ c = b. In this case c is called the sectional pseudocomplement of a with respect to b and it will be denoted by a ∗ b. The aim of this paper is to extend this concept to posets. 2 Properties of sectionally pseudocomplemented po- sets and lattices Let (P, ≤) be a poset, a, b ∈ P and A, B ⊆ P . Recall that L(A) := {x ∈ P x ≤ y for all y ∈ A}, U(A) := {x ∈ P y ≤ x for all y ∈ A}. Instead of L({a}), L({a, b}), L(A∪{a}), L(A∪B), L(U(A)) we simply write L(a), L(a, b), L(A, a), L(A, B), LU(A), respectively. Analogously we proceed in similar cases. We also put ↓(A) = {x ∈ P x ≤ y for some y ∈ A}. We start with the following definition. Definition 2.1. A poset P = (P, ≤) is called sectionally pseudocomplemented if for all a, b ∈ P there exists a greatest c ∈ P satisfying L(U(a, b), c) = L(b). This element c is called the sectional pseudocomplement a ∗ b of a with respect to b. The poset P is called strongly sectionally pseudocomplemented if it is sectionally pseudocomplemented, it has a greatest element 1 and it satisfies the condition x ≤ (x ∗ y) ∗ y (which, as we will se later, is equivalent to the identity x ∗ ((x ∗ y) ∗ y) ≈ 1). The following example shows that there really exist sectionally pseudocomplemented posets which are not strongly sectionally pseudocomplemented. Hence, we cannot expect that every sectionally pseudocomplemented poset satisfies the condition x ≤ (x ∗ y) ∗ y. Example 2.2. The poset visualized in Fig. 1 e b ✉ ❅ ✉ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ✉ ❅ ✉ g d 1 ✉ ❅ ❅ ❅ f ✉ ❅ ❅ c ✉ ❅ ✉ a Fig. 1 2 is sectionally pseudocomplemented and ∗ has the operation table ∗ a b c d e f g 1 a 1 1 c 1 1 1 1 1 b g 1 g g 1 1 g 1 c f f 1 f 1 f 1 1 d e e 1 e 1 1 1 e d f g d 1 f g 1 c g e 1 g 1 f a e g b e f e f 1 1 c d e f g 1 1 a b e b but it is not strongly sectionally pseudocomplemented since c 6≤ a = f ∗ a = (c ∗ a) ∗ a. Recall from [6] or [8] that the relative pseudocomplement of a with respect to b is the greatest d ∈ P satisfying L(a, d) ⊆ L(b). We are going to show that every sectionally pseudocomplemented lattice with 1 is strongly sectionally pseudocomplemented. The following lemma was proved in [3]. For the convenience of the reader we provide a short proof. Lemma 2.3. Every sectionally pseudocomplemented lattice L = (L, ∨, ∧, ∗) satisfies x ∨ y ≤ (x ∗ y) ∗ y. Proof. Because of (x ∨ y) ∧ y = y we have y ≤ x ∗ y and hence ((x ∗ y) ∨ y) ∧ (x ∨ y) ≈ (x ∨ y) ∧ (x ∗ y) = y whence x ∨ y ≤ (x ∗ y) ∗ y. In a lattice (P, ∨, ∧) the equation L(U(a, b), c) = L(b) is equivalent to (a ∨ b) ∧ c = b. Example 2.4. The poset visualized in Fig. 2 1 ✉ ❏ c a ✉ ✉ ❅ ❅ ❅ ❅ ✡ ✉ 0 ❏ ❏ ❏ ✡ ✡ ✡ ❏ ✡ ❏ ✡ ✉ b is a strongly sectionally pseudocomplemented lattice and ∗ has the operation table Fig. 2 ∗ 0 a b c 1 0 1 1 1 1 1 a b 1 b 1 1 c a 1 c 1 b c b a b 1 1 c 1 1 0 a b 3 but this poset is not relatively pseudocomplemented since the relative pseudocomplement of c with respect to a does not exist. It was shown in [3] that the class of sectionally pseudocomplemented lattices forms a variety. However, the defining identities given in [3] are rather complicated. We present some simpler identities as follows. Theorem 2.5. The class of sectionally pseudocomplemented lattices forms a variety which besides the lattice axioms is determined by the following identities: (i) z ∨ y ≤ x ∗ ((x ∨ y) ∧ (z ∨ y)), (ii) (x ∨ y) ∧ (x ∗ y) ≈ y. Proof. Let L = (L, ∨, ∧) be a lattice and a, b, c ∈ L. First assume L to be sectionally pseudocomplemented. If d := (a ∨ b) ∧ (c ∨ b) then a ∨ d = a ∨ ((a ∨ b) ∧ (c ∨ b)) = a ∨ (b ∨ ((a ∨ b) ∧ (c ∨ b))) = = (a ∨ b) ∨ ((a ∨ b) ∧ (c ∨ b)) = a ∨ b, c ∨ d = c ∨ ((a ∨ b) ∧ (c ∨ b)) = c ∨ (b ∨ ((a ∨ b) ∧ (c ∨ b))) = = (c ∨ b) ∨ ((a ∨ b) ∧ (c ∨ b)) = c ∨ b and hence d ≤ (a ∨ d) ∧ (c ∨ d) = (a ∨ b) ∧ (c ∨ b) = d, i.e. (a ∨ d) ∧ (c ∨ d) = d which shows c ∨ b = c ∨ d ≤ a ∗ d = a ∗ ((a ∨ b) ∧ (c ∨ b)) proving (i). Identity (ii) follows from the definition of ∗. Conversely, assume L to satisfy (i) and (ii). Then (a ∨ b) ∧ (a ∗ b) = b according to (ii). If (a ∨ b) ∧ c = b then b ≤ c and hence (a ∨ b) ∧ (c ∨ b) = (a ∨ b) ∧ c = b whence c = c ∨ b ≤ a ∗ ((a ∨ b) ∧ (c ∨ b)) = a ∗ b according to (i). This shows that a∗b is the sectional pseudocomplement of a with respect to b. We can prove that this variety has very strong congruence properties. Recall that an algebra A is called arithmetical if Θ◦Φ = Φ◦Θ for all Θ, Φ ∈ Con A and if the congruence lattice of A is distributive. (Here and in the following Con A denotes the set of all congruences on A.) Moreover, recall that an algebra A with 1 is called weakly regular (see e.g. [4]) if for arbitrary Θ, Φ ∈ Con A we have that [1]Θ = [1]Φ implies Θ = Φ. A variety V (with 1) is called arithmetical or weakly regular if every of its members has the respective property. Theorem 2.6. The variety V of sectionally pseudocomplemented lattices with 1 is arith- metical and weakly regular. 4 Proof. Since every member of V is a lattice, V is congruence distributive. Moreover, since for p(x, y, z) := ((x ∗ y) ∗ z) ∧ ((z ∗ y) ∗ x). we have p(x, x, y) ≈ ((x ∗ x) ∗ y) ∧ ((y ∗ x) ∗ x) ≈ (1 ∗ y) ∧ ((y ∗ x) ∗ x) ≈ y ∧ ((y ∗ x) ∗ x) ≈ y, p(y, x, x) ≈ ((y ∗ x) ∗ x) ∧ ((x ∗ x) ∗ y) ≈ ((y ∗ x) ∗ x) ∧ (1 ∗ y) ≈ ((y ∗ x) ∗ x) ∧ y ≈ y, V is congruence permutable. Finally, since for t1(x, y) := x ∗ y and t2(x, y) := y ∗ x we have that t1(x, y) = t2(x, y) = 1 is equivalent to x = y, V is weakly regular (cf. [4]). Evidently, every relatively pseudocomplemented lattice (L, ∨, ∧) is sectionally pseudo- complemented since for a, b ∈ L we have a ∗ b = (a ∨ b) ◦ b where ∗ and ◦ denote sectional and relative pseudocomplementation, respectively. (Observe that (a ∨ b) ∧ b = b and hence (a ∨ b) ∧ ((a ∨ b) ◦ b) = b.) The following poset is an example of a sectionally pseudocomplemented poset which is neither a lattice nor relatively pseudocomplemented. Example 2.7. The poset visualized in Fig. 3 d c a 1 ✉ ❅ ❅ ❅ ❅ ✉ ✉ ❅ ❅ ✟✟✟✟✟✟✟✟ ❅ ❅ ✉ ❅ ❅ ❅ ❅ ✉ ✉ ❅ ❅ ❅ ❅ ✉ 0 e b is strongly sectionally pseudocomplemented and the operation table of ∗ looks as follows: Fig. 3 ∗ 0 a b c d e 1 0 1 1 1 1 1 1 1 a b 1 b 1 1 1 1 c a 1 c 1 1 1 b b a b 1 1 1 1 c d 0 a b c 1 e 1 e 0 a b c d 1 1 1 0 a b c d e 1 Evidently, this poset is not a lattice. However, it is also not relatively pseudocomplemented since the relative pseudocomplement of c with respect to a does not exist. 5 In the following we list several important properties of sectionally pseudocomplemented posets. Theorem 2.8. Let P = (P, ≤, ∗, 1) be a sectionally pseudocomplemented poset with 1. Then the following hold: (i) x ≤ y if and only if x ∗ y = 1, (ii) x ∗ x ≈ x ∗ 1 ≈ 1, (iii) 1 ∗ x ≈ x, (iv) x ∗ (y ∗ x) ≈ 1, (v) x ∗ ((y ∗ x) ∗ x) ≈ 1, (vi) if x ∗ y = 1 or y ∗ x = 1 then x ∗ ((x ∗ y) ∗ y) = 1, (vii) if x ∗ y = 1 then (y ∗ z) ∗ (x ∗ z) = 1, (viii) L(U(x, y), x ∗ y) ≈ L(y). Proof. Let a, b, c ∈ P . (i) The following are equivalent: (ii) follows from (i). (iii) The following are equivalent: a ≤ b, U(a, b) = U(b), LU(a, b) = L(b), L(U(a, b), 1) = L(b), a ∗ b = 1. L(U(1, a), b) = L(a), L(b) = L(a), b = a. (iv) We have L(U(a, b), b) = L(b) implies b ≤ a ∗ b. (v) Because of (iv) we have L(U(b ∗ a, a), a) = L(b ∗ a, a) = L(a) which shows a ≤ (b ∗ a) ∗ a. (vi) If a ≤ b then a ≤ b = 1 ∗ b = (a ∗ b) ∗ b according to (iii) and (i), and if b ≤ a then L(U(a ∗ b, b), a) = L(a ∗ b, a) = L(U(a, b), a ∗ b) = L(b) and hence a ≤ (a ∗ b) ∗ b. (vii) If a ≤ b then L(c) ⊆ L(U(a, c), b ∗ c) ⊆ L(U(b, c), b ∗ c) = L(c), i.e. L(U(a, c), b ∗ c) = L(c) whence b ∗ c ≤ a ∗ c. (viii) follows from the definition of ∗. 6 Remark 2.9. Assertion (vii) of Theorem 2.8 says that ∗ is antitone in the first variable, i.e. x ≤ y implies y ∗ z ≤ x ∗ z. Contrary to the case of relatively pseudocomplemented posets, sectional pseudocomplementation is not monotone in the second variable. Namely, in Example 2.7 we have 0 ≤ a, but b ∗ 0 = c 6≤ a = b ∗ a. However, ∗ need not be monotone in the second variable also in sectionally pseudocomplemented lattices as the following example shows. Example 2.10. The lattice visualized in Fig. 4 1 ✉ ❆ ❆ ❆ ❆ ✉ e b ✉ ❅ ❅ ✉ c ❅ ❅ ✉ a ✁ ✁ ✁ ✁ ✉ 0 ❆ ❆ ✁ ✁ ❆ ❆ ✁ ✁ ✉ d is sectionally pseudocomplemented and ∗ has the operation table Fig. 4 ∗ 0 a b c d e 1 0 1 1 1 1 1 1 1 a d 1 1 1 d 1 1 b d c 1 c d 1 1 b 1 d 1 1 c d b d e a b c 1 e 1 e d a b c d 1 1 c d e 1 1 0 a b Here we have 0 < a and b ∗ 0 = d k c = b ∗ a. For every algebra (A, ∗, 1) of type (2, 0) and every subset B of A put L(B) := {x ∈ A x ∗ y = 1 for all y ∈ B}, U(B) := {x ∈ A y ∗ x = 1 for all y ∈ B} We are going to show that sectionally pseudocomplemented posets can be defined as certain groupoids. Theorem 2.11. An algebra (A, ∗, 1) of type (2, 0) can be organized into a sectionally pseudocomplemented poset with 1 if and only if the following hold: (i) x ∗ x ≈ 1 7 (ii) x ∗ y = y ∗ x = 1 ⇒ x = y, (iii) x ∗ y = y ∗ z = 1 ⇒ x ∗ z = 1, (iv) L(U(x, y), x ∗ y) = L(y), (v) L(U(x, y), z) = L(y) ⇒ z ∗ (x ∗ y) = 1. Proof. The necessity of the conditions is clear. Conversely, assume (i) -- (v) to hold. Define a binary relation ≤ on A by x ≤ y if x ∗ y = 1 (x, y ∈ A). Now (i) implies reflexivity of ≤, (ii) implies antisymmetry of ≤, (iii) implies transitivity of ≤, (iv) and (v) imply that x ∗ y is the sectional pseudocomplement of x with respect to y. Hence (A, ≤) is a sectionally pseudocomplemented poset with sectional pseudocomple- mentation ∗. Recall that a lattice (L, ∨, ∧) is called completely meet-semidistributive if the following holds: If ∅ 6= M ⊆ L, a, b ∈ L and x ∧ a = b for all x ∈ M then (_ M) ∧ a = b. For posets, we modify this concept as follows. Definition 2.12. A poset (P, ≤) is called completely L-semidistributive if the following holds: If ∅ 6= M ⊆ P, a, b ∈ P and L(x, a) = L(b) for all x ∈ M then L(U(M), a) = L(b). Theorem 2.13. Let P = (P, ≤, ∗) be a sectionally pseudocomplemented poset. Then P is completely L-semidistributive. Proof. If ∅ 6= M ⊆ P, a, b ∈ P and L(x, a) = L(b) for all x ∈ M then b ≤ a and therefore L(U(a, b), x) = L(a, x) = L(b) for all x ∈ M and hence x ≤ a ∗ b for all x ∈ M whence a ∗ b ∈ U(M) which finally implies L(b) ⊆ L(U(M), a) ⊆ L(a ∗ b, a) = L(U(a, b), a ∗ b) = L(b), i.e. L(U(M), a) = L(b). 3 Congruences in sectionally pseudocomplemented posets Theorem 2.8 (i) shows that in a sectionally pseudocomplemented poset (P, ≤, ∗, 1) with 1, ≤ is uniquely determined by ∗. Let (P, ≤, ∗, 1) be a sectionally pseudocomplemented poset with 1 and Θ ∈ Con(P, ∗). We are interested in the question when (P/Θ, ≤′) is a poset where ≤′ is defined by [x]Θ ≤′ [y]Θ if [x]Θ ∗ [y]Θ = [1]Θ (x, y ∈ P ). We will see that this is the case if Θ is convex, i.e. every class of Θ is a convex subset of (P, ≤). First we show that if (P, ≤, ∗, 1) is a finite sectionally pseudocomplemented poset with 1 then (P, ∗) has convex congruences. In the following lemma and theorem we frequently use Theorem 2.8 (vi). 8 Lemma 3.1. Let (P, ≤, ∗, 1) be a sectionally pseudocomplemented poset with 1 satisfying the Ascending Chain Condition, let a, b ∈ P and Θ ∈ Con(P, ∗) and assume a < b < (b ∗ a) ∗ a and a Θ (b ∗ a) ∗ a. Then a Θ b. Proof. Assume (a, b) /∈ Θ. Put a1 := a, a2 := b and an := (an−1 ∗ an−2) ∗ an−2 for n ≥ 3. Then a1 < a2 < a3 and a3 Θ a1. Now a4 = (a3 ∗ a2) ∗ a2 Θ (a1 ∗ a2) ∗ a2 = 1 ∗ a2 = a2. Moreover, a3 ≤ a4. Now a3 = a4 would imply a1 Θ a3 = a4 Θ a2, a contradiction. This shows a3 < a4. Now a5 = (a4 ∗ a3) ∗ a3 Θ (a2 ∗ a3) ∗ a3 = 1 ∗ a3 = a3. Moreover, a4 ≤ a5. Now a4 = a5 would imply a1 Θ a3 Θ a5 = a4 Θ a2, a contradiction. This shows a4 < a5. Going on in this way we would obtain an infinite strictly ascending chain a1 < a2 < a3 < a4 < a5 < · · · contradicting the Ascending Chain Condition. This shows a Θ b. Hence, we conclude Theorem 3.2. Let (P, ≤, ∗, 1) be a sectionally pseudocomplemented poset with 1 satisfy- ing the Ascending Chain Condition and let Θ ∈ Con(P, ∗). Then Θ is convex. Proof. Assume a, b, c ∈ P , a < b < c and (a, c) ∈ Θ. Then b ≤ (b ∗ a) ∗ a. If b = (b ∗ a) ∗ a then b = (b ∗ a) ∗ a Θ (b ∗ c) ∗ a = 1 ∗ a = a. If b < (b ∗ a) ∗ a then a Θ b according to Lemma 3.1. Corollary 3.3. If (P, ≤, ∗, 1) is a finite sectionally pseudocomplemented poset with 1 then (P, ∗) has convex congruences. For the infinite case we have the following result. Theorem 3.4. Let (P, ≤, ∗, 1) be a sectionally pseudocomplemented poset with 1 such that x, y ∈ P , x < y < 1, x 6≺ y and x < y ∗ x together imply Θ(x, y) = P 2. Then (P, ∗) has convex congruences. Proof. Let Θ ∈ Con(P, ∗) and a, b, c ∈ P and assume a < b < c and (a, c) ∈ Θ. If c = 1 then a Θ 1 = a ∗ b Θ 1 ∗ b = b. If c < 1 and a < c ∗ a then Θ(a, c) = P 2 and hence Θ = P 2 which implies a Θ b. If c < 1 and a = c ∗ a then a = 1 ∗ a = (a ∗ a) ∗ a Θ (c ∗ a) ∗ a = a ∗ a = 1 = a ∗ b = (c ∗ a) ∗ b Θ (a ∗ a) ∗ b = 1 ∗ b = b. This shows that Θ is convex. Let (P, ≤, ∗, 1) be a sectionally pseudocomplemented poset with 1 and Θ ∈ Con(P, ∗). We define a binary relation ≤′ on P/Θ by [x]Θ ≤′ [y]Θ if [x]Θ ∗ [y]Θ = [1]Θ (x, y ∈ P ). Now we can prove 9 Theorem 3.5. Let (P, ≤, ∗, 1) be a strongly sectionally pseudocomplemented poset and let a, b ∈ P and Θ a convex congruence on (A, ∗). Then the following hold: (i) If [a]Θ ≤′ [b]Θ then there exists some d ∈ [b]Θ with a ≤ d, (ii) if a ≤ b then [a]Θ ≤′ [b]Θ, (iii) (P/Θ, ≤′) is a poset. Proof. (i) Put d := (a ∗ b) ∗ b. Then d = (a ∗ b) ∗ b Θ 1 ∗ b = b and a ≤ (a ∗ b) ∗ b = d. (ii) If a ≤ b then a∗b = 1 according to Theorem 2.8 and hence a∗b Θ 1, i.e. [a]Θ ≤′ [b]Θ. (iii) Obviously, ≤′ is reflexive. Now assume [a]Θ ≤′ [b]Θ ≤′ [a]Θ. Then, by (i), there exists some d ∈ [b]Θ with a ≤ d. Because of [d]Θ = [b]Θ ≤′ [a]Θ there exists some e ∈ [a]Θ with d ≤ e. Since a ≤ d ≤ e, (a, e) ∈ Θ and Θ is convex we conclude (a, d) ∈ Θ showing [a]Θ = [d]Θ = [b]Θ proving antisymmetry of ≤′. Finally, let c ∈ P and assume [a]Θ ≤′ [b]Θ ≤′ [c]Θ. Then, by (i) there exists some f ∈ [b]Θ with a ≤ f and because of [f ]Θ = [b]Θ ≤′ [c]Θ some g ∈ [c]Θ with f ≤ g. Because of a ≤ f ≤ g we have a ≤ g which implies [a]Θ ≤′ [g]Θ = [c]Θ by (ii), proving transitivity of ≤′. Lemma 3.6. Let P = (P, ≤, ∗, 1) be a strongly sectionally pseudocomplemented poset, a ∈ P and Θ ∈ Con(P, ∗). Then [a]Θ is up-directed. Hence, if P satisfies the Ascending Chain Condition, then [a]Θ has a greatest element. Proof. If b, c ∈ [a]Θ then (b ∗ c) ∗ c ∈ [(c ∗ c) ∗ c]Θ = [1 ∗ c]Θ = [c]Θ = [a]Θ, b ≤ (b ∗ c) ∗ c since P is strongly sectionally pseudocomplemented, and c ≤ (b ∗ c) ∗ c according to Theorem 2.8 (v). From Theorem 3.5 we have: If (P, ≤, ∗, 1) is a strongly sectionally pseudocomplemented poset, Θ a convex congruence on (P, ∗), a the greatest element of [a]Θ and b the greatest element of [b]Θ then a ≤ b if and only if [a]Θ ≤′ [b]Θ. Now we solve the problem for which Θ ∈ Con(P, ∗) the quotient P/Θ is again sectionally pseudocomplemented. We can state a sufficient condition. Definition 3.7. Let P = (P, ≤, ∗) be a sectionally pseudocomplemented poset and Θ ∈ Con(P, ∗). We say that Θ is strong if the following holds: If a, b ∈ P , a is the greatest element of [a]Θ and b the greatest element of [b]Θ then a ∗ b is the greatest element of [a ∗ b]Θ. In this case we define [a]Θ ∗′ [b]Θ := [a ∗ b]Θ. It is easy to see that if Θ is strong, a, b, c, d ∈ P , a ≤ b and c is the greatest element of [a]Θ and d the greatest element of [b]Θ then c ≤ d. 10 Theorem 3.8. Let P = (P, ≤, ∗, 1) be a strongly sectionally pseudocomplemented poset and Θ ∈ Con(P, ∗) a strong congruence. If P satisfies the Ascending Chain Condition then (P/Θ, ≤′, ∗′, [1]Θ) is a strongly sectionally pseudocomplemented poset. Proof. Since P satisfies the Ascending Chain Condition we know from Theorems 3.2 and 3.5 that Θ is convex and (P/Θ, ≤′) a poset. Moreover, from Lemma 3.6 we have that any congruence class of Θ has a greatest element. Put Q := {x ∈ P x is the greatest element of [x]Θ}. Then (Q, ∗, 1) is a subalgebra of (P, ∗, 1). Assume a, b ∈ Q. Since P is a strongly sectionally pseudocomplemented poset and Θ is strong we have a, b ≤ (a ∗ b) ∗ b and a ∗ b, (a ∗ b) ∗ b ∈ Q. This implies LQ(b) ⊆ LQ(UQ(a, b), a ∗ b) ⊆ LQ(UQ((a ∗ b) ∗ b), a ∗ b) = LQ((a ∗ b) ∗ b, a ∗ b) = = L((a ∗ b) ∗ b, a ∗ b) ∩ Q = L(U(a ∗ b, b), (a ∗ b) ∗ b) ∩ Q = L(b) ∩ Q = LQ(b). Note that the first inclusion follows from the fact that b ≤ a ∗ b and b ∈ LQUQ(a, b). The second inclusion follows from the fact that LQUQ(a, b) ⊆ LQUQ((a ∗ b) ∗ b). The first equality follows from LQUQ((a ∗ b) ∗ b) = LQ((a ∗ b) ∗ b). The second equality follows from the definition of LQ. Since P is sectionally pseudocomplemented we have the next equality. Now, let c ∈ Q be such that LQ(UQ(a, b), c) = LQ(b). We have UQ(a, b) = U(a, b) ∩ Q ⊆ U(a, b). Hence LU(a, b) ⊆ LUQ(a, b) = ↓(LQUQ(a, b)). The last equality follows from the fact that LQUQ(a, b) ⊆ LUQ(a, b) which yields ↓(LQUQ(a, b)) ⊆ ↓(LUQ(a, b)) = LUQ(a, b) and from the fact that y ∈ LUQ(a, b) implies y ≤ x ∈ LQUQ(a, b) where x is the greatest element of [y]Θ. We obtain L(b) ⊆ L(U(a, b), c) ⊆ ↓(LQUQ(a, b)∩LQ(c)) = ↓(LQ(b)) = L(b) since b, c ∈ Q. Since P is a sectionally pseudocomplemented poset we have c ≤ a ∗ b, i.e., (Q, ∗, 1) is sectionally pseudocomplemented. Since (Q, ∗, 1) is a subalgebra of (P, ∗, 1) we have that (Q, ∗, 1) is strongly sectionally pseudocomplemented. Moreover, x 7→ [x]Θ an iso- morphism from (Q, ≤, ∗, 1) to (P/Θ, ≤′, ∗′, [1]Θ) and hence (P/Θ, ≤′, ∗′, [1]Θ) is also a strongly sectionally pseudocomplemented poset. The following lemma shows that in a strongly sectionally pseudocomplemented poset all principal congruences are given by the congruences of the form Θ(c, 1). Lemma 3.9. Let (P, ≤, ∗, 1) be a strongly sectionally pseudocomplemented poset, assume that every congruence on (P, ∗) is convex and let a, b ∈ P with a ≤ b . Then Θ(b ∗ a, 1) = Θ(a, b). Proof. Since (a, b) ∈ Θ(a, b) yields (b ∗ a, 1) = (b ∗ a, b ∗ b) ∈ Θ(a, b), we have Θ(b ∗ a, 1) ⊆ Θ(a, b). Conversely, (b∗a, 1) ∈ Θ(b∗a, 1) yields ((b∗a)∗a, a) = ((b∗a)∗a, 1∗a) ∈ Θ(b∗a, 1) which because of a ≤ b ≤ (b∗a)∗a and the convexity of Θ(b∗a, 1) implies (a, b) ∈ Θ(b∗a, 1), i.e. Θ(a, b) ⊆ Θ(b ∗ a, 1). 11 4 Completion of sectionally pseudocomplemented posets Now we consider the Dedekind-MacNeille completion of sectionally pseudocomplemented posets. It was shown by Y. S. Pawar ([8]) that the Dedekind-MacNeille completion DM(Q) of a relatively pseudocomplemented poset Q is relatively pseudocomplemented and that the relative pseudocomplementation in DM(Q) extends the relative pseudocomplementation in Q if Q is canonically embedded into DM(Q). In contrast to this the Dedekind- MacNeille completion of a strongly sectionally pseudocomplemented poset P need not be sectionally pseudocomplemented, even if P is finite and has a greatest element. Example 4.1. Though the poset visualized in Fig. 5 ❅ ❅ 1 ✉ ❅ ✉ b Fig. 5 ✉ a ✉ ❅ c is strongly sectionally pseudocomplemented with ∗ a b c 1 a 1 b c 1 b a 1 c 1 c a b 1 1 1 a b c 1 as the operation table for ∗, its Dedekind-MacNeille completion visualized in Fig. 6 a ✉ ❅ ❅ ✉ b ❅ ❅ ✉ c 1 ✉ ❅ ❅ ❅ ❅ ✉ 0 is not sectionally pseudocomplemented since a ∗ 0 does not exist. Fig. 6 For our next investigations, we introduce the following useful concepts. Definition 4.2. Let (I, ≤) be a chain with greatest element ⊤ and smallest element ⊥. Let Pi = (Pi, ≤i), i ∈ I, be a family of posets such that P⊤ has a greatest element 1 and such that the following hold: (i) If i ∈ I then there are a, b ∈ Pi with a < b, 12 (ii) if i, j, k ∈ I and i < k < j then Pi ∩ Pj = ∅, (iii) if i, j ∈ I and i < j then Pi ∩ Pj ≤ 1, (iv) if i, j ∈ I, i < j and Pi ∩ Pj = {a} then Pi = LPi(a) and Pj = UPj (a). Put P = Si∈I Pi. For a, b ∈ P , say a ∈ Pi and b ∈ Pj with i, j ∈ I, define a ≤ b if and only if a = b or (i = j and a ≤i b) or i < j. We call P = (P, ≤) the generalized ordinal sum of Pi, i ∈ I. It is elementary that P is a poset with a greatest element 1. Now, we can state some sufficient conditions under which the Dedekind-MacNeille com- pletion of a sectionally pseudocomplemented poset is sectionally pseudocomplemented. By [9] the Dedekind-MacNeille completion of a poset P is (up to isomorphism) any com- plete lattice Q into which P can be supremum-densely and infimum-densely embedded (i.e., for every element x ∈ Q there exist M, N ⊆ P such that x = W ϕ(M) = V ϕ(N), where ϕ : P → Q is the embedding). We usually identify P with ϕ(P ). In this sense Q preserves all infima and suprema existing in P. Now we turn our attention to a notion of a DM-yoked family of a generalized ordinal sum. The importance of this concept is based on the fact that under natural assumptions (which are e.g. satisfied for a finite index set I) the Dedekind-MacNeille completion of a generalized ordinal sum will be isomorphic to a generalized ordinal sum of the respective DM-yoked family. Definition 4.3. Let P = (P, ≤) be the generalized ordinal sum of Pi = (Pi, ≤i), i ∈ I. We say that a family Qi = (Qi, ≤i), i ∈ I, of posets is a DM-yoked family of P if the following conditions are satisfied: (y1) Pi is a sub-poset of Qi such that Qi is the Dedekind-MacNeille completion of Pi for every i ∈ I, (y2) if i, j, k ∈ I and i < k < j then Qi ∩ Qj = ∅, (y3) if i, j ∈ I and i < j then Qi ∩ Qj ≤ 1, (y4) if i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pi has a greatest element and Pj has a smallest element then Qi ∩ Qj = ∅, (y5) if i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pi does not have a greatest element and Pj has a smallest element 0Pj then 0Pj is the greatest element of Qi, (y6) if i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pj does not have a smallest element and Pi has a greatest element 1Pi then 1Pi is the smallest element of Qj, (y7) if i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pj does not have a smallest element and Pi does not have a greatest element then the greatest element 1Qi of Qi is the smallest element 0Qj of Qj, 13 (y8) if i, j ∈ I, i < j and Qi ∩ Qj = {a} then a is the greatest element of Qi and the smallest element of Qj. The question when there exists a DM-yoked family for a given poset P which is a gener- alized ordinal sum of posets Pi = (Pi, ≤i), i ∈ I, is positively answered in the following series of lemmas under the natural assumption that Pj ∩ (DM(Pi) × {i}) = ∅ for all i, j ∈ I. We will first need the following definition. Definition 4.4. Let P = (P, ≤) be the generalized ordinal sum of Pi = (Pi, ≤i), i ∈ I. We say that a family Ri = (Ri, ≤i), i ∈ I, of posets is a DM-related family of P if the following conditions are satisfied: (r1) Pi is a sub-poset of Ri such that Ri is the Dedekind-MacNeille completion of Pi for every i ∈ I, (r2) if i ∈ I and x ∈ Ri then x ∈ Ri \ Pi if and only if x = (A, i) and A is a non-principal cut of DM(Pi). Lemma 4.5. Let P = (P, ≤) be the generalized ordinal sum of Pi = (Pi, ≤i), i ∈ I, such that Pi ∩ (DM(Pi) × {i}) = ∅. Then a DM-related family Ri = (Ri, ≤i), i ∈ I, of P exists. Proof. Let i ∈ I. We put Ri := Pi ∪ ({A ∈ DM(Pi) A is not a principal cut in DM(Pi)} × {i}). We have Pi ∩ ({A ∈ DM(Pi) A is not a principal cut in DM(Pi)} × {i}) = ∅. Define a mapping κi : DM(Pi) → Ri as follows: κ(A) := (cid:26) a (A, i) if A = L(a) for some a ∈ A if A is not a principal cut in DM(Pi) for all A ∈ DM(Pi). Clearly, κi is a bijection. Let x, y ∈ Ri. We define x ≤i y if and only if κ−1 i (x). Then (Ri, ≤i), i ∈ I, is a poset containing Pi isomorphic with the poset DM(Pi). Hence the family Ri = (Ri, ≤i), i ∈ I, is DM-related. i (x) ⊆ κ−1 Lemma 4.6. Let P = (P, ≤) be the generalized ordinal sum of Pi = (Pi, ≤i), i ∈ I, such that Pj∩(DM(Pi)×{i}) = ∅ for all i, j ∈ I. Then a DM-yoked family Qi = (Qi, ≤i), i ∈ I, of P exists. Proof. Let Ri = (Ri, ≤i), i ∈ I, be the DM-related family of P which exists by Lemma 4.5. We will proceed in two steps. Step 1: Let i ∈ I. Assume that there exists some j ∈ I with i ≺ j and Pi ∩ Pj = ∅. If Pi does not have a greatest element we put Si := (Ri \ {1Ri}) ∪ {0Rj } such that 0Rj will be the greatest element of Si and the order ≤i on Si restricted to Ri \ {1Ri} will be the restriction of the order on Ri. Clearly Pi ⊆ Si and Pi is a sub-poset of Si. If Pi does have a greatest element, we put Si := Ri. If Pi ∩ Pj = {a} then a = 1Pi = 0Pj . Hence also a = 1Ri = 0Rj and we put Si := Ri. If there is no j ∈ I such that i ≺ j, we put again Si := Ri. 14 Step 2: Let j ∈ I. Assume there exists some i ∈ I with i ≺ j and Pi ∩ Pj = ∅. If Pj does not have a smallest element, we put Qj := (Sj \ {0Sj }) ∪ {1Si} such that 1Si will be the smallest element of Qj and the order ≤j on Qj restricted to Sj \ {0Sj } will be the restriction of the order on Sj. Clearly Pj ⊆ Qj and Pj is a sub-poset of Qj. Otherwise we always put Qj := Sj. Let us now check that Qi = (Qi, ≤i), i ∈ I, is a DM-yoked family of P. (y1): This follows immediately from the definition of Qi. (y2): Let i, j, k ∈ I and assume i < k < j. We always have Pi ∩ Pj = ∅ and hence also Ri ∩ Rj = ∅. Step 1: Assume that a ∈ Si ∩ Sj for some a. Then a 6∈ Ri or a 6∈ Rj. Suppose first that a 6∈ Ri. Then there exists some l ∈ I with i ≺ l ≤ k < j and Pi ∩ Pl = ∅ 6∈ Pl for some and hence a = 0Rl < 1Rl. Now either a = 0Pl or a = (b, l) = 0Rl b. Since a ∈ Sj, it can be only of a form 0Pj or 0Pm for j ≺ m. But both cases are not possible (in the first case we would obtain 0Pj = 1Pl = 0Pl, in the second Pl ∩ Pm 6= ∅). Assume now that a 6∈ Rj. Then there exists some m ∈ I such that j ≺ m and a = 0Rm. Since a ∈ Si, it can be only of a form 1Pi or 0Rn for i ≺ n. Since i ≺ n ≤ k < j ≺ m, this is again not possible. Hence Si ∩ Sj = ∅. Step 2: Suppose a ∈ Qi ∩ Qj for some a. Then a 6∈ Si or a 6∈ Sj. Assume now a 6∈ Si. Then there exists some l ∈ I with l ≺ i < k < j and Pl ∩ Pi = ∅ and hence 6∈ Pl (otherwise we would have a ∈ Pj which is not possible or a = 0Pm a = 1Sl for some m ∈ I with j ≺ m which is also not possible). We conclude that either a = (b, l) for some element b or a = 0Ri ∈ Si, a contradiction in the last case. Hence a = (b, l) = 1Sl. Since a ∈ Qj, we have a 6∈ Pj, i.e., a = (c, j) for some element c (which is not possible) or a = 1Sn for some n ≺ j with l ≺ i < k ≤ n < j (which is not possible) or a = 0Pm for some m ∈ I with j ≺ m (which is also not possible). Suppose a 6∈ Sj. Then there exists some n ∈ I with i < k ≤ n ≺ j and a = 1Sn 6∈ Pn (otherwise we would have a ∈ Pi which is not possible since then a = 0Pn or a = 1Pr for some r ∈ I with r ≺ i which is not possible since Pr ∩ Pn = ∅ or a = 0Pq for some q ∈ I with i ≺ q ≤ k ≤ n in which case 0Pq = 01q , a contradiction). Hence a = 1Sn = (c, n) for some element c. Since a ∈ Qi, we have that either there exists some r ∈ I with r ≺ i such that a = 1Sr, i.e., either a = (d, r) or a = (e, i) for some elements d, r, a contradiction to r < i < n, or there exists some q ∈ I with i ≺ q ≤ k ≤ n such that a = 0Sq in which case q = k = n and a = 1Sn = 0Sn, a contradiction. Hence Qi ∩ Qj = ∅. (y3): Let i, j ∈ I with i < j. If i 6≺ j we know from (y2) that Qi ∩ Qj = 0. Hence we may assume i ≺ j. Let a ∈ Qi ∩ Qj. It is enough to show that a = 1Qi = 0Qj (which will give us also (y8)). Assume first Pi ∩ Pj 6= ∅. Then 1Pi = 0Pj , Si = Ri and Qj = Sj. Hence either Qj = Rj or there exists some k ∈ I such that j ≺ k, Pj ∩ Pk = ∅ and Pj does not have a greatest element. Similarly, either Qi = Ri or there exists some h ∈ I such that h ≺ i, Ph ∩ Pi = ∅ and Pi does not have a smallest element. We distinguish the following four cases: Qj = Rj and Qi = Ri: Since a ∈ Qi ∩ Qj we have a ∈ Pi ∩ Pj, i.e., a = 1Pi = 0Pj . We have a = 1Qi = 0Qj . Qj = Rj and there exists some h ∈ I such that h ≺ i ≺ j, Ph ∩ Pi = ∅ and Pi does not have a smallest element: Then either a ∈ Si = Ri (which yields that a = 1Pi = 0Pj ) or a = 1Sh. If a = 1Sh we have either a = 1Ph ∈ Ph ∩ Pj = ∅ or 15 a = 0Si = 0Ri ∈ Rj, i.e., a ∈ Pi ∩ Pj, a contradiction. Qi = Ri and there exists some k ∈ I such that j ≺ k, Pj ∩ Pk = ∅ and Pj does not have a greatest element: Since a ∈ Qj = Sj we have either a ∈ Pj ∩ Pi, i.e., If a = 0Rk then either a = 0Pk ∈ Pk ∩ Pi = ∅ or a = 1Pi = 0Pj or a = 0Rk. a = 0Rk ∈ Rk \ Pk, a = (b, k) 6∈ Ri = Qi, a contradiction. There exist h, k ∈ I such that h ≺ i ≺ j ≺ k, Ph ∩ Pi = ∅ = Pj ∩ Pk, Pj does not have a greatest element and Pi does not have a smallest element: We have a ∈ Qi ∩ Sj. Hence (a ∈ Rj or a = 0Rk) and (a ∈ Ri or a = 1Rh). We have four cases. Three of them can be settled as above. So assume that (a = 0Rk) and (a = 1Rh). If a = 0Pk or a = 1Ph, we have a ∈ Pk ∩ Ph = ∅, a contradiction. Hence a = 0Rk ∈ Rk \ Pk and a = 1Rh ∈ Rh \ Ph, a contradiction to h < k. Assume now Pi ∩ Pj = ∅. Then Ri ∩ Rj = ∅. Clearly Si ∩ Sj ⊆ {0Rj }. Assume first Si ∩ Sj = ∅. Then Qi ∩ Qj ⊆ {1Si} = {1Ri}. If Qi ∩ Qj = {1Si} = {1Ri} then 1Qi = 1Si = 0Qj . Suppose now Si ∩ Sj = {0Rj }. Then 0Rj = 0Sj = 1Si. If Pj does have a smallest element then Sj = Qj and Qi = Si or Qi = (Si \ {0Si}) ∪ {1Sh} for some h ∈ I with h ≺ i ≺ j. In the first case Qi ∩ Qj = Si ∩ Sj = {0Rj } = {0Pj } = {0Qj }. But 1Si = 1Qi. Hence Qi ∩ Qj = {1Qi} as well. In the second case Qi ∩ Qj = ((Si \ {0Si) ∪ {1Sh}) ∩ Sj ⊆ (Si ∪ Sh) ∩ Sj ⊆ (Si ∪ Rh) ∩ Sj ⊆ {0Rj }. Moreover, 1Si = 1Qi ∈ Qi and 1Si = 0Qj ∈ Qj. Hence Qi ∩ Qj = {0Qj } = {1Qi}. Assume Pj does not have a smallest element. Then Qj = (Sj \ {0Sj }) ∪ {1Si} and Qi = Si or Qi = (Si \ {0Si) ∪ {1Sh} for some h ∈ I with h ≺ i ≺ j. In the first case Qi ∩ Qj = Si ∩ ((Sj \ {0Sj }) ∪ {1Si}) = {0Rj } = {1Si}. Moreover, 0Qj = 1Si = 1Qi. In the second case Qi ∩ Qj = ((Si \ {0Si) ∪ {1Sh}) ∩ ((Sj \ {0Sj }) ∪ {1Si}) = {1Si}. As above, 1Si = 1Qi ∈ Qi and 1Si = 0Qj ∈ Qj. Summarizing, we obtain that always Qi ∩ Qj = ∅ or Qi ∩ Qj = {a} in which case a = 1Qi = 0Qj . (y4): Assume i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pi has a greatest element 1Pi and Pj has a smallest element 0Pj . Then Ri ∩ Rj = ∅, Si = Ri and Qj = Sj. Assume a ∈ Qi ∩ Qj. Then a ∈ Qi ∩ Sj. We have either a = 1Sh for some h ∈ I with h ≺ i or a ∈ Si = Ri, and either a ∈ Rj or a = 0Rk for some k ∈ I with j ≺ k. We can assume a ∈ Si = Ri or a ∈ Rh ∪ Ri for some h ≺ i < j and a ∈ Sj ⊆ Rj ∪ Rk. Moreover, Rh ∩ (Rj ∪ Rk) = ∅ and Ri ∩ (Rj ∪ Rk) = ∅, a contradiction. We conclude Qi ∩ Qj = ∅. (y5): Suppose i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pi does not have a greatest element and Pj has a smallest element 0Pj . We have Ri ∩Rj = ∅, Qj = Sj and Si = {0Rj }∪(Ri \{1Ri}). Clearly, 0Rj ∈ Qi and 0Rj ∈ Sj = Qj. Hence 0Pj = 0Rj ∈ Qi ∩ Qj is the greatest element of Qi. (y6): Assume i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pj does not have a smallest element and Pi has a greatest element 1Pi. We obtain Si = Ri, Qj = (Sj \ {0Sj }) ∪ {1Si} and Ri ∩ Rj = ∅. We have 1Pi = 1Ri = 1Si ∈ Qj and 1Si ∈ Qi. Hence 1Pi is the smallest element of Qj. (y7): Assume i, j ∈ I, i ≺ j, Pi ∩ Pj = ∅, Pj does not have a smallest element and Pi does not have a greatest element. Then Si = {0Rj } ∪ (Ri \ {1Ri}) and Qj = (Sj \ {0Sj }) ∪ {1Si}. Since 1Si = 0Rj = 0Qj and 1Si = 1Qi, we obtain that the greatest element 1Qi of Qi is the smallest element 0Qj of Qj. 16 Finally, we can state our results on Dedekind-MacNeille completion of posets which are the generalized ordered sum of their parts (Pi, ≤i), i ∈ I. Theorem 4.7. Let P = (P, ≤) be the generalized ordinal sum of Pi = (Pi, ≤i), i ∈ I, and Qi = (Qi, ≤i), i ∈ I, be a DM-yoked family of P. Then the generalized ordinal sum Q = (Q, ≤) of Qi = (Qi, ⊆), i ∈ I, exists. If any non-empty subset of I has a maximal element then DM(P) ∼= Q. Proof. First let us check that the assumptions of Definition 4.2 are satisfied. Clearly, Qi, i ∈ I, is a family of posets such that Q⊤ has a greatest element 1 (we may identify 1 with 1P⊤) and condition (i) is satisfied by (y1). Condition (ii) follows from (y2) and condition (iii) from (y3). From (y8) we obtain condition (iv). Hence the generalized ordinal sum Q = (Q, ≤) of Qi = (Qi, ⊆), i ∈ I, exists. Assume now that any non-empty subset of I has a maximal element and I has a smallest element ⊥. Let us show Q ∼= DM(P). Note that we have, for every i ∈ I, order isomorphisms ϕi : Qi → DM(Pi) and ψi : DM(Pi) → Qi defined by ϕi(a) := LPiUPi({x ∈ (B ∩ Pi) for all a ∈ Qi and B ∈ DM(Pi), and ϕi ◦ ψi = Pi x ≤ a}) and ψi(B) := WQi idDM(Pi), ψi ◦ ϕi = idQi. We define mappings ϕ : Q → DM(P ) and ψ : DM(P ) → Q as follows: ϕ(a) := LU({x ∈ P x ≤ a}) and ψ(B) := (cid:26) ψ⊥(∅) WQj if B = ∅, (B ∩ Pj) otherwise where j := max B∩Pm6=∅ m (a ∈ Q, B ∈ DM(P )). Clearly, ϕ and ψ are well-defined and order-preserving. Recall also that ϕ(a) = ϕk(a) ∪ Sm<k ((ϕk(a) ∪ [m<k ψ(ϕ(a)) = ψ(ϕk(a) ∪ [m<k Pm) = _Qj Pm where k := max a∈Qm m. We have Pm) ∩ Pk) = ψk(ϕk(a)) = a for all a ∈ Q, here k := max a∈Qm m. Let B ∈ DM(P ). If B = ∅ and B = LU(B) then ϕ(ψ(∅)) = ϕ(ψ⊥(∅)) = ϕ⊥(0DM(P⊥)). Assume x ∈ ϕ⊥(0DM(P⊥)). Then x is the smallest element of P⊥, i.e., x ≤ p for all p ∈ P , i.e., x ∈ B, a contradiction. Hence ϕ(ψ(∅)) = ∅. Suppose now that B 6= ∅ and put j := max m. Then B∩Pm6=∅ ϕ(ψ(B)) = ϕ(_Qj (B ∩ Pj)) = ϕj(_Qj (B ∩ Pj)) ∪ [m<j Pm = (B ∩ Pj) ∪ [m<j Pm = B. Now we show that the construction of a generalized ordinal sum preserves the property of sectional pseudocomplementation. Theorem 4.8. Let P = (P, ≤) be a generalized ordinal sum of Pi = (Pi, ≤i), i ∈ I, assume that (Pi, ≤, ∗i) are sectionally pseudocomplemented for all i ∈ I, that j ∈ I, Lj(Pj) = ∅ implies Us(Ps) = ∅ where s := max m and that any non-empty subset of I has m<j a maximal element. Then P is sectionally pseudocomplemented. 17 Proof. Let i, j ∈ I, a ∈ Pi and b ∈ Pj such that i and j are maximal with this property. We put a ∗ b :=   1 a ∗i b b if a ≤ b, if a 6≤ b and i = j, if a 6≤ b and i > j. We prove that a ∗ b is the sectional pseudocomplement of a and b in P. Case 1. a ≤ b. We have L(U(a, b), 1) = L(b, 1) = L(b). Case 2. a 6≤ b and i = j. We have L(U(a, b), a ∗i b) = L(U(a, b) ∩ Pi, a ∗i b) = (L(U(a, b) ∩ Pi, a ∗i b) ∩ Pi) ∪ [m<i Pm = = Li(Ui(a, b), a ∗i b) ∪ [m<i Pm = Li(b) ∪ [m<i Pm = L(b). Now assume L(U(a, b), c) = L(b), c ∈ Pk, k ∈ I. Then b ≤ c which implies k ≥ i. Now k > i would imply a ∈ LU(a, b) = L(U(a, b), c) = L(b), a contradiction. Hence k = i and Li(Ui(a, b), c) = L(U(a, b), c) ∩ Pi = L(b) ∩ Pi = Li(b) which implies c ≤ a ∗i b. Case 3. a 6≤ b and i > j. We have L(U(a, b), b) = L(a, b) = L(b). Now assume L(U(a, b), c) = L(b), c ∈ Pk, k ∈ I. Then b ≤ c which implies k ≥ j and L(b) = L(a, c). Case 3a. i > k. We have c ≤ b which yields b = c. Case 3b. i < k. We have L(b) = L(a, c) = L(a), i.e., a ≤ b, a contradiction. Case 3c. i = k. We obtain k > j. Since L(b) = L(a, c), b = 1j is the greatest element of Pj. If Pi has a smallest element 0i then necessarily b = 0i ∈ Pi, a contradiction to the assumption that j is the maximal index from I with b ∈ Pj. Hence Pi does not have a smallest element. Put r := max m. Assume first that j < r < k = i. Then there exist by Definition 4.2 (i) m<k elements x, y ∈ Pr with b ≤ x < y ≤ a, c. We conclude y ∈ L(a, c), y 6≤ b, a contradiction. Suppose now that j = r < k = i. Since Li(Pi) = ∅ and b is the greatest element of Pj we obtain {b} = Ur(Pr) = ∅, a contradiction. Hence the only possible case is 3a which yields b = c and finally that P is sectionally pseudocomplemented. Altogether, we can summarize our results as follows. Corollary 4.9. Let P = (P, ≤) be the generalized ordinal sum of Pi = (Pi, ≤i), i ∈ I, such that Pj ∩ (DM(Pi) × {i}) = ∅ for all i, j ∈ I. Let (DM(Pi), ≤i, ∗i) be sectionally pseudocomplemented for all i ∈ I and assume that any non-empty subset of I has a maximal element and that j ∈ I and Lj(Pj) = ∅ imply Us(Ps) = ∅ where s := max m. m<j Then DM(P) is sectionally pseudocomplemented. 18 Proof. From Lemma 4.6 we obtain that there exists a DM-yoked family Qi = (Qi, ≤i ), i ∈ I, of P such that Qi is isomorphic to a sectionally pseudocomplemented poset DM(Pi) for every i ∈ I. From Theorem 4.7 we know that DM(P) is isomorphic to the generalized ordinal sum Q of the DM-yoked family Qi = (Qi, ≤i), i ∈ I, of P. Since every Qi is sectionally pseudocomplemented we have from Theorem 4.8 that Q and hence also DM(P) are sectionally pseudocomplemented. The situation described in Theorem 4.8 and Corollary 4.9 can be illustrated by the following example. Example 4.10. Consider the sectionally pseudocomplemented poset P from Example 2.7. It is evident that P is the generalized ordinal sum of the sectionally pseu- docomplemented posets P1 = (P1, ≤) = ({0, a, b, c}, ≤) and P2 = (P2, ≤) = ({d, e, 1}, ≤). Of course, P1 ∩ P2 = ∅. Hence the conditions of Definition 4.2 are satisfied. Then DM(P1) is the lattice N5 and DM(P2) the four-element Boolean algebra, i.e. both are sectionally pseudocomplemented lattices. The generalized ordinal sum of DM(P1) and DM(P2) is visualized in Fig. 7. According to Theorem 4.8 it is again sectionally pseu- docomplemented. The Dedekind-MacNeille completion of the poset visualized in Fig. 3 is visualized in Fig. 7 d c a 1 ✉ ❅ ❅ ❅ ❅ ✉ ✉ ❅ ❅ ✟✟✟✟✟✟✟✟ ❅ f ✉ ❅ ❅ ✉ ❅ ❅ ❅ ✉ ✉ ❅ ❅ ❅ ❅ ✉ 0 e b Here L(x) is abbreviated by x for x ∈ {0, a, b, c, d, e, 1} and f is an abbreviation of L(d, e). The operation table of ∗ in DM(P) looks as follows: Fig. 7 ∗ 0 a b c d e f 1 0 1 1 1 1 1 1 1 1 a b 1 b 1 1 1 1 1 b c a 1 c 1 1 1 1 b a b 1 1 1 1 1 c d 0 a b c 1 e e 1 e 0 a b c d 1 d 1 c 1 1 1 1 f 0 a b 1 0 a b c d e f 1 According to Corollary 4.9, DM(P) is just the generalized ordinal sum of DM(P1) and DM(P2). 19 References [1] R. Balbes, On free pseudo-complemented and relatively pseudo-complemented semi- lattices. Fund. Math. 78 (1973), 119 -- 131. [2] G. Birkhoff, Lattice Theory, Amer. Math. Soc., Providence, R. I., 1979. ISBN 0-8218- 1025-1. [3] I. Chajda, An extension of relative pseudocomplementation to non-distributive lat- tices. Acta Sci. Math. (Szeged) 69 (2003), 491 -- 496. [4] I. Chajda, G. Eigenthaler and H. Langer, Congruence Classes in Universal Algebra, Heldermann, Lemgo 2012. ISBN 3-88538-226-1. [5] I. Chajda and H. Langer, Convex congruences. Soft Computing 21 (2017), 5641 -- 5645. [6] I. Chajda and H. Langer, Relatively residuated lattices and posets. Math. Slovaca (submitted). http://arxiv.org/abs/1901.06664. [7] R. P. Dilworth, Non-commutative residuated lattices, Trans. Amer. Math. Soc. 46 (1939), 426 -- 444. [8] Y. S. Pawar, Implicative posets. Bull. Calcutta Math. Soc. 85 (1993), 381 -- 384. [9] J. Schmidt, Zur Kennzeichnung der Dedekind-MacNeilleschen Hulle einer geordneten Hulle. Arch. Math. (Basel) 7 (1956), 241 -- 249. Authors' addresses: Ivan Chajda Palack´y University Olomouc Faculty of Science Department of Algebra and Geometry 17. listopadu 12 771 46 Olomouc Czech Republic [email protected] Helmut Langer TU Wien Faculty of Mathematics and Geoinformation Institute of Discrete Mathematics and Geometry Wiedner Hauptstrasse 8-10 1040 Vienna Austria, and Palack´y University Olomouc Faculty of Science Department of Algebra and Geometry 17. listopadu 12 771 46 Olomouc Czech Republic [email protected] 20 Jan Paseka Masaryk University Brno Faculty of Science Department of Mathematics and Statistics Kotl´arsk´a 2 611 37 Brno Czech Republic [email protected] 21
1210.3389	1	1210	2012-10-11T22:57:55	Finiteness conditions on the Yoneda algebra of a monomial algebra	[ "math.RA" ]	Let A be a connected graded noncommutative monomial algebra. We associate to A a finite graph \Gamma(A) called the CPS graph of A. Finiteness properties of the Yoneda algebra Ext_A(k,k) including Noetherianity, finite GK dimension, and finite generation are characterized in terms of \Gamma(A). We show these properties, notably finite generation, can be checked by means of a terminating algorithm.	math.RA	math	FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA Andrew Conner Ellen Kirkman James Kuzmanovich W. Frank Moore Department of Mathematics Wake Forest University Winston-Salem, NC 27109 Abstract. Let A be a connected graded noncommutative monomial algebra. We associate to A a finite graph Γ(A) called the CPS graph of A. Finiteness properties of the Yoneda algebra ExtA(k, k) including Noetherianity, finite GK dimension, and finite generation are characterized in terms of Γ(A). We show these properties, notably finite generation, can be checked by means of a terminating algorithm. 1. Introduction Complete intersections are a well-studied class of commutative algebras, yet there is not an agreed upon notion of complete intersection in the case of noncommutative algebras. From the point of view of noncommutative algebraic geometry, such a generalization should be homological. A starting point for a homological definition of complete intersection is found in the results of Gulliksen [12, 13], F´elix-Thomas [8] and F´elix-Halperin-Thomas [9], which state for a graded Noetherian commutative k-algebra, the following properties are equivalent: (i) A is a graded complete intersection (ii) ExtA(k, k) is a Noetherian k-algebra (iii) ExtA(k, k) has finite Gelfand-Kirillov (GK) dimension. However, conditions (ii) and (iii) are not equivalent for graded Noetherian k- algebras, in fact, not even for algebras with monomial relations. Since such alge- bras are a tractable class of algebras with a well-understood projective resolution of the trivial module (see, for example [2],[4]), their Yoneda algebras are computable, though often complex. This paper concerns the study of conditions (ii) and (iii) as well as the finite generation of ExtA(k, k) when A is a connected graded noncom- mutative k-algebra with finitely many monomial relations. 2010 Mathematics Subject Classification. Primary: Key words and phrases. Yoneda algebra, monomial algebra. 2 CONNER, KIRKMAN, KUZMANOVICH, MOORE To a monomial algebra A, we associate a finite directed graph Γ(A) which we call the CPS graph of A. See Construction 2.1 for the definition of Γ(A). Our first result concerns the Gelfand-Kirillov dimension of the Yoneda algebra E(A) = ExtA(k, k). Theorem 1.1 (Corollary 2.8). Let A be a monomial k-algebra. If no pair of distinct circuits in Γ(A) have a common vertex, then GKdim(E(A)) is the maximal number of distinct circuits contained in any walk. Otherwise, GKdim(E(A)) = ∞. Given any connected graded k-algebra B one can use a noncommutative Grobner basis to associate to B a monomial algebra B′ with the property GKdim(E(B)) ≤ GKdim(E(B′)). Thus, Theorem 1.1 can provide an easily calculated upper bound on GKdim(E(B)), though this bound is not always finite when GKdimE(B) is. See Remark 2.9 below. The Yoneda product on E(A) can also be described combinatorially in terms of walks in the graph Γ(A). Up to a notion of equivalence described in Section 2, all nonzero Yoneda products are compositions of admissible walks in Γ(A). Using this description of the Yoneda product, we are able to characterize finite generation and the Noetherian property in E(A). See Sections 2 and 3 for definitions of terminology and notation. Theorem 1.2. Let A be a monomial k-algebra. (1) (Theorem 3.6) E(A) is finitely generated if and only if for every infinite anchored walk p in Γ(A), ep contains a dense edge or two admissible edges of opposite parity. (2) (Theorem 5.2) E(A) is left (resp. right) Noetherian if and only if every vertex of Γ(A) lying on an oriented circuit has out-degree (resp. in-degree) one and every edge of every oriented circuit is admissible. The second statement extends a theorem of Green et. al. [10] who characterized Noetherianity of E(A) in terms of the Ufnarovski relation graph of A in the case where A is quadratic. Theorem 1.2(1) describes an infinite set of criteria to be satisfied for E(A) to be finitely generated. Whether finite generation of E(A) can be determined by finitely many criteria is a problem of recent interest. Working in the more general context of monomial factor algebras of quiver path algebras, Green and Zacharia [11] describe a (potentially infinite) process by which finite generation of the Yoneda algebra can be checked. Further progress was made by Davis [6] and Cone [5] who showed finite generation can be determined by finitely many criteria when the given quiver is a cycle or an "in-spoked" cycle. In Section 4 we show the same can be said in our situation; that is, when the quiver consists of a single vertex and finitely many loops. Theorem 1.3 (Theorem 4.3). Let A be a monomial k-algebra with gl.dim A = ∞. Let N be the smallest even integer greater than or equal to 2E 2 + E + 1 where E is the number of edges in Γ(A). The Yoneda algebra E(A) is finitely generated if and only if every anchored walk of length N or N + 1 is decomposable. In our experience, determining if E(A) is finitely generated when E(A) has infinite Gelfand-Kirillov dimension can be a difficult problem, and we were unable to obtain an efficient bound in Theorem 1.3. However, the case GKdim(E(A)) < ∞ is much simpler. We describe a recursive algorithm for determining finite generation in that case in Section 4. FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 3 We sincerely thank Ed Green for the helpful conversations and illuminating examples he provided in the course of this project. 2. The CPS graph In [15], C. Phan associated a weighted digraph to any monomial graded algebra A. One important feature of Phan's graph is that a k-basis for E(A) is repre- sented by certain directed paths. After establishing some notation, we recall the unweighted version of Phan's graph - which we call the CPS graph of A - and we record a description of a minimal graded projective resolution of Ak (due to Cas- sidy and Shelton) in terms of this graph. We also prove several combinatorial facts about the CPS graph needed later. Let k be a field. Throughout this paper we use the phrase graded k-algebra or just k-algebra to mean a connected, N-graded, locally finite-dimensional k-algebra which is finitely generated in degree 1. If A is a graded k-algebra, we use the term (left or right) ideal to mean a graded (left or right) ideal of A generated by homogeneous elements of degree at least 2, unless otherwise indicated. The to denote the trivial graded A-module A/A+. The bigraded Yoneda algeba of A A (k, k). (Here i denotes the cohomology degree and j denotes the internal degree inherited from the grading on augmentation ideal is A+ = Li≥1 Ai. We abuse notation and use k (or Ak or kA) is the k-algebra E(A) = Li,j≥0 Ei,j(A) = Li,j≥0 Exti,j A.) Let Ep(A) = Lq Ep,q(A). Let s ∈ N and let V = spank{x1, . . . , xs}. We denote the tensor algebra on V by T (V ). The tensor algebra is a graded k-algebra, graded by tensor degree. We denote the tensor degree of a homogeneous element w ∈ T (V ) by deg w. By a monomial in T (V ) we mean a pure tensor with coefficient 1. We consider 1T (V ) a monomial. By a monomial algebra, we mean an algebra of the form A = T (V )/I where I is an ideal of T (V ) generated by finitely many monomials. Such an algebra A is a graded k-algebra with the grading inherited from the tensor grading on T (V ). Let M be the set of monomials in T (V ). Multiplication in T (V ) induces the structure of a monoid on M . Let I = hw1, . . . wri be an ideal in T (V ). We assume the wi form a minimal set of monomial generators for I and we let di = deg wi be the tensor degree of wi for each i. Recall that we assume every di ≥ 2. Let A = T (V )/I and let π : T (V ) → A be the natural surjection. Construction 2.1 (CPS graph). Suppose m, w ∈ M − I and w ⊗ m ∈ I. Let L(w, m) = w′ where w = w′′ ⊗ w′ for w′, w′′ ∈ M and w′ is minimal such that w′ ⊗ m ∈ I. For m ∈ M − I define Am = {w ∈ M − I : w ⊗ m ∈ I and L(w, m) = w} Then the images of elements of Am in A generate the left annihilator of π(m). Let G0 = {x1, . . . xs} and for i ≥ 1 let Gi = Sw∈Gi−1 Si≥0 Aw. Finally, let G = Gi. Define the CPS graph of A to be the directed graph Γ(A) with vertex set G, and edges m1 → m2 whenever m2 ∈ Am1. We note the graph Γ(A) is finite. The graph may have loops and parallel edges with opposite orientation, but it has no parallel edges with the same orientation. 4 CONNER, KIRKMAN, KUZMANOVICH, MOORE Example 2.2. Let A = kha, b, c, di/habc, cdabi and B = A/hbcdai The graphs Γ(A) and Γ(B) are shown below. Γ(A) / ab ) cd / cda c b ) cd Γ(B) / ab * cda * bcd c b a Remark 2.3. An obvious, but extremely important feature of the CPS graph is that there is a directed edge m1 → m2 with m1 ∈ G0 if and only if m2 ⊗ m1 is a minimal generator of I. As illustrated by Proposition 2.5 below, this correspondence parallels the standard identification of Ext2 A(k, k) with the graded dual of the space I/(V ⊗ I + I ⊗ V ). If the defining relations of a monomial algebra A are quadratic, A is Koszul [16]. In that case, Γ(A) is Ufnarovski's "relation graph" [18] for the Koszul dual algebra A!. We also note that because we consider only minimal left annihilators, the CPS graph Γ(A) is quite different from the notion of "zero-divisor graph" studied recently in [1]. We adopt some standard graph-theoretic terminology. By a walk we mean a finite or infinite sequence v0v1v2 · · · of vertices where vi → vi+1 is a directed edge for all 0 ≤ i < n. If v0v1 · · · vn is a finite walk, we say the walk has length n. A walk is called a path if it contains no repeated vertices. We will not need to distinguish walks which repeat vertices but not edges. By a closed walk of length n we mean a walk of length n such that vn = v0. A circuit of length n is a closed walk of length n such that v0, . . . , vn−1 are distinct. In the context of a weighted digraph, we abuse this terminology slightly and use "walk," "path," and "circuit" to refer to sequences of vertices in the underlying unweighted graph. If p and q are walks of length n and m respectively, we say p extends q or q is a prefix of p and write q ⊢ p if n ≥ m and pi = qi for all 0 ≤ i ≤ m. In [4, §5], Cassidy and Shelton give a combinatorial description of a minimal graded projective left A-module resolution P• of Ak in terms of monomial matrices. We briefly recount their resolution here, indexing the bases of each graded projective module by certain walks in Γ(A). each w ∈ Wn, let dw = Pn Let Wn denote the set of all walks w of length n in Γ(A) such that w0 ∈ G0. For i=0 deg wi where deg wi denotes the tensor degree of the monomial wi. Let A(−dw) be the graded free left A-module of rank 1 with grading shift A(−dw)p = Ap−dw . Choose a basis for A(−dw) and denote this element by ew. Let P0 = A be the graded free module with fixed basis element e∅ and for j > 0, let Pj = M w∈Wj−1 A(−dw) Define dj : Pj → Pj−1 on the A-basis {ew : w ∈ Wj−1} by setting dj (ew) = π(wj−1)e ¯w where ¯w = w0 · · · wj−2 if j ≥ 2 and ¯w = ∅ if j = 1. Extend dj A-linearly / ) i i / O O / ) i i * h h * i i FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 5 to all of Pj. Since wj−1 → wj is an edge in Γ(A) only if wj ∈ Awj−1 , it is clear that djdj+1 = 0 for j ≥ 0. The following lemma is a straightforward consequence of the definition of Γ(A). Lemma 2.4. The complex (P•, d•) described above is a minimal graded projective resolution of Ak. Moreover, the bases for the Pj can be ordered so the matrices of the dj with respect to the ordered bases are precisely the monomial matrices described in [4]. The next fact follows immediately from Lemma 2.4. Proposition 2.5. Let A be a monomial k-algebra and i ∈ N. Then the graded duals {εw} of the basis elements {ew} where w is a walk of length i in Γ(A) with w0 ∈ G0 form a k-basis for Exti+1 A (k, k). We make extensive use of this basis throughout the paper. For ease of exposition, we make the following defintion. Definition 2.6. A walk w in Γ(A) is called anchored if w0 ∈ G0. Remark 2.7. Anchored walks of length i in Γ(A) correspond to the sets Γi de- scribed in [11]. Several properties of A and E(A) are immediate from Proposition 2.5. We denote the Gelfand-Kirillov dimension of a k-algebra A by GKdim(A). Corollary 2.8. (1) If Γ(A) contains no circuit, then gl.dim(A) is equal to the length of the longest path in Γ(A). Otherwise, gl.dim(A) = ∞. (2) GKdim(E(A)) = ∞ if and only if Γ(A) contains distinct circuits with a common vertex. (3) If no pair of distinct circuits in Γ(A) have a vertex in common, then GKdim(E(A)) is the maximal number of circuits contained in any walk (ignoring multiplicity). (4) The Hilbert series of E(A) is a rational function. Proof. (1) is clear. (2), (3), and (4) are standard (see [18]). (cid:3) Remark 2.9. To any connected graded k-algebra B = T (V )/J (we do not assume J is generated by monomials) one can associate a monomial algebra in the usual way: Choose an ordered basis of V and induce a total ordering the monoid M via degree-lexicographic order. Let F be a noncommutative Grobner basis of J with respect to this ordering. Let ht(F ) be the set of high terms of elements of F and let B′ = T (V )/hht(F )i. Let PB(y, z) = X p,q dim Extp,q B (k, k)ypzq denote the Poincare series of B. From the well-known coefficientwise inequality PB(y, z) ≤ PB ′(y, z) (see Lemma 3.4 of [2]) we can deduce GKdim(E(B)) ≤ GKdim(E(B′)). Equality holds in the important case where the Grobner basis for J consists of homogeneous polynomials of the same degree (see Corollary 4.6 of [14]). Thus Corollary 2.8 can sometimes provide an easily calculated upper bound on GKdim(E(B)). For further examples, see Section 6. 6 CONNER, KIRKMAN, KUZMANOVICH, MOORE It is also interesting to note that E(A) has either exponential or polynomial growth - this is the case for commutative k-algebras (see [3, 12, 13]). Observe GKdim(E(A)) = GKdim(E(B)) = 1 for algebras A and B from Example 2.2. Given a minimal projective resolution of Ak, one can compute the Yoneda prod- uct of classes ε1 and ε2 in E(A) by lifting a representative of ε2 through the reso- lution to the appropriate cohomology degree and composing with a representative of ε1. For a monomial algebra A, we wish to describe the Yoneda product combi- natorially in terms of walks in the graph Γ(A). To do this, we introduce a notion of walk equivalence as a combinatorial analog of lifting a representative through a projective resolution. We call two walks p = p0 · · · pn and q = q0 · · · qm in a CPS graph Γ(A) equivalent if m = n and pn ⊗ pn−1 ⊗ · · · ⊗ p0 = qm ⊗ qm−1 ⊗ · · · ⊗ q0 as elements of M . If p and q are equivalent, we write p ∼ q. It is clear that ∼ is an equivalence relation on walks in Γ(A). Lemma 2.10. Let Γ(A) be a CPS graph, and let p and q be equivalent walks of length n > 0 in Γ(A). Then (1) the prefix walks p0 · · · p2k+1 and q0 · · · q2k+1 are equivalent for all 0 ≤ k ≤ ⌊ n 2 ⌋. (2) if n is even, then pn = qn. (3) we have p2k+1 ⊗ p2k = q2k+1 ⊗ q2k for all 0 ≤ k ≤ ⌊ n (4) if deg(p0) ≥ deg(q0), then 2 ⌋. deg(pi) ≥ deg(qi) if 0 < i ≤ n is even deg(qi) ≥ deg(pi) if 0 < i ≤ n is odd (5) the walk q is unique if it is anchored. Proof. To prove (1), we induct on k. Let k = 0. By switching the variables p and q if necessary, there is no loss of generality in assuming deg(p0) ≥ deg(q0). Since pn ⊗ · · · ⊗ p0 = qn ⊗ · · · ⊗ q0 there exists a unique monomial m ∈ M − I such that m ⊗ q0 = p0. We have q1 ⊗ q0 ∈ I, m ⊗ q0 /∈ I, and pn ⊗ · · · ⊗ p1 ⊗ m ⊗ q0 = qn ⊗ · · · ⊗ q1 ⊗ q0 so there is a unique monomial m′ ∈ M − I such that q1 = m′ ⊗ m. Now, m′ ⊗ p0 = m′ ⊗ m ⊗ q0 = q1 ⊗ q0 ∈ I and L(q1, q0) = q1, so L(m′, p0) = m′ and m′ ∈ Ap0. Thus pn ⊗ · · · ⊗ p1 ⊗ p0 = qn ⊗ · · · ⊗ q1 ⊗ q0 = qn ⊗ · · · ⊗ m′ ⊗ p0 and p1, m′ ∈ Ap0, so p1 = m′. Hence p1 ⊗ p0 = m′ ⊗ p0 = q1 ⊗ q0 as desired. For the induction step, assume p2k+1 ⊗ · · · ⊗ p0 = q2k+1 ⊗ · · · ⊗ q0 and deg(p2k+2) ≥ deg(q2k+2) and proceed as in the base case. This completes the proof of (1). FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 7 Statements (2) and (3) follow immediately from (1). We consider statement (4). In light of (2) and (3), it suffices to prove deg(qi) ≥ deg(pi) for 0 < i ≤ n odd. Since q1 ⊗ q0 = p1 ⊗ p0 and deg(p0) ≥ deg(q0), it is clear that deg(q1) ≥ deg(p1). Thus the result holds for n ≤ 2. Assume n > 2 and for 0 < i < n− 1 odd deg(qi) ≥ deg(pi). Since qi ⊗ qi−1 = pi ⊗ pi−1, there exists m ∈ M such that qi = pi ⊗ m. Suppose toward contradiction that deg(pi+1) < deg(qi+1). Since qi+2 ⊗ qi+1 = pi+2 ⊗ pi+1, there exists m′ ∈ M , deg(m′) > 0 such that qi+1 = m′ ⊗ pi+1. Since pi+1 ⊗ pi ∈ I, we have pi+1 ⊗ qi = pi+1 ⊗ pi ⊗ m ∈ I. The fact that deg(m′) > 0 contradicts the assumption that L(qi+1, qi) = qi+1. So deg(qi+1) ≤ deg(pi+1) and hence deg(qi+2) ≥ deg(pi+2). Statement (4) now follows by induction. To prove (5), suppose q′ is another walk such that p ∼ q′ and q′ 0 ∈ G0. Then 0. Since G0 consists solely of degree 1 monomials, q0 = q′ 0 1 ⊗ q′ q ∼ q′ and q1 ⊗ q0 = q′ so q1 = q′ 1. Suppose inductively that qi = q′ i for all 0 ≤ i ≤ 2k + 1 < n. If n = 2k + 2, the induction hypothesis and the definition of equivalence imply q2k+2 = q′ 2k+2. If n > 2k + 2, q2k+3 ⊗ q2k+2 = q′ 2k+3 ⊗ q′ q′ if necessary, we can assume deg(q2k+2) ≥ deg(q′ some m ∈ M . But L(q2k+2, q2k+1) = q2k+2 and 2k+2. By switching the variables q and 2k+2 for 2k+2), so q2k+2 = m ⊗ q′ q′ 2k+2 ⊗ q2k+1 = q′ 2k+2 ⊗ q′ 2k+1 ∈ I by the induction hypothesis. Thus m = 1, q2k+2 = q′ Statement (5) now follows by induction. 2k+2, and hence q2k+3 = q′ 2k+3. (cid:3) Since finite anchored walks in Γ(A) enumerate a k-basis for E(A), we make the following definition. Definition 2.11. A finite walk in Γ(A) is called admissible if it is equivalent to an anchored walk. By Lemma 2.10(5), every admissible walk is equivalent to a unique anchored walk. In Example 2.2, edge ab → cd is an admissible walk of length 1 in both Γ(A) and Γ(B), but edge cd → ab is admissible in neither graph. In Phan's original weighted digraph, the edge weighting distinguished admissible edges from their counterparts. That distinction is too coarse for our purposes, but the importance of admissible edges seems evident from the following useful facts about admissible walks. Proposition 2.12. Let Γ(A) be a CPS graph, and let p be an admissible walk of length n in Γ(A). Let q be a walk of length s such that q extends p. If either n or s − n is even, then q is admissible. Proof. An admissible walk of length 0 consists of a single vertex in G0, so the statement is trivial if n = 0. The statement is also trivial if s = n. So assume n > 0, s − n > 0, and let r be a path in Γ(A) such that p ∼ r and r0 ∈ G0. If n is even, then rn = pn by Lemma 2.10(2). It follows immediately that the i = qi for n + 1 ≤ i ≤ s is i = ri for 0 ≤ i ≤ n and r′ s given by r′ path r′ = r′ equivalent to q and has r′ 0 · · · r′ 0 ∈ G0. Suppose n and s are odd. If rn = pn, we can proceed as above, so assume rn 6= pn. By Lemma 2.10(3) and (4), there exists a monomial m ∈ M , deg(m) > 0 8 CONNER, KIRKMAN, KUZMANOVICH, MOORE such that rn = pn ⊗ m = qn ⊗ m. Thus qn+1 ⊗ rn ∈ I. Put rn+1 = L(qn+1, rn) and let m′ ∈ M such that qn+1 = m′ ⊗ rn+1. Now, qn+2 ⊗ m′ ⊗ rn+1 = qn+2 ⊗ qn+1 ∈ I Since qn+1 /∈ I and L(qn+2, qn+1) = qn+2, it follows that qn+2 ⊗ m′ ∈ Arn+1. Put rn+2 = qn+2 ⊗ m′. Then by construction, r0 · · · rn+2 is a well-defined walk equivalent to p′ = q0 · · · qn+2 and r0 ∈ G0. Thus p′ is an admissible walk of length n + 2 and q is an extension of p′ of length s. The result now follows by induction on s − n. (cid:3) We show in the next section that E(A) is finitely generated if Γ(A) has "enough" admissible walks. To make this more precise, we make the following definition. Definition 2.13. Let p be an infinite walk in Γ(A) and let e = pipi+1 be an admissible edge in p. We call e dense in p if e has an admissible even-length extension in p. In Example 2.2, c → ab → cd → ab → cd · · · is the only infinite anchored walk in Γ(A). The admissible edge ab → cd is dense in this walk since ab → cd → ab ∼ b → cda → ab However, the edge ab → cd is not dense in the same walk in Γ(B). The equivalent anchored walks corresponding to odd-length extensions of ab → cd begin at vertex b and end at either b or cda. It follows that no even length extension of ab → cd is admissible because condition (2) of Lemma 2.10 cannot be satisfied. An admissible edge e may belong to many infinite walks. The edge e may be dense in some infinite walks, but not others. Furthermore, e may not be dense in an infinite walk w, but w may contain some other dense edge. See Example 6.1. The following criterion for establishing density is immediate from Lemma 2.10(2). Lemma 2.14. Let w be a (possibly infinite) walk in Γ(A) and let e = wiwi+1 be an admissible edge in w. Let q be any odd-length extension of e in w and let q′ be the unique anchored walk equivalent to q. Then every even-length extension of q in w is admissible if and only if q′ t = wi+t for some even t ≥ 0. 3. Multiplicative Structure In this section we show certain extensions of walks in Γ(A) correspond to Yoneda products in E(A) and use the result to combinatorially characterize finite generation of E(A). Recall that if w is an anchored walk of length n in Γ(A), we denote the cor- responding A-basis element of Pn+1 by ew. We denote the graded dual of ew by εw. Fix an anchored walk q of length n. To connect the Yoneda product in E(A) to extensions of walks in Γ(A), we explicitly construct lifts of εq through the resolution (P•, d•) defined in Section 2. We need one additional definition before describing the construction. Definition 3.1 ([4]). An element r in an ideal I ⊂ T (V ) is called essential if r is not in the ideal generated by V ⊗ I + I ⊗ V . FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 9 We note that a monomial r in a monomial ideal I is essential if and only if r is a minimal generator of I. Hence a walk w0w1 in Γ(A) is admissible if and only if w1 ⊗ w0 is a minimal generator of I. For i ≥ 0, we define A-module maps fi such that the following diagram com- mutes. (1) · · · · · · / Pn+3 dn+3 / Pn+2 dn+2 / Pn+1 f2 / P2 d2 f1 / P1 d1 f0 / P0 εq !❈❈❈❈❈❈❈❈ / k For i ≥ 0 let Qn+i be the graded free submodule of Pn+i spanned by the set {er : q ⊢ r} and let Zn+i be the complement to Qn+i in Pn+i. We observe that P≥n = Q≥n ⊕ Z≥n as complexes of graded free left A-modules. For all i ≥ 0, we define fi(Zn+i+1) = 0. We define fi on the specified A-basis of Qn+i+1 in several steps. (i) Define f0(eq) = e∅. (ii) For any walk r such that er ∈ Qn+2, we have rn+1 = m ⊗ xj for a unique m ∈ M and generator xj. Define f1(er) = π(m)exj . (iii) Suppose d > 2 and r is a walk such that er ∈ Qn+d. Recall the walk rn+1rn+2 is admissible if and only if rn+2 ⊗ rn+1 is essential. (See Remark 2.3.) If rn+2 ⊗ rn+1 is not essential, we define fd−1(er) = 0. If rn+2 ⊗ rn+1 is essential, our construction depends on the partiy of d. If d is odd and rn+1rn+2 is admissible, the walk rn+1 · · · rn+d−1 of length d − 2 is admissible by Proposition 2.12. Let r′ = r′ d−2 be the unique anchored walk equivalent to rn+1 · · · rn+d−1. Then er′ ∈ Pd−1 and we define fd−1(er) = er′. 0 · · · r′ 0 · · · r′ If d is even and rn+1rn+2 is admissible, then the length d − 3 walk rn+1 · · · rn+d−2 is admissible. Let r′ d−3 be the equivalent anchored walk. Lemma 2.10(3) and (4) imply that there exists a unique monomial m ∈ M such that rn+d−2 ⊗ m = r′ d−3. Since rn+d−1 ⊗ rn+d−2 ∈ I, we have d−3) and let m′ ∈ M be the rn+d−1 ⊗ r′ d−2 = L(rn+d−1, r′ unique monomial such that rn+d−1 = m′ ⊗ r′ d−2. Then r′ = r′ d−2 is a well-defined anchored walk, er′ ∈ Pd−1, and we may define fd−1(er) = π(m′)er′. d−3 ∈ I. Put r′ 0 · · · r′ To summarize: For d > 0 and r a walk such that er ∈ Qn+d, we define fd−1(er) = e∅ π(m)exj er′ π(m′)er′ 0 if d = 1 if d = 2 and rn+1 = m ⊗ xj if rn+2 ⊗ rn+1 is essential and d > 2 odd if rn+2 ⊗ rn+1 is essential and d > 2 even else   where r′ is a uniquely determined anchored walk of length d − 2 and rn+d−1 = m′ ⊗ r′ d−2. Extending the definitions of the fi A-linearly, we obtain a sequence of A-module maps. / / / ! / / / / 10 CONNER, KIRKMAN, KUZMANOVICH, MOORE Lemma 3.2. With fi defined as above, the diagram (1) commutes. Proof. Because fi(Zn+i+1) = 0 for all i ≥ 0 and Z>n is a subcomplex of P>n, it suffices to show commutativity for the complex Q>n. We compute the first few squares explicitly. (d=1) The augmentation map ǫ : P0 → k takes e∅ 7→ 1, so εq = ǫf0. (d=2) Let r be a walk of length n + 1 in Γ(A) which extends q. Then f0dn+2(er) = f0(π(rn+1)eq) = π(rn+1)e∅ On the other hand, rn+1 = m ⊗ xj for unique m ∈ M and generator xj so d1f1(er) = d1(π(m)exj ) = π(m)π(xj )e∅ = π(rn+1)e∅ (d=3) Let r be a walk of length n + 2 in Γ(A) which extends q. If rn+2 ⊗ rn+1 is essential, then d2f2(er) = d2(er′ ) = π(r′ 1)er′ where r′ = r′ other hand, 0r′ 1 is anchored, equivalent to rn+1rn+2, and r′ = r′ 0. On the f1(dn+3(er)) = f1(π(rn+2)e¯r) = π(rn+2)π(m)exj = π(rn+2 ⊗ m)exj where ¯r = r0 · · · rn+1 and rn+1 = m ⊗ xj . In this case, since r′ r′ 1 ⊗ r′ 0 = rn+2 ⊗ rn+1, we have r′ 0 = xj and r′ 1 = rn+2 ⊗ m as desired. 0 ∈ G0 and If rn+2 ⊗ rn+1 is not essential, f2(er) = 0 and π(rn+2 ⊗ m) = 0 since rn+2 ∈ Arn+1. For d > 3, the arguments are similar to those above and are omitted. The key d−4 is equivalent to rn+1 · · · rn+d−3 by Lemma 2.10(1) d−4 by uniqueness (Lemma 2.10(5)). The definitions of the fi then d−2 if d is even, from d−3 if d is odd and rn+d−1 = m′ ⊗ r′ observation is that r′ so ¯r′ = r′ imply rn+d−2 = m′ ⊗ r′ which commutativity follows. 0 · · · r′ 0 · · · r′ If α, β ∈ E(A), we denote the Yoneda composition product by α ⋆ β. If w is an admissible walk of length m in Γ(A) (not necessarily anchored), we define the symbol εw to mean the dual basis element εq where q is the unique anchored walk equivalent to w guaranteed by Lemma 2.10. The following proposition provides a combinatorial description of the Yoneda product. (cid:3) Proposition 3.3. Let p = p0 · · · ps and q = q0 · · · qn be admissible walks in Γ(A). Then εp ⋆ εq = 0 unless there exist walks p′ ∼ p and q′ ∼ q such that q′ is anchored and q′ In that case εp ⋆ εq = εw where w ∼ q′ 0 is an edge in Γ(A). n → p′ 0 · · · q′ np′ 0 · · · p′ s. Proof. By Lemma 2.10(5) it suffices to consider the case where q is anchored. For i ≥ 0, let fi be defined as above. By definition of Yoneda composition product, εp ⋆ εq = εpfs+1. Let r be any anchored walk of length n + s + 1. If r does not extend q, then εpfs+1(er) = 0. If r extends q, then εpfs+1(er) = εp(π(m)exj ) εp(er′) εp(π(m′)er′) 0 if s = 0 and rn+1 = m ⊗ xj if s > 0 is odd and rn+2 ⊗ rn+1 is essential if s > 0 is even and rn+2 ⊗ rn+1 is essential else   FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 11 where r′ and m′ are defined as in (iii) above. Thus εpfs+1(er) = 0 unless r extends q, p ∼ r′, and, if s is even, r′ s = rn+s+1. The last condition implies r′ ∼ rn+1 · · · rn+s+1 when s is even. (This equivalence always holds when s is odd.) So if εpfs+1(er) 6= 0, we have εpfs+1(er) = 1 and it follows that εp ⋆ εq = εw where w = q0 · · · qnrn+1 · · · rn+s+1. Since p ∼ r′ ∼ rn+1 · · · rn+s+1, setting p′ = rn+1 · · · rn+s+1 gives the desired result. (cid:3) We call a class α ∈ Ei(A) for i > 0 decomposable if α is in the subalgebra of E(A) generated by Lj<i Ej(A). Otherwise we call α indecomposable. Proposition 3.3 illustrates a nice feature of our chosen k-basis for E(A). Corollary 3.4. If w is an anchored walk of length n in Γ(A), then εw is de- composable if and only if there exists an admissible walk p = p0 · · · pm such that w = w0 · · · wip0 · · · pm for some 0 ≤ i < n. We note this implies the relations of E(A) consist exclusively of monomials and binomials. (That E(A) can be presented this way was also observed by C. Phan.) We are nearly ready to give a combinatorial characterization of finite genera- indecomposable) if εw is tion. We call an admissible walk w decomposable (resp. decomposable (resp. indecomposable) in E(A). The following important fact is an application of the classical Konig's Lemma (see [7] p. 1046); its proof by induction is omitted. Lemma 3.5. If Γ(A) contains infinitely many indecomposable finite anchored walks, there exists an infinite anchored walk with infinitely many indecomposable finite pre- fixes. Our main theorem characterizes infinite walks with infinitely many indecompos- In the next section, we give a finite procedure for checking these able prefixes. conditions. If p is a walk of length n in Γ(A), let ep = p1 · · · pn be the walk p with the initial edge deleted. Recall from Section 2 that if e is an admissible edge in an infinite walk p, we call e dense in p if e has an admissible even-length extension in p. Theorem 3.6. Let A be a monomial k-algebra. The following are equivalent. (1) E(A) is finitely generated. (2) Every infinite anchored walk in Γ(A) has finitely many indecomposable pre- fixes. (3) For every infinite anchored walk p in Γ(A), ep contains a dense edge or two admissible edges of opposite parity. Here "opposite parity" means the number of edges properly between the two admissible edges is even. See Section 6 for an illustration of the theorem. Proof. The equivalence of (1) and (2) follows from Lemma 3.5. We prove (2) and (3) are equivalent. Let p be an infinite anchored walk in Γ(A). If ep contains a dense edge e, then there exists an even-length extension e ⊢ q in p such that q is admissible. By Proposition 2.12, every extension of q is admissible, so by Corollary 3.4, p has only finitely many indecomposable prefixes. 12 CONNER, KIRKMAN, KUZMANOVICH, MOORE By Proposition 2.12, every odd-length extension of an admissible edge is admis- indecomposable prefixes. sible. Hence if ep has admissible edges of opposite parity, p has only finitely many Suppose instead that ep has no dense edges and all admissible edges in ep have the same parity. If ep contains no admissible edges, then Corollary 3.4 implies that every finite prefix of p is indecomposable. If ep contains an admissible edge, let e = pipi+1 0 ≤ j < n. Since a walk of the form pi+2j · · · pi+2n has even length and ep contains be the admissible edge with i minimal. Since admissible edges have the same parity, for n > 0 the admissible edges in p0 · · · pi+2n have the form pi+2jpi+2j+1 for no dense edges, p0 · · · pi+2n is indecomposable for all n > 0 by Corollary 3.4. (cid:3) Remark 3.7. If w is an infinite walk and for j > 0, wjwj+1 is a dense edge in ew, then any admissible edge wj−2iwj−2i+1, 0 ≤ i ≤ ⌊ j 2 ⌋ is also dense in w. This follows from the fact that wj−2i · · · wj−1 is an odd-length extension of wj−2iwj−2i+1, Propositions 2.12, 2.5 and 3.3. Thus ew contains a dense edge if and only if the first admissible edge in ew is dense in w. It follows from the discussion in Section 2 that for the algebras A and B from Example 2.2, E(A) is finitely generated and E(B) is not. 4. An Upper Bound for Checking Finite Generation At first glance, verification of the conditions of Theorem 3.6 appears to require an infinite procedure, in general. The infinitude arises both from the number of infinite walks in Γ(A) and the determination of edge density. In this section we establish an upper bound on the cohomological degree of an indecomposable element if E(A) is finitely generated. For the first time, the distinction between "path" and "walk" is important. Let L be the maximal length of an anchored path p in Γ(A) with pL−1pL an admissible edge, and no edge pipi+1 admissible for 0 < i < L − 1. Let M be the size of the largest edge equivalence class, and let E be the number of edges of Γ(A). First we show that if gl.dim A = ∞ and L = 1, E(A) is not finitely generated. Thus in the sequel, we will focus our attention on the case L > 1. In that case, the existence of an admissible edge m1 → m2 with m1 /∈ G0 implies M > 1. Lemma 4.1. Let A be a monomial k-algebra with gl.dim A = ∞ and L = 1. Then E(A) is finitely generated if and only if every circuit in Γ(A) contains a vertex in G0. Proof. Every anchored walk is admissible. If every circuit in Γ(A) contains a vertex in G0, then for any infinite walk w, the walk ew contains a vertex in G0. Thus ew contains a dense edge, and because w was arbitrary, E(A) is finitely generated by Theorem 3.6. Since gl.dim A = ∞, the graph Γ(A) contains a circuit. If Γ(A) contains a circuit C missing G0, let p be an anchored path of length n such that pn is in C. Since L = 1, neither ep nor C contains an admissible edge. Let q be the infinite extension of p defined by repeatedly traversing C. Then eq contains no admissible edge, hence every prefix of q is indecomposable by Corollary 3.4 and E(A) is not finitely generated by Theorem 3.6. (cid:3) FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 13 Next we establish an important upper bound. Lemma 4.2. Suppose q is a walk of length N > 2E(M − 1) + L in Γ(A). Assume if N − j is even and p = qjqj+1 · · · qN −1 if N − j is odd. Then L > 1 and eq contains admissible edge qjqj+1 for 0 < j < L. Let p = qjqj+1 · · · qN 2i+1 where p′ ∼ p (1) p is admissible (2) for all 0 ≤ i ≤ 2E(M − 1), we have q2i+j q2i+j+1 ∼ p′ 2ip′ and p′ is anchored. (3) Either (a) q2i+jq2i+j+1 = p′ (b) there exist 0 ≤ c < d ≤ 2E(M − 1) such that 2i+1 for some 0 ≤ i ≤ 2E(M − 1) or 2ip′ q2c+jq2c+j+1 = q2d+jq2d+j+1 and p′ 2cp′ 2c+1 = p′ 2dp′ 2d+1 Proof. Statement (1) is immediate from Proposition 2.12 since the length of p is odd. Statement (2) then follows from Lemma 2.10. To prove (3), let E denote the set of edges of Γ(A) and let S = {(e1, e2) ∈ E × E : e1 ∼ e2, e1 6= e2} Then S ≤ E(M − 1). Since the walks p and p′ consist of at least 2E(M − 1) edges of Γ(A), either one of the pairs of equivalent edges (q2i+j q2i+j+1, p′ 2ip′ 2i+1) 0 ≤ i ≤ E(M − 1) is not in S, in which case (a) holds, or some element of S appears twice, in which case (b) holds. (cid:3) Theorem 4.3. Let A be a monomial k-algebra with gl.dim A = ∞ and L > 1. Let N be the smallest even integer greater than or equal to 2E(M − 1) + L + 1. The Yoneda algebra E(A) is finitely generated if and only if every anchored walk q of length N or N + 1 is decomposable. Since M − 1 and L are at most E, we obtain the weaker, but more easily stated bound of 2E 2 + E + 1 mentioned in the Introduction. Proof. Suppose every anchored walk of length N or N + 1 is decomposable. Let q be any anchored walk of length N + 1. Then q and q′ = q0 · · · qN are both decom- posable. By Corollary 3.4, eq′ contains an admissible edge qiqi+1. By Proposition 2.12, every odd length extension of qiqi+1 is admissible. This can account for the decomposability of only one of q and q′. Since both are decomposable, either q contains an admissible edge whose parity is opposite qiqi+1 or an even length ex- tension of qiqi+1 is admissible, making qiqi+1 dense in any infinite walk with prefix q. Since q was arbitrary, E(A) is finitely generated by Theorem 3.6. Conversely, suppose Γ(A) contains an indecomposable anchored walk q of length N or N + 1. We will construct an infinite anchored walk w in Γ(A) in which all admissible edges have the same parity, but none are dense in w. That E(A) is not finitely generated will then follow from Theorem 3.6. By the discussion preceeding Lemma 4.1, we have M > 1, hence eq contains a circuit. If eq contains no admissible edge, or if the first admissible edge of eq follows a circuit in eq, we can construct an infinite walk in Γ(A) in which every prefix is such that qjqj+1 is an admissible edge of eq. indecomposable as in the proof of Lemma 4.1. Otherwise, let 0 < j < L be minimal 14 CONNER, KIRKMAN, KUZMANOVICH, MOORE Since q is indecomposable, Proposition 2.12 implies the length of q and the index j must have opposite parity. We consider only the case where q has length N , the other case being identical after the obvious necessary index shift. Let p = qj · · · qN −1. By Lemma 4.2(1), p is admissible, so let p′ be the unique anchored walk equivalent to p. The walk qj · · · qN is not admissible, so by Lemma 2.14 we must have q2i+j 6= p′ 2i for all 0 ≤ i ≤ 2E(M − 1). Therefore, we have q2i+j q2i+j+1 6= p′ 2i+1 for all 0 ≤ i ≤ 2E(M − 1). By Lemma 4.2(3), there exist 0 ≤ c < d ≤ 2E(M − 1) such that q2c+jq2c+j+1 = q2d+jq2d+j+1 and p′ 2ip′ 2c+1 = p′ 2cp′ 2d+1. 2dp′ Let z = q2c+j · · · q2d+j−1, let z′ = p′ 2c · · · p′ 2d−1 and let w be the infinite walk Since q2c+j = q2d+j and q2d+j−1 → q2d+j is an edge in Γ(A), the walk w is indeed well-defined. Likewise, the walk q0 · · · q2c+j−1zzz · · · w′ = p′ 0 · · · p′ 2d−1z′z′z′ · · · of w′. Since q2i+j 6= p′ is well-defined. Since all admissible edges of eq have the same parity as qjqj+1, the same is true for ew. Moreover, every admissible extension of qjqj+1 in w is a prefix dense in w by Lemma 2.14. By Remark 3.7, ew contains no dense edges. Therefore, 2i for all 0 ≤ i ≤ d as noted above, the edge qjqj+1 is not E(A) is not finitely generated by Theorem 3.6. (cid:3) In many cases, one can determine if E(A) is finitely generated well before the upper bound above. Indeed if GKdim E(A) = 1, there are finitely many infinite If d = GKdim E(A) < ∞, it is easy to describe a recursive anchored walks. procedure: (1) Analyze the (finite number of) subgraphs of Γ(A) with at most d−1 distinct circuits (ignoring multiplicity) in any walk. (2) If no anchored walk with infinitely many indecomposable prefixes is found, let M be the maximal multiplicity of a circuit in an indecomposable walk. Analyze the (finite number of) infinite walks w containing d distinct circuits (ignoring multiplicity) such that the first d − 1 circuits of w occur with multiplicity ≤ M . 5. The Noetherian Property Green et. al. [10] observed that if A is a monomial quadratic algebra, it is possible to determine if E(A) is Noetherian by considering its Ufnarovski relation graph. As noted in Section 2, if A is a monomial quadratic algebra, then Γ(A) is precisely the Ufnarovski graph of the Koszul dual A! ∼= E(A) with edge orientations reversed. In this section we prove an analog of Green et. al.'s "Noetherianity" theorem (Theorem 5.4 of [10]) holds for Γ(A). The following Lemma illustrates an important difference between quadratic mono- mial algebras and monomial algebras with defining relations in higher degrees. The Lemma also conveys the sense in which Theorem 5.2 below generalizes the result in [10]. Lemma 5.1. Let A be a monomial k-algebra such that the defining ideal of A is generated by quadratic monomials. Then G = G0 and every edge of Γ(A) is admissible. FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 15 Proof. Since the minimal generators of I are quadratic, for any generator xj , Axj consists of linear monomials. Thus G1 ⊂ G0 and G = G0. It follows (see Remark 2.3) that every edge of Γ(A) is admissible. (cid:3) For our discussion of the Noetherian property, we discard the assumption that ideals in a graded k-algebra are generated by homogeneous elements of degrees ≥ 2. To establish the main theorem of this section in the left Noetherian case, we filter a left ideal by defining a total order on the path basis of Proposition 2.5. We invoke this total order only when Γ(A) has the property that every vertex lying on an oriented circuit has out-degree 1. To handle the right Noetherian case, one first defines the analogous total order under the assumption that every vertex on an oriented circuit has in-degree 1. In the interest of brevity, we provide details only for the left Noetherian case. We define the order in several steps. We first fix a total ordering of the s circuits of Γ(A): C1 < C2 < · · · < Cs. An in-path p for a circuit Ci is an anchored path p with the final vertex of p on Ci and no other vertex of p on Ci. The set of in-paths to a particular circuit is finite, so we fix a total ordering on each set of in-paths. Of the maximal paths in Γ(A), finitely many terminate on no circuit. We fix a total ordering on these paths as well and define them to be less than any in-path. If p and q are in-paths of lengths n and m for circuits Ci and Cj respectively, we define p < q if n < m or n = m and i < j or n = m and i = j and p < q in the fixed ordering on in-paths of Ci. If w is any anchored walk in Γ(A), then there exists a unique path w such that exactly one of the following holds: • w is an in-path terminating on Ci with i minimal and w extends w or • w is an in-path terminating on Ci and is a proper prefix of w or • w is a maximal extension of w terminating on no circuit If p and q are anchored walks of lengths n and m respectively, we define εp < εq if n < m or if n = m and p < q. Theorem 5.2. For a monomial k-algebra A, the Yoneda algebra E(A) is left (resp. right) Noetherian if and only if (1) every vertex of Γ(A) lying on an oriented circuit has out-degree (resp. in- degree) 1, and (2) every edge of every oriented circuit is admissible. Proof. If Γ(A) contains no circuit, E(A) is finite dimensional, hence Noetherian, by Corollary 2.8. Assume Γ(A) contains a circuit. First suppose Γ(A) satisfies conditions (1) and (2). Let J be any left ideal of E(A). We claim J is finitely generated. Order the path basis as described above. For any class ε ∈ E(A), let h(ε) be the largest basis element appearing with nonzero coefficient when ε is expressed in the path basis. Let F • be the natural filtration on J inherited from the cohomology grading on E(A). For n > 0 let Ln be the left ideal generated by {h(ε) : ε ∈ F nJ}. Conditions (1) and (2) guarantee the existence of a largest integer d such that the final edge in a path p of length d is not admissible. Then by Corollary 3.4, εq is decomposable for any anchored walk q of length > d. It follows that L/Ld is finitely generated as a left ideal (if not, one could find anchored walks p and q with Let L = Sn Ln. 16 CONNER, KIRKMAN, KUZMANOVICH, MOORE p ⊢ q and εp and εq algebraically independent), hence the ascending chain of left ideals L1 ⊂ L2 ⊂ · · · stabilizes. The fact that J is finitely generated then follows by the standard Hilbert Basis argument. Conversely, let C = c0 · · · cn be a circuit of length n in Γ(A). First suppose vertex ci has out-degree > 1, where 0 ≤ i < n. Let v 6= ci+1 be a vertex such that ci → v is an edge in Γ(A). Let p be any anchored path of length m in Γ(A) such that pm = cn−1. For ℓ ≥ 0, define qℓ = pC ℓc0 · · · civ where C ℓ indicates the circuit C is traversed ℓ times. Let J be the left ideal of E(A) generated by {εqℓ : ℓ ≥ 0}. We claim that J is not finitely generated. If J is finitely generated, there exists L > 0 such that εqℓ is in the left ideal generated by εq0 , . . . , εqL for all ℓ > L. Fix ℓ0 > L. Then by Proposition 2.5 and Corollary 3.4, there exists a walk w and an index 0 ≤ d ≤ L such that εqℓ = εw ⋆ εqd and qℓ ∼ qdw. Since v 6= ci+1 and since pC dc0 · · · ci is a prefix of qℓ, Lemma 2.10(2) implies qd is an even-length walk. But by Lemma 2.10(3) and (4), w0 ⊗ v = ci+2 ⊗ ci+1 (where, if i = n − 1, ci+2 = c1) and deg(v) = deg(ci+1), implying v = ci+1, a contradiction. Thus if Γ(A) contains a vertex of out-degree > 1 lying on a circuit, E(A) is not left Noetherian. Suppose instead that every vertex of C has out-degree 1 and C contains an edge cj → cj+1 which is not admissible. Let K be the left ideal of E(A) generated by εqi for i ≥ 0 where qi = pC ic0 · · · cj−1 (if j = 0, then since c0 = cn we take qi = pC ic0 · · · cn−1). Since cjcj+1 is not admissible, and since cj is the only successor of cj−1 in Γ(A), it follows from Lemma 2.10(1) and Proposition 3.3 that εw ⋆ εqi = 0 for any admissible walk w. Thus K is an infinitely-generated trivial left ideal and E(A) is not left Noetherian. We omit the analogous proof for the right Noetherian case. (cid:3) For the algebras A and B of Example 2.2, E(A) and E(B) are neither left nor right Noetherian. Comparing our graph-theoretic characterizations of GK dimen- sion and the Noetherian property, we have the following immediate corollary. Corollary 5.3. Let A be a monomial k-algebra. If E(A) is left or right Noetherian, then GKdimE(A) ≤ 1. If E(A) is Noetherian then Γ(A) consists of finitely many disjoint circuits and paths. 6. Examples The following example suggests that the GKdim(E(A)) = ∞ case can be quite complicated; edges that are dense in one infinite walk need not be dense in another. Example 6.1. Let S = khw, x, y, z, W, X, Y, Z, p, qi be a free algebra and let I be the ideal generated by pqwxyz wxyzp Y Zpqwx pqW XY Z W XY Zp XY ZpqW X yzpqW X xyzpqwx Let A be the factor algebra A = S/I. The graph Γ(A) has two components and is shown in Figure 1 below. Admissible edges are indicated by solid arrows; dashed arrows are non-admissible edges. Vertex pq is common to two oriented circuits, so GKdim(E(A)) = ∞. There are many infinite walks in Γ(A). The walk p → wxyz → pq → wxyz → pq → · · · FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 17 is anchored and wxyz → pq is dense in this walk since the even-length walk wxyz → pq → wxyz → pq → wxyz z → pqwxy → xyz → pqw → wxyz is equivalent to However, the walk p → wxyz → pq → W XY Z → pq → wxyz → pq → W XY Z → · · · contains no dense edge. To see this, observe that the equivalent anchored walks cor- responding to odd-length admissible extensions of wxyz (and likewise of W XY Z) terminate on the circuit yz → pqwx → Y Z → pqW X → yz It follows that no even-length extension of wxyz → pq is admissible because condi- tion (2) of Lemma 2.10 cannot be satisfied. By Theorem 3.6, E(A) is not finitely generated. Example 6.2. Let A = khx, yi/hx3 − x2y, xy2, y3i and observe x4 = 0 in A. The degree-lexicographic ordering on monomials in khx, yi with x < y yields the as- sociated monomial algebra A′ = khx, yi/hx2y, xy2, y3, x4i. Although dim Ei(A) ≤ dim Ei(A′) for all i, one can check that equality does not always hold. The graph Γ(A′) is shown below. By Corollary 2.8, we have GKdim(E(A′)) = 2. It follows that GKdim(E(A)) ≤ 2. x3 ( x xy y2 y Γ(A′) x2 We leave to the reader the straightforward verification that GKdim(E(A)) > 1, hence GKdim(E(A)) = 2 by Bergman's gap theorem. In many cases of interest, knowing GKdim(E(A′)) provides little or no infor- maiton about GKdim(E(A)). Consider the algebra A = khx, y, zi hxy − z2, zx − y2, yz − x2i The algebra A is a 3-dimensional Sklyanin algebra, hence GKdim(E(A)) = 0. Using lexicographic ordering with z > y > x, the associated monomial algebra of A is A′ = khx, y, zi hz2, zx, yz, y3, zy2, yxy, yx3, y2x2, zyx2i Constructing the CPS graph of A′ reveals GKdim(E(A′)) = ∞. ( i i o o O O o o / / I I f f 18 CONNER, KIRKMAN, KUZMANOVICH, MOORE xyzpqw /❴❴❴❴ w o❴ ❴ ❴ ❴ ❴ ❴ ❴ ❴ ❴ ❴ xyzpq x Y Zpqw /❴❴❴❴ W X Y Zpq t t t t yt %❏❏❏❏❏❏❏❏❏❏❏ ✉ yzpq d■■■■■■■■■■ :✉ ✉ ✉ ✉ wx o❴ ❴ ❴ ❴ yzpqW X XY Zpq /❴❴❴❴❴❴❴❴❴ W o❴ ❴ ❴ ❴ XY ZpqW / pqwxy ❍ r ❍ ❍ ❍ r r r yr z Y Z $❍ xyz ;✇✇✇✇✇✇✇✇✇✇ pqwx e❑❑❑❑❑❑❑❑❑❑❑ 9ssssssssss ysssssssss ❑ ❑ ❑ XY Z e❑ pqW X ❑ pqW XY xrrrrrrrrrrr &▲▲▲▲▲▲▲▲▲▲▲ 8rrrrrr yz Z ✤✕ ✮ pq ✮ pqW W XY Z 8qqqqqqqqqqqq 4❤❤❤❤❤❤❤❤❤❤ &▼▼▼▼▼▼▼▼▼▼▼▼▼ ❱❱❱❱❱❱❱❱❱❱❱ p ✕ wxyz j❚ ❚ ❚ ❚ ❚ ❚ ❚ ❚ ❚ ❚ t✐ ✐ ✐ ✐ ✐ ✐ ✐ ✐ ✐ ✐ pqw Figure 1. The graph Γ(A) for Example 6.1. References 1. S. Akbari and A. Mohammadian, Zero-divisor graphs of non-commutative rings, Journal of Algebra 296 (2006), no. 2, 462 -- 479. 2. David J. Anick, On monomial algebras of finite global dimension, Trans. Amer. Math. Soc. 291 (1985), no. 1, 291 -- 310. MR 797061 (86k:16002) / o O O y / % d O O o : O O / o / $ y x & y ; e 9 8 e o o 4 8 & H H ✤ j t I I FINITENESS CONDITIONS ON THE YONEDA ALGEBRA OF A MONOMIAL ALGEBRA 19 3. Luchezar L. Avramov, Local algebra and rational homotopy, Algebraic homotopy and local algebra (Luminy, 1982), Ast´erisque, vol. 113, Soc. Math. France, Paris, 1984, pp. 15 -- 43. MR 749041 (85j:55021) 4. Thomas Cassidy and Brad Shelton, Generalizing the notion of Koszul algebra, Math. Z. 260 (2008), no. 1, 93 -- 114. MR MR2413345 (2009e:16047) 5. Randall E. Cone, Finite generation of ext-algebras for monomial algebras, Ph.D. Thesis, Virginia Polytechnic Institute and State University (2010). 6. Gabriel Davis, Finiteness conditions on the Ext-algebra of a cycle algebra, J. Algebra 310 (2007), no. 2, 526 -- 568. MR 2308170 (2008c:16013) 7. Yu. L. Ershov, S. S. Goncharov, A. Nerode, J. B. Remmel, and V. W. Marek (eds.), Handbook of recursive mathematics. Vol. 2, Studies in Logic and the Foundations of Mathematics, vol. 139, North-Holland, Amsterdam, 1998, Recursive algebra, analysis and combinatorics. MR 1673582 (99k:03004) 8. Y. F´elix and J.-C. Thomas, The radius of convergence of Poincar´e series of loop spaces, Invent. Math. 68 (1982), no. 2, 257 -- 274. MR 666163 (84f:55007) 9. Yves F´elix, Stephen Halperin, and Jean-Claude Thomas, Elliptic Hopf algebras, J. London Math. Soc. (2) 43 (1991), no. 3, 545 -- 555. MR 1113392 (92i:57033) 10. E. L. Green, N. Snashall, O. Solberg, and D. Zacharia, Noetherianity and Ext, J. Pure Appl. Algebra 212 (2008), no. 7, 1612 -- 1625. MR 2400732 (2009c:16024) 11. E. L. Green and D. Zacharia, The cohomology ring of a monomial algebra, Manuscripta Math. 85 (1994), no. 1, 11 -- 23. MR 1299044 (95j:16012) 12. T. H. Gulliksen, A homological characterization of local complete intersections, Compositio Math. 23 (1971), 251 -- 255. MR 0301008 (46 #168) 13. , On the deviations of a local ring, Math. Scand. 47 (1980), no. 1, 5 -- 20. MR 600076 (82c:13022) 14. Michael Jollenbeck and Volkmar Welker, Minimal resolutions via algebraic discrete Morse theory, Mem. Amer. Math. Soc. 197 (2009), no. 923, vi+74. MR 2488864 (2009m:13017) 15. Christopher Phan, Koszul and generalized Koszul properties for noncommutative graded al- gebras, Ph.D. Thesis, University of Oregon (2009). 16. Alexander Polishchuk and Leonid Positselski, Quadratic algebras, University Lecture Series, vol. 37, American Mathematical Society, Providence, RI, 2005. MR MR2177131 17. Darin R. Stephenson and James J. Zhang, Growth of graded Noetherian rings, Proc. Amer. Math. Soc. 125 (1997), no. 6, 1593 -- 1605. MR 1371143 (97g:16033) 18. V. A. Ufnarovskij, Combinatorial and asymptotic methods in algebra, Algebra, VI, Ency- clopaedia Math. Sci., vol. 57, Springer, Berlin, 1995, pp. 1 -- 196. MR 1360005
1111.3975	2	1111	2012-01-18T08:30:15	A Generalized Next-Closure Algorithm -- Enumerating Semilattice Elements from a Generating Set	[ "math.RA" ]	We present a generalization of the well known Next-Closure algorithm working on semilattices. We prove the correctness of the algorithm and apply it on the computation of the intents of a formal context.	math.RA	math	A Generalized Next-Closure Algorithm -- Enumerating Semilattice Elements from a Generating Set Daniel Borchmann TU Dresden Faculty of Mathematics and Sciences Institute for Algebra [email protected] October 25, 2018 Abstract We present a generalization of the well known Next-Closure algorithm working on semilattices. We prove the correctness of the algorithm and apply it on the computation of the intents of a formal context. 1 Introduction Next-Closure is one of the best known algorithms in Formal Concept Analysis [7] to compute the concepts of a formal context. In its general form it is able to efficiently enumerate the closed sets of a given closure operator on a finite set. This generality might be a drawback concerning efficiency compared to other algorithms like Close- by-One [8, 1], but widens its field of applications. However, there are still applications where Next-Closure might be useful, but is not applicable, because an closure operator on a finite set is not explicitly available. One such example might be the computation of concepts of a fuzzy formal context [3]. Even worse, if a closure operator is given, but not on a finite set, Next-Closure is not directly applicable as well. In those cases most often an ad hoc variation of Next-Closure can be constructed. The aim of this paper is to provide a generalization of Next-Closure which covers those cases, and even goes beyond them. As it turns out, Next-Closure is not about enumerating closed sets of a closure oper- ator, even not on an abstract ordered set. The algorithm is merely about enumerating elements of a certain semilattice, given as an operation together with a generating set. This observation is somewhat surprising, but, as we shall see, quite natural. This paper is organized as follows. First of all we shall revisit the original version of Next-Closure, together with the basic definitions. Then we present our generalized 1 version working on semilattices, together with a complete proof of its correctness. Then we show how this generalized form is indeed a generalization of the original Next-Closure. Additionally, we discuss a new algorithm for enumerating the intents of a given formal context. Finally, we give some outlook on further questions which might be interesting within this line of research. 2 The Next-Closure Algorithm Before we are going to discuss our generalized form of Next-Closure, let us revisit the original version as it is given in [7, 6]. To make our discussion a bit more consistent, we shall make one minor modifications to the presentation given here, which will be explicitly mentioned. Let M be a finite set and let c : P(P ) −→ P(P ) be a function such that a) c is idempotent, i.e. c(c(A)) = c(A) for all A ⊆ M , b) c is monotone, i.e. if A ⊆ B, then c(A) ⊆ c(B) for all A, B ⊆ M , and c) c is extensive, i.e. A ⊆ c(A) for all A ⊆ M . A set A ⊆ M is called closed (with respect to c) if A = c(A), and the image of c is defined as c[P(M )] := { c(A) A ⊆ M }. Without loss of generality, let M = { 1, . . . , n } for some n ∈ N. For two sets A, B ∈ c[P(M )] with A 6= B and i ∈ M we say that A is lectically smaller than B at position i if and only if i = min(A ∆ B) and i ∈ B, where A ∆ B = (A \ B) ∪ (B \ A) is the symmetric difference of A and B. We shall write A ≺i B if A is lectically smaller than B at position i. Finally, we say that A is lectically smaller than B, for A, B ∈ c[P(M )], if A = B or A ≺i B for some i ∈ M and we shall write A (cid:22) B in this case. It has to be noted that, in contrast to our definition, the lectic order is normally defined for all sets A, B ⊆ M in the very same spirit as given above. However, as we shall see, this is not necessary, wherefore we have restricted our definition to closed sets only. Now let us define for A ∈ c[P(M )] and i ∈ M A ⊕ i := c({ j ∈ A j < i } ∪ { i }). Then we have the following result. Theorem 1 (Next-Closure) Let A ∈ c[P(M )]. Then the next closed set A+ ∈ c[P(M )] after A with respect to the lectic order (cid:22), if it exists, is given by with i ∈ M being maximal with A <i A ⊕ i. A+ = A ⊕ i 2 This is the original version of Next-Closure, as it is given in [7]. Now let us have a closer look on the definition of ⊕. The set A ⊕ i can be seen as the smallest closed set containing both { j ∈ A j < i } and { i }, or equivalently, both c({ j ∈ A j < i }) and c({ i }). This means that we can rewrite A ⊕ i as A ⊕ i = c({ j ∈ A j < i }) ∨ c({ i }), where X ∨Y is the smallest closed sets containing both X, Y ∈ c[P(M )], the supremum of X and Y , which is simply given by X ∨ Y = c(X ∪ Y ). This observation allows us to consider Next-Closure on abstract algebraic structures with a binary operation ∨ with some certain properties. To do so we need a more general notion of c({ i }), since we do not necessarily deal with subsets, and a more general notion of { j ∈ A j < i }, which likewise might not be expressible in a more general setting. Finally, we need to find a starting point for our enumeration, which is c(∅) in the original description of Next-Closure, but may vary in other cases. Luckily, all this is possible and quite natural, as we shall see in the next section. 3 Generalizing Next-Closure for Semilattices The aim of this section is to present a generalization of the Next-Closure algorithm that works on semilattices. For this recall that a semilattice L = (L, ∨) is an algebraic structure with a binary operation ∨ which is associative, commutative and idempotent. It is well known that with x ≤L y : ⇐⇒ x ∨ y = y, with x, y ∈ L. An order relation on L is defined such that for every two elements x, y the element x ∨ y is the least upper bound of both x and y with respect to ≤L. For the remainder of this section let L = (L, ∨) be an arbitrary but fixed semilattice. Furthermore, let (xi i ∈ I) be an enumeration of a finite generating set { xi i ∈ I } ⊆ L of L. Finally, let ≤I be a total order on I. Definition 2 Let a, b ∈ L and let i ∈ I. Set ∆a,b := { j ∈ I (xj ≤L a and xj 6≤L b) or (xj 6≤L a and xj ≤L b) }. We then define a <i b : ⇐⇒ i = min ∆a,b and xi ≤L b. Furthermore we write a < b if a <i b for some i ∈ I and write a ≤ b if a = b or a < b.♦ One can see the similarity of this definition to the one of the lectic order. Here, the set ∆a,b generalizes a ∆ b and xi ≤L b somehow represents the fact that i ∈ b, or equivalently { i } ⊆ b, in the special case of L = P(M ) and i ∈ M . Note that if a <i b and k ∈ I with k <I i, then xk ≤L a ⇐⇒ xk ≤L b. 3 This observation is quite useful and will be used in some of the proofs later on. The first thing we want to consider now are two easy results stating that ≤ is a total order relation on L extending ≤L. Lemma 3 The relation < is irreflexive and transitive. Furthermore, for every two elements a, b ∈ L with a 6= b, it is either a < b or b < a. Proof If a = b, then the set ∆a,b defined above is empty, therefore we cannot have a <i a for some i ∈ I. This shows the irreflexivity of <. Let us now consider the transitivity of <. For this let a, b, c ∈ L, i, j ∈ I and suppose that a <i b and b <j c. We have to show that a < c. Let us consider the following cases. Case i <I j. We have xi 6≤L a and xi ≤L b because of a <i b. Due to i <I j it follows that xi ≤L c. Suppose that there exists k ∈ I, k <I i with xk ≤L a and xk 6≤L c. Then if xk 6≤L b we would have xk 6≤L a because k <I i, a contradiction. But if xk ≤L b, then xk ≤L c because of k <I i <I j, again a contradiction. Thus we have shown that a < c. Case j <I i. We have xj 6≤L b, xj ≤L c because of b <j c. Due to j <I i it follows that xj 6≤L a. Now if there were a k ∈ I, k <I j with xk ≤L a and xk 6≤L c, then xk ≤L b would imply xk ≤L c and xk 6≤L b would imply xk 6≤L a, analogously to the first case, a contradiction. Hence such a k cannot exist and a < c. Case i = j. This cannot occur since otherwise xi ≤L b, because of a <i b, and xi 6≤L b, because of b <i c, a contradiction. Overall we have shown that a < c in any case and therefore < is a transitive relation. Finally let a, b ∈ L with a 6= b. Then because { xi i ∈ I } is a generating set, the set ∆a,b is not empty, since otherwise a = b, and with i := min ∆a,b we either have a <i b if xi ≤L b and b <i a otherwise. (cid:3) Lemma 4 Let a, b ∈ L with a ≤L b. Then a ≤ b. In particular, if a ≤L c and b ≤L c for a, b, c ∈ L, then a ∨ b ≤ c. Proof We show xi ≤L a =⇒ xi ≤L b for all i ∈ I. This shows b 6< a, hence a ≤ b by Lemma 3. Now if xi ≤L a, then because of a ≤L b we see that xi ≤L b and the claim is proven. (cid:3) The next step towards a general notion of Next-Closure is to provide a generaliza- tion of ⊕. Definition 5 Let a ∈ L and i ∈ I. Then define a ⊕ i := _ xj ∨ xi. j<I i xj ≤La ♦ With all these definitions at hand we are now ready to prove the promised gener- alization. For this, we generalize the proof of Next-Closure as it is given in [7, page 67]. 4 Lemma 6 Let a, b ∈ L and i, j ∈ I. Then the following statements hold: i) a <i b, a <j c, i <I j =⇒ c <i b. ii) a < a ⊕ i if xi 6≤L a. iii) a <i b =⇒ a ⊕ i ≤ b. iv) a <i b =⇒ a <i a ⊕ i. Proof i) It is xi 6≤L a and due to i <I j we get xi 6≤L c as well. Furthermore, xi ≤L b because of a <i b. Now if there would exist a k ∈ I with k <I i such that xk ≤L c, xk 6≤L b, then xk ≤L a because of k <I i and xk 6≤L a because of k <I i <I j, a contradiction. With the same argumentation a contradiction follows from the assumption that there exists a k ∈ I, k <I i with xk 6≤L c, xk ≤L b. In sum we have shown c <i b, as required. ii) We have xi 6≤L a and xi ≤L a ⊕ i. Furthermore, for k ∈ I, k <I i and xk ≤L a we have xk ≤L a ⊕ i by definition. This shows a < a ⊕ i. iii) Let a <k b for some k ∈ I. Then Wj<I k,xj ≤La xj ≤L b and xk ≤L b, hence with Lemma 4 we get a ⊕ k ≤ b. iv) Let a <i b. Then xi 6≤L a and with (ii) we get a < a ⊕ i. By (iii), a ⊕ i ≤ b. If for k ∈ I, k <I i it holds that xk ≤L a ⊕ i and xk 6≤L a, then we also have xk ≤L a ⊕ i ≤L b, i.e. xk ≤L b, contradicting the minimality of i. (cid:3) Theorem 7 (Next-Closure for Semilattices) Let a ∈ L. Then the next element a+ ∈ L with respect to <, if it exists, is given by with i ∈ I being maximal with a <i a ⊕ i. a+ = a ⊕ i Proof Let a+ be the next element after a with respect to <. Then a <i a+ for some i ∈ I and by Lemma 6.iv we get a <i a ⊕ i and with Lemma 6.iii we see a ⊕ i ≤ a+, hence a ⊕ i = a+. The maximality of i follows from Lemma 6.i. (cid:3) To find the correct element i ∈ I such that a+ = a ⊕ i can be optimized with Lemma 6.ii. Because of this result, only elements i ∈ I with xi 6≤L a have to be considered, a technique which is also known for the original form of Next-Closure. However, to make the above theorem practical for enumerating the elements of a certain semilattice, one has to start with some element, preferably the smallest element in L with respect to ≤. This element must also be minimal in L with respect to ≤L, by Lemma 4. Since { xi i ∈ I } is a generating set of L, and a ≤ a ∨ b for all a, b ∈ L, the minimal elements of L with respect to ≤L must be among the elements xi, i ∈ I. So to find the first element of L with respect to ≤, find all minimal elements in { xi i ∈ I } and choose the smallest element with respect to ≤ from them. But because of xi ≤ xj 5 if and only if j <I i, one just has to take the largest index j of all minimal elements among the xi to find the smallest element in L with respect to ≤. As a final remark for this section note that the set { xi i ∈ I } must always include the ∨-irreducible elements of L. These are all those elements a ∈ L that cannot be represented as a join of other elements, or, equivalently, { b ∈ L b <L a } = ∅ or _ b <L a. b<La It is also easy to see that the ∨-irreducible elements of L are also sufficient, i.e. they are a generating set of L. 4 Computing the Intents of a Formal Context We have seen an algorithm that is able to enumerate the elements of a semilattice from a given generating set. We have also claimed that this is a generalization of Next-Closure, which we want to discuss in this section. Furthermore, we want to give another example of an application of this algorithm, namely the computation of the intents of a given formal context. Firstly, let us reconstruct the original Next-Closure algorithm from Theorem 7 and the corresponding definitions. For this let M be a finite set and let c be a closure operator on M = { 0, . . . , n − 1 }, say. We then apply Theorem 7 to the semilattice P = (c[P(M )], ∨). We immediately see that ≤P = ⊆ and that <i is the usual lectic order on P . Then the set { c({ i }) i ∈ M } ∪ { c(∅) } is a finite generating set of P and we can define xi := c({ i }) and xn := c(∅), i.e. I = { 0, . . . , n }. For a closed set A ⊆ M and i ∈ I then follows A ⊕ i = _ xi ∨ xi j<i xj ⊆A = c( [ xj) ∨ xi j<i xj ⊆A = c([{ c({ j }) j < i, j ∈ A }) ∨ c({ i }) = c({ j j < i, j ∈ A }) ∨ c({ i }) = c({ j j < i, j ∈ A } ∪ { i }) which is the original definition of ⊕ for Next-Closure. Furthermore, it is A ⊕ n = A since c(∅) ⊆ A for each closed set A. We therefore do not need to consider xn when looking for the next closed set, and indeed, the only reason why xn = c(∅) has been included is that it is the smallest closed set in P . All in all, we see that Next-Closure is a special case of Theorem 7. 6 However, for a closure operator c on a finite set M it seems more natural to consider the semilattice P = (c[P(M )], ∩), because the intersection of two closed sets of c again yields a closed set of c. One sees that ≤P = ⊇. As a generating set we take the set of ∩-irreducible elements { Xi i ∈ G } for some index set G. Let A, B ∈ c[P(M )] and let <G be a linear ordering on G. Then A < B if and only if there exists i ∈ G such that i = min{ j ∈ G (Xi ⊇ A, Xi 6⊇ B) or (Xi 6⊇ A, Xi ⊇ B) } and Xi ⊇ B and ⊕ is just given by A ⊕ i = \ Xj ∩ Xi. j<Gi Xj ⊇A Now note that ⊕ does not need the closure operator c anymore. This means that if the computation of c is very costly and the ∩-irreducible elements (or a superset thereof) is known, this approach might be much more efficient. In general, however, it is not known how to efficiently determine the ∩-irreducible closed sets of c. But if c is given as the ·′′ operator of a formal context, these irreducible elements can be determined quickly [7]. Let G and M be two finite sets and let J ⊆ G × M . We then call the triple K := (G, M, J) a formal context, G the objects of the formal context and M the attributes of the formal context. For g ∈ G and m ∈ M we write g J m for (g, m) ∈ J and say that object g has attribute m. Let A ⊆ G and B ⊆ M . We then define the derivations of A and B to be A′ := { m ∈ M ∀g ∈ A : g J m } B′ := { g ∈ G ∀m ∈ B : g J m }. Then the ·′′ operator is just the twofold derivation of a given set of attributes. It turns out that this is indeed a closure operator, and that every closure operator can be represented as a ·′′ operator of a suitable formal context [7, 4]. The closed sets of ·′′, i.e. all sets B ⊆ M with B = B′′, are called the intents of K and shall be denoted by Int(K). It is clear from the previous remarks that (Int(K), ∩) is a semilattice. The advantage of representing a closure operator is that the ∩-irreducible elements of (Int(K), ∩) can be directly read off from the format context. As discussed in [7], the set { { g }′ g ∈ G } contains the irreducible elements we are looking for, except M . Furthermore, it is possible to omit certain objects g from K without changing Int(K). Every object g ∈ G can be omitted from K for which the set { g }′ is either equal to M or can be represented as a proper intersection of other sets { g1 }′, . . . , { gn }′ for some elements g1, . . . , gn ∈ G. It is also clear that if there exist two distinct objects g1 and g2 with { g1 }′ = { g2 }′, that we can remove one of them without changing Int(K). A formal context for which no such objects exist is called object clarified and object reduced. If 7 Listing 1: Compute Next Intent of a Formal Context define next−intent (K = (G, M, J) , A) for g ∈ G , descending i f { g }′ 6⊇ A then let (B := A ⊕ g) i f ∀h ∈ G, h <G g, { h }′ 6⊇ A : { h }′ 6⊇ B then return B end i f end let end i f end for return nil end K = (G, M, J) is an object clarified and object reduced formal context, then the set { { g }′ g ∈ G } is exactly the set of ∩-irreducible intents of K, except for the set M . The above described algorithm now takes the following form when applied to (Int(K), ∩). As index set we choose the set G of object of the given formal context, ordered by <G. For every object g ∈ G we set xg := { g }′. Then the set { xg g ∈ G } is a generating set of the semilattice (Int(K) \ { M }, ∩), which we want to enumerate (since we get the set M for free). For A begin an intent of K and g ∈ G we have A ⊕ g := \ { h }′ ∩ { g }′ h∈G h<Gg and as the first intent we take M . For an intent A ⊆ M of K we then have to find the maximal object (with respect to <G) g ∈ G such that A <g A⊕g. This is equivalent to g being maximal with { g }′ 6⊇ A and ∀h ∈ G, h <G g : { h }′ ⊇ A ⇐⇒ { h }′ ⊇ A ⊕ g. However, the direction "=⇒" is clear, hence we only have to ensure ∀h ∈ G, h <G g : { h }′ 6⊇ A =⇒ { h }′ 6⊇ A ⊕ g. All these considerations yield the algorithm shown in Listing 1. Of course, the deriva- tions of the form { g }′ should not be computed every time they are needed but rather stored somewhere for reuse. In contrast to the original version of Next-Closure used to compute the intents of a given formal context, our version only traverses the formal context K once, when computing { g }′ for every g ∈ G. This might be an advantage if accessing the context is computationally expensive, but also be a disadvantage if there are much more objects than attributes (which occurs quite often in practice). However, in this case one might better compute the extents of the formal context, i.e. all sets A ⊆ G with A = A′′, which also form a closure system. Then the algorithm in Listing 1 would run through the set of attributes to compute the extent. 8 Finally, let us consider the time complexity of the new algorithm to compute the intent after A for a given formal context K = (G, M, J). If we assume the operations ∩ and ⊇ to be constant (i.e. independent of the size of both G and M , a very optimistic assumption), then the algorithm from Listing 1 roughly needs G × (G + G) steps to compute the next intent, i.e. has worst time complexity O(G2). On the other hand, if one assumes a naive implementation of ∩ and ⊇ taking time O(M ), one has time complexity G × (M + G × M + G × M ), hence worst case complexity O(G2 × M ), which is the time complexity of Next-Closure when used to compute the extents of K. 5 Conclusion We have seen a natural generalization of the Next-Closure algorithm to enumerate elements of a semilattice from a generating set. We have proven the algorithm to be correct and applied it to the standard task of computing the intents of a given formal context, yielding a new algorithm to accomplish this. However, there are still some interesting ideas one might want to look at. Firstly, a variation of the original Next-Closure algorithm is able to compute the stem base of a formal context, a very compact representation of its implicational knowledge. It would be interesting to know whether the generalization given in this paper gives more insight into the computation, and therefore into the nature of the stem base. This is, however, quite a vague idea. Secondly, the new algorithm discussed above to compute the intents of a formal context enumerates them in a certain order, which might or might not be a lectic one. Understanding this order relation might be fruitful, especially with respect to the complexity results obtained recently, which consider enumerating pseudo-intents of a formal context in lectic order [5]. Finally, and more technically, it might be interesting to look for other applications of our general algorithm. As already mentioned in the introduction, the enumeration of fuzzy concepts of a fuzzy formal context might be worth investigating (and it might be interesting comparing it to [2]), but there might be other applications. References [1] Simon Andrews. In Sebastian Rudolph, Frithjof Dau, and Sergei O. Kuznetsov, editors, ICCS Supple- mentary Proceedings, Moscow, 2009. In-Close, a fast algorithm for computing formal concepts. [2] Radim Belohlávek, Bernard De Baets, Jan Outrata, and Vilém Vychodil. Com- puting the Lattice of All Fixpoints of a Fuzzy Closure Operator. IEEE T. Fuzzy Systems, 18(3):546 -- 557, 2010. 9 [3] Radim Belohlávek and Vilém Vychodil. Attribute Implications in a Fuzzy Setting. In Rokia Missaoui and Jürg Schmid, editors, ICFCA, volume 3874 of Lecture Notes in Computer Science, pages 45 -- 60. Springer, 2006. [4] Daniel Borchmann. Decomposing Finite Closure Operators by Attribute Explo- ration. In ICFCA 2011, Supplementary Proceedings, 2011. [5] Felix Distel. Hardness of Enumerating Pseudo-intents in the Lectic Order. In Léonard Kwuida and Baris Sertkaya, editors, ICFCA, volume 5986 of Lecture Notes in Computer Science, pages 124 -- 137. Springer, 2010. [6] Bernhard Ganter. Two basic algorithms in concept analysis. FB4-Preprint Nr. 831, 1984. [7] Bernhard Ganter and Rudolph Wille. Formal Concept Analysis: Mathematical Foundations. Springer, Berlin-Heidelberg, 1999. [8] Vilém Vychodil, Petr Krajča, and Jan Outrata. Advances in algorithms based on CbO. In Marzena Kryszkiewicz and Sergei Obiedkov, editors, Concept Lattices and Their Application, pages 325 -- 337, 2010. 10
1112.3835	1	1112	2011-12-16T14:50:36	On the smoothness of centres of rational Cherednik algebras in positive characteristic	[ "math.RA", "math.RT" ]	In this article we study rational Cherednik algebras at $t = 1$ in positive characteristic. We study a finite dimensional quotient of the rational Cherednik algebra called the restricted rational Cherednik algebra. When the corresponding pseudo-reflection group belongs to the infinite series $G(m,d,n)$, we describe explicitly the block decomposition of the restricted algebra. We also classify all pseudo-reflection groups for which the centre of the corresponding rational Cherednik algebra is regular for generic values of the deformation parameter.	math.RA	math	ON THE SMOOTHNESS OF CENTRES OF RATIONAL CHEREDNIK ALGEBRAS IN POSITIVE CHARACTERISTIC GWYN BELLAMY AND MAURIZIO MARTINO Abstract. In this article we study rational Cherednik algebras at t = 1 in positive characteristic. We study a finite dimensional quotient of the rational Cherednik algebra called the restricted rational Cherednik algebra. When the corresponding pseudo-reflection group belongs to the infinite series G(m, d, n), we describe explicitly the block decomposition of the restricted algebra. We also classify all pseudo-reflection groups for which the centre of the corresponding rational Cherednik algebra is regular for generic values of the deformation parameter. Dedicated, with admiration and thanks, to Ken Brown and Toby Stafford on their 60th birthdays 1. Introduction 1.1. Rational Cherednik algebras were introduced by Etingof and Ginzburg in 2002. Since their introduction they have been extensively studied and have been shown to be related to many other branches of mathematics such as integrable systems, symplectic algebraic geometry and algebraic combinatorics. In this article we continue the study, initiated in [BFG], [BC3], [BC2] and [BC1], of these algebras at t = 1 and over a field of positive characteristic. We focus on the representation theoretic aspects of the story. In particular, we examine the block structure of certain finite-dimensional quotient algebras called restricted rational Cherednik algebras. We also look at the question of when the centre of the rational Cherednik algebra is smooth. Analogous problems have already been solved for rational Cherednik algebras at t = 0 in characteristic zero, see [EG], [Gor1], [Bel2], [Mar2], [Mar3], [Bel1], and our results are very similar in nature. The methods we develop, however, are new, and in fact can be used to reprove many of the characteristic zero results. 1.2. Let us review our results. Further details can be found in the main body of the paper. Let W be a pseudo-reflection group. Let k be an algebraically closed field of characteristic p with p ∤ W . Let V denote the reflection representation of W over k. Let S(W ) denote the set of reflections in W and let c : S(W ) → k be a W -invariant function. To this data we can attach a k-algebra Hc(W ) called the rational Cherednik algebra. Let V (1) denote the Frobenius twist of V . Let Zc(W ) denote the centre of Hc(W ). There is an injective algebra homomorphism k[V (1)]W ⊗ k[(V ∗)(1)]W ֒→ Zc(W ). Factoring out by the unique graded, maximal ideal in this central subalgebra, one gets a finite dimensional, graded quotient of the rational Cherednik algebra. This factor algebra is called the restricted rational Cherednik algebra and is denoted Hc(W ). Simple modules for this finite dimensional algebra are in natural bijection with the simple modules of the group W . Thus, the blocks of Hc(W ) give us a partition of the set 1 IrrW . Our first main result is an explicit combinatorial description of this block partition when W belongs to the infinite series G(m, 1, n). In this case, the irreducible representations of W are naturally labeled by P(m, n), the set of m-multipartitions of n. For each λ ∈ P(m, n), let Lc(λ) denote the corresponding simple Hc(W )-module. For the definitions in the following statement and the proof of the following theorem, see 4.5. Theorem. Let λ, µ ∈ P(m, n). Let a = (0, H1, H1 + H2, . . . , H1 + · · · + Hm−1). Then Lc(λ) and Lc(µ) belong to the same block of Hc(W ) if and only if x(ap i −ai)Res λi (x−(κp−κ)) = m−1Xi=0 x(ap i −ai)Res µi (x−(κp−κ)). m−1Xi=0 Using this description of the blocks of the restricted rational Cherednik algebra Hc(G(m, 1, n)), some Clif- ford theory also allows one to describe the blocks of the restricted rational Cherednik algebra Hc(G(m, d, n)), see 4.6. Here G(m, d, n) denotes (in the Shepard-Todd classification) the normal subgroup of G(m, 1, n) where we impose the restriction that d m and either n > 2 or n = 2 and d is odd. 1.3. Our remaining results concern the smoothness of Zc(W ). In section 5 we relate the smoothness of Zc(W ) to the representation theory of Hc(W ). Taking as our starting point the fact that the smooth and Azumaya loci of Zc(W ) are equal, we use the restriction functors of [BE] to establish the following. Theorem. The following are equivalent: (1) Zc(W ) is smooth; (2) the blocks of Hc ′ (W ′) are singletons for all parabolic subgroups W ′ ⊆ W . Here, c′ denotes the restriction of c to S(W ) ∩ W ′. In the case that W = G(m, 1, n), we can apply the theorem together with the description of blocks from Theorem 1.2 to determine for which parameters c the centre Zc(W ) is smooth. Corollary. The centre of Hc(G(m, 1, n)) is smooth if and only if c does not lie on the finitely many hyper- planes in C defined by κ ∈ Fp and ai − aj ± Cκ ∈ Fp, ∀ 0 ≤ i 6= j ≤ m − 1, 0 ≤ C ≤ n − 1. We should note that an analogous version of the theorem and its corollary is true for rational Cherednik algebras at t = 0 over the complex numbers. Using [Mar3, Theorem 5.5], this gives an alternative way to describe the parameter values where the centre is smooth, cf. [Gor2, Lemma 4.3]. The theorem also clarifies the relationship between restricted Cherednik algebras and the smoothness of the centres Zc(W ). 1.4. The corollary shows that the centre of Hc(G(m, 1, n)) is a smooth algebra for generic values of the parameter c. One can ask more generally: for which pseudo-reflection groups W is Zc(W ) smooth for generic values of the parameter c? Our final result answers this question. Theorem. The centre of the rational Cherednik algebra Hc(W ) is never smooth if W is not isomorphic to G(m, 1, n) or G4. If W is isomorphic to G(m, 1, n) for some m and n or to the exceptional group G4, then the centre of Hc(W ) is smooth for generic values of c. 2 Our approach to proving Theorem 1.4 follows the same path as in [Bel2] - we calculate the Poincar´e polynomial of the graded Hc(W )-modules Lc(λ) under the assumption that the dimension of these modules is maximal, namely dimk Lc(λ) = pnW . This leads to a contradiction for many choices of λ. However, the naive argument of loc. cit. is not sufficient in our case to calculate this Poincar´e polynomial. Instead, we show that Lc(λ) can be deformed (flatly) to a graded H0(W ) = D(V ) ⋊ W -module. Using a result of Cartier's on D-modules with zero p-curvature, we study the graded W -character of this module at c = 0. Remark. Theorem 1.4 and Corollary 1.2 provide a complete answer to [BC3, Question D] for rational Cherednik algebras. 1.5. The paper is structured as follows. In section 2 we introduce notation and recall some facts about pseudo-reflection groups. In section 3 we define rational Cherednik algebras and state their main properties. The Dunkl-Opdam operators are introduced in Section 4 and are used to prove Theorem 1.2. Section 5 is devoted to parabolic restriction and induction and their application in the proof of Theorem 1.3. Finally, in section 6 we establish certain properties of D-modules in characteristic p and use these to prove Theorem 1.4. 1.6. Acknowledgements. The first author is supported by the EPSRC grant EP-H028153. The second author was supported by the SFB/TR 45 "Periods, Moduli Spaces and Arithmetic of Algebraic Varieties" of the DFG (German Research Foundation). The authors would like to thank Ulrich Thiel for suggesting that the Euler element should distinguish the blocks for G4 at generic parameters, and for showing us an early version of [Thi]. 2. Basics 2.1. Definitions and notation. Let k be an algebraically closed field of characteristic p > 0. Let V be a k-vector space of finite dimension and W a finite group acting linearly on V . An element s ∈ W is called a pseudo-reflection if the fixed space of s has co-dimension one. Let S(W ) denote the set of all pseudo- reflections in W . Then W is said to be a pseudo-reflection group if W = hS(W )i, a good reference on the theory of pseudo-reflection groups is [LT]. We assume throughout that char(k) does not divide W . This assumption on the characteristic of k implies that there are no transvections in S(W ). One can also check from the classification of pseudo-reflection groups, as recalled in [KM], that this assumption implies that W is the reduction mod p of a complex reflection group. 2.2. Frobenius twists and group actions. Let V be a finite dimensional vector space over k. Let W be a finite group acting linearly on V and assume that p does not divide W . The Frobenius morphism Fr : k[V ] → k[V ] is the ring homomorphism f 7→ f p. Denote by k[V (1)] the image of Fr but with twisted linear structure z ⋆ f = zpf for all f ∈ k[V (1)] and z ∈ k. Then Fr : k[V ] → k[V (1)] is a k-linear isomorphism. Note that k[V ] is a finite free k[V (1)]-module of rank (dim V )p. It is easy to check that Fr is a W -equivariant ring homomorphism, so we have an isomorphism FrW : k[V ]W → k[V (1)]W . In particular, the p-th powers of a generating set for k[V ]W form a generating set for k[V (1)]W . 3 2.3. Representations of pseudo-reflection groups. Let us retain the notation from above. Let K be a finite field extension of Q containing all the W th roots of unity, let A be the localisation of the ring of integers of K at the prime ideal generated by p ∈ Z and let L be the residue field of A. Note that L is a finite field of order a power of p. With this setup we can define a decomposition map on characters of irreducible representations as follows, see [GP, § 7]. Let M be an irreducible KW -module. We can choose an A-lattice MA in M so that the action of W on MA has structure constants in A. Let ML denote the reduction of MA to L, which is naturally a LW -module. The decomposition map is the assignment χ 7→ χL, where χ, χL denote the characters of M and ML, respectively. By [CR, Corollary 17.2] and Tits' deformation theorem, [GP, Theorem 7.4.6], we have the following. Theorem. Both KW and LW are split algebras and the decomposition map defines a bijection between the irreducible characters of KW and LW . In particular, it follows that the irreducible characters of kW are given by reducing the irreducible char- acters of CW to k. 2.4. p-coinvariant rings. Let W be a pseudo-reflection group and let K(W ) denote the Grothendieck group of finite dimensional W -modules. Let IrrW be a complete set of isomorphism classes of simple W - modules. We denote by IrrZW a complete set of isomorphism classes of graded, simple W -modules. For a graded W -module M , we write cht,W (M ) ∈ K(W )[t, t−1] for its graded character. The shift M [i] of M is the graded W -module such that M [i]j = Mj−i. We endow the algebra k[V ] with its usual N-grading. Let d1, . . . , dn be the degrees of a set of fundamental homogeneous algebraically independent generators of k[V ]W . The coinvariant ring of W is k[V ]coW , and for each λ ∈ IrrW , we denote by fλ(t) the corresponding fake polynomial, defined by cht,W (k[V ]coW ) = Xλ∈IrrW fλ(t) · [λ]. Definition. Let k[V (1)]W is defined to be the finite dimensional graded algebra k[V ]pcoW := k[V ]/hk[V (1)]W + denote the invariant polynomials with zero constant term. The p-coinvariant ring + i. Lemma. Keep notation as above. Then there is an isomorphism of graded W -modules k[V ]pcoW ≃ (k[V ]/hk[V ]W + i) ≃ (k[V ]/hk[V (1)]+i) ⊗ (k[V (1)]/hk[V (1)]W + i) ⊗ (k[V ]W /hk[V (1)]W + i). (1) (2) Proof. Consider the inclusions of algebras k[V (1)]W 9ttttttttt %❏❏❏❏❏❏❏❏❏ #●●●●●●●●● ;①①①①①①①① k[V ] k[V ]W k[V (1)] 4 # 9 % ; Choose a basis x1, . . . , xn of V ∗, homogeneous generators f1, . . . , fn of k[V ]W and a free homogeneous basis b1, . . . , bW of k[V ] as a k[V ]W -module. Then there are free bases for k[V ] as a k[V (1)]W -module given by: and {f αbj 0 ≤ αi ≤ p − 1, 1 ≤ j ≤ W }, {xαbp j 0 ≤ αi ≤ p − 1, 1 ≤ j ≤ W }. Taking the images of these bases in k[V ]pcoW yields the lemma. (cid:3) Remark. The isomorphism (1) implies that for all λ ∈ IrrW . [k[V ]pcoW i : λ]ti = fλ(t) · Xi∈Z 1 − tpdi 1 − tdi nYi=1 3. Rational Cherednik algebras 3.1. For s ∈ S(W ), fix αs ∈ V ∗ to be a basis of the one dimensional space Im(s − 1)V ∗ and α∨ s ∈ V a basis of the one dimensional space Im(s − 1)V , normalised so that αs(α∨ s ) = 1. Let C denote the space of W -equivariant functions S(W ) → k and choose c ∈ C and t ∈ k. The rational Cherednik algebra, Ht,c(W ), as introduced by Etingof and Ginzburg [EG, page 250], is the quotient of the skew group algebra of the tensor algebra, T (V ⊕ V ∗) ⋊ W , by the ideal generated by the relations [x, x′] = 0, [y, y′] = 0, [y, x] = tx(y) −Xs∈S c(s)αs(y)x(α∨ s )s, (3) for all x, x′ ∈ V ∗ and y, y′ ∈ V . We define a filtration F• on Ht,c(W ) via F0 = kW , F1 = kW ⊗ (V ⊕ V ∗) and Fi = F i 1 for i > 1. By [EG, Theorem 1.3], there is an isomorphism of algebras grF Ht,c(W ) ∼= S(V ⊕ V ∗) ⋊ W. As a consequence, there is a vector space isomorphism Ht,c(W ) ∼= k[V ] ⊗ kW ⊗ k[V ∗]. (4) (5) There is also a Z-grading on Hc(W ) given by setting deg(W ) = 0, deg(V ) = −1 and deg(V ∗) = 1. Through- out this article we assume that t 6= 0. Therefore, without loss of generality t ≡ 1 and we write Hc(W ) for H1,c(W ). Let x1, . . . , xn be a basis of V ∗ and y1, . . . , yn ∈ V the dual basis. Define the Euler element in Hc(W ) to be h = nXi=1 xiyi −Xs∈S cs 1 − λs s. One can easily check that [h, x] = x, [h, y] = −y and [h, w] = 0 for all x ∈ V ∗, y ∈ V and w ∈ W . Therefore the element hp − h belongs to the centre of Hc(W ). 5 3.2. Below we summarize fundamental the properties of Hc(W ). Proofs of all these statements can be found in [BC3]. Proposition. Let H := Hc(W ) be a rational Cherednik algebra associated to (V, W ). (1) The P.I. degree of H equals pnW . (2) The centre Zc(W ) of H is an affine domain and the algebra H is a finite module over its centre. (3) The smooth locus of Zc(W ) equals the Azumaya locus of H. (4) The commutative subalgebras k[V (1)]W and k[(V ∗)(1)]W of H are central. 3.3. The restricted rational Cherednik algebra. Let L be a simple, graded Hc(W )-module. The centre of Hc(W ) acts as a scalar on L and the grading on L implies that (k[V (1)]W ⊗ k[(V ∗)(1)]W )+ annihilates L. Therefore to study these simple, graded modules it suffices to consider a certain graded, finite dimensional quotient of Hc(W ). Definition. The restricted rational Cherednik algebra Hc(W ) is the finite dimensional quotient of Hc(W ) by the central ideal generated by (k[V (1)]W ⊗ k[(V ∗)(1)]W )+. The algebra Hc(W ) is Z-graded and has dimension p2nW 3. The PBW property (5) implies that Hc(W ) ∼= k[V ]pcoW ⊗ kW ⊗ k[V ∗]pcoW as vector spaces. Since k[V ]pcoW is a complete intersection, [Eis, Corollary 21.19] implies that it is Gorenstein and thus equipped with a non-degenerate bilinear form. Then the proof of [BGS, Theorem 3.6] shows that the algebra Hc(W ) is symmetric. Definition. Let λ ∈ IrrZ(W ). The baby Verma module ¯∆(λ) associated to λ is the induced module ¯∆(λ) = Hc(W ) ⊗A λ, where A = k[V ∗]pcoW ⋊ W and the natural action of W on λ extends to A by making V act by zero. As was done in [Gor1], we can apply the theory developed in [HN] to the category Hc(W )-grmod of finite- dimensional, graded, left Hc(W )-modules. The forgetful functor Hc(W )-grmod → Hc(W )-mod is denoted F . Proposition. Let λ, µ ∈ IrrZW . (1) The baby Verma module ¯∆(λ) has a simple head L(λ). (2) ¯∆(λ) is isomorphic to ¯∆(µ) if and only if λ = µ in IrrZW . (3) The set {L(λ) λ ∈ IrrZW } is a complete set of isomorphism classes of simple modules in Hc(W )-grmod. (4) The set {F (L(λ)) λ ∈ IrrZW } is a complete set of isomorphism classes of simple Hc(W )-modules. Proposition 3.3 shows that there is a natural bijection, λ 7→ L(λ), between IrrW and IrrHc(W ). Therefore the blocks of Hc(W ) define a partition, which we call the block partition, of the set IrrW . 6 4. Blocks for G(m, d, n) 4.1. Let m ≥ 1, n > 1 be integers. Let Cm be the cyclic group of order m. We fix a generator g ∈ Cm and let sij ∈ Sn denote the transposition which swaps i and j. We denote by sj the simple transposition swapping j and j + 1. The group W = G(m, 1, n) is the semidirect product Sn ⋉ (Cm)n. We write gl i for the element (1, . . . , gl, . . . 1) ∈ G(m, 1, n) with gl in the ith place. Let V = kn be the reflection representation of G(m, 1, n). Recall that we assume that p ∤ W , so in particular p 6= 2. Let η ∈ k be a primitive mth root of unity. We fix a basis {y1, . . . , yn} of V so that gi(yj) = ηyj if i = j yj otherwise and σ(yj) = yσ(j), for all i, j and all σ ∈ Sn. Let {x1, . . . , xn} ∈ V ∗ be the dual basis. The conjugacy classes of reflections in W are given by {sijg−l i gl j : 0 ≤ l ≤ m − 1 and i 6= j}, and, for each 1 ≤ l ≤ m − 1, {gl j : 1 ≤ j ≤ n}. (6) (7) The parameter c is represented by (κ, c1, . . . , cm−1) ∈ kn, where c(sij g−l i gl j) = κ and c(gl j ) = cl. Using this notation, the definition of the rational Cherednik algebra becomes the following. Definition. Let W = G(m, 1, n). Then Hc(W ) is the quotient of T (V ⊕ V ∗) ⋊ W by the relations: [xi, xj] = 0, [yi, yj] = 0, sij g−l i gl j − cl(1 − η−l)gl i, m−1Xl=1 [yi, xi] = 1 − κ m−1Xl=0 Xj6=i m−1Xl=0 [yi, xj ] = κ η−lsij g−l i gl j. 4.2. The Dunkl-Opdam operators. For all 1 ≤ i ≤ n, we define elements in Hc(W ): zi = yixi − = xiyi + 1 2 1 2 + κ − κ m−1Xl=0 X1≤j<i m−1Xl=0 Xi<j m−1Xl=1 m−1Xl=1 sijgl ig−l j − clη−lgl i sij gl ig−l j − clgl i. (8) (9) By [Mar1, Lemma 3.2], [zi, zj] = 0 for all i, j. Let k[z1, . . . , zn] denote the algebra generated by the zis. By (4), this is a polynomial algebra. We denote by Er and Pr the rth elementary symmetric polynomial and rth power sum in the zi, respectively. By convention, E0 = P0 = 1. We will use the following result from [Mar1, 4.4]. 7 Theorem. Keep notation as above. Then [Er, x1] = X1<j2<···<jr ≤n x1zj2 . . . zjr . 4.3. Let E′ r = d dz1 (Er) and P ′ r = d dz1 (Pr) = rzr−1 1 . In this notation Theorem 4.2 reads Recall Newton's formula: Applying d dz1 to (11) we get [Er, x1] = x1E′ r. rEr = rXi=1 (−1)i−1PiEr−i. rE′ r = rXi=1 (−1)i−1(P ′ i Er−i + PiE′ r−i). A straightforward induction argument using the fact that E′ r = Er−1 − z1E′ r−1 shows that E′ r = r−1Xi=0 (−1)r−i−1zr−i−1 1 Ei. We will need a simple preparatory lemma. Let Qn := (z1 + 1)n − zn 1 . Lemma. We have Qn+1 = nXi=1 zn−i 1 Qi + (n + 1)zn 1 . Proof. The proof is by induction. The case n = 0 is clear. For n > 0, we have Qn+1 = z1Qn + Qn + zn 1 , (10) (11) (12) (13) and the induction step follows by a simple calculation which we leave to the reader. (cid:3) Proposition. For 1 ≤ r ≤ n, [Pr, x1] = x1Qr. Proof. The proof goes by induction on r. For r = 1 we have P1 = E1 and E′ [P1, x1] = x1. 1 = 1. By Theorem 4.2, Suppose r > 1. Then (−1)r−1[Pr, x1] = [rEr + (−1)iPiEr−i, x1] r−1Xi=1 r−1Xi=1 8 = rx1E′ r + (−1)i(Pix1E′ r−i + x1QiEr−i). (14) The first line follows from rewriting Pr using (11), and the second line follows from the induction hypothesis and (10). Now (14) equals r−1Xi=1 rx1E′ r + =x1" rXi=1 =x1" rXi=1 =x1" rXi=1 (−1)i(x1PiE′ r−i + x1Qi(E′ r−i + Er−i)) (−1)i−1(P ′ i Er−i + PiE′ r−i) + r−i + Qi(E′ r−i + Er−i))# (−1)i(PiE′ r−1Xi=1 r−i + Er−i)# r−i−1Xk=0 r−1Xi=1 r−1Xi=1 (−1)i−1P ′ i Er−i + (−1)i(Qi(E′ (−1)i−1izi−1 1 Er−i + (−1)i(Qi( (−1)r−i−k−1zr−i−k−1 1 Ek) + QiEr−i)# . (15) Here the first line follows from the induction hypothesis, the second from (12) and the fourth from (13). For a fixed 1 ≤ l ≤ r − 1, the coefficient of Er−l in (15) is equal to (−1)l−1lzl−1 1 + l−1Xi=1 (−1)l−1zl−i−1 1 Qi + (−1)lQl, which equals zero by the lemma above. By the lemma above, the coefficient of E0 is (−1)r−1rzr−1 + r−1Xi=1 (−1)r−1zr−i−1 1 Qi = (−1)r−1Qr. Therefore, [Pr, x1] = x1Qr. (cid:3) 4.4. We can now provide some central elements in Hc(W ). Theorem. For all 1 ≤ r ≤ n, Pr(zp 1 − z1, . . . , zp n − zn) = (zp i − zi)r ∈ Zc(W ). nXi=1 Thus, k[zp 1 − z1, . . . , zp n − zn]Sn ⊂ Zc(W ). Proof. We continue to write Pm for the power sum in the variables z1, . . . , zn. We have (zp i − zi)r = nXi=1 nXi=1 rXj=0(cid:18)r j(cid:19)(−1)r−jzpj 9 i zr−j i = rXj=0(cid:18)r j(cid:19)(−1)r−jPpj+r−j . Therefore by Proposition 4.3, (zp i − zi)r, x1] = x1 [ nXi=1 = x1 = x1 rXj=0(cid:18)r rXj=0(cid:18)r rXj=0(cid:18)r j(cid:19)(−1)r−jQpj+r−j j(cid:19)(−1)r−j((z1 + 1)pj+r−j − zpj+r−j j(cid:19)(−1)r−j((zp 1 + 1)j(z1 + 1)r−j − (zp ) 1 = x1 [(zp 1 + 1) − (z1 + 1)]r − x1(zp 1 − z1)r = 0. 1 )jzr−j 1 ) The theorem now follows as in [Mar1, Theorem 3.4]. The element Pr(zp n − zn) is symmetric in the zis so it commutes with any σ ∈ Sn, see [Gri, Lemma 5.1]. Therefore this power sum commutes with σx1σ−1 = σ(x1) for all σ, and in particular with each xi. To see that the power sum also commutes ′ (W ) from [Mar1, 3.3], where c′ is also defined. with the yis, we use the isomorphism ψ : Hc(W ) → Hc By definition ψ(xi) = yn−i+1 and ψ(yi) = xn−i+1, so that ψ(zi) = zn−i+1 for all i. Applying ψ−1 to [Pr(zp (cid:3) n − zn), xi] shows that [Pr(zp n − zn), yi] = 0 for all i. 1 − z1, . . . , zp 1 − z1, . . . , zp 1 − z1, . . . , zp 4.5. Blocks. We use an identical argument to the proof of [Mar1, Theorem 5.5], to determine the blocks of Hc(W ). We use freely the notation from [Mar1, §5]. Let us first change our parameters. We define h, H0, . . . , Hm−1 ∈ k via h = −κ and − cl(1 − η−l) = η−ljHj. (16) m−1Xj=0 We denote by P(m, n) the set of m-multipartitions of n, P(m, n) := {(λ0, . . . , λm−1) :Pm−1 i=0 λi = n}. The simple representations of kW are labeled by the set P(m, n); over the complex numbers this is standard and the same holds for kW by reduction, cf. 2.3. In fact the construction from [Pus] is also valid over k, since p does not divide W , so that we can find bases of irreducible kW -modules given by eigenvectors of Jucys-Murphy elements. Given a partition λ, the residue Res λ(x−(κp−κ)) is an element in Z[k], the group ring of the additive group (k, +). For λ = (λ0, . . . , λm−1) ∈ P(m, n) and a = (a1, . . . , am) ∈ km, let Res aλ :=Pm−1 i=0 x(ap i −ai)Res λi(x−(κp−κ)). Theorem. Let λ, µ ∈ P(m, n). Let a = (0, H1, H1 + H2, . . . , H1 + · · · + Hm−1). Then Lc(λ) and Lc(µ) belong to the same block of Hc(W ) if and only if Res aλ = Res aµ. Proof. Since this follows the proof of [Mar1, Theorem 5.5] closely, we shall sketch the argument, pointing out the necessary changes to loc. cit. Using Theorem 4.4 together with Weyl's Theorem for invariants of symmetric groups (which is valid whenever p ∤ W , see [Smi, Theorem 3.3.1]), it is enough to calculate the characters of Pr(zp n − zn) on each Lc(λ). The characters are determined by choosing a simultaneous eigenvector vλ ∈ Lc(λ) for z1, . . . , zn. The eigenvalues for zp i − zi are evaluated as in [Mar1, 5.4], and produce the desired combinatorial description. (cid:3) 1 − z1, . . . , zp 10 4.6. Blocks for G(m, d, n). The blocks of Hc(W ) for W = G(m, d, n), where d m and either n > 2 or n = 2 and d is odd, can be calculated from Theorem 4.5 by using Clifford theory as in [Bel1]. The resulting description of the blocks is then analogous to that given in characteristic zero, see [Mar1, 5.6]. 5. Smoothness of centres for G(m, 1, n) be the algebra of crystalline differential operators on Vreg. For y ∈ V , let Dy denote the Dunkl operator 5.1. Dunkl embedding. Let α = Qs∈S αs. Let Vreg denote the set {v ∈ V α(v) 6= 0}. Let D(Vreg) ∂y +Ps∈S (s − 1) ∈ D(Vreg) ⋊ W . By [EG, page 280] there is an injective algebra morphism c(s) 1−λs αs(y) αs Hc(W ) ֒→ D(Vreg) ⋊ W ; w 7→ w, x 7→ x, y 7→ Dy, for all w ∈ W , x ∈ V ∗ and y ∈ V . This embedding becomes an isomorphism after localizing Hc(W ) at the Ore set {αm}m≥0. 5.2. Completions. Let us recall the setup of [BE]. Let W ′ ⊂ W be the stabilizer of a point b ∈ V and let V = V /V W ′ . For any point b ∈ V we write k[[V ]]b for the completion of k[V ] at b, and we write dk[V ]b for the completion of k[V ] at the W -orbit of b in V . Note that we have k[[V ]]0 = dk[V ]0. For any finitely generated k[V ]-module M , let cMb = dk[V ]b ⊗k[V ] M. The completion bHc(W, V )b is the algebra generated by kc[V ]b, the Dunkl operators Dy for y ∈ V , and the group W . Let c′ denote the restriction of c to S ∩ W ′. The algebra bHc similarly. Let P = FunW ′ (W,bHc(W ′, V )0) be the set of W ′-equivariant maps from W to bHc(W ′, V )0. Let Z(W, W ′,bHc(W ′, V )0) be the ring of endomorphisms of the right bHc(W ′, V )0-module P . The following proposition is proved over C in [BE, Theorem 3.2], and has an identical proof over k. ′(W ′, V )0 is then defined Proposition. There is an isomorphism of algebras defined as follows: for f ∈ P , α ∈ V ∗, a ∈ V , u ∈ W , Θ :bHc(W, V )b → Z(W, W ′,bHc ′(W ′, V )0) (Θ(u)f )(w) = f (wu), (Θ(xα)f )(w) = (x(b) wα + α(w−1b))f (w), (Θ(ya)f )(w) = y(b) waf (w) + Xs∈S,s /∈W ′ cs αs(wa) 1 − λs x(b) αs + αs(b) (f (sw) − f (w)), where xα ∈ V ∗ ⊂ Hc(W, V ), x(b) α ∈ V ∗ ⊂ Hc ′ (W ′, V ), ya ∈ V ⊂ Hc(W, V ) and y(b) a ∈ V ⊂ Hc ′ (W ′, V ). ′(W ′, V )0) be defined by (e11 · f )(w) = f (w) if w ∈ W ′ and (e11 · f )(w) = 0 oth- ′ (W ′, V )0)e11. The bimodule Let e11 ∈ Z(W, W ′,bHc erwise. Then e11 is a primitive idempotent and bHc e11Z(W, W ′,bHc ′ (W ′, V )0) yields a Morita equivalence between bHc ′ (W ′, V )0 ∼= e11Z(W, W ′,bHc ′ (W ′, V )0 and Z(W, W ′,bHc ′(W ′, V )0). 11 5.3. Suppose that M is a finite dimensional Hc(W, V )-module and let Supp M denote the support of M as a k[V ]W -module. If Supp M is the W -orbit of some point b ∈ V , then M is naturally a bHc(W, V )b-module. In particular, the support of any simple Hc(W, V )-module is a single W -orbit and so is abHc(W, V )b-module. Let M be a bHc(W, V )b-module, then denote by Θ∗M the Z(W, W ′,bHc the bHc(W, V )b-action on M via Θ. Definition. An Hc(W, V )-module M is called small if dim M < W pn. ′ (W ′, V )0)-module given by transporting Proposition. A Hc(W, V )-module M is small if and only if e11Θ∗(M ) is a small Hc ′ (W ′, V )-module. Proof. Note first that e11Θ∗(M ) is a Hc 1, . . . , gt be right coset representatives of W ′ in W . Then Θ∗(M ) =Lt ′ (W ′, V )-module by restricting the bHc i=1 gie11g−1 j Θ∗M for all i, j. In particular, dim M < W pn if and only if dim e11Θ∗(M ) < W pn/t = (cid:3) ′ (W ′, V )0-action. Let g1 = i Θ∗M , with dimk gie11g−1 i Θ∗M = dimk gje11g−1 W ′pn. Let Zc(W, V ) denote the centre of Hc(W, V ). Corollary. Keep notation as above. Then Zc(W, V ) is smooth if and only if Hc M with Supp M = 0 for all parabolic subgroups W ′ ⊆ W . ′ (W ′, V ) has no small modules Proof. By Proposition 3.2 (3), Zc(W, V ) is smooth if and only if Hc(W, V ) has no small modules. Let M be a small, simple Hc(W, V )-module and choose b ∈ V such that the support of M equals the W -orbit of b. Then ′(W ′, V )-module, where W ′ is the stabilizer of b in W . Moreover, the support of e11Θ∗(M ) is 0. Conversely, if there exists a ′ (W ′, V )0 small Hc M extends to a bHc(W, V )b-module and Proposition 5.3 says that e11Θ∗(M ) is a small Hc is Morita equivalent to Z(W, W ′,bHc ′ (W ′, V )0) ∼= bHc(W, V )b, there exists some bHc(W, V )b-module M such ′ (W ′, V )-module M ′ supported on 0 then this extends to a bHc that M ′ ≃ e11Θ∗(M ). Proposition 5.3 says that M is actually a small Hc(W, V )-module. ′ (W ′, V )0-module. Since bHc (cid:3) 5.4. Let M be a simple Hc(W, V )-module with Supp M = 0. We can define an induced module (cf. 3.3) as follows. Let B = k[V ] ⋊ W and let E ∈ IrrW . Extend the W -action on E to a B-action by letting f ∈ k[V ] act by f (0). Define ∆(E) = Hc(W, V ) ⊗B E. By the support condition on M , the subspace M0 = {m ∈ M V ∗ · m = 0} of M is non-zero. It is a W - submodule of M and we may assume without loss of generality that E ⊆ M0. Therefore there is a surjective homomorphism ∆(E) ։ M , which maps 1 ⊗ E ⊂ ∆(E) to E ⊆ M0 in the obvious way. Recall that Hc(W, V ) has a Z-grading. We make E into a graded B-module by setting E0 = E and Ei = 0 for all i 6= 0. We give ∆(E) a Z-graded module structure by inducing the graded structure on E. Since M is simple, there is some a ∈ (V ∗)(1)/W such that ma · M = 0, where ma denotes the maximal ideal corresponding to a. Define ∆(a, E) = (Hc(W, V ) ⊗B E)/ma · (Hc(W, V ) ⊗B E). By our choice of a, there is also a surjective homomorphism ∆(a, E) ։ M . The grading on ∆(E) induces filtrations on ∆(a, E) and M . We use the notation grZ to denote the associated graded objects with respect to these filtrations. Recall that a morphism f : M → N between filtered Hc(W, V )-modules is called a strictly filtered mor- phism if f (FiM ) = FiN ∩ f (M ) for all i ∈ Z. The functor grZ is exact on short exact sequences of strictly 12 filtered morphisms. Note also that the surjection ∆(a, E) ։ M is strictly filtered by definition. The filtration on ∆(E) is both exhaustive and separating, therefore the same is true of ∆(a, E). However, this module is finite dimensional therefore we have Fi∆(a, E) = 0 and F−i∆(a, E) = ∆(a, E) for i ≫ 0. Proposition. The Hc(W, V )-module grZM has dimension dimk M and is annihilated by (k[V (1)]W ⊗k[(V ∗)(1)]W )+. Proof. Denote by Fi the ith piece of the filtration on M . By construction, F• is a decreasing filtration with F1 = 0 and F−i = M for all i ≫ 0. Thus dimk grZM =P∞ For the second statement, we show that grZ∆(a, E) ∼= ∆(0, E). The claim then follows since there is a i=0 dimk F−i/F−i+1 = dimk M . surjective map ∆(0, E) → grZM . Consider the short exact sequence: 0 → ma · (Hc(W, V ) ⊗B E) → ∆(E) → ∆(a, E) → 0, where the left-hand term is given the induced filtration. By the PBW theorem and the Nullstellensatz, grZma · (Hc(W, V ) ⊗B E) = k[(V ∗)(1)]W + · (Hc(W, V ) ⊗B E). Since ∆(E) is graded, grZ∆(E) = ∆(E). Thus grZ∆(a, E) ∼= ∆(0, E). (cid:3) Corollary. The following are equivalent: (1) Zc(W, V ) is smooth; (2) Hc (3) Hc ′ (W ′, V ) has no small modules M with Supp M = 0 for all parabolic subgroups W ′ ⊆ W ; ′ (W ′, V ) has no small modules M for all parabolic subgroups W ′ ⊆ W ; (4) The blocks of Hc ′ (W ′, V ) are singletons for all parabolic subgroups W ′ ⊆ W . Proof. There are isomorphisms of algebras Hc D(V W ′ alent to ), where D(V W ′ ) := D(V W ′ )/h(k[(V W ′ ′ (W ′, V ) ≃ Hc )(1)] ⊗ k[((V W ′ ′ (W ′, V )⊗D(V W ′ ′ (W ′, V )⊗ )∗)(1)])+i. In particular, (2) and (3) are equiv- ′ (W ′, V ) ≃ Hc ) and Hc (2') Hc (3') Hc ′ (W ′, V ) has no small modules M with Supp M = 0 for all parabolic subgroups W ′ ⊆ W ; ′ (W ′, V ) has no small modules M for all parabolic subgroups W ′ ⊆ W , respectively, where we have used the fact that D(V W ′ ) is an Azumaya algebra. The equivalence of (1) and ′ (W ′, V ) has a small module (2') is Corollary 5.3. Clearly (2') implies (3'). For the converse, suppose that Hc M with Supp M = 0 for some parabolic subgroup W ′ ⊆ W . Then by the proposition, grZM is a small module for Hc ′ (W ′, V ). The equivalence of (3) and (4) follows from an identical argument to [Gor1]. (cid:3) Remark. The proofs above are valid, mutando mutandae, for t = 0 and char k = 0. 5.5. Smoothness of centres of Hc(G(m, 1, n)). Let m, n be positive integers and assume that n > 1. Let W = G(m, 1, n). In this section we give a proof of Corollary 1.3. The idea is to use Corollary 5.4 (4) together with the results of section 4 to determine for which c the centre Zc(W ) is smooth. Recall that the parabolic subgroups of W are of the form Sk1 × · · · × Skt × G(m, 1, n′), k1 + · · · + kt + n′ ≤ n. Here Ski denotes the symmetric group on ki letters, and S0 = G(m, 1, 0) = {id} by definition. For such a parabolic subgroup, the representation V is the reflection representation. Recall that c = (κ, c1, . . . , cm−1). 13 Recall the parameters a1, . . . , am from Theorem 4.5. Let Cm,n denote the set of all c such that either κ ∈ Fp, or ai − aj ± Cκ ∈ Fp, for some 0 ≤ i 6= j ≤ m − 1 and integer C such that 0 ≤ C ≤ n − 1. Thus Cm,n is a finite union of hyperplanes in km. Theorem. The algebra Zc(W ) is smooth if and only if c /∈ Cm,n. Proof. Let us first suppose that c ∈ Cm,n. Recall that, for λ ∈ P(m, n), Res aλ :=Pm−1 Suppose that κ ∈ Fp. Then κp − κ = 0 and so Res a((n), ∅, . . . , ∅) = Res a((n)t, ∅, . . . , ∅). Thus by Theo- rem 4.5, Hc(W, V ) has a non-singleton block, and so Zc(W ) is singular, Corollary 5.4. If m > 1 and ai − aj − Cκ ∈ Fp for some 0 ≤ C ≤ n − 1, then i=0 x(ap i −ai)Res λi (x−(κp−κ)). Without loss of generality i = 1 and j = 2, and then ap i − ai = ap j − aj + C(κp − κ). Res a((n), ∅, ∅, . . . , ∅) = Res a(∅, (n − C, 1C), ∅, . . . , ∅) = xap 1 −a1 x−i(κp−κ). n−1Xi=0 Thus Hc(W, V ) has a non-singleton block and Zc(W ) is singular. A similar argument applies in the case −n + 1 ≤ C ≤ 0. Suppose now that c /∈ Cm,n. We first show that Hc(W, V ) has only singleton blocks. For a contradiction, suppose that there exist distinct λ, µ ∈ P(m, n) such that Res aλ = Res aµ. Since κ /∈ Fp, x−(κp−κ) 6= 1. Each box b in the Young diagram of λi contributes x(ap i −ai)x−cont(b)(κp−κ) to Res aλ. Since λ 6= µ, there exist 1 ≤ i < j ≤ N and boxes b ∈ λi, b′ ∈ µj such that x(ap j −aj )x−cont(b′)(κp−κ). Now cont(b) and cont(b′) are integers such that −λi + 1 ≤ cont(b) ≤ λi − 1 and −µj + 1 ≤ cont(b′) ≤ µj − 1. Thus ap j − aj + C(κp − κ) for some −λi − µi + 1 ≤ C ≤ λi + µi − 1. This means that −ai + aj − Cκ ∈ Fp, and so c ∈ Cm,n, a contradiction. i −ai)x−cont(b)(κp−κ) = x(ap i − ai = ap We now want to prove the stronger statement that c /∈ Cm,n implies that Zc(W ) is smooth. Note that Cm′,n′ ⊂ Cm,n for all m′ = 1 or m, n′ < n. Therefore c /∈ Cm,n implies that c′ /∈ Cm′,n′ for all m′ = 1 or m, n′ < n. By the description of parabolic subgroups W ′ ⊂ W given above, the previous paragraph implies Hc(W ′, V ) has only singleton blocks for all W ′. By Corollary 5.4 this implies that Zc(W ) is smooth. (cid:3) Remark. (1) In the case m = 1, W is just the symmetric group Sn. Although V = kn is not the reflection representation, we have Hκ(Sn, kn) ∼= Hκ(Sn, kn−1) ⊗ D(A1), where kn−1 now denotes the reflection representation of Sn. The set C1,n is then identified with Fp ⊂ k. (2) Note that the proof of the above theorem shows that for W = G(m, 1, n), the centre Zc(W ) is smooth if and only if Hc(W, V ) has only singleton blocks. 6. Degenerations In this section we describe, based on Cartier's Theorem, the category of D(V )⋊W -module with p-curvature zero. This will allow us to prove Theorem 1.4. 14 6.1. p-curvature. Fix a basis x1, . . . , xn of V ∗ and ∂1, . . . , ∂n of V such that ∂i(xj ) = δi,j. Let A = k[∂1, . . . , ∂n, xp 1, . . . , xp n] and Spec A = T ∗,(1)V. The centre of D(V ) embeds in A and we write π : T ∗,(1)V → (T ∗V )(1) for the corresponding finite morphism. The group W acts on T ∗,(1)V and (T ∗V )(1) such that the map π is W -equivariant and satisfies Wζ := StabW (ζ) = StabW (π(ζ)), ∀ ζ ∈ T ∗,(1)V. For fixed ζ ∈ T ∗,(1)V and λ ∈ Irrk(Wζ ), we define Vζ (λ) := Ind ζ. D(V )⋊W A⋊Wζ λ, where A acts on λ via the character Lemma. Fix ζ ∈ T ∗,(1)V and λ ∈ Irrk(Wζ ). Then (1) the D(V ) ⋊ W -module Vζ (λ) is simple; (2) Vζ1 (λ1) ≃ Vζ2 (λ2) if and only if ζ2 ∈ W · ζ1 and, moreover, if wζ1 = ζ2 then λ1 ≃ λ2 via the conjugation isomorphism w : Wζ1 ∼−→ Wζ2 ; (3) Every simple D(V ) ⋊ W -module is isomorphic to Vζ (λ) for some ζ and λ. Proof. Considered as an A-module, Vζ (λ) =Lη∈W ·ζ(Vζ (λ))η where (Vζ (λ))η is set-theoretically supported at η. Each (Vζ (λ))wζ is a D(V )⋊(wWζ w−1)-submodule of Vζ (λ) and Vζ (λ) will be a simple D(V )⋊W -module λ is a simple D(V ) ⋊ Wζ -module. If ζ = (a, α) with a ∈ V (1) then we if and only if (Vζ (λ))ζ = Ind write (b, α) with b ∈ V for the unique closed point in π−1(ζ). Applying the Wζ -equivariant automorphism xi 7→ xi − hxi, bi and ∂j 7→ ∂j − h∂j, αi to D(V ), we may assume without loss of generality that ζ = 0 and Wζ = W . Let δ0 = Ind k be the unique simple D(V )-module supported at 0 ∈ (T ∗V )(1); simplicity of δ0 follows from the fact that D(V ) is Azumaya of rank p2n and the dimension of δ0 is pn. Then V0(λ) = δ0 ⊗ λ, with W acting diagonally. The module V0(λ) is simple: let v1, . . . , vpn be a basis of δ0 such that v1 = 1 and D(V )⋊Wζ A⋊Wζ choose any 0 6= l =Pi vi ⊗ li ∈ V0(λ). Choose an i such that li 6= 0. Since D(V ) surjects onto Endk(δ0), there is some D ∈ D(V ) such that D · vj = 0 for all j 6= i and D · vi = v1. Then D · l = 1 ⊗ li and we have W · (D · l) = 1 ⊗ λ. Hence D(V ) ⋊ W · l = V0(λ). To show that the various V0(µ) are non-isomorphic, note that D(V ) A dimk HomD(V )⋊W (V0(λ), V0(µ)) = dimk HomA⋊W (λ, V0(µ)) = δλ,µ because the space {v ∈ δ0 ∂i · v = 0 ∀ i} is one-dimensional. Arguing geometrically as above shows the second claim of the statement. The final statement is clear just by considering the socle, as a A-module, of an simple D(V ) ⋊ W -module. (cid:3) 6.2. Now we require a special case, Proposition 6.2, of a classical result by Cartier on D-module with zero p-curvature. We follow the presentation given in [Kat, §5]. Recall that if D ∈ Der(V ) ⊂ D(V ) is a derivation, then Dp also acts as a derivation and we write D[p] for this derivation so that Dp − D[p] acts trivially on k[V ]. Definition. Let M be a finitely generated D(V ) ⋊ W -module. The p-curvature of M is the map ψ : Der(V ) → Endk(M ) given by ψ(D) = ρ(D)p − ρ(D[p]), where ρ : D(V ) ⋊ W → Endk(M ) is the action map. We say that M has zero p-curvature if ψ = 0. Denote by D0 the full subcategory of D(V ) ⋊ W -mod consisting of modules with zero p-curvature. 15 Proposition. Let M ∈ D(V ) ⋊ W -mod. Let V (1) ⊂ (T ∗V )(1) be the zero section (defined by ∂p ∂p n = 0). 1 = · · · = (1) The module M has zero p-curvature if and only if M is scheme-theoretically supported on V (1) when considered as a Z(D(V ))-module i.e. ∂p i · M = 0 for all i. (2) The "horizontal sections" functor DR : D0 → k[V (1)] ⋊ W -mod, DR(M ) := M ∇ = {m ∈ M ∂i · m = 0 ∀ i}, is an equivalence of categories with quasi-inverse DR⊥ : k[V (1)] ⋊ W -mod → D0, DR⊥(N ) = Ind D(V )⋊W A⋊W N, where A acts on N via the morphism A → k[V (1)], ∂i 7→ 0. (3) The equivalence DR restricts to an equivalence of graded categories grD0 → k[V (1)] ⋊ W -grmod. The following operators where introduced in [Kat, (5.1.2)]. Their properties can be verified by direct calculation. Lemma. Let D(x) be the first Weyl algebra and M a D(x)-module with zero p-curvature. (1) Define P =Pp−1 M ∇, P M ∇ = id and P 2 = P . (2) Define the map T : M → M by i=0 (−x)i i! ∂i ∈ D(x). Then P defines a k[xp]-linear operator on M such that P (M ) ⊂ Then T = idM . m 7→ xi i! p−1Xi=0 P (∂i · m). Proof of Proposition 6.2. f p 1 ψ(D1) + f p necessary that ρ(∂i)p = 0, i.e. ∂p (1) It is shown in [Kat, Proposition 5.2] that ψ is p-linear i.e. ψ(f1D1+f2D2) = 2 ψ(D2) for all fi ∈ k[V ] and Di ∈ Der(V ). Since ψ(∂i) = ρ(∂i)p for all i it is clearly i · M = 0, in order for M to have zero p-curvature. On the other hand, every element D ∈ Der(V ) can be expressed as D =Pn ψ(D) = i=1 fi∂i for some fi ∈ k[V ]; thus and ψ(D) · M = 0 if ∂p i · M = 0 for all i. (2) As in Lemma 6.2, define Pi =Pp−1 i=1 Pi. Let M ∈ D0. Then Lemma 6.2 implies P defines a k[V (1)]-linear operator on M such that DR(M ) = P (M ), PM ∇ = id and P 2 = P . Note that the subspace M ∇ of M is a W -submodule of M . Therefore we define j! ∂j (−xi)j j=0 i f p i ∂p nXi=1 i and set P =Qn W Xw∈W w · P, 1 P = so that P is a W -equivariant projection onto M ∇. Let 0 → M1 → M2 φ → M3 → 0 be a short exact sequence in D0. Since P ∈ D(V ), we have P ◦ φ = φ ◦ P . Therefore DR(φ) : DR(M2) → DR(M3) is surjective and the left exact functor DR is actually exact. Similarly, since 16 D(V ) ⋊ W is flat over A ⋊ W , DR⊥ is also an exact functor. D(V ) ⋊ W generated by ∂1, . . . , ∂n, then DR⊥(k[V (1)] ⋊ W ) = D(V ) ⋊ W/(∂1, . . . , ∂n) and If (∂1, . . . , ∂n) is the left ideal of DR(cid:18) D(V ) ⋊ W (∂1, . . . , ∂n)(cid:19) = A ⋊ W + (∂1, . . . , ∂n) (∂1, . . . , ∂n) ≃ k[V (1)] ⋊ W as a k[V (1)] ⋊ W -module. Since DR ◦ DR⊥(k[V (1)] ⋊ W ) = k[V (1)] ⋊ W and the functor DR ◦ DR⊥ is exact, for each N ∈ k[V (1)] ⋊ W -mod we get the standard diagram k[V (1)] ⋊ W k k[V (1)] ⋊ W l N DR ◦ DR⊥(k[V (1)] ⋊ W )k / DR ◦ DR⊥(k[V (1)] ⋊ W )l / DR ◦ DR⊥(N ) 0 / 0 where the first two vertical morphisms are isomorphisms. This implies that the third vertical mor- phism is also an isomorphism. Hence the natural transformation 1 → DR ◦ DR⊥, coming from the fact that DR⊥ is left adjoint to DR, is an isomorphism. Now take M ∈ D0 and consider the natural morphism DR⊥ ◦ DR(M ) → M . Since DR is conservative, the fact that DR ◦ DR⊥ = 1 implies i=1 Ti, where Ti : M → M , that this morphism is injective. On the other hand, if we define T =Qn Ti(m) = Pi(∂j i · m), xj i j! p−1Xj=0 then Lemma 6.2 implies that T = idM . This proves that DR⊥ ◦ DR(M ) → M is surjective. (3) It is straight-forward to see that DR and DR⊥ send graded modules to graded modules. (cid:3) 6.3. Let 1 denote the trivial W -module. The following observation will be required later. Lemma. Let N be a k[V (1)] ⋊ W -module such that DR⊥(N ) is isomorphic to pn copies of the regular representation as a W -module. Then N is isomorphic to the regular representation as a W -module. Proof. As a W -module, DR⊥(N ) ≃ V0(1) ⊗ N . Therefore it suffices to show that the Brauer character χ of V0(1) satisfies χ(w) 6= 0 for all w in W . Recall that V0(1) ≃ k[V ∗]/hk[(V ∗)(1)]+i. Let λ1, . . . , λn be the eigenvalues of w on V . Since we are calculating the Brauer character of V0(1), we assume that λi ∈ C for all i. Then χ(g) = Tr(g, V0(1)) = 1 − λp i 1 − λi . nYi=1 Since χ(g) ∈ C and p does not divide W , the product on the right hand side is non-zero. (cid:3) 6.4. Lattices. In this section let k be an arbitrary algebraically closed field. We let H denoted a Z- graded k[x]-algebra such that H is a finite, free k[x]-module, where k[x] is graded with deg(x) = 0. Write ℓ = Spec k[x] and K = k(x). For α ∈ k, we denote by Hα the specialization H ⊗k[x] kα of H at α. We assume that there exists a finite set I and H-modules {∆(λ) λ ∈ I} such that (1) The module ∆(λ) is graded and free as a k[x]-module. (2) For all α ∈ ℓ, there is a bijection IrrHα ≃ I such that the simple, graded Hα-module Lα(λ) is the unique simple quotient of ∆α(λ), the specialization of ∆(λ) at α. 17 / / / / / / / / / (3) There is a bijection I ≃ IrrHK such that the simple, graded HK-module LK(λ) is the unique simple quotient of ∆K(λ). Let M be a H-module. We say that M is a H-lattice if it is free, of finite rank, as a k[x]-module. Note that if M is a H-lattice and N a H-submodule, then N is a H-lattice because k[x] is a principal ideal domain. Lemma. There exists a composition series 0 = ∆0(λ) ⊂ · · · ⊂ ∆r(λ) = ∆(λ) of graded H-lattices such that if Li(λ) = ∆i(λ)/∆i−1(λ), then Li(λ) is a graded H-lattice and Li(λ)K ≃ LK(λi) for some λi ∈ I. Proof. Fix a graded composition series 0 = ∆0 K (λ) ⊂ · · · ⊂ ∆r K(λ) = ∆K(λ) such that ∆i an inclusion. We set K(λ)/∆i−1 K (λ) ≃ LK(λi) for some λi ∈ I. Write φ : ∆(λ) → ∆K (λ) for the natural map. It is ∆i(λ) = φ−1(∆i K(λ)) = ∆i K(λ) ∩ ∆(λ), ∀ i. K(λ)/∆i−1 Then ∆i(λ) is a H-submodule of the H-lattice ∆(λ), hence it is a H-lattice. We have a H-morphism φi : Li(λ) := ∆i(λ)/∆i−1(λ) → ∆i K (λ) = LK(λi). We claim that Li(λ) is a H-lattice. It suffice to show that it is torsion-free with respect to k[x]. Let ¯a ∈ Li(λ) and 0 6= f (x) ∈ k[x] such that f (x)·¯a = 0. Then f · a ∈ ∆i−1(λ) = ∆i−1 K (λ) ∩ ∆(λ) = ∆i−1(λ). Hence ¯a = 0. Since Li(λ) is a H-lattice and LK(λi) is a simple HK-module, φi will induce an isomorphism Li(λ)K ≃ LK(λi) provided Li(λ) 6= 0. Let 0 6= b ∈ ∆i K (λ), then there exists some 0 6= g(x) ∈ k[x] such that K(λ) ∩ ∆(λ) and g(x)b /∈ ∆i−1 g(x)b ∈ ∆i (cid:3) K(λ) − ∆i−1 K (λ). Therefore g(x)b ∈ ∆i(λ) − ∆i−1(λ). K (λ) ∩ ∆(λ), which implies that a ∈ ∆i−1 6.5. By Lemma 6.4, for each λ ∈ I we have a graded H-lattice L(λ) := Lr(λ) such that ∆(λ) ։ L(λ). The specialization of L(λ) to x = α ∈ k is denoted L(λ)x=α to distinguish it from the simple module Lα(λ). Lemma. There exists a prime ideal p ⊳ H such that p · L(λ) = 0. Proof. Let Ann1 = {I ⊳ H GKdim (∆(λ)/I∆(λ)) = 1}. Since GKdim (∆(λ)) = 1, the set Ann1 is non- empty. Since H is Noetherian, we can choose p ∈ Ann1 to be maximal with respect to inclusion. The claim is that p is prime. Assume otherwise, so that there exist ideals I, J with IJ ⊂ p but I, J /∈ Ann1. Note that ∆/p · ∆ has GK-dimension one and is a finitely generated H/p-module. Therefore H/p has GK-dimension one too. Since ∆/J∆ has GK-dimension zero, the short exact sequence 0 → J∆/p∆ → ∆/p∆ → ∆/J∆ → 0 implies, by [MR, Proposition 8.3.11], that GKdim(J∆/p∆) = 1. However, J∆/p∆ is clearly a torsion H/p-module. Therefore [MR, Corollary 8.3.6] implies that the GK-dimension of J∆/p∆ is zero. This contradiction implies that p is prime. Let M = ∆/p · ∆. The H-submodule of M consisting of elements that are torsion with respect to k[x] is a proper submodule of GK-dimension zero. Therefore quotienting out by this submodule we may assume that M is torsion free and hence free, and that p · M = 0. Moreover, MK is a non-zero quotient of ∆K(λ) which implies that MK ։ LK(λ). This implies that p · LK(λ). Since L(λ) is a H-submodule of LK(λ), we have p · L(λ) = 0 as required. (cid:3) Proposition. Let mλ = maxα∈ℓ(dimk Lα(λ)). (1) There exists a finite set ℓ0 ⊂ ℓ such that dimk Lα(λ) = mλ ⇔ α /∈ ℓ0. 18 (2) The rank of L(λ) equals mλ. (3) The specialization of L(λ) at α is isomorphic to the simple Hα-module Lα(λ) for all α ∈ ℓ − ℓ0. Proof. Let p be a prime ideal in H such that p · L(λ) = 0. Its existence is guaranteed by Lemma 6.5. Set R := H/p so that L(λ) is an R-module. Since R is a prime ring, the image of k[x] in R is a domain. Therefore the fact that R is a finite k[x]-module of GK-dimension one implies that R is a free k[x]-module. The map π : Spec Z(R) → Spec k[x] is finite. Let A denote the Azumaya locus of R. Since A is open and dense in the irreducible variety Spec Z(R) and the map π is finite, the set π(Spec Z(R) − A) is a proper closed subset of ℓ. Therefore the set ℓ1 := ℓ − π(Spec Z(R) − A) = {α ∈ ℓ π−1(α) ⊂ A} is open and dense in ℓ. Since the specialization L(λ)x=α is a quotient of ∆(λ)x=α = ∆α(λ), the simple module Lα(λ) is a quotient of L(λ)x=α. This implies that every Lα(λ) is an R-module and hence mλ is bounded above by the P.I. degree of R. By definition, this bounded is achieved for all α ∈ ℓ1. Fix some α ∈ ℓ1. Recall that LK(λ) = K · L(λ). Hence dimK LK(λ) = rank k[x]L(λ) ≥ dimk Lα(λ). On the other hand, since Lα(λ) is supported on the Azumaya locus of R and LK(λ) is a simple module for the central localization RK of R, [MR, Posner's Theorem 13.6.5] together with [MR, Kaplansky's Theorem 13.3.8] imply that dimk(Lα(λ)) = P.I. − degree(R) ≥ dimK dimK LK(λ) = rank k[x]L(λ). Therefore we get the required equality. (cid:3) 6.6. We return to the case where k has positive characteristic. Recall that C is the space of parameters for H. We fix a parameter c. Let c be a variable and denote by Hc(W ) the rational Cherednik algebra over k[c] such that the specialization of Hc(W ) at c = 1 recovers Hc(W ). Let K = k(c) be the field of fractions of k[c]. Given λ ∈ Irrk(W ), let ¯∆c(λ) = Hc(W ) ⊗k[V ]⋊W λ be the corresponding baby Verma module. It is a free k[c]-module of finite rank. Let ¯∆K(λ) = HK(W ) ⊗K[V ]⋊W K ⊗k λ be the corresponding baby Verma module for HK(W ). Lemma. (1) Every simple module λ ∈ IrrK(W ) is absolutely irreducible. (2) The module ¯∆K(λ) has a unique simple quotient LK(λ). (3) We have a natural identification of HK(W )-modules K ⊗k[c] ¯∆c(λ) = ¯∆K(λ). Proof. (1) The k-algebra kW is already split semi-simple since k is assumed to be algebraically closed and the characteristic of k does not divide the order of W . Therefore the K-algebra KW = K ⊗kkW is also split semi-simple. (2) Let µ ∈ IrrK(W ), considered as a K[V ] ⋊ W -module such that K[V ]+ acts as zero. Adjunction implies that HomK[V ]⋊W ( ¯∆K (λ), µ) = HomW (Kλ, µ) = Kδλ,µ since Kλ and µ are absolutely irreducible. Therefore, if φ : ¯∆K(λ) → S is any simple quotient of ¯∆K(λ), we must have a K[V ] ⋊ W -surjection S → Kλ. Hence S = HK(W ) · Kλ is the unique graded quotient of ¯∆K(λ). 19 (3) The space K ⊗k λ ⊂ K ⊗k[c] a surjective map ¯∆K(λ) → K ⊗k[c] both modules have the same dimension over K, this surjection must be an isomorphism. ¯∆c(λ) (since K ⊗k λ is a generating set for K ⊗k[c] ¯∆c(λ) is a K[V ] ⋊ W such that K[V ]+ acts as zero. Therefore we have ¯∆c(λ)). Since 6.7. A classification. Let ev : K(W )[t, t−1] → Z[t, t−1] be the map sending [λ] to dim λ. Let 1 − t(cid:19)n I(t) := ev(cht,W (V0(1))) =(cid:18) 1 − tp . Lemma. If c ∈ C such that dimk Lc(λ) = W pn, then the Poincar´e polynomial of Lc(λ) is given by P (Lc(λ), t) = dim(λ) · tpbλ∗ P (k[V ∗]coW , tp) · I(t) fλ∗ (tp) . (cid:3) (17) (18) Proof. Let L(λ) be the Hc(W )-module described in 6.5. Since it is free as a k[c]-module, the graded W - module structure of all specializations of L(λ) are the same. Therefore it suffices to prove that the Poincar´e c(λ) of ¯∆c(λ) as in polynomial of the specialization L(λ)c=0 has the desired form. We fix a filtration ¯∆i Lemma 6.4. Specializing to c = 0 gives a filtration ¯∆i (λ) ≃ L(λ)c=0. All the D(V ) ⋊ W -modules ¯∆i 0(λ) = Ind 0(λ) have p-curvature zero. Therefore Proposition 6.2 says that ¯∆i 0(λ)(1) for some graded k[V (1)] ⋊ W -module ¯∆i 0(λ) of ¯∆0(λ) such that ¯∆r 0(λ)(1) and we have D(V )⋊W A⋊W 0(λ)/ ¯∆r−1 0 ¯∆i ¯∆r 0(λ)(1)/ ¯∆r−1 0 (λ)(1) := L(λ)(1) c=0, with L(λ)c=0 = Ind D(V )⋊W A⋊W L(λ)(1) c=0. This implies that Lc(λ) ≃ V0(1) ⊗ L0(λ)(1) as graded W -modules. Let Hc(W )-latt denote the category of Hc(W )-lattices i.e. the category of finitely generated, graded Hc(W )-modules that are free over k[c]. It is an exact, extension closed subcategory of Hc(W )-grmod. Therefore it makes sense to consider the Grothendieck group K0(Hc(W )-latt) of Hc(W )-latt. By our assumption on dimk Lc(λ), the support of Lc(λ) as a Zc(W )- module is contained in the Azumaya locus of Hc(W ). Therefore Muller's Theorem, [BG, Proposition 2.7], implies that each graded composition factor of the indecomposable module ¯∆c(λ) is of the form Lc(λ)[mi] for some integer mi. Hence Lemma 6.4 says that each Li c(λ) specializes at c = 1 to some Lc(λ)[mi]. Thus, in K0(Hc(W )-latt) we have an equality [ ¯∆c(λ)] = h(t)[Lc(λ)] for some Laurent polynomial h(t). This implies that [ ¯∆0(λ)(1)] = h(t)[L(λ)(1) c=0] in the Grothendieck group of graded k[V (1)] ⋊ W -modules. By Lemma 6.3, the fact that L(λ)c=0 is isomorphic to pn copies of the regular representation implies that L(λ)(1) c=0 is isomorphic to a graded copy of the regular representation. On the other hand, Lemma 2.4 and Proposition 6.2 imply that DR⊥( ¯∆0(λ)) ≃ k[V (1)]coW ⊗ λ, hence [k[V (1)]coW ⊗ λ] = h(t)[L(λ)(1) c=0]. The proof of [Bel2, Lemma 3.3] implies that h(t) = t−pbλ∗ fλ∗ (tp). From this one can deduce formula (18). (cid:3) Now we are finally in a position to prove Theorem 1.4. Proof of Theorem 1.4. Assume that W and c are chosen so that the centre of Hc(W ) is smooth. Then every simple Hc(W )-module has dimension W pn. This implies that the graded character of Lc(λ) is given by the formula of Lemma 6.7. Since Lc(λ) is finite dimensional, the rational function on the right hand side of (18) must be a Laurent polynomial. This means that fλ∗ (tp) divides P (k[V ∗]coW , tp)I(t) in C[t, t−1]. The formula (17) for I(t) shows that every root of I is a primitive pth root of unity. Therefore, if fλ∗ (tp) and I(t) have a non-trivial common factor, we must have fλ∗ (ζp) = 0 for some primitive pth root of unity ζ. But 20 fλ∗ (1) = dim λ∗ 6= 0. Hence fλ∗ (tp) must divide P (k[V ∗]coW , tp) in C[tp, t−p]. It was shown in [Bel2] that if W is not isomorphic to G(m, 1, n) or G4 then one can always find some λ for which fλ∗ (tp) does not divide P (k[V ∗]coW , tp). Since c has played no part in this argument, we conclude that the centre of Hc(W ) is never smooth in these cases. Conversely, it follows from Corollary 1.3 that the centre of Hc(G(m, 1, n)) is regular for generic c. Now we consider the group G4; we have p 6= 2, 3 in this case. By Corollary 5.4, it suffices to show that the blocks of ′ (W ′) are singletons for generic parameters c and all parabolic subgroups W ′ of G4. We begin by showing Hc that this is true for W ′ = G4. Recall from 3.1 the central element hp − h of Hc(G4). For the remainder of the proof we follow the notation of [Bel2, §4]. The function c is defined by c(si) = c1 and c(ti) = c2 for some c1, c2 ∈ k. To show that the block partition of IrrG4 is trivial, it suffices to show that the scalars by which the element hp − h acts the L(λ)'s are pairwise distinct. Let z1 = s1 + · · · + s4 and z2 = t1 + · · · + t4 ∈ Z(G4). The scalar µi by which the central element zi acts on the simple G4-module µ is given by µ T z1 4 4ω2 4ω z2 4 4ω V1 4ω2 V2 W −2 −2 h −2ω2 −2ω h∗ −2ω −2ω2 U 0 0 We have 1 − λsi = 1 − ω2 and 1 − λti = 1 − ω. Define di = −ci 1−λi dp 1µp 1 − d1µ1 + dp 2 − d2µ2, hence it will act by the same scalar on L(µ) and L(ρ) if and only if ∈ k. Then hp − h acts on L(µ) by 2µp d1(µ1 − ρ1) + d2(µ2 − ρ2) ∈ Fp. (19) Since k = ∞ and the list of values µi − ρi is finite, we can always choose d1, d2 ∈ k such that (19) does not hold, provided there is no pair µ 6= ρ such that µ1 = ρ1 and µ2 = ρ2. This can be checked directly: e.g. −2 = 4 if and only if 6 = 0; −2 = 4ω if and only if p − 1 = 2ω which implies that (p − 1)3 = p − 1 = 8 and hence p = 9 - both clearly contradictions. Up to conjugacy, there is only one proper parabolic subgroup of G4, it is Z3. We may assume that Z3 = hs1i = {1, s1, t1}. Repeating the above argument for Z3 and noting that IrrZ3 = {T Z3, V1Z3, V2Z3}, one sees that the block partition of IrrZ3 will not consist of singletons precisely if equation (19) is satisfied for ρ, µ ∈ {T, V1, V2}. (cid:3) 6.8. The Kac-Weisfeiler conjecture. The following is a result in the spirit of the Kac-Weisfeiler conjec- ture. Proposition. The subset Creg of C consisting of all parameters c such that the dimension of every simple Hc(W )-module is divisible by pn is open and dense. Note that, since the P.I. degree of Hc(W ) is pnW , pn is also the largest power of p dividing the dimension of any simple Hc(W )-module. Also, one can show that Creg is always a proper subset of C. 21 Proof. First, we show that the subset Creg(W ) of C consisting of all c such that the dimension of every simple Hc(W )-module is divisible by pn is open and dense in C. Consider the k[C]-algebra Hc(W ). It is free as a k[C]-module. Therefore it is a continuous family of L-algebras in the sense of [PS, Definition 2.2] (for L trivial). Then [PS, Lemma 2.3] says that, for each integer d, the set of points c ∈ C such that Hc(W ) contains a two-sided ideal of dimension d is closed in C. The annihilator of a simple Hc(W )-module M has codimension (dimk M )2 (recall that k is assumed to be algebraically closed). Therefore the set of all points in C for which there exists an ideal in Hc(W ) whose codimension belongs to {(pir)2 0 ≤ i < n, 1 ≤ r ≤ W } is closed and its compliment is Creg(W ). We just need to show that this set is non-empty. Take any line ℓ in C. Then Lemma 6.5 implies that ℓ ∩ Creg(W ) is dense in ℓ. Note that this actually shows that there is no linear subspace of C contained in C − Creg(W ). Now we treat the general case. Let W ′ be a parabolic subgroup of W and CW ′ the space of parameters for Hc(W ′). Restriction of parameters defines a linear map ρ : C → CW ′, its image is a non-zero linear subspace of CW ′ . We have shown that the set of points in CW ′ for which all simple Hc(W ′)-module have "maximum p-dimension" is open and dense. If Creg(W ′) denotes the pre-image of this set under ρ then it is open and dense in C. Then it follows from Propositions 5.3 and 5.4 together with Corollary 5.4 that Creg =\W ′ Creg(W ′), where the intersection is over all conjugacy classes of parabolic subgroups of W . (cid:3) Remark. We remark that our reduction method (combining Propositions 5.3 and 5.4 together with Corollary 5.4) also gives a different proof of the result [Tik, Corollary 4.2]: If M is a simple Hc(W ) and b ∈ V such that the support of M equals the image of b in V /W then pW/Wb divides dimk M . Example. When W = Sn+1, Creg = k − F× p . One can show this as follows: by remark 5.5, the algebra Hc(Sn+1) is Azumaya for all c ∈ k − Fp and hence k − Fp ⊆ Creg. On the other hand, direct calculations show that C − Creg(S2) = F× p ⊂ C − Creg. Finally, when c = 0, Hc(Sn+1) = D(V ) ⋊ Sn+1 and, even though this is not an Azumaya algebra, it follows from Lemma 6.1 that pn does divide dimk L for all simple D(V ) ⋊ Sn+1-modules L. p . Therefore Propositions 5.3 implies that F× References [BC1] M. Balagovic and H. Chen. Category O for Rational Cherednik Algebras Ht,c(GL2(Fp), h) in Characteristic p. arXiv:1107.5996v1. [BC2] M. Balagovic and H. Chen. Representations of Rational Cherednik Algebras in Positive Characteristic. arXiv:1107.0504v1. [BC3] K. A. Brown and K. Changtong. Symplectic reflection algebras in positive characteristic. Proc. Edinb. Math. Soc. (2), 53(1):61 -- 81, 2010. [BE] R. Bezrukavnikov and P. Etingof. Parabolic induction and restriction functors for rational Cherednik algebras. Selecta Math. (N.S.), 14(3-4):397 -- 425, 2009. [Bel1] G. Bellamy. The Calogero-Moser partition for G(m, d, n). arXiv:0911.0066v1. [Bel2] G. Bellamy. On singular Calogero-Moser spaces. Bull. Lond. Math. Soc., 41(2):315 -- 326, 2009. [BFG] R. Bezrukavnikov, M. Finkelberg, and V. Ginzburg. Cherednik algebras and Hilbert schemes in characteristic p. Repre- sent. Theory, 10:254 -- 298, 2006. With an appendix by Pavel Etingof. [BG] K. A. Brown and I. G. Gordon. The ramification of centres: Lie algebras in positive characteristic and quantised enveloping algebras. Math. Z., 238(4):733 -- 779, 2001. 22 [BGS] K. A. Brown, I. G. Gordon, and C. H. Stroppel. Cherednik, Hecke and quantum algebras as free Frobenius and Calabi- Yau extensions. J. Algebra, 319(3):1007 -- 1034, 2008. [CR] C. W. Curtis and I. Reiner. Methods of representation theory. Vol. I. John Wiley & Sons Inc., New York, 1981. With applications to finite groups and orders, Pure and Applied Mathematics, A Wiley-Interscience Publication. [EG] P. Etingof and V. Ginzburg. Symplectic reflection algebras, Calogero-Moser space, and deformed Harish-Chandra ho- momorphism. Invent. Math., 147(2):243 -- 348, 2002. [Eis] D. Eisenbud. Commutative Algebra, volume 150 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1995. With a view toward algebraic geometry. [Gor1] I. G. Gordon. Baby Verma modules for rational Cherednik algebras. Bull. London Math. Soc., 35(3):321 -- 336, 2003. [Gor2] I. G. Gordon. Quiver varieties, category O for rational Cherednik algebras, and Hecke algebras. Int. Math. Res. Pap. IMRP, (3):Art. ID rpn006, 69, 2008. [GP] M. Geck and G. Pfeiffer. Characters of finite Coxeter groups and Iwahori-Hecke algebras, volume 21 of London Math- ematical Society Monographs. New Series. The Clarendon Press Oxford University Press, New York, 2000. [Gri] S. Griffeth. Towards a combinatorial representation theory for the rational Cherednik algebra of type G(r, p, n). arXiv:math/0612733, to appear in Proceedings of the Edinburgh Mathematical Society, 2006. [HN] R. R. Holmes and D. K. Nakano. Brauer-type reciprocity for a class of graded associative algebras. J. Algebra, 144(1):117 -- 126, 1991. [Kat] N. M. Katz. Nilpotent connections and the monodromy theorem: Applications of a result of Turrittin. Inst. Hautes ´Etudes Sci. Publ. Math., (39):175 -- 232, 1970. [KM] G. Kemper and G. Malle. The finite irreducible linear groups with polynomial ring of invariants. Transform. Groups, 2(1):57 -- 89, 1997. [LT] G. I. Lehrer and D. E. Taylor. Unitary reflection groups, volume 20 of Australian Mathematical Society Lecture Series. Cambridge University Press, Cambridge, 2009. [Mar1] M. Martino. Blocks of restricted rational Cherednik algebras for G(m, d, n). arXiv:1009.3200v1. [Mar2] M. Martino. Stratifications of Marsden-Weinstein reductions for representations of quivers and deformations of sym- plectic quotient singularities. Math. Z., 258(1):1 -- 28, 2008. [Mar3] M. Martino. The Calogero-Moser partition and Rouquier families for complex reflection groups. J. Algebra, 323:193 -- 205, 2010. [MR] J. C. McConnell and J. C. Robson. Noncommutative Noetherian Rings, volume 30 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, revised edition, 2001. With the cooperation of L. W. Small. [PS] A. Premet and S. Skryabin. Representations of restricted Lie algebras and families of associative L-algebras. J. Reine Angew. Math., 507:189 -- 218, 1999. [Pus] I. A. Pushkarev. On the theory of representations of the wreath products of finite groups and symmetric groups. Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI), 240(Teor. Predst. Din. Sist. Komb. i Algoritm. Metody. 2):229 -- 244, 294 -- 295, 1997. [Smi] L. Smith. Polynomial invariants of finite groups, volume 6 of Research Notes in Mathematics. A K Peters Ltd., Wellesley, MA, 1995. [Thi] U. Thiel. Generic Calogero-Moser blocks for exceptional complex reflection groups, in preparation. [Tik] A. Tikaradze. An analogue of the Kac-Weisfeiler conjecture. arXiv:1007.2387v3. School of Mathematics, Room 2.233, Alan Turing Building, University of Manchester, Oxford Road, Manch- ester, M13 9PL E-mail address: [email protected] Mathematisches Institut, Endenicher Allee 60, 53115 Bonn, Germany E-mail address: [email protected] 23
1509.07883	2	1509	2016-02-22T21:02:16	Determinantal representations of W-weighted Drazin inverse solutions of some quaternion matrix equations	[ "math.RA" ]	By using determinantal representations of the W-weighted Drazin inverse previously obtained by the author within the framework of the theory of the column-row determinants, we get explicit formulas for determinantal representations of the W-weighted Drazin inverse solutions (analogs of Cramer's rule) of the quaternion matrix equations $ {\bf W}{\bf A}{\bf W}{\bf X}={\bf D}$, $ {\bf X}{\bf W}{\bf A}{\bf W}={\bf D} $, and ${\bf W}_{1}{\bf A}{\bf W}_{1}{\bf X}{\bf W}_{2}{\bf B}{\bf W}_{2}={\bf D} $.	math.RA	math	Determinantal representations of W-weighted Drazin inverse solutions of some quaternion matrix equations Ivan I. Kyrchei ∗ Pidstrygach Institute for Applied Problems of Mechanics and Mathematics NAS of Ukraine, Lviv, 79060, Ukraine Abstract: By using determinantal representations of the W-weighted Drazin inverse previously obtained by the author within the framework of the theory of the column-row determinants, we get explicit formulas for determinantal rep- resentations of the W-weighted Drazin inverse solutions (analogs of Cramer's rule) of the quaternion matrix equations WAWX = D, XWAW = D, and W1AW1XW2BW2 = D. Keywords: W-weighted Drazin inverse, Quaternion matrix, Cramer rule 2000 AMS subject classifications: 15A15, 16W10. 1 Introduction Throughout the paper, we denote the real number field by R, the set of all m×n matrices over the quaternion algebra H = {a0 + a1i + a2j + a3k i2 = j2 = k2 = −1, a0, a1, a2, a3 ∈ R} by Hm×n, and by Hm×n the set of all m × n matrices over H with a rank r. Let M (n, H) be the ring of n × n quaternion matrices. For A ∈ Hn×m, the symbols A∗ stands for the conjugate transpose (Hermitian adjoint) matrix of A. The matrix A = (aij ) ∈ Hn×n is Hermitian if A∗ = A. r In the past, researches into the quaternion skew field had more a theoretical importance, but now a growing number of investigations give wide practical applications of quaternions. In particular through their attitude orientation, the quaternions arise in various fields such as quaternionic quantum theory [1], fluid mechanics and particle dynamics [2, 3], computer graphics [4], aircraft orientation [5], robotic systems [6], life science [7, 8] and etc. Research on quaternion matrix equations and generalized inverses, which are usefulness tools used to solve matrix equations, has been actively ongoing for more recent years. We mention only some recent papers. Yuan, Wang and Duan [9] derived solutions of the quaternion matrix equation AX = B and their applications in color image restoration. Wang and Yu [10] studied extreme ranks of real matrices in solution of the quaternion matrix equation AXB = C. Yuan, Liao and Lei [11] obtained the expressions of least squares Hermitian solution ∗E-mail address: [email protected] 1 with minimum norm of the quaternion matrix equation (AXB, CXD) = (E, F ). Feng and Cheng [12] gave a clear description of the solution set to the quaternion matrix equation AX − ¯XB = 0. Jiang and Wei [13] derived the explicit solution of the quaternion matrix equation X − A XB = C. Song, Chen and Wang [14] obtained the expressions of the explicit solutions of quaternion matrix equations XF −AX = BY and XF −A X = BY . Yuan and Wang [15] gave the expressions of the least squares η-Hermitian solution with the least norm of the quaternion matrix equation AXB + CXD = E. Zhang, Wei, Lia and Zhao derived [16] the expressions of the minimal norm least squares solution, the pure imaginary least squares solution, and the real least squares solution for the quaternion matrix equation AX = B. The definitions of the generalized inverse matrices have been extended to quaternion matrices as follows. The Moore-Penrose inverse of A ∈ Hm×n, denoted by A†, is the unique matrix X ∈ Hn×m satisfying the following equations 1) AXA = A; 2) XAX = X; 3) (AX)∗ = AX; 4) (XA)∗ = XA. For A ∈ Hn×n with k = Ind A the smallest positive number such that rank Ak+1 = rank Ak, the Drazin inverse of A, denoted by AD, is defined to be the unique matrix X that satisfying Eq. 2) and the equations 5) AX = XA; 6) Ak+1X = Ak. In particular, when Ind A = 1, then X is called the group inverse of A and is denoted by X = Ag. If Ind A = 0, then A is nonsingular, and AD ≡ A† = A−1. Cline and Greville [17] extended the Drazin inverse of square matrix to rect- angular matrix, that has been generalized to the quaternion algebra as follows. For A ∈ Hm×n and W ∈ Hn×m, the W-weighted Drazin inverse of A with respect to W is the unique solution to the following equations 7) (AW)k+1XW = (AW)k; 8) XWAWX = X; 9) AWX = XWA, where k = max{Ind(AW), Ind(WA)}. The Drazin inverse and weighted Drazin inverse has several important appli- cations such as, applications in singular differential and difference equations [18], signal processing [19], Marckov chains and statistic problems [20, 21], descrip- tor continuous-time systems [22], numerical analysis and Kronecker product systems [23], solving singular fuzzy linear system [24, 25], constrained linear systems [26] and etc. Cramer's rule for the W-weighted Drazin inverse solutions, in particular, have been derived in [27] for singular linear equations and in [26] for a class of restricted matrix equations. Recently, within the framework of the theory of the column-row determinants Song [28] has first obtained a determinantal representation of the W-weighted Drazin inverse and Cramer's rule of a class of restricted matrix equations over the quaternion algebra. But in obtaining, he has used auxiliary matrices other than that are given. In [29], we have obtained new determinantal representations of the W-weighted Drazin inverse over the quaternion skew field without any auxiliary matrices. An important application of determinantal representations of generalized inverses is the Cramer rule for generalized inverse solutions of matrix equations. 2 But, when there is a need for a W-weighted Drazin inverse solution? Con- sider for example the following matrix equation, A1X = D. If A1 is rectangular and we can represent it as A1 = WAW, where WA and AW are quadratic and singular, then its W-weighted Drazin inverse solution is needed. In the paper we investigate analogs of Cramer's rule for W-weighted Drazin inverse solutions of the following matrix equations over the quaternion skew field H, WAWX = D, XWAW = D, W1AW1XW2BW2 = D. (1) (2) (3) The paper is organized as follows. We start with some basic concepts and results from the theory of the row-column determinants in Subsection 2.1. In Subsection 2.2, we give determinantal representations of the Moore-Penrose and Drazin inverses for a quaternion matrix. Determinantal representations of the W-weighted Drazin inverse and its properties we consider in Subsection 2.3. In Subsection 3.1, we give the background of the problem of Cramer's rule for the W-weighted Drazin inverse solution. In Subsection 3.2 we obtain explicit representation formulas of the W-weighted Drazin inverse solutions (analogs of Cramer's rule) of the quaternion matrix equations (1-3). In Section 4, we give numerical examples to illustrate the main result. 2 Preliminaries 2.1 Elements of the theory of the column and row deter- minants The theory of the row-column determinants over the quaternion skew field has been introduced in [30 -- 32], and later it has been applied to research generalized inverses and generalized inverse solutions of matrix equations. In particular, determinantal representations of the Moore-Penrose [33, 34] and explicit repre- sentation formulas for the minimum norm least squares solutions of some quater- nion matrix equations [35], and determinantal representations of the Drazin [36], and W-weighted Drazin inverses [29] have been obtained by the author. Song at al. derived determinantal representation of the generalized inverse A2 T,S [37], Bott-Duffin inverse [38] and the Cramer rule for the solutions of restricted ma- trix equations [39], and the generalized Stein quaternion matrix equation [40], etc. For A = (aij) ∈ M (n, H) we define n row determinants and n column determinants as follows. Suppose Sn is the symmetric group on the set In = {1, . . . , n}. Definition 2.1 The i-th row determinant of A = (aij ) ∈ M (n, H) is defined for all i = 1, n by putting rdetiA = Xσ∈Sn (−1)n−r ai ik1 aik1 ik1 +1 . . .aik1 +l1 i . . . aikr ikr +1 . . . aikr +lr ikr , σ = (i ik1 ik1+1 . . . ik1+l1) (ik2 ik2+1 . . . ik2+l2) . . . (ikr ikr+1 . . . ikr+lr ) , with conditions ik2 < ik3 < . . . < ikr and ikt < ikt+s for t = 2, r and s = 1, lt. 3 Definition 2.2 The j-th column determinant of A = (aij) ∈ M (n, H) is de- fined for all j = 1, n by putting cdetj A = Xτ ∈Sn (−1)n−r ajkr jkr +lr . . . ajkr +1ikr . . .aj jk1 +l1 . . . ajk1 +1jk1 ajk1 j, τ = (jkr +lr . . . jkr +1jkr ) . . . (jk2+l2 . . . jk2+1jk2 ) (jk1+l1 . . . jk1+1jk1 j) , with conditions, jk2 < jk3 < . . . < jkr and jkt < jkt+s for t = 2, r and s = 1, lt. Suppose Ai j denotes the submatrix of A obtained by deleting both the ith row and the jth column. Let a.j be the j-th column and ai. be the i-th row of A. Suppose A.j (b) denotes the matrix obtained from A by replacing its j-th column with the column b, and Ai. (b) denotes the matrix obtained from A by replacing its i-th row with the row b. The following theorem has a key value in the theory of the column and row determinants. Theorem 2.1 [30] If A = (aij ) ∈ M (n, H) is Hermitian, then rdet1A = · · · = rdetnA = cdet1A = · · · = cdetnA ∈ R. Since all column and row determinants of a Hermitian matrix over H are equal, we can define the determinant of a Hermitian matrix A ∈ M (n, H). By defi- nition, we put, det A := rdeti A = cdeti A, for all i = 1, n. The determinant of a Hermitian matrix has properties similar to a usual determinant. They are completely explored in [30, 31] by its row and column determinants. They can be summarized by the following theorems. Theorem 2.2 If the i-th row of a Hermitian matrix A ∈ M (n, H) is replaced with a left linear combination of its other rows, i.e. ai. = c1ai1 . + . . . + ckaik ., where cl ∈ H for all l = 1, k and {i, il} ⊂ In, then rdeti Ai . (c1ai1. + . . . + ckaik .) = cdeti Ai . (c1ai1. + . . . + ckaik .) = 0. Theorem 2.3 [30] If the j-th column of a Hermitian matrix A ∈ M (n, H) is replaced with a right linear combination of its other columns, i.e. a.j = a.j1 c1 + . . . + a.jk ck, where cl ∈ H for all l = 1, k and {j, jl} ⊂ Jn, then cdetj A.j (a.j1 c1 + . . . + a.jk ck) = rdetj A.j (a.j1 c1 + . . . + a.jk ck) = 0. The determinant of a Hermitian matrix also has a property of expansion along arbitrary rows and columns using row and column determinants of submatrices. We have the following theorem on the determinantal representation of the inverse matrix over H. Theorem 2.4 [30] The necessary and sufficient condition of invertibility of A ∈ M(n, H) is ddetA 6= 0. Then there exists A−1 = (LA)−1 = (RA)−1, where (LA)−1 = (A∗A)−1 A∗ = L11 L21 L12 L22 . . . . . . L1n L2n . . . Ln1 . . . Ln2 . . . . . . . . . Lnn ,   1 ddetA   4 (4) (RA)−1 = A∗ (AA∗)−1 = 1 ddetA∗   R11 R21 R12 R22 . . . . . . R1n R2n . . . Rn1 . . . Rn2 . . . . . . . . . Rnn   (5) and Lij = cdetj(A∗A).j (a∗ . i) , R ij = rdeti(AA∗)i.(cid:0)a∗ j.(cid:1) , for i, j = 1, n, and ddetA = det(AA∗) = det(A∗A). 2.2 Determinantal representations of the Moore-Penrose and Drazin inverses over the quaternion skew field We shall use the following notations. Let α := {α1, . . . , αk} ⊆ {1, . . . , m} and β := {β1, . . . , βk} ⊆ {1, . . . , n} be subsets of the order 1 ≤ k ≤ min {m, n}. By Aα β denote the submatrix of A determined by the rows indexed by α and the columns indexed by β. Then A α α denotes the principal submatrix determined by the rows and columns indexed by α. If A ∈ M (n, H) is Hermitian, then by Aα α denote the corresponding principal minor of det A. For 1 ≤ k ≤ n, the collection of strictly increasing sequences of k integers chosen from {1, . . . , n} is denoted by Lk,n := { α : α = (α1, . . . , αk) , 1 ≤ α1 ≤ . . . ≤ αk ≤ n}. For fixed i ∈ α and j ∈ β, let Ir, m{i} := { α : α ∈ Lr,m, i ∈ α}, Jr, n{j} := { β : β ∈ Lr,n, j ∈ β}. i. the j-th column and the i-th row of A∗ and by a(m) Denote by a∗ .j and a∗ .j and a(m) i. the j-th column and the i-th row of Am, respectively. The following theorem give determinantal representations of the Moore- Penrose inverse over the quaternion skew field H. Theorem 2.5 [32] If A ∈ Hm×n , then the Moore-Penrose inverse A+ = r (cid:0)a+ ij(cid:1) ∈ Hn×m possess the following determinantal representations: ij = Pβ∈Jr, n{i} a+ ij = Pα∈Ir,m{j} a+ β .j(cid:1)(cid:1) β i. )) α α (A∗A) β cdeti(cid:0)(A∗A) . i(cid:0)a∗ Pβ∈Jr, n(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) Pα∈Ir, m rdetj ((AA∗)j . (a∗ (AA∗) α α , . (6) (7) or for all i = 1, n, j = 1, m. Proposition 2.1 [20] If Ind(A) = k, then AD = Ak(A2k+1)+Ak. Denote by a.s and at. the s-th column of (A2k+1)∗Ak =: A = (aij) ∈ Hn×n and the t-th row of Ak(A2k+1)∗ =: A = (aij ) ∈ Hn×n, respectively for all s, t = 1, n. Using the determinantal representations of the Moore-Penrose inverse (6) and (7), and Proposition 2.1 the following determinantal representations of the Drazin inverse for an arbitrary square matrix over H have been obtained in [33]. 5 Theorem 2.6 [33] If A ∈ M (n, H) with Ind A = k and rank Ak+1 = rank Ak = r, then the Drazin inverse AD possess the determinantal representations, a(k) n Pt=1 aD ij = and aD ij = n Ps=1 Pα∈Ir, n{s} β (A2k+1)∗ (A2k+1) β cdett(cid:16)(cid:0)A2k+1(cid:1)∗(cid:0)A2k+1(cid:1). t (a. j)(cid:17) β it Pβ∈Jr, n{t} Pβ∈Jr, n(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) rdets(cid:16)(cid:16)A2k+1(cid:0)A2k+1(cid:1)∗(cid:17) .s Pα∈Ir, n(cid:12)(cid:12)(cid:0)A2k+1 (A2k+1)∗(cid:1) α α(cid:12)(cid:12) (ai .)(cid:17) α α! a(k) sj (8) . (9) In the special case, when A ∈ M (n, H) is Hermitian, we can obtain simpler determinantal representations of the Drazin inverse. Theorem 2.7 [33] If A ∈ M (n, H) is Hermitian with Ind A = k and rank Ak+1 = rank Ak = r, then the Drazin inverse AD =(cid:0)aD determinantal representations, ij(cid:1) ∈ Hn×n possess the following ij = Pβ∈Jr, n{i} aD β .j(cid:1)(cid:1) β (k) i. )(cid:17) α α (Ak+1) β cdeti(cid:0)(cid:0)Ak+1(cid:1) . i(cid:0)ak Pβ∈Jr, n(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) rdetj(cid:16)(Ak+1)j . (a Pα∈Ir, n (Ak+1) α α , . (10) (11) or ij = Pα∈Ir,n{j} aD 2.3 Determinantal representations of the W-weighted Drazin inverse Definition 2.3 For an arbitrary matrix over the quaternion skew field, A ∈ Hm×n, we denote by Rr(A) = {y ∈ Hm : y = Ax, x ∈ Hn} , the column right space of A, Nr(A) = {y ∈ Hn : Ax = 0}, the right null space of A, Rl(A) = {y ∈ Hn : y = xA, x ∈ Hm}, the column left space of A, Nr(A) = {y ∈ Hm : xA = 0}, the left null space of A. We introduce some mathematical background from the theory of the W- weighted Drazin inverse [27, 41, 42] that can be generalized to H. Lemma 2.1 Let A ∈ Hm×n and W ∈ Hn×m with k = max{Ind(AW), Ind(WA)}. 6 Than we have: A; (b) Ad,WW = (AW)D; WAd,W = (WA)D; (a) Ad,W = A(cid:0)(WA)D)(cid:1)2 (c) Ad,W =n(AW)k(cid:2)(AW)2k+1(cid:3)+ =(cid:0)(AW)D)(cid:1)2 Ad,W = W+n(WA)k(cid:2)(WA)2k+1(cid:3)+ (AW)ko W+; (WA)ko ; (d) WAWAd,W = PRr ((WA)k),Nr((WA)k); Ad,WWAW = PRl((AW)k),Nl((AW)k), where PRr((WA)k),Nr ((WA)k) is the projector on Rr((WA)k) along Nr((WA)k), and PRl((AW)k),Nl((AW)k) is the projector on Rl((AW)k) along Nl((AW)k). In particular, the point (a) of Lemma 2.1 due to Cline and Greville [17] is generalized [28] to H. Using this proposition, we have obtained [29] the following determinantal representations W-weighted Drazin inverse. Denote WA =: U = (uij) ∈ Hn×n and AW =: V = (vij ) ∈ Hm×m. Due to Theorem 2.6, we denote an entry of the Drazin inverse UD by n Pt=1 uD,1 ij = u(k) or uD,2 ij = n Ps=1 Pα∈Ir, n{s} (12) (13) (14) (15) (16) β (U2k+1)∗ (U2k+1) β cdett(cid:16)(cid:0)U2k+1(cid:1)∗(cid:0)U2k+1(cid:1). t (u. j)(cid:17) β it Pβ∈Jr, n{t} Pβ∈Jr, n(cid:12)(cid:12)(cid:12) rdets(cid:16)(cid:16)U2k+1(cid:0)U2k+1(cid:1)∗(cid:17) .s Pα∈Ir, n(cid:12)(cid:12)(cid:0)U2k+1 (U2k+1)∗(cid:1) α α(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) (ui .)(cid:17) α α! u(k) sj where u.s and ut. are the s-th column of (U2k+1)∗Uk =: U = (uij) ∈ Hn×n and the t-th row of Uk(U2k+1)∗ =: U = (uij) ∈ Hn×n, respectively for all s, t = 1, n, r = rank Uk+1 = rank Uk. Then we have the following determinantal representations of Ad,W = (ad,W ) ∈ ij Hm×n, where ad,W ij = n Xq=1 aiq(uD qj)(2) (uD qj)(2) = qp uD,f uD,l pj n Xp=1 for all l, f = 1, 2, and uD,1 from (12) and uD,2 ij from (13). ij Similarly using V = (vij ) ∈ Hm×m, ad,W ij = m Xq=1 (vD iq )(2)aqj. 7 where the first factor is one of the four possible equations (vD iq )(2) = vD,l ip vD,f pq m Xp=1 for all l, f = 1, 2, and an entry of the Drazin inverse VD is denoting by m Pt=1 vD,1 ij = v(k) or vD,2 ij = m Ps=1 Pα∈Ir, m{s} β (V2k+1)∗ (V2k+1) β cdett(cid:16)(cid:0)V2k+1(cid:1)∗(cid:0)V2k+1(cid:1). t (v. j)(cid:17) β it Pβ∈Jr, m{t} Pβ∈Jr, m(cid:12)(cid:12)(cid:12) rdets(cid:16)(cid:16)V2k+1(cid:0)V2k+1(cid:1)∗(cid:17) .s Pα∈Ir, m(cid:12)(cid:12)(cid:0)V2k+1 (V2k+1)∗(cid:1) α α(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) (vi .)(cid:17) α α! v(k) sj (17) (18) , (19) where v.s and vt. are the s-th column of (V2k+1)∗Vk =: V = (vij ) ∈ Hm×m and the t-th row of Vk(V2k+1)∗ =: V = (vij ) ∈ Hm×m, respectively for all s, t = 1, m, r = rank Vk+1 = rank Vk. The point (c) of Lemma 2.1 due to [23] has been generalized to H in [33]. Using this proposition, we have obtained the following two determinantal rep- resentations of the W-weighted Drazin inverse. Theorem 2.8 [29] Let A ∈ Hm×n and W ∈ Hn×m with k = Ind(AW) and r = rank(AW)k+1 = rank(AW)k. Then the W-weighted Drazin inverse of A with respect to W possesses the determinantal representations r1 ad,W ij = m Pt=1 Pα∈Ir, m{t} rdett(cid:16)(cid:16)V2k+1(cid:0)V2k+1(cid:1)∗(cid:17)t. (vi .)(cid:17) α Pα∈Ir, m(cid:12)(cid:12)(cid:0)V2k+1 (V2k+1)∗(cid:1) α α Pα∈Ir1 , n{j} α(cid:12)(cid:12) Pα∈Ir1 , n rdetj(cid:16)(WW∗)j . ( wt.)(cid:17) α α (WW∗) α α and ad,W ij = n Pt=1 Pβ∈Jr1, m{i} cdeti ((W∗W).t ( w.t)) β β Pβ∈Jr, n{t} β(cid:12)(cid:12)(cid:12) Pβ∈Jr, n(cid:12)(cid:12)(cid:12)(cid:0)(U2k+1)∗ cdett(cid:16)(cid:16)(cid:0)U2k+1(cid:1)∗ β(cid:12)(cid:12)(cid:12) U2k+1(cid:1) β (W∗W) β Pβ∈Jr1 , m(cid:12)(cid:12)(cid:12) (20) U2k+1(cid:17).t (u.j )(cid:17) β β (21) where V = Vk(V2k+1)∗, W = VkW∗, and U = (U2k+1)∗Uk, W = W∗Uk. In the special cases, when AW ∈ Hm×m and WA ∈ Hn×n are Hermitian, we can obtain simpler determinantal representations of the W-weighted Drazin inverse. 8 Theorem 2.9 [29] If A ∈ Hm×n, W ∈ Hn×m, and AW ∈ Hm×m is Hermi- tian with k = max{Ind(AW), Ind(WA)} and rank(AW)k+1 = rank(AW)k = W possess the following determinantal representations: r, then the W-weighted Drazin inverse Ad,W =(cid:16)ad,W ij (cid:17) ∈ Hm×n with respect to (¯v.j)(cid:17) β cdeti(cid:16)(AW)k+2 Pβ∈Jr, m(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) where ¯v.j is the jth column of ¯V = (AW)kA for all j = 1, m. ij = Pβ∈Jr, m{i} (AW)k+2 β ad,W (22) . i β , Theorem 2.10 [29] If A ∈ Hm×n, W ∈ Hn×m, and WA ∈ Hn×n is Hermi- tian with k = max{Ind(AW), Ind(WA)} and rank(WA)k+1 = rank(WA)k = W possess the following determinantal representations: r, then the W-weighted Drazin inverse Ad,W =(cid:16)ad,W rdetj(cid:16)(WA)k+2 Pα∈Ir, n(cid:12)(cid:12)(cid:12) α(cid:12)(cid:12)(cid:12) ij = Pα∈Ir,n{j} where ¯ui. is the ith row of ¯U = A(WA)k for all i = 1, n. (WA)k+2 α j . (¯u(k) ad,W ij (cid:17) ∈ Hm×n with respect to i. )(cid:17) α (23) α . 3 Cramer's rule for the W-weighted Drazin in- verse solution 3.1 Background of the problem In [27] Wei has established a Cramer's rule for solving of a general restricted equation WAWx = b, x ∈ R(cid:2)(AW)k1(cid:3) , (24) where A ∈ Cm×n, W ∈ Cn×m with Ind (AW) = k1, Ind (WA) = k2 and rank (AW)k1 = r1, rank (WA)k2 = r2. He proofed that the restricted matrix equation (24) has a unique solution, x = Ad,W b, and presented its Cramer's rule as follows, xj = det(cid:18)WAW(j −→ b) U1 0 (cid:19) , 0 (cid:19)(cid:30) det(cid:18)WAW U1 V1 (25) are matrices whose columns form bases where U1 ∈ Cn×n−r2 for N ((WA)k2 ) and N ((AW)k∗ , V∗ n−r2 m−r1 1 ), respectively. V1(j −→ 0) 1 ∈ Cm×m−r1 Recently, within the framework of a theory of the column and row determi- nants Song [28] has considered the characterization of the W-weighted Drazin inverse over the quaternion skew and presented a Cramer's rule of the restricted matrix equation, W1AW1XW2BW2 = D, Rr(X) ⊂ Rr(cid:0)(AW1)k1(cid:1) , Nr(X) ⊃ Nr(cid:0)(W2B)k2(cid:1) , Rl(X) ⊂ Rl(cid:0)(BW2)k2(cid:1) , Nl(X) ⊃ Nl(cid:0)(W1A)k1(cid:1) , (26) (27) 9 where A ∈ Hm×n, W1 ∈ Hn×m, B ∈ Hp×q, W2 ∈ Hq×p, and D ∈ Hn×p with k1 = max {Ind(AW1), Ind (W1A)}, k2 = max {Ind(BW2), Ind (W2B)}, and rank (AW1)k1 = s1, rank (BW2)k2 = s2. He proofed that if D ∈ Rr(cid:0)(W1A)k1 , (W2B)k2(cid:1) , D ∈ Rl(cid:0)(AW1)k1 , (BW2)k2(cid:1) and there exist auxiliary matrices of full column rank, L1 ∈ Hn×n−s1 1 ∈ Hm×m−s1 with additional terms of their ranges and null spaces, then the restricted matrix equation (26) has a unique solution, , L2 ∈ Hq×q−s2 2 ∈ Hp×p−s2 p−s2 , M∗ q−s2 , M∗ m−s1 n−s1 X = Ad,W1 DBd,W2. Using auxiliary matrices, L1, M1, L2, M2, Song presented its Cramer's rule by analogy to (25). In this paper we have avoided such approach and have obtained explicit formulas for determinantal representations of the W-weighted Drazin inverse solutions of matrix equations by using only given matrices. 3.2 A Cramer's rule for the W-weighted Drazin inverse solutions of some matrix equations Consider the following restricted matrix equation, WAWX = D, Rr(X) ⊂ Rr(cid:0)(AW)k(cid:1) , Nl(X) ⊃ Nl(cid:0)(WA)k(cid:1) , with k = max {Ind(AW), Ind (WA)}, and where A ∈ Hm×n, W ∈ Hn×m D ∈ Hn×p . r1 (28) (29) Theorem 3.1 If D ⊂ Rr(cid:0)(AW)k(cid:1) and D ⊃ Nl(cid:0)(WA)k(cid:1), then the restricted matrix equation (28) has a unique solution, X = Ad,W D, (30) which possess the following determinantal representations for all i = 1, m, j = 1, p, i) xij = n Pt=1 Pβ∈Jr1, m{i} cdeti ((W∗W).t ( w.t)) β β Pβ∈Jr, n{t} β(cid:12)(cid:12)(cid:12) Pβ∈Jr, n(cid:12)(cid:12)(cid:12)(cid:0)(U2k+1)∗ cdett(cid:16)(cid:16)(cid:0)U2k+1(cid:1)∗ β(cid:12)(cid:12)(cid:12) U2k+1(cid:1) β U2k+1(cid:17).t (d.j )(cid:17) β β (31) (W∗W) β Pβ∈Jr1 , m(cid:12)(cid:12)(cid:12) where d.j is the j-th column of D = UD = (U2k+1)∗UkD, U = WA, W = W∗Uk, and r = rank(WA)k+1 = rank(WA)k. ii) xij = m Xq=1 (vD iq )(2)rqj , 10 (32) where (vD iq )(2) can be obtained by (17) and AD = R = (rqj ) ∈ Hm×p. iii) If AW ∈ Hm×m is Hermitian, then xij = Pβ∈Jr, m{i} β . i cdeti(cid:16)(AW)k+2 (f.j)(cid:17) β Pβ∈Jr, m(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) (AW)k+2 β where f.j is the j-th column of F = ¯VD = (AW)kAD. , (33) Proof. The proof contains two parts. We first shall establish that the unique solution of (28) can be represented as (30). By the definition of the right range, we have for some matrix Y ∈ Hn×p. It follows that D = (WA)kY Then by Lemma 2.1 (d), Rr(D) ⊂ Rr(cid:0)(WA)k(cid:1) . WAWAd,W D = PRr ((WA)k),Nr((WA)k)D = D. It means that (30) is a solution of (28) and satisfies the restricted conditions (29). Now we prove the uniqueness of (30). Let X0 is a solution of (28). Then it satisfies the restricted conditions (29), and Ad,W D = Ad,W WAWX0 = PRr ((WA)k),Nr ((WA)k)X0 = X0. To derive a Cramer's rule (31), we use the determinantal representation (21) for Ad,W . Then p xij = ad,W is dsj = n Xs=1 Pt=1 Pβ∈Jr1, m{i} p Xs=1 cdeti(W∗W).t ( w.t)β  U2k+1(cid:17).t cdett(cid:16)(cid:0)U2k+1(cid:1)∗  β(cid:12)(cid:12)(cid:12) U2k+1(cid:1) β Denote D = UD = (U2k+1)∗UkD, where D =(cid:16) dsj(cid:17) ∈ Hn×p. Since β Pβ∈Jr, n{t} β(cid:12)(cid:12)(cid:12) Pβ∈Jr, n(cid:12)(cid:12)(cid:12)(cid:0)(U2k+1)∗ Pβ∈Jr1, m(cid:12)(cid:12)(cid:12) (W∗W) β p β (u.s) β   dsj (34) u.sdsj = d.j, Xs=1 where d.j is the j-th column of D, then (31) follows from (34). Similarly, we derive the analogs of Cramer's rule (32) and (33) by using the determinantal representations for the W-weighted Drazin inverse (16), and (22), respectively. (cid:3) 11 Remark 3.1 In the complex case, i.e. A ∈ Cm×n, W ∈ Cn×m , and D ∈ Cn×p, we substitute usual determinants for all corresponding row and column determinants in (31), (32), and (33). r1 Note that in the case iii), the condition AW ∈ Cm×m be Hermitian is not necessary, then in the complex case (33) will have the form xij = Pβ∈Jr, m{i}(cid:12)(cid:12)(cid:12)(cid:16)(AW)k+2 (AW)k+2 β . i Pβ∈Jr, m(cid:12)(cid:12)(cid:12) , β(cid:12)(cid:12)(cid:12) (f.j)(cid:17) β β(cid:12)(cid:12)(cid:12) where f.j is the j-th column of F = ¯VD = (AW)kAD. Now, consider the following restricted matrix equation, XWAW = D, (35) (36) Rl(X) ⊂ Rl(cid:0)(AW)k(cid:1) , Nr(X) ⊃ Nr(cid:0)(WA)k(cid:1) , with k = max {Ind(AW), Ind (WA)}, and where A ∈ Hm×n, W ∈ Hn×m D ∈ Hq×m. r1 Theorem 3.2 If D ⊂ Rl(cid:0)(AW)k(cid:1) and D ⊃ Nr(cid:0)(WA)k(cid:1), then the restricted matrix equation (35) has a unique solution, X = DAd,W , (37) which possess the following determinantal representations for i = 1, q, j = 1, n, i) xij = m Pl=1 Pα∈Ir, m{t} rdetl(cid:16)(cid:16)V2k+1(cid:0)V2k+1(cid:1)∗(cid:17)l. (di .)(cid:17) α Pα∈Ir, m(cid:12)(cid:12)(cid:0)V2k+1 (V2k+1)∗(cid:1) α α Pα∈Ir1 , n{j} α(cid:12)(cid:12) Pα∈Ir1 , n rdetj(cid:16)(WW∗)j . ( wl.)(cid:17) α α (WW∗) α α (38) where di. is the i-th row of D = D V = DVk(V2k+1)∗, V = AW and r = rank(AW)k+1 = rank(AW)k. ii) n xij = lit(uD tj)(2), (39) where (uD qj )(2) can be obtained by (15) and DA = L = (lit) ∈ Hq×n. iii) If AW ∈ Hm×m is Hermitian, then xij = Pα∈Ir,n{j} α rdetj(cid:0)(WA)k+2 j . (gi. )(cid:1) α Pα∈Ir, n(cid:12)(cid:12)(cid:12) α(cid:12)(cid:12)(cid:12) (WA)k+2 α where gi. is the i-th row of G = DA(WA)k for all i = 1, n. . (40) Xt=1 12 Proof. The proof is similar to the proof of Theorem 3.1. (cid:3) Remark 3.2 In the complex case, i.e. A ∈ Cm×n, W ∈ Cn×m , and D ∈ Cn×p, we substitute usual determinants for all corresponding row and column determinants in (38), (39), and (40). Herein the condition WA ∈ Cn×n be Hermitian is not necessary, then in the complex case (40) can be represented as follows, r1 xij = Pα∈Ir,n{j}(cid:12)(cid:12)(cid:0)(WA)k+2 (WA)k+2 α . α(cid:12)(cid:12) j . (gi. )(cid:1) α α(cid:12)(cid:12)(cid:12) Pα∈Ir, n(cid:12)(cid:12)(cid:12) where gi. is the i-th row of G = DA(WA)k for all i = 1, n. Now we consider the matrix equation (26) with the constraints (3.1). Denote ADB =: D = (cid:16) dlf(cid:17) ∈ Hm×q, and ¯VD ¯U =: ¯D = (cid:0) ¯dlf(cid:1) ∈ Hm×q, where ¯V := (AW1)k1 A , ¯U := B(W2B)k2 . Theorem 3.3 Suppose D ∈ Hn×p, A ∈ Hm×n, W1 ∈ Hn×m with k1 = max {Ind(AW1), Ind (W1A)}, where rank (AW1)k1 = s1, and B ∈ Hp×q, W2 ∈ Hq×p r2 with k2 = max {Ind(BW2), Ind (W2B)}, rank (BW2)k2 = s2. If D ∈ Rr(cid:0)(W1A)k1 , (W2B)k2(cid:1), D ∈ Rl(cid:0)(AW1)k1 , (BW2)k2(cid:1), then the re- stricted matrix equation (26) has a unique solution, r1 X = Ad,W1 DBd,W2, (41) which possess the following determinantal representations for all i = 1, m, j = 1, q. i) xij = m Xl=1 q (vD il )(2) dlf (uD f j)(2), Xf =1 (42) where (vD obtained by (17), and (uD (uD il ) = VD is the Drazin inverse of V = AW1 and (vD il )(2) can be f j) = UD is the Drazin inverse of U = W2B and qj)(2) can be obtained by (15). ii) If AW1 ∈ Hm×m and W2B ∈ Hq×q are Hermitian, then xij = xij = or where . j = dB  Xα∈Is2 ,q{j} . j(cid:1)(cid:17) β (cid:0)dB β (W2B)k2+2 α (dA i .)(cid:17) α α (W2B)k2+2 α α(cid:12)(cid:12)(cid:12) α(cid:12)(cid:12)(cid:12) . i (AW1)k1+2 β Pβ∈Js1, m{i} Pβ∈Js1, m(cid:12)(cid:12)(cid:12) Pα∈Is2 ,q{j} Pβ∈Js1, m(cid:12)(cid:12)(cid:12) rdetj(cid:16)(W2B)k2+2 cdeti(cid:16)(AW1)k1+2 β(cid:12)(cid:12)(cid:12) Pα∈Is2 , q(cid:12)(cid:12)(cid:12) rdetj(cid:16)(W2B)k2+2 β(cid:12)(cid:12)(cid:12) Pα∈Is2 , q(cid:12)(cid:12)(cid:12) α (¯dt.)(cid:17)α  (AW1)k1+2 β j . j . 13 , , (43) (44) ∈ Hn×1, t = 1, n (45) i . = dA  Xβ∈Js1, m{i} cdeti(cid:16)(AW1)k1+2 . i β (cid:0)¯d. l(cid:1)(cid:17) β  ∈ H1×q, are the column vector and the row vector, respectively. ¯di. and ¯d.j are the i-th row and the j-th column of ¯D for all i = 1, n, j = 1, p. l = 1, q (46) Proof. The existence and uniqueness of the solution (41) can be proved similar as in ( [28], Theorem 5.2). To establish a Cramer's rule of (26) we note that we shall not use the deter- minantal representations (22) and (22) for (41) because corresponding determi- nantal representations of it's solution will be too cumbersome. To derive a Cramer's rule (42) we use the sentence (a) from Lemma 2.1. Then we obtain X =(cid:0)(AW1)D)(cid:1)2 ADB(cid:0)(W2B)D)(cid:1)2 . (47) Denote ADB =: D = (cid:16) dlf(cid:17) ∈ Hm×q, V := AW1, and U := W2B. Then the equation (47) will be written component-wise as follows xij = p n Xs=1 Xt=1 (ad,W 1 it )dts(bd,W2 sj ) = p Xs=1 n Xt=1 m Xl=1 (vD il )(2)alt! dts  q Xf =1 bsf (uD f j)(2)  By changing the order of summation, from here it follows (42). ii) If A ∈ Hm×n and AW1 ∈ Hm×m and W2B ∈ Hq×q are Hermitian, then by Theorems 2.9 and 2.10 the W-weighted Drazin inverses , B ∈ Hp×q r2 r1 Ad,W1 = (cid:16)ad,W1 ij determinantal representations respectively, ij (cid:17) ∈ Hm×n and Bd,W2 = (cid:16)bd,W2 ij = Pβ∈Js1 , m{i} ad,W1 (cid:17) ∈ Hq×p posses the following , (48) where ¯v.j is the j-th column of ¯V = (AW1)k1 A for all j = 1, m; β . i (AW1)k1+2 β cdeti(cid:16)(AW1)k1+2 (¯v.j)(cid:17) β Pβ∈Jr, m(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) (¯ui. )(cid:17) α rdetj(cid:16)(W2B)k2+2 Pα∈Is2 , q(cid:12)(cid:12)(cid:12) α(cid:12)(cid:12)(cid:12) (W2B)k2+2 α j . α , ij = Pα∈Is2 ,q{j} bd,W2 where ¯ui. is the i-th row of ¯U = B(W2B)k2 for all i = 1, p. By component-wise writing (41) we obtain, xij = p Xs=1 n Xt=1 ad,W 1 it dts! · bd,W2 sj 14 (49) (50) Denote by d.s the s-th column of ¯VD = (AW1)k1 AD =: D = ( dij ) ∈ Hm×p ¯v. tdts = d. s that for all s = 1, p. It follows from Pt Pβ∈Js1, m{i} ad,W 1 it dts = n n Xt=1 Xt=1 Pβ∈Js1 , m{i} β n . i . i · dts = β · dts (AW1)k1+2 β cdeti(cid:16)(AW1)k1+2 (¯v.t)(cid:17) β Pβ∈Js1, m(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) (¯v.t)(cid:17) β cdeti(cid:16)(AW1)k1+2 Pt=1 Pβ∈Js1, m(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) cdeti(cid:16)(AW1)k1+2 (cid:16)d. s(cid:17)(cid:17) β Pβ∈Js1, m(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) Pβ∈Js1 , m{i} (AW1)k1+2 β (AW1)k1+2 β = . i β (51) Suppose es. and e. s are respectively the unit row-vector and the unit column- vector whose components are 0, except the s-th components, which are 1. Sub- stituting (51) and (49) in (50), we obtain xij = Pβ∈Js1, m{i} p Xs=1 Since β . i (cid:16)d. s(cid:17)(cid:17) β cdeti(cid:16)(AW1)k1+2 Pβ∈Js1 , m(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) (AW1)k1+2 β d. s = n Xt=1 e. t dts, ¯us. = ¯uslel., q Xl=1 Pα∈Is2 ,q{j} . α j . (W2B)k2+2 α rdetj(cid:16)(W2B)k2+2 (¯us. )(cid:17) α Pα∈Is2 , q(cid:12)(cid:12)(cid:12) α(cid:12)(cid:12)(cid:12) Xs=1 dts ¯usl = ¯dtl, (52) p then we have xij = p n q Ps=1 Pt=1 Pl=1 Pβ∈Js1 , m{i} n q Pt=1 Pl=1 Pβ∈Js1 , m{i} (el.)(cid:17) α α = (W2B)k2+2 α β . i (AW1)k1+2 β cdeti(cid:16)(AW1)k1+2 Pβ∈Js1, m(cid:12)(cid:12)(cid:12) cdeti(cid:16)(AW1)k1+2 Pβ∈Js1, m(cid:12)(cid:12)(cid:12) (e. t)(cid:17) β dts ¯usl Pα∈Is2 ,q{j} β(cid:12)(cid:12)(cid:12) Pα∈Is2 , q(cid:12)(cid:12)(cid:12) (e. t)(cid:17) β ¯dtl Pα∈Is2 ,q{j} β(cid:12)(cid:12)(cid:12) Pα∈Is2 , q(cid:12)(cid:12)(cid:12) (AW1)k1+2 β (W2B)k2+2 α . i β α(cid:12)(cid:12)(cid:12) j . rdetj(cid:16)(W2B)k2+2 α(cid:12)(cid:12)(cid:12) rdetj(cid:16)(W2B)k2+2 j . (el.)(cid:17) α α . (53) 15 Denote by dA il := Xβ∈Js1, m{i} cdeti(cid:16)(AW1)k1+2 . i β = (cid:0)¯d. l(cid:1)(cid:17) β n Xt=1 Xβ∈Js1, m{i} cdeti(cid:16)(AW1)k1+2 . i (e. t)(cid:17) β β ¯dtl the l-th component of a row-vector dA ing it in (53), we have i . = (dA i1, ..., dA iq) for all l = 1, q. Substitut- q dA Pl=1 Pβ∈Js1, m(cid:12)(cid:12)(cid:12) il Pα∈Is2 ,q{j} (AW1)k1+2 β j . rdetj(cid:16)(W2B)k2+2 β(cid:12)(cid:12)(cid:12) Pα∈Is2 , q(cid:12)(cid:12)(cid:12) α (el.)(cid:17) α α(cid:12)(cid:12)(cid:12) (W2B)k2+2 α . xij = q Pl=1 If we denote by Since il el. = dA dA i ., then it follows (44). dB tj := q Xl=1 ¯dtl Xα∈Is2 ,q{j} rdetj(cid:16)(W2B)k2+2 j . (el.)(cid:17)α α = Xα∈Is2 ,q{j} rdetj(cid:16)(W2B)k2+2 j . (¯dt.)(cid:17)α α (54) the t-th component of a column-vector dB substituting it in (53), we obtain . j = (dB 1j, ..., dB nj)T for all t = 1, n and xij = n Pt=1 Pβ∈Js1, m{i} Pβ∈Jr1, n(cid:12)(cid:12)(cid:12) . i cdeti(cid:16)(AW1)k1+2 β(cid:12)(cid:12)(cid:12) Pα∈Ir2 ,p (A∗A) β β dB tj (e. t)(cid:17) β . (BB∗)α α Since n Pt=1 e.tdB tj = dB . j, then it follows (43). (cid:3) 4 Examples In this section, we give examples to illustrate our results. 1. Let us consider the matrix equation WAWX = D (55) with the restricted conditions (29), where i 1 0 0 0 i k 1 0 1 −k −j A =    , W =  k −j 0 16 0 i k 0 1 0 −k 0 1  , D =  i k i −j 1 −i  . Then V = AW =  −k −1 − j k −j i + k 0 −i + k 1 − j 0 j i i i 1 + j 0 i − k   , U = WA =  j 0 i 0 k 0 0 0 0  , and rank W = 3, rank V = 3, rank V3 = rank V2 = 2, rank U2 = rank U = 2. So, Ind V = 2, Ind U = 1, and k = max{Ind(AW), Ind(WA)} = 2. We shall find the W-weighted Drazin inverse solution of (55) by it's deter- minantal representation (31). We have −1 i + k 0 0 0 −1 0 U2 =  −2i − 3k 0 0 0  , 14 0 0 0 1 k 0 i 0 0 2 + 3j 2i + 3k 0  , U5 =  0 U5 =  ,(cid:0)U5(cid:1)∗   , W∗ = 0   , W = W∗U2 =    −j 0 j −k 1 − 2j i + k 0 1 + j i −1 0 −k j 0 −k −i 0 1 0 , 0 1 0 k 0  ,     0 0 0 0 . i − j − k 1 + 3i + 6j − 2k 4i − 2k (U5)∗ =  −i 0 2 − 3j −k 0 0 D = (U5)∗U2D =  W∗W =  Since by (31) 2 −i j −j i −j 2 0 2k 0 −2k 1 0 0 2 x11 = 3 Pt=1 Pβ∈I3, 4{1} where cdet1 ((W∗W).1 ( w.t)) β cdett(cid:16)(cid:16)(cid:0)U5(cid:1)∗ β Pβ∈J2, 3{t} β(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) Pβ∈J2, 3(cid:12)(cid:12)(cid:12)(cid:0)(U5)∗ U5(cid:1) β (W∗W) β Pβ∈J3, 4(cid:12)(cid:12)(cid:12) U5(cid:17).t (d.1)(cid:17) β β , cdet1 ((W∗W).1 ( w.1)) β β = Xβ∈I3, 4{1} cdet1  k 0 i i −j 2 0 0  + cdet1 1   k 0 0 j i 2 −2k 2k  + cdet1 1   k −j 1 i 0 0 j 0 2  = 0, cdet1 ((W∗W).1 ( w.2)) β β = −2j, Xβ∈I3, 4{1} cdet1 ((W∗W).1 ( w.3)) β Xβ∈I3, 4{1} β = 0, Xβ∈J3, 4(cid:12)(cid:12)(cid:12) = 2, (W∗W) β β(cid:12)(cid:12)(cid:12) 17 and Xβ∈J2, 3{1} cdet1(cid:18) Xβ∈J2, 3{2} then x11 = x12 = x21 = x22 = x31 = x32 = We finally get, cdet1(cid:16)(cid:16)(cid:0)U5(cid:1)∗ i − j − k U5(cid:17).1 −2i − 3k 1 + 3i + 6j − 2k cdet2(cid:16)(cid:16)(cid:0)U5(cid:1)∗ U5(cid:17).2 cdet3(cid:16)(cid:16)(cid:0)U5(cid:1)∗ Xβ∈J2, 3{3} 0 β = (d.1)(cid:17) β 14 (cid:19) + cdet1(cid:18)i − j − k 0 (d.1)(cid:17) β U5(cid:17).3 (d.1)(cid:17) β β = j, β = 0, Xβ∈J2, 3(cid:12)(cid:12)(cid:12)(cid:16)(cid:0)U5(cid:1)∗ = 1, β(cid:12)(cid:12)(cid:12) U5(cid:17) β 0(cid:19) = −2i − j − k, 0 · (−2i − j − k) + (−2j) · j + 0 · 0 2 · 1 0 · (−2 + 2j) + (−2j) · i + 0 · 0 2 · 1 = k, = 1, 2j · (−2i − j − k) + (10i − 4k) · j + 0 · 0 2j · (−2 + 2j) + (10i − 4k) · i + 0 · 0 2 · 1 2 · 1 10i · (−2i − j − k) + j · j + 0 · 0 2 · 1 10i · (−2 + 2j) + j · i + 0 · 0 2 · 1 = −10i + 9.5k, = 1 + i + 7k, = −7 − 4j, = 9.5 + 5j − 5k, 1 X = 9.5 + 5j − 5k −10i + 9.5k  .  1 + i + 7k −7 − 4j k 2. Let now we consider the matrix equation with the constraints (3.1), where W1AW1XW2BW2 = D, (56) (57) , W2 =  k −i 0 j 0 1   , k −j 0 A =  0 i k 0 1 0 −k 0 1 B =(cid:18)k j   k −j 0 1 0 k 0 0 i 1 −k 0  , W1 =  0 1(cid:19) , D =  0 j i −1 0 k 0 j −1 0   0  , U = W2B =  . 18 Since the following matrices are Hermitian V = AW1 =  i −2 0 −i −1 0 0 0 0 −i −i i −1 0 i 0 −1  , then we can find the W-weighted Drazin inverse solution of (57) by it's deter- minantal representation (43). We have k1 = max {Ind(AW1), Ind (W1A)} = 1, k2 = max {Ind(BW2), Ind (W2B)} = 1, and s1 = rank (AW1) = 2, s2 = rank (W2B) = 2. Since then We have 0 0 8i (AW1)3 β −13 0 −8i −5 0 (AW1)3 =  Xβ∈J2, 3(cid:12)(cid:12)(cid:12) β(cid:12)(cid:12)(cid:12) ¯D = AW1ADBW2B =  0 −3i −3i 3i −3 3i 0 0 3   , 0  , (W2B)3 =  = 1, Xα∈I2, 3(cid:12)(cid:12)(cid:12) α(cid:12)(cid:12)(cid:12) (W2B)3 α = −27. 2i + j −7 + k −5 + 2k −1 + k −5i − j −4i − 2j 0 0 0   , By (45), we can get 36i − 9j −27 − 9k dB .1 =  Since (AW1)3 0  .2 =  , dB  . 1(cid:1) = . 1(cid:0)dB  cdet1(cid:16)(AW1)3 β(cid:12)(cid:12)(cid:12) Pα∈I2, 3(cid:12)(cid:12)(cid:12) (AW1)3 β then finally we obtain Similarly, x11 = Pβ∈J2, 3{1} Pβ∈J2, 3(cid:12)(cid:12)(cid:12) cdet1(cid:18) −27 cdet2(cid:18)−13 −27 x12 = x21 = = 8i 1 3 36i − 9j −18i −5(cid:19) −8i −27 − 9k(cid:19) cdet2(cid:18)−13 −9 − 9k 9i + 3j (cid:19) −27 −8i = −27 x23 = 3 = 19 −27 −18i 0   , dB 9 − 9k 9i + 3j 0   . .3 =  4  , 0 36i − 9j 0 −27 − 9k −5 0 8i 0 36i − 27j −27 = −4i + 3j 3 , . 1(cid:0)dB (W2B)3 α β = . 1(cid:1)(cid:17) β α(cid:12)(cid:12)(cid:12) cdet1(cid:18) 9 − 9k , x13 = −7 − 5k , x22 = −9 − 7k , 8i −27 9i − 3j −5(cid:19) cdet2(cid:18)−13 −27 −8i −18i(cid:19) = 9 −27 = −2i 3 , 15i − 11j 9 , x31 = x32 = x33 = 0. So, the W-weighted Drazin inverse solution of (57) are X = −9 − 7k −12i + 9j −21 − 15k −6i 15i − 11j 3 0 0 0 1 9    . Note that we used Maple with the package CLIFFORD in the calculations. References [1] S. L. Adler, Quaternionic Quantum Mechanics and Quantum Fields. Ox- ford University Press, New York, 1995. [2] J.D. Gibbon, A quaternionic structure in the three-dimensional Euler and ideal magnetohydrodynamics equation. Physica D 166 (2002) 17-28. [3] J.D. Gibbon, D.D. Holm, R.M. Kerr, I. Roulstone, Quaternions and particle dynamics in the Euler fluid equations. Nonlinearity 19 (2006) 1969 -- 1983. [4] A. Handson, H. Hui, Quaternion frame approach to streamline visualiza- tion. IEEE Tran. Vis. Comput. Grap. 1 (1995) 164-172. [5] B.L.Stevens, F.L.Lewis, E.N.Johnson, Aircraft Control and Simulation: Dynamics, Controls Design, and Autonomous Systems. 3rd Edition, Wiley, 2015. [6] A. Perez, J. M. McCarthy, Dual quaternion synthesis of constrained robotic systems. J. Mech. Des. (2003) 126(3) 425 -- 435. [7] T. Isokawa, T. Kusakabe, N. Matsui, F. Peper, Quaternion neural network and its application. Lecture Notes in Computer Science 2774 (2003) 318- 324. [8] J. Proskova, Description of protein secondary structure using dual quater- nions. Journal of Molecular Structure 1076 (2014) 89-93. [9] S.F. Yuan, Q.W. Wang, X.F. Duan, On solutions of the quaternion matrix equation AX = B and their applications in color image restoration. Appl. Math. Comput. 221 (2013) 10 -- 20. [10] Q. W. Wang, S. W. Yu, Extreme ranks of real matrices in solution of the quaternion matrix equation AXB = C with applications. Algebra Colloq. 17(2) (2010) 345 -- 360. [11] S. Yuan, A. Liao, Y. Lei, Least squares Hermitian solution of the matrix equation (AXB, CXD) = (E, F ) with the least norm over the skew field of quaternions. Math. Comput. Model. 48 (2008) 91 -- 100. [12] L. G. Feng, W. Cheng, The solution set to the quaternion matrix equation AX − ¯XB = 0. Algebra Colloq. 19(1) (2012) 175 -- 180. [13] T. S. Jiang, M. S. Wei, On a solution of the quaternion matrix equation X − A XB = C. Acta Math. Sin. (Engl. Ser.) 21(3) (2005) 483-490. 20 [14] C. Q. Song, G. L. Chen, X. D. Wang, On solutions of quaternion matrix equations XF − AX = BY and XF − A X = BY . Acta Math. Sci. 32(5) (2012) 1967 -- 1982. [15] S.Yuan, Q. W. Wang, Two special kinds of least squares solutions for the quaternion matrix equation AXB +CXD = E. Electron. J. Linear Algebra 23 (2012) 57 -- 74. [16] F. Zhang, M. Wei, Y. Lia, J. Zhao, Special least squares solutions of the quaternion matrix equation AX=B with applications. Appl. Math. Comput. 270 (2015) 425 -- 433. [17] R.E. Cline, T.N.E. Greville, A Drazin inverse for rectangular matrices. Linear Algebra Appl. 29 (1980) 53 -- 62. [18] S. L. Campbell, C. D. Meyer Jr., N. J. Rose, Applications of the Drazin inverse to linear systems of differential equations with singular constant coefficients SIAM J. Appl. Math. 31 (1976) 411-425. [19] S.L. Campbell, Comments on 2-D descriptor systems. Automatica 27(1) (1991) 189 -- 192. [20] S. L. Campbell, C.D. Meyer, Generalized inverse of linear transformations. Corrected reprint of the 1979 original. Dover Publications, Inc., New York, 1991. [21] D. J. Spitzner, T. R Boucher, Asymptotic variance of functionals of discrete-time Markov chains via the Drazin inverse. Elect. Comm. in Probab. 12 (2007) 120 -- 133. [22] T. Kaczorek, Application of the Drazin inverse to the analysis of descriptor fractional discrete -- time linear systems with regular pencils. Int. J. Appl. Math. Comput. Sci. 23(1) (2013) 29 -- 33. [23] Z. Al-Zhour, A. Kili¸cman, M. H. Abu Hassa, New representations for weighted Drazin inverse of matrices. Int. Journal of Math. Analysis 1(15) (2007) 697 -- 708. [24] M. Nikuie, M. Z. Ahmad, New Results on the W-weighted Drazin Inverse. AIP Conference Proceedings 1602 (2014) 157. [25] M. Nikuie, Singular fuzzy linear systems. App. Math. and Comp. Intel. 2(2) (2013) 157-168. [26] G. Wang, J. Sun. A Cramer rule for solution of the general restricted matrix equation. Appl. Math. Comput. 154 (2004) 415 -- 422. [27] Y. Wei. A characterization for the W-weighted Drazin inverse and a Cramer rule for the W-weighted Drazin inverse solution. Appl. Math. Comput. 125 (2002) 303 -- 310. [28] G. Song, Characterization of the W-weighted Drazin inverse over the quaternion skew field with applications. Electron. J. Linear Algebra 26 (2013) 1 -- 14. 21 [29] I. Kyrchei, Determinantal representations of the W-weighted Drazin inverse over the quaternion skew field. Appl. Math. Comput. 264 (2015) 453 -- 465. [30] I. Kyrchei, Cramer's rule for quaternion systems of linear equations. J. Math. Sci. 155(6) (2008) 839 -- 858. [31] I. Kyrchei, The theory of the column and row determinants in a quaternion linear algebra. In: Albert R. Baswell (Eds.), Advances in Mathematics Research 15, Nova Sci. Publ., New York, pp. 301 -- 359, 2012. [32] I. Kyrchei, Cramer's rule for some quaternion matrix equations. Appl. Math. Comput. 217(5) (2010) 2024 -- 2030. [33] I. Kyrchei, Determinantal representations of the Moore-Penrose inverse over the quaternion skew field and corresponding Cramer's rules. Linear Multilinear A. 59 (2011) 413 -- 431. [34] I. Kyrchei, Determinantal representation of the Moore -- Penrose inverse ma- trix over the quaternion skew field. J. Math. Sci. 180(1) (2012) 23 -- 33. [35] I. Kyrchei, Explicit representation formulas for the minimum norm least squares solutions of some quaternion matrix equations. Linear Algebra Appl. 438 (2013) 136 -- 152. [36] I. Kyrchei, Determinantal representations of the Drazin inverse over the quaternion skew field with applications to some matrix equations. Appl. Math. Comput. 238 (2014) 193 -- 207. [37] G. Song, Determinantal representation of the generalized inverses over the quaternion skew field with applications. Appl. Math. Comput. 219 (2012) 656 -- 667. [38] G. Song, Bott-Duffin inverse over the quaternion skew field with applica- tions. J. Appl. Math. Comput. 41 (2013) 377 -- 392. [39] G. Song, Q. Wang, H. Chang, Cramer rule for the unique solution of re- stricted matrix equations over the quaternion skew field. Comput. Math. Appl. 61 (2011) 1576 -- 1589. [40] G. Song, X. Wang, X. Zhang, On solutions of the generalized Stein quater- nion matrix equation. J. Appl. Math. Comput. 43 (2013) 115 -- 131. [41] V. Rakocevi´c, Y. Wei, A weighted Drazin inverse and applications. Linear Algebra Appl. 350 (2002) 25 -- 39. [42] Y. Wei. Integral representation of the W-weighted Drazin inverse. Appl. Math. Comput. 144 (2003) 3 -- 10. 22
0812.0080	3	0812	2013-03-13T18:08:31	$\omega$-Lie algebras	[ "math.RA" ]	We study a certain generalization of Lie algebras where the Jacobian of three elements does not vanish but is equal to an expression depending on a skew-symmetric bilinear form.	math.RA	math	ω-LIE ALGEBRAS PASHA ZUSMANOVICH Abstract. We study a certain generalization of Lie algebras where the Jacobian of three elements does not vanish but is equal to an expression depending on a skew-symmetric bilinear form. Introduction An anticommutative algebra L with multiplication [ · , · ] over a field K is called an ω-Lie algebra if there is a bilinear form ω : L × L → K such that (1) [[x, y], z] + [[z, x], y] + [[y, z], x] = ω(x, y)z + ω(z, x)y + ω(y, z)x for any x, y, z ∈ L. We will refer to this identity as the ω-Jacobi identity. These algebras were introduced by Nurowski in a recent interesting paper [N]†. Nurowski was motivated by some physical considerations, but our treatment here is a purely mathe- matical one. ω-Lie algebras are obvious generalizations of Lie algebras, the latter corresponding to the case ω = 0. It follows immediately from the definition that ω is skew-symmetric. As noted in [N], there are no 1- and 2-dimensional ω-Lie algebras which are not Lie algebras. Nurowski exhibited nontrivial examples of 3-dimensional ω-Lie algebras (actually, he fully classified them over the field of real numbers). It seems that no structures like this were studied before. Of course, altered Jacobi identities appeared previously in the literature, the closest things we are aware of are, first, algebras studied by Sagle in a series of papers started in the 1960s (see, for example, [S] and references therein), second, structures which, as we suspect, started to appear a long time ago in the literature (see, for example, [L]), and recently were advertised and systematically studied by Hartwig, Larsson and Silvestrov in [HLS] under the name of Hom-Lie algebras, and, third, L∞-algebras and their relatives. Sagle's algebras are obtained by taking the direct sum decomposition L = H ⊕ M of a Lie algebra L, where H is a subalgebra, [H, M] ⊆ M, and defining a new algebra structure on H as the projection of the Lie bracket on it. Such algebras satisfy the condition [[x, y], z] + [[z, x], y] + [[y, z], x] = [h(x, y), z] + [h(z, x), y] + [h(y, z), x] where h : H × H → M is the projection of the Lie bracket on M. Hom-Lie algebras satisfy the condition [[x, y], z] + [[z, x], y] + [[y, z], x] = [[x, y], σ(z)] + [[z, x], σ(y)] + [[y, z], σ(x)] where σ : L → L is a linear map. In both of these cases, the Jacobi identity is altered by maps to the underlying algebra, while the ω-Jacobi identity is altered by the map ω to the ground field, so their similarity is probably too superficial. In a sense, the ω-Jacobi identity should be much more restrictive. Date: last revised November 12, 2012. J. Geom. Phys. 60 (2010), 1028 -- 1044; arXiv:0812.0080. † In [N], the term ω-deformed Lie algebra was used. We find this term somewhat misleading, as these are not deformations of Lie algebras in the usual strict sense. See, however, Question 2 in §11. 1 2 PASHA ZUSMANOVICH L∞-algebras are much more general structures encompassing the notion of a (co)chain complex and a Lie algebra. The Jacobi identity in these structures is valid "up to homotopy" (see for example, the conditions defining the so-called 2-term L∞-algebras in [BC, Lemma 4.3.3], especially the condition (g)). However, a tedious but straightforward computation which we will omit here, shows that this "homotopy", in general, cannot take the form of the right-hand side of (1), so ω-Lie algebras cannot serve as initial terms of L∞-algebras, except for some degenerate trivial cases. Unlike most of the classes of algebras studied, the ω-Jacobi identity does not single out a variety of algebras. In fact, the class of ω-Lie algebras is not closed under the usual construc- tions employed in structural theory of algebras, such as taking the direct sum or tensoring with commutative associative algebra. (It is however closed under taking subalgebras and quotients. The first fact is obvious, the second one is not and comes after a bit of additional work, as shown below). Moreover, the ω-Jacobi identity suggests that any ω-Lie algebra with nontrivial ω should be close to a perfect one (L = [L, L]), thus largely excluding phenomena related to nilpotency and solvability. Such algebras cannot be graded with a large number of graded components, so an analog of the root space decomposition with respect to a Cartan subalgebra, if it exists, should have properties different from the Lie-algebraic case. The main result of this paper roughly says: finite-dimensional ω-Lie algebras which are not Lie algebras are either low-dimensional, or possess a very "degenerate" structure -- in particular, have an abelian subalgebra of small codimension with further restrictive condi- tions. In the first two short sections of this paper we observe some elementary, but useful facts about ground field extension and modules over ω-Lie algebras, needed in subsequent sections. In §3 we establish a sort of analog of the ω-Jacobi identity in 4 variables (Lemma 3.3) which will serve as our main working tool, and establish with its help some auxiliary facts about ideals of ω-Lie algebras. §4 contains a treatment of a rudimentary analog of the root space decomposition. §5 contains results about quasi-ideals, and establishes a preliminary division of finite-dimensional ω-Lie algebras (Lemma 5.4) into the following three classes: those having a Lie subalgebra of codimension 1, those having Ker ω of codimension 2, and those of the form of an abelian extension of a simple ω-Lie algebra with a non-degenerate ω. The next three sections contain treatments of these three classes. Though we are unable to achieve a complete classification (and doubt a reasonable classification exists), we show that all ω-Lie algebras under consideration are "degenerate" in the sense that they contain an abelian subalgebra of a small codimension. In §8 we prove that ω-Lie algebras with a nondegenerate ω do not exist, thus completing the classification. §9 contains formulations of two main theorems, which describe the structure of finite-dimensional ω-Lie algebras, and claim that there are no semisimple ω-Lie algebras (which are not Lie algebras) in high dimensions. In §10 we discuss what identities may be satisfied by ω-Lie algebras, and the last §11 contains some further questions and speculations. Appendix contains description of the GAP code used in analysis of low-dimensional algebras in §8. We note that in the course of the study of ω-Lie algebras many notions in the Lie algebras theory -- derivations, second cohomology, quasi-ideals -- arise naturally. Notation and conventions The ground field K is assumed to be an arbitrary field of characteristic different from 2 and 3, unless stated otherwise. Our terminology concerning bilinear forms is standard. Let ω be a skew-symmetric bilinear form on a linear space V . A subspace W ⊆ V is called isotropic if ω(W, W ) = 0. Let ω-LIE ALGEBRAS 3 W ⊥ = {x ∈ V ω(x, W ) = 0} denote the orthogonal complement to a subspace W . Let Ker ω = V ⊥ denote the kernel of ω. For the standard results from linear algebra we use, see, for example, [B]. The Lie-algebraic notions that do not involve the form ω in their definitions are extended verbatim to ω-Lie algebras: for example, we speak about commutators, commutant, adjoint endomorphisms (which are right multiplications), subalgebras, ideals, simple, semisimple and abelian algebras, and nilpotent elements. 1. Extension of the ground field Sometimes in the subsequent reasonings, we, naturally, would like to have the luxury to work over an algebraically closed ground field. For this, one should to be sure that the property of being an ω-Lie algebra is preserved under the ground field extension. This is indeed the case, as the following elementary proposition shows. Proposition 1.1. Let L be an algebra over a field K, K ⊂ F a field extension. Then L is an ω-Lie algebra over K for some bilinear form ω on L if and only if L ⊗K F is an Ω-Lie algebra over F for some bilinear form Ω on L ⊗K F . Proof. The "only if" part is obvious: L ⊗K F is an Ω-Lie algebra where Ω is a bilinear form on L ⊗K F extended from ω by linearity. To see the validity of the "if" part, note that if dim L ≤ 2, the statement is trivially true (both L and L ⊗K F are Lie algebras and both ω and Ω can be chosen arbitrary), and assume dim L ≥ 3. Take any linearly independent elements x, y, z ∈ L, and apply the ω-Jacobi identity to the triple x ⊗ 1, y ⊗ 1, z ⊗ 1 ∈ L ⊗F K. Then the left-hand side of the ω-Jacobi identity lies in L ⊗ 1, hence all coefficients on the right-hand side belong to K. Hence Ω(L ⊗ 1, L ⊗ 1) ⊆ K, and we may take ω to be a restriction of Ω to L ⊗ 1. (cid:3) 2. Modules Let L be an ω-Lie algebra. Consider a vector space M over K and a linear homomorphism ϕ : L → End(M). It is natural to assume that M is an L-module, if the semidirect product L ⊕ M, with multiplication extended from L by [x, m] = ϕ(x)m, x ∈ L, m ∈ M, and [M, M] = 0, and a skew-symmetric bilinear form Ω extended from L, is an Ω-Lie algebra. One immediately sees that, provided dim L ≥ 2, this is the case if and only if (2) ϕ([x, y])m = ϕ(x)ϕ(y)m − ϕ(y)ϕ(x)m + ω(x, y)m for any x, y ∈ L, m ∈ M, and Ω that trivially extends ω: M ⊆ Ker Ω. This suggests the following Definition. A vector space M is called a module over an ω-Lie algebra L, if there exists a homomorphism ϕ : L → End(M) such that (2) holds. Note that the very existence of a module over an ω-Lie algebra could impose severe restrictions on it. For example, consider the case of a 1-dimensional module M = Km. Then (3) ϕ(x) = λ(x)m for some linear form λ : L → K, any two endomorphisms ϕ(x), where x ∈ L, commute, and (2) reduces to (4) ω(x, y) = λ([x, y]). This is an important case we will encounter below, so it deserves a special 4 PASHA ZUSMANOVICH Definition. An ω-Lie algebra is called multiplicative if there is a linear form λ : L → K such that (4) holds, i.e., [[x, y], z] + [[z, x], y] + [[y, z], x] = λ([x, y])z + λ([z, x])y + λ([y, z])x for any x, y, z ∈ L. So, the previous observation could be rephrased as Lemma 2.1. An ω-Lie algebra L has a 1-dimensional module if and only if L is multiplica- tive, in which case the module structure is given by (3). Note that, unless L is a Lie algebra, L is not a module over itself under the adjoint action. As in the case of Lie algebras, we may consider extensions of an ω-Lie algebra L by means of an L-module M: (5) 0 → M → E → L → 0 where M is considered as an abelian algebra, and ω is extended from L to E trivially by putting ω(M, L) = 0. In what follows, we will need only the following case which we distinguish by the following Definition. An abelian extension of an ω-Lie algebra L is an extension of L by the direct sum of several copies of a 1-dimensional L-module†. Given an ω-Lie algebra and an L-module M, one may define cohomology groups H n(L, M) precisely by the same formula for the differential as for ordinary Lie algebras. Direct, but tedious calculation shows that the square of the differential is zero, so this cohomology is well-defined. As in the case of Lie algebras, direct verification shows that H 2(L, M) describes nonequivalent classes of extensions of kind (5), and, consequently, abelian extensions of L are described by the direct sum of an appropriate number of copies of H 2(L, K), where K is understood as a 1-dimensional L-module. 3. Ideals In this section we show that ideals of non-Lie ω-Lie algebras are either "large", or have a very simple structure. Lemma 3.1. Let I be a proper ideal of an ω-Lie algebra L. Then ω(I, I) = 0. If, addition- ally, I is of codimension > 1, then I ⊆ Ker ω. Proof. Apply the ω-Jacobi identity to x, y ∈ I and z /∈ I, z 6= 0. All the terms on the left-hand side belong to I, and the terms ω(z, x)y and ω(y, z)x on the right-hand side also belong to I. Hence the remaining term ω(x, y)z belongs to I. Hence, ω(x, y) = 0 for any x, y ∈ I. Now look again at the ω-Jacobi identity with x ∈ I. All the terms on the left-hand side still belong to I, as well as the term ω(y, z)x on the right-hand side. Hence, ω(x, y)z−ω(x, z)y ∈ I for any y, z ∈ L. If codimension of I is > 1, this obviously implies ω(I, V ) = 0 for any subspace V of L complementary to I. Together with ω(I, I) = 0, this implies ω(I, L) = 0. (cid:3) Corollary 3.2. A proper ideal of an ω-Lie algebra is a Lie algebra. † Note that this definition does not match the case of Lie algebras, where any extension of type (5) is called abelian. The closest case in Lie algebras would be central extensions, but the term central is obviously inappropriate here as a 1-dimensional module is necessarily non-trivial in the non-Lie case. I was not imaginative enough to devise a new term. As we consider in this paper only extensions of ω-Lie algebras which are not Lie algebras, this should not lead to confusion. ω-LIE ALGEBRAS 5 Note that condition I ⊆ Ker ω ensures that one can define an induced form ω on the quotient space L/I, which obviously satisfies the ω-Jacobi identity, so a quotient of an ω-Lie algebra by an ideal of codimension > 1 is an ω-Lie algebra. Lemma 3.1 suggests to consider the cases of ideals of codimension 1 and of codimension > 1 separately. The former case will be considered in §6. We continue with the following Lemma, which, together with the ω-Jacobi identity, will be our main tool in deriving properties of ω-Lie algebras. Lemma 3.3. Let L be an ω-Lie algebra. Then, for any x, y, z, t ∈ L, the following holds: (6) ω(z, t)[x, y] + ω(t, y)[x, z] + ω(y, z)[x, t] + ω(x, t)[y, z] + ω(z, x)[y, t] + ω(x, y)[z, t] = dω(t, z, y)x + dω(z, t, x)y + dω(y, x, t)z + dω(x, y, z)t, where dω(x, y, z) = ω([x, y], z) + ω([z, x], y) + ω([y, z], x). Proof. Write the ω-Jacobi identity for triples x, y, [z, t] and [x, y], z, t: [[x, y], [z, t]] + [[[z, t], x], y] − [[[z, t], y], x] = ω(x, y)[z, t] + ω([z, t], x)y + ω(y, [z, t])x [[[x, y], z], t] − [[[x, y], t], z] − [[x, y], [z, t]] = ω([x, y], z)t + ω(t, [x, y])z + ω(z, t)[x, y] and sum the two equalities obtained: (7) [[[z, t], x], y] − [[[z, t], y], x] + [[[x, y], z], t] − [[[x, y], t], z] = ω(x, y)[z, t] + ω([z, t], x)y + ω(y, [z, t])x + ω([x, y], z)t + ω(t, [x, y])z + ω(z, t)[x, y]. Multiply the ω-Jacobi identity for x, y, z by t: (8) [[[x, y], z], t] + [[[z, x], y], t] + [[[y, z], x], t] = ω(x, y)[z, t] + ω(z, x)[y, t] + ω(y, z)[x, t]. Subtract (8) from (7): [[[z, t], x], y] − [[[z, t], y], x] − [[[x, y], t], z] − [[[z, x], y], t] − [[[y, z], x], t] = ω([z, t], x)y + ω(y, [z, t])x + ω([x, y], z)t + ω(t, [x, y])z Perform cyclic permutations of x, y, t in the last equality and sum the three equalities so obtained: + ω(z, t)[x, y] − ω(z, x)[y, t] − ω(y, z)[x, t]. − [[[x, y], t] + [[t, x], y] + [[y, t], x], z] = dω(t, z, y)x + dω(z, t, x)y + dω(y, x, t)z + dω(x, y, z)t − ω(y, z)[x, t] − ω(z, t)[x, y] − ω(x, z)[t, y]. Combining the last equality with the ω-Jacobi identity for x, y, t, we get the equality desired. (cid:3) An alternative way to derive the identity (6), based on superalgebras, is outlined in §10. We need also the following auxiliary technical Lemma 3.4. Let L be an ω-Lie algebra and I a nonzero linear subspace of Ker ω such that for any x, y, z ∈ L, h ∈ I. Then one of the following holds: [ω(y, z)x + ω(z, x)y + ω(x, y)z, h] ∈ Kh (i) L is multiplicative and I is an abelian ideal of L which, as an L/I-module, is iso- morphic to the direct sum of 1-dimensional modules. (ii) I is contained in a Lie subalgebra of L of codimension 1. 6 PASHA ZUSMANOVICH (iii) Ker ω is a Lie subalgebra of L of codimension 2, Ker ω = {x ∈ L [x, h] ∈ Kh} for some h ∈ I, and [[Ker ω, Ker ω], h] = 0. (iv) L is a Lie algebra. Proof. Denote N(h) = {x ∈ L [x, h] ∈ Kh}. Writing the ω-Jacobi identity for x, y ∈ N(h) and h ∈ I, we get: (9) [[x, y], h] = ω(x, y)h. Hence N(h) is a subalgebra of L for any h ∈ I. We have: (10) ω(y, z)x + ω(z, x)y + ω(x, y)z ∈ N(h) for any x, y, z ∈ L and h ∈ I. Letting here z ∈ N(h), we get (11) ω(y, z)x + ω(z, x)y ∈ N(h) for any x, y ∈ L, and letting further y ∈ N(h), we get ω(y, z)x ∈ N(h) for any x ∈ L. The last inclusion implies that either N(h) = L, or ω(N(h), N(h)) = 0. If N(h) = L for all h ∈ I, then [x, h] = λ(x, h)h for any x ∈ L, h ∈ I and some map λ : L × I → K. Obviously I is an ideal of L. By linearity, λ is linear in the first argument and constant in the second one, so we may write λ(x, ·) = λ(x). By (9), ω(x, y) = λ([x, y]) for any x, y ∈ L, so L is multiplicative. As 0 = [h, h] = λ(h)h for any h ∈ I, we have λ(I) = 0 and I is abelian, so we are in case (i). Assume now there is h ∈ I such that ω(N(h), N(h)) = 0, so N(h) is a proper Lie subal- gebra of L. By (11), either N(h) is of codimension 1, or ω(N(h), L) = 0. Let N(h) be of codimension 1. If I ⊆ N(h), we are in (ii). If I * N(h), then L = N(h) + I. But then ω(N(h), N(h)) = 0 and I ⊆ Ker ω imply ω(L, L) = 0, hence L is a Lie algebra and we are in case (iv). If ω(N(h), L) = 0, (10) implies that either N(h) is of codimension 2, or ω(L, L) = 0, i.e. L is a Lie algebra again. So the only case remained to consider is when N(h) is of codimension 2 and lies in Ker ω for some h ∈ I. If L is not a Lie algebra, i.e., Ker ω is proper, then N(h) = Ker ω. By (9), [[N(h), N(h)], h] = 0, and we are in case (ii). (cid:3) Corollary 3.5. Let L be an ω-Lie algebra and I a nonzero ideal of L of codimension > 1. Then the conclusion of Lemma 3.4 holds. Proof. By Lemma 3.1, I ⊆ Ker ω. Write (6) for x, y, z ∈ L, h ∈ I: ω(y, z)[x, h] + ω(z, x)[y, h] + ω(x, y)[z, h] = dω(x, y, z)h. Thus Lemma 3.4 is applicable. (cid:3) 4. Rudimentary root space decomposition We start this section with another application of Lemma 3.3. Lemma 4.1. Let L be an ω-Lie algebra and H an abelian subalgebra of Ker ω of dimension > 1. Then: (i) ω([x, h], y) + ω(x, [y, h]) = 0 (ii) [ω(y, z)x + ω(z, x)y + ω(x, y)z, h] = dω(x, y, z)h for any x, y, z ∈ L, h ∈ H. ω-LIE ALGEBRAS 7 Proof. Write (6) for x, y ∈ L, h, h′ ∈ I: (ω([h′, y], x) + ω([x, h′], y))h + (ω([h, x], y) + ω([y, h], x))h′ = 0. Choosing h and h′ to be linearly independent, we arrive at case (i). Now writing (6) for x, y, z ∈ L, h ∈ I, and taking into account (i), we arrive at case (cid:3) (ii). In particular, (ii) shows that Lemma 3.4 is applicable: Corollary 4.2. Let L be an ω-Lie algebra and I an abelian subalgebra of Ker ω of dimension > 1. Then the conclusion of Lemma 3.4 holds. We see that Ker ω in the ω-Lie algebra satisfies, in general, quite restrictive conditions. However, to treat the cases (ii) and (iii) of Lemma 3.4 in an uniform way, we continue to consider some generalities about Ker ω. The following two Lemmata are analogs of the facts used in the proof of the well-known properties of root space decompositions in Lie algebras. Not surprisingly, they feature very similar inductive proofs involving binomial coefficients. Lemma 4.3. Let L be an ω-Lie algebra, H an abelian Lie subalgebra of Ker ω, and dim H > 1. Then (12) n Xi=0 i(cid:19)ω(cid:16)(ad(h) + α)n−i(x), (ad(h) + β)i(y)(cid:17) = (α + β)nω(x, y) (cid:18)n for any n ∈ N, x, y ∈ L, h ∈ H, α, β ∈ K. Proof. Induction on n. The case n = 1 follows easily from Lemma 4.1(i). Writing (12) for a given n for pairs (ad(h) + α)x, y and x, (ad(h) + β)y and summing the two equalities obtained, we get on the left-hand side: n Xi=0 i(cid:19)ω(cid:16)(ad(h) + α)n+1−i(x), (ad(h) + β)i(y)(cid:17) (cid:18)n n + Xi=0 i(cid:19) +(cid:18) n Xi=0 (cid:16)(cid:18)n i(cid:19)ω(cid:16)(ad(h) + α)n−i(x), (ad(h) + β)i+1(y)(cid:17) (cid:18)n i − 1(cid:19)(cid:17)ω(cid:16)(ad(h) + α)n+1−i(x), (ad(h) + β)i(y)(cid:17) n+1 = = n+1 Xi=0 (cid:18)n + 1 i (cid:19)ω(cid:16)(ad(h) + α)n+1−i(x), (ad(h) + β)i(y)(cid:17) and on the right-hand side: (α + β)nω((ad(h) + α)x, y) + (α + β)nω(x, ((ad(h) + β))y) = (α + β)n+1ω(x, y). This provides the induction step. (cid:3) Lemma 4.4. Under the same conditions as in the previous Lemma, (13) n Xi=0 i(cid:19)h(ad(h) + α)n−i(x), (ad(h) + β)i(y)i (cid:18)n for any n ∈ N, x, y ∈ L, h ∈ H, α, β ∈ K. = (ad(h) + α + β)n([x, y]) − n(α + β)n−1ω(x, y)h 8 PASHA ZUSMANOVICH Proof. Induction on n. The case n = 1 is verified directly using the ω-Jacobi identity for x, y, h and Lemma 4.1(i). The induction step runs as follows. Writing (13) for a given n for pairs (ad(h) + α)x, y and x, (ad(h) + β)y, and summing the two equalities obtained, we get on the left-hand side: n Xi=0 i(cid:19)h(ad(h) + α)n+1−i(x), (ad(h) + β)i(y)i (cid:18)n n + Xi=0 i(cid:19) +(cid:18) n Xi=0 (cid:16)(cid:18)n i(cid:19)h(ad(h) + α)n−i(x), (ad(h) + β)i+1(y)i (cid:18)n i − 1(cid:19)(cid:17)h(ad(h) + α)n+1−i(x), (ad(h) + β)i(y)i n+1 = = n+1 Xi=0 (cid:18)n + 1 i (cid:19)h(ad(h) + α)n+1−i(x), (ad(h) + β)i(y)i and on the right-hand side: ((ad(h) + α + β)n([(ad(h) + α)x, y] + [x, (ad(h) + β)y]) − n(α + β)n−1(ω((ad(h) + α)x, y) + ω(x, (ad(h) + β)y))h = ((ad(h) + α + β)n+1([x, y]) − ω(x, y)((ad(h) + α + β)nh − n(α + β)nω(x, y)h = ((ad(h) + α + β)n+1([x, y]) − (n + 1)(α + β)nω(x, y)h. (cid:3) Assume the ground field K is algebraically closed. As ad(H) is a space of commuting endomorphisms of L, we may consider the root space decomposition of L with respect to ad(H): (14) L = L0 ⊕Mα Lα. As in the Lie algebras case, we will write Lα for any α ∈ H ∗, assuming it being zero if α is not a root. Let Lα = {x ∈ L [x, h] = α(h)x for all h ∈ H} denotes the simple subspace of the root space Lα. Lemma 4.5. Let L be a finite-dimensional ω-Lie algebra over an algebraically closed field, H an abelian subalgebra of Ker ω, dim H > 1, and (14) is the root space decomposition of L with respect to H. Then: (i) ω(Lα, Lβ) = 0 for any two roots α, β such that α + β 6= 0. (ii) [Lα, Lβ] ⊆ Lα+β for any two roots α, β. (iii) If for some nonzero root α, there is a root −α, then either there are no more nonzero roots, or both Lα and L−α lie in Ker ω. (iv) If L0 = H, then H ⊕Lα Lα is a Lie subalgebra of L. Proof. (i) Take x ∈ Lα and y ∈ Lβ. Then (12) implies that for a sufficiently large n and any h ∈ H, Hence ω(Lα, Lβ) = 0 if α + β 6= 0. (ii) In its turn, (13) shows that for a sufficiently large n, (−α(h) − β(h))nω(x, y) = 0. (ad(h) − (α(h) + β(h)))n([x, y]) = n(−α(h) − β(h))n−1ω(x, y)h. The right-hand side here vanishes for any α, β, as by just proved if α + β 6= 0, then ω(x, y) = 0. Hence [Lα, Lβ] ⊆ Lα+β. ω-LIE ALGEBRAS 9 (iii) Suppose there are three distinct nonzero roots α, −α, β, and take x ∈ Lα, y ∈ L−α, z ∈ Lβ. Applying Lemma 4.1(ii), we see that all summands in the corresponding equality, lying in different root spaces, vanish. In particular, [z, h]ω(x, y) = 0. Choosing z to be an eigenvector from the corresponding root space, i.e., [z, h] = β(h)z, we see that ω(Lα, L−α) = 0. Together with (i) this implies that both Lα and L−α lie in Ker ω. y ∈ L−α, h ∈ H, we get ω(x, y) = 0. This shows that ω(Lα, L−α) = 0, which together with (cid:3) (iv) It is clear that H ⊕Lα Lα is a subalgebra. Writing the ω-Jacobi identity for x ∈ Lα, (i) implies that ω vanishes on H ⊕Lα Lα. It is possible to ponder this situation further to get more exotic-looking properties of root systems in non-Lie ω-Lie algebras, but no need in that: after all, this machinery will be applied below only to quite degenerate situations when codimension of H is small. 5. Kernel and quasi-ideals The aim of this section is to show that in nontrivial cases, the form ω should satisfy very strong vanishing conditions, and establish a preliminary classification of ω-Lie algebras. Lemma 5.1. Let L be a finite-dimensional ω-Lie algebra, and x, y ∈ L. Then [x, y] ∈ Kx + Ky in each of the following cases: (i) x, y ∈ Ker ω and rank(ω) ≥ 2. (ii) x ∈ Ker ω and rank(ω) ≥ 4. (iii) ω(x, y) = 0 and rank(ω) ≥ 6. Proof. All the cases follow the same format with slight modifications. We use (6) for suitably chosen z and t. The condition of vanishing of ω ensures that all but one terms on the left- hand side vanish, and, applying ω(·, z) to both sides, we derive further vanishing of the corresponding terms on the right-hand side. (i) Choose z, t ∈ L such that ω(z, t) = 1. In that case (6) gives: (15) [x, y] = dω(t, z, y)x + dω(z, t, x)y + ω([y, x], t)z + ω([x, y], z)t. Applying to both sides of this equality ω(·, z), we get ω([x, y], z) = −ω([x, y], z), whence ω([x, y], z) = 0. Similarly, ω([y, x], t) = 0 and (15) reduces to the desired condition. (ii) We may assume y /∈ Ker ω, otherwise we are covered by case (i). Let y ∈ V for a certain linear complement V of Ker ω in L, ω being nondegenerate on V . Since Ky is an 1-dimensional isotropic subspace of V , it lies in a certain maximal isotropic subspace W . Then there is a symmetric nondegenerate bilinear form on W such that V = W ⊕ W ∗, W ∗ is a conjugate of W with respect to that form, and for a, a′ ∈ W, f, f ′ ∈ W ∗. ω(a + f, a′ + f ′) = f ′(a) − f (a′) As dim W = 1 2rank(ω) ≥ 2, we may take z ∈ W linearly independent with y. Take t = z∗. In that case (6) gives: (16) [x, y] = dω(t, z, y)x + dω(z, t, x)y + (ω([y, x], t) + ω([x, t], y))z + (ω([x, y], z) + ω([z, x], y))t. Applying to both sides of this equality ω(·, z), we get: ω([x, y], z) = −ω([x, y], z) − ω([z, x], y), 10 whence PASHA ZUSMANOVICH By symmetry considerations, interchanging y and z, we get 2ω([x, y], z) − ω([x, z], y) = 0. whence 2ω([x, z], y) − ω([x, y], z) = 0, ω([x, y], z) = ω([x, z], y) = 0, and the condition desired follows again from (16). (iii) We may assume x, y /∈ Ker ω, otherwise we are covered by cases (i) and (ii). We reason as in the previous case, enlarging the isotropic subspace Kx + Ky to a maximal isotropic subspace W in a linear complement V of Ker ω. Since dim W = 1 2rank(ω) ≥ 3, we may take z ∈ W linearly independent with x, y. Take t = z∗. Then (6) gives: [x, y] = dω(t, z, y)x + dω(z, t, x)y + dω(y, x, t)z + dω(x, y, z)t. Applying ω(·, z) to both sides of this equality, we get Permuting x, y, z, we get ω([x, y], z) = −dω([x, y], z). ω([x, y], z) = dω([x, y], z) = 0, and the condition desired readily follows. (cid:3) The just proved Lemma shows, in particular, that for a sufficiently large rank(ω), Ker ω is a quasi-ideal of an ω-Lie algebra (recall that a subspace I of an algebra L is called quasi-ideal if [I, A] ⊆ I +A for any subspace A ⊆ L). Quasi-ideals of Lie algebras were studied by Amayo in [A, Part I]. It is possible to develop a parallel theory of quasi-ideals of ω-Lie algebras, but it turns out that it will largely coincide with the Lie algebras case (which follows, a posteriori, also from the structural results about ω-Lie algebras obtained below). Thus we restrict ourselves to the immediate case we need, namely, of 1-dimensional quasi-ideals. Lemma 5.2. Let L be an ω-Lie algebra, I an 1-dimensional quasi-ideal of L, I ⊆ Ker ω. Then either I is an ideal of L, or L is a Lie algebra. Proof. We chiefly follow the line of reasoning in [A, pp. 31 -- 32]. If dim L = 2, the Lemma is trivially true, so assume dim L ≥ 3. Let I = Ka, a ∈ Ker ω. Then (17) [x, a] = λ(x)x + µ(x)a for some functions λ, µ : L → K and for any x ∈ L. By linearity, λ(x) = λ is constant. If λ = 0, then Ka is an ideal of L, so assume λ 6= 0. Replacing a by 1 λa, we may set λ = 1. Writing the ω-Jacobi identity for x, y, a, and taking into account (17), we get: [x, y] = µ(x)y − µ(y)x + (ω(x, y) − µ([x, y]))a. Multiplying the last equality by a, we get: [x, y] = µ(x)y − µ(y)x − µ([x, y])a. Comparing the last two equalities, we get ω(x, y) = 0 for any x, y ∈ L linearly independent with a. Hence ω vanishes and L is a Lie algebra. (cid:3) Lemma 5.1 shows also that any isotropic subspace of an ω-Lie algebra is a Lie subalgebra in which every 1-dimensional subspace is a quasi-ideal, provided the rank of ω is large enough. Such Lie algebras have a fairly trivial structure. ω-LIE ALGEBRAS 11 Definition. A semidirect sum A ⊕ Kx where A is an abelian Lie algebra and ad x acts on A as the identity map, is called an almost abelian Lie algebra, and A is called its abelian part. Lemma 5.3. A finite-dimensional Lie algebra such that every two its linearly independent elements generate a two-dimensional subalgebra, is either abelian or almost abelian. Proof. This is implicit in [A, Part I, Theorem 3.6 and proof of Theorem 3.8]. As the proof is very simple and we will need a similar reasoning later, we will reproduce it here. Let L be a Lie algebra with the property specified in the condition of the Lemma. We may assume dim L > 1. Write [x, y] = λ(x, y)x + µ(x, y)y for any two elements x, y ∈ L. By anti-commutativity, µ(x, y) = −λ(y, x), and by linearity λ is constant in the first argument, so [x, y] = λ(y)x − λ(x)y for some linear form λ : L → K. If λ = 0, then L is abelian. If λ 6= 0, write L = Ker λ ⊕ Kx for x ∈ L such that λ(x) = 1, and then L is almost abelian. (cid:3) Putting all this together, we get Lemma 5.4. Let L be a finite-dimensional ω-Lie algebra which is not a Lie algebra. Then one of the following holds: (i) L has a Lie subalgebra of codimension 1. (ii) Ker ω is an abelian or almost abelian Lie subalgebra of L of codimension 2. (iii) L is an abelian extension of a simple ω-Lie algebra with nondegenerate ω. Proof. By Lemmata 5.1(i) and 5.3, Ker ω is an abelian or almost abelian Lie algebra. If rank(ω) = codim Kerω = 2, we are in case (ii), so assume rank(ω) ≥ 4. Then by Lemma 5.1(ii), Ker ω is an ideal. If L is simple, then Ker ω = 0, which is covered by case (iii). So suppose L is not simple and consider a nonzero maximal ideal I of L. By Corollary 3.2, I is a Lie algebra. If codim I = 1, we are in case (i), so let codim I > 1. Then by Corollary 3.5 either L is an abelian extension of a simple ω-Lie algebra L/I, or I is contained in a Lie subalgebra of codimension 1. In the former case, as rank(ωL/I) = rank(ω) ≥ 4, by already noted, ω is nondegenerate on L/I, so we are in case (iii). In the latter case we are in case (i). (cid:3) We will treat the cases of Lemma 5.4 subsequently in the next three sections. 6. (α, λ)-derivations In the previous sections we had encountered repeatedly a situation when an ω-Lie algebra (As, by has a Lie subalgebra of codimension 1. Corollary 3.2, proper ideals are necessarily Lie subalgebras, this includes also the case of ideals of codimension 1). In this section we study this situation. Let L be an ω-Lie algebra and A a subalgebra of L of codimension 1. Write L = A ⊕ Kv for some v ∈ L. Then (18) [x, v] = D(x) + λ(x)v for x ∈ A, and some linear maps D : A → A and λ : A → K. Straightforward calculation shows that the ω-Jacobi identity for L is equivalent to the following three conditions: first, A is an ω-Lie algebra, second, (19) D([x, y]) − [D(x), y] + [D(y), x] = λ(y)D(x) − λ(x)D(y) + ω(y, v)x − ω(x, v)y, 12 and third, (20) for any x, y ∈ A. In particular, we have PASHA ZUSMANOVICH ω(x, y) = λ([x, y]) Lemma 6.1. A subalgebra of codimension 1 in an ω-Lie algebra is a multiplicative ω-Lie algebra. Equation (19) suggests the following Definition. A linear map D : A → A of an anticommutative algebra A is called (α, λ)- derivation of A if there are linear forms α, λ : A → K such that (21) D(ab) = D(a)b + aD(b) + λ(b)D(a) − λ(a)D(b) + α(b)a − α(a)b holds for any a, b ∈ A. So, given a multiplicative ω-Lie algebra A (with ω given by (20)) and its (α, λ)-derivation D, we get an ω-Lie algebra as a vector space A ⊕ Kv, with multiplication and ω extended from A, and defining the rest by (18) and ω(x, v) = α(x). Conversely, every ω-Lie algebra with a subalgebra of codimension 1 occurs in that way. An ω-Lie algebra with a subalgebra A of codimension 1 is a Lie algebra if and only if α = 0 and λ([A, A]) = 0. Unfortunately, the space of all (α, λ)-derivations of a given noncommutative algebra A for a fixed λ is, generally, not closed under operation of commutation. There is, however, a remarkable case where it does. Proposition 6.2. The space of all (α, 0)-derivations of an anticommutative algebra forms a Lie algebra under operation of commutation. Proof. Direct calculation shows that if D1 is an (α1, 0)-derivation and D2 is an (α2, 0)- derivation, then [D1, D2] is an (α1 ◦ D2 − α2 ◦ D1, 0)-derivation. (cid:3) This Lie algebra contains an algebra of (ordinary) derivations of A. (α, 0)-derivations correspond to the case where A is an ideal of codimension 1. Note that our definition of (α, λ)-derivations looks somewhat similar to some other defi- nitions of generalized derivations of Lie and associative algebras: generalized derivations in the sense of Leger and Luks, i.e., triples (D1, D2, D3) of endomorphisms of an algebra A such that D1(ab) = D2(a)b + aD3(b) for any a, b ∈ A (see [LL]) and generalized derivations in the sense of Nakajima, i.e., pairs (D, u) of an endomorphism D of an algebra A and an element u ∈ A such that D(ab) = D(a)b + bD(a) + aub for any a, b ∈ A (see, for example, [KN] and [AA]). However, this does not go much beyond superficial similarity in formulae: in general, (α, λ)-derivations seem to intersect trivially with generalized derivations in either sense. We are interested in (α, λ)-derivations of Lie algebras. In that case, due to (20), λ vanishes on the commutant of an algebra. Lemma 6.3. Let L be a finite-dimensional Lie algebra and D its (α, λ)-derivation. Then one of the following holds: (i) dim L ≤ 3. (ii) α = 0. (iii) Ker α is a subalgebra of L of codimension 1 and one of the following holds: (a) L = A ⊕ Kx, A is abelian, ad x : A → A is any linear map; Ker α = A. (b) L is the direct sum of an abelian Lie algebra A and the two-dimensional non- abelian Lie algebra hx, y [x, y] = yi; Ker α = A ⊕ Kx. ω-LIE ALGEBRAS 13 (c) L = A⊕Kx⊕Ky, A is abelian, ad x : A → A is the identity map, ad y : A → A is any linear map, and [x, y] ∈ A; Ker α = A ⊕ Kx. (d) L = A⊕Kx⊕Ky, A is abelian, ad x : A → A is the identity map, ad y : A → A is the zero map, [x, y] = a+σy for some a ∈ A, σ ∈ K, σ 6= 0; Ker α = A⊕Kx. Proof. Applying D to the Jacobi identity, we get: α(z)[x, y] + α(x)[y, z] + α(y)[z, x] (22) for any x, y, z ∈ L. + (α([y, z]) + λ(y)α(z) − λ(z)α(y))x + (α([z, x]) + λ(z)α(x) − λ(x)α(z))y + (α([x, y]) + λ(x)α(y) − λ(y)α(x))z = 0 Assuming in (22) x, y, z ∈ Ker α, we get that either dim Ker α ≤ 2 and hence dim L ≤ 3, or α([Ker α, Ker α]) = 0 and hence Ker α is a subalgebra of L. Thus, assuming dim L > 3, either α = 0 or Ker α is a subalgebra of L of codimension 1. Now, taking in (22) x, y ∈ Ker α, z ∈ L such that α(z) = 1, we get: [x, y] = (α([z, y]) − λ(y))x + (α([x, z]) + λ(x))y. Thus Ker α is a Lie algebra such that any two linearly independent elements in it generate a two-dimensional subalgebra. By Lemma 5.3, Ker α is either abelian of almost abelian. Now straightforward computations produce the list (iii) of Lie algebras having a subalgebra of codimension 1 which is either abelian or almost abelian (note that not all algebras in this list are pairwise non-isomorphic; we have accounted also for different possibilities of Ker α). (cid:3) Lie algebras listed in part (iii) may have many (α, λ)-derivations, and they do not appear to allow description in a nice compact form. For example, consider the algebra in (iiia) with non- nilpotent ad x and suppose the ground field is algebraically closed. Let F be an eigenvector corresponding to a nonzero eigenvalue σ of ad x in a Lie algebra End(A): [F, ad x] = σF . Then D ∈ End(L) defined by D(a) = F (a) + a, a ∈ A and D(x) = 0, is an (α, λ)-derivation for α defined by α(A) = 0, α(x) = σ and λ defined by λ(A) = 0, λ(x) = −σ. This provides an example of an ω-Lie algebra which is not a Lie algebra in any finite dimension ≥ 3. Nevertheless, writing, for each of these cases, the generic (α, λ)-derivations in terms of linear functions on the corresponding space A, the elementary but tedious linear-algebraic considerations, much in the spirit of Lemmata 5.2, 5.3 above, and Lemma 7.3 below, show that one always can choose a nonzero abelian ideal in the corresponding ω-Lie algebra. That leads to Corollary 6.4. A finite-dimensional semisimple ω-Lie algebra with a Lie subalgebra of codi- mension 1 is either a Lie algebra, or has dimension ≤ 4. An alternative proof of this statement might be obtained following the approach which was used by Amayo in a description of simple Lie algebras with a subalgebra of codimension 1 in [A, Part II], but goes back to Weisfeiler at the end of 1960s. Let L be an ω-Lie algebra of dimension > 3 with a Lie subalgebra L0 of codimension 1. Since dim L0 ≥ 3, ω(L0, L0) = 0. Put L−1 = L and define a filtration on L by (23) Li+1 = {x ∈ Li [x, L] ⊆ Li}, i ≥ 0. The same reasonings as in the Lie-algebraic case, allow one to establish some of the properties of this filtration (for example, that each nonzero term has codimension 1 in the preceding term), and the additional condition [[L0, L0], [L0, L0]] = 0, which holds by inspection of all Lie algebras listed in Lemma 6.3(iii), could be used to finish the proof. 14 PASHA ZUSMANOVICH Now let us consider some examples of (α, 0)-derivations of low-dimensional Lie algebras. It is obvious that for the 2-dimensional abelian Lie algebra, all (α, 0)-derivations are ordinary derivations. Direct easy calculation shows that if L is the 2-dimensional nonabelian Lie algebra, then the Lie algebra of its (α, 0)-derivations coincides with End(L). If x, y are basic elements of L such that [x, y] = x, then the linear transformation given by matrix (cid:18)a b c d(cid:19) in that basis is not a derivation if and only if bd 6= 0 (and then α(x) = −b, α(y) = −d). Direct calculation shows that the Lie algebra of (α, 0)-derivations of sl(2) is 5-dimensional and isomorphic to the semidirect sum of sl(2) and its 2-dimensional standard module. In the basis {e, f, h [e, h] = −e, [f, h] = f, [e, f ] = h}, the basic (α, 0)-derivations which are not ordinary derivations can be chosen as the corresponding α's being e 7→ 0, f 7→ 0, h 7→ e e 7→ 0, f 7→ 0, h 7→ f, e 7→ 0, f 7→ −1, h 7→ 0 e 7→ 1, f 7→ 0, h 7→ 0, respectively. This provides examples of 4-dimensional ω-Lie algebras which are not Lie algebras. Note that since λ = 0, these algebras are not simple. On the other hand, any nonzero (α, λ)-derivation with λ 6= 0 of a 3-dimensional simple ω-Lie non-Lie algebra A, gives rise to a simple 4-dimensional ω-Lie algebra. Indeed, if L = A ⊕ Kv is such an algebra with multiplication (18), and I is a proper ideal of L, then, due to simplicity of A, I ∩ A = 0, hence we may write I = Kv, which implies D = 0, a contradiction. There are many such derivations of 3-dimensional simple ω-Lie algebras. This can be verified with the aid of a simple computer program described in Appendix. One such example is given in §9. 7. Kernel of codimension 2 Now we can see that in Lemma 5.4, the case (ii) essentially covers, up to algebras of small dimension, the case (i): Lemma 7.1. Let L be a finite-dimensional ω-Lie algebra with a Lie subalgebra of codimen- sion 1. Then one of the following holds: (i) L is a Lie algebra. (ii) dim L = 3. (iii) codim Ker ω = 2. Proof. This follows immediately from the results of the previous section. Indeed, such ω- Lie algebras are described by (α, λ)-derivations of Lie algebras listed in Lemma 6.3, with Ker ω = Ker α. (cid:3) In the opposite direction we have: Lemma 7.2. Let L be a finite-dimensional ω-Lie algebra with Ker ω of codimension 2. Then one of the following holds: (i) dim L = 3. ω-LIE ALGEBRAS 15 (ii) L has a Lie subalgebra of codimension 1. (iii) Ker ω is almost abelian, with the abelian part acting nilpotently on L. Proof. By §5, Ker ω is abelian or almost abelian. In the latter case, write Ker ω = H ⊕ Ka, H is abelian, and [h, a] = h for any h ∈ H. For notational convenience, we will assume H = Ker ω in the case of abelian Ker ω. If dim H = 1, we are in case (i) or (ii), so let dim H > 1. Consider the Fitting decomposi- tion of L with respect to H: L = L0 ⊕ L1, and its refinement -- the root space decomposition of L = L ⊗K K over an algebraic closure K of the ground field K with respect to H ⊗K K (note that L0 = L0 ⊗K K). Obviously, Ker ω ⊆ L0. By Lemma 4.5(ii), L0 is a subalgebra of L, hence L0 is a subalgebra of L. Assume first that L0 $ L. If Ker ω $ L0, then L0 is a Lie subalgebra of L of codimension 1, and we are in case (ii). Hence we may assume L0 = Ker ω and L0 = Ker ω = Ker ω ⊗K K. There is either one nonzero root space Lα of dimension 2, or two nonzero root spaces of dimension 1. In the former case, by Lemma 4.5(i), ω(Lα, Lα) = 0, hence Lα ⊆ Ker ω, a contradiction. In the latter case, both root spaces are simple. If Ker ω = H is abelian, then by Lemma 4.5(iv), L is a Lie algebra, hence L is a Lie algebra, a contradiction. Suppose Ker ω is almost abelian and let Lα = Kx be one of the root spaces. By Lemma 4.5(ii), we may write [x, a] = λx for some λ ∈ K. Writing the ω-Jacobi identity for x, a and h ∈ H, we get α(h) = 0, a contradiction. Now consider the case where L0 = L, i.e. H acts on L nilpotently. Suppose Ker ω = H is abelian. Ker ω also acts nilpotently on any module which is a quotient of L, in particular, on L/Ker ω. Consequently, there is x /∈ Ker ω such that the whole Ker ω maps x+ Ker ω ∈ L/Ker ω to zero, and hence [x, Ker ω] ⊆ Ker ω. Then Ker ω ⊕ Kx is a Lie subalgebra of L of codimension 1, and we are again in case (ii). The remaining case is when Ker ω is almost abelian and H acts on L nilpotently, and this (cid:3) is exactly the case (iii). Note that the case (iii) does not seem to be amenable to any compact classification. However, like in §6, we able to deal with simple algebras: Lemma 7.3. A finite-dimensional semisimple ω-Lie algebra satisfying the condition (iii) of Lemma 7.2, has dimension ≤ 4. Proof. Note that in all ω-Jacobi identities considered below, the right-hand side vanishes, so, essentially, this proof consists of tedious but elementary Lie-algebraic considerations. Let L be such an algebra. Write, as previously, Ker ω = H ⊕ Ka, H abelian, [a, h] = h for any h ∈ H. The set ad h for all h ∈ H, being a set of nilpotent commuting operators on the 2-dimensional vector space L/Ker ω, can be brought simultaneously to the upper triangular form, i.e., one can choose a basis {x, y} of the vector space complementary to Ker ω in L, and linear functions λ, µ, η : H → K and f, g : H → H such that [x, h] = λ(h)y + µ(h)a + f (h) [y, h] = η(h)a + g(h) for any h ∈ H. The ω-Jacobi identity for triple h1, h2 ∈ H, y (in other words, the condition that ad h commute for all h ∈ H), yields η(h1)h2 = η(h2)h1 for any h1, h2 ∈ H. The latter condition implies η = 0 provided dim H > 1, i.e., dim L > 4. Similarly, the ω-Jacobi identity for triple h1, h2 ∈ H, x, yields λ(h1)g(h2) − λ(h2)g(h1) + µ(h1)h2 − µ(h2)h1 = 0 16 PASHA ZUSMANOVICH for any h1, h2 ∈ H. Then elementary linear-algebraic considerations imply that, provided dim H > 1, one of the following holds: (1) λ 6= 0, µ = tλ for some t ∈ K, and g(h) = −th for h ∈ Ker λ; (2) λ = µ = 0. In the second case H is an abelian ideal of L. In the first case, replacing y by y + ta and g(h) by g(h) + th, we may assume that t = 0 and g(Ker λ) = 0. The ω-Jacobi identity for triple y, a, h ∈ H yields [[y, a], h] = 0. It is easy to see that the semisimplicity of L implies that the centralizer of H in L coincides with H, so the latter equality implies [y, a] ∈ H. Further, writing [x, a] = αx + βy + γa + h′ for certain α, β, γ ∈ K, h′ ∈ H, and collecting terms in the ω-Jacobi identity for triple x, a, h ∈ H, lying in Ky and H, we get that α = 1 and f (h) = −γh for any h ∈ Ker λ. But then Ker λ is a nonzero abelian ideal of L. (cid:3) In this section we treat the final, third case of Lemma 5.4. 8. Nondegenerate ω Lemma 8.1. If L is a finite-dimensional ω-Lie algebra with nondegenerate ω, then dim L = 2. Proof. Since ω is nondegenerate, dim L = rank(ω) is even. First consider the case dim L ≥ 6. To treat this case, we will adopt the coordinate notation. Though perhaps less elegant, it will make computations easier. L can be written as the direct sum of two maximal isotropic subspaces A and B, each of dimension n = dim L 2 ≥ 3. We may choose a basis {a1, . . . , an} of A and a basis {b1, . . . , bn} of B such that ω(ai, bi) = 1 and ω(ai, bj) = 0 if i 6= j. Then by Lemmata 5.1(iii) and 5.3, each that isotropic subspace is either abelian or almost abelian Lie subalgebra, and it follows from the proof of Lemma 5.3 that we may write multiplications in them as [ai, aj] = αjai − αiaj and [bi, bj] = βjbi − βibj for some αi, βi ∈ K. Again, by Lemma 5.1(iii), [ai, bj] = λijai + µijbj if i 6= j, for some λij, µij ∈ K. Writing (6) for elements bi, bj, bk, ai, where i, j, k are pairwise distinct, and collecting co- efficients of bj, we get (24) ω([ai, bi], bk) = λik for any i 6= k. Similarly, writing (6) for elements ai, aj, bi, bk, where i, j, k are pairwise distinct, and collecting coefficients of aj, we get Comparing these two equalities, we get λjk = βk for any j 6= k. In a completely symmetric way, we also get ω([ai, bi], bk) = λik − λjk + βk. (25) for any i 6= k. ω([ai, bi], ak) = −µki = αk (24) and (25) give all coefficients in the decomposition of [ai, bi] by elements of the sym- plectic basis {a1, . . . , an, b1, . . . , bn}, except those of ai, bi, so for any 1 ≤ i ≤ n we may write [ai, bi] = X1≤k≤n k6=i (βkak − αkbk) + λiai + µibi for some λi, µi ∈ K. ω-LIE ALGEBRAS 17 Finally, writing (6) for elements ai, aj, bi, bj, where i 6= j, taking into account all multipli- cation formulas between elements of A and B obtained so far, and collecting coefficients of ai and aj, we get respectively: λi = 2βi and λj = −2βj. Consequently, λi = βi = 0 for any 1 ≤ i ≤ n. Analogously, collecting coefficients of bi and bj, we get µi = αi = 0. Therefore, L is abelian. But then the ω-Jacobi identity implies that for any 3 linearly independent elements, the values of ω on their pairwise arguments vanish, which implies that ω vanishes, a contradiction. The case dim L = 4 requires a bit more cumbersome computations. Note that we may assume that the ground field is algebraically closed, as nondegeneracy of ω is obviously preserved under the ground field extension. Lemma 8.2. A 4-dimensional ω-Lie algebra over an algebraically closed field contains a 3-dimensional subalgebra. Proof. According to [KK, Corollary 2], any 4-dimensional anticommutative algebra all whose elements are nilpotent, contains a 3-dimensional subalgebra. Consequently, we may assume that L contains a non-nilpotent element x. We cannot invent anything better than proceed by boring case-by-case computations ac- cording to the Jordan normal form of ad x in a certain basis {x, y, z, t} of L. Structure constants in that basis will be denoted as C w uv, the latter being the coefficient of w in the decomposition of [u, v], where u, v, w ∈ {x, y, z, t}. 0 0 0 0 0 α 0 0 0 0 β 0 0 0 0 γ     Case 1. ad x = . Writing the ω-Jacobi identity for triple x, y, z and collecting coefficients of t, we get C t yz = 0, then Kx ⊕ Ky ⊕ Kz forms a 3-dimensional subalgebra. Otherwise, α + β − γ = 0. Repeating this argument for triples x, y, t and x, z, t, we get another two equalities: α − β + γ = 0 and −α + β + γ = 0 respectively. The obtained homogeneous system of 3 linear equations in 3 unknowns has only trivial solution, whence α = β = γ = 0 and ad x is zero, a contradiction. yz(α + β − γ) = 0. If C t   Case 2. ad x = 0 0 0 0 0 α 1 0 0 0 α 0 0 0 0 β ing coefficients of z, we get C z yt = 0, then Kx ⊕ Ky ⊕ Kt forms a 3-dimensional subalgebra, otherwise β = 0. Now writing the ω-Jacobi identity for triple x, y, z and collect- ing coefficients of t, we get C t yz = 0, and Kx ⊕ Ky ⊕ Kz forms a 3-dimensional subalgebra. yzα = 0. Since ad x is not nilpotent, α 6= 0, hence C t . Writing the ω-Jacobi identity for triple x, y, t and collect- ytβ = 0. If C z   Case 3. ad x = . Writing the ω-Jacobi identity for triple x, y, z and collecting coefficients of t, we get C t a 3-dimensional subalgebra. yzα = 0. Since α 6= 0, C t yz = 0, and Kx ⊕ Ky ⊕ Kz forms     0 0 0 0 0 α 1 0 0 0 α 1 0 0 0 α   0 1 0 0 0 0 0 0 0 0 α 0 0 0 0 β   Case 4. ad x = . Writing the ω-Jacobi identity for triple x, y, z and col- lecting coefficients of t, we get C t yz(α − β) = 0. If C t yz = 0, then Kx ⊕ Ky ⊕ Kz forms 18 PASHA ZUSMANOVICH a 3-dimensional subalgebra, otherwise α = β. Now writing the ω-Jacobi identity for triple x, z, t and collecting coefficients of y, we get 2C y ztα = 0. Since ad x is not nilpotent, α 6= 0, hence C y zt = 0, and Kx ⊕ Kz ⊕ Kt forms a 3-dimensional subalgebra. zt = 0, Case 5. ad x = . Writing the ω-Jacobi identity for triple x, z, t and collect-  0 1 0 0 0 0 0 0  0 0 α 1 0 0 0 α ing coefficients of y, we get 2C y ztα = 0. Since ad x is not nilpotent, α 6= 0, hence C y and Kx ⊕ Kz ⊕ Kt forms a 3-dimensional subalgebra.   0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 α     Case 6. ad x = . Finally, writing the ω-Jacobi identity for triple x, y, z and collecting the coefficients of t, we get C t C t yz = 0, and Kx ⊕ Ky ⊕ Kz forms a 3-dimensional subalgebra. yzα = 0. Since ad x is not nilpotent, α 6= 0, hence (cid:3) Continuation of the proof of Lemma 8.1. By Lemma 8.2, L has a 3-dimensional subalgebra M. Clearly, dim Ker ωM is equal to 1 or 3. In the latter case M is a Lie algebra, and ω is necessarily degenerate on the whole L. Hence M is a 3-dimensional ω-Lie algebra which is not a Lie algebra. All isomorphism classes of 3-dimensional ω-Lie algebras are listed in [N, Theorem 2.1]†, and following the scheme of §6, our task amounts to finding (α, λ)-derivations of these algebras. λ can be found from (20), which in all cases amounts to a linear system of 3 equations in 3 unknowns (values of λ on the basic elements), having either single or a 1-parametric solution. We used a primitive GAP code, described in Appendix, to find that for any (α, λ)- derivation of any 3-dimensional ω-Lie algebra M, α vanishes on Ker ωM . Consequently, for the appropriate 4-dimensional ω-Lie algebra L, Ker ω ⊇ Ker ωM 6= 0, i.e., ω is degen- erate. (cid:3) To summarize results of Lemmata 5.4, 6.3, 7.2 and 8.1: 9. Main theorems Theorem 1. Let L be a finite-dimensional ω-Lie algebra which is not a Lie algebra. Then one of the following holds: (i) dim L = 3. (ii) L has a Lie subalgebra of codimension 1 whose structure is described by Lemma 6.3(iii). (iii) Ker ω is an almost abelian Lie algebra of codimension 2 in L with the abelian part acting nilpotently on L. In all the cases, L has an abelian subalgebra of codimension ≤ 3. To summarize further results of Corollary 6.4 and Lemma 7.3: Theorem 2. A finite-dimensional semisimple ω-Lie algebra is either a Lie algebra, or has dimension ≤ 4. † In [N], algebras are classified over the field of real numbers, but the classification readily extends to any ground field of characteristic 6= 2. Over an algebraically closed field, types (V III)a and (IX)a are isomorphic, and over any field all parametric types are isomorphic for parameters a and −a. Note also that the definition of the ω-Jacobi identity adopted here differs from those in [N] by the sign of ω. ω-LIE ALGEBRAS 19 Thus, the structure of ω-Lie algebras beyond dimension 3 turns out to be quite "degener- ate", and, in a sense, all the interesting cases are already presented in [N]. In principle, it is possible to enumerate on computer all isomorphism classes of 4-dimen- sional ω-Lie algebras, and, among them, of all simple algebras, but we will not venture into this: as noted at the end of §6, there are a lot of them, sometimes with quite cumbersome multiplication tables. Here is just one example of a 4-dimensional simple ω-Lie algebra, obtained via appropriate (α, λ)-derivation from the algebra of type (IV )T in the Nurowski's list: [e1, e2] = e2, [e1, e3] = e3, [e2, e3] = e1, [e1, e4] = −e3 + 2e4, [e2, e4] = e1, [e3, e4] = 0, and the only nonzero values of ω on the pairs of elements from the basis are: ω(e2, e3) = ω(e2, e4) = 2. 10. Identities In this section we address the following natural question: what identities are satisfied by ω-Lie algebras? Note that one should distinguish identities of ω-Lie algebras as ordi- nary algebras, i.e., in the signature consisting of one binary operation which is an algebra multiplication, and as ω-algebras, i.e., in the signature consisting of one binary operation representing multiplication, and one binary operation with values in the ground field, rep- resenting the form ω. Let us call identities in the latter sense ω-identities. Clearly, the ω-Jacobi identity is an ω-identity, and we are primarily interested in (ordinary) identities in the former sense. One of the fruitful methods to study identities of algebras is to superize the situation, and consider the Grassmann envelopes of corresponding superalgebras. But to be able to apply this method, the class of algebras under consideration should be closed under tensoring with an associative commutative algebra. It is easy to see that this is not so for ω-Lie algebras. To this end, we enlarge the definition of ω-Lie algebras by allowing ω to take values in the centroid of the algebra, instead of merely in the ground field: Definition. An anticommutative algebra L is called an extended ω-Lie algebra if there is a bilinear map ω : L × L → Cent(L) such that the ω-Jacobi identity (1) holds. Here Cent(L) denotes the centroid of an algebra L. For any algebra L and associative commutative algebra A, we have an obvious inclusion Cent(L) ⊗ A ⊆ Cent(L ⊗ A). If L is an extended ω-Lie algebra, then, defining a bilinear map (26) Ω : (L ⊗ A) × (L ⊗ A) → Cent(L) ⊗ A (x ⊗ a, y ⊗ b) 7→ ω(x, y) ⊗ ab, where x, y ∈ L, a, b ∈ A, we see that L ⊗ A becomes an extended Ω-Lie algebra. It is clear that the class of ω-Lie algebras and the class of extended ω-Lie algebras satisfy the same identities and ω-identities. The notion of (extended) ω-Lie algebra can also be generalized to the super case. For a superalgebra L, let Cent(L) denote a supercentroid of L (which is a generalization of the ordinary centroid). 20 PASHA ZUSMANOVICH Definition. A super-anticommutative superalgebra L = L0 ⊕ L1 is called an extended ω-Lie superalgebra if there is a super-skew-symmetric bilinear map ω : L × L → Cent(L) such that (27) (−1)deg x deg z[x, [y, z]] + (−1)deg z deg y[z, [x, y]] + (−1)deg y deg x[y, [z, x]] + (−1)deg x deg zω(y, z)x + (−1)deg z deg yω(x, y)z + (−1)deg y deg xω(z, x)y = 0 holds for any homogeneous elements x, y, z ∈ L. As in the ordinary case, one easily sees that if L = L0 ⊕ L1 is an extended ω-Lie superalge- bra, and A = A0 ⊕ A1 is a commutative superalgebra, then the algebra (L0 ⊗ A0) ⊕ (L1 ⊗ A1) is an extended Ω-Lie algebra for Ω defined by formula (26), for the respective homogeneous elements x, y, a, b. In particular, the Grassmann envelope of an extended ω-Lie superalgebra is an Ω-Lie algebra for a suitable Ω. Now, again, like in the case of ordinary algebras, we have the well-known correspondence between ω-identities of an ω-Lie superalgebra and ω-identities of its Grassmann envelope (realized, on the level of multilinear identities, by injecting appropriate signs at appropriate places). This provides a compact method to write identities and ω-identities of (extended) ω-Lie algebras. For example, the ω-Jacobi superidentity (27), written for triples x, x, x and x, x, [x, x], implies respectively (28) (29) [[x, x], x] = ω(x, x)x 2ω([x, x], x)x + ω(x, x)[x, x] = 0 for any (odd) x ∈ L. It is possible to show that the full linearization of the ω-identity (29) is equivalent to the super-analog of the identity (6). In its turn, (28) implies the identity [[[[x, x], x], x], x] + [[[x, x], x], [x, x]] = 0 Linearizing the latter identity and taking its "ordinary" part, one arrives at the following identity of degree 5 satisfied by all ω-Lie algebras: Xσ∈S5 (−1)σ(cid:16)[[[[xσ(1), xσ(2)], xσ(3)], xσ(4)], xσ(5)] + [[[xσ(1), xσ(2)], xσ(3)], [xσ(4), xσ(5)]](cid:17) = 0. This is identity of the smallest possible degree: Proposition 10.1. The minimal degree of identity which is satisfied by all ω-Lie algebras is 5. Proof. It is sufficient to show that no identity of degree ≤ 4 is satisfied by all ω-Lie alge- bras. Irreducible identities of degree ≤ 4 of anticommutative algebras were described in [K, §2] and [KW, Theorem 3]. According to these results, every anticommutative algebra with multiplication [ · , · ], satisfying an identity of degree ≤ 4, satisfies one of the following identities: (30) (31) (32) (33) [[[y, x], x], x] = 0 α[[x, y], [x, z]] + β(cid:16)[[[x, y], x], z] − [[[x, z], x], y](cid:17) +γ(cid:16)[[[x, y], z], x] − [[[x, z], y], x](cid:17) + (β + γ)[[[y, z], x], x] = 0 J(x, y, [x, y]) = 0 [J(x, y, z), t] − [J(t, x, y), z] + [J(z, t, x), y] − [J(y, z, t), x] = 0, ω-LIE ALGEBRAS 21 where J(x, y, z) = [[x, y], z] + [[z, x], y] + [[y, z], x] is the Jacobian, and α, β, γ in (31) are some fixed elements of the ground field. Identities (30) and (31) are not satisfied even in the narrower class of Lie algebras: (30) is the 3rd Engel condition, and (31) is not satisfied, for example, in the free 4-generated Lie algebra. Identity (32) defines binary-Lie algebras. Being coupled with the ω-Jacobi identity (1), it implies ω(x, y)[x, y] = ω([x, y], y)x − ω([x, y], x)y The latter condition is violated, for example, for most of the 3-dimensional algebras in Nurowski's list [N]. Similarly, (33) together with (1) implies ω(z, t)[x, y] + ω(t, y)[x, z] + ω(y, z)[x, t] + ω(x, t)[y, z] + ω(z, x)[y, t] + ω(x, y)[z, t] = 0. In view of Lemma 3.3, for ω-Lie algebras of dimension > 3 this is equivalent to ω([x, y], z) + ω([z, x], y) + ω([y, z], x) = 0. The last condition is not fulfilled, for example, for 4-dimensional ω-Lie algebras obtained by extending sl(2) by its (α, 0)-derivations, described at the end of §6. (cid:3) Question 1. What happens in characteristics 2 and 3? 11. Further questions In the case of characteristic 2 an entirely different approach (and, perhaps, a different definition of an ω-Lie algebra) would be needed. On the contrary, the assumption that the characteristic of the ground field is different from 3, was used only twice, in the key Lemma 5.1 and when performing calculations with 3-dimensional ω-Lie algebras described at the end of §8. (Also, in characteristic 3 the ω-identity (28) no longer follows from the ω-Jacobi superidentity, and one needs to include it in the definition of ω-Lie superalgebra). More accurate reasonings could show that a statement similar to Lemma 5.1 still holds in characteristic 3, with stronger conditions on the dimension of rank ω (basically, shifted to 2). So, probably the case of characteristic 3 could be treated along the lines of the present paper. Question 2. Which Lie algebras are deformed into non-Lie ω-Lie algebras? As we learned from Rutwig Campoamor-Stursberg, there was a hope to get some physically meaningful contractions of ω-Lie algebras into simple Lie algebras. From Theorem 1 it is clear that in dimension > 3 this is impossible -- contracted Lie algebras should be not less degenerate than ω-Lie algebras, close to abelian ones. Nevertheless, one can still ask which Lie algebras could arise as such contractions, which is, essentially, equivalent to the question: which Lie algebras could be deformed into ω-Lie algebras? Let us try to develop a rudimentary deformation theory of ω-Lie algebras, following the standard nowadays format suggested by Gerstenhaber in [G]. A deformation of an ω-Lie algebra L (which can be just a Lie algebra with ω = 0) is an ωt-Lie algebra Lt defined over a power series ring K[[t]] whose multiplication [ · , · ]t and form ωt satisfy the conditions [x, y]t = [x, y] + ϕ1(x, y)t + ϕ2(x, y)t2 + . . . ωt(x, y) = ω(x, y) + ω1(x, y)t + ω2(x, y)t2 + . . . 22 PASHA ZUSMANOVICH for certain bilinear maps ϕn : L × L → L and ωn : L × L → K. Anticommutativity of [ · , · ]t and skew-symmetricity of ωt imply that each ϕn is anticom- mutative and each ωn is skew-symmetric. The ω-Jacobi identity for Lt is equivalent to: d ϕn(x, y, z) + Xi+j=n i,j>0 [ϕi, ϕj](x, y, z) = ωn(x, y)z + ωn(z, x)y + ωn(y, z)x for each n = 1, 2, . . . and x, y, z ∈ L, where d is the second-order Chevalley-Eilenberg differential in the Lie algebra (= ω-Lie algebra) cohomology, and [ · , · ] is the usual Massey product of 2-cochains. The first of these equalities (n = 1) reads: (34) ϕ1([x, y], z) + ϕ1([z, x], y) + ϕ1([y, z], x) + [ϕ1(x, y), z] + [ϕ1(z, x), y] + [ϕ1(y, z), x] = ω1(x, y)z + ω1(z, x)y + ω1(y, z)x. Thus, the question reduces to: which Lie algebras admit infinitesimal deformations (34) with nontrivial ω1? Question 3. Are there "interesting" examples of infinite-dimensional ω-Lie algebras and of ω-Lie superalgebras? Could it be that in the super or, more general, color case, new phenomena will arise making the structure theory more colorful, for example, allowing the existence of some interesting simple objects? Question 4. What would be analogs of ω-Lie algebras for other classes of algebras? By analogy with the ω-Jacobi identity, one may to alter the associative identity as follows: (35) (xy)z − x(yz) = ω(x, y)z − ω(y, z)x. One of the main features of associative algebras is that they are Lie-admissible, and one may wish to preserve this relationship for their ω-variants: namely, that if A is an ω-associative algebra, then its "minus" algebra with multiplication [x, y] = xy − yx for x, y ∈ A, would be ω-Lie. An easy calculation shows that the "minus" algebra of an algebra satisfying the identity (35) is Lie, and not just ω-Lie. That indicates that (35) is probably not an adequate definition of ω-associativity, and one may wish to alter it further as follows: (36) (xy)z − x(yz) = ω1(x, y)z − ω2(y, z)x for some two bilinear forms ω1, ω2 : L × L → K (one may argue that, unlike in the Lie case, to reflect the difference of order in multiplication in two terms on the left-hand side, two different bilinear forms ω1 and ω2 are required). A "minus" algebra of an algebra satisfying the latter identity is indeed an ω-Lie algebra, with ω(x, y) = (ω1 − ω2)(x, y) − (ω1 − ω2)(y, x). Is (36) a "correct" definition of an ω-associative algebra? Does it lead to anything interesting? Similarly, what would be "correct" definitions for ω-Leibniz algebras, ω-Novikov algebras, ω-left-(or right-)symmetric algebras, etc.? Acknowledgements Thanks are due to Rutwig Campoamor-Stursberg, Zhiqi Chen and Friedrich Wagemann for interesting discussions, and to Dimitry Leites for suggestions which improved the presen- tation. A previous version of this manuscript was rejected by Journal of Algebra, but I owe to the anonymous referee of that submission a great deal of useful comments: in particular, he urged formulating Theorem 2 in §9 explicitly, and most of the material in §10 of the present version stems from his suggestions. ω-LIE ALGEBRAS 23 Appendix Here we describe a simple-minded GAP code, available at http://justpasha.org/math/alder.gap, for calculating (α, λ)-derivations of an ω-Lie al- gebra, mentioned in §§6 and 8. Let L be an anticommutative algebra with the basis {e1, . . . , en} defined over a field K, and with multiplication table [ei, ej] = n Xk=1 C k ijek, and let D be an (α, λ)-derivation of L. Writing D(ei) = Pn j=1 dijej, λ(ei) = λi, and α(ei) = αi for certain dij, λi, αi ∈ K, the condition (19), written for each pair of basic elements, is equivalent to the system of n2(n−1) linear equations in n2 + n unknowns dij and αi: 2 (37) n Xk=1 C k ijdkl − n Xk=1 C l kjdik + n Xk=1 C l kidjk − λjdil + λidjl − δilαj + δjlαi = 0 for 1 ≤ i < j ≤ n, 1 ≤ l ≤ n (δij is the Kronecker symbol). If, additionally, L is an ω-Lie algebra, λi can be found from (20), which is equivalent to the system of n(n−1) 2 linear equations in n unknowns: n Xk=1 C k ijλk = ω(ei, ej) for 1 ≤ i < j ≤ n. So, taking the structure constants of an algebra, as well λ as an input (possibly involving parameters), we just solve the linear homogeneous system (37). As GAP (version 4.4.12 as of time of this writing) does not support transcendental field extensions -- which would be the natural way to work with parameters -- we are cheating by using cyclotomic fields instead. However, this cheating could be made rigorous by picking a cyclotomic extension of a prime degree (of course, any other field extension by an irre- ducible polynomial will do) larger than the highest possible power of a parameter involved in computation. For example, if we deal with 3-dimensional algebras, the system (37) is of size 9 × 12, so if a parameter enters linearly into the initial data, any of its powers occuring in the solution of the system does not exceed 9, so the cyclotomic extension of order 11 will be enough. References R.K. Amayo, Quasi-ideals of Lie algebras. I, II, Proc. London Math. Soc. 33 (1976), 28 -- 36, 37 -- 64. [A] [AA] N. Arga¸c and E. Alba¸s, On generalized (σ, τ )-derivations, Siber. Math. J. 43 (2002), 977 -- 984. [BC] J. Baez and A.S. Crans, Higher-dimensional algebra VI: Lie 2-algebras, Theory Appl. Categor. 12 [B] (2004), 492 -- 528; arXiv:math/0307263. N. Bourbaki, Alg`ebre. Chapitre 9. Formes sesquilin´eaires et formes quadratiques, Springer, 2007 (reprint of Hermann, Paris, 1959). [G] M. Gerstenhaber, On the deformation of rings and algebras, Ann. Math. 79 (1964), 59 -- 103. [HLS] J. Hartwig, D. Larsson and S. Silvestrov, Deformations of Lie algebras using σ-derivations J. Algebra 295 (2006), 314 -- 361; arXiv:math/0408064. [KW] S. Kass and W.G. Whitthoft, Irreducible polynomial identities in anticommutative algebras, Proc. Amer. Math. Soc. 26 (1970), 1 -- 9. 24 PASHA ZUSMANOVICH [KN] H. Komatsu and A. Nakajima, Generalized derivations of associative algebras, Quaest. Math. 26 (2003), 213 -- 235. [KK] N.A. Koreshkov and D.Yu. Kharitonov, About nilpotency of Engel algebras, Formal Power Series and Algebraic Combinatorics (ed. D. Krob et al.), Springer, 2000, 461 -- 467; also published in Russ. Math. (Izv. VUZ) 45 (2001), No. 11, 15 -- 18. E.N. Kuzmin, On anticommutative algebras satisfying the Engel condition, Siber. Math. J. 8 (1967), 779 -- 785. [K] [LL] G.F. Leger and E.M. Luks, Generalized derivations of Lie algebras, J. Algebra 228 (2000), 165 -- 203. [L] [N] K.-Q. Liu, Characterizations of the quantum Witt algebra, Lett. Math. Phys. 24 (1992), 257 -- 265. P. Nurowski, Deforming a Lie algebra by means of a 2-form, J. Geom. Phys. 57 (2007), 1325 -- 1329; arXiv:math/0605748. A.A. Sagle, On homogeneous spaces, holonomy, and non-associative algebras, Nagoya Math. J. 32 (1968), 373 -- 394. [S] E-mail address: [email protected]
1512.02277	2	1512	2015-12-09T12:59:17	Nil Clean Involutions	[ "math.RA" ]	We prove that if an involution in a ring is the sum of an idempotent and a nilpotent then the idempotent in this decomposition must be 1. As a consequence, we completely characterize weakly nil-clean rings introduced recently in [Breaz, Danchev and Zhou, Rings in which every element is either a sum or a difference of a nilpotent and an idempotent, J. Algebra Appl., DOI: 10.1142/S0219498816501486].	math.RA	math	Nil Clean Involutions Janez Šter Faculty of Mechanical Engineering University of Ljubljana [email protected] December 7, 2015 Abstract We prove that if an involution in a ring is the sum of an idempotent and a nilpotent then the idempotent in this decomposition must be 1. As a consequence, we completely characterize weakly nil-clean rings introduced recently in [Breaz, Danchev and Zhou, Rings in which every element is either a sum or a difference of a nilpotent and an idempotent, J. Algebra Appl., DOI: 10.1142/S0219498816501486]. In this note rings are unital. U (R), Id(R), Nil(R) and Nil∗(R) stand for the set of units, the set of idempotents, the set of nilpotents and the upper nilradical of a ring R, respectively. Zn stands for the set of integers modulo n. An involution in a ring means an element a satisfying a2 = 1. Following [2], we say that an element in a ring is nil clean if it is the sum of an idempotent and a nilpotent, and a ring is nil clean if every element is nil clean. The main result in this note is the following: Proposition 1. Let R be a ring with an involution a ∈ R. If a is the sum of an idempotent e and a nilpotent q then e = 1. In particular, every nil clean involution in a ring is unipotent (i.e. 1 plus a nilpotent). Proof. Write a = e + q with e ∈ Id(R) and q ∈ Nil(R), and denote f = 1 − e ∈ Id(R) and r = q(1 + q) ∈ Nil(R). From f q = f (a − e) = f a we compute f r = f q(1 + q) = f a(1 + a − e) = f a(f + a) = f af + f a2 = f af + f , and similarly rf = f af + f . Hence f r = rf , so that r is a nilpotent which commutes with f , e, q and a. Accordingly, f = f a2 = f qa = f r(1 + q)−1a = f (1 + q)−1a · r is a nilpotent and hence f = 0, as desired. Following [1], we say that a ring is weakly nil-clean if every element is either a sum or a difference of a nilpotent and an idempotent. Lemma 2. If R is a weakly nil-clean ring with 2 ∈ U (R) then R/ Nil∗(R) ∼= Z3. Proof. Choose any idempotent e ∈ Id(R), and set a = 1 − 2e. By assumption, either a or −a is nil clean. If a is nil clean then, since a2 = 1, Proposition 1 gives that a − 1 = −2e is a nilpotent, so that e is a nilpotent and hence e = 0. Similarly, if −a is nil clean then, since (−a)2 = 1, Proposition 1 gives that −a − 1 = −2(1 − e) is a nilpotent, so that 1 − e is a nilpotent and hence e = 1. This proves that R has only trivial idempotents. Accordingly, since R is weakly nil clean, every element of R must be either q or 1 + q or −1 + q for some q ∈ Nil(R). From this, one quickly obtains that Nil(R) must actually form an ideal in R, so that R/ Nil∗(R) can have only 3 elements and hence R/ Nil∗(R) ∼= Z3, as desired. (Alternatively, considering that R is abelian, R/ Nil∗(R) ∼= Z3 can be also obtained from [1, Theorem 12].) 1 Using the above lemma, we have: Theorem 3. A ring is weakly nil clean if and only if it is either nil clean or isomorphic to R1 ×R2 where R1 is nil clean and R2/ Nil∗(R2) ∼= Z3. Proof. Follows from Lemma 2 together with [1, Theorem 5]. Remark 4. Proposition 1 can be generalized to arbitrary algebraic elements of order 2 as follows. Let R be an algebra over a commutative ring k, and let a ∈ R be an element satisfying αa2 + βa + γ = 0, with α, β, γ ∈ k, and suppose that a = e + q with e ∈ Id(R) and qn = 0. Then one can show that r = q(αq + α + β) is a nilpotent commuting with e, which yields, similarly as in Proposition 1, that (α + β)ne + (α + β)n−1γ is also a nilpotent. Note that this result indeed generalizes Proposition 1 (taking α = 1, β = 0 and γ = −1 yields that e − 1 is a nilpotent, so that e = 1). However, for orders of algebraicity higher than 2 this argument no longer seems to work. Acknowledgements The author is indebted to Professor T.Y. Lam for a helpful discussion on the previous version of this work. References [1] S. Breaz, P. Danchev, and Y. Zhou. Rings in which every element is either a sum or a difference of a nilpotent and an idempotent. Journal of Algebra and Its Applications, 0(0):1650148, 0. [2] A. J. Diesl. Nil clean rings. J. Algebra, 383:197 -- 211, 2013. 2
1612.03021	1	1612	2016-12-09T13:45:51	A complete radical formula and 2-primal modules	[ "math.RA" ]	We introduce a complete radical formula for modules over non-commutative rings which is the equivalence of a radical formula in the setting of modules defined over commutative rings. This gives a general frame work through which known results about modules over commutative rings that satisfy the radical formula are retrieved. Examples and properties of modules that satisfy the complete radical formula are given. For instance, it is shown that a module that satisfies the complete radical formula is completely semiprime if and only if it is a subdirect product of completely prime modules. This generalizes a ring theoretical result: a ring is reduced if and only if it is a subdirect product of domains. We settle in affirmative a conjecture by Groenewald and the current author given in \cite{2p} that a module over a 2-primal ring is 2-primal. More instances where 2-primal modules behave like modules over commutative rings are given. This is in tandem with the behaviour of 2-primal rings of exhibiting tendencies of commutative rings. We end with some questions about the role of 2-primal rings in algebraic geometry.	math.RA	math	A complete radical formula and 2-primal modules David Ssevviiri Department of Mathematics Makerere University, P.O BOX 7062, Kampala Uganda E-mail: [email protected], [email protected] Abstract We introduce a complete radical formula for modules over non-commutative rings which is the equivalence of a radical formula in the setting of modules defined over commutative rings. This gives a general frame work through which known results about modules over commutative rings that satisfy the radical formula are retrieved. Examples and properties of modules that satisfy the complete radical formula are given. For instance, it is shown that a module that satisfies the complete radical formula is completely semiprime if and only if it is a subdirect product of completely prime modules. This generalizes a ring theoretical result: a ring is reduced if and only if it is a subdirect product of domains. We settle in affirmative a conjecture by Groenewald and the current author given in [5] that a module over a 2-primal ring is 2-primal. More instances where 2-primal modules behave like modules over commutative rings are given. This is in tandem with the behaviour of 2-primal rings of exhibiting tendencies of commutative rings. We end with some questions about the role of 2-primal rings in algebraic geometry. Keywords: 2-primal rings, 2-primal modules, modules that satisfy the radical formula, modules that satisfy the complete radical formula. MSC 2010 Subject Classification: 16S90; 08B26; 13C05; 16D80; 16N60; 13C10; 14A22 1 Introduction A proper ideal P of a ring R is a prime (resp. completely prime) ideal if for all ideals A, B (resp. elements a, b) of R such that AB ⊆ P (resp. ab ∈ P), we have either A ⊆ P or B ⊆ P (resp. a ∈ P or b ∈ P ). The prime radical (resp. completely prime radical) of a ring R is the intersection of all prime (resp. completely prime) ideals of R. Let N (R), β(R) and βco(R) denote the set of all nilpotent elements of R, the prime radical of R and the completely prime radical of R (also called the generalized nilradical of R). Let N be a submodule of an R-module M. The envelope of a submodule N of an R-module M is the set EM (N) := {rm : r ∈ R, m ∈ M and rkm ∈ N for some k ∈ N}. 1 In general, EM (N) is not a submodule of M. We denote a submodule of M generated by the envelope of N by hEM (N)i. The elements of hEM (0)i are called nilpotent elements of M. If R is a commutative ring (or a 2-primal ring), then ER(0) = N (R) = β(R) = βco(R). (1) A desire to have Equation (1) (or parts of Equation (1)) in the module setting, forms the basis of our study in this paper. In literature, there are studies about the equivalences hEM (N)i = βs(N) and β(M) = βco(M) in which case one says the submodule N of a module M satisfies the radical formula and a module M is 2-primal respectively. βs(N) denotes the intersection of all prime submodules of a module M containing a submodule N and β(M) (resp. βco(M)) denotes the intersection of all prime (resp. completely prime) submodules of a module M. Whereas we supplement studies of these two equivalences, we also introduce submodules that satisfy the complete radical formula, i.e., submodules N of modules M for which hEM (N)i = βs co(N) is the intersection of all completely prime submodules of a module M containing a submodule N. This, naturally generalizes the notion of submodules that satisfy the radical formula in modules over com- mutative rings to modules over non-commutative rings. For an arbitrary ring, Levitzki showed that the set of all strongly nilpotent elements coincides with the prime radical of that ring. We give examples of modules that satisfy the module analogue of Levitzki result. co(N), where βs All rings are unital and associative. The modules are left modules defined over rings. Paper road map This paper has seven sections. We give the introduction in Section 1. Preliminary results which are needed later in the sequel are given in Sections 2 and 3. Section 2 focuses on 2-primal rings and some of their properties, whereas in Section 3 we give relevant information about module analogues of well known notions in ring theory. They include: prime modules, completely prime modules, modules that satisfy the radical formula and 2-primal modules. It is in Section 4 that we introduce the complete radical formula of modules. As examples, it is shown that the following modules satisfy the complete radical formula: a projective and 2-primal module [Theorem 4.8], a finitely generated module over a 2-primal ring [Theorem 4.9], a completely prime module and the regular module RR when R is a 2-primal ring [Theorem 4.14]. In Corollaries 4.6 and 4.7, we have given several modules which are projective and 2-primal. Furthermore, all the modules given above also satisfy both the radical formula as well as the module analogue of Levitzki result for rings. A ring R satisfies a complete radical formula if every R-module satisfies the complete radical formula. It is shown that every semisimple 2-primal ring satisfies the complete radical formula [Corollary 4.13]. We give a new characterization of 2-primal 2 rings. A ring R is 2-primal if and only if β(R) = ER(0) [Theorem 4.2]. In Section 5, we give an application of modules that satisfy the complete radical formula. If a module satisfies the complete radical formula, then it is completely semiprime if and only if it is a subdirect product of completely prime modules [Theorem 5.1]. In Section 6, we prove in affirmative a conjecture posed in [5]; it states that a module over a 2-primal ring is 2-primal. In Section 7, which is the last section, we give some information which inhibits the use of non-commutative rings in algebraic geometry. However, given the behaviour of 2-primal rings, i.e., having behavioural tendencies of commutative rings, we pose some questions on possible of using 2-primal rings in algebraic geometry. 2 2-primal rings Definition 2.1. A ring R is 2-primal if N (R) = β(R). All commutative rings and all reduced rings are 2-primal. The class of 2-primal rings has been widely studied, see for instance [2, 14, 15, 16, 21] among others. Proposition 1. [2, Proposition 2.1] Let R be a ring, the following statements are equiv- alent: 1. R is 2-primal, 2. βco(R) = β(R). An ideal I of a ring R is 2-primal if βco(R/I) = β(R/I). (2) It follows that a ring is 2-primal if and only if its zero ideal is 2-primal. Equality (2) is the basis for the definition of 2-primal submodules, see Definition 3.5. The class of 2-primal rings is large. It contains many classes of generalizations of commu- tative rings: symmetric rings, IFP/SI rings, reversible rings, PSI rings, semi-symmetric rings, etc. For examples and chart of implications among these classes, see [2] and [14]. 2-primal rings behave like commutative rings. For instance, just like commutative rings, they possess the following properties: 1. their sets of all nilpotent elements are ideals; 2. they are Dedekind finite, i.e., if R is a 2-primal ring and a, b ∈ R such that ab = 1, then ba = 1; 3 3. if R is a 2-primal ring, then the ring R/β(R) is reduced, and hence it is IFP (i.e., if a, b ∈ R, then ab ∈ β(R) implies that aRb ⊆ β(R)), reversible (i.e., if a, b ∈ R, then ab ∈ β(R) implies that ba ∈ β(R)) and symmetric (i.e., if a, b, c ∈ R, then abc ∈ β(R) implies that acb ∈ β(R)); 4. they satisfy Kothe conjecture, i.e., the sum of two nil left ideals is nil; 5. prime ideals are completely prime and hence are strongly prime, strictly prime, l-prime and s-prime; 6. they cannot be full matrix rings, [14, p. 495]; 7. the equality Ns(R) = ER(0) = N (R) = β(R) = βco(R) holds, where Ns(R) is the set of all strongly nilpotent elements of R, see Corollary 4.3; 8. semisimple 2-primal rings satisfy the radical formula, see Corollaries 4.12 and 4.13, and Proposition 8. Definition 2.2. [25, Definition 19], [8, p. 742] A filtered ring A is said to be almost commutative if the associated graded ring, grA = Li∈I(Ai/Ai−1) is commutative. Basic examples of almost commutative rings involve rings of differential operators and universal enveloping algebras. Example 1. The almost commutative rings: the universal enveloping algebra of any Lie algebra over a field and the ring of differential operators are reduced rings and hence 2-primal. We are then led to ask the following question: Question 2.1. Is every almost commutative ring 2-primal? 3 The module analogues In this section, we introduce module analogues of prime rings, completely prime rings (domains), 2-primal rings and modules that mimic the equivalence N (R) = β(R) in commutative rings. 4 3.1 Prime and completely prime modules Definition 3.1. [4] A submodule P of an R-module M is a prime submodule if RM 6⊆ P and for all ideals A of R and submodules N of M such that AN ⊆ P , we have N ⊆ P or AM ⊆ P . Definition 3.2. [6, Definition 2.1] A submodule P of an R-module M is a completely prime submodule if RM 6⊆ P and for all elements r ∈ R and m ∈ M such that rm ∈ P , we have m ∈ P or rM ⊆ P . Definition 3.3. A proper submodule P of an R-module M is a semiprime (resp. com- pletely semiprime) submodule of M, if RM 6⊆ P and for all a ∈ R and m ∈ M, aRam ⊆ P (resp. a2m ∈ P ) implies that am ∈ P . A module is prime (resp. completely prime, semiprime, completely semiprime) if its zero submodule is a prime (resp. completely prime, semiprime, completely semiprime) sub- module. The prime (resp. completely prime) radical of a submodule N of an R-module M is the intersection of all prime (resp. completely prime) submodules of M contain- ing N. We denote the prime (resp. completely prime) radical of a nonzero submodule N by βs(N) (resp. βs co(N)). Otherwise, if N = 0, we write β(M) (resp. βco(M)) and call β(M) (resp. βco(M)) the prime (resp. completely prime) radical of M. If M has no prime (resp. completely prime) submodules, we write β(M) = M (resp. βco(M) = M). A ring R is prime (resp. completely prime, semiprime, completely semiprime) if and only if the R-module R is (resp. completely prime, semiprime, completely semiprime). Any completely prime submodule is prime. Every maximal submodule is a prime submodule but it need not be completely prime. A torsion-free module, a simple module which is Lee-Zhou reduced and a projective module over a domain are completely prime modules, see [6, Examples 2.2], [6, Example 2.3] and [23, Example 3.10]. An indecomposable pro- jective module over a hereditary Artin algebra is a completely prime module. To see this, if M is an indecomposable projective module over a hereditary Artin algebra R, then by [7, Proposition 5.1.1] every nonzero map f ∈ EndR(M) is a monomorphism. It follows from [23, Proposition 3.1] that M is a completely prime module. Let N be a submodule of an R-module M, by (N : M) we denote the ideal of R which is the annihilator of the factor R-module M/N. {r ∈ R : rM ⊆ N} Proposition 2. A submodule N of an R-module M is completely prime (resp. prime) if and only if P = (N : M) is a completely prime (resp. prime) ideal of R and the R/P -module M/N is torsion-free. 5 3.2 The radical formula of modules The equality ER(0) = β(R) from Equation (1) for commutative rings, motivated Mc- Casland and Moore in [17] to introduce (sub)modules that satisfy the radical formula. On the other hand, the equality β(R) = βco(R) for 2-primal rings motivated Groenewald and the current author in [5] to define 2-primal modules. In this subsection, we define, give examples and compare these two types of modules. Definition 3.4. A submodule N of an R-module M satisfies the radical formula if hEM (N)i = βs(N). A module M satisfies the radical formula if every submodule of M satisfies the radical formula. A ring R satisfies the radical formula if every R-module satisfies the radical formula. Modules and rings that satisfy the radical formula have been widely studied, see [3, 9, 12, 13, 17, 18, 19, 20] among others. A projective module over a commutative ring [9, Corollary 8], a module over a Dedekind integral domain [9, Theorem 9] and a representable module (and hence an Artinian module) over a commutative ring satisfy the radical formula [19, Theorem 9]. A semisimple commutative ring and an Artinian commutative ring [20] satisfy the radical formula. Not all modules defined over commutative rings satisfy the radical formula. 3.3 2-primal modules Definition 3.5. A submodule N of an R-module M is 2-primal if A module M is 2-primal if its zero submodule is 2-primal, i.e., if β(M/N) = βco(M/N). β(M) = βco(M). Proposition 3. [5, Proposition 2.1] A ring R is 2-primal if and only if the module RR is 2-primal. Example 2. A Lee-Zhou reduced module (see [11]) and hence a completely prime module is 2-primal. Projective modules over 2-primal rings, IFP modules, symmetric modules and modules over commutative rings are 2-primal, see [5]. Proposition 4. If the prime radical of a module M is a completely prime submodule of M, then M is a 2-primal module. 6 Since for any completely prime submodule P of M, β(M) ⊆ βco(M) ⊆ P Proof: when β(M) is a completely prime submodule of M, we get βco(M) ⊆ β(M) such that βco(M) = β(M). We observe that "2-primal modules" is a better generalisation than "modules that satisfy the radical formula". This is because, all modules over commutative rings are 2-primal just like all commutative rings are 2-primal. On the contrary, not all modules over com- mutative rings satisfy the radical formula. There was considerable effort aimed at getting examples of modules that satisfy the radical formula. Now that there is a generalisation better than the notion of modules that satisfy the radical formula, i.e., that of 2-primal modules, it is hoped that there will be interest by different researchers to search for more examples of 2-primal modules, in addition to those pointed out in this paper and in [5]. 4 The complete radical formula The inequality βco(M) ⊆ hEM (0)i (3) which is equivalent to saying that hEM (0)i = βco(M), (see Lemma 1) is a necessary and sufficient condition for a zero submodule of a module M to satisfy the radical formula if and only if M is 2-primal, see [24, Corollary 2.21]. The motivation for studying (sub)modules that satisfy the complete radical formula is three fold. Firstly, it generalizes the notion of modules over commutative rings that satisfy the radical formula to modules over not necessarily commutative rings. Secondly, it is a necessary and sufficient condition for modules to be 2-primal if and only if their zero submodules satisfy the radical formula. Lastly, it allows every completely semiprime submodule to be an intersection of completely prime submodules which is not true in general. Definition 4.1. Let R be a ring and M an R-module. A submodule N of an R-module M satisfies a complete radical formula if hEM (N)i = βs co(N). A module satisfies the complete radical formula if every submodule of M satisfies the complete radical formula. A ring R satisfies the complete radical formula if every R- module satisfies the complete radical formula. Lemma 1. [24, Lemma 2.1] If N is a submodule of an R-module M, then hEM (N)i ⊆ βs co(N). 7 Corollary 4.1. If I is a left ideal of a ring R, then hER(I)i ⊆ βs co(I). We now give a new characterization of 2-primal rings. Theorem 4.2. A ring R is 2-primal if and only if Proof: For any ring R, it is easy to see that β(R) = ER(0). β(R) ⊆ N (R) ⊆ ER(0) ⊆ βco(R). If R is 2-primal, β(R) = βco(R) and hence β(R) = ER(0). For the converse, β(R) = ER(0) implies that β(R) = N (R), which shows that R is 2-primal. Corollary 4.3. If R is a 2-primal ring, then β(R) = N (R) = ER(0) = βco(R). Proof: Follows from the fact that for any ring R, β(R) ⊆ N (R) ⊆ ER(0) ⊆ βco(R) and for 2-primal rings, β(R) = βco(R). Proposition 5. If N is a submodule of an R-module M, then the following statements are equivalent: 1. EM (N) = N, 2. N is a completely semiprime submodule of M. Proof: Elementary. Corollary 4.4. A submodule N of an R-module M is Lee-Zhou reduced if and only if it is both IFP and satisfies EM (N) = N. Corollary 4.4 allows us to paraphrase Question 2.1 posed in paper [24] as: Question 4.1. Is there a prime (resp. semiprime) module which is not completely prime (resp. Lee-Zhou reduced) but it is completely semiprime? A positive answer to this question would lead to an example of a module which satisfies the radical formula but not 2-primal. An element m of an R-module M is strongly nilpotent [1] if m = Pr i=1 aimi for some ai ∈ R, mi ∈ M and r ∈ N, such that for every i (1 ≤ i ≤ r) and every sequence ai1, ai2, ai3, · · · where ai1 = ai and ain+1 ∈ ainRain (for all n), we have aikRmi = 0 for some k ∈ N. The set of all strongly nilpotent elements of a module M is a submodule and is denoted by Ns(M). 8 Lemma 2. For any R-module M, the following inequalities hold: Ns(M) ⊆ hEM (0)i ⊆ βco(M). Let m ∈ Ns(M), then m = Pr i=1 aimi for some ai ∈ R, mi ∈ M and r ∈ N Proof: such that for every i (1 ≤ i ≤ r) and every sequence a1i, a2i, a3i, · · · where ai1 = ai and an+1i ∈ aniRani for all n, we have akiRmi = 0 for some k ∈ N. For some i, choose the sequence ai, a2 i , a4 i , a8 i , · · · = {a2r−1 i }∞ r=1 then a1i = ai and an+1i ∈ aniRani for all n. By hypothesis, there exists k ∈ N such that akiRmi = 0. Since aki = a2k−1 i mi = 0. This implies aimi ∈ EM (0) so , it follows that a2k−1 i i=1 aimi ∈ hEM (0)i. The second inequality follows from Lemma 1. that m = Pr If R is a 2-primal ring, then we know that N (R) = Ns(R) = ER(0) = βco(R) = β(R). For modules, we have Theorem 4.5 below. Theorem 4.5. If M is a projective and 2-primal R-module, then Ns(M) = hEM (0)i = βco(M) = β(M). (4) (5) Hence, the zero submodule of M satisfies both the complete radical formula as well as the radical formula; and M satisfies the module analogue of Levitzki result for rings. If M is a projective and 2-primal module, then Ns(M) = βco(M) = β(M), by Proof: [1, Theorem 3.8] and the definition of 2-primal modules. Apply Lemma 2 to complete the proof. Corollary 4.6. For a projective module M over any one of the following rings: reduced rings, commutative rings, left-duo rings, symmetric rings, reversible rings, IFP rings, PSI rings, semi-symmetric rings and 2-primal rings; the equality Ns(M) = hEM (0)i = βco(M) = β(M) holds. Hence, the zero submodule of M satisfies both the complete radical formula as well as the radical formula; and M satisfies the module analogue of Levitzki result for rings. Proof: Any of the above mentioned rings is 2-primal, see a chart of implications in [14]. By [5, Corollary 2.1], a projective module over a 2-primal ring is 2-primal. The rest follows from Theorem 4.5. Definition 4.2. An R-module M is 9 1. Lee-Zhou reduced [11] if for all a ∈ R and every m ∈ M, am = 0 implies that Rm ∩ aM = 0. This is equivalent to saying that: for all a ∈ R and every m ∈ M, a2m = 0 implies that aRm = 0; 2. symmetric if for a, b ∈ R and m ∈ M, abm = 0 implies that bam = 0; 3. semi-symmetric if for all a ∈ R and every m ∈ M, a2m = 0 implies that (a)2m = 0 where (a) is the ideal of R generated by a ∈ R; 4. IFP (i.e., it has the insertion-of-factor-property) if whenever am = 0 for a ∈ R and m ∈ M, we have aRm = 0. Corollary 4.7. For each of the following modules: 1. M is 2-primal and free, 2. M is semi-symmetric and free, 3. M is semi-symmetric and projective, 4. M is IFP and projective, 5. M is IFP and free, 6. M is symmetric and projective, 7. M is symmetric and free, 8. M is reduced and projective, 9. M is reduced and free, 10. R is commutative and M is projective, 11. R is commutative and M is free; the equality Ns(M) = hEM (0)i = βco(M) = β(M) holds. Hence, the zero submodule of M satisfies both the complete radical formula as well as the radical formula; and M satisfies the module analogue of Levitzki result for rings. Proof: By [5, Theorems 2.2 and 2.3], and the fact that every free module is projective, each of these modules is 2-primal and projective. The rest follows from Theorem 4.5. Lemma 3 below can be proved with appropriate modification of methods used to prove [17, Theorem 1.5] for modules over commutative rings. 10 Lemma 3. Let φ : M → M ′ be an R-module epimorphism and let N be a submodule of M such that N ⊇ Ker φ. (i) If βs co(N) = hEM (N)i, then βs co(φ(N)) = hEM ′(φ(N))i; (ii) If N ′ is a submodule of M ′ and βs co(N ′) = hEM ′(N ′)i, then βs co(φ−1(N ′)) = hEM (φ−1(N ′))i. (iii) If βs(N) = hEM (N)i, then βs(φ(N)) = hEM ′(φ(N))i; (iv) If N ′ is a submodule of M ′ and βs(N ′) = hEM ′(N ′)i, then βs(φ−1(N ′)) = hEM (φ−1(N ′))i. Theorem 4.8. If the R-module M is any one of the modules given in Theorem 4.5 and Corollaries 4.6 and 4.7, then M satisfies both the complete radical formula as well as the radical formula. Proof. Let N be a submodule of M. The modules M given in Theorem 4.5 and Corollaries 4.6 and 4.7 are 2-primal and projective. Hence, βco(M) = hEM (0)i. When we apply Lemma 3, by letting M ′ = M/N and N ′ = N, we get βs co(N) = hEM/N (N)i = hEM (N)i, i.e., every submodule of M satisfies the complete radical formula. A similar argument starting with β(M) = hEM (0)i shows that every submodule of M satisfies the radical formula. Note that Theorem 4.8 retrieves the well known result that a projective module over a commutative ring satisfies the radical formula, see [9, Corollary 8]. Theorem 4.9. A finitely generated (and hence a cyclic) module over a 2-primal ring satisfies both the radical formula as well as the complete radical formula. Hence, it is 2-primal, satisfies the module analogue of Levitzki result for rings and equality (5) holds. Let R be a 2-primal ring. Since 2-primal rings are closed under direct sums Proof: (see [2]), the ring Rn for some n ∈ N is also 2-primal. By Corollary 4.3, Rn considered as an R-module satisfies both the complete radical formula as well as the radical formula. By Lemma 3, every homomorphic image of a module that satisfies the (complete) radical formula also satisfies the (complete) radical formula. Since a finitely generated R-module is a homomorphic image of Rn, it must also satisfy the (complete) radical formula. Theorem 4.9 retrieves an already known result: a finitely generated module over a prin- cipal ideal domain (resp. over a Dedekind domain) satisfies the radical formula, see [17, Theorem 2] (resp. [9, Theorem 9]). For rings, every semiprime (resp. completely semiprime) ideal of R is an intersection of prime (resp, completely prime) ideals. For modules, this is not true in general, see [9, p. 3600]. However, for modules that satisfy the complete radical formula, we have Proposition 6. 11 Proposition 6. An R-module M satisfies the complete radical formula if and only if every completely semiprime submodule N of M is an intersection of completely prime submodules of M. Proof: Suppose hEM (N)i = βs co(N) for every submodule N of M. If K is a completely semiprime submodule of M, then by Proposition 5, hEM (K)i = K such that by hypothe- sis, βs co(K) = K. This shows that K is an intersection of completely prime submodules of M. Conversely, suppose that K is an intersection of completely prime submodules of M. Then K is a completely semiprime submodule of M. By Proposition 5, hEM (K)i = K. It follows that K = hEM (K)i ⊆ βs co(K) ⊆ K and hence hEM (K)i = βs co(K). The property of the module being 2-primal allows the module to behave as though it is defined over a commutative ring. For modules over commutative rings, there is no distinction between modules that satisfy the radical formula and those that satisfy the complete radical formula. For 2-primal modules, we have Proposition 7. Proposition 7. If M is a 2-primal module, then the following statements are equivalent: 1. M satisfies the complete radical formula, 2. M satisfies the radical formula. Proof: If β(M) = βco(M), then hEM (0)i = β(M) if and only if hEM (0)i = βco(M). Definition 4.3. A ring R is left hereditary if every submodule of a projective R-module is projective. Semisimple rings, domains and path algebras over a quiver are examples of left heredi- tary rings. A ring R is semisimple if the regular module RR is a direct sum of simple submodules. Lemma 4. Let M be a projective module defined over a left hereditary ring R. Then for any submodule N of M, β(N) ⊆ hEN (0)i ⊆ βco(N). Proof: hereditary ring. N projective, Since M is projective, so is every submodule N of M by definition of a left implies β(N) = β(R)N. Let m ∈ β(N), then m = i=1 aini where ai ∈ β(R), ni ∈ N and k ∈ N. Since β(R) is nil, aini ∈ EN (0) ⊆ hEN (0)i. This shows that β(N) ⊆ hEN (0)i. The second inequality follows from Lemma 1. Pk Theorem 4.10. Let M be a projective module over a hereditary ring R. If a submodule N of M is 2-primal considered as a module, then N satisfies both the radical formula as well as the complete radical formula and hence β(N) = hEN (0)i = βco(N). 12 If N is 2-primal (as a module), then β(N) = βco(N). Now apply Lemma 4 Proof: to get β(N) = hEN (0)i = βco(N) and Lemma 3 to see that N satisfies both the radical formula as well as the complete radical formula. Corollary 4.11. Let M be a module over a semisimple ring R. If a submodule N of M is 2-primal considered as a module, then N satisfies both the radical formula as well as the complete radical formula. Proof: from Theorem 4.10. If R is a semisimple ring, then every R-module is projective. The rest follows Corollary 4.12. A semisimple commutative ring satisfies the radical formula. Since every module over a commutative ring R is 2-primal, applying Corollary Proof: 4.11 when R is semisimple shows that every submodule of an R-module satisfies the radical formula and hence R satisfies the radical formula. Corollary 4.13. A semisimple 2-primal ring satisfies the complete radical formula. Proof: A module M over a semisimple 2-primal ring R is projective and 2-primal by properties of semisimple rings and [5, Theorem 1] respectively. Since a semisimple ring is hereditary, every submodule N of such a module is also projective. N is 2-primal considered as a module by [5, Theorem 1]. By Theorem 4.10, N satisfies the complete radical formula. So, M satisfies the complete radical formula. Corollaries 4.12 and 4.13 give us another situation where 2-primal rings behave like com- mutative rings. Proposition 8. The following statements are equivalent: 1. a semisimple 2-primal ring satisfies the complete radical formula, 2. a semisimple 2-primal ring satisfies the radical formula. Let R be a semisimple 2-primal ring. Then any R-module M is 2-primal. By Proof: Proposition 7, M satisfies the complete radical formula if and only if it satisfies the radical formula. Example 3 shows that it is possible for a submodule to satisfy the complete radical formula when it neither satisfies the radical formula nor 2-primal. Example 3. Let M = (cid:26)(cid:18)¯0 ¯0 ¯1 ¯1(cid:19)(cid:27) where entries of matrices in M are from Z2 = {¯0, ¯1} and R = M2(Z). The zero submodule of the R-module M satisfies the complete radical formula, but M is neither 2-primal nor its zero submodule satisfies the radical formula. ¯1 ¯1(cid:19) ,(cid:18)¯1 ¯1 ¯0 ¯0(cid:19) ,(cid:18)¯0 ¯0 ¯0 ¯0(cid:19) ,(cid:18)¯1 ¯1 13 Proof: It suffices to show that 0 = β(M) $ βco(M) = hEM (0)i = M. Let r = (cid:18)a b c d(cid:19) ∈ R, rM = (cid:26)(cid:18)¯0 ¯0 ¯0 ¯0(cid:19) ,(cid:18)a a c c(cid:19) ,(cid:18)b d d(cid:19) ,(cid:18)a + b a + b c + d c + d(cid:19)(cid:27) ⊆ M b for any a, b, c, d ∈ Z. The would be non-trivial proper submodules, namely; N1 = (cid:26)(cid:18)¯0 ¯0 N3 = (cid:26)(cid:18)¯0 ¯0 ¯0 ¯0(cid:19) ,(cid:18)¯1 ¯1 ¯0 ¯0(cid:19) ,(cid:18)¯1 ¯1 ¯0 ¯0(cid:19)(cid:27), N2 = (cid:26)(cid:18)¯0 ¯0 ¯1 ¯1(cid:19)(cid:27) are not closed under multiplication by R since, for a and c ¯0 ¯0(cid:19) ,(cid:18)¯0 ¯0 ¯1 ¯1(cid:19)(cid:27) and odd, rN1 6⊆ N1, for b and d odd, rN2 6⊆ N2 and for a odd but b, c, d even, rN3 6⊆ N3. This shows that M is simple and hence prime. So, we have β(M) = 0. However, if we take a = (cid:18)3 3 2 2(cid:19)(cid:18)¯1 ¯1 (cid:18)3 3 2 2(cid:19) ∈ R and m = (cid:18)¯1 ¯1 ¯1 ¯1(cid:19) ∈ M, am = 0 but aM 6= 0 since a = ¯0 ¯0(cid:19) 6= 0. This shows that M is not completely prime. So, M has ¯0 ¯0(cid:19) = (cid:18)¯1 ¯1 no completely prime submodules, i.e., βco(M) = M. Note that 1. m0 = (cid:18)¯0 ¯0 2. m1 = (cid:18)¯1 ¯1 3. m2 = (cid:18)¯0 ¯0 4. m3 = (cid:18)¯1 ¯1 ¯0 ¯0(cid:19) = (cid:18)1 0 ¯0 ¯0(cid:19) = (cid:18)2 1 ¯1 ¯1(cid:19) = (cid:18)2 2 ¯1 ¯1(cid:19) = (cid:18)0 1 0 1(cid:19)(cid:18)¯0 ¯0 2 2(cid:19)(cid:18)¯0 ¯0 1 1(cid:19)(cid:18)¯1 ¯1 1 0(cid:19)(cid:18)¯0 ¯0 0 1(cid:19)2(cid:18)¯0 ¯0 2 2(cid:19)2(cid:18)¯0 ¯0 1 1(cid:19)2(cid:18)¯1 ¯1 ¯0 ¯0(cid:19) = (cid:18)¯0 ¯0 ¯0 ¯0(cid:19) and (cid:18)1 0 ¯0 ¯0(cid:19) ¯1 ¯1(cid:19) = (cid:18)¯0 ¯0 ¯0 ¯0(cid:19) ¯1 ¯1(cid:19) and (cid:18)2 1 ¯0 ¯0(cid:19) = (cid:18)¯0 ¯0 ¯0 ¯0(cid:19) ¯0 ¯0(cid:19) and (cid:18)2 2 ¯1 ¯1(cid:19) a linear combination of m1. This shows that hEM (0)i = M. Theorem 4.14. The following modules satisfy the complete radical formula: 1. a module with βco(M) = 0, (e.g., when M is completely prime); 2. a module with hEM (0)i = M (e.g., a module given in Example 3); 3. the regular module RR when R is 2-primal. Moreover, a module with βco(M) = 0 (and the regular module RR when R is 2-primal) is 2-primal, satisfies both the radical formula and a module analogue of Levitzki result for rings. 14 For each the three modules, βco(M) = hEM (0)i. Now, apply Lemma 3 to Proof: get the desired result. For the second part, since β(M) ⊆ βco(M) = 0 and Ns(M) ⊆ hEM (0)i ⊆ βco(M) = 0, we have β(M) = Ns(M) = hEM (0)i = βco(M). For the regular modules RR, apply Corollary 4.3 with the fact that β(R) = β(RR), βco(R) = βco(RR) and Ns(R) = Ns(RR). 5 Application of modules that satisfy the complete radical formula We know that a ring R is reduced if and only if it is a subdirect product of domains. In general, this structure theorem is not true in the module setting. This is due to the fact that not every completely semiprime submodule is an intersection of completely prime submodules. However, it holds when a module satisfies the complete radical formula, see Theorem 5.1. Definition 5.1. A module M is a subdirect product of the modules Sλ, λ ∈ Λ if there is an injective module homomorphism σ : M → S = Qλ∈Λ Sλ such that σ ◦ πλ is surjective for all λ ∈ Λ and for every canonical surjection πλ : S → Sλ. Theorem 5.1. Suppose that a module M satisfies the complete radical formula, then the following statements are equivalent: 1. M is completely semiprime, 2. hEM (0)i = 0, 3. βco(M) = 0, 4. M is a subdirect product of completely prime modules. Proof: 1 ⇔ 2. Follows from Proposition 5. 2 ⇔ 3. Since M satisfies the complete radical formula, hEM (0)i = βco(M). So, hEM (0)i = 0 if and only if βco(M) = 0. 3 ⇒ 4. Suppose that βco(M) = 0. Let {Nλ}λ∈Λ be a collection of all completely prime submodules of M. Then ∩λ∈ΛNλ = 0 and M is a subdirect product of modules M/Nλ, λ ∈ Λ which are completely prime. To see this, define σ : M → Qλ∈Λ(M/Nλ) by σ(m) = (m + Nλ)λ∈Λ. Then Ker σ = ∩λ∈Λ Ker πλ = ∩λ∈ΛNλ, σ ◦ πλ is surjective for every λ ∈ Λ and σ is injective if and only if ∩λ∈ΛNλ = 0. 15 4 ⇒ 3. Let M be a subdirect product of completely prime modules {Sλ}λ∈Λ, i.e., there is an injection σ : M → Qλ∈Λ Sλ with σ ◦ πλ surjective, where πλ : Qλ∈Λ Sλ → Sλ is the canonical surjection. Then, Ker(σ ◦ πλ) is a completely prime submodule of M. Hence, βco(M) ⊆ ∩λ∈ΛKer (σ ◦ πλ) = Ker (σ) = 0 and βco(M) = 0. Corollary 5.2. Let M be a module over a commutative ring. If M satisfies the radical formula, then the following statements are equivalent: 1. M is semiprime, 2. hEM (0)i = 0, 3. β(M) = 0, 4. M is a subdirect product of prime modules. Proof: For modules defined over commutative rings, prime (resp. semiprime) is indis- tinguishable from completely prime (resp. completely semiprime). Also, modules that satisfy the radical formula are indistinguishable from those that satisfy the complete rad- ical formula. Corollary 5.3. If a module M satisfies the complete radical formula, then the submodule βco(M) is the smallest completely semiprime submodule of M. If M satisfies the complete radical formula, then by Theorem 5.1, the zero Proof: submodule of M is a completely semiprime submodule of M if and only if βco(M) = 0. In this case, there is no completely semiprime submodule of M smaller than βco(M). Corollary 5.4. Suppose that M is a module defined over a commutative ring. If M sat- isfies the radical formula, then the submodule β(M) is the smallest semiprime submodule of M. Proof: For modules over commutative rings, β(M) = βco(M) and for a module M to satisfy the radical formula is equivalent to having M satisfy the complete radical formula. The notion of semiprime is indistinguishable from that of completely semiprime. Question 5.1. Can we have Corollary 5.2 for a module over a not necessarily commuta- tive ring? i.e., a module to be semiprime if and only if it is a subdirect product of prime modules. Note that a not necessarily commutative ring is semiprime if and only if it is a subdirect product of prime rings. Whereas we do not know the answer, we hasten to mention that 2 ⇔ 3 ⇔ 4 ⇒ 1 is easy to prove. So, the question reduces to checking whether for modules M that satisfy the radical formula, M semiprime implies β(M) = 0. Note that, this is not true in general, see Example 3. 16 6 A module over a 2-primal ring is 2-primal We have seen that a projective module over a 2-primal ring is 2-primal [5, Corollary 2.1], a finitely generated module (and hence a cyclic module) over a 2-primal ring is 2-primal [Theorem 4.9] and a module over a commutative ring (a commutative ring is 2-primal) is 2-primal. This further compels ones belief in the conjecture [5, Conjecture 2.1] which states that a module defined over a 2-primal ring is 2-primal. We show in Theorem 6.1 that this conjecture is true. Theorem 6.1. A module over a 2-primal ring is 2-primal. Let M be a module over a 2-primal ring R. We know that every module is Proof: a homomorphic image of a projective module. So, there exists a projective R-module P such that M is a homomorphic image of P . By [5, Corollary 2.1], P is a 2-primal module (since it is a projective module over a 2-primal ring). To complete the proof, it is enough to show that every homomorphic image of a 2-primal module is 2-primal. But this is easy to see since for every R-module epimorphism φ : M → M ′, we have φ(β(M)) = β(φ(M)) = β(M ′) and φ(βco(M)) = βco(φ(M)) = βco(M ′). Theorem 6.1 shows that 2-primal modules are abundant. 7 Questions in algebraic geometry Much of the algebraic geometry is done using commutative rings. Naturally, one wonders whether the algebraic geometry already known for commutative rings can be developed for non-commutative rings. However, there are two challenges in trying to achieve this objective. 1) Unlike commutative rings, non-commutative rings have fewer ideals and hence fewer prime ideals. As such, there is not usually a good topological space that reflects the ideal structure and representation theory of a given ring. Hence, defining a projective scheme as a ringed topological space on the homogeneous primes of a ring would not be useful. 2) There isn't a good theory of localization for non-commutative rings. So, any attempt to develop a non-commutative algebraic geometry based on rings and their localizations will not work, see [10] and [22]. Against this background together with the behaviour of 2-primal rings having tendencies of commutative rings, some questions come to mind. Question 7.1. Do 2-primal rings have as many ideals (and hence as many prime ideals) as the commutative rings so that it is possible and useful to define a projective scheme as a ringed topological space on its homogeneous prime ideals? Question 7.2. Can one develop a good theory of localization for 2-primal rings? In other words, is the theory of localization of 2-primal rings close to that of commutative rings 17 that one can be able to do with 2-primal rings almost all that is done with commutative rings as regards localization? An affirmative answer to any one of the two questions above will increase on the class of rings for which certain algebraic geometry can be done. Acknowledgement The author acknowledges support from Sida phase IV bilateral program with Makerere university, 2015–2020, project: 316-2014 and wishes to thank Prof. Rikard Bogvad for the hospitality while at Stockholm university and for introducing him to almost commutative rings. References [1] M. Behboodi, On the prime radical and Baer's lower nilradical of modules, Acta Math. Hungar., 122(3), (2009), 293–306. [2] G. F. Birkenmeier, H. E. Heatherly and E. K. Lee, Completely prime ideals and associated radicals, in Proc. Biennial Ohio State-Denison Conf., 1992, eds. S. K. Jain and S. T. Rizvi, World Scientific, Singapore, 102–129, 1993. [3] D. Buyruk and D. Pusat-Yilmaz, Modules over prufer domains which satisfy the radical formula, Glasgow Math. J. 49, (2007), 127–131. [4] J. Dauns, Prime modules, Reine Angew. Math. 298, (1978), 156–181. [5] N. J. Groenewald and D. Ssevviiri, 2-primal modules, J. Algebra Appl., 12, (2013), DOI: 10.1142/S021949881250226X. [6] N. J. Groenewald and D. Ssevviiri, Completely prime submodules, Int. Elect. J. Algebra, 13, (2013), 1–14. [7] L. A. Hugel, An introduction to Auslander-Reiten theory, Lecture notes, Advanced school on representation theory and related topics, ICTP Trieste, 2006.????? [8] L. Huishi, Structure sheaves on almost commutative algebras, J. Algebra, 226, (1999), 742–764. [9] J. Jenkins and F. Patrick Smith, On the prime radical of a module over a commutative ring, Comm. Algebra, 20, (1992), 3593–3602. [10] D. S. Keeler, The rings of non-commutative projective geometry, arXiv:math/0205005, 2002. 18 [11] T. K. Lee and Y. Zhou, Reduced Modules, Rings, Modules, Algebra and Abelian group, Lectures in Pure and Applied Mathematics, Vol. 236 Marcel Decker, New York, 365–377, 2004 [12] K. H. Leung and S. H. Man, On commutative Noetherian rings which satisfy the radical formula, Glasgow Math. J, 39 (1997), 285–293. [13] S. H. Man, On commutative Noetherian rings which satisfy the generalized radical formula, Comm. Algebra, 27(8) (1999), 4075–4088. [14] G. Marks, A taxonomy of 2-primal rings, J. Algebra, 266, (2003), 494–520. [15] G. Marks, On 2-primal Ore extensions, Comm. Algebra, 29(5), (2001), 2113–2123. [16] G. Marks, Skew polynomial rings over 2-primal, Comm. Algebra, 27(9), (1999), 4411– 4423. [17] R. L. McCasland and M. E. Moore, On radicals of submodules, Comm. Algebra, 19(5) (1991), 1327–1341. [18] A. Nikseresht and A. Azizi, On radical formula in modules, Glasgow Math. J., 53, (2011), 657–668, DOI:10.1017/S0017089511000243. [19] B. Sarac and Y. Tiras, On modules which satisfy the radical formula, Turk. J. Math. doi:10.3906/mat-1103-9. [20] H. Sharif, Y. Shari and S. Namazi, Rings satisfying the radical formula, Acta Math. Hungar., 71, (1996), 103–108. [21] G. Shin, Prime ideals and sheaf representation of a pseudo symmetric ring, Trans. Amer. Math. Soc., 32, (1972), 68–73. [22] S. P. Smith, An introduction to non-commutative projective algebraic geometry, Lecture notes under preparation (2016). [23] D. Ssevviiri, On completely prime submodules, Int. Elect. J. Algebra, 19, (2016), 77–90. [24] D. Ssevviiri, A relationship between 2-primal modules and modules that satisfy the radical formula, Int. Elect. J. Algebra, 18, (2015), 34–45. [25] O. Ungermann, The commutant of simple modules over almost commutative algebras, (2013), https://arxiv.org/abs/1311.5741v1. 19
1511.02732	1	1511	2015-11-09T16:04:58	The radius in matrix algebras--Examples and remarks	[ "math.RA" ]	The main purpose of this note is to illustrate how the radius in a finite-dimensional power-associative algebra over a field $\mathbb{F}$, either $\mathbb{R}$ or $\mathbb{C}$, may change when the multiplication in this algebra is modified. Our point of departure will be $\mathbb{F}^{n \times n}$, the familiar algebra of $n \times n$ matrices over $\mathbb{F}$ with the usual matrix operations, where it is known that the radius is the classical spectral radius. We shall alter the multiplication in $\mathbb{F}^{n \times n}$ in three different ways and compute, in each case, the radius in the resulting algebra.	math.RA	math	THE RADIUS IN MATRIX ALGEBRAS -- EXAMPLES AND REMARKS MOSHE GOLDBERG To my daughter Maya on her 40th birthday Abstract. The main purpose of this note is to illustrate how the radius in a finite-dimensional power-associative algebra over a field F, either R or C, may change when the multiplication in this algebra is modified. Our point of departure will be Fn×n, the familiar algebra of n × n matrices over F with the usual matrix operations, where it is known that the radius is the classical spectral radius. We shall alter the multiplication in Fn×n in three different ways and compute, in each case, the radius in the resulting algebra. 1. Introduction: the radius and its basic properties Let A be a finite-dimensional algebra over a field F, either R or C. We shall assume that A is power-associative, i.e., that the subalgebra of A generated by any one element is associative; thus ensuring that powers of each element in A are unambiguously defined. As usual, by a minimal polynomial of an element a in A we mean a monic polynomial of lowest positive degree with coefficients in F that annihilates a. With this familiar definition, we may cite: Theorem 1.1 ([G1, Theorem 1.1]). Let A be a finite-dimensional power-associative algebra over F. Then: (a) Every element a ∈ A possesses a unique minimal polynomial. (b) The minimal polynomial of a divides every other polynomial over F that annihilates a. Denoting the minimal polynomial of an element a ∈ A by pa, we follow [G1] and define the radius of a to be the nonnegative quantity r(a) = max{λ : λ ∈ C, λ is a root of pa}. The radius has been computed for elements in several well-known finite-dimensional power- associative algebras. For instance, it was recently shown in [GL1] that the radius in the Cayley -- Dickson algebras is given by the corresponding Euclidean norm. Another example emerged in [G1, page 4060], where it was established that if A is an arbitrary finite-dimensional matrix algebra over F with the usual matrix operations, then the radius of a matrix A ∈ A is given by the classical spectral radius, ρ(A) = max{λ : λ ∈ C, λ is an eigenvalue of A}. 2010 Mathematics Subject Classification. Primary 11C08, 16P10, 17A05, 17D05. Key words and phrases. Radius of an element in a finite-dimensional power-associative algebra, matrix alge- bras, standard matrix multiplication, Hadamard product in matrices, Jordan product in matrices. 1 2 MOSHE GOLDBERG With this last example in mind, we recall the following theorem which asserts that the radius retains some of the most basic properties of the spectral radius not only in finite-dimensional matrix algebras with the usual matrix operations, but in the general finite-dimensional power- associative case as well. Theorem 1.2 ([G1, Theorems 2.1 and 2.4]). Let A be a finite-dimensional power-associative algebra over F. Then: (a) The radius r is a nonnegative function on A. (b) The radius is homogeneous, i.e., for all a ∈ A and α ∈ F, r(αa) = αr(a). (c) For all a ∈ A and all positive integers k, r(ak) = r(a)k. (d) The radius vanishes only on nilpotent elements of A. (e) The radius is a continuous function on A.1 As a final introductory remark we mention that an analysis of the relevance of the radius to stability of subnorms and to the Gelfand formula can be found in [G1], [G2], [G3], and [GL1]. 2. Examples of radii in matrix algebras Our main purpose in this note is to illustrate how the radius in a finite-dimensional power- associative algebra may change when the multiplication in this algebra is modified. Selecting a positive integer n, n ≥ 2, our point of departure will be Fn×n, the familiar algebra of n × n matrices over F, either R or C, with the usual matrix operations. By what we already know about the radius in arbitrary finite-dimensional matrix algebras over F with the usual operations, we may register the following result which can also be derived directly from the fact that the roots of the minimal polynomial of a matrix A in Fn×n are the eigenvalues of A. Theorem 2.1. The radius of a matrix A in Fn×n is given by r(A) = ρ(A), where ρ denotes the spectral radius. The multiplication in Fn×n can be altered, of course, in a myriad of ways. Often, however, computing the radius in the newly obtained algebra will remain out of reach. In what follows, we shall modify the multiplication in Fn×n in three different ways, and calculate the radius in each case. 1Naturally, a real-valued function on a finite-dimensional algebra A is said to be continuous if it is continuous with respect to the (unique) finite-dimensional topology on A. THE RADIUS IN MATRIX ALGEBRAS -- EXAMPLES AND REMARKS 3 We embark on our plan by replacing the standard multiplication in Fn×n by the well-known Hadamard product which, for any two n × n matrices A = (αij) and B = (βij), is defined entry-wise by The resulting algebra, denoted by Fn×n A ◦ B = (αijβij). H , has been extensively studied in the literature (see for example Chapter 5 in [HJ] and the references at the end of that chapter). Obviously, Fn×n is distributive, commutative, and associative; and its unit element is given by E, the n × n matrix all of whose entries are 1. H Denoting the k-th power of a matrix A = (αij) in Fn×n H by A[k], we see that (2.1) A[k] = (ak ij), k = 1, 2, 3, . . . . Assisted by this observation, we can now post: Theorem 2.2. The radius of a matrix A = (αij) in Fn×n H is given by the sup norm of A, i.e., r(A) = max i,j αij. Proof. Select a matrix A = (αij) in Fn×n distinct entries of A (so that each αij equals precisely one of the ζl's). Let H , and let ζ1, . . . , ζs (1 ≤ s ≤ n2) be a list of all the be the minimal polynomial of A in Fn×n H , hence pA(t) = tm + αm−1tm−1 + · · · + α1t + α0 By (2.1), this can be equivalently written as A[m] + αm−1A[m−1] + · · · + α1A[1] + α0E = 0. l + αm−1ζ m−1 ζ m l + · · · + α1ζl + α0 = 0, l = 1, . . . , s. It follows that the ζl are roots of pA; and since these roots are distinct, we infer that the monic polynomial must divide pA. On the other hand, we notice that q(t) = (t − ζ1)(t − ζ2) · · · (t − ζs) (A − ζ1E) ◦ (A − ζ2E) ◦ · · · ◦ (A − ζsE) = 0; H . Appealing to Theorem 1.1(b), we conclude that pA must divide q; (cid:3) Another way of altering the standard multiplication in Fn×n is to replace it by the familiar so q annihilates A in Fn×n hence pA = q, and the rest of the proof follows without difficulty. Jordan product A · B = 1 2 (AB + BA), which turns Fn×n into the special Jordan algebra Fn×n+ (e.g., [J, page 4, Definition 2]). Since Fn×n is distributive, so is Fn×n+. Further, both Fn×n+ and Fn×n share the same unit element, the n × n identity matrix I. We observe, however, that Fn×n+, unlike Fn×n, is commutative. 4 MOSHE GOLDBERG Moreover, in contrast with Fn×n, the algebra Fn×n+ is not associative, nor even alternative.2 Indeed, consider the matrices 0 0(cid:19) ⊕ On−2, B = (cid:18)0 0 where On−2 is the (n − 2) × (n − 2) zero matrix. Then, A = (cid:18)0 1 1 0(cid:19) ⊕ On−2, 1 2 1 2 B 6= 0 = A · (B · B), (A · B) · B = 1 4 (AB2 + 2BAB + B2A) = BAB = and alternativity is shattered. Despite the fact that Fn×n+ is not alternative, it is power-associative. This is so because powers of matrices in Fn×n+ coincide with those in Fn×n, and hence are uniquely defined. Turning to compute the radius in Fn×n+, we realize that it is a simple task: Since Fn×n+ and Fn×n have an identical linear structure, and since raising to powers in Fn×n+ and Fn×n coincide, the minimal polynomials of a matrix A in Fn×n+ and in Fn×n are one and the same; thus, the radii of A in Fn×n+ and in Fn×n come to the same thing, yielding: Theorem 2.3. The radius of a matrix A in Fn×n+ is given by r(A) = ρ(A). This result is of particular interest, precisely because it tells us that altering the multiplication in Fn×n does not necessarily result in a different radius. In our last example, we shall modify the multiplication in Fn×n in a more intricate way, by introducing the product where A′ is the matrix obtained from A by replacing α1n, the (1, n) entry of A, by its negative, and where A′B′ is the usual product of A′ and B′ in Fn×n. A ∗ B = (A′B′)′, Denoting our new algebra by Fn×n ∗ , we remark that it is distributive and associative. Indeed, for all A, B, C ∈ Fn×n ∗ we have, A ∗ (B + C) = (A′(B + C)′)′ = (A′B′ + A′C ′)′ = (A′B′)′ + (A′C ′)′ = A ∗ B + A ∗ C and similarly, so the distributive laws are in the bag. Furthermore, since (A + B) ∗ C = A ∗ C + B ∗ C; (A ∗ B) ∗ C = (A′B′)′ ∗ C = ((A′B′)C ′)′ = (A′(B′C ′))′ = A ∗ (B′C ′)′ = A ∗ (B ∗ C), associativity holds as well. We also observe that the identity matrix I constitutes the unit element in Fn×n ∗ , since for all A ∈ Fn×n ∗ , A ∗ I = (A′I ′)′ = (A′I)′ = A, 2As usual, we call an algebra A alternative if the subalgebra generated by any two elements of A is associative. THE RADIUS IN MATRIX ALGEBRAS -- EXAMPLES AND REMARKS 5 and analogously, I ∗ A = A. Lastly, we note that since Fn×n is not commutative, neither is Fn×n . ∗ It seems interesting to mention that the algebra Fn×n possesses certain exotic properties which are not shared by either Fn×n, Fn×n contains nilpotent matrices which have nonzero eigenvalues. To substantiate this statement, let A = (αij) be the n × n matrix all of whose entries are zero except for α11, α1n, αn1, and αnn which are given by 1, −i, i, and -1, respectively. It is not hard to verify that A ∗ A = 0, so A is a nilpotent matrix of index 2 in Fn×n H , or Fn×n+. For instance, Fn×n . At the same time, we have ∗ ∗ ∗ so √2 and −√2 are eigenvalues of A. det(tI − A) = tn−2(t2 − 2), ∗ ∗ Another property of Fn×n which is not shared by our previous matrix algebras lies in the admits positive matrices whose squares are negative.3 For example, consider the fact that Fn×n n × n matrix A = (αij) where α1n = αn1 = 1 and the rest of the entries vanish. While A is positive, its squaring in Fn×n provides the negative matrix all of whose entries are zero, except for the first and last entries along its diagonal which equal -1. ∗ Turning to compute the radius in Fn×n ∗ , we denote the k-th power of a matrix A in this algebra by Ahki, and offer the following elementary observation. Lemma 2.1. If A ∈ Fn×n (2.2) ∗ , then Ahki = ((A′)k)′, k = 1, 2, 3, . . . , where (A′)k is the usual k-th power of A′ in Fn×n. Proof. For k = 1 the assertion is trivial. So assuming (2.2) for k, we get Ahk+1i = A ∗ Ahki = A ∗ ((A′)k)′ = (A′(A′)k)′ = ((A′)k+1)′, and we are done. (cid:3) With the above lemma in our grip, we may now proceed to record: Theorem 2.4. The minimal polynomial of a matrix A in Fn×n polynomial of A′ in Fn×n. ∗ coincides with the minimal Proof. Let be a polynomial over F that annihilates A in Fn×n ; that is, ∗ p(t) = αmtm + · · · + α1t + α0 By (2.2), this is equivalent to αmAhmi + · · · + α1Ah1i + α0I = 0. αm((A′)m)′ + · · · + α1(A′)′ + α0I ′ = 0; 3By a positive matrix we mean here a nonzero matrix all of whose entries are nonnegative. Similarly, a negative matrix is a nonzero matrix whose entries are all non-positive. 6 MOSHE GOLDBERG or in other words, to It follows that p annihilates A in Fn×n Theorem 1.1, the proof follows. ∗ αm(A′)m + · · · + α1A′ + α0I = 0. if and only if p annihilates A′ in Fn×n; so aided by (cid:3) An immediate consequence of Theorem 2.4 reads: Corollary 2.1. The radii of A in Fn×n ∗ and of A′ in Fn×n coincide. Finally, since the radius in Fn×n is the spectral radius, we get: Theorem 2.5. The radius of a matrix A in Fn×n ∗ is given by r(A) = ρ(A′). We conclude this note by pointing out that all our findings regarding Fn×n ∗ hold verbatim when the product is defined by where now, A′ is obtained from A by negating αn1, the (n, 1) entry of A. The author is truly grateful to Thomas Laffey for helpful discussions. A ∗ B = (A′B′)′ References [G1] Moshe Goldberg, Minimal polynomials and radii of elements in finite-dimensional power-associative algebras, Trans. Amer. Math. Soc. 359 (2007), no. 8, 4055 -- 4072. [G2] Moshe Goldberg, Radii and subnorms on finite-dimensional power-associative algebras, Linear Multilin- ear Algebra, 55 (2007), no. 5, 405 -- 415. [G3] Moshe Goldberg, Stable subnorms on finite-dimensional power-associative algebras, Electron. J. Linear Algebra 17 (2008), 359 -- 375. [GL1] Moshe Goldberg and Thomas J. Laffey, On the radius in Cayley -- Dickson algebras, Proc. Amer. Math. [HJ] [J] Soc. 143, (2015), no. 11, 4733 -- 4744. Roger A. Horn and Charles R. Johnson, Topics in Matrix Analysis, Cambridge Univ. Press, Cambridge 1991. Nathan Jacobson, Structure and representations of Jordan algebras, Amer. Math. Soc. Colloquium Publications, Vol. XXXIX, Amer. Math. Soc., Providence, R.I. 1968. Department of Mathematics, Technion -- Israel Institute of Technology, Haifa 32000, Israel E-mail address: [email protected]
1911.04193	1	1911	2019-11-11T11:36:48	Asymptotics for Capelli Polynomials with Involution	[ "math.RA" ]	Let $F\langle X, \ast \rangle$ be the free associative algebra with involution $\ast$ over a field $F$ of characteristic zero. We study the asymptotic behavior of the sequence of $\ast$-codimensions of the T-$\ast$-ideal $\Gamma_{M+1,L+1}^\ast$ of $F\langle X, \ast \rangle$ generated by the $\ast$-Capelli polynomials $Cap^\ast_{M+1} [Y,X]$ and $Cap^\ast_{L+1} [Z,X]$ alternanting on $M+1$ symmetric variables and $L+1$ skew variables, respectively. It is well known that, if $F$ is an algebraic closed field of characteristic zero, every finite dimensional $\ast$-simple algebra is isomorphic to one of the following algebras: \begin{itemize} \item [$\cdot$]$(M_{k}(F),t)$ the algebra of $k \times k$ matrices with the transpose involution; \item [$\cdot$]$(M_{2m}(F),s)$ the algebra of $2m \times 2m$ matrices with the symplectic involution; \item [$\cdot$]$(M_{h}(F)\oplus M_{h}(F)^{op}, exc)$ the direct sum of the algebra of $h \times h$ matrices and the opposite algebra with the exchange involution. \end{itemize} We prove that the $\ast$-codimensions of a finite dimensional $\ast$-simple algebra are asymptotically equal to the $\ast$-codimensions of $\Gamma_{M+1,L+1}^\ast$, for some fixed natural numbers $M$ and $L$. In particular: $$ c^{\ast}_n(\Gamma^{\ast}_{\frac{k(k+1)}{2} +1,\frac{k(k-1)}{2} +1})\simeq c^{\ast}_n((M_k(F),t)); $$ $$ c^{\ast}_n(\Gamma^{\ast}_{m(2m-1)+1,m(2m+1)+1})\simeq c^{\ast}_n((M_{2m}(F),s)); $$ and $$ c^{\ast}_n(\Gamma^{\ast}_{h^2+1,h^2+1})\simeq c^{\ast}_n((M_{h}(F)\oplus M_{h}(F)^{op},exc)). $$	math.RA	math	ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION F. S. BENANTI AND A. VALENTI Abstract. Let F hX, ∗i be the free associative algebra with involution ∗ over a field F of characteristic zero. We study the asymptotic behavior of the sequence of ∗- M +1,L+1 of F hX, ∗i generated by the ∗-Capelli poly- codimensions of the T-∗-ideal Γ∗ L+1[Z, X] alternanting on M + 1 symmetric variables M +1[Y, X] and Cap∗ nomials Cap∗ and L + 1 skew variables, respectively. It is well known that, if F is an algebraic closed field of characteristic zero, every finite dimensional ∗-simple algebra is isomorphic to one of the following algebras: · (Mk(F ), t) the algebra of k × k matrices with the transpose involution; · (M2m(F ), s) the algebra of 2m × 2m matrices with the symplectic involution; · (Mh(F ) ⊕ Mh(F )op, exc) the direct sum of the algebra of h × h matrices and the opposite algebra with the exchange involution. We prove that the ∗-codimensions of a finite dimensional ∗-simple algebra are asymp- M +1,L+1, for some fixed natural numbers M totically equal to the ∗-codimensions of Γ∗ and L. In particular: n(Γ∗ c∗ k(k+1) 2 +1, k(k−1) 2 +1 ) ≃ c∗ n((Mk(F ), t)); and c∗ n(Γ∗ m(2m−1)+1,m(2m+1)+1 ) ≃ c∗ n((M2m(F ), s)); c∗ n(Γ∗ h2+1,h2+1) ≃ c∗ n((Mh(F ) ⊕ Mh(F )op, exc)). 9 1 0 2 v o N 1 1 ] . A R h t a m [ 1 v 3 9 1 4 0 . 1 1 9 1 : v i X r a 1. Introduction 1, x2, x∗ 1, · · · , xn, x∗ Let (A, ∗) be an algebra with involution ∗ over a field F of characteristic zero and let F hX, ∗i = F hx1, x∗ 2, . . .i denote the free associative algebra with involution ∗ gen- erated by the countable set of variables {x1, x∗ 2, . . .} over F . Recall that an ele- ment f (x1, x∗ n) of F hX, ∗i is a ∗-polynomial identity (or ∗-identity) for A if f (a1, a∗ n) = 0, for all a1, . . . , an ∈ A. We denote by Id∗(A) the set of all ∗- polynomial identities satisfied by A which is a T-∗-ideal of F hX, ∗i, i.e., an ideal invariant under all endomorphisms of F hX, ∗i commuting with the involution of the free algebra. For Γ = Id∗(A) we denote by var∗(Γ) = var∗(A) the variety of ∗-algebras having the elements of Γ as ∗-identities. 1, · · · , an, a∗ 1, x2, x∗ It is well known that in characteristic zero Id∗(A) is completely determinated by the multilinear ∗-polynomials it contains. To the T-∗-ideal Γ = Id∗(A) one can associates a numerical sequence called the sequence of ∗-codimensions c∗ n(A) which is the main tool for the quantitative investigation of the ∗-polynomial identities of A. Recall that c∗ n(A), n = 1, 2, . . ., is the dimension of the space of multilinear polynomial in n-th variables in the corresponding relatively free algebra with involution of countable rank. Thus, if we denote by P ∗ n the space of all multilinear polynomials of degree n in x1, x∗ 1, · · · , xn, x∗ n(Γ) = c∗ n then n(A) = dimP ∗ c∗ n (A) = dim P ∗ n P ∗ n ∩ Id∗(A) . 2000 Mathematics Subject Classification. Primary 16R10; Secondary 16P90. Key words and phrases. Algebras with involution, Capelli polynomials, Codimension, Growth. The authors were partially supported by INDAM-GNSAGA of Italy. 1 2 BENANTI AND VALENTI A celebrated theorem of Amitsur [2] states that if an algebra with involution satisfies a ∗-polynomial identity then it satisfies an ordinary polynomial identity. At the light of this result in [15] it was proved that, as in the ordinary case, if A satisfies a non trivial ∗-polynomial identity then c∗ n(A) is exponentially bounded, i.e. there exist constants a and b such that c∗ n(A) ≤ abn, for all n ≥ 1. Later (see [3]) an explicit exponential bound for c∗ n(A) was exhibited and in [18] a characterization of finite dimensional algebras with invo- lution whose sequence of ∗-codimensions is polynomial bounded was given. This result was extended to non-finite dimensional algebras (see [12]) and ∗-varieties with almost polynomial growth were classified in [11] and [21]. The asymptotic behavior of the ∗-codimensions was determined in [6] in case of matrices with involution. Recently (see [14]), for any algebra with involution, it was studied the exponential be- havior of c∗ n(A), and it was showed that the ∗-exponent of A exp∗(A) = lim n→∞ n(A) npc∗ exists and is a non negative integer. ∗-exponent was proved in [17] for finite dimensional algebra with involution. It should be mentioned that the existence of the Now, if f ∈ F hX, ∗i we denote by hf i∗ the T-∗-ideal generated by f . Also for a set of polynomials V ⊂ F hX, ∗i we write hV i∗ to indicate the T-∗-ideal generated by V . An interesting problem in the theory of PI-algebras with involution ∗ is to describe the T-∗-ideals of ∗-polynomial identities of ∗-simple finite dimensional algebras. Recall that, if F is an algebraically closed field of characteristic zero, then, up to isomorphisms, all finite dimensional ∗-simple are the following ones (see [22], [16]): · (Mk(F ), t) the algebra of k × k matrices with the transpose involution; · (M2m(F ), s) the algebra of 2m × 2m matrices with the symplectic involution; · (Mh(F ) ⊕ Mh(F )op, exc) the direct sum of the algebra of h × h matrices and the opposite algebra with the exchange involution. The aim of this paper is to find a relation among the asymptotics of the ∗-codimensions of the finite dimensional ∗-simple algebras and the T-∗-ideals generated by the ∗-Capelli polynomials Cap∗ M+1[Y, X] and Cap∗ L+1[Z, X] alternanting on M + 1 symmetric variables and L + 1 skew variables, respectively. More precisely, if (A, ∗) is any algebra with involution ∗, let A+ = {a ∈ A a∗ = a} and A− = {a ∈ A a∗ = −a} denote the subspaces of symmetric and skew elements of A, respectively. Since charF =0, we can regard the free associative algebra with involution F hX, ∗i as generated by symmetric and skew variables. In particular, for i = 1, 2, . . ., we let yi = xi + x∗ i , then we write X = Y ∪ Z as the disjoint union of the set Y of symmetric variables and the set Z of skew variables and F hX, ∗i = F hY ∪ Zi. Hence a polynomial f = f (y1, . . . , ym, z1, . . . , zn) ∈ F hY ∪ Zi is a ∗-polynomial identity of A if and only if f (a1, . . . , am, b1, . . . , bn) = 0 for all ai ∈ A+, bi ∈ A−. i and zi = xi − x∗ Let us recall that, for any positive integer m, the m-th Capelli polynomial is the element of the free algebra F hXi defined as Capm(t1, . . . , tm; x1, . . . , xm−1) = = Xσ∈Sm (sgnσ)tσ(1)x1tσ(2) · · · tσ(m−1)xm−1tσ(m) where Sm is the symmetric group on {1, . . . , m}. In particular we denote by Cap∗ m[Y, X] = Capm(y1, . . . , ym; x1, . . . , xm−1) and Cap∗ m[Z, X] = Capm(z1, . . . , zm; x1, . . . , xm−1) ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 3 the m-th ∗-Capelli polynomial in the alternating symmetric variables y1, . . . , ym and skew variables z1, . . . , zm, respectively (x1, . . . , xm−1 are arbitrary variables). Let Cap+ m denote the set of 2m−1 polynomials obtained from Capm[Y, X] by deleting any subset of variables xi (by evaluating the variables xi to 1 in all possible way). Similarly, we define by Cap− m the set of 2m−1 polynomials obtained from Capm[Z, X] by deleting any subset of variables xi. If L and M are two natural numbers, we denote by Γ∗ M+1, Cap− M+1,L+1. the T-∗-ideal generated by the polynomials Cap+ = var∗(ΓM+1,L+1) for the ∗-variety generated by Γ∗ L+1i M+1,L+1 = hCap+ L+1. We also write U ∗ M+1, Cap− M+1,L+1 an bn U ∗ In this paper we study the asymptotic behavior of the sequence of ∗-codimensions of M+1,L+1. Recall that two sequences an, bn, n = 1, 2, . . ., are asymptotically equal, an ≃ bn, if = 1. In the ordinary case (no involution) (see [19]) it was proved the asymptotic limn→+∞ equality between the codimensions of the Capelli polynomials Capk2+1 and the codimensions of the matrix algebra Mk(F ). In [4] these result was extended to finite dimensional simple superalgebras proving that the graded codimensions of the T2-ideal generated by the graded Capelli polynomials ΓM+1,L+1, for some fixed M , L, are asymptotically equal to the graded codimensions of a simple finite dimensional superalgebra. The link between the asymptotic of the codimensions of the Amitsur's Capelli-type polynomials and the verbally prime algebras was studied in [5]. Here we characterize the T-∗-ideal of ∗-identities of any ∗-simple finite dimensional algebra showing that Γ∗ k(k+1) 2 +1, = Id∗((Mk(F ), t) ⊕ D′); k(k−1) 2 +1 m(2m−1)+1,m(2m+1)+1 = Id∗((M2m(F ), s) ⊕ D′′); Γ∗ h2+1,h2+1 = Id∗((Mh(F ) ⊕ Mh(F )op, exc) ⊕ D′′′) Γ∗ where D′, D′′ and D′′′ are finite dimensional ∗-algebra with exp∗(D′) < k2, exp∗(D′′) < (2m)2 and exp∗(D′′′) < 2h2. It follows that asymptotically c∗ n(Γ∗ k(k+1) 2 +1, k(k−1) 2 +1 ) ≃ c∗ n((Mk(F ), t)); c ∗ n(Γ∗ m(2m−1)+1,m(2m+1)+1) ≃ c∗ n((M2m(F ), s)); c ∗ n(Γ∗ h2+1,h2+1) ≃ c∗ n((Mh(F ) ⊕ Mh(F )op, exc)). 2. Preliminaries Let F be a field of characteristic zero and let G be the Grassmann algebra over F generated by the elements e1, e2, . . . subject to the following condition eiej = −ejei, for all i, j ≥ 1. Recall that G has a natural Z2-grading G = G0 ⊕ G1 where G0 (resp. G1) is the span of the monomials in the e′ is of even length (resp. odd length). If B = B0 ⊕ B1 is a superalgebra, then the Grassmann envelope of B is defined as G(B) = (G0 ⊗ B0) ⊕ (G1 ⊗ B1). The relevance of G(A) relies in a result of Kemer ([20, Theorem 2.3]) stating that if B is any PI-algebra, then its T-ideal of polynomial identities coincides with the T-ideal of identities 4 BENANTI AND VALENTI of the Grassmann envelope of a suitable finite dimensional superalgebra. This result has been extended to algebras with involution in [1] and the following result holds Theorem 1. If A is a P I-algebra with involution over a field F of characteristic zero, then there exists a finite dimensional superalgebra with superinvolution B such that Id∗(A) = Id∗(G(B)). 0 L B− 0 L B+ 1 L B− 1 . Recall that a superinvolution ∗ of B is a linear map of B of order two such that (ab)∗ = (−1)abb∗a∗, for any homogeneous elements a, b ∈ B, where a denotes the homogeneous degree of a. 1 ⊆ B1 and we decompose B = B+ It is well known that in this case B∗ 0 ⊆ B0, B∗ We can define a superinvolution ∗ on G by requiring that e∗ i = −ei, for any i ≥ 1. Then it is easily checked that G0 = G+ and G1 = G−. Now, if B is a superalgebra one can perform its Grassmann envelope G(B) and in [1] it was shown that if B has a superinvolution ∗ we can regard G(B) as an algebra with involution by setting (g ⊗a)∗ = g∗ ⊗a∗, for homogeneous elements g ∈ G, a ∈ B. By making use of the previous theorem, in [14] it was proved the existence of the ∗- exponent of a P I-algebra with involution A and also an explicit way of computing exp∗(A) was given. More precisely if B is a finite dimensional algebra with superinvolution over an algebraic closed field of characteristic zero, then by [13] we write B = ¯B + J where ¯B is a maximal semisimple superalgebra with induced superinvolution and J = J(B) = J ∗. Also we can write ¯B = B1 ⊕ · · · ⊕ Bk where B1, · · · , Bk are simple superalgebras with induced superinvolution. We say that a subalgebra Bi1 ⊕· · ·⊕Bit, where Bi1 , . . . , Bit are distinct simple components, is admissible if for some permutation (l1, . . . , lt) of (i1, . . . , it) we have that Bl1JBl2 J · · · JBlt 6= 0. Moreover if Bi1 ⊕ · · · ⊕ Bit is an admissible subalgebra of B then B′ = Bi1 ⊕ · · · ⊕ Bit + J is called a reduced algebra. In [14] it was proved that exp∗(A) = exp∗(G(B)) = d where d is the maximal dimension of an admissible subalgebra of B. It follows immediately that Remark 1. If A is a ∗-simple algebra then exp∗(A) = dimF A. We next prove that the reduced algebras are basic elements of any ∗-variety. We start with the following Lemma 1. Let A and B be algebras with involution satisfying a ∗-polynomial identity. Then c∗ n(A), c∗ n(B) ≤ c∗ n(A ⊕ B) ≤ c∗ n(A) + c∗ n(B). Hence the exp∗(A ⊕ B) = max{exp∗(A), exp∗(B)}. Proof. It follows easily from the proof of the Lemma 1 in [19]. If V = var∗(A) is the variety of ∗-algebras generated by A we write Id∗(V) = Id∗(A), c∗ n(V) = c∗ n(A) and exp∗(V) = exp∗(A). We have the following Theorem 2. Let V be a proper variety of ∗-algebras. Then there exists a finite number of reduced superalgebras with superinvolution B1, . . . , Bt and a finite dimensional superalgebra with superinvolution D such that with exp∗(V) = exp∗(G(B1)) = · · · = exp∗G((Bt)) and exp∗(G(D)) < exp∗(V). V = var(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)) ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 5 Proof. The proof follows closely the proofs given in [19, Theorem 1] and in [4, Theorem 3]. Let A be a ∗-PI-algebra such that V = var∗(A). By Theorem 1, there exists a finite dimensional superalgebra with superinvolution B such that Id∗(A) = Id∗(G(B)). Also, by [13, Theorem 4.1], we may assume that B = ¯B1 ⊕ · · · ⊕ ¯Bs + J(B), where ¯Bi are simple ∗-superalgebras and J ∗ = J is the Jacobson radical of B. Let exp∗(A) = d. Then, since d is the maximal dimension of an admissible subalgebra of A there exist distinct ∗-simple superalgebras ¯Bj1 , . . . ¯Bjk such that ¯Bj1 J · · · J ¯Bjk 6= 0 and dimF ( ¯Bj1 ⊕ · · · ⊕ ¯Bjk ) = d. Let Γ1, . . . , Γt be all possible subset of {1, . . . , s} such that, if Γj = {j1, . . . , jk}, then dimF ( ¯Bj1 ⊕ · · · ⊕ ¯Bjk ) = d and ¯Bσ(j1)J · · · J ¯Bσ(jk) 6= 0 for some permutation σ ∈ Sk. For any such Γj, j = 1, . . . t, then we put Bj = ¯Bj1 ⊕ · · · ⊕ ¯Bjk + J. It follows, by the characterization of the ∗-exponent, that exp∗(G(B1)) = · · · = exp∗(G(Bt)) = d = exp∗(G(B)). Let D = D1 ⊕ · · · ⊕ Dp, where D1, . . . , Dp are all subsuperalgebras of B with superinvolution of the type ¯Bi1 ⊕ · · · ⊕ ¯Bir + J, with 1 ≤ i1 < · · · < ir ≤ s and dimF ( ¯Bi1 ⊕ · · · ⊕ ¯Bir ) < d. Then, by Lemma 1, we have exp∗(G(D)) < exp∗(G(B)). Now, we want to prove that var∗(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)) = var∗(G(B)). Since G(D), G(Bi) ∈ var∗(A), ∀i = 1, . . . , t, it follows that var∗(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)) ⊆ var∗(G(B)). 1 , . . . , y+ 1 , . . . , y− 1 , . . . , z− 1 , . . . , z+ m, z+ p , z− n , y− Now, let f = f (y+ q ) be a multilinear polyno- mial such that f 6∈ Id∗(G(B)). We shall prove that f 6∈ Id∗(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)). Since f 6∈ Id∗(G(B)), there exist a+ 1,0 ⊗ g1,0, . . . , a+ a− 1,0 ⊗ h1,0, . . . , a− b− 1,1 ⊗ g1,1, . . . , b− n,0 ⊗ gn,0 ∈ G(B)+ m,0 ⊗ hm,0 ∈ G(B)− p,1 ⊗ gp,1 ∈ G(B)+ 0 = B+ 0 = B− 1 = B− 0 ⊗ G0, 1 ⊗ G1, 0 ⊗ G0, b+ 1,1 ⊗ h1,1, . . . , b+ q,1 ⊗ hq,1 ∈ G(B)− 1 = B+ 1 ⊗ G1 1,0 ⊗ g1,0, . . . , a+ f (a+ b− 1,1 ⊗ g1,1, . . . , b− m,0 ⊗ hm,0, q,1 ⊗ hq,1) 6= 0. n,0 ⊗ gn,0, a− 1,0 ⊗ h1,0, . . . , a− p,1 ⊗ gp,1, b+ 1,1 ⊗ h1,1, . . . , b+ and such that It follows that 0 6= f (a+ 1,0 ⊗ g1,0, . . . , a+ n,0 ⊗ gn,0, a− 1,0 ⊗ h1,0, . . . , a− m,0 ⊗ hm,0, b− 1,1 ⊗ g1,1, . . . , b− f (a+ 1,0, . . . , a+ n,0, a− p,1 ⊗ gp,1, b+ 1,0, . . . , a− 1,1 ⊗ h1,1, . . . , b+ p,1, b+ 1,1, . . . , b− m,0, b− q,1 ⊗ hq,1) = 1,1, . . . , b+ q,1)⊗ g1,0 · · · gn,0h1,0 · · · hm,0g1,1 · · · gp,1h1,1 · · · hq,1 where f is the multilinear polynomial introduced in [14, Lemma 1]. Clearly f 6= 0. From the linearity of f we can assume that a+ j,1 ∈ B1 ∪ · · · ∪ Bs ∪ J. Since BiBj = 0 for i 6= j, from the property of the ∗-exponent described above, we have that i,0 and b+ j,1, a− i,0, b− a+ 1,0, . . . , a+ 1,1, . . . , b+ for some Bj1 , . . . , Bjk such that dimF (Bj1 ⊕ · · · ⊕ Bjk ) ≤ d. 1,0, . . . , a− 1,1, . . . , b− m,0, b− n,0, a− p,1, b+ q,1 ∈ Bj1 ⊕ · · · ⊕ Bjk + J 6 BENANTI AND VALENTI Thus f is not an identity for one of the algebras G(B1), . . . , G(Bt), G(D). Hence f 6∈ Id∗(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)). In conclusion var∗(G(B)) ⊆ var∗(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)) and the proof is complete. An application of Theorem 2 is given in terms of ∗-codimensions. Corollary 1. Let V = var∗(A) be a proper variety of ∗-algebras. Then there exists a finite number of reduced superalgebras with superinvolution B1, . . . , Bt and a finite dimensional superalgebra with superinvolution D such that c∗ n(A) ≃ c∗ n(G(B1) ⊕ · · · ⊕ G(Bt)). Proof. By Theorem 2, there is a finite number of reduced superalgebras with superinvolu- tion B1, . . . , Bt such that V = var∗(A) = var∗(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)) with exp∗(A) = exp∗(G(B1)) = · · · = exp∗(G(Bt)) and exp∗(G(D)) < exp∗(A). Then, by Lemma 1 c∗ n(G(B1) ⊕ · · · ⊕ G(Bt)) ≤ c∗ n(G(B1) ⊕ · · · ⊕ G(Bt) ⊕ G(D)) ≤ Recalling that exp∗(G(D)) < exp∗(G(B1)) = exp∗(G(B1) ⊕ · · · ⊕ G(Bt)) we have that c∗ n(G(B1) ⊕ · · · ⊕ G(Bt)) + c∗ n(G(D)). and the proof of the corollary is complete. c∗ n(A) ≃ c∗ n(G(B1) ⊕ · · · ⊕ G(Bt)) If A is a finite dimensional ∗-algebra we obtain a simplified form of the previous theorem and corollary. Let us recall that an algebra with involution can be regarded as a superalgebra with superinvolution with trivial grading. We have the following Corollary 2. Let A be a finite dimensional ∗-algebra. Then there exists a finite number of reduced ∗-algebras B1, . . . , Bt and a finite dimensional ∗-algebra D such that var∗(A) = var∗(B1 ⊕ · · · ⊕ Bt ⊕ D) n(A) ≃ c∗ c∗ n(B1 ⊕ · · · ⊕ Bt) and exp∗(A) = exp∗(B1) = · · · = exp∗(Bt), exp∗(D) < exp∗(A). The following results give us a characterization of the ∗-varieties satisfying a Capelli identity. Let's start with the Remark 2. Let M and L be two natural numbers. satisfying the ∗-Capelli polynomials Cap∗ identity CapM+L(x1, . . . , xM+L; ¯x1, . . . , ¯xM+L−1). M [Y, X] and Cap∗ If A is an algebra with involution L[Z, X], then A satisfies the Capelli ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 7 Proof. To obtain the thesis it is sufficient to observe that CapM+L(x1, . . . , xM+L; ¯x1, . . . , ¯xM+L−1) = CapM+L( x1 + x∗ 1 2 + x1 − x∗ 1 2 , . . . , xM+L + x∗ M+L 2 + xM+L − x∗ M+L 2 ; ¯x1, . . . , ¯xM+L−1) is a linear combinations of ∗-Capelli polynomials alternating either in m ≥ M symmetric variables or in l ≥ L skew variables. The proof of the next result follows closely the proof given in [16, Theorem 11.4.3] Theorem 3. Let V be a variety of ∗-algebras. If V satisfies the Capelli identity of some rank then V = var∗(A), for some finitely generated ∗-algebra A. Let M , L be two natural numbers. Let A = A+ ⊕ A− be a generating ∗-algebra of U ∗ M+1,L+1. By remark 2, A satisfies a Capelli identity. Hence by the previous theorem, we may assume that A is a finitely generated ∗-algebra. Moreover by [23, Theorem 1] we may consider A as a finite-dimensional ∗-algebra. Since any polynomial alternating on M + 1 symmetric variables vanishes in A (see [16, Proposition 1.5.5]), we get that dim A+ ≤ M . Similarly we get that dim A− ≤ L and exp∗(A) ≤ dim A ≤ M + L. Thus we have the following Lemma 2. exp∗(U ∗ M+1,L+1) ≤ M + L. 3. The algebra U T ∗(A1, . . . , An) In this section we recall the construction of the ∗-algebra U T ∗(A1, . . . , An) given in Section 2 of [7]. Let A1, . . . , An be a n-tuple of finite dimensional ∗-simple algebras, then Ai = (Mdi, µi), where µi is the transpose or the symplectic involution, or Ai = (Mdi ⊕ M op di , exc), where exc is the exchange involution. Let γd be the orthogonal involution defined on the matrix algebra Md(F ) by putting, for all a ∈ Md(F ), where aγd = g−1atg = gatg, g = 0 1 . . . · . . . · 1 0 ·     and at is the transposed of the matrix a. γd acts on matrix units epq of Md by sending it to eγd pq = ed−q+1,d−p+1 (it is the reflection along the secondary diagonal). i=1 dimF Ai, then we have an embedding of ∗-algebras If d = Pn defined by ∆ : n Mi=1 Ai → (M2d(F ), γ2d) 8 BENANTI AND VALENTI (a1, . . . , an) → ¯a1 . . . ¯an ¯bn   . . . µiγdi i ¯b1 where, if ai ∈ Ai = (Mdi, µi), then ¯ai = ai and ¯bi = a (Mdi ⊕ M op di , exc), then ¯ai = ai and ¯bi = bi. , and if ai = (ai, bi) ∈ Ai = Let denote by D = D(A1, . . . , An) ⊆ M2d(F ) the ∗-algebra image of Ln let U be the subspace of M2d(F ) so defined: i=1 Ai by ∆ and     0 U12 . . . · · · . . . 0 U1t ... Ut−1t 0 0 Utt−1 . . . · · · Ut1 ... . . . U21 0 0   where, for 1 ≤ i, j ≤ n, i 6= j, Uij denote the vector space of the rectangular matrices of dimensions di × dj. Let define U T ∗(A1, . . . , An) = D ⊕ U ⊆ M2d(F ) (see section 2 of [7]). It is easy to show that U T ∗(A1, . . . , An) is a subalgebra with involution of (M2d(F ), γ2d) in which the algebras Ai are embedded as ∗-algebras and whose ∗-exponent is given by exp∗(U T ∗(A1, . . . , An)) = n Xi=1 dimF Ai. In [10] and [8] the link between the degrees of ∗-Capelli polynomials and the ∗-polynomial identities of U T ∗(A1, . . . , An) was investigated. If we set d+ := Pn (see [9]) i=1 dimF A+ i and d− := Pn i=1 dimF A− i , then the following result applies Lemma 3. Let R = U T ∗(A1, . . . , An). Then Cap∗ and only if M ≥ d+ + n and L ≥ d− + n. M [Y, X] and Cap∗ L[Z, X] are in Id∗(R) if 4. Asymptotics for U ∗ k(k+1) 2 +1, k(k−1) 2 +1 and (Mk(F ), t) Let A = ¯A ⊕ J where ¯A is a ∗-simple finite dimensional algebra and J = J(A) is its Jacobson radical. It is well known that the Jacobson radical J is a ∗-ideal of A. We start with the following key lemmas that hold for any ∗-simple finite dimensional algebra. ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 9 Lemma 4. Let A = ¯A ⊕ J where ¯A is a ∗-simple finite dimensional algebra and J = J(A) is its Jacobson radical. Then J can be decomposed into the direct sum of four ¯A-bimodules J = J00 ⊕ J01 ⊕ J10 ⊕ J11 where, for p, q ∈ {0, 1}, Jpq is a left faithful module or a 0-left module according to p = 1, or p = 0, respectively. Similarly, Jpq is a right faithful module or a 0-right module according to q = 1 or q = 0, respectively. Moreover, for p, q, i, l ∈ {0, 1}, JpqJql ⊆ Jpl, JpqJil = 0 for q 6= i and there exists a finite dimensional nilpotent ∗-algebra N such that J11 ∼= ¯A ⊗F N (isomorphism of ¯A-bimodules and of ∗-algebras). Proof. It follows from the proof of Lemma 2 in [19]. Notice that J00 and J11 are stable under the involution whereas J ∗ 01 = J10. Lemma 5. Let ¯A be a ∗-simple finite dimensional algebra. Let M = dimF ¯A+ and L = dimF ¯A−. Then ¯A does not satisfy Cap∗ M [Y, X] and Cap∗ L[Z, X]. Proof. The result follows immediately from [9, Lemma 3.1]. From now on we assume that A = Mk(F ) + J, where J = J(A) is the Jacobson radical of the finite dimensional ∗-algebra A and (Mk(F ), t) is the ∗-algebra of matrices with transpose involution. Lemma 6. Let M = k(k + 1)/2 and L = k(k − 1)/2 with k ∈ N, k > 0. If Γ∗ Id∗(A), then J10 = J01 = (0). M+1,L+1 ⊆ Proof. By Lemma 5, Mk(F ) does not satisfy the ∗-Capelli polynomial Cap∗ there exist elements a+ M [Y, X]. Then, 1 , . . . , a+ Cap∗ M ∈ Mk(F )+ and b1, . . . , bM−1 ∈ Mk(F ) such that M (a+ M ; b1, . . . , bM−1) = e1,k, 1 , . . . , a+ where the ei,j's are the usual matrix units. Let d ∈ J01, then d∗ ∈ J10 and d + d∗ ∈ (J01 ⊕ J10)+. Since Γ∗ M+1,L+1 ⊆ Id∗(A) we have 0 = Cap∗ M+1(a+ 1 , . . . , a+ M , d + d∗; b1, . . . , bM−1, ek,k) = de1,k ± e1,kd∗. Hence de1,k ± e1,kd∗ = 0 and, so, de1,k = ∓e1,kd∗ ∈ J01 ∩ J10 = (0). Then d = 0, for all d ∈ J01. Thus J01 = (0) and J10 = (0). Lemma 7. Let M = k(k + 1)/2 and L = k(k − 1)/2 with k ∈ N, k > 0. Let J11 ∼= Mk(F ) ⊗F N , as in Lemma 4. If ΓM+1,L+1 ⊆ Id∗(A), then N is commutative. Proof. Let N be the finite dimensional nilpotent ∗-algebra of the Lemma 4. Write N = N + ⊕ N −, where N + and N − denote the subspaces of symmetric and skew elements of N respectively. Let e+ M be an ordered basis of Mk(F )+ consisting of symmetric matrices e+ 1 = e1,1 and let a0, a1, . . . , aM ∈ Mk(F ) be such that h ∈ {ei,i i = 1, . . . k} ∪ {ei,j + ej,i i < j, i, j = 1, . . . k} such that e+ 1 , . . . , e+ a0e+ 1 a1 · · · aM−1e+ M aM = e1,1 and for any σ ∈ SM , σ 6= id. a0e+ σ(1)a1 · · · aM−1e+ σ(M)aM = 0 10 BENANTI AND VALENTI Consider d+ 1 , d+ 2 ∈ N + and set c+ 1 = e1,1d+ 1 and c+ 2 = e1,1d+ 2 . Notice that, since N commutes with Mk(F ), c+ 1 , c+ Cap∗ 2 ; a0, . . . , aM ) = 2 ∈ A+ and M+2(c+ 1 , e+ c+ 1 e1,1c+ 2 − c+ 2 c+ c+ 1 , . . . , e+ 2 e1,1c+ 2 c+ 1 ]e1,1 = [d+ M , c+ 1 − e1,1c+ 1 c+ 1 − c+ 2 , d+ 1 ]e1,1. 1 e1,1 + e1,1c+ [c+ 2 , c+ 2 + 1 c+ 2 e1,1 = M+1[Y ; X] ⊆ Id∗(A) we have [d+ 1 , d+ 2 ] = 0. Thus d+ 1 d+ 2 = d+ 2 d+ 1 , for all d+ 1 , d+ 2 ∈ Since Cap∗ N +. Now, let e− 1 , . . . , e− L be an ordered basis of Mk(F )− consisting of skew matrices e− h ∈ 1 = e1,2 − e2,1. We consider b0, . . . , bL ∈ Mk(F ) {ei,j − ej,i i < j, i, j = 1, . . . k} such that e− such that b0 = e1,1, bL = ek,k, 2 = (e1,2 + e2,1)d− 2 . Since N 1 , d− 2 ]. 2 d− 2 = d− 1 , for all d− 1 , d− 2 ∈ and b0e− 1 b1 · · · bL−1e− L bL = e1,k b0e− τ (1)b1 · · · bL−1e− τ (L)bL = 0 for all τ ∈ SL, τ 6= id. 1 , d− Let d− commutes with Mk(F ) then c− Cap∗ 2 ∈ N − and put c− 1 , c− L+2(c− 1 and c− 1 = (e1,2 + e2,1)d− 2 ∈ A−. As above we compute 2 ; b0, . . . , bL) = 1 , d− 2 ] = e1,k[d− 1 d− L , c− 2 ]e1,k = (e1,1 + e2,2)e1,k[d− 2 ] = 0, then d− 1 , . . . , e− 1 , d− 1 , e− [c− 1 , c− Since Cap∗ N −. L+1[Z; X] ⊆ Id∗(A) we get that [d− Next we show that N + commutes with N −. Take e+ 1 , . . . , e+ 1 = e1,1 and let a1, . . . , aM ∈ Mk(F ) be such that such that e+ M an ordered basis of Mk(F )+ and a1e+ 1 a2 · · · aM e+ M = e1,k a1e+ ρ(1)a2 · · · aM e+ ρ(M) = 0 for any ρ ∈ SM , ρ 6= id. Notice that a1 = a2 = e1,1. Let d+ 1 = (e1,2 + e2,1)d+ c+ Id∗(A), we obtained 1 and ¯a2 = e1,1d− 2 . Notice that c+ 1 ∈ N + and d− 1 ∈ A+. Then, since Cap∗ 2 ∈ N −. We set M+1[Y ; X] ⊆ Thus d+ 1 d− 2 = d− 2 d+ 0 = Cap∗ M+1(c+ 1 , for all d+ 1 , e+ 1 , . . . , e+ 1 ∈ N +, d− M ; a1, ¯a2, a3, . . . , aM ) = [d+ 2 ∈ N − and we are done. 1 , d− 2 ]e2,k. Now we are able to prove the main result about Id∗((Mk(F ), t)) and the T-∗-ideal gen- erated by the ∗-Capelli polynomials Cap+ , Cap− k(k+1) 2 +1 . k(k−1) 2 +1 First we prove the following Lemma 8. Let M = k(k + 1)/2 and L = k(k − 1)/2 with k ∈ N, k > 0. Then exp∗(U ∗ M+1,L+1) = M + L = k2 = exp∗((Mk(F ), t)). ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 11 M+1,L+1 is equal to the exponent of some minimal variety lying in M+1,L+1 ( for the definition of minimal variety see [16]). By [10, Theorem 1.2] and Lemma Proof. The exponent of U ∗ U ∗ 3 we have that exp∗(U ∗ M+1,L+1) = max{exp∗(U T ∗(A1, . . . , An)) U T ∗(A1, . . . , An) satisfies Cap+ M+1 and Cap− L+1}. i=1 dimF A− i , then Let d+ := Pn i=1 dimF A+ i and d− := Pn exp∗(U ∗ M+1,L+1) = max{exp∗(U T ∗(A1, . . . , An)) d+ + n ≤ M + 1 and d− + n ≤ L + 1} ≥ Since by Lemma 2, exp∗(U ∗ exp∗(U T ∗(Mk(F ))) = k2 = M + L. M+1,L+1) ≤ M + L then the proof is completed. Theorem 4. Let M = k(k + 1)/2 and L = k(k − 1)/2 with k ∈ N, k > 0. Then U ∗ M+1,L+1 = var∗(Γ∗ M+1,L+1) = var∗(Mk(F ) ⊕ D′), where D′ is a finite dimensional ∗-algebra such that exp∗(D′) < M + L. In particular c∗ n(Γ∗ M+1,L+1) ≃ c∗ n(Mk(F )). Proof. By Lemma 8 we have that exp∗(U ∗ M+1,L+1) = M + L. Let A = A+ ⊕ A− be a generating ∗-algebra of U ∗ M+1,L+1. As remarked before we can assume that A is finite dimensional. Thus, by Corollary 2, there exists a finite number of reduced ∗-algebras B1, . . . , Bs and a finite dimensional ∗-algebra D′ such that U ∗ M+1,L+1 = var∗(A) = var∗(B1 ⊕ · · · ⊕ Bs ⊕ D′). (1) exp∗(B1) = · · · = exp∗(Bs) = exp∗(U ∗ M+1,L+1) = M + L Moreover and Next, we analyze the structure of a finite dimensional reduced ∗-algebra R such that exp∗(D′) < exp∗(U ∗ M+1,L+1) = M + L. exp∗(R) = M + L = exp∗(U ∗ M+1,L+1) and Γ∗ M+1,L+1 ⊆ Id∗(R). We can write R = R1 ⊕ · · · ⊕ Rq + J, where Ri are simple ∗-subalgebras of R, J = J(R) is the Jacobson radical of R and R1J · · · JRq 6= 0. Recall that every ∗-algebra Ri is isomorphic to one of the following algebras :(Mki (F ), t) or (M2mi(F ), s) or (Mhi(F ) ⊕ Mhi(F )op, exc). Let t1 be the number of ∗-algebras Ri of the first type, t2 the number of ∗-algebras Ri of the second type and t3 the number of Ri of the third type, with t1 + t2 + t3 = q. By [7, Theorem 4.5] and [7, Proposition 4.7] there exists a ∗-algebra R isomorphic to the ∗-algebra U T ∗(R1, . . . , Rq) such that Let observe that exp∗(R) = exp∗(R) = exp∗(U T ∗(R1, . . . , Rq)). k2 = M + L = exp∗(R) = exp∗(R) = exp∗(U T ∗(R1, . . . , Rq)) = dimF R1 + · · · + dimF Rq = k2 1 + · · · + k2 t1 + (2m1)2 + · · · + (2mt2)2 + 2h2 1 + · · · + 2h2 t3. 12 BENANTI AND VALENTI Let d± = dimF (R1 ⊕ · · · ⊕ Rq)± then d+ + d− = d = dimF (R1 ⊕ · · · ⊕ Rq) = exp∗(R) = M + L. d−+q−1[Z; X], but R satisfies Cap∗ By [10, Lemma 3.2] R does not satisfy the ∗-Capelli polynomials Cap∗ d++q−1[Y ; X] and L+1[Z; X]. Thus d+ + q − 1 ≤ M Cap∗ and d− + q − 1 ≤ L. Hence d+ + d− + 2q − 2 ≤ M + L. Since d+ + d− = M + L we obtain that 2q − 2 = 0 and so 1 = q = t1 + t2 + t3. Since t1, t2 and t3 are nonnegative integers, we have the following three possibilities M+1[Y ; X] and Cap∗ (1) t1 = 1 and t2 = t3 = 0; (2) t2 = 1 and t1 = t3 = 0 (3) t3 = 1 and t1 = t2 = 0. If t2 = 1, then R = (M2m(F ), s) + J and exp∗(R) = 4m2. Thus k2 = M + L = exp∗(R) = 4m2 and so k = 2m. By hypothesis R satisfies Cap∗ does not satisfy Cap∗ d−[Z; X], where d− = m(2m + 1). It follows that L+1[Z; X] but, since Id∗(R) ⊆ Id∗(R), R L + 1 = k(k − 1)/2 + 1 = m(2m − 1) + 1 < m(2m + 1) = d−, for m ≥ 1, and this is a contradiction. Let assume t3 = 1. Then R = (Mh(F ) ⊕ Mh(F )op) + J and exp∗(R) = 2h2. Thus k2 = M + L = exp∗(R) = 2h2 and this is impossible. Then t1 = 1 and in this case From Lemmas 4, 6, 7 we obtain R ∼= Mk(F ) + J. R ∼= (Mk(F ) + J11) ⊕ J00 ∼= (Mk(F ) ⊗ N ♯) ⊕ J00 where N ♯ is the algebra obtained from N by adjoining a unit element. Since N ♯ is com- mutative, it follows that Mk(F ) + J11 and Mk(F ) satisfy the same ∗-identities. Thus var∗(R)=var∗(Mk(F ) ⊕ J00) with J00 a finite dimensional nilpotent ∗-algebra. Hence, re- calling the decomposition given in (1), we get U ∗ M+1,L+1 = var∗(ΓM+1,L+1) = var∗(Mk(F ) ⊕ D′), where D′ is a finite dimensional ∗-algebra with exp∗(D′) < M + L. Then, from Corollary 1 we have and the theorem is proved. n(ΓM+1,L+1) ≃ c∗ c∗ n(Mk(F )) 5. Asymptotics for U ∗ m(2m−1)+1,m(2m+1)+1 and (M2m(F ), s) Throughout this section we assume that A = M2m(F ) + J, where J = J(A) is the Jacobson radical of the finite dimensional ∗-algebra A and (M2m(F ), s) is the algebra of matrices with symplectic involution. Lemma 9. Let M = m(2m − 1) and L = m(2m + 1) with m ∈ N, m > 0. If Γ∗ Id∗(A), then J10 = J01 = (0). M+1,L+1 ⊆ ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 13 Proof. By Lemma 5, (M2m(F ), s) does not satisfy the ∗-Capelli polynomial Cap∗ Also there exist elements a+ M ∈ M2m(F )+ and b1, . . . , bM−1 ∈ M2m(F ) such that 1 , . . . , a+ M [Y, X]. Cap∗ M (a+ 1 , . . . , a+ M ; b1, . . . , bM−1) = e1,2m, where the ei,j's are the usual matrix units. Let d ∈ J01, then d + d∗ ∈ (J01 ⊕ J10)+. Since Γ∗ M+1,L+1 ⊆ Id∗(A) we have 0 = Cap∗ M+1(a+ 1 , . . . , a+ M , d + d∗; b1, . . . , bM−1, e2m,2m) = de1,2m ± e1,2md∗. Hence de1,2m = ∓e1,2md∗ ∈ J01 ∩ J10 = (0) and so d = 0, for all d ∈ J01. Thus J01 = (0) = J10 and we are done. Lemma 10. Let M = m(2m − 1) and L = m(2m + 1) with m ∈ N, m > 0. Let J11 ∼= M2m(F ) ⊗F N , as in Lemma 4. If ΓM+1,L+1 ⊆ Id∗(A), then N is commutative. Proof. Let N be the finite dimensional nilpotent ∗-algebra of the Lemma 4. As in Lemma 7 we can consider an ordered basis of M2m(F )+ consisting of symmetric matrices e+ h ∈ {ei,j + em+j,m+i i, j = 1, . . . m} ∪ {ei,m+j − ej,m+i 1 ≤ i < j ≤ m} ∪ {em+i,j − em+j,i 1 ≤ i < j ≤ m} with e+ 1 = e1,1 + em+1,m+1 and a0, a1, . . . , aM ∈ M2m(F ) such that a0e+ 1 a1 · · · aM−1e+ M aM = e1,1 a0e+ σ(1)a1 · · · aM−1e+ σ(M)aM = 0 2 ∈ N + and set c+ for any σ ∈ SM , σ 6= id. Consider d+ c+ 2 = (e1,1 + em+1,m+1)d+ 1 , e+ 2 . Since N commutes with M2m(F ), c+ 1 , . . . , e+ 2 ; a0, . . . , aM ) = [c+ M+2(c+ M , c+ 1 , d+ 2 , c+ Cap∗ M+1[Y ; X] ⊆ Id∗(A) we have d+ 1 d+ 2 = d+ 2 d+ 1 , for all d+ 1 , d+ Since Cap∗ 1 ]e1,1 = [d+ 1 ]e1,1. 2 , d+ 2 ∈ N +. 1 = (e1,1 + em+1,m+1)d+ 1 and 1 , c+ 2 ∈ A+ and we obtain 1 , . . . , e− Now, let e− L be an ordered basis of M2m(F )− consisting of skew matrices, e− h ∈ {ei,j −em+j,m+i i, j = 1, . . . m}∪{ei,m+j +ej,m+i 1 ≤ i < j ≤ m}∪{ei,m+i i = 1, . . . , m}∪ {em+i,j + em+j,i 1 ≤ i < j ≤ m} ∪ {em+i,i i = 1, . . . , m} such that e− 1 = e1,1 − em+1,m+1 . We consider b0, . . . , bL ∈ M2m(F ) such that b0 = e1,1, bL = e2m,2m and and b0e− 1 b1 · · · bL−1e− L bL = e1,2m b0e− 1 , d− τ (1)b1 · · · bL−1e− 2 ∈ N −. Let c− τ (L)bL = 0 for all τ ∈ SL, τ 6= id. Let d− (e1,1 + em+1,m+1)d− obtain 2 . Since N commutes with M2m(F ) then c− 1 , c− 1 = (e1,1 + em+1,m+1)d− 2 = 2 ∈ A−. As above we 1 and c− Cap∗ L+2(c− 1 , e− 1 , . . . , e− L , c− 2 ; b0, . . . , bL) = [c− 1 , c− 2 ]e1,2m = (e1,1 + em+1,m+1)e1,2m[d− 1 , d− 2 ] = e1,2m[d− 1 , d− 2 ]. 2 ] = 0 and so d− 1 d− 2 = d− 1 , d− 2 d− 1 , for all 2 ∈ N −. L+1[Z; X] ⊆ Id∗(A) we get that [d− Since Cap∗ d− 1 , d− Next we show that N + commutes with N −. Take e+ 1 , . . . , e+ a1, . . . , aM ∈ M2m(F ) be such that a1e+ 1 a2 · · · aM e+ M = e1,2m and a1e+ ρ(1)a2 · · · aM e+ ρ(M) = 0 M an ordered basis of M2m(F )+ such that e+ 1 = e1,1 + em+1,m+1. Let 14 BENANTI AND VALENTI for any ρ ∈ SM , ρ 6= id. Notice that a1 = a2 = e1,1. Let d+ 1 = (e1,2 + em+2,m+1)d+ c+ we obtain 1 and ¯a2 = e1,1d− 2 , then c+ 1 ∈ A+. Since Cap∗ 1 ∈ N + and d− 2 ∈ N −. We set M+1[Y ; X] ⊆ Id∗(A), Thus d+ 1 d− 2 = d− 2 d+ 0 = Cap∗ M+1(c+ 1 , for all d+ 1 , e+ 1 , . . . , e+ 1 ∈ N +, d− M ; a1, ¯a2, a3, . . . , aM ) = [d+ 2 ∈ N − and we are done. 1 , d− 2 ]e2,2m. Lemma 11. Let M = m(2m − 1) and L = m(2m + 1) with m ∈ N, m > 0. Then exp∗(U ∗ M+1,L+1) = M + L = 4m2 = exp∗((M2m(F ), s)). Proof. The proof is the same of that of Lemma 8. Now we are able to prove the following Theorem 5. Let M = m(2m − 1) and L = m(2m + 1) with m ∈ N, m > 0. Then U ∗ M+1,L+1 = var∗(ΓM+1,L+1) = var∗(M2m(F ) ⊕ D′′), where D′′ is a finite dimensional ∗-algebra such that exp∗(D′′) < M + L. In particular c∗ n(ΓM+1,L+1) ≃ c∗ n(M2m(F )). Proof. The first part of the proof follows step by step that of Theorem 4 and we obtain U ∗ M+1,L+1 = var∗(B1 ⊕ · · · ⊕ Bs ⊕ D′′), where B1, . . . , Bs are reduced ∗-algebras and D′′ is a finite dimensional ∗-algebra such that exp∗(D′′) < exp∗(Bi) = M + L, for all i = 1, . . . s. Let R be a finite dimensional reduced ∗-algebra with exp∗(R) = M +L = exp∗(UM+1,L+1) and Γ∗ M+1,L+1 ⊆ Id∗(R). We can write R = R1 ⊕ · · · ⊕ Rq + J, where Ri are simple ∗- subalgebras of R, J = J(R) is the Jacobson radical of R. Let t1 be the number of ∗-algebras Ri isomorphic to (Mki (F ), t), t2 the number of ∗-algebras Ri isomorphic to (M2mi (F ), s) and let t3 be the number of Ri isomorphic to (Mhi(F ) ⊕ Mhi(F )op, exc), where t1 + t2 + t3 = q. Hence, as in Theorem 4, there exists a ∗-algebra R isomorphic to the ∗-algebra U T ∗(R1, . . . , Rq) such that exp∗(R) = exp∗(R) = exp∗(U T ∗(R1, . . . , Rq)) and 4m2 = M + L = exp∗(R) = exp∗(R) = exp∗(U T ∗(R1, . . . , Rq)) = k2 1 + · · · + k2 t1 + (2m1)2 + · · · + (2mt2)2 + 2h2 1 + · · · + 2h2 t3. As in the proof of the Theorem 4 we have q = 1 and so we obtain only three possibilities: t1 = 1 and t2 = t3 = 0 or t2 = 1 and t1 = t3 = 0 or t3 = 1 and t1 = t2 = 0. If t1 = 1, then R = (Mk(F ), t)+J and exp∗(R) = k2. Thus 4m2 = M +L = exp∗(R) = k2 M+1[Y ; X] and, since Id∗(R) ⊆ Id∗(R), R does not d+ [Y ; X], where d+ = k(k + 1)/2. It follows that M + 1 = m(2m − 1) + 1 < and so k = 2m. Hence R satisfies Cap∗ satisfy Cap∗ m(2m + 1) = k(k + 1)/2 = d+ for m ≥ 1, and this is a contradiction. Let assume t3 = 1. Then R = (Mh(F ) ⊕ Mh(F )op) + J and exp∗(R) = 2h2. Thus 4m2 = M + L = exp∗(R) = 2h2 and this is impossible. Finally, let t2 = 1 and t1 = t3 = 0. Then R ∼= M2m(F ) + J. By Lemmas 4, 9, 10 we obtain R ∼= (M2m(F ) + J11) ⊕ J00 ∼= (M2m(F ) ⊗ N ♯) ⊕ J00 ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 15 where N ♯ is the algebra obtained from N by adjoining a unit element. Since N ♯ is com- mutative, it follows that M2m(F ) + J11 and M2m(F ) satisfy the same ∗-identities. Thus var∗(R)=var∗(M2m(F ) ⊕ J00) with J00 a finite dimensional nilpotent ∗-algebra. Hence by (1) we get U ∗ M+1,L+1 = var∗(ΓM+1,L+1) = var∗(M2m(F ) ⊕ D′′), where D′′ is a finite dimensional ∗-algebra with exp∗(D′′) < M + L. Then, from Corollary 1, we have c∗ n(ΓM+1,L+1) ≃ c∗ n(M2m(F )) and the theorem is proved. 6. Asymptotics for U ∗ h2+1,h2+1 and (Mh(F ) ⊕ Mh(F )op, exc) Throughout this section we assume that A = (Mh(F ) ⊕ Mh(F )op) + J, where J = J(A) is the Jacobson radical of the finite dimensional ∗-algebra A and (Mh(F ) ⊕ Mh(F )op, exc) is the direct sum of the algebra of h × h matrices and the opposite algebra with the exchange involution. We start with the following lemmas Lemma 12. Let M = L = h2 with h ∈ N, h > 0. If Γ∗ (0). M+1,L+1 ⊆ Id∗(A), then J10 = J01 = M [Y, X]. Also, there exist elements a+ Proof. By Lemma 5, (Mh(F ) ⊕ Mh(F )op, exc) does not satisfy the ∗-Capelli polyno- M ∈ (Mh(F ) ⊕ Mh(F )op)+ and mial Cap∗ b1, . . . , bM−1 ∈ Mh(F ) ⊕ Mh(F )op such that 1 , . . . , a+ M ; b1, . . . , bM−1) = e1,h, 1 , . . . , a+ M (a+ Cap∗ where the ei,j = (ei,j , ej,i) and ei,j's are the usual matrix units. Let d ∈ J01, then d + d∗ ∈ (J01 ⊕ J10)+. Since Γ∗ 0 = Cap∗ M+1,L+1 ⊆ Id∗(A) it follows M+1(a+ M , d + d∗; b1, . . . , bM−1, eh,h)eh,h = de1,h. 1 , . . . , a+ Hence d = 0, for all d ∈ J01. Thus J01 = (0) and J10 = (0). Lemma 13. Let M = L = h2 with h ∈ N, h > 0. Let J11 ∼= (Mh(F ) ⊕ Mh(F )op) ⊗F N , as in Lemma 4. If ΓM+1,L+1 ⊆ Id∗(A), then N is commutative. Proof. Let N be the finite dimensional nilpotent ∗-algebra of the Lemma 4. Let now v+ 1 , . . . , v+ i,j = (ei,j, ei,j) such that v+ M an ordered basis of (Mh(F ) ⊕ Mh(F )op)+ consisting of all e+ 1,1 and let a0, . . . , aM ∈ Mh(F ) ⊕ Mh(F )op be such that 1 = e+ a0v+ 1 a1 · · · aM−1v+ M aM = e+ 1,1 and a0v+ σ(1)a1 · · · aM−1v+ σ(M)aM = (0, 0) for any σ ∈ SM , σ 6= id. Now let d+ 1 = e+ with Mh(F ) ⊕ Mh(F )op we have that c+ 2 ∈ N + and set c+ 1 , d+ 1,1d+ 1 , c+ 1 , v+ M+2(c+ Cap∗ M+1[Y ; X] ⊆ Id∗(A) it follows d+ 1 , . . . , v+ M , c+ Since Cap∗ 2 . Recalling that N commutes 1,1d+ 2 = e+ 1 and c+ 2 ∈ A+ and, as in Lemma 7, 1,1 = [d+ 1 , d+ 1 , for all d+ 2 = d+ 2 , c+ 1 ]e+ 2 d+ 1 d+ 2 ; a0, . . . , aM ) = [c+ 1,1. 2 , d+ 1 ]e+ 2 ∈ N +. 16 BENANTI AND VALENTI Now, let v− 1 , . . . , v− L be an ordered basis of (Mh(F ) ⊕ Mh(F )op)− consisting of all e− i,j = (ei,j, −ei,j) such that v− 1 = e− 1,1 and let b0, . . . , bL ∈ Mh(F ) ⊕ Mh(F )op be such that b0v+ 1 b1 · · · bL−1v+ L bL = (cid:26) e+ 1,1, e− 1,1, if L is even; if L is odd. and b0v+ σ(L)bL = (0, 0) 2 ∈ N − and set c− 2 . Since N commutes with Mh(F ) ⊕ Mh(F )op, we have that c− L+1[Z, X] ⊆ Id∗(A), we obtain for any σ ∈ SM , σ 6= id. Now, we take d− e+ 1,1d− Cap∗ σ(1)b1 · · · bL−1v+ 1 , d− 1,1d− 1 = e+ 1 , c− 2 = 2 ∈ A−. Thus, since 1 and c− 0 = Cap∗ L+2(c− 1 , v− 1 , . . . , v− L , c− 2 ; b0, . . . , bL) =(cid:26) [c− [c− 2 , c− 2 , c− 1 ]e+ 1 ]e− 1,1 = [d− 1,1 = [d− 2 , d− 2 , d− 1 ]e+ 1 ]e− 1,1, 1,1, if L is even; if L is odd. In conclusion [d− 2 , d− Finally, we consider d+ 1 ] = 0, for all d− 1 ∈ N +, d− 2 ∈ N −. 1 , d− 2 ∈ N − and set c+ 1 = e+ 1,1d+ 1 and c+ 2 = e− 1,1d− 2 . As above, we have that c+ 1 , c+ 2 ∈ A+. Then M+2(c+ 0 = Cap∗ 1 ] = 0 for all d+ 1 , v+ 1 , . . . , v+ 1 ∈ N +, d− and [d− 2 , d+ 2 ; a0, . . . , aM ) = [c+ M , c+ 2 ∈ N − and the lemma is proved. 2 , c+ 1 ]e+ 1,1 = [d− 2 , d+ 1 ]e− 1,1 Lemma 14. Let M = L = h2 with h ∈ N, h > 0. Then exp∗(U ∗ M+1,L+1) = M + L = 2h2 = exp∗((Mh(F ) ⊕ Mh(F )op, exc)). Proof. The proof is the same of that of Lemma 8. Theorem 6. Let M = L = h2 with h ∈ N, h > 0. Then U ∗ M+1,L+1 = var∗(ΓM+1,L+1) = var∗((Mh(F ) ⊕ Mh(F )op) ⊕ D′′′), where D′′′ is a finite dimensional ∗-algebra such that exp∗(D′′′) < M + L. In particular c∗ n(ΓM+1,L+1) ≃ c∗ n(Mh(F ) ⊕ Mh(F )op). Proof. The proof proceeds as in that of Theorem 4. Let R be a finite dimensional reduced ∗-algebra such that exp∗(R) = M + L = exp∗(U ∗ M+1,L+1 ⊆ Id∗(R). We can write R = R1 ⊕ · · · ⊕ Rq + J, where Ri are simple ∗-subalgebras of R, J = J(R) is the Jacobson radical of R. If t1 denotes the number of algebras Ri isomorphic to (Mki(F ), t), t2 the number of algebras Ri isomorphic to (M2mi(F ), s) and t3 the number of Ri isomorphic to (Mhi(F ) ⊕ Mhi(F )op, exc), then we have t1 + t2 + t3 = q. As in Theorem 4, there exists a ∗-algebra R isomorphic to the ∗-algebra U T ∗(R1, . . . , Rq) such that exp∗(R) = exp∗(R) = exp∗(U T ∗(R1, . . . , Rq)). Then M+1,L+1) and Γ∗ 2h2 = M + L = exp∗(R) = exp∗(R) = exp∗(U T ∗(R1, . . . , Rq)) = k2 1 + · · · + k2 t1 + (2m1)2 + · · · + (2mt2)2 + 2h2 1 + · · · + 2h2 t3 and we have only three possibilities: t1 = 1 and t2 = t3 = 0 or t2 = 1 and t1 = t3 = 0 or t3 = 1 and t1 = t2 = 0. If t1 = 1, then R = (Mk(F ), t)+ J and exp∗(R) = k2. Thus 2h2 = M + L = exp∗(R) = k2 and this is a contradiction. If t2 = 1. Then R = (M2m(F ), s) + J and exp∗(R) = 4m2. Thus 2h2 = M + L = exp∗(R) = 4m2 and so h2 = 2m2, contradiction. ASYMPTOTICS FOR CAPELLI POLYNOMIALS WITH INVOLUTION 17 Finally, let t3 = 1 and t1 = t2 = 0. Then R ∼= (Mh(F ) ⊕ Mh(F )op) + J. As in Theorem 4, by Lemmas 4, 12, 13 we obtain R ∼= ((Mh(F ) ⊕ Mh(F )op) + J11) ⊕ J00 ∼= ((Mh(F ) ⊕ Mh(F )op) ⊗ N ♯) ⊕ J00 where N ♯ is the algebra obtained from N by adjoining a unit element. Since N ♯ is com- mutative, we have that (Mh(F ) ⊕ Mh(F )op) + J11 and Mh(F ) ⊕ Mh(F )op satisfy the same ∗-identities. Thus var∗(R)=var∗((Mh(F ) ⊕ Mh(F )op) ⊕ J00) with J00 a finite dimensional nilpotent ∗-algebra. We get U ∗ M+1,L+1 = var∗(ΓM+1,L+1) = var∗((Mh(F ) ⊕ Mh(F )op) ⊕ D′′′), where D′′′ is a finite dimensional ∗-algebra with exp∗(D′′′) < M + L. Then, from Corollary 1, we have c∗ n(ΓM+1,L+1) ≃ c∗ and the proof is completed. n((Mh(F ) ⊕ Mh(F )op)) References 1. E. Aljadeff, A. Giambruno and Y. Karasik, Polynomial identities with involution, superinvolutions and the Grassmann envelope, Proc. Amer. Math. Soc. 145 (2017), 1843-1857. 2. S. A. Amitsur, Identities in rings with involution, Israel J. Algebra 7 (1969), 63 -- 68. 3. Y. Bahturin, A. Giambruno and M. Zaicev, G-identities on associative algebras, Proc. Amer. Math. Soc. 127 (1999), 63 -- 69. 4. F. Benanti, Asymptotics for Graded Capelli Polynomials, Algebra Repres. Theory 18 (2015), 221 -- 233. 5. F. Benanti and I. Sviridova Asymptotics for Amitsur's Capelli-type polynomials and verbally prime PI-algebras, Israel J. Math. 156 (2006), 73 -- 91. 6. A. Berele, A. Giambruno and A. Regev, Involution codimensions and trace codimensions of matrices are asymptotically equal, Israel J. Math. 96 (1996), 49 -- 62. 7. O. M. Di Vincenzo and R. La Scala Minimal algebras with respect to their ∗-exponent, J.Algebra 317 (2007), 642-657. 8. O. M. Di Vincenzo and E. Spinelli, Some results of ∗-minimal algebras with involution, in Group, Rings and Group Rings 2008 (A. Giambruno et al.eds.), International Conference, Ubatuba, Brazil, July 27 August 2, 2008, Contemporary Mathematics 499, American Mathematical Society, Providence, RI, (2009), 75 -- 88. 9. O. M. Di Vincenzo and E. Spinelli, On ∗-minimality of algebras with involution, J. Algebra 323 (2010), 121 -- 131. 10. O. M. Di Vincenzo and E. Spinelli, A characterization of ∗-minimal algebras with involution, Israel J. Math. 186 (2011), 381 -- 400. 11. A. Giambruno and S. Mishchenko, On star-varieties with almost polynomial growth, Algebra Colloq. 8 (2001), 33 -- 42. 12. A. Giambruno and S. Mishchenko, Polynomial growth of the ∗-codimensions and Young diagrams, Comm. Algebra 29 (2001), 277 -- 284. 13. A. Giambruno, A. Ioppolo and D. La Mattina, Varieties of Algebras with Superinvolution of Almost Polynomial Growth, Algebr. Represent. Theory, 19 (2016), 599 -- 611. 14. A. Giambruno, C. Polcino Milies and A. Valenti, Star-polynomial identities: Computing the exponential growth of the codimensions, J. Algebra 469 (2017), 302 -- 322. 15. A. Giambruno and A. Regev, Wreath products and P.I.algebras, J. Pure Appl. Algebra 35 (1985), 133 -- 149. 16. A. Giambruno and M. Zaicev, Polynomial Identities and Asymptotics Methods, Surveys, vol. 122, Amer- ican Mathematical Society, Providence, RI, 2005. 17. A. Giambruno and M. Zaicev, Involution codimensions of finite dimensional algebras and exponential growth, J. Algebra 222 (1999), 471 -- 484. 18. A. Giambruno and M. Zaicev, A characterization of algebras with polynomial growth of the codimensions, Proc. Amer. Math. Soc. 129 (2001), 59 -- 67. 19. A. Giambruno and M. Zaicev, Asymptotics for the Standard and the Capelli Identities, Israel J. Math. 135 (2003), 125 -- 145. 18 BENANTI AND VALENTI 20. A. R. Kemer, Ideals of Identities of Associative Algebra, Amer. Math. Soc. Translations of Math. Mono- graphs 87, Providence, RI, 1991. 21. S. Mishchenko and A. Valenti A star-variety with almost polynomial growth, J. Algebra 223 (2000), 66 -- 84. 22. L. H. Rowen, Polynomial Identities in Ring Theory, Academic Press, New York, 1980. 23. I. Sviridova, Finitely generated algebras with involution and their identities, J. Algebra 383 (2013), 144 -- 167. Dipartimento di Matematica e Informatica, Universit´a di Palermo, via Archirafi, 34, 90123 Palermo,Italy E-mail address: [email protected] Dipartimento di Ingegneria, Universit`a di Palermo, Viale delle Scienze, 90128 Palermo, Italy E-mail address: [email protected]
1509.02682	1	1509	2015-09-09T09:04:15	Non-Noetherian generalized Heisenberg algebras	[ "math.RA" ]	In this note we classify the non-Noetherian generalized Heisenberg algebras H(f) introduced by Rencai L\"u and Kaiming Zhao [Linear Algebra Appl., 2015]. In case the polynomial f has degree greater than 1, we determine all locally finite and also all locally nilpotent derivations of H(f) and describe the automorphism group of these algebras.	math.RA	math	Non-Noetherian generalized Heisenberg algebras Samuel A. Lopes∗ Abstract In this note we classify the non-Noetherian generalized Heisenberg algebras H(f ) introduced in [8]. In case deg f > 1, we determine all locally finite and also all locally nilpotent derivations of H(f ) and describe the automorphism group of these algebras. MSC Numbers (2010): Primary 16W25, 16W20; Secondary 16P40, 16S36. Keywords: generalized Heisenberg algebra; ambiskew polynomial ring; weak general- ized Weyl algebra; automorphism group; derivation; Noetherian. 1 Introduction Fix a polynomial f ∈ C[h]. The generalized Heisenberg algebra H(f ) is the unital associative C-algebra with generators x, y, h satisfying the relations: hx = xf (h), yh = f (h)y, yx − xy = f (h) − h. (1.1) See [8] and the references therein for information on how these algebras first appeared and on their applications to theoretical physics. Ambiskew polynomial rings were introduced by Jordan over a series of papers (see the references in [5]), but for our purposes the best suited definition is the one found in [5], which we briefly recall. Let σ be an endomorphism of a commutative C-algebra B, c ∈ B and p ∈ C. The ambiskew polynomial ring R(B, σ, c, p) is the C-algebra generated by B and two indeterminates, x and y, subject to the relations bx = xσ(b), yb = σ(b)y, yx − pxy = c, for all b ∈ B. On comparing these relations with those in (1.1), one immediately sees that H(f ) ∼= R(C[h], σ, f (h) − h, 1), (1.2) where σ : C[h] → C[h] is the algebra endomorphism given by σ(h) = f (h). In particular, one can see that there is an overlap between the generalized Heisenberg algebras defined above and (generalized) down-up algebras (see Corollary 2.7 below). ∗The author was partially supported by CMUP (UID/MAT/00144/2013), which is funded by FCT (Portugal) with national (MEC) and European structural funds through the programs FEDER, under the partnership agreement PT2020. 1 The algebras H(f ) can also be seen as weak generalized Weyl algebras over a poly- nomial algebra in two variables, in the sense of [7], a construction which includes the generalized Weyl algebras introduced by V.V. Bavula in [1]. In [8] the authors determine a basis for H(f ) over C, compute the center of H(f ), solve the isomorphism problem for this family of algebras and classify all the finite-dimensional irreducible representations of H(f ). In this note we show that H(f ) is (right or left) Noetherian if and only if deg f = 1 and that H(f ) is isomorphic to a generalized down-up algebra if and only if deg f ≤ 1. For this reason, we then concentrate on the case where deg f > 1 and determine the locally nilpotent and the locally finite derivations of H(f ), all Z-gradings of H(f ) and describe the automorphism group of H(f ). In particular, we obtain the following results in case deg f > 1: (i) H(f ) in neither right nor left Noetherian (Proposition 2.4); (ii) H(f ) admits a unique (up to an integer multiple) nontrivial Z-grading, in which x has degree 1, y has degree −1 and h has degree 0 (Corollary 4.10); (iii) the automorphism group of H(f ) is abelian: it is isomorphic to C∗ × C, where C is a finite cyclic group whose order divides ( deg f ) − 1 (Theorem 5.5). In Section 2 of the paper we review some properties of H(f ) which have been established in [8], determine when H(f ) is Noetherian and when it is isomorphic to a generalized down-up algebra, while in Section 3 we introduce a useful commutative subalgebra of H(f ), which is a maximal commutative subalgebra if deg f > 1. Assuming that deg f > 1, we then investigate the locally finite and the locally nilpotent derivations of H(f ) and also its Z-gradings in Section 4, and in the final section, Section 5, we describe the automorphism group of H(f ) and show that it is always an abelian group generated by the automorphisms which fix h and the automorphisms which fix x. We make use of the commutator notation [a, b] = ab − ba. The sets of integers, nonnegative integers and positive integers are denoted by Z, Z≥0 and Z>0, respectively. The field of complex numbers is denoted by C, and the multiplicative group of nonzero complex numbers is denoted by C∗. For a polynomial g ∈ C[h], deg g will always denote the degree of g as a polynomial in h. Throughout the paper, σ : C[h] → C[h] is the algebra endomorphism given by σ(h) = f (h). For any function φ : X → X, we will use the notation φk to mean the k-th power of φ with respect to composition. In particular, φ0 denotes the identity on the set X. 2 The Noetherian property Below we record a few results from [8] which will be useful in the course of this paper. As usual, Z(H(f )) denotes the center of H(f ). Lemma 2.1 ([8, Lemma 1, Lemma 2, Theorem 4]). Let f ∈ C[h]. Then: 2 (a) The set {xihj yk i, j, k ∈ Z≥0} is a basis of H(f ). (b) The algebra H(f ) is a domain if and only if deg f ≥ 1. (c) The center of H(f ) contains the polynomial algebra C[z], where z = xy − h = yx − f (h). If deg f 6= 1, then Z(H(f )) = C[z]. Remarks 2.2. 1. Identifying H(f ) with the ambiskew polynomial ring R(C[h], σ, f (h) − h, 1) as in (1.2), it follows that H(f ) is conformal, as defined in [5, Section 2.3], and the corresponding Casimir element is precisely the central element z = xy − h defined above. 2. Suppose f ∈ C. Then by considering the generators −x, y and h − f , we see that H(f ) ∼= R(C[h], σ, h, 1), with σ = 0, and from [5, Theorem 7.10] we conclude that H(f ) is a prime ring. Thus by Lemma 2.1(b), H(f ) is a prime ring for any f ∈ C[h]. 3. Since the center of H(f ) contains the polynomial algebra C[z] and H(f ) has countable dimension over C, it follows from Dixmier's version of Schur's Lemma that H(f ) is never primitive. There is an order two anti-automorphism of H(f ), denoted by ι, that fixes h and interchanges x and y: ι : H(f ) → H(f ), x 7→ y, y 7→ x, h 7→ h. (2.3) Hence H(f ) is isomorphic to its opposite algebra H(f )op. Proposition 2.4. The algebra H(f ) is right (respectively, left) Noetherian if and only if deg f = 1. Proof. If deg f = 1 then H(f ) is a generalized Weyl algebra over a polynomial ring in two variables, and thus it is right and left Noetherian. So assume that deg f 6= 1. In particular, f (h) − h has some root α ∈ C. Let F (h) = f (h + α) − α. Then deg F = deg f (here we assume the zero polynomial has degree 0) and F (h) ∈ hC[h]. Moreover, F (h − α) = f (h) − α and then H(f ) ∼= H(F ) by [8, Lemma 3]. So there is no loss in assuming that f (h) ∈ hC[h]. By the isomorphism H(f ) ∼= H(f )op it will be enough to show that H(f ) is not left Noetherian. For each n ∈ Z≥0 define the left ideal In = n X i=0 H(f )hyi. Then In ⊆ In+1 for all n ≥ 0 and we finish the proof by showing that these inclusions are strict. Note that by Lemma 2.1(a), H(f ) = M j,k≥0 xj C[h]yk. 3 Given j, k ≥ 0 and g(h) ∈ C[h], we have xj g(h)ykhyi = xj g(h)σk(h)yk+i. Assume, by way of contradiction, that hyn+1 ∈ In. Then there exist gi(h) ∈ C[h], i = 0, . . . , n, such n that hyn+1 = X i=0 gi(h)σn+1−i(h)yn+1. It follows by Lemma 2.1(a) that h = n X i=0 gi(h)σn+1−i(h). (2.5) As by hypothesis σ(h) = f (h) ∈ hC[h], one can deduce that σn+1−i(h) ∈ f (h)C[h] for all 0 ≤ i ≤ n and (2.5) then implies that h ∈ f (h)C[h], which is a contradiction since under our hypothesis either f (h) = 0 or deg f > 1. This proves that hyn+1 6∈ In for any n ≥ 0 and hence {In}n≥0 is a strict ascending chain of left ideals of H(f ). Remark 2.6. The case f ∈ C of Proposition 2.4 follows also from [5, Corollary 7.3], which applies when σ is not injective. In terms of the endomorphism σ, Proposition 2.4 could be restated as: The algebra H(f ) is right (respectively, left) Noetherian if and only if σ is an automorphism. We recall that a generalized down-up algebra L(g, r, s, γ), given by the parameters g ∈ C[H] and r, s, γ ∈ C, is defined as the unital associative C-algebra generated by d, u and H, subject to the relations: dH − rHd + γd = 0, Hu − ruH + γu = 0, du − sud + g(H) = 0. Generalized down-up algebras were defined in [4] as generalizations of the down-up algebras introduced by Benkart and Roby in [2]. Generalized down-up algebras include all down-up algebras, the algebras similar to the enveloping algebra of sl2 defined by Smith [11], Le Bruyn's conformal sl2 enveloping algebras [6] and Rueda's algebras similar to the enveloping algebra of sl2 [10]. Corollary 2.7. The algebra H(f ) is isomorphic to a generalized down-up algebra if and only if deg f ≤ 1. Proof. Suppose first that deg f ≤ 1, say f (h) = ah + b for a, b ∈ C. Then it is straightforward to verify that H(f ) ∼= L(H − f (H), a, 1, −b), under an isomorphism that sends x, y and h to u, d and H, respectively. Conversely, suppose that deg f > 1. Then by Proposition 2.4 and Lemma 2.1(b), H(f ) is a non-Noetherian domain. Hence H(f ) cannot be isomorphic to a generalized down-up algebra, as a generalized down-up algebra is a domain if and only if it is Noetherian, by Propositions 2.5 and 2.6 of [4]. In view of this result, we will henceforth focus most of our attention on the gener- alized Heisenberg algebras H(f ) with f ∈ C[h] such that deg f > 1. 4 3 The commutative algebra H(f )0 In this short section we record a few useful formulas for computing in H(f ) and then explore an interesting commutative subalgebra of H(f ). Lemma 3.1. Let k ∈ Z≥0 and g ∈ C[h]. Then the following hold: (a) [y, xk] = xk−1(σk(h) − h); (b) [yk, x] = (σk(h) − h)yk−1; (c) (xkgyk)x = x(xkσ(g)yk + xk−1(σk(h) − h)gyk−1); (d) y(xkgyk) = (xkσ(g)yk + xk−1(σk(h) − h)gyk−1)y; (e) xkgyk commutes with xj gyj for all g ∈ C[h] and all j ∈ Z≥0. Proof. Parts (a) and (b) have been established in [5], formulas (6a) -- (6b). We prove part (c) using (b): (xkgyk)x = xkgxyk + xkg[yk, x] = xk+1σ(g)yk + xkg(σk(h) − h)yk−1 = x(xkσ(g)yk + xk−1(σk(h) − h)gyk−1). Formula (d) follows from applying the anti-automorphism ι of (2.3) to (c). Finally, we prove (e) by induction on k, the case k = 0 being trivial: g(xj gyj) = xj σj(g)gyj = xj gσj(g)yj = (xj gyj)g. Now suppose (e) holds for a certain k ≥ 0. Thus we have: (xk+1gyk+1)(xj gyj) = (xk+1gyk)y(xj gyj) = x(xkgyk)(xj σ(g)yj + xj−1(σj(h) − h)gyj−1)y = x(xj σ(g)yj + xj−1(σj(h) − h)gyj−1)(xkgyk)y = (xj gyj)x(xkgyk)y = (xj gyj)(xk+1gyk+1), by (c) by (d) (∗) where (∗) follows from the induction hypothesis. So (e) holds for all k ∈ Z≥0. There is an obvious grading of H(f ) relative to which x has degree 1, y has degree −1 and h has degree 0. We denote the corresponding homogeneous subspaces by H(f )ℓ, for ℓ ∈ Z, so that H(f ) = M H(f )ℓ, ℓ∈Z with H(f )ℓ = M i−k=ℓ C xiC[h]yk. (3.2) We call this the standard grading of H(f ), and, whenever we mention a homogeneous component or element of H(f ), we will always be referring to this standard grading. 5 The subalgebra H(f )0 has basis {xkhj yk k, j ≥ 0} and H(f )ℓ = xℓ H(f )0 if ℓ ≥ 0; H(f )ℓ = H(f )0 y−ℓ if ℓ ≤ 0. Thus we have the decomposition H(f ) = M ℓ>0 xℓ H(f )0 ⊕ H(f )0 ⊕ M ℓ>0 H(f )0 yℓ. Proposition 3.3. The subalgebra H(f )0 is commutative. If deg f > 1, then H(f )0 is a maximal commutative subalgebra of H(f ) which strictly contains C[z, h], the polynomial subalgebra of H(f ) generated by h and the central element z = xy − h. Proof. The first statement is a direct consequence of Lemma 3.1(e). Assume now that deg f > 1. Then σ is injective and has infinite order. For any i, k ∈ Z≥0 and g ∈ C[h], [h, xigyk] = xi(σi(h) − σk(h))gyk. Hence, if g 6= 0, we deduce that [h, xigyk] = 0 ⇐⇒ i = k, and from this it is straightforward to conclude that H(f )0 is the centralizer of h, hence a maximal commutative subalgebra of H(f ). The commuting elements h and z are homogeneous of degree 0 and are easily seen to be algebraically independent, as zk − xkyk is in the span of {xigyi i < k, g ∈ C[h]}. Suppose, by contradiction, that there exist gk ∈ C[h] such that xhy = Pk≥0 gkzk. Then by the argument above we must have gk = 0 for all k > 1 and σ(g1) = h, which is possible only if deg f = 1. Therefore xhy ∈ H(f )0 \ C[z, h]. By Lemma 3.1(c) -- (d), it is possible to extend σ to a C-linear endomorphism σ of H(f )0 so that σ(xkgyk) = xkσ(g)yk + xk−1(σk(h) − h)gyk−1, for all k ∈ Z≥0 and g ∈ C[h]. For simplicity, we still denote this endomorphism by σ instead of σ. By Lemma 3.1(c) -- (d) and Lemma 2.1(a), σ is defined by the relations: θx = xσ(θ), yθ = σ(θ)y, for all θ ∈ H(f )0. (3.4) In particular, (3.4) implies that σ is an algebra endomorphism of H(f )0. 4 locally finite derivations of H(f ) when deg f > 1 Henceforth we will assume that deg f > 1. By Corollary 2.7 we are assuming that H(f ) is not a generalized down-up algebra. Most of our subsequent results do not hold if deg f ≤ 1. Our goal in this section is to determine all locally finite derivations of H(f ). In particular, we will classify all Z-gradings of H(f ) and show that H(f ) has no nontrivial locally nilpotent derivations. Our methods are akin to those used in [12]. Let δ be a C-linear endomorphism of H(f ). We recall the following standard defi- nitions: • δ is a derivation of H(f ) if δ(ab) = δ(a)b + aδ(b); • δ is locally finite if for every a ∈ H(f ) the C-linear span of {δk(a) k ∈ Z≥0} is finite dimensional; • δ is locally nilpotent if for every a ∈ H(f ) there is k ∈ Z≥0 such that δk(a) = 0; 6 • δ is homogeneous of degree r ∈ Z if δ (H(f )ℓ) ⊆ H(f )ℓ+r for all ℓ ∈ Z. Assume δ is any derivation of H(f ). Since H(f ) is finitely generated, there exist homogeneous derivations δ1, . . . , δk of strictly increasing degrees such that δ = δ1 + · · ·+ δk. Moreover, as seen in [12, Lemma 1.1], if δ is locally finite, then so are δ1 and δk, and if δ1 (respectively, δk) is of nonzero degree, then it must be locally nilpotent. We need one final definition. Given a locally nilpotent derivation δ and a ∈ H(f ), define degδ(a) = max{k ∈ Z≥0 δk(a) 6= 0} if a 6= 0; It can be easily checked (see for example [9]) that for define also degδ(0) = −∞. a, b ∈ H(f ), degδ(a + b) ≤ max{degδ(a), degδ(b)}, with equality if degδ(a) 6= degδ(b). Since H(f ) is a domain and C has characteristic 0, we also have, from the Leibniz rule, degδ(ab) = degδ(a) + degδ(b). In particular, kerδ is factorially closed: if δ(ab) = 0 for some nonzero a, b ∈ H(f ), then δ(a) = 0 = δ(b). Lemma 4.1. Assume that deg f > 1. Then all locally finite derivations of H(f ) are homogeneous of degree 0. Proof. Let δ be a locally finite derivation of H(f ). By the above decomposition δ = δ1 + · · · + δk of δ into homogeneous derivations of strictly increasing degrees, it will be enough to show that there are no nonzero homogeneous locally nilpotent derivations of H(f ) of degree r 6= 0. So assume δ is a homogeneous locally nilpotent derivation of H(f ) of degree r 6= 0. Let d = degδ(h) and suppose that d > 0. Then degδ(f (h)) = d deg f and the relation hx = xf (h) yields d + degδ(x) = degδ(x) + d deg f, so deg f = 1, which contradicts our hypothesis. Hence d = 0 and δ(h) = 0. By replacing δ with ιδι−1, where ι is the anti-automorphism defined in (2.3), we can assume that r > 0. Then kerδ contains some nonzero homogeneous element of positive degree. Since elements of H(f ) of positive degree lie in xH(f ) and kerδ is factorially closed, we deduce that δ(x) = 0. Any derivation maps the center of an algebra into itself, so δ restricts to a locally nilpotent derivation of C[z], by Lemma 2.1(c), and thus δ(z) ∈ C. On the other hand, since z = xy − h is homogeneous of degree 0 and δ has positive degree, it must be that δ(z) = 0, and from 0 = δ(z) = xδ(y), we conclude that δ(y) = 0. Then δ = 0 and the lemma is proved. The next theorem, our main result on derivations of H(f ) when deg f > 1, shows that the space of locally finite derivations of H(f ) is one-dimensional over C, spanned by the derivation ∂ defined by ∂(xihj yk) = (i − k)xihj yk, for all i, j, k ∈ Z≥0. (4.2) Theorem 4.3. Assume that deg f > 1. If δ is a locally finite derivation of H(f ), then there is λ ∈ C such that δ(x) = λx, δ(y) = −λy and δ(h) = 0. 7 Proof. Let δ be a locally finite derivation of H(f ). By Lemma 4.1, we know that δ is homogeneous of degree 0, so there are θx, θh, θy ∈ H(f )0 so that δ(x) = xθx, δ(h) = θh, and δ(y) = θyy. In particular, since h commutes with θh, we have δ(g(h)) = g′(h)θh for all g(h) ∈ C[h], where g′(h) denotes the derivative of g(h) with respect to h. Claim 1: θh = 0 and θx + θy = 0. Proof of Claim 1: Write θh = X k≥0 xkgk(h)yk, (4.4) with gk(h) ∈ C[h] and gk(h) = 0 except for finitely many indices k. As observed in the proof of Lemma 4.1, δ restricts to a locally finite derivation of C[z], the center of H(f ), and thus δ(z) ∈ C ⊕ Cz, say δ(z) = µz − λ, with λ, µ ∈ C. Since µz − λ = δ(xy − h) = x(θx + θy)y − θh, we have θh = x(θx + θy)y − µz + λ = x(θx + θy − µ)y + µh + λ. (4.5) In particular, g0(h) = µh + λ. We now apply δ to the relation yh = f (h)y and get θyyh+yθh = f ′(h)θhy +f (h)θyy. As h and θy commute, and yθh = σ(θh)y, by (3.4), we obtain σ(θh) = f ′(h)θh. Now combining (4.4) and (4.6) we deduce that, for every k ≥ 0: σk(f ′(h))gk(h) = σ(gk(h)) + (σk+1(h) − h)gk+1(h). (4.6) (4.7) Setting k = 0 in (4.7) we obtain (f (h) − h)g1(h) = f ′(h)g0(h) − σ(g0(h)). Since we have already established that deg g0 ≤ 1, we deduce now from the latter equation that deg (f (h) − h)g1(h) ≤ deg f , and thus g1 ∈ C. Combining this with the k = 1 case of (4.7), σ(f ′(h))g1(h) = σ(g1(h)) + (σ(f (h)) − h)g2(h), yields g2 = 0, and in turn the latter gives gk = 0 for all k ≥ 2. Using again the relation σ(f ′(h))g1(h) = σ(g1(h)) + (σ(f (h)) − h)g2(h) with g2 = 0 and g1 ∈ C gives g1 = 0. Therefore we have σ(g0) = f ′(h)g0. (4.8) Suppose g0 6= 0, and let a be the leading coefficient of f (h). Then µ 6= 0 and compar- ing leading coefficients in (4.8) yields µa = a( deg f )µ, whence deg f = 1, which is a contradiction. Thus g0 = 0. From the above we conclude that θh = Pk≥0 xkgk(h)yk = 0 and finally by (4.5) we get θx + θy = 0, establishing Claim 1. So far we have shown that δ(x) = xθx, δ(h) = 0, and δ(y) = −θxy, 8 so it remains to be inferred that θx ∈ C. Claim 2: δ(θ) = 0, for all θ ∈ H(f )0. Proof of Claim 2: Since δ(g) = 0 for all g ∈ C[h], it suffices to show that if θ ∈ H(f )0 and δ(θ) = 0, then also δ(xθy) = 0. This follows easily using the fact that H(f )0 is commutative, as proved in Proposition 3.3. From Claim 2 it follows that, for all k ≥ 0, δ(θk x) = 0, which implies that δk(x) = xθk x. x k ∈ Z≥0} must then be finite dimensional. This As δ is locally finite, the span of {θk is possible only if θx ∈ C, thus finishing the proof of the theorem. Since locally nilotent derivations are locally finite, we derive the following corollary. Corollary 4.9. Assume that deg f > 1. Then H(f ) has no nonzero locally nilpotent derivations. Suppose that H(f ) = Lα∈C Vα is a grading. Define the C-linear endomorphism δ of H(f ) by δ(vα) = αvα for all vα ∈ Vα and all α ∈ C. It is immediate to check that δ is a diagonalizable derivation of H(f ) whose eigenvalues are those α ∈ C such that Vα 6= (0). Conversely, if δ is a diagonalizable derivation, then δ determines a grading where Vα is the α-eigenspace of δ. Furthermore, diagonalizable derivations are clearly locally finite. Thus, we deduce from Theorem 4.3 that, except for the trivial grading in which every element of H(f ) has degree 0, H(f ) only admits the standard grading defined in (3.2), up to scaling by some integer. More precisely, we have: Corollary 4.10. Assume that deg f > 1. Then for any Z-grading of H(f ), there is an integer ℓ ∈ Z such that, relative to that grading, x has degree ℓ, y has degree −ℓ and h has degree 0. 5 Automorphisms of H(f ) when deg f > 1 When deg f = 1 the algebra H(f ) is a Noetherian generalized down-up algebra, by Corollary 2.7, and the automorphisms of the latter have been investigated in [3]. We continue to assume that deg f > 1 and note again that our results do not generalize to the cases with deg f ≤ 1. Since H(f ) has no nonzero locally nilpotent derivations, it seems natural to conjec- ture that the automorphism group of H(f ) is somewhat small. However, over C we can consider also the exponential of a diagonalizable derivation. Specifically, let c ∈ C and let ∂ be the derivation of H(f ) defined in (4.2). Then the expression exp(c∂) := ∞ X k=0 (c∂)k k! defines an automorphism of H(f ) satisfying exp(c∂)(x) = ∞ X k=0 ck k! x = exp(c)x, exp(c∂)(y) = exp(−c)y, exp(c∂)(h) = h, 9 with inverse exp(−c∂). The above motivates the following definition. For each λ ∈ C∗, let φλ be the automorphism of H(f ) defined by φλ(x) = λx, φλ(y) = λ−1y, φλ(h) = h. (5.1) The group of algebra automorphisms of H(f ) will be denoted by AutC(H(f )). We have a first description of AutC(H(f )) below. Proposition 5.2. Assume deg f > 1. Then the following hold: (a) Any automorphism of H(f ) restricts to an automorphisms of C[h], and x and y are eigenvectors. (b) {φ ∈ AutC(H(f )) φ(h) = h} = {φλ λ ∈ C∗} ∼= C∗, and this is a central subgroup of AutC(H(f )). (c) {φ ∈ AutC(H(f )) φ(x) = x} is a finite cyclic subgroup whose order divides ( deg f ) − 1. Proof. Let φ be an automorphism of H(f ). Then as argued in Claim 4 of the proof of [8, Theorem 5], the relation φ(h)φ(x) = φ(x)f (φ(h)) with deg f > 1 implies that φ(h) ∈ C[h]; applying this result to φ−1 gives that φ(h) = ah + b, for some a, b ∈ C with a 6= 0. Now writing φ(x) as a sum of terms of the form xigi,j yj with i, j ∈ Z≥0 and gi,j ∈ C[h], and comparing the corresponding expressions for φ(h)φ(x) and φ(x)f (φ(h)), we obtain φ(x) ∈ H(f )1. Similarly, φ(y) ∈ H(f )−1, so φ is homogeneous of degree 0. Thus, there exist θx, θy ∈ H(f )0 such that φ(x) = xθx and φ(y) = θyy. Applying the same reasoning to φ−1, we deduce that θx, θy ∈ C∗, which proves (a). Now assume φ ∈ AutC(H(f )) and φ(h) = h. By (a) there exist λ, µ ∈ C∗ such that φ(x) = λx and φ(y) = µy. Applying φ to the relation [y, x] = f (h) − h yields λµ = 1, so φ = φλ. This proves the equality in (b), and the isomorphism {φλ λ ∈ C∗} ∼= C∗ is clear, as φλ ◦ φµ = φλµ for all λ, µ ∈ C∗. Next, we show that the subgroup {φλ λ ∈ C∗} is central in AutC(H(f )). Let λ ∈ C∗, and suppose ψ ∈ AutC(H(f )) is arbitrary. By (a) we know that ψ(h) ∈ C[h], which implies that φλ ◦ ψ(h) = ψ ◦ φλ(h). But as x and y are eigenvalues for any automorphism of H(f ), φλ ◦ ψ and ψ ◦ φλ also agree on these generators, and thus φλ ◦ ψ = ψ ◦ φλ. To prove part (c), suppose that φ ∈ AutC(H(f )) and φ(x) = x. We know already that φ(h) = ah + b and φ(y) = cy, for some a, b, c ∈ C with a, c 6= 0. Then xf (ah + b) = xφ(f (h)) = φ(h)x = (ah + b)x = x(af (h) + b), and we obtain f (ah + b) = af (h) + b. (5.3) Therefore, c(f (h) − h) = c[y, x] = φ([y, x]) = φ(f (h) − h) = af (h) + b − (ah + b) = a(f (h) − h), 10 and we conclude that c = a. d Write f (h) = Pn k=0 akhk, where n = deg f and ak ∈ C. Applying the derivation dh to (5.3) n − 1 times yields an−1f (n−1)(ah + b) = af (n−1)(h), as n − 1 ≥ 1. As f (n−1)(h) = (n − 1)!(nanh + an−1), we obtain an−1 = 1 and b = (a − 1)an−1 nan . (5.4) Let Un−1 = {ζ ∈ C∗ ζ n−1 = 1} be the cyclic group of order n − 1, and define a map Ψ : {φ ∈ AutC(H(f )) φ(x) = x} −→ Un−1, φ 7→ a, where φ(h) = ah + b. Then Ψ is well defined by (5.4), and it is a group homomorphism. If Ψ(φ) = 1 for some φ ∈ AutC(H(f )) with φ(x) = x, then the above shows that φ(y) = y and φ(h) = h + b. Again by (5.4) we deduce that b = 0, so φ is the identity on H(f ). This shows that Ψ is an injective group homomorphism and thus {φ ∈ AutC(H(f )) φ(x) = x} is isomorphic to a subgroup of Un−1; hence it is a finite cyclic group whose order divides n − 1. It is now an easy matter to determine the structure of AutC(H(f )). The symbol × used below denotes the internal direct product of subgroups of a group. Theorem 5.5. Assume deg f > 1. Then AutC(H(f )) = {φλ λ ∈ C∗} ×{φ ∈ AutC(H(f )) φ(x) = x} (5.6) is an abelian group, where: • {φλ λ ∈ C∗} ∼= C∗ and φλ is defined in (5.1); • {φ ∈ AutC(H(f )) φ(x) = x} is a finite cyclic group whose order divides ( deg f )− 1 and which, as a set, can be identified with {(a, b) ∈ C∗×C f (ah+b) = af (h)+b} via the correspondence φ 7→ (a, b), where φ(h) = ah + b. Proof. Since we have already seen in Proposition 5.2 that {φλ λ ∈ C∗} is central, in order to prove the direct product decomposition in (5.6), it remains to show that the two subgroups have trivial intersection, which is clear, and generate AutC(H(f )). Let ψ ∈ AutC(H(f )). Then there is λ ∈ C∗ such that ψ(x) = λx, whence φ−1 λ ◦ ψ ∈ {φ ∈ AutC(H(f )) φ(x) = x}, and this shows the latter claim. Moreover, since {φ ∈ AutC(H(f )) φ(x) = x} is abelian, by Proposition 5.2(c), the group AutC(H(f )) must also be abelian. The remaining parts of the theorem have already been proved, except for the ob- servation that {φ ∈ AutC(H(f )) φ(x) = x} can be identified with the set {(a, b) ∈ C∗ × C f (ah + b) = af (h) + b}. Indeed, if φ(x) = x, then we have seen in the proof of Proposition 5.2 that φ(h) = ah + b and φ(y) = ay, for some a, b ∈ C with a 6= 0, and (5.3) must hold. This shows that the correspondence φ 7→ (a, b) is well defined and one-to-one. Conversely, given (a, b) ∈ C∗ × C satisfying f (ah + b) = af (h) + b, it is routine to check that there is an automorphism of H(f ) defined by the conditions φ(x) = x, φ(y) = ay, φ(h) = ah + b, and this shows the correspondence is onto. 11 Remark 5.7. Any pair (a, b) ∈ C∗ × C satisfying f (ah + b) = af (h) + b must also satisfy (5.4), where n = deg f , although the conditions in (5.4) are not sufficient (see the examples below). Thus, for each (n − 1)-th root of unity a, the corresponding scalar b is determined by (5.4), but one still needs to check the relation f (ah + b) = af (h) + b for the pair (a, b). Examples 5.8. (a) If deg f = 2, then n = 2 in (5.4), so a = 1 and b = 0, and the pair (1, 0) corresponds to the identity map. It follows that the group {φ ∈ AutC(H(f )) φ(x) = x} is trivial and AutC(H(f )) ∼= C∗. (b) Let f (h) = h3 + h. Then n = 3 in (5.4), so a = ±1. If a = 1, then b = 0, and the corresponding automorphism is the identity. If a = −1, then b = 0, and in fact f (−h) = −f (h). Therefore there is an automorphism φ of H(f ) such that φ(x) = x, φ(y) = −y, φ(h) = −h, and AutC(H(f )) ∼= C∗ × Z2. (c) Let f (h) = h3 + h + 1. Then n = 3 in (5.4), so a = ±1. If a = −1, then (5.4) yields b = 0, but f (−h) 6= −f (h), so the group {φ ∈ AutC(H(f )) φ(x) = x} is trivial and AutC(H(f )) ∼= C∗. (d) Let f (h) = hn, for n > 1. Then (5.4) says that a is a (n − 1)-th root of unity and b = 0. Moreover, f (ah) = anf (h) = af (h) for any (n − 1)-th root of unity a. Hence AutC(H(f )) ∼= C∗ × Zn−1. (e) Let f (h) = hn + 1, for n > 1. Then, as before, a is a (n − 1)-th root of unity and b = 0. However, in this case, f (ah) = ahn + 1 whereas af (h) = ahn + a, so equality holds if and only if a = 1. Hence AutC(H(f )) ∼= C∗. (f) Let f (h) = hn + hk+1, for some n ≥ 4, and take any 1 ≤ k < n − 1 such that k n − 1. Then a is a (n − 1)-th root of unity and b = 0. In this case f (ah) = ahn+ak+1hk+1 whereas af (h) = ahn+ahk+1, so equality holds if and only if ak = 1. By the hypothesis that k n − 1, we deduce that AutC(H(f )) ∼= C∗ × Zk. References [1] V. V. Bavula. Generalized Weyl algebras and their representations. Algebra i Analiz, 4(1):75 -- 97, 1992. [2] G. Benkart and T. Roby. Down-up algebras. J. Algebra, 209(1):305 -- 344, 1998. [3] Paula A. A. B. Carvalho and Samuel A. Lopes. Automorphisms of generalized down-up algebras. Comm. Algebra, 37(5):1622 -- 1646, 2009. [4] Thomas Cassidy and Brad Shelton. Basic properties of generalized down-up alge- bras. J. Algebra, 279(1):402 -- 421, 2004. 12 [5] David A. Jordan. Down-up algebras and ambiskew polynomial rings. J. Algebra, 228(1):311 -- 346, 2000. [6] Lieven Le Bruyn. Conformal sl2 enveloping algebras. Comm. Algebra, 23(4):1325 -- 1362, 1995. [7] Rencai Lu, Volodymyr Mazorchuk, and Kaiming Zhao. Simple weight modules over weak generalized Weyl algebras. J. Pure Appl. Algebra, 219(8):3427 -- 3444, 2015. [8] Rencai Lu and Kaiming Zhao. Finite-dimensional simple modules over generalized Heisenberg algebras. Linear Algebra Appl., 475:276 -- 291, 2015. [9] L. Makar-Limanov. Locally nilpotent derivations, a new ring invariant and appli- cations. Available at http://math.wayne.edu/~lml/lmlnotes.pdf. [10] Sonia Rueda. Some algebras similar to the enveloping algebra of sl(2). Comm. Algebra, 30(3):1127 -- 1152, 2002. [11] S. P. Smith. A class of algebras similar to the enveloping algebra of sl(2). Trans. Amer. Math. Soc., 322(1):285 -- 314, 1990. [12] Mariano Su´arez-Alvarez and Quimey Vivas. Automorphisms and isomorphisms of quantum generalized Weyl algebras. J. Algebra, 424:540 -- 552, 2015. Samuel A. Lopes CMUP, Faculdade de Ciencias, Universidade do Porto Rua do Campo Alegre 687 4169-007 Porto, Portugal [email protected] 13
1605.03880	2	1605	2017-03-27T17:59:15	Fiat categorification of the symmetric inverse semigroup IS_n and the semigroup F^*_n	[ "math.RA", "math.CT", "math.GR", "math.RT" ]	Starting from the symmetric group $S_n$, we construct two fiat $2$-categories. One of them can be viewed as the fiat "extension" of the natural $2$-category associated with the symmetric inverse semigroup (considered as an ordered semigroup with respect to the natural order). This $2$-category provides a fiat categorification for the integral semigroup algebra of the symmetric inverse semigroup. The other $2$-category can be viewed as the fiat "extension" of the $2$-category associated with the maximal factorizable subsemigroup of the dual symmetric inverse semigroup (again, considered as an ordered semigroup with respect to the natural order). This $2$-category provides a fiat categorification for the integral semigroup algebra of the maximal factorizable subsemigroup of the dual symmetric inverse semigroup.	math.RA	math	FIAT CATEGORIFICATION OF THE SYMMETRIC INVERSE SEMIGROUP ISn AND THE SEMIGROUP F ∗ n PAUL MARTIN AND VOLODYMYR MAZORCHUK Abstract. Starting from the symmetric group Sn, we construct two fiat 2- categories. One of them can be viewed as the fiat "extension" of the natural 2-category associated with the symmetric inverse semigroup (considered as an ordered semigroup with respect to the natural order). This 2-category provides a fiat categorification for the integral semigroup algebra of the symmetric in- verse semigroup. The other 2-category can be viewed as the fiat "extension" of the 2-category associated with the maximal factorizable subsemigroup of the dual symmetric inverse semigroup (again, considered as an ordered semi- group with respect to the natural order). This 2-category provides a fiat categorification for the integral semigroup algebra of the maximal factorizable subsemigroup of the dual symmetric inverse semigroup. 1. Introduction and description of the results Abstract higher representation theory has its origins in the papers [BFK, CR, Ro1, Ro2] with principal motivation coming from [Kh, Str]. For finitary 2-categories, ba- sics of 2-representation theory were developed in [MM1, MM2, MM3, MM4, MM5, MM6] and further investigated in [GrMa1, GrMa2, Xa, Zh1, Zh2, Zi, CM, MZ, MaMa, KMMZ], see also [KiMa1] for applications. For different ideas on higher rep- resentation theory, see also [FB, BBFW, El, Pf, Ka] and references therein. The major emphasis in [MM1, MM2, MM3, MM4, MM5, MM6] is on the study of so-called fiat 2-categories, which are 2-categorical analogues of finite dimensional algebras with involution. Fiat 2-categories appear naturally both in topology and representation theory. They have many nice properties and the series of papers mentioned above develops an essential starting part of 2-representation theory for fiat categories. Many examples of 2-categories appear naturally in semigroup theory, see [KuMa2, GrMa1, GrMa2, Fo]. The easiest example is the 2-category associated to a monoid with a fixed admissible partial order, see Subsection 4.1 for details. Linear ana- logues of these 2-categories show up naturally in representation theory, see [GrMa1, GrMa2]. A classical example of an ordered monoid is an inverse monoid with re- spect to the natural partial order. There is a standard linearization procedure, which allows one to turn a 2-category of a finite ordered monoid into a finitary 2-category, see Subsection 3.2 for details. One serious disadvantage with linearizations of 2-categories associated to finite ordered monoids is the fact that they are almost never fiat. The main reason for that is lack of 2-morphisms which start from the identity 1-morphism. In the present paper we construct two natural "extensions" of the symmetric group to 2-categories whose linearizations are fiat. One of them becomes a nice 2-categorical analogue (categorification) for the symmetric inverse semigroup ISn. The other one 1 2 PAUL MARTIN AND VOLODYMYR MAZORCHUK becomes a nice 2-categorical analogue for the maximal factorizable subsemigroup n in the dual symmetric inverse semigroup I ∗ F ∗ n. The main novel component of the present paper is in the definitions and construc- tions of the main objects. To construct our 2-categories, we, essentially, have to define three things: • sets of 2-morphisms between elements of Sn; • horizontal composition of 2-morphisms; • vertical composition of 2-morphisms. In the case which eventually leads to ISn, we view elements of Sn as binary relations in the obvious way and define 2-morphisms between two elements of Sn as the set of all binary relations contained in both these elements. We chose vertical composition to be given by intersection of relations and horizontal composition to be given by the usual composition of relations. Although all these choices are rather natural, none of them seems to be totally obvious. Verification that this indeed defines a 2-category requires some technical work. In the case which eventually leads to F ∗ n , we do a similar thing, but instead of binary relations, we realize Sn inside the partition monoid. For 2-morphisms between elements σ and τ in Sn, we use those partitions which contain both σ and τ . All details on both constructions and all verifications can be found in Section 2. Section 3 recalls the theory of k-linear 2-categories and gives explicit constructions for a finitary k-linear 2-category starting from a finite 2-category. In Section 4 we establish that our constructions lead to fiat 2-categories. We also recall, in more details, the standard constructions of finitary 2-categories, starting from ISn and F ∗ n , considered as ordered monoids, and show that the 2-categories obtained in this way are not fiat. In Section 5 we make the relation between our constructions and ISn and F ∗ n precise. In fact, we show that the decategorification of our first construction is isomorphic to the semigroup algebra Z[ISn], with respect to the so-called Mobius basis in Z[ISn], cf. [Ste, Theorem 4.4]. Similarly, we show that the decategorification of our second construction is isomorphic to the semigroup algebra Z[F ∗ n ], with respect to a similarly defined basis. We complete the paper with two explicit examples in Section 6. Acknowledgment. The main part of this research was done during thew visit of the second author to University of Leeds in October 2014. Financial support of EP- SRC and hospitality of University of Leeds are gratefully acknowledged. The first author is partially supported by EPSRC under grant EP/I038683/1. The second author is partially supported by the Swedish Research Council and Goran Gustafs- son Foundation. We thank Stuart Margolis for stimulating discussions. 2. Two 2-categorical "extensions" of Sn 2.1. 2-categories. A 2-category is a category enriched over the monoidal category Cat of small categories. This means that a 2-category C consists of • objects i, j, . . . ; • small morphism categories C(i, j); • bifunctorial compositions; FIAT CATEGORIFICATION OF ISn AND F ∗ n 3 • identity objects 1i ∈ C (i, i); which satisfy the obvious collection of (strict) axioms. Objects in morphism cate- gories are usually called 1-morphisms (for example, all 1i are 1-morphisms) while morphisms in morphism categories are usually called 2-morphisms. Composition of 2-morphisms inside a fixed C (i, j) is called vertical and denoted ◦1. Composition of 2-morphisms coming from the bifunctorial composition in C is called horizontal and denoted ◦0. We refer the reader to [Mac, Le] for more details on 2-categories. The main example of a 2-category is Cat itself, where • objects are small categories; • 1-morphisms are functors; • 2-morphisms are natural transformations; • composition is the usual composition; • identity 1-morphisms are identity functors. 2.2. First 2-category extending Sn. For n ∈ N := {1, 2, 3, . . . }, consider the set n = {1, 2, . . . , n} and let Sn denote the symmetric group of all bijective transforma- tions of n under composition. We consider also the monoid Bn = 2n×n of all binary relations on n which is identified with the monoid of n×n-matrices over the Boolean semiring B := {0, 1} by taking a relation to its adjacency matrix. Note that Bn is an ordered monoid with respect to usual inclusions of binary relations. We identify Sn with the group of invertible elements in Bn in the obvious way. We now define a 2-category A = A n. To start with, we declare that • A has one object i; • 1-morphisms in A are elements in Sn; • composition · of 1-morphisms is induced from Sn; • the identity 1-morphism is the identity transformation idn ∈ Sn. It remains to define 2-morphisms in A and their compositions. • For π, σ ∈ Sn, we define HomA (π, σ) as the set of all α ∈ Bn such that α ⊆ π ∩ σ. • For π, σ, τ ∈ Sn, and also for α ∈ HomA (π, σ) and β ∈ HomA (σ, τ ), we define β ◦1 α := β ∩ α. • For π ∈ Sn, we define the identity element in HomA (π, π) to be π. • For π, σ, τ, ρ ∈ Sn, and also for α ∈ HomA (π, σ) and β ∈ HomA (τ, ρ), we define β ◦0 α := βα, the usual composition of binary relations. Proposition 1. The construct A above is a 2-category. Proof. Composition · of 1-morphisms is associative as Sn is a group. The vertical composition ◦1 is clearly well-defined. It is associative as ∩ is associative. If we have α ∈ HomA (π, σ) or α ∈ HomA (σ, π), then α ⊆ π and thus α ∩ π = α. Therefore π ∈ HomA (π, π) is the identity element. Let us check that the horizontal composition ◦0 is well-defined. From α ⊆ π and β ⊆ τ and the fact that Bn is ordered, we have βα ⊆ τ α ⊆ τ π. Similarly, from α ⊆ σ and β ⊆ ρ and the fact that Bn is ordered, we have βα ⊆ ρα ⊆ ρσ. It follows 4 PAUL MARTIN AND VOLODYMYR MAZORCHUK that βα ∈ HomA (τ π, ρσ) and thus ◦0 is well-defined. Its associativity follows from the fact that usual composition of binary relations is associative. It remains to check the interchange law, that is the fact that, for any 1-morphisms π, σ, ρ, τ, µ, ν and for any α ∈ HomA (π, σ), β ∈ HomA (τ, µ), γ ∈ HomA (σ, ρ) and δ ∈ HomA (µ, ν), we have (2.1) (δ ◦0 γ) ◦1 (β ◦0 α) = (δ ◦1 β) ◦0 (γ ◦1 α). Assume first that σ = µ = idn. In this case both α, β, γ and δ are subrelations of the identity relation idn. Note that, given two subrelations x and y of the identity relation idn, their product xy as binary relations equals x ∩ y. Hence, in this particular case, both sides of (2.1) are equal to α ∩ β ∩ γ ∩ δ. Before proving the general case, we will need the following two lemmata: Lemma 2. Let π, σ, τ, ρ ∈ Sn. (i) Left composition with π induces a bijection from HomA (σ, τ ) to HomA (πσ, πτ ). (ii) For any α ∈ HomA (σ, τ ) and β ∈ HomA (τ, ρ), we have π ◦0 (β ◦1 α) = (π ◦0 β) ◦1 (π ◦0 α). Proof. Left composition with π maps an element (y, x) of α ∈ HomA (σ, τ ) to (π(y), x) ∈ HomA (πσ, πτ ). As π is an invertible transformation of n, multiplying with π−1 returns (π(y), x) to (y, x). This implies claim (i). Claim (ii) follows from claim (i) and the observation that composition with invertible maps commutes with taking intersections. (cid:3) Lemma 3. Let π, σ, τ, ρ ∈ Sn. (i) Right composition with π induces a bijection from HomA (σ, τ ) to HomA (σπ, τ π). (ii) For any α ∈ HomA (σ, τ ) and β ∈ HomA (τ, ρ), we have (β ◦1 α) ◦0 π = (β ◦0 π) ◦1 (α ◦0 π). Proof. Analogous to the proof of Lemma 2. (cid:3) Using Lemmata 2 and 3 together with associativity of ◦0, right multiplication with σ−1 and left multiplication with µ−1 reduces the general case of (2.1) to the case σ = µ = idn considered above. This completes the proof. (cid:3) 2.3. Second 2-category extending Sn. For n ∈ N, consider the corresponding partition semigroup Pn, see [Jo, Mar1, Mar2, Maz1]. Elements of Pn are partitions of the set n := {1, 2, . . . , n, 1′, 2′, . . . , n′} into disjoint unions of non-empty subsets, called parts. Alternatively, one can view elements of Pn as equivalence relations on n. Multiplication (ρ, π) 7→ ρπ in Pn is given by the following mini-max algorithm, see [Jo, Mar1, Mar2, Maz1] for details: • Consider ρ as a partition of {1′, 2′, . . . , n′, 1′′, 2′′, . . . , n′′} using the map x 7→ x′ and x′ 7→ x′′, for x ∈ n. • Let τ be the minimum, with respect to inclusions, partition of {1, 2, . . . , n, 1′, 2′, . . . , n′, 1′′, 2′′, . . . , n′′}, such that each part of both ρ and π is a subset of a part of τ . FIAT CATEGORIFICATION OF ISn AND F ∗ n 5 • Let σ to be the maximum, with respect to inclusions, partition of {1, 2, . . . , n, 1′′, 2′′, . . . , n′′}, such that each part of σ is a subset of a part of τ . • Define the product ρπ as the partition of n induced from σ via the map x′′ 7→ x′, for x ∈ n. Note that Sn is, naturally, a submonoid of Pn. Moreover, Sn is the maximal subgroup of all invertible elements in Pn. A part of ρ ∈ Pn is called a propagating part provided that it intersects both sets {1, 2, . . . , n} and {1′, 2′, . . . , n′}. Partitions in which all parts are propagating are called propagating partitions. The set of all propagating partitions in Pn is denoted by PPn, it is a submonoid of Pn. The monoid Pn is naturally ordered with respect to inclusions: ρ ≤ τ provided that each part of ρ is a subset of a part in τ . With respect to this order, the partition of n with just one part is the maximum element, while the partition of n into singletons is the minimum element. This order restricts to PPn. As elements of Pn are just equivalence relations, the poset Pn is a lattice and we denote by ∧ and ∨ the corresponding meet and join operations, respectively. The poset PPn is a sublattice in Pn with the same meet and join. As Sn ⊂ PPn, all meets and joints in Pn of elements from Sn belong to PPn. We now define a 2-category B = B n. Similarly to Subsection 2.2, we start by declaring that • B has one object i; • 1-morphisms in B are elements in Sn; • composition of 1-morphisms is induced from Sn; • the identity 1-morphism is the identity transformation idn. It remains to define 2-morphisms in B and their compositions. • For π, σ ∈ Sn, we define HomB (π, σ) as the set of all α ∈ PPn such that we have both, π ≤ α and σ ≤ α. • For π, σ, τ ∈ Sn, and also any α ∈ HomB (π, σ) and β ∈ HomB (σ, τ ), we define β ◦1 α := β ∨ α. • For π ∈ Sn, we define the identity element in HomB (π, π) to be π. • For π, σ, τ, ρ ∈ Sn, and also any α ∈ HomB (π, σ) and β ∈ HomB (τ, ρ), we define β ◦0 α := βα, the usual composition of partitions. Proposition 4. The construct B above is a 2-category. Proof. The vertical composition ◦1 is clearly well-defined. It is associative as ∨ is associative. If α ∈ HomB (π, σ) or α ∈ HomB (σ, π), then π ≤ α and thus α∨π = α. Therefore π ∈ HomB (π, π) is the identity element. Let us check that the horizontal composition ◦0 is well-defined. From π ≤ α and τ ≤ β and the fact that Pn is ordered, we have τ π ≤ τ α ≤ βα. Similarly, from σ ≤ α and ρ ≤ β and the fact that Pn is ordered, we have ρσ ≤ ρα ≤ βα. It follows that βα ∈ HomB (τ π, ρσ) and thus ◦0 is well-defined. Its associativity follows from the fact that usual composition of partitions is associative. 6 PAUL MARTIN AND VOLODYMYR MAZORCHUK It remains to check the interchange law (2.1). For this we fix any 1-morphisms π, σ, ρ, τ, µ, ν and any α ∈ HomB (π, σ), β ∈ HomB (τ, µ), γ ∈ HomB (σ, ρ) and δ ∈ HomB (µ, ν). Assume first that σ = µ = idn. In this case both α, β, γ and δ are partitions containing the identity relation idn. Note that, given two partitions x and y containing the identity relation idn, their product xy as partitions equals x ∨ y. Hence, in this particular case, both sides of (2.1) are equal to α ∨ β ∨ γ ∨ δ. Before proving the general case, we will need the following two lemmata: Lemma 5. Let π, σ, τ, ρ ∈ Sn. (i) Left composition with π induces a bijection between the sets HomB (σ, τ ) and HomB (πσ, πτ ). (ii) For any α ∈ HomB (σ, τ ) and β ∈ HomB (τ, ρ), we have π ◦0 (β ◦1 α) = (π ◦0 β) ◦1 (π ◦0 α). Proof. Left composition with π simply renames elements of {1′, 2′, . . . , n′} in an invertible way. This implies claim (i). Claim (ii) follows from claim (i) and the observation that composition with invertible maps commutes with taking unions. (cid:3) Lemma 6. Let π, σ, τ, ρ ∈ Sn. (i) Right composition with π induces a bijection between the sets HomB (σ, τ ) and HomB (σπ, τ π). (ii) For any α ∈ HomB (σ, τ ) and β ∈ HomB (τ, ρ), we have (β ◦1 α) ◦0 π = (β ◦0 π) ◦1 (α ◦0 π). Proof. Analogous to the proof of Lemma 5. (cid:3) Using Lemmata 5 and 6 together with associativity of ◦0, right multiplication with σ−1 and left multiplication with µ−1 reduces the general case of (2.1) to the case σ = µ = idn considered above. This completes the proof. (cid:3) 3. 2-categories in the linear world For more details on all the definitions in Section 3, we refer to [GrMa2]. 3.1. Finitary 2-categories. Let k be a field. A k-linear category C is called fini- tary provided that it is additive, idempotent split and Krull-Schmidt with finitely many isomorphism classes of indecomposable objects and finite dimensional homo- morphism spaces. A 2-category C is called prefinitary (over k) provided that (I) C has finitely many objects; (II) each C (i, j) is a finitary k-linear category; (III) all compositions are biadditive and k-linear whenever the notion makes sense. Following [MM1], a prefinitary 2-category C is called finitary provided that (IV) all identity 1-morphisms are indecomposable. FIAT CATEGORIFICATION OF ISn AND F ∗ n 7 3.2. k-linearization of finite categories. For any set X, let us denote by k[X] the vector space (over k) of all formal linear combinations of elements in X with coefficients in k. Then we can view X as the standard basis in k[X]. By convention, k[X] = {0} if X = ∅. Let C be a finite category, that is a category with a finite number of objects and a finite number of morphisms. Define the k-linearization Ck of C as follows: • the objects in Ck and C are the same; • we have Ck(i, j) := k[C(i, j)]; • composition in Ck is induced from that in C by k-bilinearity. 3.3. k-additivization of finite categories. Assume that objects of the category C are 1, 2,. . . , k. For C as in Subsection 3.2, define the additive k-linearization C⊕ of C in the following way: k • objects in C⊕ k are elements in Zk ≥0, we identify (m1, m2, . . . , mk) ∈ Zk ≥0 with the symbol 1 ⊕ · · · ⊕ 1 ⊕ 2 ⊕ · · · ⊕ 2 ⊕ · · · ⊕ k ⊕ · · · ⊕ k ; m1 times m2 times mk times {z } {z } {z } • the set C⊕ k (i1 ⊕ i2 ⊕ · · · ⊕ il, j1 ⊕ j2 ⊕ · · · ⊕ jm) is given by the set of all matrices of the form   f11 f21 ... fm1 f12 f22 ... fm2 . . . . . . . . . . . . f1l f2l ... fml   where fst ∈ Ck(it, js); • composition in C⊕ k is given by the usual matrix multiplication; • the additive structure is given by addition in Zk ≥0. One should think of C⊕ k as the additive category generated by Ck. 3.4. k-linearization of finite 2-categories. Let now C be a finite 2-category. We define the k-linearization C k of C over k as follows: • C k and C have the same objects; • we have C k(i, j) := C(i, j)⊕ k ; • composition in C k is induced from composition in C by biadditivity and k-bilinearity. By construction, the 2-category C k satisfies conditions (I) and (III) from the def- inition of a finitary 2-category. A part of condition (II) related to additivity and finite dimensionality of morphism spaces is also satisfied. Therefore, the 2-category C k is finitary if and only if, the 2-endomorphism k-algebra of every 1-morphism in C k is local. 8 PAUL MARTIN AND VOLODYMYR MAZORCHUK 3.5. k-finitarization of finite 2-categories. Let C be a finite 2-category. Con- sider the 2-category C k. We define the finitarization kC of C k as follows: • kC and C k have the same objects; • kC(i, j) is defined to be the idempotent completion of C k(i, j); • composition in kC is induced from composition in C. By construction, the 2-category kC is prefinitary. Therefore, the 2-category kC is finitary if and only if, the 2-endomorphism k-algebra of every identity 1-morphism in kC is local. 3.6. Idempotent splitting. Let C be a prefinitary 2-category. If C does not satisfy condition (IV), then there is an object i ∈ C such that the endomorphism algebra EndkC (1i) is not local, that is, contains a non-trivial idempotent. In this subsection we describe a version of "idempotent splitting", for all EndkC (1i), to turn C into a finitary 2-category which we denote by C . For i ∈ C, the 2-endomorphism algebra of 1i is equipped with two unital associative operations, namely, ◦0 and ◦1. These two operations satisfy the interchange law. By the classical Eckmann-Hilton argument (see, for example, [EH] or [Ko, Subsec- tion 1.1]), both these operations, when restricted to the 2-endomorphism algebra of 1i, must be commutative and, in fact, coincide. Therefore we can unambiguously speak about the commutative 2-endomorphism algebra EndC (1i). Let ε(j) i , where j = 1, 2, . . . , ki, be a complete list of primitive idempotents in EndC (1i). Note that the elements ε(j) are identities in the minimal ideals of EndC (1i) and hence are canonically determined (up to permutation). i We now define a new 2-category, which we denote by C , in the following way: • Objects in C are i(s), where i ∈ C and s = 1, 2, . . . , ki. • 1-morphisms in C(i(s), j(t)) are the same as 1-morphisms in C (i, j). • for 1-morphisms F, G ∈ C (i(s), j(t)), the set HomC (F, G) equals j ◦0 HomC (F, G) ◦0 ε(s) ε(t) i . • The identity 1-morphism in C (i(s), i(s)) is 1i. • All compositions are induced from C . Lemma 7. Let C be a prefinitary 2-category. Then the construct C is a finitary 2-category. Proof. The fact that C is a 2-category follows from the fact that C is a 2-category, by construction. For C , conditions (I), (II) and (III) from the definition of a prefinitary 2-category, follow from the corresponding conditions for the original category C . It remains to show that C satisfies (IV). By construction, the endomorphism alge- bra of the identity 1-morphism 1i in C (i(s), i(s)) is ε(s) i ◦0 EndC (1i) ◦0 ε(s) i . The latter algebra is local as ε(s) tion (IV) is satisfied and completes the proof. is a minimal idempotent. This means that condi- (cid:3) i FIAT CATEGORIFICATION OF ISn AND F ∗ n 9 Starting from C and taking, for each i ∈ C , a direct sum of i(s), where s = 1, 2, . . . , ki, one obtains a 2-category biequivalent to the original 2-category C . The 2-categories C and C are, clearly, Morita equivalent in the sense of [MM4]. Warning: Despite of the fact that C (i(s), j(t)) and C(i, j) have the same 1- morphisms, these two categories, in general, have different indecomposable 1-mor- phisms as the sets of 2-morphisms are different. In particular, indecomposable 1-morphisms in C(i, j) may become isomorphic to zero in C (i(s), j(t)). 4. Comparison of kA n and kB n to 2-categories associated with ordered monoids ISn and F ∗ n 4.1. 2-categories and ordered monoids. Let (S, ·, 1) be a monoid and ≤ be an admissible order on S, that is a partial (reflexive) order such that s ≤ t implies both sx ≤ tx and xs ≤ xt, for all x, s, t ∈ S. Then we can associate with S a 2-category S = S S = S (S,·,1,≤) defined as follows: • S has one object i; • 1-morphisms are elements in S; • for s, t ∈ S, the set HomS (s, t) is empty if s 6≤ t and contains one element (s, t) otherwise; • composition of 1-morphisms is given by ·; • both horizontal and vertical compositions of 2-morphism are the only pos- sible compositions (as sets of 2-morphisms are either empty or singletons); • the identity 1-morphism is 1. Admissibility of ≤ makes the above well-defined and ensures that S becomes a 2-category. A canonical example of the above is when S is an inverse monoid and ≤ is the natural partial order on S defined as follows: s ≤ t if and only if s = et for some idempotent e ∈ S. 4.2. (Co)ideals of ordered semigroups. Let S be a semigroup equipped with an admissible order ≤. For a non-empty subset X ⊂ S, let X ↓ := {s ∈ S : there is x ∈ X such that s ≤ x} denote the lower set or ideal generated by X. Let X ↑ := {s ∈ S : there is x ∈ X such that x ≤ s} denote the upper set or coideal generated by X. Lemma 8. For any subsemigroup T ⊂ S, both T ↓ and T ↑ are subsemigroups of S. Proof. We prove the claim for T ↓, for T ↑ the arguments are similar. Let a, b ∈ T ↓. Then there exist s, t ∈ T such that a ≤ s and b ≤ t. As ≤ is admissible, we have ab ≤ sb ≤ st. Now, st ∈ T as T is a subsemigroup, and thus ab ∈ T ↓. (cid:3) 10 PAUL MARTIN AND VOLODYMYR MAZORCHUK 4.3. The symmetric inverse monoid. For n ∈ N, we denote by ISn the symmet- ric inverse monoid on n, see [GaMa]. It consists of all bijections between subsets of n. Alternatively, we can identify ISn with S↓ n inside the ordered monoid Bn. The monoid ISn is an inverse monoid. The natural partial order on the inverse monoid ISn coincides with the inclusion order inherited from Bn. The group Sn is the group of invertible elements in ISn. 4.4. The dual symmetric inverse monoid. For n ∈ N, we denote by I ∗ n the dual symmetric inverse monoid on n, see [FL]. It consists of all bijections between quotients of n. Alternatively, we can identity I ∗ n with PPn in the obvious way. The monoid I ∗ n is an inverse monoid. The natural partial order on the inverse monoid I ∗ n coincides with the order inherited from Pn. The group Sn is the group of invertible elements in I ∗ n. We also consider the maximal factorizable submonoid F ∗ n, that is the sub- monoid of all elements which can be written in the form σε, where σ ∈ Sn and ε is an idempotent in I ∗ n are exactly the identity transformations of quotient sets of n, equivalently, idempotents in I ∗ n coincide with the principal coideal in Pn generated by the identity element. n. Idempotents in I ∗ n of I ∗ Lemma 9. The monoid F ∗ n coincides with the subsemigroup S↑ n of PPn. n contains both Sn and all idempotents of I ∗ Proof. As S↑ other hand, let ρ ∈ S↑ Hence ρσ−1 is an idempotent and ρ = (ρσ−1)σ ∈ F ∗ n . n. On the n. Then σ ≤ ρ for some σ ∈ Sn. This means that idn ≤ ρσ−1. (cid:3) n, we have F ∗ n ⊂ S↑ 4.5. Fiat 2-categories. Following [MM1], we say that a finitary 2-category C is fiat provided that there exists a weak anti-involution ⋆ : C → C which reverses the direction of both 1-morphisms and 2-morphisms, such that, for any objects i, j ∈ C and any 1-morphism F ∈ C (i, j), there are 2-morphisms α : 1i → F⋆F and β : FF⋆ → 1j such that (β ◦0 idF) ◦1 (idF ◦0 α) = idF and (idF⋆ ◦0 β) ◦1 (α ◦0 idF⋆ ) = idF⋆ . This means that, in any functorial action of C, the 1-morphisms F and F⋆ are represented by adjoint functors. The above property is usually called existence of adjunction 2-morphisms. 4.6. Comparison of fiatness. Theorem 10. Let n ∈ N. (i) Both 2-categories, kA n and kB n, are fiat. (ii) Both 2-categories, kS ISn and kS F ∗ n , are finitary but not fiat. Proof. The endomorphism algebra of any 1-morphism in kS ISn is k, by definition. Therefore kS ISn is finitary by construction. The category kS ISn cannot be fiat as it contains non-invertible indecomposable 1-morphisms but it does not contain any non-zero 2-morphisms from the identity 1-morphism to any non-invertible in- decomposable 1-morphism. Therefore adjunction 2-morphisms for non-invertible indecomposable 1-morphisms cannot exist. The same argument also applies to kS F ∗ n , proving claim (ii). By construction, the 2-category kA n satisfies conditions (I), (II) and (III) from the definition of a finitary 2-category. Therefore the 2-category kA n is a finitary 2-category by Lemma 7. Let us now check existence of adjunction 2-morphisms. FIAT CATEGORIFICATION OF ISn AND F ∗ n 11 Recall that an adjoint to a direct sum of functors is a direct sum of adjoints to components. Therefore, as kA n is obtained from (A n)k by splitting idempotents in 2-endomorphism rings, it is enough to check that adjunction 2-morphisms exist in (A n)k. Any 1-morphism in (A n)k is, by construction, a direct sum of σ ∈ Sn. Therefore it is enough to check that adjunction 2-morphisms exist in A n. In the latter category, each 1-morphism σ ∈ Sn is invertible and hence both left and right adjoint to σ−1. This implies existence of adjunction 2-morphisms in A n. The above shows that the 2-category kA n is fiat. Similarly one shows that the 2-category kB n is fiat. This completes the proof. (cid:3) 5. Decategorification 5.1. Decategorification via Grothendieck group. Let C be a finitary 2-category. A Grothendieck decategorification [C ] of C is a category defined as follows: • [C] has the same objects as C . • For i, j ∈ C, the set [C ](i, j) coincides with the split Grothendieck group [C(i, j)]⊕ of the additive category C (i, j). • The identity morphism in [C ](i, i) is the class of 1i. • Composition in [C] is induced from composition of 1-morphisms in C . We refer to [Maz1, Lecture 1] for more details. For a finitary 2-category C , the above allows us to define the decategorification of C as the Z-algebra AC := Mi,j∈C [C](i, j) with the induced composition. The algebra AC is positively based in the sense of [KiMa2] with respect to the basis corresponding to indecomposable 1-morphisms in C. 5.2. Decategorifications of kA n and kS ISn . Theorem 11. We have AkA n ∼= AkS ISn ∼= Z[ISn]. Proof. Indecomposable 1-morphisms in kS ISn correspond exactly to elements of ∼= Z[ISn] where an indecomposable ISn, by construction. This implies that AkS ISn 1-morphism σ on the left hand side is mapped to itself on the right hand side. So, we only need to prove that AkA n ∼= Z[ISn]. For σ ∈ ISn, set (5.1) σ := Xρ⊂σ (−1)σ\ρρ ∈ Z[ISn]. Then {σ : σ ∈ ISn} is a basis in Z[ISn] which we call the Mobius basis, see, for example, [Ste, Theorem 4.4]. The endomorphism monoid EndA n (idn) is, by construction, canonically isomorphic to the Boolean 2n of n with both ◦0 and ◦1 being equal to the operation on 2n of taking the intersection. We identify elements in EndA n (idn) and in 2n in the 12 PAUL MARTIN AND VOLODYMYR MAZORCHUK obvious way. With this identification, in the construction of kA n, we can take, for X ⊂ n, (5.2) ε(X) i = XY ⊆X (−1)X−Y Y. For σ ∈ Sn and X, Y ⊂ n, consider the element (5.3) ε(Y ) i ◦0 σ ◦0 ε(X) i ∈ EndkA n(σ) and write it as a linear combination of subrelations of σ (this is the standard basis in EndkA n (σ)). A subrelation ρ ⊂ σ may appear in this linear combination with a non-zero coefficient only if ρ consist of pairs of the form (y, x), where x ∈ X and y ∈ Y . Assume that σ(X) = Y . Then the relation ρσ = [x∈X {(σ(x), x)}, clearly, appears in the linear combination above with coefficient one. Moreover, the idempotent properties of ε(X) imply that the element in (5.3) is exactly ρσ. and ε(Y ) i i Assume that σ(X) 6= Y . Then the inclusion-exclusion formula implies that any subrelation of σ appears in the linear combination above with coefficient zero. This means that the 1-morphism σ ∈ kA n(i(Y ), i(X)) is zero if and only if σ(X) 6= Y . If X = Y and σ, π ∈ Sn are such that σ(x) = π(x) ∈ Y , for all x ∈ X, then ρσ = ρπ ∈ HomkA n (σ, π) ∩ HomkA n (π, σ) gives rise to an isomorphism between σ and π in kA n(i(Y ), i(X)). If σ(x) 6= π(x), for some x ∈ X, then any morphism in HomkA n (σ, π) is a linear combination of relations which are properly contained in both ρσ and ρπ. Therefore σ and π are not isomorphic in kA n. Consequently, isomorphism classes of indecomposable 1-morphisms in the category kA n(i(Y ), i(X)) correspond precisely to elements in ISn with domain X and image Y . Composition of these indecomposable 1-morphism is inherited from Sn. By comparing formulae (5.1) and (5.2), we see that composition of 1-morphisms in kA n corresponds to multiplication of the Mobius basis elements in Z[ISn]. This completes the proof of the theorem. (cid:3) Theorem 11 allows us to consider kA n and S ISn as two different categorifications of ISn. The advantage of kA n is that this 2-category is fiat. The construction we use in our proof of Theorem 11 resembles the partialization construction from [KuMa1]. 5.3. Decategorifications of kB n and kS F ∗ n . ∼= Z[F ∗ n ]. Theorem 12. We have AkB n ∼= AkSF ∗ n Proof. Using the Mobius function for the poset of all quotients of n with respect to ≤ (see, for example, [Rot, Example 1]), Theorem 12 is proved mutatis mutandis Theorem 11. (cid:3) Theorem 12 allows us to consider kB n and S F ∗ of F ∗ n . The advantage of kB n is that this 2-category is fiat. n as two different categorifications FIAT CATEGORIFICATION OF ISn AND F ∗ n 13 6. Examples for n = 2 6.1. Example of F ∗ as follows: 2 . The monoid F ∗ 2 consists of three elements which we write ǫ := (cid:18) 1 1 2 2 (cid:19) , σ := (cid:18) 1 2 2 1 (cid:19) , τ := (cid:18) {1, 2} {1, 2} (cid:19) . These are identified with the following partitions of {1, 2, 1′, 2′}: ǫ ↔ (cid:8){1, 1′}, {2, 2′}(cid:9), σ ↔ (cid:8){1, 2′}, {2, 1′}(cid:9), The symmetric group S2 consists of ǫ and σ. τ ↔ (cid:8){1, 2, 1′, 2′}(cid:9). Here is the table showing all 2-morphisms in B 2 from x to y: y \ x ǫ σ τ ǫ ǫ, τ τ τ σ τ σ, τ τ τ τ τ τ The 2-endomorphism algebra of both ǫ and σ in (B 2)k is isomorphic to k ⊕ k where the primitive idempotents are τ and ǫ − τ , in the case of ǫ, and τ and σ − τ , in the case of σ. The 2-category kB 2 has three isomorphism classes of indecomposable 1-morphisms, namely τ , ǫ − τ and σ − τ . The 2-category kB 2 has two objects, iτ and iǫ−τ . The indecomposable 1-morphisms in kB 2 give indecomposable 1-morphisms in kB 2 from x to y as follows: y \ x iǫ−τ iτ iǫ−τ iτ ǫ − τ, σ − τ ∅ τ ∅ 6.2. Example of IS2. We write elements of IS2 as follows: ǫ := (cid:18) 1 2 1 ∅ (cid:19) , 1 2 (cid:19) , β := (cid:18) 1 σ := (cid:18) 1 2 2 ∅ (cid:19) , 2 1 (cid:19) , γ := (cid:18) 1 τ := (cid:18) 1 ∅ 1 (cid:19) , ∅ ∅ (cid:19) , δ := (cid:18) 1 2 2 2 2 α := (cid:18) 1 2 ∅ 2 (cid:19) . The symmetric group S2 consists of ǫ and σ. Here is the table showing all 2-morphisms in A 2 from x to y: y \ x ǫ ǫ σ τ α β γ δ ǫ, α, δ, τ τ τ α, τ τ τ α, τ σ τ σ, β, γ, τ τ τ β, τ β, τ τ τ α τ τ τ α, τ τ τ τ α, τ τ τ τ τ τ τ β τ γ τ β, τ γ, τ τ τ β, τ τ τ τ τ τ γ, τ τ δ δ, τ τ τ τ τ τ δ, τ The 2-endomorphism algebra of ǫ in (A 2)k is isomorphic to k ⊕ k ⊕ k ⊕ k where the primitive idempotents are τ , α − τ , δ − τ and ǫ − α − δ + τ . Similarly one can describe the 2-endomorphism algebra of σ in (A 2)k. The 2-endomorphism algebra of α in (A 2)k is isomorphic to k ⊕ k where the primitive idempotents are τ and α−τ . Similarly one can describe the 2-endomorphism algebras of β, γ and δ. 14 PAUL MARTIN AND VOLODYMYR MAZORCHUK The 2-category kA 2 has seven isomorphism classes of indecomposable 1-morphisms, namely τ, α − τ, β − τ, γ − τ, δ − τ, ǫ − α − δ + τ, σ − β − γ + τ. The 2-category kA 2 has four objects, iτ , iα−τ , iδ−τ and iǫ−α−δ+τ . The indecom- posable 1-morphisms in kA 2 give indecomposable 1-morphisms in kA 2 from x to y as follows: y \ x iǫ−α−δ+τ iα−τ iδ−τ iτ iǫ−α−δ+τ ǫ − α − δ + τ, σ − β − γ + τ ∅ iα−τ iδ−τ iτ ∅ ∅ ∅ α − τ β − τ ∅ This table can be compared with [MS, Figure 1]. ∅ ∅ γ − τ ∅ δ − τ ∅ τ ∅ References [BBFW] [BFK] [CM] [CR] [EH] [El] [FL] [FB] [Fo] [GaMa] for finitary 2- J. Baez, A. Baratin, L. Freidel, D. Wise. Infinite-dimensional representations of 2- groups. Mem. Amer. Math. Soc. 219 (2012), no. 1032, vi+120 pp. J. Bernstein, I. Frenkel, M. Khovanov. A categorification of the Temperley-Lieb algebra and Schur quotients of U (sl2) via projective and Zuckerman functors. Selecta Math. (N.S.) 5 (1999), no. 2, 199–241. A. Chan, V. Mazorchuk. Diagrams and discrete extensions representations. Preprint arXiv:1601.00080 J. Chuang, R. Rouquier. Derived equivalences for symmetric groups and sl2-categori- fication. Ann. of Math. (2) 167 (2008), no. 1, 245–298. B. Eckmann, P. Hilton. Group-like structures in general categories. I. Multiplications and comultiplications. Math. Ann. 144/145 (1962), 227–255. J. Elgueta. Representation theory of 2-groups on Kapranov and Voevodsky's 2-vector spaces. Adv. Math. 213 (2007), 53-92. D. FitzGerald, J. Leech. Dual symmetric inverse monoids and representation theory. J. Austral. Math. Soc. Ser. A 64 (1998), no. 3, 345–367. M. Forrester-Barker. Representations of crossed modules and Cat1-groups. Ph.D. The- sis, University of Wales, Bangor, 2003. L. Forsberg. Multisemigroups with multiplicities and complete ordered semi-rings. Preprint arXiv:1510.01478 O. Ganyushkin, V. Mazorchuk. Classical finite transformation semigroups. An intro- duction. Algebra and Applications, 9. Springer-Verlag London, Ltd., London, 2009. [GrMa1] A.-L. Grensing, V. Mazorchuk. Categorification of the Catalan monoid. Semigroup Forum 89 (2014), no. 1, 155–168. [GrMa2] A.-L. Grensing, V. Mazorchuk. Finitary 2-categories associated with dual projection [Jo] [Ka] [Kh] functors. Preprint arXiv:1501.00095. V. Jones. The Potts model and the symmetric group. In: Subfactors: Proceedings of the Taniguchi Symposium on Operator Algebras (Kyuzeso, 1993), River Edge, NJ, World Sci. Publishing, 1994, pp. 259–267. K. Kapranov, V. Voevodsky. 2-categories and Zamolodchikov tetrahedra equations. Algebraic groups and their generalizations: quantum and infinite-dimensional methods (University Park, PA, 1991), 177–259, Proc. Sympos. Pure Math., 56, Part 2, Amer. Math. Soc., Providence, RI, 1994. M. Khovanov. A categorification of the Jones polynomial. Duke Math. J. 101 (2000), no. 3, 359–426. [KMMZ] T. Kildetoft, M. Mackaay, V. Mazorchuk, J. Zimmermann. Simple transitive 2- [KiMa1] [KiMa2] [Ko] representations of small quotients of Soergel bimodules. Preprint arXiv:1605.01373 T. Kildetoft, V. Mazorchuk. Parabolic projective functors in type A. Preprint arXiv:1506.07008 T. Kildetoft, V. Mazorchuk. Special modules over positively based algebras. Preprint arXiv:1601.06975 J. Kock. Note on commutativity in double semigroups and two-fold monoidal cate- gories. J. Homotopy Relat. Struct. 2 (2007), no. 2, 217–228. FIAT CATEGORIFICATION OF ISn AND F ∗ n 15 [KuMa1] G. Kudryavtseva, V. Mazorchuk. Partialization of categories and inverse braid- permutation monoids. Internat. J. Algebra Comput. 18 (2008), no. 6, 989–1017. [KuMa2] G. Kudryavtseva, V. Mazorchuk. On multisemigroups. Port. Math. 72 (2015), no. 1, 47–80. T. Leinster. Basic Bicategories. Preprint arXiv:math/9810017 [Le] [MaMa] M. Mackaay, V. Mazorchuk. Simple transitive 2-representations for some 2- [Mac] [Mar1] [Mar2] [Maz1] [Maz2] [MM1] [MM2] [MM3] [MM4] [MM5] [MM6] [MS] [MZ] [Pf] [Rot] [Ro1] [Ro2] [Ste] [Str] [Xa] [Zh1] [Zh2] [Zi] subcategories of Soergel bimodules. Preprint arXiv:1602.04314 S. Mac Lane. Categories for the working mathematician. Second edition. Graduate Texts in Mathematics 5. Springer-Verlag, New York, 1998. P. P. Martin. Potts models and related problems in statistical mechanics. World Sci- entific, Singapore, 1991. P. P. Martin. Temperley-Lieb algebras for non-planar statistical mechanics - the partition algebra construction. Journal of Knot Theory and its Ramifications 3 (1994), no. 1, 51–82. V. Mazorchuk. Endomorphisms of Bn, PBn and Cn. Comm. Algebra 30 (2002), no. 7, 3489–3513. V. Mazorchuk. Lectures on algebraic categorification. QGM Master Class Series. Eu- ropean Mathematical Society (EMS), Zurich, 2012. x+119 pp. V. Mazorchuk, V. Miemietz. Cell 2-representations of finitary 2-categories. Compositio Math. 147 (2011), 1519–1545. V. Mazorchuk, V. Miemietz. Additive versus abelian 2-representations of fiat 2-cate- gories. Moscow Math. J. 14 (2014), no. 3, 595–615. V. Mazorchuk, V. Miemietz. Endomorphisms of cell 2-representations. Preprint arXiv:1207.6236. To appear in IMRN. V. Mazorchuk, V. Miemietz. Morita theory for finitary 2-categories. Quantum Topol. 7 (2016), no. 1, 1–28. V. Mazorchuk, V. Miemietz. Transitive 2-representations of finitary 2-categories. Preprint arXiv:1404.7589. To appear in Trans. Amer. Math. Soc. V. Mazorchuk, V. Miemietz. Isotypic faithful 2-representations of J -simple fiat 2- categories. Math. Z. 282 (2016), no. 1-2, 411–434. V. Mazorchuk, C. Stroppel. G(l, k, d)-modules via groupoids. J. Algebraic Combin. 43 (2016), no. 1, 11–32. V. Mazorchuk, X. Zhang. Simple transitive 2-representations for two non-fiat 2- categories of projective functors. Preprint arXiv:1601.00097 H. Pfeiffer. 2-Groups, trialgebras and their Hopf categories of representations. Adv. Math. 212 (2007), 62-108. G.-C. Rota. On the foundations of combinatorial theory. I. Theory of Mobius functions. Z. Wahrscheinlichkeitstheorie und Verw. Gebiete 2 (1964), 340–368. R. Rouquier. 2-Kac-Moody algebras. Preprint arXiv:0812.5023. R. Rouquier. Quiver Hecke algebras and 2-Lie algebras. Algebra Colloquium 19 (2012), 359–410. B. Steinberg. Mobius functions and semigroup representation theory. J. Combin. The- ory Ser. A 113 (2006), no. 5, 866–881. C. Stroppel. Categorification of the Temperley-Lieb category, tangles, and cobordisms via projective functors. Duke Math. J. 126 (2005), no. 3, 547–596. Q. Xantcha. Gabriel 2-Quivers for Finitary 2-Categories. Preprint arXiv:1310.1586, to appear in J. London Math. Soc. X. Zhang. Duflo involutions for 2-categories associated to tree quivers. Preprint arXiv:1501.03052. To appear in J. Algebra Appl. X. Zhang. Simple transitive 2-representations and Drinfeld center for some finitary 2-categories. Preprint arXiv:1506.02402. J. Zimmermann. Simple transitive 2-representations of Soergel bimodules in type B2. Preprint arXiv:1509.01441. P. M.: Department of Pure Mathematics, University of Leeds, Leeds, LS2 9JT, UK, e-mail: [email protected] V. M: Department of Mathematics, Uppsala University, Box. 480, SE-75106, Upp- sala, SWEDEN, email: [email protected]
1807.10070	2	1807	2018-12-03T21:27:14	Construction of a quotient ring of $\mathbb{Z}_2\mathcal{F}$ in which a binomial $1 + w$ is invertible using small cancellation methods	[ "math.RA" ]	We apply small cancellation methods originating from group theory to investigate the structure of a quotient ring $\mathbb{Z}_2\mathcal{F} / \mathcal{I}$, where $\mathbb{Z}_2\mathcal{F}$ is the group algebra of the free group $\mathcal{F}$ over the field $\mathbb{Z}_2$, and the ideal $\mathcal{I}$ is generated by a single trinomial $1 + v + vw$, where $v$ is a complicated word depending on $w$. In $\mathbb{Z}_2\mathcal{F} / \mathcal{I}$ we have $(1 + w)^{-1} = v$, so $1 + w$ becomes invertible. We construct an explicit linear basis of $\mathbb{Z}_2\mathcal{F} / \mathcal{I}$ (thus showing that $\mathbb{Z}_2\mathcal{F} / \mathcal{I}\neq 0$). This is the first step in constructing rings with exotic properties.	math.RA	math	Construction of a Quotient Ring of Z2F in which a Binomial 1 + w is Invertible Using Small Cancellation Methods A. Atkarskaya Department of Mathematics, Bar-Ilan University, 5290002 Ramat Gan, Israel [email protected] Department of Mathematics, Bar-Ilan University, 5290002 Ramat Gan, Israel A. Kanel-Belov ∗ [email protected] E. Plotkin Department of Mathematics, Bar-Ilan University, 5290002 Ramat Gan, Israel [email protected] E. Rips Department of Mathematics, The Hebrew University of Jerusalem, Givat Ram, 9190401 Jerusalem, Israel [email protected] Dedicated to Professor Boris I. Plotkin with love and respect on the occa- sion of his 90th Anniversary. ∗The paper was supported by Russian Science foundation grant №17-11-01337. Abstract 1 We apply small cancellation methods originating from group the- ory to investigate the structure of a quotient ring Z2F/I, where Z2F is the group algebra of the free group F over the field Z2, and the ideal I is generated by a single trinomial 1 + v + vw, where v is a complicated word depending on w. In Z2F/I we have (1 + w)−1 = v, so 1 + w becomes invertible. We construct an explicit linear basis of Z2F/I (thus showing that Z2F/I 6= 0). This is the first step in constructing rings with exotic properties. 1 Introduction This paper describes the first step in a construction of a skew field with a finitely generated multiplicative group. We will construct this skew field as a quotient ring of a group algebra of a finitely generated free group F . The full construction of such a skew field would involve an iterative procedure, in which every step is similar to the one described in this paper, in a more complicated situation. The resulting injective limit will be either 0 or a skew field. Our main objective is to show that the resulting skew field is non trivial, and, moreover, to obtain a skew field of infinite dimension over its center in which every non zero element is equal to a monomial, that is, an element of the free group F . Hence, when F is finitely generated, the multiplicative group of this skew field is finitely generated. In order to show that the resulting skew field is non trivial, we need to develop its structure theory. In this paper we consider the following problem. Let Z2F be the group algebra of the free group F over the field Z2, and let 1 + w be a binomial. We would like to find a quotient ring of Z2F in which the image of 1 + w is invertible, without losing control of the quotient ring relations. Our way to deal with this problem is influenced by an analogue with small cancellation in groups. Namely, we equate (1 + w)−1 to a complicated word v = xαyxα+1y · · · xβ−1y with w ≪ α ≪ β, where w is a word length of w in F . The monomial v exhibits small cancellation properties because every subword of v that contains at least two letters y appears in v only once ([7]). In the case of groups we consider semicanonical monomials, namely, monomials that do not contain parts of the relations (or of their cyclic conju- 2 gates and inverses) that are too big. It turns out that semicanonical monomi- als representing the same element of the group are connected by a so-called one-layer diagram. The transition between semicanonical monomials repre- senting the same element is done by substituting subwords of the relations by the inverses of their complements repeatedly (such transitions are called turns). Here we imitate this process with inevitable complications. Since our relations are not binomial but polynomial, we substitute a subword by the sum of the rest of the monomials of the relation. We call such transitions multi-turns. The simple-minded picture goes like this a(1) h1 a(2) h2 . . . a(k) hk U where U is a monomial, and we have polynomial relations a(i) ji = 0, i = 1, . . . , k, hi ∈ {1, . . . , li}. liXji=1 Performing a number of multi-turns, we obtain sums of monomials of the form a(1) j1 a(2) j2 . . . a(k) jk We have l1 · . . . · lk monomials with linear dependencies induced by the multi- turns. In our (simple-minded) case we obtain a linear space of dimension (l1 − 1) · . . . · (lk − 1). The actual situation is more complicated, due to a number of factors. Let us list some of them: • Some of the words a(i) ji might be short, causing certain degeneracies. • There can be highly non-trivial interactions between the neighbour oc- currences, so that the order of performing the multi-turns might matter. This explains why pursuing this program turns out not entirely simple. We have to introduce a number of elaborate concepts to deal with the monomials and the subwords of the relations appearing in them (for this, see especially Section 3). 3 To control the degenerations, we introduce a (decreasing) filtration and study its properties (Section 4). Then we need to study the interaction between the linear dependencies and the filtration. Because of the properties like transversality and non-degeneracy, this interaction has nice properties (Section 4.2, Section 5). Our final result (Theorem 5.1) supplies an explicit linear basis for the quotient ring. In particular, the quotient ring does not collapse to 0. Then we introduce a linear generating set of our quotient ring with par- ticularly nice properties. For this generating set, the multiplication can be expressed as a sum of monomials forming thin triangles with the factors (Section 6, Theorem 6.1). Small cancellation groups are hyperbolic, so their multiplication is expressed by thin triangles. While we do not possess the concept of a hyperbolic ring (even for algebras over a field), the quotient ring we construct does display some features of what might be expected for a "hyperbolic ring". 2 Basic definitions Consider the group ring Z2F , where F is a free group with at least 4 free generators. We will call elements of F words or monomials. We deal only with reduced monomials unless the converse is explicitly stated. For our purposes we can use any field k, but we choose the field Z2 to simplify our calculations. Let us fix w ∈ F , an arbitrary cyclically reduced primitive (not a proper power) word from F . Fix positive integers α and β such that w ≪ α ≪ β, where w is a word length of w in F . Our aim is inverting of the binomial 1 + w. Consider the word v = xαyxα+1y · · · xβ−1y (1) such that w does not start or end with the letters x, y, x−1, y−1. We are going to study a structure of the quotient ring Z2F /I, where I = h1 + v + vwi. (2) Clearly, in the quotient ring we have v(1 + w) = 1. Multiplying it by v on the right side, we obtain v(1 + w)v = v. Since v is a word, it is invertible in Z2F , so, it is also invertible in Z2F /I or the quotient ring is trivial. Multiplying the last equation by v−1, we obtain (1 + w)v = 1. Thus, v−1 = 1 + w in the ring Z2F /I. 4 Definition 2.1. Suppose M(v, w) is a non-commutative monomial over the words v, w. We call arbitrary subwords of monomials M(v, w) (v, w)-generalized fractional powers (or simply generalized fractional powers). Let v1 be an arbitrary subword of v (or of v−1). We define an additive (length) measure (aka Λ-measure) on subwords of v (or v−1) in the following way Λ(v1) = my(v1) β − α , (3) where my(v1) is the number of occurrences of the letter y (or of the letter y−1) in v1. Obviously, Λ(v) = 1 (Λ(v−1) = 1). For any subword w1 of w we put Λ(w1) = 0. From now on, let us call Λ-measure of a generalized fractional power the sum of Λ-measure of its parts. Let U be a word and U1 be its subword. We call the triple that consists of U, U1 and the position of U1 in U an occurrence of U1 in U. An arbitrary word from F may contain occurrences of generalized frac- tional powers. While the notion of a maximal occurrence seems to be intu- itively simple, it turns out that the notion we actually need has considerable subtleties. So, we define a maximal occurrence in the following way. Let a be a generalized fractional power, that is, a is an occurrence in a monomial M(v, w) = MLaMR. Assume that ML = M1(v, w)M ′ L, where M1(v, w) is a monomial in v, w, and M ′ L has no initial subword equal to a monomial in v, w. In the same way, assume that MR = M ′ RM2(v, w), where M2(v, w) is a monomial in v, w, and M ′ R has no final subword equal to a monomial in v, w. Then we consider a as an occurrence in the monomial M0(v, w) = M1(v, w)−1M(v, w)M −1 2 (v, w) = M ′ LaM ′ R. Consider a set of monomials A = {M(v, w) M(v, w) = Mi(v, w)M0(v, w)Mj(v, w), i, j ∈ N}, where Mi(v, w) and Mj(v, w) are arbitrary monomials in v, w. Then the occurrence of a in M0(v, w) can be considered as the occurrence of a in M(v, w) ∈ A with the same position in M0(v, w). Suppose a is an occurrence in a word U ∈ F , a letter a1 prolongs a in U from the left side, a letter a2 prolongs a in U from the right side. If a is an initial subword of U, a1 is absent; if a is a final subword of U, a2 is absent. If 5 for any M(v, w) ∈ A the occurrence a can be prolonged neither by a1 on the left nor by a2 on the right in M(v, w), then we call a a maximal occurrence of a generalized fractional power in U. Let U = ULaUR. The above definition in fact means the following. The occurrence a is a maximal occurrence of a generalized fractional power in U if and only if it satisfies one of the conditions: (L1) UL is the empty word; (L2) UL = U ′ La1, M ′ L = M ′′ La′ 1, a1 6= a′ 1; (L3) UL = U ′ La1, M ′ L is the empty word, the last letter in each Mi(v, w) is different from a1; and one of the conditions (R1) UR is the empty word; (R2) UR = a2U ′ R, M ′ R = a′ 2M ′′ R, a2 6= a′ 2; (R3) UR = a2U ′ R, M ′ R is the empty word, the first letter in each Mj(v, w) is different from a2. Definition 2.2. Let τ be some small value, U be an arbitrary word from F . We call the set of maximal occurrences of generalized fractional powers in U of Λ-measure greater or equal to a given threshold τ the chart of the word U, and we call the corresponding occurrences of generalized fractional powers members of the chart. β−α. Further we will assume that τ > 10ε. Let us put ε = 1 Any subword of v containing at least two letters y appears in a unique way in v. Moreover, a subword of vl containing at least two letters y appears in vl uniquely modulo the period v. Therefore, if the Λ-measure of some maximal occurrence of a generalized fractional power is greater or equal to 2ε, it can not be properly contained in another occurrence of a generalized fractional power (because otherwise it would not be maximal). In particular, given the chart of some word, one member of the chart can not be properly contained in another. However, members of the chart may have overlaps. Consider an overlap of two members of the chart. If the overlap is a common part of two subwords of vl, then according to the last remark, the Λ-measure of the overlap is not greater than ε. If the overlap contains a 6 subword of wk, it may have a more complicated form. Since α ≫ w, w does not contain subwords yδ1xγyδ2, where γ > α. Hence, since the Λ-measure of subwords of w is equal to zero, the Λ-measure of the overlap still is not greater than ε for both members of the chart. In particular, since τ > 10ε, we never have a situation when one member of the chart is fully covered by others. Definition 2.3. Consider oriented graphs with edges marked by generators of the group F . Take such a graph of the form O wk w w−1 w−k v (4) Here to each integer power of w corresponds a separate arc. That is, there are infinitely many arcs that correspond to different wk. We call this graph a v-diagram. Assume that we have an oriented path in the graph (4). When we go along this path, we can write down the mark of an edge if we pass the edge in the positive direction, and we can write down the inverse to the mark of an edge if we pass the edge in the negative direction. As a result, we obtain a generalized fractional power (possibly after cancellations if there are any). It is easy to see that to each monomial over v, w there corresponds a path in the graph (4) with the initial and the final vertex O. By the definition, every generalized fractional power is a subword in a monomial over v, w. Hence, given a generalized fractional power M, one can specify two points I and F on the graph (4) such that M corresponds to the unique path starting at I and ending at F . For the sake of uniqueness, we assume that we always 7 choose an arc with maximal absolute value of degree of w along the path. So, one can see that M corresponds to the path in the graph of one of the types (5) -- (7). Note that if a generalized fractional power is of Λ-measure strictly greater than ε, then the positions of the initial and the final points that belong to the v-arc are uniquely determined. In what follows we use for v the notation v = vivmvf , where vi is some initial part, vm is some middle part and vf is some final part of v (any part is allowed to be empty). Analogously to the notion of v-diagram, in this paper we represent all monomials (not only generalized fractional powers) as segments that consist of oriented edges marked by generators of the group F ; such segments we always read from left to right. So, for v we have = v vi vm vf Similarly, for w we use the notation w = wiwmwf , where wi is some initial part, wm is some middle part and wf is some final part of w (any part is allowed to be empty). Let us enumerate all possible positions of points I and F on the v- diagram. Thereby, we enumerate all possible types of generalized fractional powers. In every case, we explicitly write the corresponding forms of gen- eralized fractional powers. As above, M(v, w) is a monomial over v and w. First, both points I and F may lie on the v-arc. Then they divide the v-arc into three parts and, according to the notations introduced above, we denote 8 them vi, vm and vf and obtain the pictures wk wk O vi w−k I vf F O vi w−k F vf I vm vmvf M(v, w)v−1 f , vmvf M(v, w)vivm, v−1 i M(v, w)vivm, v−1 i M(v, w)v−1 f ; vm vf M(v, w)vi, vf M(v, w)v−1 v−1 m v−1 m v−1 v−1 f v−1 m , i M(v, w)vi, i M(v, w)v−1 f v−1 m . (5) The next configuration is when the point I lies on the v-arc and the point F lies on a w-arc or vice versa. Then, according to the above notations, we 9 put v = vivf , w = wiwf and obtain the pictures wk I O wk F O vi w−k vi w−k vf vf F I wf M(v, w)vi, wf M(v, w)v−1 f , w−1 i M(v, w)vi, i M(v, w)v−1 w−1 f ; vf M(v, w)wi, v−1 i M(v, w)wi, vf M(v, w)w−1 f , v−1 i M(v, w)w−1 f . (6) In picture (6) points I and F lie on a positive w-arc, but they may lie on a negative w-arc within this type of paths as well. The last configuration is when both points I and F lie on a w-arc. Then, according to the above notations, we put w = wiwmwf and obtain the pic- 10 tures wk I F O wk F I O w−k v w−k v wf M(v, w)wi, wf M(v, w)w−1 w−1 w−1 f w−1 m , i M(v, w)wi, i M(v, w)w−1 m w−1 m w−1 f w−1 m ; wmwf M(v, w)w−1 f , wmwf M(v, w)wiwm, w−1 w−1 i M(v, w)wiwm, i M(v, w)w−1 f . (7) In picture (7) points I and F lie on a positive w-arc, but each of them may lie on a negative w-arc within this type of paths as well. Also the point I and the point F may lie on different w-arcs. Given a set of generalized fractional powers Mj, j = 1, . . . , k, that cor- respond to the paths with the same initial and the same final point in the j=1 Mj corresponds to the collection diagram of type (5) -- (7), their sumPk Let Z2(w) be the field of rational functions in one variable w over Z2. Consider a non-commutative Laurent polynomial P (x1, x2) over Z2 such that of these paths. P ((1 + w)−1, w) = 0 as an element of Z2(w). (8) Notice that if P (x1, x2) satisfies condition (8), then P (v, w) ∈ h1 + v + vwi = I. Example 2.1. Let us give an example of such polynomials. The first example comes from the obvious equality in Z2(w) w · 1 1 + w 1 1 + w · w. = 11 From this equality it follows that the polynomial satisfies condition (8) and P (x1, x2) = x1x2 + x2x1 P (v, w) = vw + wv ∈ I. Recalling the reduction of rational functions of the form w±k 1+w in Z2(w) to elementary fractions, we obtain the following equalities in Z2(w) for k > 0: wk 1 + w = wk + wk−1 + wk−1 + . . . + w + w + 1 + 1 1 + w = wk−1 + wk−2 + . . . + w + 1 + 1 1 + w and w−k 1 + w = w−k + w−k+1 + w−k+1 + . . . + w−1 + w−1 + 1 + 1 1 + w = w−k + w−k+1 + . . . + w−1 + 1 1 + w . Hence, the polynomials P1(x1, x2) = x1xk P2(x1, x2) = x1x−k 2 + xk−1 2 + x−k 2 + xk−2 2 + x−k+1 2 + . . . + x2 + 1 + x1, 2 + . . . + x−1 2 + x1 satisfy condition (8) and P1(v, w) = vwk + wk−1 + wk−2 + . . . + w + 1 + v ∈ I, P2(v, w) = vw−k + w−k + w−k+1 + . . . + w−1 + v ∈ I. According to types (5) -- (7) of v-diagrams that correspond to possible forms of generalized fractional powers, we consider the list of expressions (9) -- (14), where P (x1, x2) is a non-commutative Laurent polynomial such that P ((1 + w)−1, w) = 0 as an element of Z2(w). We allow possibility of cancellations in monomials in (9) -- (14). First, consider expressions vf P (v, w)vi, vf P (v, w)v−1 v−1 m v−1 v−1 m v−1 f v−1 m , i P (v, w)vi, i P (v, w)v−1 f v−1 m . 12 (9) To each expression of (9), there corresponds a collection of paths in the graph (5). Similarly, to each expression vmvf P (v, w)v−1 f , vmvf P (v, w)vivm, v−1 i P (v, w)vivm, i P (v, w)v−1 v−1 f , (10) there corresponds a collection of paths in the graph (5). To each expression wf P (v, w)vi, wf P (v, w)v−1 f , w−1 i P (v, w)vi, w−1 i P (v, w)v−1 f , (11) there corresponds a collection of paths in the graph (6). Similarly, to each expression vf P (v, w)wi, v−1 i P (v, w)wi, vf P (v, w)w−1 f , i P (v, w)w−1 v−1 f , (12) there corresponds a collection of paths in the graph (6). Finally, consider the expressions wf P (v, w)wi, wf P (v, w)w−1 w−1 w−1 f w−1 m , i P (v, w)wi, i P (v, w)w−1 m w−1 m w−1 f w−1 m . (13) To each expression of (13) there corresponds a collection of paths in the graph (7). Similarly, to each expression wmwf P (v, w)w−1 f , wmwf P (v, w)wiwm, w−1 w−1 i P (v, w)wiwm, i P (v, w)w−1 f , 13 (14) there corresponds a collection of paths in the graph (7). Since P (v, w) ∈ I, the expressions (9) -- (14) vanish in Z2F /I. fractional powers. For any linear combinationPk Every expression (9) -- (14) is, in fact, a linear combination of generalized j=1 Mj of such type (where the cancellations in monomials are already performed) we call the monomials Mj1, Mj2, j1, j2 ∈ {1, . . . , k}, incident monomials. Notice that the paths in the v-diagram corresponding to incident monomials always have the same initial points and the same final points. Clearly, one generalized fractional power has an infinite number of incident monomials because they may contain different powers of v and w. Suppose we have a group G with a relator R = M1M −1 2 . Let U = LM1R. Recall that the transition from U = LM1R to LM2R representing the same element of G L M2 M1 R is called a turn of an occurrence of a subrelation M1 (to its complement M2). We generalize this notion to a multi-turn. Multi-turns will play a central role in our work. Definition 2.4. LetPk j=1 Mj be one of the expressions (9) -- (14) (where the cancellations in monomials are already performed). We call the transition Mh 7−→ Mj, kXj=1 j6=h an elementary multi-turn of Mh. Let Uh = LMhR Uh = L Mh R be a monomial in Z2F , where Mh is a generalized fractional power, and 14 j=1 Mj be one of the expressions (9) -- (14), then the transition Pk Uh 7−→ Uj, kXj=1 j6=h where Uj = LMjR, Uj = L Mj R j = 1, . . . , k, j 6= h, is called a multi-turn of the occurrence Mh in Uh. Notice that sincePk I, Uh =Pk We call the corresponding sum Pk j=1 Uj the support of the multi-turn. Then the expressions (9) -- (14) are all types of supports of elementary multi-turns. j=1 Mj ∈ Uj in the ring Z2F /I. j=1 j6=h 3 Linear dependencies on Z2F induced by multi- turns. The description of the ideal I as a linear subspace of Z2F 3.1 How multi-turns influence the chart Let U be a word and a and b be members of its chart, i.e., maximal occur- rences of generalized fractional powers. Since the monomial U is represented as a segment, it is natural to represent members of the chart of U as its subsegments. Let us describe possible configurations of the members a and b. 1. We say that the members of the chart a and b are separated if there exists some non empty subword between them in the word U. non empty subword U a b 15 2. We say that the members a and b touch at a point if a and b are adjacent and have no common non empty subword. U a b 3. We say that the members a and b have an overlap if the members have non empty common subword. Recall that the Λ-measure of the overlap is not greater than ε. overlap a b U We will call the member of the chart of U closest to a from the left side the left neighbour of a in the chart of U. Similarly, we will call the member of the chart of U closest to a from the right side the right neighbour of a in the chart of U. Recall that maximal occurrences of generalized fractional powers are considered as members of the chart only when they have Λ-measure greater or equal to a given threshold τ and we assume that τ > 10ε. The restriction on the Λ-measure allows us to avoid situations when one member of the chart is fully covered by other members. So, we avoid the situation when there exist two members that are not separated from a on one side. For example, we never obtain configurations like the following: c c a a b b U U Λ(c) 6 2ε Λ(c) 6 2ε Suppose Uh = LahR is a word, ah is a member of its chart, in particular Uj is a multi-turn of Uh that comes from an Let us study how the chart of Uj = LajR, j = 1, . . . , k, j 6= h, is related to the chart of Uh = LahR. We distinguish three types of monomials in the aj. So, Uh =Pk j=1 j6=h Uj in the ring Z2F /I. j=1 j6=h j=1 j6=h Λ(ah) > τ . Suppose Uh 7→Pk elementary multi-turn ah 7→Pk sumPk j=1 j6=h Uj: 16 1. LajR, where Λ(aj) > ε; 2. LajR, where aj = 1. 3. LajR, where Λ(aj) 6 ε but aj 6= 1; Let Uj = LajR be a monomial of type 1, that is, Λ(aj) > ε. Let us study the chart of LajR. First of all notice that since ah is a maximal occurrence of a generalized fractional power in LahR, the monomial Uj = LajR has no cancellations and aj is a maximal occurrence of a generalized fractional power in Uj. Clearly, all members of the chart of Uh = LahR that are separated from ah remain unchanged in the chart of Uj = LajR. Let bh be the left neighbour of ah in the chart of Uh. Assume bh is not separated from ah, and b′ h is its initial subword such that L = L′b′ hahR. Then there is a corresponding occurrence of b′ h in Uj. Let bj be the member of the chart of Uj that prolongs b′ h. Clearly, the member of the chart of Uh that prolongs b′ h is precisely bh. The member bj may differ from bh when j 6= h. There are four possibilities: h, Uh = L′b′ 1.1 The members bh and ah touch at a point in Uh. The members bj and aj have a non-empty overlap in Uj. Assume c is the overlap between bj and aj. In this case bj = bhc. For example, we may obtain this effect when a beginning of vm is replaced by an end of v−1 or vice versa. i c aj ah bh 1.2 The members bh and ah have an overlap in Uh. The member bh is enlarged in Uj with the use of aj and the overlap between bj and aj increases in Uj. If the overlap increases by a piece c, then bj = bhc. bh Uh a′ a′′ a′′′ ah 17 bj bh c a′ a′′′ aj Uj 1.3 The members bh and ah have an overlap in Uh. The member bh is shortened in Uj such that bj and aj touch at a point in Uj. For example, we obtain this effect when a beginning of vm is replaced by an end of v−1 i or vice versa. In this case bh = bjc. c ah aj bj 1.4 The members bh and ah have an overlap in Uh. The member bh is shortened in Uj and the overlap between bj and aj becomes shorter in Uj. If the overlap decreases on a piece c, then bh = bjc. bh c a′ a′′′ ah bj a′ a′′ a′′′ aj Uh Uj The essential observation is that either bh = bjc, or bj = bhc; in both cases the Λ-measure of c can not exceed ε. The possibilities for being changed for the right neighbour of ah in the chart are the same. The cases described in 1.1 -- 1.4 may occur even if Λ(bh) < τ or Λ(bj) < τ . Therefore, the chart of Uj also may change as a result of the following effect. 18 Assume τ − ε 6 Λ(bh) < τ . Then we may obtain Λ(bj) > τ in Uj, that is, bj is counted as a member of the chart of Uj. So, a new member appears in the chart of Uj. It can also happen that Λ(bj) < τ , while Λ(bh) > τ . Consider the chart of a monomial Uj = LR of type 2. Cancellations between L and R may occur. Suppose L = L′C, R = C −1R′ and L′R′ does not have further cancellations. Let P be the meeting point of L′ and R′. Members of the chart of L′R′ that are separated from the point P remain unchanged. The possible configurations are as follows: 2.1 L′ does not have a terminal subword equal to some generalized frac- tional power and R′ does not have an initial subword equal to some generalized fractional power. Then the chart of L′R′ consists of mem- bers that lie left to P in L′ and right to P in R′. L′ L′ a P P b R′ R′ 1a, R′ = bR′ Suppose L′ = L′ 1, a and b are maximal occurrences of generalized fractional powers in L′ and in R′, respectively, possibly with Λ-measure less than τ . We allow one of a or b to be equal to 1. The occurrence of a in L′R′ does not have to be maximal. Let us denote by a′ the maximal occurrence of a generalized fractional power that prolongs a in L′R′. Similarly, let us denote by b′ the maximal occurrence of a generalized fractional power that prolongs b in L′R′. Then interactions between a′ and b′ are as follows: 2.2 a′ and b′ touch at a point in L′R′. Then like in the case 2.1 the chart of L′R′ consists precisely of the members contained in L′ and the members contained in R′. L′ Uj a′ b′ R′ P 2.3 a′ and b′ in L′R′ have an overlap. Then a′ is at most slightly enlarged with respect to a, and b′ is at most slightly enlarged with respect to b. Since the Λ-measure of their overlap does not exceed ε, their total increase does not exceed ε. 19 L′ Uj a′ a P b b′ R′ 2.4 a′ and b′ merge to one generalized fractional power ab in L′R′. L′ Uj a P b R′ merged generalized fractional power Consider the chart of monomials Uj = LajR of type 3. First of all notice that since ah is a maximal occurrence of a generalized fractional power in Uh, the monomial Uj = LajR has no cancellations. The possible configurations are as follows: 3.1 L does not have a terminal subword equal to some generalized fractional power and R does not have an initial subword equal to some generalized fractional power. Then all members of the chart of LajR are separated from aj. In this case no new members are added; hence, the chart of Uj consists of occurrences that lie strictly left to aj in L and strictly right to aj in R. Suppose L = L1a, R = bR1, a and b are maximal occurrences of gener- alized fractional powers in L and R, respectively, possibly with Λ-measure less than τ . We allow one of a or b to be equal to 1. As above, denote by a′ the maximal occurrence of a generalized fractional power that prolongs a in LajR, and denote by b′ the maximal occurrence of a generalized fractional power that prolongs b in LajR. We have the possibilities similar to the above 2.2 -- 2.4. 3.2 a′ remains equal to a and b′ remains equal to b in Uj. Hence, a′ and b′ are separated in LajR. L a′ b′ R aj 20 3.3 a′ slightly enlarges from the right side with respect to a, and b′ slightly enlarges from the left side with respect to b. Let us illustrate possible configurations of a′ and b′. a′ a′ a′ aj aj aj b′ b′ b′ aj aj aj We may obtain a combination of any two configurations for a′ and for b′ in LajR; hence, a′ and b′ may have any mutual position in LajR (i.e., remain either separate, or touch at a point, or have an overlap). Let us illustrate one possibility as an example. a′ aj b′ Since Λ-measure of a possible overlap of a′ and b′ does not exceed ε, the total increase of a′ and b′ with respect to a and b does not exceed 2ε. Hence, if Λ-measure of both a′ and b′ increases, Λ-measure of each of them increases at most by ε. 3.4 a′ enlarges with respect to a, b′ remains the same, or b′ enlarges with respect to b, a′ remains the same (the case when both of them enlarge is treated in 3.3). As above, a′ and b′ may have any mutual position in LajR (i.e.,either remain separate, or touch at a point, or have an overlap). In this case Λ-measure of a′ or b′ increases at most by 2ε. Let us illustrate one important possibility, the case when Λ-measure of a′ increases precisely by 2ε. L a′ aj 21 R b′ The most important case is the following: 3.5 a′ and b′ merge to one generalized fractional power aajb in Uj. L a aj b R merged generalized fractional power Remark 3.1. Notice that, using the explicit description of generalized frac- tional powers (5) -- (7) and the explicit description of elementary multi-turns (9) -- (14), one can show that the case 3.5 may take place in at most one monomial of the corresponding sumPk j=1 j6=h LajR. 3.2 Linear dependencies on Z2F induced by multi-turns First we prove the important property of the supports of elementary multi- turns, that are the expressions (9) -- (14). Then we will show that all the supports of multi-turns linearly generate the ideal I. This fact plays the key role for the description of the linear structure of the quotient ring Z2F /I. Proposition 3.1 (Transversality). Let Q1, . . . , Qn ∈ Z2F be linear combi- nations of generalized fractional powers of the form (9) -- (14). Consider l=1 Ql. If in this sum, after additively cancelling out the identical mono- mials, we obtain non-zero, then at least one of the remaining monomials (generalized fractional powers) has Λ-measure > τ . Pn Proof. Monomials of Ql can be represented as paths in the corresponding v-diagram with the same initial and the same final point. If some initial or some final point lies on a w-arc, we fix it on the arc marked by the first power l=1 Ql into subsums corresponding to the same initial of w. Divide the sumPn and final points (I, F ) in the v-diagram nXl=1 Let a be a member of Pn ber of Pn(I,F ) l=1 Q(I,F ) l l=1 Ql such that Λ(a) > τ . Notice that since Λ(a) > τ > ε, a has uniquely defined points I and F . So, a is a mem- with Q(I ′,F ′) and a is not a member of any Pn(I ′,F ′) l=1 l Ql = X(I,F ) n(I,F )Xl=1 Q(I,F ) l . 22 l=1 Q(I,F ) l (I ′, F ′) 6= (I, F ). Hence, if a is not additively cancelled out in the sum l=1 Ql. It l=1 Q(I,F ) Pn(I,F ) , then a is not additively cancelled out in the sumPn follows that it is enough to prove Proposition 3.1 for a sum Pn(I,F ) Assume the contrary. Suppose all members ofPn(I,F ) with some fixed (I, F ). after addi- tively cancelling out the identical monomials are of Λ-measure less than τ . Let us study possible forms of monomials in this sum. Suppose b is a mono- mial in this sum, Λ(b) < τ , I and F are the initial and the final points of the path corresponding to b. There are the following possible positions of I and F : l l l=1 Q(I,F ) 1. The points I and F lie on the v-arc. 2. The points I and F lie on a w-arc. 3. The point I lies on the v-arc, the point F lies on a w-arc. 4. The point I lies on a w-arc, the point F lies on the v-arc. Since Λ(b) < τ , the smallest path between I and F is necessarily of Λ-measure less than τ . Consider case 1. Let us denote the smallest path from I to F by b′, the smallest path from I to O by b1, and the smallest path from O to F by b2. Recall our notion that v = vivmvf . Then there are the following possibilities: 1.1 b′ contains the point O; 23 wk wk O vi I vf F w−k vm , b2 = v−1 b1 = v−1 f , i b′ = v−1 i v−1 f ; O vi F vf I w−k vm b1 = vf , b2 = vi, b′ = vf vi. 1.2 b′ does not contain the point O and Λ(b1) + Λ(b2) < τ wk O vi I vm F wk O vi F vm I w−k w−k vf b1 = v−1 i b′ = vm; vf , b2 = vivm, b1 = v−1 m v−1 b′ = v−1 m ; i , b2 = vi, 24 wk O w−k vf F I vm wk O w−k vf I F vm vi vi b1 = vmvf , b2 = v−1 f , b′ = vm; b1 = vf , b2 = v−1 b′ = v−1 m ; f v−1 m , 1.3 b′ does not contain the point O and Λ(b1) + Λ(b2) > τ . wk O w−k vf wk O w−k vf vi vi F vm I I vm F b′ = vm; b′ = v−1 m ; sum Pn(I,F ) l In case 1.3 there is only one path of Λ-measure < τ (b = b′). Hence, if the contains several different monomials of Λ-measure < τ , l=1 Q(I,F ) 25 then configuration 1.3 is not possible. Consider configurations 1.1 and 1.2. Since all members of the sumPn(I,F ) correspond to paths in the v-diagram with the initial point I and the final point F and of Λ-measure less than τ , the members are of the form l=1 Q(I,F ) l b1wkj b2. Notice that the monomial with kj = 0 may have cancellations and equals to b′ after cancellations. Let us return to the consideration of case 1 in general. Recall that all by from the right side, we obtain the Laurent such that P (v, w) satisfies (8). But, are of one of the types (9) -- (14). So, if we multiplyPn(I,F ) Q(I,F ) l b−1 1 j=1 Qjb−1 polynomial P (v, w) = b−1 on the other hand, in cases 1.1 and 1.2 from the left side and by b−1 2 l=1 Q(I,F ) 2 l 1 Pn P (v, w) = b−1 1 n(I,F )Xl=1 Q(I,F ) l b−1 2 = wkl. sXj=1 In case 1.3, the polynomial consists of one monomial P (v, w) = b−1 1 n(I,F )Xl=1 Q(I,F ) l 2 = b−1 b−1 1 b′b−1 2 . Evidently, this polynomials does not satisfy (8). This contradiction completes the proof. The remaining cases are considered in the same way. Let us define a subspace of Z2F of linear dependencies induced by multi- turns. For every monomial from F , we do all possible multi-turns of all members of the chart. As a result, we obtain a set of expressions Uj Uh ∈ F , Uh = LahR, where ah is a member of the chart of Uh, Uh 7→ Uj is a multi-turn coming from an elementary multi-turn ah 7→ T =( kXj=1 kXj=1 j6=h aj). kXj=1 j6=h (15) That is, T consists of the supports of all the multi-turns of members of the chart. We denote by hT i the linear span of this set. 26 Proposition 3.2. The linear subspace hT i ⊆ Z2F is equal to the ideal I = h1 + v + vwi. j=1 j6=h j=1 j6=h j=1 Uj)z ∈ T . aj, that is, T =Pk Proof. First let us show that hT i is an ideal in Z2F . Assume that T is the support of a multi-turn of Mh = LahR that comes from an elementary multi- j=1 Uj, Uj = LajR. We have to check that if Z is a monomial, then ZT ∈ hT i and T Z ∈ hT i. Clearly, it is sufficient to check this property only for a monomial Z that consists of only one letter j=1 Uj) ∈ T , j=1 Uj), Uj = LajR. Suppose Λ(ah) < τ . From Proposi- j=1 aj there exists a monomial ah′ such that Λ(ah′) > τ . Then, by definition, ah′ is a member of the chart of Lah′R. j=1 LajR can be considered as the support of LajR. So, in the sequel we can assume turn ah 7→Pk z. We will show an even stronger property, namely that z(Pk (Pk Consider z(Pk tion 3.1, it follows that in the sumPk Hence, the sumPk j=1 Uj =Pk the multi-turn Mh′ = Lah′R 7→Pk when it does. Since z(LajR) = (zL)ajR, Pk j=1(zL)ajR. Then, clearly, Pk Pk (zL)ahR 7→Pk (zL)ajR. Hence,Pk First consider the case when L is not empty. Since Λ(ah) > τ , ah is a member of the chart of (zL)ahR both when z does not cancel out with L, or j=1 z(LajR) = is the support of the multi-turn j=1 zUj = Pk that Λ(ah) > τ . j=1 zUj ∈ T . j=1 zUj j=1 j6=h Consider the case when L is empty, that is, Uj = ajR. If z does not cancel with ah and z does not prolong ah to a generalized fractional power from the left, then ah is a maximal occurrence of a generalized fractional power in zUh = zahR. Since Λ(ah) > τ , ah is a member of the chart of zahR. zajR is a multi-turn of a member of the chart and its Hence, zahR 7→Pk supportPk power), or z cancels with aj. Hence, zah 7→ Pk Assume that z does not cancel with ah and z prolongs ah to a generalized fractional power from the left. Then for every j 6= h either z does not cancel with aj (and since Λ(ah) > τ > ε, z prolongs it to a generalized fractional zaj is an elementary multi-turn (after the cancellations from the right hand side). Clearly, zah is a maximal occurrence of a generalized fractional power in zUh = zahR. Since Λ(zah) > Λ(ah) > τ , zah is a member of the chart of zahR. Therefore, j=1 zajR =Pk j=1 zUj belongs to T . j=1 j6=h j=1 j6=h′ 27 j6=h j=1 j=1 zUj belongs to T . zajR is a multi-turn of a member of the chart and its support zahR 7→Pk j=1 zajR =Pk Pk tional power), or z cancels with aj. Hence, zah 7→Pk Assume that z cancels with ah. Then for every j 6= h either z does not cancel with aj (and since Λ(ah) > τ > ε, z prolongs it to a generalized frac- zaj is an elementary multi-turn (after the cancellations). Since Λ(zah) 6 Λ(ah), we distinguish the following two possibilities: j=1 j6=h j=1 j6=h j=1 zUj belongs to T . j=1 zajR =Pk 1. Λ(zah) > τ . Then zah is a member of the chart of zahR. Therefore, zajR is a multi-turn of a virtual member of the chart 2. Λ(zah) > τ . Then zah is not a member of the chart of zahR. Then we argue as at the beginning of the proof. From Proposition 3.1 it j=1 zaj there exists a monomial zah′ such that Λ(zah′) > τ . So, zah′ is a member of the chart of zah′R. Hence, j=1 zajR can be considered as the support of the multi-turn j=1 zajR belongs j6=h′ j=1 zahR 7→ Pk and its supportPk follows that in the sum Pk the sumPk zah′R 7→Pk Summarising all of the above, we obtain zT = z(Pk the same reason we obtain T z = (Pk Evidently, every Pk j=1 LajR belongs to I for everyPk Pk j=1 aj of the form (9) -- (14) belongs to I. Hence, j=1 aj of the form (9) -- (14). So, T ⊆ I and hT i ⊆ I. Since vw + v + 1 is the support of the multi-turn vw 7→ v + 1 and Λ(vw) = 1 > τ , vw + v + 1 ∈ T . Therefore, since hT i is an ideal in Z2F , we obtain I ⊆ hT i. Thus, I = hT i. j=1) ∈ T . Clearly, for j=1)z ∈ T . Hence, hT i is an ideal in zajR. Thus, the sumPk j=1 zUj =Pk to T . Z2F . 3.3 Virtual members of the chart Recall that the chart of a word consists of maximal occurrences of generalized fractional powers with Λ-measure > τ (τ is our threshold). When we perform multi-turns, the measure of the occurrences may increase on the right or decrease on the right by at most ε in the resulting monomials of type 1. Similarly, it may increase on the left by at most ε or decrease on the left 28 by at most ε. It follows that occurrences with measure > τ (above the threshold) may turn into occurrences < τ (below the threshold) and vice versa. Therefore we need to modify our notion of a member of the chart to a notion with more stable properties with respect to multi-turns. We call such a notion a virtural member of the chart. In this section, we define this notion and study its properties. Namely, we may have the following effect. As above, let Uh be a mono- Uj be mial, ah be a member of its chart, and Uh = LahR. Let Uh 7→Pk a multi-turn of ah that comes from an elementary multi-turn ah 7→Pk Uj = LajR. Assume bh is a maximal occurrence of a generalized fractional power in Uh different from ah, Λ(bh) < τ , that is, bh is not counted as a member of the chart of Uh. According to the previous section, the element bh may be prolonged in Uj and the Λ-measure of the corresponding prolonged element bj in Uj may increase and become > τ . So, bj may become a member of the chart of Uj and the number of members of the chart of Uj may become greater than the number of members of the chart of Uh. We may obtain this effect for both neighbours of ah simultaneously. j=1 j6=h aj, j=1 j6=h Uh b(1) h b(1) j Uj b(1) h b(2) h b(2) j b(2) h ah aj Λ(b(1) Λ(b(2) h ) < τ h ) < τ Λ(b(1) Λ(b(2) j ) > τ j ) > τ In this case, the number of members of the chart of Uj become greater than the number of members of the chart of Uh even if Λ(aj) < τ . Assume that Uj = LajR is a resulting monomial of a multi-turn such that, roughly speaking, aj is of small Λ-measure (for example Λ(aj) < τ , so aj is not counted as a member of the chart). In Section 4, we will use an inductive argument for such resulting monomials of multi-turns. It looks natural to use induction by the number of members of the chart. But the example described above shows that the number of members of the chart may increase in Uj; hence, it is not an appropriate parameter for the induction. 29 In order to prove that the induction is nevertheless finite, we refine a notion of a member of the chart in this section. After that we introduce a function that guarantees finiteness of the inductive process (see Corollary 3.6). In this section, we consider the set M(Uh) of all maximal occurrences of generalized fractional powers in Uh, regardless of their Λ-measure, such that they are not properly contained in other occurrences of generalized fractional powers in Uh. There are two types of such occurrences: occurrences that are not fully covered by other occurrences from M(Uh) and occurrences that are fully covered by other occurrences from M(Uh). Denote the first set by Mnfc(Uh) and the second set by Mfc(Uh). Clearly, all maximal occurrences of a generalized fractional powers in Uh of Λ-measure greater than ε are contained in the set M(Uh). It is also clear that elements of Mfc(Uh) are of Λ-measure not greater than 2ε. For any maximal occurrence of a generalized fractional power in Uh we define a set of its images in a resulting monomial of a multi-turn. Definition 3.1. Let Uh = LahR be a monomial, and ah be a maximal occurrence of a generalized fractional power such that Λ(ah) > τ − 2ε. Let aj and ah be incident monomials. Assume bh is a maximal occurrence of a generalized fractional power in Uh that is different from ah and is not properly contained in ah. Since Λ(ah) > τ − 2ε > ε, ah can not be properly contained in any other occurrence, particularly in bh. To be precise, assume that the end of bh lies strictly left to the end of ah. We give the definition in the cases aj 6= 1 and aj = 1 separately. 1. Assume aj 6= 1, Uj = LajR. Then we call an element of M(Uj) that contains aj we call an image of ah in Uj. We call the set of all these elements the set of images of ah in Uj. Denote by b′ h a subword of bh that is an intersection of bh and L. We hL′ 1b′ have L = L′ 2 is possibly equal to 1. Hence, 2, where L′ hL′ 1b′ Uh = LahR = L′ hL′ 1b′ Uj = LajR = L′ 2ahR, 2ajR. We call an element of M(Uj) that contains b′ We call the set of all these elements the set of images of bh in Uj. h an image of bh in Uj. 2. Assume aj = 1. Then we say that the set of images of ah in Uj is empty. 30 Let L = L′C, R = C −1R′ and L′R′ has no further cancellations. If there is no non-empty subword of bh that is contained in L′, we say that the set of images of bh in Uj is empty. Otherwise, let b′ h be a subword of bh that is an intersection of bh and L′. We have L′ = L′ 2, where L′ 2 is possibly equal to 1. Hence, hL′ 1b′ Uh = LahR = L′CahC −1R′ = L′ Uj = L′R′ = L′ 2R′. hL′ 1b′ 1b′ hL′ 2CahC −1R′, We call an element of M(Uj) that contains b′ We call the set of all these elements the set of images of bh in Uj. h an image of bh in Uj. Clearly, we have a similar correspondence for bh when its beginning lies strictly right from the beginning of ah. If Λ(b′ h) 6 ε, it can be contained in several elements of M(Uj) and the set of images of bh may contain several elements. h) > ε, then its prolongation in Uj is uniquely determined and the set of images of bh consists of one element. In fact, in the previous section we described in detail all possible images of maximal occurrences of generalized fractional powers. If Λ(b′ Example 3.1. We observe the following effect for elements of Mfc(Uj). Let a maximal occurrence bh touch ah at a point from the left side. Suppose bh = b(1) h, where c′ h c′ h is a maximal occurrence of a generalized fractional power, Λ(c′ h) 6 ε. Suppose aj = c′ j is a maximal occurrence of a generalized fractional power, Λ(c′ j is a maximal occurrence of a generalized fractional power in Uj, then we obtain a new element in Mfc(Uj), which grows from two maximal occurrences c′ h /∈ M(Uh) and c′ ja(1) j j) 6 ε. Assume cj = c′ , where c′ hc′ j /∈ M(Uj). Uh bh c′ h cj ah Uj bh c′ h c′ j aj This example shows that we can not use just the size of M(Uj) as an in- ductive parameter. We will introduce special notions in order to control the behaviour of elements both from Mnfc(Uj) and from Mfc(Uj). 31 All elements of M(Uh) can be divided into sets such that every set covers a part of Uh and different parts are separated from each other. We denote them by M1(Uh), . . . , Mk(Uh)(Uh). We consider subcoverings of all parts, that are subsets of M1(Uh), . . . , Mk(Uh)(Uh), and call the union of this subcoverings a covering of Uh. There are finitely many different coverings of Uh, we denote them by {Ci(Uh) i = 1, . . . , n(Uh)}. If Z is a subword of Uh and Ci(Uh) is a covering of Uh, we consider elements of Ci(Uh) that have non empty intersection with Z. We call the set of this elements a covering of a subword Z and denote it by Ci(Z, Uh). Let us consider a subcovering of every set Mk(Uh) that consists of the smallest number of elements and denote this number by Nk(Uh). Let us call the union of this subcoverings a minimal covering of Uh. Clearly, a minimal covering of Uh may not be uniquely defined. We define N(U) = Nk(Uh), k(Uh)Xk=1 that is, the number of elements in a minimal covering of Uh. Notice that all elements of Mnfc(Uh) are contained in any covering of Uh. If some element of Mfc(Uh) is fully covered by elements of Mnfc(Uh), it is never contained in a minimal covering of Uh. Let us study a general structure of positions of elements of Mfc(Uh). Consider c ∈ Mfc(Uh), that is, c is fully covered by other elements of M(Uh). Consider the set of all maximal occurrences {ck} from M(Uh) that have non empty intersection with c. Then for every ck either its beginning point or its end point belong to c, since c is not properly contained in another occurrence. Consider ck1 with the leftmost beginning with respect to the beginning of c and ck2 with the rightmost beginning with respect to the beginning of c. ck2 c ck1 By the definition of M(Uh) neither of ck is contained in another one. Hence, if k 6= k1, then ck begins after ck1, if k 6= k2, then ck ends before ck2. Therefore, if k 6= k1, k 6= k2, then ck is fully covered by c, ck1, ck2, so, ck ∈ Mfc(Uh). ck2 ck1 c ck 32 ck2 ck c ck1 If ck1 is also fully covered by other elements of M(Uh), we consider an ele- ment of M(Uh) that has the end at the leftmost side of ck1 and repeat the argument. If ck2 is fully covered by other elements of M(Uh), we consider an element of M(Uh) that has non-empty intersection with ck2 with the right- most beginning with respect to the beginning of ck2 and repeat the argument. We continue the process until we find an element of Mnfc(Uh) that has the beginning from the left of c and an element of Mnfc(Uh) that has the end from the right of c. Denote the first element by d1 and the second element by d2. We see that all elements of M(Uh) that have the beginning after the beginning of d1 and the end before the end of d2 belong to Mfc(Uh). d2 · · · · · · c d1 Lemma 3.3. Let Uh be a monomial, ah ∈ Mnfc(Uh), Uh = LahR. Assume ah and aj are incident monomials, Uj = LajR. Then we have N(Uj) 6 N(Uh). If, moreover, aj is fully covered by images of elements of Mnfc(Uh) \ {ah} (for instance, if aj = 1), then N(Uj) < N(Uh). Proof. Let Ci1(Uh) be a minimal covering of Uh. First consider the case aj 6= 1. Since ah ∈ Mnfc(Uh), it necessarily belongs to Ci1(Uh). Hence, the covering Ci1(Uh) can be written as Ci1(Uh) = Ci1(L, Uh) ⊔ {ah} ⊔ Ci1(R, Uh), j an image of ah. Denote by C′ the where ⊔ is a disjoint union. Denote by a′ set of images of elements of Ci1(L, Uh), by C′′ the set of images of elements of Ci1(R, Uh). For elements that have more than one image, we take one arbitrary image. Every letter of L that is covered by Ci1(Uh) is covered by C′ in Uj. Every letter of R that is covered by Ci1(Uh) is covered by C′′ in Uj. The occurrence of aj in Uj is covered by a′ j. Therefore, C′ ∪ {a′ j} ∪ C′′ is a covering of Uj. 33 Denote it by Ci2(Uj). Since we take one image for every element of Ci1(Uh), C′ = Ci1(L, Uh), C′′ = Ci1(R, Uh), where · is the number of elements in a set. Hence, Ci2(Uj) 6 Ci1(Uh) = N(Uh), so we obtain N(Uj) 6 N(Uh). Let us show that bh ∈ Mnfc(Uh) \ {ah} has one image in Uj. If bh is separated from ah, it is obvious. Suppose bh is not separated from ah from the left. Let b′ h be a subword of bh that is an intersection of bh and L. We have L = L′ 1b′ h. Since bh is not fully covered by elements of M(Uh), b′ h is also not fully covered by elements of M(Uh). This means that b′ h can not be prolonged from the left neither in Uh nor in Uj. Hence, b′ h in Uj may be fully contained only in some element of M(Uj) that is contained in the word b′ hajR. Since b′ hajR, there exists only one such element of M(Uj) that contains b′ h itself). Hence, bh has one image in Uj. The case of bh being not separated from ah from the right is considered similarly. h is an initial subword of b′ h (possibly b′ Every element of Mnfc(Uh) is contained in any covering of Uh. Therefore, every element of Mnfc(Uh) \ {ah} is contained either in Ci1(L, Uh), or in Ci1(R, Uh). Assume aj is fully covered by images of elements of Mnfc(Uh) \ {ah}. Since every element of Mnfc(Uh) \ {ah} has only one image, aj is fully covered by C′ ∪ C′′. Therefore, C′ ∪ C′′ is a covering of Uj that has strictly less elements than Ci1(Uh). Thus, we obtain N(Uj) < N(Uh). Now consider the case aj = 1. Assume Uj = LR = L′CC −1R′ = L′R′, where L′R′ has no further cancellations. We have Uh = L′CahC −1R′, hence the covering Ci1(Uh) can be written as Ci1(Uh) = Ci1(L′, Uh) ⊔ {Ci1(L, Uh) \ Ci1(L′, Uh)} ⊔ {ah} ⊔ {Ci1(R, Uh) \ Ci1(R′, Uh)} ⊔ Ci1(R′, Uh). Denote by C′ the set of images of elements of Ci1(L′, Uh), by C′′ the set of images of elements of Ci1(R′, Uh). Again, for elements that have more than one image, we take one arbitrary image. Every letter of L′ that is covered by Ci1(Uh) is covered by C′ in Uj. Every letter of R′ that is covered by Ci1(Uh) is covered by C′′ in Uj. Hence, C′ ∪ C′′ is a covering of Uj. Denote it by Ci2(Uj). 34 Since we take one image for every element of Ci1(Uh), we have C′ = Ci1(L′, Uh), C′′ = Ci1(R′, Uh). Hence, Ci2(Uj) < Ci1(Uh) = N(Uh), so we obtain N(Uj) < N(Uh). Assume ah ∈ Mnfc(Uh). Consider the set of elements of M(Uh) that are not separated from ah from the left side. Evidently, there exists at most one element of Mnfc(Uh) in this set. Similarly, there exists at most one element of Mnfc(Uh) that is not separated from ah from the right side. Let us call them the essential left neighbour of ah and the essential right neighbour of ah, respectively. We define a special sequence of transformations of monomials. Definition 3.2. Let U be a word, b ∈ Mnfc(U). We denote U by U (1) and b by b(1) and define a sequence of transformations inductively. Assume monomials U (i), maximal occurrence of a generalized fractional powers b(i) in U (i) and transformations ri : U (i−1) 7→ U (i), 1 6 i 6 k, are already defined (we assume that the transformation r1 is identical). Let a(k) h be a maximal occurrence of a generalized fractional power in U (k) = L(k)a(k) h R(k) such that 1. a(k) h differs from b(k); 2. Λ(a(k) h ) > τ − 2ε (hence, a(k) h ∈ Mnfc(U (k))); 3. if the beginning of b(k) lies from the right of the beginning a(k) h has the right essential neighbour; if the beginning b(k) lies from the left of the beginning a(k) h , a(k) h has the left essential neighbour. h , a(k) are incident monomials. If a(k) in L(k)a(k) j R(k) is not j Suppose a(k) covered by images of Mnfc(U (k)) \ {a(k) h and a(k) j h }, then we say that j R(k), U (k+1) = L(k)a(k) b(k+1) is an image of b(k) in U (k+1), rk+1 : U (k) 7→ U (k+1) is a replacement of a(k) h by a(k) j . Lemma 3.4. Let U be a word, b ∈ Mnfc(U). Assume we have a sequence of transformations defined above that starts from U. If U (K) is a monomial in this sequence, then b(K) can increase or decrease at most by a piece of Λ-measure ε from the right side and at most by a piece of Λ-measure ε from the left side with respect to b. 35 Proof. In this proof, we use notations from Definition 3.2. Let us prove the lemma by induction on K. If a(1) h Assume K = 2, that is, the sequence of transformations has only one is separated from b(1), then the occurrence b(2) in U (2) is step r1. equal to b(1). Let a(1) h be not separated from b(1). Since b(1) = b ∈ Mnfc(U), b(1) is an essential neighbour of a(1) h . According to the classification given in Section 3.1, b(2) can decrease at most by a piece of Λ-measure ε after the replacement a(1) is h not covered by the image of b(1), by Definition 3.2. Hence, according to the classification given in Section 3.1, b(2) can increase at most by a piece of Λ-measure ε after the replacement a(1) h . Since b(1) is an essential neighbour of a(1) h , a(1) 7→ a(1) j 7→ a(1) j . j 7→ a(1) j Let us prove the step of the induction. Consider the replacement r2 : U (1) 7→ U (2), a(1) . Suppose b(2) ∈ Mfc(U). Then any element of h Mnfc(U) that is not separated from b(2) does not have the essential neighbour from the necessary side. Hence, there are no more possible replacements U (i) 7→ U (i+1), i > 2, of occurrences that are not separated from b(i). So, b(i) remains equal to b(2) after any further replacement. Further we consider only the case b(2) ∈ Mnfc(U). If a(1) h is separated from b(1), then the occurrence b(2) in U (2) is equal to b(1). The sequence of transformations starting from U (2) has a fewer number of steps. Hence, b(K) can increase or decrease at most by a piece of Λ-measure ε from the right and the left sides with respect to b(2) (= b(1)) by the induction hypothesis. Suppose a(1) h is not separated from b(1). If b(2) remains equal to b(1), then as above b(K) can increase or decrease at most by a piece of Λ-measure ε from the right and the left sides with respect to b(1) by the induction hypothesis. Assume Λ(b(2)) changes with respect to Λ(b(1)). Then there are the fol- lowing cases: 1. The beginning of b(1) lies from the left of the beginning of a(1) h , Λ(b(2)) < Λ(b(1)). 2. The beginning of b(1) lies from the right of the beginning of a(1) h , Λ(b(2)) < Λ(b(1)). 3. The beginning of b(1) lies from the left of the beginning of a(1) h , Λ(b(2)) > Λ(b(1)). 4. The beginning of b(1) lies from the right of the beginning of a(1) h , Λ(b(2)) > Λ(b(1)). 36 Let us consider case 1. There are the following two configurations for b(2): (α) a(1) j ∈ Mnfc(U (2)) b(2) U (2) a(1) j (β) a(1) j ∈ Mfc(U (2)) Λ(c) 6 4ε U (2) b(2) c a(1) j If a(1) j ∈ Mfc(U (2)), then an element c ∈ M(U (2)) covers its terminal subword. Since a(1) h }, c j is not an image of an element of Mnfc(U (1)) \ {a(1) h }. Hence, c is an image of some generalized fractional power of Λ-measure not greater than 2ε. Then, by the classification from Section 3.1, Λ(c) 6 4ε. is not covered by images of Mnfc(U (1)) \ {a(1) Let us study case (β) first. Since Λ(c) 6 4ε < τ −2ε, c can not be used for the next replacement r3. The sequence of replacements starting from U (2) has K − 1 steps. Hence, if c ∈ Mnfc(U (2)), then by the induction hypothesis the Λ-measure of an image of c can increase at most by 2ε with respect to Λ(c) in every U (i), 3 6 i 6 K. So, an image of c is of less Λ-measure than τ − 2ε and can not be used for a replacement in any U (i), 3 6 i 6 K. Hence, all the next replaced occurrences are separated from b(i) from the right side. So, b(i) has no more changes from the right side with respect to b(2), 3 6 i 6 K. Suppose c ∈ Mfc(U (2)). Denote the element closest to c from the right such that it belongs to Mnfc(U (2)) by d. Then d does not have the left essential neighbour. By the induction hypothesis, b(i) can decrease at most by a piece of Λ-measure ε from the left side. Hence, the image of d can not have the left essential neighbour in every U (i). Thus, d can not be used for a replacement in any U (i), 3 6 i 6 K. Therefore, all the next replaced occurrences are separated from the image of c from the right side. Hence, b(i) has no more changes from the right side with respect to b(2), 3 6 i 6 K. 37 Consider case (α). Consider the next replacement r3 : U (2) 7→ a(2) j 7→ U (3), a(2) . With a slight abuse of language, we use the same indices in h different incident monomials. By the induction hypothesis, b(i), 3 6 i 6 K, may increase or decrease from the left at most by a piece of Λ-measure ε with respect to b(2). This means that it remains to consider only replacements from the right of b(2). Therefore, without loss of generality, we can assume that the beginning a(2) lies from the right of the beginning of b(2). Then there are h the following possibilities: is separated from a(1) j 1. a(2) h b(3), that is, the image of a(1) . Then we again obtain configuration (α) for j belongs to Mnfc(U (3)). . coincides with a(1) j 2. a(2) j ∈ Mnfc(U (3)), we obtain h configuration (α) for b(3), possibly with a smaller overlap or without an overlap between b(3) and a(2) j ∈ Mfc(U (3)), we obtain config- j uration (β) for b(3). In this case if a(2) . If a(2) 3. a(2) j belongs to Mnfc(U (3)), belongs to If the image of a(1) j h is not separated from a(1) we obtain configuration (α) for b(3). Mfc(U (3)), we obtain configuration (β) for b(3). . If the image of a(1) j So, we may obtain for b(3) either configuration (β) that was already consid- ered, or again configuration (α). If we obtain configuration (α), b(3) increases or decreases at most by a piece of Λ-measure ε from the right side with respect to b(1). Cases 2 -- 4 are studied in the same way as case 1. So, if we continue the description for the next transformations ri, we observe that possible changes of b(i) with respect to b(1) are restricted by a piece of Λ-measure not greater than ε from each side. Remark 3.2. In fact, in Lemma 3.4 we also proved the following. Assume that some maximal occurrence a ∈ Mnfc(U) has an overlap of measure ε with its left essential neighbour. Then after a sequence of replacements from Definition 3.2, the Λ-measure of the image of a can not increase from the left with respect to the Λ-measure of a. Similarly, if a has an overlap of measure ε with its right essential neighbour, then after a sequence of replacements from Definition 3.2 the Λ-measure of the image of a can not increase from the right with respect to the Λ-measure of a. 38 Definition 3.3. Let U be a word, b ∈ Mnfc(U). If there exists a sequence of replacements constructed in Definition 3.2 such that U (K) is the last mono- mial in the sequence and Λ(b(K)) > τ , we call b a virtual member of the chart of U. We call the set of such maximal occurrences from M(U) the virtual τ -chart of U. We denote the number of virtual members of the chart of U by Kτ (U). From Lemma 3.4 it follows that if b is a virtual member of the chart of U, then Λ(b) > τ − 2ε. Clearly, every member of the chart of U is also a virtual member of the chart of U. From now on, when we speak about the chart of some word, we use only virtual members of the chart even if the qualification virtual is omitted. Let us prove important properties of virtual members of the chart that we will use in the further argument. Lemma 3.5. Let Uh be a monomial, ah be a virtual member of its chart, Uh = LahR. Assume ah and aj are incident monomials, aj 6= 1, Uj = LajR. If aj is not fully covered by images of Mnfc(Uh) \ {ah}, then Kτ (Uj) 6 Kτ (Uh). If, moreover, aj is not a virtual member of the chart of Uj, then Kτ (Uj) < Kτ (Uh). Proof. Let Uh = LahR, Uj = LajR. Suppose b is a virtual member of the chart of Uj different from aj. Since aj is not fully covered by images of Mnfc(Uh) \ {ah}, aj is not fully contained in b. To be precise, assume that the beginning of b lies from the left side of the beginning of aj, that is, b is an occurrence in the subword Laj. Since b is a virtual member of the chart of Uj, there exists a sequence of transformations Uj = U (1) j 7→ . . . 7→ U (K) j (16) defined above such that the image of b in U (K) by b′ the inverse image of b in Uh. j is of Λ-measure > τ . Denote First assume that ah in Uh has the left essential neighbour. Since aj is not fully covered by images of Mnfc(Uh) \ {ah}, the replacement ah 7→ aj can be used in the definition of a virtual member of the chart. Hence, we can add the transformation Uh 7→ Uj to the beginning of the sequence (16). Therefore, we obtain the sequence that starts from Uh and defines b′ as a virtual member of the chart of Uh. 39 Now assume that ah in Uh does not have the left essential neighbour. Then it is possible that there exist elements of Mfc(Uh) that are not separated from ah from the left side. The other possibility is that there are no elements of M(Uh) that are not separated from ah from the left side. Let us study the first case. Since ah in Uh does not have the left essential neighbour, b′ is separated from ah. b′ ah Uh Consider the first possibility, namely, there exist elements of Mfc(Uh) that are not separated from ah from the left side. Let us show that there exists an element of M(Uj) of Λ-measure 6 5ε such that its beginning lies between the beginning of b and the beginning of aj and it is not fully covered by its essential neighbours. Let d′ ∈ M(Uh) be not separated from ah from the left and has the leftmost beginning among such elements. Let d be an image of d′ in Uj. Since ah does not have the left essential neighbour, d′ ∈ Mfc(Uh). Hence, Λ(d′) 6 2ε, so, Λ(d) 6 5ε. If d is not covered by its essential neighbours, then d is the necessary element. Suppose d is covered by its essential neighbours. In particular, this means that d has both the left and the right essential neighbour. First assume that aj is contained in d. Then aj is covered by essential neighbours of d. Since aj is not covered by images of Mnfc(Uh) \ {ah}, at least one essential neighbour of d is not equal to an image of the element of Mnfc(Uh) \ {ah}. Hence it is of Λ-measure 6 5ε and it satisfies necessary conditions. Assume that aj is not contained in d. Since d′ has the rightmost beginning among elements that are not separated from ah from the left, its image d in Uj has the same beginning position. Therefore, the left essential neighbour of d in Uj is the image of the left essential neighbour of d′ in Uh. Hence, the left essential neighbour of d is separated from aj, since ah does not have the left essential neighbour. So, d can not be covered by its left essential neighbour and aj. Then consider the right essential neighbour of d and denote it by c. Since d is not covered by its left essential neighbour and aj, the occurrence c is not equal to the occurrence aj. If some element of M(Uj) has the beginning from the right of the beginning of d and the end from the left of the end of aj, it is covered by d and aj. Hence, it does not belong to Mnfc(Uj). Therefore, since c ∈ Mnfc(Uj), aj is contained in c. Since aj is not covered by images of Mnfc(Uh) \ {ah}, c is not equal to an image of an element of Mnfc(Uh) \ {ah}. 40 Since the occurrence c is not equal to the occurrence aj, Λ(c) 6 5ε and it satisfies the necessary conditions. We denote the obtained element by g. We put g = glgmgr, where gl is an overlap with the left essential neighbour (if there is any) and gr is an overlap with the right essential neighbour (if there is any). Since g is not covered by its left and right essential neighbours, gm is not empty. We put Uj = U (1) j = L(1)gmR(1). Then b is an occurrence in the subword L(1). From Lemma 3.4 it follows that the Λ-measure of an image of g in every U (i) is not greater than 7ε < τ − 2ε. Therefore, an image of g can not j be used in any replacement U (i) in (16). Then every replacement j in (16) is of the form 7→ U (i+1) j L(i)gmR(i) 7→ L(i+1)gmR(i+1), where either L(i+1) = L(i), or R(i+1) = R(i). The possibility of a transformation L(i)gmR(i) 7→ L(i+1)gmR(i+1), where R(i+1) = R(i), does not depend on R(i). Since b is an occurrence in the subword L(1), it is sufficient to do only transformations L(i)gmR(i) 7→ L(i+1)gmR(i+1), where R(i+1) = R(i), in a sequence that defines b. So, we obtain the sequence U (1) j = L(1)gmR(1) 7→ L(i1)gmR(1) . . . 7→ L(is)gmR(1) that also defines b. We have Uh = L(1)R′. Thus, the same replacements can be done starting with Uh, so, we obtain the sequence Uh = L(1)R′ 7→ L(i1)R′ . . . 7→ L(is)R′ (17) such that the image of b′ in L(is)R′ is of Λ-measure > τ . Therefore, b′ is a virtual member of the chart of Uh. The case when there are no elements of M(Uh) that are not separated from ah from the left side is considered in a similar (but easier) way. So, we have proved that every virtual member of the chart of Uj is the image of a virtual member of the chart of Uh. Hence, Kτ (Uj) 6 Kτ (Uh). If aj is not a virtual member of the chart of Uj, then the image of ah is not a virtual member of the chart of Uj, because aj is not fully covered by images of Mnfc(Uh) \ {ah}. Thus, in this case Kτ (Uj) < Kτ (Uh). Corollary 3.6. Let Uh be a monomial, ah be a virtual member of its chart, Uh = LahR. Assume ah and aj are incident monomials, Uj = LajR. If 41 aj is a virtual member of the chart of Uj, then N(Uj) = N(Uh). If aj is not a virtual member of the chart of Uj, then either N(Uj) < N(Uh), or N(Uh) = N(Uj) and Kτ (Uj) < Kτ (Uh). Proof. From Lemma 3.3 it follows that N(Uj) 6 N(Uh). Let aj be a virtual member of the chart of Uj, then aj ∈ Mnfc(Uj). Hence, we can apply Lemma 3.3 to the opposite replacement aj 7→ ah and obtain N(Uh) 6 N(Uj). Thus, N(Uj) = N(Uh). Assume aj is not a virtual member of the chart of Uj. From Lemma 3.3 it follows that if aj is fully covered by images of Mnfc(Uh) \ {ah}, then N(Uj) < N(Uh). Therefore, if N(Uh) = N(Uj), then aj is not fully covered by images of Mnfc(Uh) \ {ah}. Hence, from Lemma 3.5 it follows that Kτ (Uj) < Kτ (Uh). Corollary 3.7. Let Uh be a monomial, ah be a virtual member of its chart, Uh = LahR. Assume ah and aj are incident monomials, aj is a virtual member of the chart of Uj = LajR. Assume bh ∈ Mnfc(Uh), bh is different from ah and bj is the image of bh in Uj. Then bh is a virtual member of the chart of Uh if and only if bj is a virtual member of the chart of Uj. In particular, in this case Kτ (Uh) = Kτ (Uj). Proof. Since aj is a virtual member of the chart of Uj, Λ(aj) > τ − 2ε and aj ∈ Mnfc(Uj). Hence, the conditions of Lemma 3.5 hold for the replacement Uh 7→ Uj, ah 7→ aj. In the proof of Lemma 3.5 we actually show that if bj is a virtual member of the chart of Uj, then bh is a virtual member of the chart of Uh. Since aj is a virtual member of the chart of Uj, we can consider the opposite transformation Uj 7→ Uh, aj 7→ ah. Then, obviously, the image of bj in Uh is bh. If we apply Lemma 3.5 in this case, we obtain that if bh is a virtual member of the chart of Uh, then bj is a virtual member of the chart of Uj. In the same way as (15) we define the set of all the supports of multi-turns 42 of virtual members of the chart T ′ =( kXj=1 Uj Uh ∈ F , Uh = LahR, whereah is a virtual member (18) of the chart of Uh, Uh 7→ Uj is a multi-turn kXj=1 j6=h that comes from an elementary multi-turn ah 7→ aj). kXj=1 j6=h Proposition 3.8. The linear subspace hT ′i ⊆ Z2F is equal to the ideal I = h1 + v + vwi. Proof. From Proposition 3.1 it easily follows that T ′ = T . Hence, by Propo- sition 3.2, hT ′i = I. 3.4 Sequences of transformations of a given monomial Let U be a monomial, ah and bh be virtual members of the chart of U = LaahRa = LbbhRb. To be precise, assume that the beginning of ah lies from the left of the beginning of bh. Assume ah and aj, bh and bj are incident monomials and consider transformations ah 7→ aj, bh 7→ bj. If aj or bj is equal to 1 and the resulting monomial LaRa or LbRb have cancellations, we do not perform them right after the replacement. of bh in LaajRa. fractional power that cancels or prolongs bh (d2 may be empty). Leteah be the image of ah in LbbjRb. If bj = 1 and LbRb has cancellations, we say thateah is the intersection of ah and Lb (notice that this slightly differs from Definition 3.1). Then eah = ahd1, where d1 is a generalized fractional power that cancels or prolongs ah (d1 may be empty). Letebh be the image If aj = 1 and LaRa has cancellations, we say thatebh is the intersection of bh and Ra. Thenebh = d2bh, where d2 is a generalized Since the monomials ah and aj are incident, the monomials ahd1 = eah and ajd1 are also incident. Hence, we consider the replacement eah 7→ eaj in LbbhRb, where eaj = ajd1. Similarly, since the monomials bh and bj are incident, the monomials d2bh =ebh and d2bj are also incident. So, we consider the replacementebh 7→ebj in LaahRa, whereebj = d2bj. The next lemma states an important property of these transformations. 43 Lemma 3.9. The result of the replacementeah 7→eaj in LbbjRb and the result of the replacementebh 7→ebj in LaajRa are equal. Proof. First assume that ah and bh are separated, U = LaahMbhRb. Then we obtain two sequences of transformations U = LaahMbhRb 7→ LaajMbhRb 7→ LaajMbjRb and U = LaahMbhRb 7→ LaahMbjRb 7→ LaajMbjRb. Obviously, the results are equal. Assume ah and bh touch at a point, U = LaahbhRb. After the replacement ah 7→ aj we obtain the monomial LaajbhRb. The image of bh in LaajbhRb may have different forms depending on aj (see classification in Section 3.1). But in any case the replacementebh 7→ebj in LaajbhRb can be represented as a replacement bh 7→ bj and the further cancellations (if there are any). So, we obtain U = LaahbhRb 7→ LaajbhRb 7→ LaajbjRb. After the replacement bh 7→ bj we obtain the monomial LaahbjRb. Similarly, ah 7→ aj and the further cancellations (if there are any). So, we obtain the replacement eah 7→eaj in LaahbjRb can be represented as a replacement U = LaahbhRb 7→ LaahbjRb 7→ LaajbjRb and results of both sequences are equal. Assume ah and bh have an overlap c. Denote by a′ ah and Lb, by b′ the replacement ah 7→ aj we obtain the monomial Laajb′ h the intersection of bh and Ra, then U = Laa′ h the intersection of hRb. After hRb. As above, hRb can be represented as a replacement hcb′ b′ h 7→ c−1bj and the further cancellations (if there are any). So, we obtain the replacementebh 7→ebj in Laajb′ U = Laa′ hcb′ hRb 7→ Laajb′ hRb 7→ Laajc−1bjRb. After the replacement bh 7→ bj we obtain the monomial Laa′ hbjRb. As above, hbjRb can be represented as a replacement a′ h 7→ ajc−1 and the further cancellations (if there are any). So, we obtain the replacement eah 7→eaj in Laa′ U = Laa′ hcb′ hRb 7→ Laa′ hbjRb 7→ Laajc−1bjRb and results of both sequences are equal. 44 Lemma 3.10. Let U be a monomial, ah and bh be virtual members of the chart of U, U = LaahRa = LbbhRb. Let ah and aj be incident monomials, bh and bj be incident monomials, and consider transformations ah 7→ aj, bh 7→ bj. Assume aj is a virtual member of the chart of LaajRa, bj is a bh 7→ bj by the corresponding generalized fractional power. Let us apply this virtual member of the chart of LbbjRb. Denote by ebh the image of bh in LaajRa and consider the transformationebh 7→ebj, obtained by multiplying of replacement to the monomial LaajRa and denote the result by Ua,b. Thenebj is a virtual member of the chart of Ua,b. Proof. To be precise, assume that the beginning of ah lies from the left of the beginning of bh. We will consider only the most interesting case when ah and bh have an overlap c. The cases when ah and bh are separate or touch at a point are considered similarly. Denote by a′ h the intersection of ah and Lb, that is, ah = a′ hc. Denote by h the intersection of bh and Ra, that is, bh = cb′ b′ h. ah U a′ h c b′ h bh Let us perform the replacement ah 7→ aj in U = LaahRa. Then the image of bh in LaajRa is equal to c1b′ h, where c1 is an overlap with aj (possibly empty). We put aj = a′ Since aj is a virtual member of the chart of LaajRa, from Corollary 3.7 it follows that ebh is also a virtual member of the chart of LaajRa. The monomials bh and bj are incident. Hence, multiplying them by c1c−1 from h and c1c−1bj. So, we have the the left, we obtain incident monomials c1b′ replacement c1b′ h =ebh 7→ebj, whereebj = c1c−1bj. 45 jc1 andebh = c1b′ h. aj LaajRa a′ j c1 b′ h ebh Let us perform the replacement bh 7→ bj in U = LbbhRb. Then the image hc2, where c2 is an overlap with bj (possibly of ah in LbbjRb is equal to a′ empty). We put bj = c2b′ hc2. j andeah = a′ eah a′ h b′ j bj LbbjRb c2 Since bj is a virtual member of the chart of LbbjRb, from Corollary 3.7 it fol- ah and aj are incident. Hence, multiplying them by c−1c2 from the right, we obtain incident monomials a′ hc2 and ajc−1c2. So, we have the replacement a′ lows thateah is also a virtual member of the chart of LbbjRb. The monomials hc2 =eah 7→eaj, whereeaj = ajc−1c2. From Lemma 3.9 it follows that the sequences of replacements and give the same final result. So, we have U = LaahRa 7→ LaajRa 7→ Laa′ jebjRb U = LbbhRb 7→ LbbjRb 7→ Laeajb′ jRb eaj a′ j b′ j c3 Ua,b where the overlap c3 is possibly empty. One can easily calculate that c3 = c1c−1c2. ebj Recall that the Λ-measure of any overlap of two maximal occurrences of generalized fractional powers is equal either to ε, or to zero. Since aj = a′ jc1 is a virtual member of the chart of LaajRa, Λ(aj) > τ − 2ε. If Λ(c1) = ε, then, according to Remark 3.2, Λ(aj) > τ − ε. Hence, Λ(a′ j) > τ − 2ε. If Λ(c1) = 0, then, clearly, Λ(a′ j) = Λ(aj) > τ − 2ε. j) > τ − 2ε. Since bj is a virtual member of the chart of LbbjRb, there exists a sequence of replacements from Definition 3.2 starting from LbbjRb such that the Λ-measure of the image of bj in the last We have Λ(eaj) > Λ(a′ 46 a result we obtain the sequence of replacements starting from Ua,b such that obtain the transformation Ua,b 7→ LbbjRb. We add this transformation to the monomial of the sequence is not less than τ . Since Λ(eaj) > τ − 2ε, we can consider the replacement eaj 7→ a′ hc2 = eah in the monomial Ua,b. Then we beginning of the sequence. Since the image ofebj in LbbjRb is equal to bj, as the Λ-measure of the image ofebj in the last monomial is not less than τ . Hence,ebj is a virtual member of the chart of Ua,b. Applying Lemma 3.9 and Lemma 3.10, we obtain the following statement, that we will use in the next section. Corollary 3.11. Let U be a monomial. (1) Assume we have a sequence of replacements starting from U such that every replacement transforms a virtual member of the chart into a vir- tual member of the chart. Then any replacement can be moved to any position in the sequence and the final result remains the same. More- over, after changing of the order of the replacements, every replacement in the obtained sequence still transforms a virtual member of the chart into a virtual member of the chart. (2) Let a(1) h , . . . , a(n) h 7→ a(i) j be virtual members of the chart of U. Consider a number of replacements a(i) is a vir- h tual member of the chart of the resulting monomial. Suppose these transformation are applied consecutively. Then every transformation in the chain also transforms a virtual member of the chart into a vir- tual member of the chart. Moreover, we can apply the replacements in any order. in U such that each a(i) j 4 The structure of certain subspaces of Z2F/I: filtration, grading and tensor products First let us introduce the notion of derived monomials. Definition 4.1 (derived monomials). Let U be a monomial. Consider the following transformations of U: 47 (1) Replacements of a virtual member of the chart by an incident monomial non-equal to 1. Recall that in this case the result is always a reduced monomial. (2) Replacements of a virtual member of the chart by incident monomial equal to 1 and further cancellations (in order to obtain the reduced monomial). Starting with a certain monomial U we consecutively apply transformations (1), (2). All the monomials that we obtain after some sequence of transforma- tions (1), (2) (including the monomial U itself) are called derived monomials of U. Definition 4.2. Let {Ui} be either finite or countable set of monomials. By hU1, . . . , Uk, . . .id, we denote a subspace of Z2F generated by all the derived monomials of the monomials {Ui}. Remark 4.1. Assume U is a monomial and U ∈ hU1, . . . , Uk, . . .id, where Ui . So, we obtain that U ∈ hU1, . . . , Uk, . . .id if and only if U is a derived monomial of some Ui. j=1eUij , where eUij is a derived monomial of Uij . are monomials. Then U =Pl Since monomials form a basis of Z2F , we obtain U = eUij0 We have shown that I = hT ′i. Consider Pl Uh 7→Pk by definition of hUhid, we obtain T ∈ hUhid. j=1 Uj = T ∈ T ′, where Uj is a multi-turn of a virtual member of the chart of Uh. Then, j=1 j6=h In this section, we show that hU1, . . . , Ukid ∩ I is generated by supports of multi-turns of monomials from hUiid for i = 1, . . . , k. This enables us to construct a linear basis of hU1, . . . , Ukid/(hU1, . . . , Ukid ∩ I) and a linear basis of Z2F /I (and therefore show that Z2F /I is non-trivial). Our approach can be compared with the more standard one, that uses the Diamond Lemma to control the ring relations ([1]). Instead, we intro- duce a filtration (and the corresponding grading) on hU1, . . . , Ukid and show its compatibility with the subspaces of linear dependencies (see Lemma 4.7 and Theorem 4.1). This allows us to deal with linear dependencies in each graded component independently. The structure of each graded component is described in Proposition 4.10 and Proposition 4.12. Notice that the Diamond Lemma can also be reformulated in the language of grading. 48 4.1 The filtration on spaces hU1, . . . , Ukid Let Z be a monomial. We introduce the following numerical characteristics of Z (f -characteristics of monomials): f (Z) = (N(Z), Kτ (Z)), (19) where N(Z) is the number of elements in a minimal covering of Z, Kτ (Z) is the number of virtual members of the chart of Z. If Z1 and Z2 are monomials, we say that f (Z1) < f (Z2) if and only if N(Z1) < N(Z2) or N(Z1) = N(Z2) and Kτ (Z1) < Kτ (Z2). The characteristics f satisfies the following property. Lemma 4.1. Assume U and Z are monomials, where Z is a derived mono- mial of U. Then f (Z) 6 f (U). Moreover, f (Z) < f (U) if and only if in the corresponding sequence of replacements there exists at least one replacement of the form LahR → LajR such that ah is a virtual member of the chart of LahR and aj is not a virtual member of the chart of LajR. Proof. It follows directly from Corollary 3.6 and Corollary 3.7. In this section, let us set V = hU1, . . . , Ukid, U1, . . . , Uk ∈ F . We will define a decreasing filtration on V . First consider a space generated by one monomial and its derived mono- mials, namely, W = hUid, U ∈ F . Let us define a subspace L(W ) ⊆ W . We put L(W ) = hZ ∈ F Z is a derived monomial of U such that f (Z) < f (U)i. (20) If the set of monomials with strictly smaller f -characteristics than f (U) is empty, by definition, we put L(W ) = 0. Let a monomial Z ′ ∈ L(W ), Z ′′ be a derived monomial of Z ′. Then, by definition of L(W ) and Lemma 4.1, f (U) > f (Z ′) > f (Z ′′). Hence, Z ′′ also belongs to L(W ), that is, L(W ) is closed under taking derived monomials. Now consider a space Y = hZ1, . . . , Zk, . . .id, where {Zi} is either a finite or infinite set of monomials. By definition, we put L(hZiid). (21) L(Y ) =Xi 49 Since every L(hZiid) is generated by monomials and closed under taking derived monomials, the space L(Y ) is generated by monomials and closed under taking derived monomials as well. Hence, L(Y ) = hZ ′ 1, . . . , Z ′ k, . . .id for some Z ′ i ∈ F . (22) Notice that one generalized fractional power may have an infinite number of incident monomials of Λ-measure less than τ − 2ε, because they may contain different powers of w. Recall that such generalized fractional powers are never counted as a virtual members of the chart. Hence, even if the space Y is generated by derived monomials of finite number of monomials, the space L(Y ) might be generated by derived monomials of countably many monomials. In the sequel, we will widely use the following simple properties of derived monomials. Lemma 4.2. Let Z1 be a monomial, Z be a derived monomial of Z1, Y ⊆ Z2F be a space generated by monomials and closed under taking derived monomials. Then the following statements hold: (1) If Z ∈ hZ1id \ L(hZ1id), then hZ1id = hZid. (2) If Z ∈ Y and Z ∈ hZ1id \ L(hZ1id), then hZ1id ⊆ Y . Proof. Suppose Z ′ 1 is a result of transformation (1) (Definition 4.1) such that Z ′ 1 is contained in hZ1id \ L(hZ1id). Then this transformation is invertible, hence Z1 is also a derived monomial of Z ′ 1. Repeating this argument for every transformation in a sequence that connects Z1 and Z, we obtain that Z1 is a derived monomial of Z. Now assume U is a derived monomial of Z1. Since Z1 is a derived monomial of Z, U is also a derived monomial of Z. Hence hZ1id = hZid. The first statement of the lemma is proved. Now let us prove the second statement. Since Z ∈ Y and Y is generated by monomials and closed under taking derived monomials, we obtain hZid ⊆ Y . But above we proved hZid = hZ1id, hence, hZ1id ⊆ Y . We defined the subspace L(Y ) using a set of generators of Y (see for- mula (21)). Let us show that, in fact, L(Y ) does not depend on the set of generators of Y . 50 Proposition 4.3. Assume Y = hZ1, . . . , Zk, . . .id = hZ ′ 1, . . . , Z ′ k, . . .id, where {Zi} and {Z ′ i} are either finite or infinite sets of monomials. Then Xi L(hZiid) =Xi L(hZ ′ iid). Proof. Let us enumerate all monomials from Y . Let {Xj} be the set of all monomials of the space Y . Then, evidently, we obtain the following description of Y : Y = hX1, . . . , Xk, . . .i = hX1, . . . , Xk, . . .id. Clearly, it is sufficient to show that Xi L(hZiid) =Xj L(hXjid) for an arbitrary set of monomials {Zi} such that Y = hZ1, . . . , Zk, . . .id. Since {Z1, . . . , Zk, . . .} ⊆ {X1, . . . , Xk, . . .}, clearly, Xi L(hZiid) ⊆Xj L(hXjid). Assume Z ∈ Pj L(hXjid). Monomials form a basis of Z2F and every L(hXjid) is generated by monomials. Hence, Z is a derived monomial of some Xj0 such that we have Z ∈ L(hXj0id) for some j0. Therefore, there exists a sequence of transformations α1, . . . , αs1 of type (1), (2) such that Xj0 α1 7−→ . . . αs1 7−→ Z. (23) Moreover, Yl−1 → Yl such that f (Yl) < f (Yl−1). in the sequence, there exists at least one transformation αl : Since Xj0 ∈ Y = hZ1, . . . , Zk, . . .id, we have Xj0 ∈ hZi0id for some i0. Hence, there exists a sequence of transformations β1, . . . , βs2 of type (1), (2) such that βs2 7−→ Xj0. Zi0 β1 7−→ . . . 51 (24) Gluing (24) and (23), we obtain Zi0 β1 7−→ . . . βs2 7−→ Xj0 α1 7−→ . . . αs1 7−→ Z. Consequently, since there exists αl that decreases the value of the function f , we have Z ∈ L(hZi0id). Hence, Z ∈Pi L(hZiid) and L(hZiid). Xj L(hXjid) ⊆Xi This completes the proof. Now we define a decreasing filtration on V = hU1, . . . , Ukid in the following way. By definition, put F0V = V, Fn+1V = L(FnV ). (25) Since for every Y = hZ1, . . . , Zk, . . .id, where {Zi} is either a finite or in- finite set of monomials, we have description (22), therefore formula (25) is applicable for every n > 0. Here we mean L(0) = 0. Proposition 4.4. The filtration defined above has finitely many levels, that is, there exists a number N such that FN V = 0. Moreover, we never have a situation Fn+1V = FnV for FnV 6= 0. Proof. Recall that V = hU1, . . . , Ukid, U1, . . . , Uk ∈ F . We put Nmax = max (N(Ui)). i∈{1,...,k} Consider f -characteristics of monomials in V . Let U be an arbitrary mono- mial from V . From Lemma 4.1 it follows that N(U) 6 Nmax. Since every virtual member of the chart is of Λ-measure not less than τ −2ε, it is contained in every covering of U. Hence, we evidently obtain Kτ (U) 6 N(U) 6 Nmax. Therefore, there are finitely many different values of f -characteristics of monomials in V . Then is finite for any n such that FnV 6= 0. m(n) = max U ∈FnV f (U) 52 Recall that, by definition, Fn+1V = L(FnV ). Then from (20) and (21) it follows that m(n) > m(n+1) if Fn+1V 6= 0. So, we have a strictly decreasing sequence of f -characteristics m(0) > . . . > m(n) > . . . (26) that corresponds to the decreasing sequence of non-trivial spaces F0V ⊇ . . . ⊇ FnV ⊇ . . . . The sequence (26) can not be infinite, so, it ends up at some step n0. This means that Fn0V is the last non-trivial subspace of the filtration. That is, the filtration has finitely many non-zero levels. Assume Fn+1V = FnV for FnV 6= 0. That is, FnV = L(FnV ). Then by induction we have Fn+kV = FnV 6= 0 for any k ∈ N. But we already proved that the filtration has finitely many non-zero levels. This contradiction com- pletes the proof. Definition 4.3. Suppose Y is a subspace of Z2F linearly generated by an arbitrary set of monomials and closed under taking derived monomials. Ev- ery linear dependence from T ′ is, in fact, a linear dependence between a monomial U and its derived monomials. Hence, any multi-turn of a virtual member of the chart of a monomial from Y generates a linear dependence between the monomials from Y , because Y is closed under taking derived monomials. We consider the subspace of Y Uj Uh is a monomial from Y, Dp(Y ) =* kXj=1 Uh 7→ kXj=1 j6=h Uh = LahR, where ah is a virtual member of the chart of Uh, Uj is a multi-turn that comes from an elementary multi-turn ah 7→ aj+ kXj=1 j6=h and call this subspace the subspace of dependencies on Y . Using this notion, I = hT ′i = Dp(Z2F ). 53 Note that if a monomial U ∈ W has the empty chart, then there are no multi-turns of U. Hence, when W consists only of monomials with the empty chart, by definition we put Dp(W ) = 0. Since V is closed under taking derived monomials, we consider its sub- space Dp(V ). Evidently, for every monomial U the space L(hUid) is linearly generated by monomials and closed under taking derived monomials. There- fore, by the definition, every FnV is generated by monomials and closed under taking derived monomials. Hence, we can consider its subspace Dp(FnV ) and define the filtration on Dp(V ) in the following way FnDp(V ) = Dp(FnV ), that is, FnDp(V ) is the vector space generated by linear dependencies coming from monomials of FnV . Lemma 4.5. Suppose U is a monomial, U ∈ FnV . If U /∈ Fn+1V , then Fn+1V ∩ hUid = L(hUid). Proof. Let FnV = hZ1, . . . , Zk, . . .id, where {Zi} is either a finite or infinite set of monomials. Since U ∈ FnV and FnV is closed under taking derived monomials, we can assume that FnV = hU, Z1, . . . , Zk, . . .id. Since the defi- nition of Fn+1V = L(FnV ) does not depend on the set of generators of FnV (see Proposition 4.3), we have Fn+1V = L(FnV ) = L(hUid) +Xi L(hZiid). Let Z ∈ L(hUid). Then, using the last equality, we immediately obtain Z ∈ Fn+1V . So, L(hUid) ⊆ Fn+1V ∩ hUid. Assume a monomial Z ∈ Fn+1V ∩ hUid and Z /∈ L(hUid). Since U /∈ Fn+1V , we have U /∈ L(hZiid) for all i. Monomials form a basis of Z2F and every L(hZiid) is generated by monomials. So, since Z ∈ Fn+1V = Pi L(hZiid), we have Z ∈ L(hZi0id) for some i0. We assumed that Z ∈ hUid \ L(hUid), hence, by Lemma 4.2, hUid ⊆ L(hZi0id). But that contradicts the assumption that U /∈ Fn+1V . Therefore, Fn+1V ∩ hUid ⊆ L(hUid). Thus, finally we obtain Fn+1V ∩ hUid = L(hUid). In Definition 4.1. we defined derived monomials with the use of a special set of transformations (1) and (2). However, in the next lemma we will use 54 a slightly wider class of transformations. Let us prove that, using this wider class of transformations of a given monomial, we still will have its derived monomials. Namely, let U be a monomial, ah and bh be virtual members of the chart of U. We consider two replacements ah 7→ aj, bh 7→ bj (ah and aj are incident monomials, bh and bj are incident monomials). Assume that first we apply the transformation ah 7→ aj, namely, U = LaahRa 7→ LaajRa. To be precise, we suppose that the beginning of bh lies from the right of the beginning of ah. Denote by b′ h the intersection of bh and Ra. If ah and bh are separated or touch at a point, b′ h = bh. ah b′ h = bh U U ah b′ h = bh If ah and bh has an overlap c, then bh = cb′ h. U ah c b′ h Ra Ra Ra In any case, we can apply the transformation b′ j to LaajRa, where either b′ j = c−1bj if ah and j = bj if ah and bh are separated or touch at a point, or b′ bh has an overlap c. Here we mean that if aj = 1 and the monomial LaRa has cancellations, we do not perform them. Instead, we perform a replacement b′ h 7→ b′ j. h 7→ b′ Assume aj 6= 1 or aj = 1 and LaRa has no further cancellations. The h in LaajRa (see Definition 3.1) can be repre- h 7→ b′ j and the further cancellations if there are h is not a maximal occurrence of a generalized h is a virtual member of j in LaajRa and the further replacement of the image of b′ sented as the replacement b′ any (they may occur when b′ fractional power in LaajRa). So, if the image of b′ LaajRa, as a result of the transformation b′ h 7→ b′ cancellations, we obtain a derived monomial of U. 55 If aj is a virtual member of the chart of LaajRa, then, by Corollary 3.7, the h is always a virtual member of LaajRa. If aj is not virtual member h may not be a virtual member of h 7→ b′ j image of b′ of the chart of LajR, then the image of b′ LaajRa. The following lemma states that if we apply the replacement b′ to LaajRa in this case, we, nevertheless, obtain a derived monomial of U. Lemma 4.6. Let U be a monomial, ah and bh be virtual members of the chart of U. We consider two replacements ah 7→ aj, bh 7→ bj (ah and aj are incident monomials, bh and bj are incident monomials). Assume U = LaahRa, aj is not a virtual member of the chart of LaajRa, b′ h is an intersection of Ra and bh. Then the result of the replacement b′ h 7→ b′ j (corresponding to bh 7→ bj) in LaajRa is a derived monomial of U, possibly after the further cancellations. Moreover, its f -characteristics is strictly less than f (U). Proof. We will consider only the most interesting case when ah and bh has an overlap c. Two other cases are similar. U ah c b′ h First we consider the case when aj 6= 1. Then the resulting monomial LaajRa is reduced. We have the transformation bh 7→ bj. Since bh = cb′ h, we also h 7→ c−1bj. So, we have a sequence of replacements have the transformation b′ U = LaahRa 7→ LaajRa = Laajb′ hRb 7→ Laajc−1bjRb. (27) h) > τ . The image of b′ h in LaajRa prolongs b′ If Λ(bh) > τ + ε, then Λ(b′ h. Therefore, its Λ-measure is not less than the Λ-measure of b′ h. So, the image of b′ h in LaajRa is of Λ-measure > τ . Therefore, it is a virtual member of the chart of LaajRa and the result of (27) is a derived monomial of U, possibly after the further cancellations. Suppose Λ(bh) < τ +ε. Consider an elementary multi-turn bh 7→Ps where bj is one of the resulting monomials. In Proposition 3.1, we proved that there exists bk0 such that Λ(bk0) > τ . However, using the same argument, one can easily prove a slightly stronger statement, namely, that there exists bk0 such that Λ(bk0) > τ + ε (and, hence, k0 6= h). k=1 k6=h bk, Assume U = LbbhRb. Consider the transformation U = LbbhRb 7→ Lbbk0Rb. 56 h the intersection of ah and Lb in LbbhRb. Then ah = a′ Denote by a′ the image of ah in Lbbk0Rb be equal to a′ let us put bk0 = c′b′ from Corollary 3.7 it follows that a′ Lbbk0Rb. hc. Let hc′, where c′ is possibly empty. Also k0. Since bk0 is a virtual member of the chart of Lbbk0Rb, hc′ is a virtual member of the chart of Lb b′ k0 Rb a′ h c′ La hc′ 7→ a′ j, where a′ We have a transformation ah 7→ aj. Multiplying it by c−1c′ from the j = ajc−1c′. The monomials bk0 and bj right, we obtain a′ are incident, so, we have a transformation bk0 7→ bj. Multiplying it by c′−1 j = c′−1bj. Since Λ(bk0) > τ + ε, from the left, we obtain b′ clearly, Λ(b′ k0 is always a virtual member of the chart. Hence, we obtain a sequence of replacements of virtual members of the chart k0) > τ . Therefore, a prolongation of b′ j, where b′ k0 7→ b′ Lbbk0Rb = Laa′ hc′b′ k0Rb 7→ Laa′ jRb = Laajc−1c′c′−1bjRb = Laajc−1bjRb. jb′ 7→ Laa′ k0Rb jb′ (28) So, the results of (27) and of (28) are equal. But the result of sequence (28) is a derived monomial of U, by definition. Since aj is not a virtual member of the chart of LaajRa, from Lemma 3.10 it follows that a′ j is not a virtual member of the chart of Laa′ k0Rb. Hence, f (Laajc−1bjRb) < f (U). jb′ Consider the case when aj = 1. If there are no cancellations in LaRa, then we argue as above. Suppose there are cancellations in LaRa, namely, bh = cbh,mbh,f , La = L′ h,m, where bh,f is possibly empty. For simplicity, assume that L′ abh,f Rb has no further cancellations. ab−1 L′ a ah bh,m bh,f U b−1 h,m c h be an intersection of bh and Ra, that is, bh = cb′ As above, let b′ h = bh,mbh,f . We have a replacement bh 7→ bj and the corresponding replacement b′ h = c−1bh 7→ c−1bj. After the replacement LaahRa 7→ LaRa there are h, b′ 57 cancellations of the resulting monomial. Let us not perform them, instead, we perform the replacement b′ h 7→ c−1bj. So, we have a sequence of replacements U = LaahRa 7→ LaRa = Lab′ hRb = Lac−1bjRb. (29) of the resulting monomials. Multiplying it by b−1 Again consider an elementary multi-turn bh 7→Ps we obtain an elementary multi-turn bh,f 7→Ps bk, where bj is one h,mc−1 from the left side, b−1 h,mc−1bk. If Λ(bh,f ) > τ , then its image in L′ abh,f Rb is a virtual member or the chart. Then we again can argue as in the beginning of the proof. Assume Λ(bh,f ) < τ . Then, by Proposition 3.1, there exists bk1 such that Λ(b−1 h,mc−1bk1) > τ . k=1 k6=h k=1 k6=h We have the replacement ah 7→ 1. As above, let a′ ah and Lb, that is, ah = a′ possibly empty). That is, c′′ is an overlap of a′ Corollary 3.7 it follows that a′ hc. Let a′ h be an intersection of hc′′ be the image of ah in Lbbk1Rb (c′′ is k1. From hc′′ is a virtual member of the chart of Lbbk1Rb. hc′′ and bk1, bk1 = c′′b′ L′ a U a′ h b−1 h,m c′′ b′ k1 So, multiplying ah 7→ 1 by c−1c′′ from the right side, we obtain the replace- ment a′ hc′′ 7→ c−1c′′. The monomials bk1 and bj are incident, hence, we have a transformation bk1 7→ bj. So, we also have b−1 h,mc−1bk1) > τ its prolongation is always a virtual member of the chart. Therefore, we obtain the sequence of replacements of virtual members of the chart: h,mc−1bj. Since Λ(b−1 h,mc−1bk1 7→ b−1 U = LbbhRb 7→ Lbbk1Rb = Laa′ hc′′b′ k1Rb 7→ h,mc−1bk1Rb 7→ L′ 7→ Lac−1c′′b′ k1Rb = L′ ab−1 ab−1 h,mc−1bjRb = Lac−1bjRb. (30) So, the result of (29) is equal to the result of (30). But the result of se- quence (30) is a derived monomial of U, by definition. In (30) we had the hc′′ 7→ c−1c′′, where Λ(c−1c′′) 6 2ε. Hence, c−1c′′ can not be a replacement a′ virtual member of the chart. Therefore, f (Lac−1bjRb) < f (U). Lemma 4.7 (Main Lemma). Let V = hU1, . . . , Ukid, U1, . . . , Uk ∈ F . Then we have Dp(FnV ) ∩ Fn+1V = Dp(Fn+1V ). 58 Proof. We have Dp(FnV ) = Dp(Fn+1V ) + hT (n) 1 , . . . , T (n) k , . . .i, i where T (n) are linear dependencies coming from monomials of FnV \ Fn+1V . Evidently, Dp(FnV ) ∩ Fn+1V ⊇ Dp(Fn+1V ). So, we need to show that Dp(FnV ) ∩ Fn+1V ⊆ Dp(Fn+1V ). Since Dp(Fn+1V ) ⊆ Fn+1V , it is sufficient to prove that hT (n) 1 , . . . , T (n) k , . . .i ∩ Fn+1V ⊆ Dp(Fn+1V ). First let us show that it is sufficient to prove Lemma 4.7 only for the case when FnV is generated by one monomial and its derived monomials. Indeed, suppose Lemma 4.7 is proved for this case, then let us prove it for the general case. Assume T1, . . . Tl are linear dependencies coming from monomials of i=1 Ti ∈ Fn+1V . So, all monomials from FnV \ Fn+1V FnV \ Fn+1V andPl have to cancel out in this sum. Let Ti be generated by a multi-turn of a monomial Zi ∈ FnV \ Fn+1V , i = 1, . . . , l. We choose a subset {Zi1, . . . , Zik} ⊆ {Z1, . . . , Zl} such that hZi1, . . . , Zikid = hZ1, . . . , Zlid and hZij id * hZij′ id for j 6= j′. The de- pendencies {T1, . . . Tl} can be split into groups {T j lj }, j = 1, . . . k, of dependencies coming from monomials of hZij id. So, we have 1 , . . . T j Ti = lXi=1 kXj=1 ljXi=1 T j i ∈ Fn+1V. (31) Every linear dependence Ti comes from a monomial that belongs to FnV \ Fn+1V ; hence, by the definition of Fn+1V , every T j i comes from a mono- mial that belongs to hZij id \ L(hZij id). From Lemma 4.5 it follows that all monomials from hZij id \ L(hZij id) are contained in FnV \ Fn+1V and L(hZij id) ⊆ Fn+1V . Hence, the monomials from every hZij id \ L(hZij id), j = 1, . . . , k, have to cancel out in the sum (31). Since hZij id * hZij′ id for j 6= j′, from Lemma 4.2 it follows that if a monomial Z belongs to hZij id \ L(hZij id), then Z /∈ hZij′ id for j 6= j′. There- fore, the monomials from FnV \ Fn+1V in the sum (31) have to cancel out in i ∈ L(hZij id). Since we assumed that the statement of the lemma holds for the spaces hZij id, we i separately. Hence, we obtainPlj every sumPlj i=1 T j i=1 T j 59 T j i ∈ Dp(L(hZij id)) ⊆ Dp(Fn+1V ). Ti = T j i ∈ Dp(Fn+1V ). kXj=1 ljXi=1 have Thus, ljXi=1 lXi=1 So, it remains to prove the lemma only for the case when FnV is generated by one monomial and its derived monomials. We start with proving the following statement. Lemma 4.8. Suppose A1, . . . , Ak are vector spaces, L(Ai) ⊆ Ai is a subspace of Ai. Suppose Di is a subspace of Ai such that Di ∩ L(Ai) = 0. Consider A1 ⊗ . . . ⊗ Ak and its subspaces A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ Ak. Define the subspaces L(A1 ⊗ . . . ⊗ Ak) = kXi=1 A1 ⊗ . . . ⊗ L(Ai) ⊗ . . . ⊗ Ak, and L(A1⊗. . .⊗Di⊗. . .⊗Ak) = kXi′=1 i6=i′ A1⊗. . .⊗Di⊗. . .⊗L(Ai′)⊗. . .⊗Ak, i = 1, . . . , k. Then kXi=1 A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ Ak!\ L(A1⊗. . .⊗Ak) = kXi=1 L(A1⊗. . .⊗Di⊗. . .⊗Ak). Proof. From the condition Di ∩ L(Ai) = 0 it follows that the sum of this subspaces is a direct sum Di ⊕ L(Ai) ⊆ Ai. Since Ai is a vector space, the subspace Di ⊕ L(Ai) has a direct complement Ji and we obtain Ai = Ji ⊕ Di ⊕ L(Ai). Hence, L(A1 ⊗ . . . ⊗ Ak) = = kXi=1 kXi=1 (J1 ⊕ D1 ⊕ L(A1)) ⊗ . . . ⊗ L(Ai) ⊗ . . . ⊗ (Jk ⊕ Dk ⊕ L(Ak)) Bj 1 ⊗ . . . ⊗ Bj i−1 ⊗ L(Ai) ⊗ Bj i+1 ⊗ . . . ⊗ Bj k), (Mj 60 where every Bj is either Ji, or Di, or L(Ai) and we encounter all possible i combinations in the sum. Tensor products that contain more then one mem- ber L(Ai) repeat in the sum. If we take every repeating space only one time, k, where every is either Ji, or Di, or L(Ai), we encounter all combinations in the sum, Bj i such that at least one Bj we obtain the direct sum L(A1 ⊗ . . . ⊗ Ak) =Lj Bj kXi=1 A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ Ak i = L(Ai). Similarly, 1 ⊗ . . . ⊗ Bj (J1 ⊕ D1 ⊕ L(A1)) ⊗ . . . ⊗ Di ⊗ . . . ⊗ (Jk ⊕ Dk ⊕ L(Ak)) = = kXi=1 kXi=1 C j (Mj 1 ⊗ . . . ⊗ C j i−1 ⊗ Di ⊗ C j i+1 ⊗ . . . ⊗ C j k), where every C j is either Ji, or Di, or L(Ai) and we encounter all possible i combinations in the sum. Tensor products that contain more then one mem- ber Di repeat in the sum. If we take every repeating space only one time, 1 ⊗ . . . ⊗ C j k, is either Ji, or Di, or L(Ai), we encounter all combinations we obtain the direct sumPk kXi=1 i=1 A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ Ak =Lj C j A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ Ak) ∩ L(A1 ⊗ . . . ⊗ Ak! =Mj where every C j i in the sum, such that at least one C j 1 ⊗ . . . ⊗ K j k, i = Di. Hence, K j where every K j i in the sum, such that at least one K j Therefore, is either Ji, or Di, or L(Ai), we encounter all combinations i = L(Ai). i = Di and at least one K j kXi=1 A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ Ak) ∩ L(A1 ⊗ . . . ⊗ Ak! = = kXi=1 kXi=1 A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ L(Ai′) ⊗ . . . ⊗ Ak kXi′=1 i6=i′ L(A1 ⊗ . . . ⊗ Di ⊗ . . . ⊗ Ak). 61 Using Lemma 4.8, we continue the proof of Lemma 4.7. Assume U is a monomial. Derived monomials of U are defined with the use of certain sequences of replacements of virtual members of the chart. When we perform replacements that preserve f -characteristics of monomials, they preserve, roughly speaking, the structure of the chart. Moreover, there is no interaction between the replaced occurrence and the separated virtual members of the chart and there is a very small interaction between the replaced occurrence and its neighbours. This kind of behaviour provides the idea to consider a tensor product of linear spaces that correspond to each place of the chart of U. Assume the monomial U has k virtual members of the chart, that is, Kτ (U) = k. Let a(i) be the virtual member of the chart of U placed on the i-th position from the beginning of U. Let Ai[U] be a subspace of Z2F such that a(i) and a(i) j j are incident monomials, j ∈ NE . (32) Suppose U = Lia(i)Ri. We define a subspace Li[U] ⊆ Ai[U] by the following rule: a(i) and a(i) j are incident monomials, j ∈ N, (33) Ai[U] =Da(i) Li[U] =Da(i) j Lia(i) j Ri ∈ L(hUid)E . Remark 4.2. Suppose U ′ is a derived monomial of U such that U ′ ∈ hUid \ L(hUid). Let us construct the spaces Ai[U ′] and Li[U ′] ⊆ Ai[U ′] cor- responding to U ′ as above. Then, obviously, the spaces Ai[U ′] are generated by the same sets of monomials as the spaces Ai[U] up to shifts of the initial and the final points in the corresponding v-diagram by ε. Moreover, using Lemma 3.10, one can prove that the spaces Li[U ′] are also generated by the same sets of monomials as the spaces Li[U] up to shifts of the initial and the final points in the corresponding v-diagram by ε. Although the precise forms of the spaces Ai[U ′] and Li[U ′] ⊆ Ai[U ′] de- pend on the monomial U, in the sequel, we will omit [U] in the denotation of this spaces when it does not lead to ambiguity. We construct a linear mapping µ[U] : A1 ⊗ . . . ⊗ Ak → hUid. 62 Elements b(1) ⊗ . . . ⊗ b(k) ∈ A1 ⊗ . . . ⊗ Ak, where b(i) ∈ Ai are generalized fractional powers incident to a(i), form a basis of A1 ⊗ . . . ⊗ Ak, because generalized fractional powers incident to a(i) form a basis of Ai. We will define µ[U] on these basis elements. We distinguish between four possibilities: Case 1 First we define µ[U] on elements b(1)⊗. . .⊗b(k) such that all b(i) ∈ Ai\Li. It encodes a sequence of replacements that starts from the monomial U. Recall that a(1), . . . , a(k) are the virtual members of the chart of U enumerated from left to right. Let U = L1a(1)R1. Then we start with a(2), where Λ(c(2)) 6 ε. The next step is the replacement of the transformation L1a(1)R1 7→ L1b(1)R1. Denote byba(2) the image of ba(2) in L1b(1)R1. Since Λ(b(1)) > ε, eitherba(2) = a(2), orba(2) = c(2)a(2), or ba(2) = c(2)−1 ba(2) in L1b(1)R1 corresponding to the transformation a(2) 7→ b(2). That is, the replacement ba(2) 7→bb(2) in L1b(1)R1, where eitherbb(2) = b(2) if ba(2) = a(2), or bb(2) = c(2)b(2) if ba(2) = c(2)a(2), or bb(2) = c(2)−1 ba(2) = c(2)−1 b(2) if In the same way, we continue transformations for every position in the chart. Notice that, by Corollary 3.11, we actually can perform the replacements in any order. a(2). Case 2 Assume 1 6 i0 6 k is a place in the chart of U. We define µ[U] on elements b(1) ⊗ . . . ⊗ b(k) such that b(i0) ∈ Li0 and b(i) ∈ Ai \ Li for i 6= i0. First we replace virtual members of the chart of U in positions different from i0 as we described above. Then we perform a correspond- ing replacement in the position i0 and all the further cancellations if necessary. Case 3 Assume 1 6 i′ 0 < i0 6 n are two places in the chart of U. We define µ[U] on elements b(1) ⊗ . . . ⊗ b(k) such that b(i) ∈ Ai \ Li for i 6= i′ 0, i0 and b(i′ 0, b(i0) ∈ Li0. First we replace virtual members of the chart of U in positions different from i′ 0, i0 as we described above. Then we perform the corresponding replacement in the position i0. Assume 0) ∈ Li′ that as a result we obtain a monomial Lbb(i0)R. If there are any further cancellations in Lbb(i0)R (when bb(i0) = 1), we do not perform them Instead we perform the corresponding right after the replacement. replacement in the subword L or R in the place corresponding to i′ 0. Then we perform the cancellations if there are any. From Lemma 3.9 it follows that we can perform two last transforma- tions starting from any position i′ 0 or i0 and obtain the same final 63 result. From Lemma 4.6 it follows that the resulting monomial is a derived monomial U. Case 4 For elements b(1) ⊗. . . ⊗b(k) such that there are more than two b(i) ∈ Li, we could continue in a similar way. But, in fact, we do not need to preserve full information about these elements. So, by definition, we put µ[U](b(1) ⊗ . . . ⊗ b(k)) = 0 in this case. From Corollary 3.11, it follows that in Case 1 an element µ[U](b(1) ⊗ . . . ⊗ b(k)) ∈ hUid \ L(hUid). From Lemma 4.1 and Lemma 4.6, it follows that in Case 2 and Case 3 an element µ[U](b(1) ⊗ . . . ⊗ b(k)) ∈ L(hUid). Let Z be a derived monomial of U such that Z does not belong to L(hUid). By Lemma 4.1, this means that there exists a sequence of replace- ments of virtual members of the chart such that every replacement preserves f -characteristics of monomials. From Corollary 3.11, it follows that the re- placements can be performed in any order and after changing the order every replacement still preserves f -characteristics of monomials. Namely, we can first perform all replacements in the first position of the chart, then in the members of the chart of Z enumerated from left to right. Then the element second position, etc, in the k-th position. Assumebb(1), . . . ,bb(k) are the virtual bb(i) may not be incident to a(i) (where a(i) is the corresponding member of an incident monomial of a(i), corresponding tobb(i). Then the chart of U). But, clearly, its ends differ only by a shift by ε. Let b(i) be Z = µ[U](b(1) ⊗ . . . ⊗ b(k)). From Corollary 3.11 it follows that every replacement can be moved to the beginning of the sequence, which starts from U, and it still preserves the f -characteristic of the monomials. That is, every replacement a(i) 7→ b(i) in the monomial U does not decrease the f -characteristic of the resulting monomial. Hence, b(i) ∈ Ai \ Li. Suppose b(1) ⊗ . . . ⊗ b(k) and d(1) ⊗ . . . ⊗ d(k) are different elements such that b(i), d(i) are generalized fractional powers, b(i), d(i) ∈ Ai \ Li. Since there exists b(i0) 6= d(i0) and two different incident generalized fractional powers can not differ only by a piece equal to a possible overlap, we obtain µ[U](b(1) ⊗ . . . ⊗ b(k)) 6= µ[U](d(1) ⊗ . . . ⊗ d(k)). Thus, we have showed that µ[U] gives a bijective correspondence between elements b(1) ⊗ . . . ⊗ b(k) such that generalized fractional powers b(i) ∈ Ai \ Li and monomials from hUid \ L(hUid). 64 As above, we denote by Dp(Ai) the subspace of Ai generated by supports of corresponding elementary multi-turns (see Definition 4.3). Recall that Ai is generated by generalized fractional powers incident to one generalized fractional power a(i). Therefore, from the definition of an elementary multi- turn, it follows that a sum of two supports from Dp(Ai) is again the support of elementary multi-turn. So, actually Dp(Ai) consists precisely of supports of elementary multi-turns. Recall that a maximal occurrence of a generalized fractional power of Λ-measure > τ is always a virtual member of the chart. Hence, applying together (33) and Proposition 3.1, we obtain Dp(Ai)∩Li = 0. In order to prove Lemma 4.7 it remains to show that Dp(hUid) ∩ L(hUid) = Dp(L(hUid)). (34) We already proved that Dp(hUid)∩L(hUid) ⊇ Dp(L(hUid)). So, we will show that Dp(hUid) ∩ L(hUid) ⊆ Dp(L(hUid)). Suppose Ts, s = 1, . . . , m, are the supports of multi-turns coming from monomials of hUid \ L(hUid) and Ts ∈ L(hUid). mXs=1 Assume Ts comes from a monomial U (s) an element b(1) h ⊗ . . . ⊗ b(k) h ∈ A1 ⊗ . . . ⊗ An such that h ∈ hUid \ L(hUid). Then there exists µ[U](b(1) hs ⊗ . . . ⊗ b(k) hs ) = U (s) h , hs are generalized fractional powers such that b(i) where b(i) hs /∈ Li. Assume this multi-turn comes from an elementary multi-turn of the virtual member of the chart of U (s) hs ∈ Ais corresponds to this virtual member of the chart. By the definition of µ[U] in Case 1 and Case 2, we obtain h placed on the is-th position. The element b(is) Ts = µ[U](b(1) hs ⊗ . . . ⊗ t(is) s ⊗ . . . ⊗ b(k) hs ), where t(is) s ∈ Dp(Ais) is the support of the corresponding elementary multi- turn of b(is) hs . So, we have Ts = µ[U] mXs=1 mXs=1 b(1) hs ⊗ . . . ⊗ t(is) s ⊗ . . . ⊗ b(k) hs! . 65 By the assumption, all monomials in the left-hand sum that do not belong to L(hUid) cancel out. Since µ[U] gives a bijective correspondence between the elements b(1) ⊗ . . . ⊗ b(k) such that generalized fractional powers b(i) ∈ Ai \ Li and the monomials from hUid \ L(hUid), we obtain that all such elements cancel out in the right-hand sum as well. Then from Lemma 4.8 it follows that s ⊗. . . ⊗b(k) hs ∈ kXi=1 kXi′=1 i6=i′ A1 ⊗. . .⊗Dp(Ai)⊗. . . ⊗Li′ ⊗. . . ⊗Ak, s=1 b(1) hs ⊗ . . . ⊗ t(is) s ⊗ . . . ⊗ b(k) hs is equal to a sum of elements b(1) hs ⊗. . . ⊗t(is) mXs=1 that is, the sumPm 0−1) ⊗ed(i′ d(1) ⊗ . . . ⊗ d(i′ of the form 0) ⊗ d(i′ 0+1) ⊗ . . . ⊗ d(i0−1) ⊗et(i0) ⊗ d(i0+1) ⊗ . . . ⊗ d(k), 0+1) ⊗ . . . ⊗ d(k) 0) ⊗ d(i′ 0, where generalized fractional powers d(i) ∈ Ai \ Li, a 0, and the support of an elementary 0) ∈ Li′ for different i0 6= i′ d(1) ⊗ . . . ⊗ d(i0−1) ⊗ed(i0) ⊗ d(i0+1) ⊗ . . . ⊗ d(i′ 0−1) ⊗et(i′ generalized fractional power ed(i′ multi-turnet(i0) ∈ Dp(Ai0). To be precise, assume i′ Assumeet(i0) =Pm0 . Let us calculate j=1 e(i0) j 0 < i0. µ[U](d(1) ⊗ . . . ⊗ed(i′ 0) ⊗ . . . ⊗ e(i0) j ⊗ . . . ⊗ d(k)). m0Xj=1 From Lemma 3.9 and Corollary 3.11, it follows that in the definition of µ[U] first we can do replacements of all positions of the chart except i0 and i′ 0, then a replacement in the position i0 and then a replacement in the position corresponding to i′ ∈ Li0). Suppose 0 (the exact position may shift if e(i0) j Z is a result after the replacement in the position i0. Then Z = Lbe(i0)R, wherebe(i0) may differ from e(i0) by shifts of its ends by ε. For simplicity, we illustrate the case when the virtual members of the chart on the positions i′ 0 and i0 are separated, but the other cases are analogous to this one. 0) ba(i′ L be(i0) R 66 Then the last replacement in Z can be represented as a replacement of the corresponding maximal occurrence in L and the further cancellations if there are any. Denote the result of the transformation of L by L′. (i′ 0) bed L′ be(i0) R Hence, we obtain µ[U](d(1) ⊗ . . . ⊗ed(i′ where the word L′bej µ[U](d(1) ⊗ . . . ⊗ed(i′ (i0)R, 0) ⊗ . . . ⊗ 0) ⊗ . . . ⊗ e(i0) m0Xj=1 e(i0) j ⊗ . . . ⊗ d(k)) = (i0)R is possibly non-reduced. So, j ⊗ . . . ⊗ d(k)) = L′bej m0Xj=1 L′be(i0) = L′ m0Xj=1be(i0) j ! R. is the support of an elementary multi-turn, the sumPm0 j=1be(i0) j R j is the support of an elementary multi-turn as well (possibly with shifts of the ends of the monomials by ε). Therefore, from Proposition 3.2 it follows that j R is the support of a multi-turn (after the cancellation in the j j=1 e(i0) monomials). SincePm0 Pm0 j=1 L′be(i0) Since ed(i′ µ[U](d(1) ⊗ . . . ⊗ed(i′ Then, since µ[U](d(1) ⊗ . . . ⊗ed(i′ of a multi-turn, we have 0) ∈ L(Ai), we have µ[U](d(1) ⊗ . . . ⊗ed(i′ Thus, 0) ⊗ . . . ⊗ 0) ⊗ . . . ⊗ e(i0) j ⊗ . . . ⊗ d(k)) ∈ L(hUid). 0) ⊗ . . . ⊗ e(i0) j ⊗ . . . ⊗ d(k)) is the support m0Pj=1 e(i0) j ⊗ . . . ⊗ d(k)) ∈ Dp(L(hUid)). m0Xj=1 and the equality (34) holds. This concludes the proof of Lemma 4.7. Ts ∈ Dp(L(hUid)) mXs=1 67 Using Lemma 4.7, we obtain the following proposition. Proposition 4.9. Suppose X, Y are subspaces of Z2F generated by monomi- als and closed under taking derived monomials, Y ⊆ X. Then Dp(X) ∩ Y = Dp(Y ). Proof. Clearly, Dp(Y ) ⊆ Dp(X) ∩ Y . Let us show that Dp(X) ∩ Y ⊆ Dp(Y ). Suppose T1 + . . . + Tm ∈ Dp(X) ∩ Y , where every Ti, i = 1, . . . m, belongs to the set of generating supports of multi-turns of Dp(X). Consider Ti0 ∈ Y . Since Y is generated by monomials, we obtain that every monomial of Ti0 belongs to Y . So, Ti0 is a linear dependence generated by a monomial from Y , that is, Ti0 ∈ Dp(Y ). So, we may suppose that every Ti /∈ Y , i = 1, . . . , m. Denote by X ′ the subspace of X generated by the monomials of Ti, i = 1, . . . , m, and their derived monomials. Suppose the filtration on X ′ has N non-zero levels. Let us prove that T1 + . . . + Tm ∈ Dp(Y ) by induction on N. First let us do the step of induction. Assume Ti is generated by a multi- turn of a monomial Zi. If Zi ∈ L(X ′), then all the monomials of Ti belong to L(X ′) because L(X ′) is closed under taking derived monomials. Assume Zi ∈ X ′ \ L(X ′). Let Z be an arbitrary monomial of Ti such that Z ∈ X ′ \ L(X ′). Since Z ∈ X ′ \ L(X ′), we have Z ∈ hZiid \ L(hZiid). Therefore, from Lemma 4.2 it follows that Zi is a derived monomial of Z. Assume Z ∈ Y , then all its derived monomials belong to Y because Y is closed under taking derived monomials. So, we have Zi ∈ Y ; hence, all the monomials of Ti are contained in Y . This contradicts our assumption that Ti /∈ Y . Therefore, the monomials of Ti that are contained in X ′ \ L(X ′) are not contained in Y . Since T1 + . . . + Tm ∈ Y , the monomials that are contained in X ′ \ L(X ′) cancel in the sum T1 + . . . + Tm; hence, T1 + . . . + Tm ∈ L(X ′). Then from Lemma 4.7 it follows that T1 + . . . + Tm ∈ Dp(L(X ′)). Therefore, T1 + . . . + Tm = T ′ are supports of multi-turns that come from monomials of L(X ′). We have i ∈ L(U ′), i = 1, . . . , m′, m′, where T ′ 1 + . . . + T ′ T1 + . . . + Tm = T ′ 1 + . . . + T ′ m′ = XT ′ i ∈Y T ′ i + XT ′ i /∈Y T ′ i . 68 i belongs to Dp(Y ). The space L(X ′) has N − 1 non-zero levels of the filtration; hence the second sum i ∈Y T ′ i belongs to Dp(Y ) by the induction hypothesis, and therefore, As above, every element of the first sum PT ′ PT ′ i /∈Y T ′ T1 + . . . + Tm ∈ Dp(W ). Let us prove the basis of induction. Consider N = 1. As above, we obtain T1 + . . . + Tm ∈ L(X ′). But since N = 1, we have L(X ′) = 0; therefore T1 + . . . + Tm = 0, and so, it belongs to Dp(Y ). Thus, we obtain Dp(X) ∩ Y ⊆ Dp(Y ). This concludes the proof. Using the mapping µ[U] that was constructed in the second part of the proof of Lemma 4.7, we obtain the following statement. Proposition 4.10. Let U be a monomial with k virtual members of the chart. Suppose Ai and Li ⊆ Ai, i = 1, . . . k, are defined above by (32) and (33) subspaces of Z2F corresponding to U. Then we have hUid/L(hUid) ∼= A1/L1 ⊗ . . . ⊗ Ak/Lk. Moreover, hUid/(Dp(hUid) + L(hUid)) ∼= A1/(Dp(A1) + L1) ⊗ . . . ⊗ Ak/(Dp(Ak) + Lk). Proof. Recall that in the proof of Lemma 4.7 we constructed a linear mapping µ[U] : A1 ⊗ . . . ⊗ Ak → hUid (see page 62). We define a linear mapping µ1[U] : A1/L1 ⊗ . . . ⊗ Ak/Lk → hUid/L(hUid) by the following rule: µ1[U]((b(1) + L1) ⊗ . . . ⊗ (b(k) + Lk)) = µ[U](b(1) ⊗ . . . ⊗ b(k)) + L(hUid), where b(i) ∈ Ai are generalized fractional powers. In Lemma 4.7, we proved (using Lemma 4.1 and Lemma 4.6) that µ[U] kXi=1 A1 ⊗ . . . ⊗ Li ⊗ . . . ⊗ Ak! ⊆ L(hUid). (35) 69 Hence, the mapping µ1[U] is well-defined. Since µ[U] gives a bijective corre- spondence between elements b(1) ⊗ . . . ⊗ b(k), where b(i) are generalized frac- tional powers such that b(i) ∈ Ai \Li, and the monomials from hUid \L(hUid), the mapping µ1[U] is bijective. So, µ1[U] is an isomorphism of linear spaces. In the same way, we define a linear mapping µ2[U] : A1/(Dp(A1)+L1)⊗. . .⊗Ak/(Dp(Ak)+Lk) → hUid/(Dp(hUid)+L(hUid) by the following rule: µ2[U]((b(1) + Dp(A1) + L1) ⊗ . . . ⊗ (b(k) + Dp(Ak) + Lk)) = µ[U](b(1) ⊗ . . . ⊗ b(k)) + Dp(hUid) + L(hUid), (36) where b(i) ∈ Ai are generalized fractional powers. Arguing in the same way as in Lemma 4.7, it is easy to show that µ[U] kXi=1 A1 ⊗ . . . ⊗ Dp(Ai) ⊗ . . . ⊗ Ak! ⊆ Dp(hUid) + L(hUid). Using this together with (35), we see that the mapping µ2[U] is well-defined. Since µ[U] gives a bijective correspondence between elements b(1) ⊗ . . . ⊗ b(k), where b(i) are generalized fractional powers such that b(i) ∈ Ai \ Li, and the monomials from hUid \ L(hUid), one can show that the mapping µ2[U] is bijective. So, µ2[U] is an isomorphism of linear spaces. Remark 4.3. In Proposition 4.10, in fact, we used the following construc- tion. We consider a composition of linear mappings A1 ⊗ . . . ⊗ Ak A1 ⊗ . . . ⊗ Ak µ[U ] −→ hUid µ[U ] −→ hUid π1−→ hUid/L(hUid), π2−→ hUid/(Dp(hUid) + L(hUid)), where π1 and π2 are the canonical homomorphisms. Then, by properties of µ[U] established in Lemma 4.7, ker(π1 ◦ µ[U]) = ker(π2 ◦ µ[U]) = A1 ⊗ . . . ⊗ Li ⊗ . . . ⊗ Ak, A1 ⊗ . . . ⊗ Li ⊗ . . . ⊗ Ak + kXi=1 kXi=1 Then, using the isomorphism theorem and the definition of a tensor product of vector spaces, we obtain the result of Proposition 4.10. A1 ⊗ . . . ⊗ Dp(Ai) ⊗ . . . ⊗ Ak. kXi=1 70 4.2 The structure of quotient spaces hU1, . . . , Ukid/Dp(hU1, . . . , Ukid) The following statement easily follows from Proposition 4.9. Corollary 4.11. Let V = hU1, . . . , Ukid, where U1, . . . , Uk are monomi- als. Let bV be the corresponding subspace in Z2F /hT ′i, that is, bV = (V + hT ′i)/hT ′i. Then bV ∼= V /Dp(V ). Proof. From the Isomorphism Theorem, it follows that bV ∼= V /(V ∩ hT ′i). bV ∼= V /Dp(V ). According to Definition 4.3, we have hT ′i = Dp(Z2F ). Therefore, from Proposition 4.9 it follows that V ∩ hT ′i = V ∩ Dp(Z2F ) = Dp(V ); hence, The quotient space V /Dp(V ) inherits the filtration from V , Fn(V /Dp(V )) = (FnV + Dp(V ))/Dp(V ). Suppose V has N non-zero levels of the filtration V = F0V ⊇ F1V ⊇ . . . ⊇ FN −1V ⊇ FN V = 0. We have the corresponding graded space Gr(V /Dp(V )) = (Fn(V /Dp(V ))/Fn+1(V /Dp(V ))). N −1Mn=0 For any n = 0, . . . , N − 1, we have the mapping grn : FnV → FnV /Fn+1V . The following theorem establishes the compatibility of the filtration and the corresponding grading with the linear dependencies on the space V . Theorem 4.1. Let V = hU1, . . . , Ukid, U1, . . . , Uk ∈ F . Then where Gr(V /Dp(V )) ∼= N −1Mn=0 Grn(V )/grn(Dp(FnV )), Grn(V ) = FnV /Fn+1V, grn(Dp(FnV )) = (Dp(FnV ) + Fn+1V )/Fn+1V. 71 Proof. Using isomorphism theorems, we obtain Fn(V /Dp(V ))/Fn+1(V /Dp(V )) ∼= (FnV + Dp(V ))/(Fn+1V + Dp(V )) = (FnV + Fn+1V + Dp(V ))/(Fn+1V + Dp(V )) ∼= FnV /(FnV ∩ (Fn+1V + Dp(V ))) = FnV /(Fn+1V + FnV ∩ Dp(V )). Hence, Gr(V /Dp(V )) = FnV /(Fn+1V + FnV ∩ Dp(V )). On the other hand, we have GrnV /grn(Dp(FnV )) = N −1Mn=0 ∼= N −1Mn=0 N −1Mn=0 N −1Mn=0 (FnV /Fn+1V )/((Dp(FnV ) + Fn+1V )/Fn+1V ) FnV /(Dp(FnV ) + Fn+1V ). From Corollary 4.9 it follows that FnV ∩ Dp(V ) = Dp(FnV ). Consequently, GrnV /grn(Dp(FnV )) ∼= Gr(V /Dp(V )). N −1Mn=0 Proposition 4.12. Let V = hU1, . . . , Ukid, U1, . . . , Uk ∈ F . Assume FnV is non-zero subspace of the filtration, FnV = hZ1, . . . , Zs, . . .id, where {Zi}i∈I is either a finite or infinite set of monomials, and let Vi = hZiid. Then (1) there exists a subset of indices I ′ ⊆ I such that FnV /(Dp(FnV ) + Fn+1V ) ∼=Mi∈I ′ Vi/(Dp(Vi) + L(Vi)), where Vi * Fn+1V and the spaces Vi, i ∈ I ′ are pairwise different; (2) if Z ∈ FnV is a monomial such that Z /∈ Fn+1V , then Z belongs to precisely one space Vi, i ∈ I ′, and the isomorphism acts as follows Z + Dp(FnV ) + Fn+1V 7→ (0, . . . , 0, Z + Dp(Vi) + L(Vi), 0, . . .). 72 Proof. Since FnV = hZ1, . . . , Zs, . . .id, FnV =P∞ many Vi and FnV = Pm i=1 Vi if there are infinitely i=1 Vi otherwise. In what following we only discuss the case of infinitely many Vi, i ∈ I. The other case is similar. So, we have FnV /Fn+1V = ∞Xi=1 ((Vi + Fn+1V )/Fn+1V ) . Let us choose a special subset of {Vi}i∈I. We take all the different spaces from {Vi}i∈I and remove Vi such that Vi * Fn+1V . Since FnV 6= 0, from Proposition 4.4 it follows that FnV 6= Fn+1V . Hence, we do not remove all Vi in this process and obtain a non-empty set {Vi}i∈I ′ such that FnV /Fn+1V =Xi∈I ′ ((Vi + Fn+1V )/Fn+1V ) . Assume Vj is one of the spaces {Vi}i∈I ′, that is, j ∈ I ′, and ((Vj + Fn+1V )/Fn+1V )\Xi6=j i∈I ′ ((Vi + Fn+1V )/Fn+1V ) 6= 0. (37) Since every Vi, i ∈ I ′, and Fn+1V are generated by monomials, it follows from (37) that there exist monomials Z ′ j /∈ Fn+1V and a space Vi0, i0 ∈ I ′, i0 6= j, such that Z ′ j /∈ Fn+1V , we have Z ′ j /∈ L(Vi0). Then from Lemma 4.2 it follows that Vj = Vi0. But this contradicts the assumption that all spaces Vi, i ∈ I ′, are pairwise different. Hence, j ∈ Vj such that Z ′ j ∈ Vi0. Since Z ′ j /∈ L(Vj) and Z ′ ((Vj + Fn+1V )/Fn+1V )\Xi6=j i∈I ′ ((Vi + Fn+1V )/Fn+1V ) = 0. Thus, we obtain a direct sum FnV /Fn+1V =Mi∈I ′ ((Vi + Fn+1V )/Fn+1V ) . (38) Let Z ∈ FnV be a monomial, Z /∈ Fn+1V . Notice that, by the same argument as above, we obtain that Z belongs to precisely one of the spaces Vi, i ∈ I ′. 73 Uj, where Uh ∈ FnV \ Fn+1V . Hence, Uh belongs to a space Vi0, i0 ∈ I ′. Then every Uj also belongs to Vi0 and T ∈ Dp(Vi0). Therefore, we have Let T be the support of a multi-turn Uh 7→Pk (Dp(FnV ) + Fn+1V )/Fn+1V =Mi∈I ′ ((Dp(Vi) + Fn+1V )/Fn+1V ) . j=1 j6=h Hence, FnV /(Dp(FnV ) + Fn+1V ) ∼= (FnV /Fn+1V )/ ((Dp(FnV ) + Fn+1V )/Fn+1V ) ((Vi + Fn+1V )/Fn+1V )! / Mi∈I ′ (39) ((Dp(Vi) + Fn+1V )/Fn+1V )! ((Vi + Fn+1V )/Fn+1V )/((Dp(Vi) + Fn+1V )/Fn+1V ) (Vi + Fn+1V )/(Dp(Vi) + Fn+1V ). = Mi∈I ′ ∼=Mi∈I ′ ∼=Mi∈I ′ By the Isomorphism Theorem, we obtain (Vi + Fn+1V )/(Dp(Vi) + Fn+1V ) = (Vi + Dp(Vi) + Fn+1V )/(Dp(Vi) + Fn+1V ) (40) ∼= Vi/((Dp(Vi) + Fn+1V ) ∩ Vi) = Vi/(Dp(Vi) + Fn+1V ∩ Vi) Since Vi, i ∈ I ′, is not contained in Fn+1V , from Lemma 4.5 it follows that Fn+1V ∩ Vi = L(Vi). Thus, from (39) and (40) it follows that FnV /(Dp(FnV ) + Fn+1V ) ∼=Mi∈I ′ So, the first statement is proved. Vi/(Dp(Vi) + L(Vi)). Assume Z ∈ FnV is a monomial, Z /∈ Fn+1V . We noticed above that Z belongs to one space Vi0, i0 ∈ I ′. Then from the definition of the canon- ical isomorphisms in (39) and (40), it easily follows that the final isomor- phism maps Z + Dp(FnV ) + Fn+1V to the corresponding quotient space Vi0/(Dp(Vi0) + L(Vi0)), namely, Z + Dp(FnV ) + Fn+1V 7→ (0, . . . , 0, Z + Dp(Vi0) + L(Vi0), 0, . . .). So, the second statement is proved. 74 5 Description of a basis in Z2F/I Let us enumerate all the monomials from F , namely, F = {U1, U2, . . . , Uk, . . .}. Consider an increasing sequence of subspaces of Z2F hU1id ⊆ hU1, U2id ⊆ . . . ⊆ hU1, . . . , Ukid ⊆ . . . Clearly, the union of all these subspaces gives the whole ring Z2F . Consider the corresponding increasing sequence of subspaces of Z2F /hT ′i = Z2F /I (hU1id + hT ′)/hT ′ii ⊆ (hU1, U2id + hT ′i)/hT ′i ⊆ . . . (41) ⊆ (hU1, . . . , Ukid + hT ′i)/hT ′i ⊆ . . . . The union of all these subspaces gives the whole ring Z2F /hT ′i. By Corol- lary 4.11, we obtain (hU1, . . . , Ukid + hT ′i)/hT ′i ∼= hU1, . . . , Ukid/Dp(hU1, . . . , Ukid). In Section 5.2, we will construct a basis Bk in every subspace (hU1, . . . , Uki+ hT ′i)/hT ′i such that we obtain the increasing sequence B1 ⊆ B2 ⊆ . . . ⊆ Bk ⊆ . . . . Since the union of the subspaces (41) gives the whole ring Z2F /hT ′i, the unionSk Bk is a basis of Z2F /hT ′i = Z2F /I. As above, assume V = hU1, . . . , Ukid, U1, . . . , Uk ∈ F . Let the filtration on the space V have N non-zero levels. Putting together the results from Section 4.2, we obtain (V + hT ′i)/hT ′i ∼= N −1Mn=0Mi∈I ′ n V (n) i /(Dp(V (n) i ) + L(V (n) i )), id, Z (n) i i = hZ (n) where V (n) In Section 5.1 we will characterize non-trivial quotient spaces V (n) In Section 5.2 we will construct a basis in every space V (n) )) and, using this, we will explicitly describe a basis in the whole ring Z2F /I. is a monomial from FnV . ) + L(V (n) /(Dp(V (n) )+L(V (n) /(Dp(V (n) i ). i i i i i i 75 5.1 λ-semicanonical words Fix a constant 1 2 ≪ λ ≪ 1 (≪ on the ε-scale). For example, one can use λ > 2 3. We introduce λ-forbidden words. Recall our notation v = vivmvf , where vi is some initial part, vm is some middle part and vf is some final part of v (any part is allowed to be empty). A word of the form v−1 m is called λ-forbidden if Λ(vm) > λ. v−1 m , Λ(vm) > λ. (42) Definition 5.1. A word is called λ-semicanonical occurrences of λ-forbidden words. if it does not contain We are motivated by an informal analogue for small cancellation groups. Let G = hX Ri, where R satisfies small cancellation conditions [7] with a certain measure Λ on the relators Ri. That is Λ(Ri) = 1 and Λ(P ) 6 ε for any small piece P . We call a subword S of R±1 i λ-forbidden if Λ(S) > λ, where λ as above is between 1 2 and 1. A word U is λ-semicanonical if it does not contain λ-forbidden words. In the Cayley graph of G any λ-semicanonical word is a quasigeodesic [3], [4]. A small cancellation group (with appropriate constants) is hyperbolic. Recall that if the elements of a hyperbolic group are represented by quasi- geodesics, then their product takes the form of a thin triangle ([5]). In our case, first we will construct a special basis of Z2F /I with the use of λ- semicanonical words. Then in Section 6, we will construct a special set of linear generators of Z2F /I (not linearly independent) such that it contains the basis and we can express the product of two elements of this set as a sum of elements of this set that form thin triangles with the factors. Proposition 5.1. Let ah be a generalized fractional power, W = hahid. Then ah is either λ-semicanonical, or is equal modulo Dp(W ) to a sum of λ-semicanonical generalized fractional powers from W . Namely, either ah aj is λ-semicanonical, or there exists an elementary multi-turn ah 7→Pk such that aj, j = 1, . . . , k, j 6= h, are λ-semicanonical generalized fractional powers and they have shorter word length in F and less Λ-measure than ah. j=1 j6=h Proof. Let ah contain a λ-forbidden subword bh, ah = LbhR. Returning to forbidden words (42), we consider the following relation in Z2F /I: v−1 m = vf wvi + vf vi. (43) 76 Multiplying the relation v−1 = 1 + w by vf on the left side and by vi on the right side, we get (43). By definition, the words vf wvi and vf vi are non λ-forbidden. If v−1 m ) > λ, then we obtain m is a λ-forbidden word, that is, Λ(v−1 Λ(vf vi) = Λ(vf wvi) = Λ(vf ) + Λ(vi) < 1 − λ < λ < Λ(v−1 m ). Moreover, since w ≪ v, we obtain vf vi < vf wvi < v−1 word length in F . m , where · is By Definition 2.4, the transformation v−1 m 7→ vf wvi + vf vi is an elemen- tary multi-turn and according to the above, it gives the reduction of λ- forbidden words to sums of shorter (in word length in F and in Λ-measure) λ-semicanonical words. So, bh can be substituted by a sum of λ-semicanonical generalized frac- bj = vf wvi +vf vi. Since Λ(bh) > ε, it appears in v−1 only tional powersPn j=1 j6=h ah 7→ Pk j=1 j6=h once and, therefore, has fixed initial and fixed final point in the corresponding v-diagram. Since ah = LbhR is itself a generalized fractional power, we obvi- ously obtain that every word LbjR is a generalized fractional power, possibly after cancellations, with the same initial point and the same final point in the corresponding v-diagram as the word ah has. So, since v−1 m 7→ vf wvi + vf vi is LbjR is an elementary multi-turn as an elementary multi-turn, ah 7→Pn well. Hence, LbjR are derived monomials of ah and ah =Pn LbjR modulo Dp(W ). j=1 j6=h j=1 j6=h If the obtained word LbjR contains a λ-forbidden subword, we repeat the process and reduce it using the transformation (43). Successive elementary multi-turns match to addition of the corresponding expressions (9) -- (14). Thus, applying several elementary multi-turns consecutively, we obtain an elementary multi-turn as a result. So, we obtain an elementary multi-turn of ah after each step of the reduction process. From (43) it follows that after every transformation that reduces λ- forbidden subword of a monomial, the word length in F of the resulting monomials is strictly smaller than of the initial monomial. Thus, the reduc- tion process finishes and we obtain the result as an elementary multi-turn aj, where j = 1, . . . , k, j 6= h, are λ-semicanonical generalized fractional powers with shorter word length in F and less Λ-measure than ah. Corollary 5.2. Assume V = hU1, . . . , Ukid, U1, . . . , Uk ∈ F . Let U ∈ V be a monomial. Then U is equal to a sum of λ-semicanonical words from V 77 modulo Dp(V ). In particular, U is equal to a sum of λ-semicanonical words from hUid modulo Dp(hUid). Proof. Suppose U ∈ F and ah is a maximal occurrence of a generalized fractional power in U, U = LahR. Suppose ah contains a λ-forbidden word. Then, clearly, ah is a virtual member of the chart of U. From Proposition 5.1, ai, where it follows that there exists an elementary multi-turn ah 7→ Pk j = 1, . . . , k, j 6= h, are λ-semicanonical generalized fractional powers with the word length in F strictly smaller than the word length of ah. Then LajR are derived monomials of U and j=1 j6=h U = LahR = kXj=1 j6=h LajR mod Dp(hUid), where the word length in F of all monomials appearing in the right-hand side is strictly smaller than the word length of U. We continue the reduction process for the obtained monomials if necessary. Since word length of the monomials strictly decreases after each step of the reduction, the process finishes and we obtain the result as a sum of λ-semicanonical words. Assume U ∈ V = hU1, . . . , Ukid. Since V is generated by monomials and closed under taking derived monomials, Dp(hUid) ⊆ Dp(V ). Hence, U is equal to the same resulting sum of λ-semicanonical monomials as above modulo Dp(V ). The most important property of λ-semicanonical word for us is the fol- lowing. Proposition 5.3 (Non-degeneracy). Suppose ah is a λ-semicanonical gen- aj is an elementary multi-turn. Then eralized fractional power, ah 7→Pl there exists aj, j 6= h, such that Λ(aj) > τ . j=1 j6=h Proof. Let I and F be the initial and the final points of the paths correspond- ing to aj, j = 1, . . . , l, in the v-diagram. There are the following possible positions of I and F : Case 1 The points I and F lie on the v-arc. Case 2 The points I and F lie on a w-arc. 78 Case 3 The point I lies on the v-arc, the point F lies on a w-arc. Case 4 The point I lies on a w-arc, the point F lies on the v-arc. Consider Case 1 for a1, . . . , al. Assume the contrary, that is, Λ(aj) < τ , j = 1, . . . , l, j 6= h. Recall possible forms of monomials of Λ-measure less than τ (we enumerated them in the proof of Proposition 3.1). We denote the smallest path from I to F by b′, the smallest path from I to O by b1, and the smallest path from O to F by b2. Since Λ(aj) < τ for j 6= h, the smallest path between I and F is necessarily of Λ-measure less than τ . Then there are the following possibilities (for a graphical illustration see Proposition 3.1). 1.1 b′ contains the point O; 1.2 b′ does not contain the point O and Λ(b1) + Λ(b2) < τ ; 1.3 b′ does not contain the point O and Λ(b1) + Λ(b2) > τ . We will study the following particular configuration in 1.1 in detail. Recall that we use the notation v = vivmvf . wk O vi I vf F w−k vm , b2 = v−1 f , i v−1 f ; b1 = v−1 i b′ = v−1 79 Returning to pictures (5), we obtain that the following list of generalized fractional powers corresponds to the above picture: vmvf M(v, w)v−1 f , vmvf M(v, w)vivm, v−1 i M(v, w)vivm, v−1 i M(v, w)v−1 f . So, every aj, j = 1, . . . , l, is of the above form. Denote possible forms of ah by vmvf Mh(v, w)v−1 f , vmvf Mh(v, w)vivm, v−1 i Mh(v, w)vivm, v−1 i Mh(v, w)v−1 f . Since ah is λ-semicanonical, the monomial Mh(v, w) does not contain negative powers of v. By the assumption, Λ(aj) < τ for j 6= h; hence, aj = v−1 i wkj v−1 f . So, there are the following possible forms of the elementary multi-turn ah 7→ aj depending on the form of ah: Pl j=1 j6=h lXj=1 lXj=1 lXj=1 lXj=1 vmvf Mh(v, w)v−1 f 7→ i wkjv−1 v−1 f , vmvf Mh(v, w)vivm 7→ i wkjv−1 v−1 f , v−1 i Mh(v, w)vivm 7→ i wkj v−1 v−1 f , i wkj v−1 v−1 f . v−1 i Mh(v, w)v−1 f 7→ 80 We transform every elementary multi-turn above to an elementary multi- turn of a monomial over v, w, multiplying every expression above by corre- sponding generalized fractional powers: vMh(v, w) 7→ wkj , vMh(v, w)v 7→ wkj , Mh(v, w)v 7→ wkj , Mh(v, w) 7→ wkj . lXj=1 lXj=1 lXj=1 lXj=1 That is, we obtain an elementary multi-turn of the form fMh(v, w) 7→ wkj , lXj=1 (44) all monomials on the right-hand side are of zero Λ-measure, by Proposi- where the monomial fMh(v, w) does not contain negative powers of v. Since tion 3.1, Λ(fMh(v, w)) > τ . Hence, the monomial fMh(v, w) contains at least one positive power of v. Then, clearly, the polynomial P (v, w) = fM (v, w) + wkj lXj=1 does not satisfy condition (8), so, (44) can not be a multi-turn. We obtain a contradiction. All the rest of the configurations in Case 1 -- Case 4 are processed in the same way. Using Corollary 5.2, Proposition 4.10 and Proposition 5.3, we obtain the following important statement. Corollary 5.4. Assume W = hZid, where Z is a monomial. Then the space W/(Dp(W )+L(W )) is non-trivial if and only if there exists a λ-semicanonical 81 monomial eZ ∈ W such that W = heZid. Moreover, if X is a λ-semicanonical monomial such that X ∈ W \L(W ) (and then W = hXid), then X +Dp(W )+ L(W ) 6= 0 as an element of W/(Dp(W ) + L(W )). Proof. Assume the quotient space W/(Dp(W ) + L(W )) is non-trivial. Con- sider the set of all monomials from W \ L(W ). If every monomial from this set belongs to Dp(W ) + L(W ), then the quotient space W/(Dp(W ) + L(W )) is trivial. Hence, there exists a monomial Z ′ ∈ W \ L(W ) such that Z ′ /∈ Dp(W ) + L(W ). From Corollary 5.2 it follows that Z ′ = lXi=1 Zi mod Dp(W ), where Zi ∈ W are λ-semicanonical monomials. Since Z ′ /∈ Dp(W ) + L(W ), there exists a monomial Zi0 ∈ W \L(W ) in this sum. Hence, from Lemma 4.2 it follows that W = hZi0id. Proposition 4.10, we have Assume W = heZid, where eZ is λ-semicanonical monomial. Then, by W/(Dp(W ) + L(W )) ∼= A1/(Dp(A1) + L1) ⊗ . . . ⊗ Ak/(Dp(Ak) + Lk), where A1, . . . , Ak are spaces generated by generalized fractional powers and corresponding to each place in the chart of eZ by formula (32). In particular, if a(1), . . . , a(k) are all virtual members of the chart of eZ enumerated from left Recall that Dp(Ai) consists of supports of elementary multi-turns. There- to right, then a(1) ∈ A1, . . . , a(k) ∈ Ak. fore, by Proposition 5.3, we obtain a(i) /∈ Dp(Ai) + Li, i = 1, . . . , k. Hence, (a(1) + Dp(A1) + L1) ⊗ . . . ⊗ (a(k) + Dp(Ak) + Lk) 6= 0. (45) Then, by definition of the isomorphism we have µ2[eZ] : A1/(Dp(A1) + L1) ⊗ . . . ⊗ Ak/(Dp(Ak) + Lk) → W/(Dp(W ) + L(W )), µ2[eZ]((a(1) + Dp(A1) + L1) ⊗ . . . ⊗ (a(k) + Dp(Ak) + Lk)) = µ[eZ](a(1) ⊗ . . . ⊗ a(k)) + Dp(W ) + L(W ) = eZ + Dp(W ) + L(W ). 82 Then, by (45), we obtain that eZ + Dp(W ) + L(W ) 6= 0. Thus, the quotient Let X be a λ-semicanonical monomial from W \L(W ). Then, by Lemma 4.2, space W/(Dp(W ) + L(W )) is non-trivial. we have W = hXid. Therefore, by the same argument as above, we obtain that X + Dp(W ) + L(W ) 6= 0. This completes the proof. 5.2 How to calculate a basis in Z2F /I using λ-semicanonical words As before, we denote the space hU1, . . . , Ukid, where U1, . . . , Uk are monomi- als, by V . Putting together the results of Corollary 4.11, Theorem 4.1 and Proposition 4.12, we obtain (V + hT ′i)/hT ′i ∼= N −1Mn=0Mi∈I ′ n V (n) i /(Dp(V (n) i ) + L(V (n) i )), (46) i = hZ (n) i id, Z (n) i is a monomial from FnV , V (n) i * Fn+1V and the where V (n) spaces V (n) i are pairwise different. Proposition 5.5. Assume (i,n) j W (i,n) j ∈ V (n) i /(Dp(V (n) i ) + L(V (n) i )), j ∈ No , nW j i is a basis of V (n) ∈ V (n) Let W (i,n) i ) + L(V (n) /(Dp(V (n) be an arbitrary representative of the coset W )), where n = 0, . . . , N − 1 and i ∈ I ′ n. . Then i i (i,n) j N −1[n=0 [i∈I ′ nnW (i,n) j + hT ′i j ∈ No is a basis of (V + hT ′i)/hT ′i. Proof. While the statement is pretty obvious, we prefer to give a proof to recollect the previously stated facts. In Proposition 4.12 we proved that there exists an isomorphism ϕ : FnV /(Dp(FnV ) + Fn+1V ) →Mi∈I ′ n V (n) i /(Dp(V (n) i ) + L(V (n) i )) 83 that acts as follows ϕ(U + Dp(FnV ) + Fn+1V ) = (0, . . . , 0, U + Dp(V (n) i ) + L(V (n) i ), 0, . . .), (47) where U is a monomial, U ∈ FnV \ Fn+1V . Suppose (i,n) j W (i,n) j ∈ V (n) i /(Dp(V (n) i ) + L(V (n) i )), j ∈ No nW is a basis of V (n) i /(Dp(V (n) i ) + L(V (n) i )). Then from (47) it easily follows that [i∈I ′nW (i,n) j + Dp(FnV ) + Fn+1V j ∈ No , where W (i,n) basis of FnV /(Dp(FnV ) + Fn+1V ). j ∈ FnV is an arbitrary representative of the coset W (i,n) j , is a Recall that Fn(V /Dp(V )) = (FnV + Dp(V ))/Dp(V ) and, by the isomorphism theorems, we have the following canonical isomor- phisms: π1 : Fn(V /Dp(V ))/Fn+1(V /Dp(V )) → (FnV + Dp(V ))/(Fn+1V + Dp(V )), π2 : (FnV + Dp(V ))/(Fn+1V + Dp(V )) → FnV /(Fn+1V + Dp(V ) ∩ FnV ). Let us recall that from Proposition 4.9 it follows that Dp(V ) ∩ FnV = Dp(FnV ). Therefore, we have Fn(V /Dp(V ))/Fn+1(V /Dp(V )) π1−→ (FnV + Dp(V ))/(Fn+1V + Dp(V )) π2−→ FnV /(Fn+1V + Dp(FnV )). Let W be an arbitrary element of FnV (not necessarily a monomial). Then, by the well known construction of the canonical isomorphisms in the isomor- phism theorems, we have π2 ◦ π1(W + Dp(V ) + Fn+1(V /Dp(V ))) = π2(W + Dp(V ) + Fn+1V ) = W + Fn+1V + Dp(FnV ). So, since [i∈I ′nW (i,n) j + Dp(FnV ) + Fn+1V j ∈ No 84 is a basis of FnV /(Fn+1V + Dp(FnV )), we obtain that [i∈I ′nW (i,n) j + Dp(V ) + Fn+1(V /Dp(V )) j ∈ No (48) is a basis of Fn(V /Dp(V ))/Fn+1(V /Dp(V )). We have the graded space Gr(V /Dp(V )) = N −1Mn=0 Fn(V /Dp(V ))/Fn+1(V /Dp(V )), where Fn(V /Dp(V )) = (FnV + Dp(V ))/Dp(V ). Assume that (n) j Z nZ (n) j ∈ Fn(V /Dp(V ))/Fn+1(V /Dp(V )), j ∈ No is a basis of Fn(V /Dp(V ))/Fn+1(V /Dp(V )), n = 0, 1, . . . , N − 1. Then one can easily show that N −1[n=0nZ (n) j Z (n) j ∈ Fn(V /Dp(V )), j ∈ No , j is an arbitrary representative of the coset Z where Z (n) , forms a basis of V /Dp(V ). Hence, using this observation together with (48) and Corol- lary 4.11, we obtain that (n) j N −1[n=0 [i∈I ′ nnW (i,n) j + hT ′i j ∈ No is a basis of (V + hT ′i)/hT ′i. Proposition 5.6. Let {Xj} be all the λ-semicanonical monomials of V . Then there exists one-to-one correspondence between all the different spaces hXjid and all the spaces V (n) )) is non-trivial. Namely, (1) every space V (n) from (46) such that V (n) such that V (n) )) 6= 0 is equal to ) + L(V (n) /(Dp(V (n) i i /(Dp(V (n) ) + L(V (n) i i i i i i some space hXjid; (2) every space hXjid is equal to some space V (n) such that V (n) i /(Dp(V (n) i )+ i L(V (n) i )) 6= 0. 85 Proof. Consider a space V (n) )) 6= 0. Then from Corollary 5.4 it follows that there exists a λ-semicanonical monomial Xj0 ∈ V (n) such that V (n) such that V (n) /(Dp(V (n) ) + L(V (n) i = hXj0id. i i i i i Let Xj ∈ V be a λ-semicanonical monomial. Since FN V = 0, there exists a number n0 such that Xj ∈ Fn0V and Xj /∈ Fn0+1V . Then, by Proposition 4.12, Xj ∈ V (n0) n0. Since Xj /∈ Fn0+1V , we have Xj /∈ L(V (n0) = hXjid. It remains to notice that, by Corollary 5.4, we have V (n0) for some i ∈ I ′ ). Then, by Lemma 4.2, V (n0) ) + L(V (n0) /(Dp(V (n0) )) 6= 0. i i i i i i The next theorem follows from Proposition 5.5 and Proposition 5.6. Theorem 5.1. Let {Xj}j∈N be all λ-semicanonical monomials from F . Let {Vi}i∈N be all the different spaces {hXjid}j∈N (Vi1 6= Vi2 for i1 6= i2). If is a basis of Vi/(Dp(Vi) + L(Vi)), then (i) j W nW (i) j ∈ Vi/(Dp(Vi) + L(Vi)), j ∈ No [i∈NnW (i) j + I j ∈ No is a basis of Z2F /I, where W (i) j W (i) j . In particular, is an arbitrary representative of the coset Z2F /I ∼=Mi∈N Vi/(Dp(Vi) + L(Vi)) as vector spaces, and the right-hand side is explicitly described in Proposi- tion 4.10. Proof. Let us enumerate all the monomials from F , namely, F = {U1, U2, . . . , Uk, . . .}. Then, evidently, Z2F /I = ∞[k=1 ((hU1, . . . , Ukid + hT ′i)/hT ′i) . Consider spaces hU1, . . . , Ukid ⊆ hU1, . . . , Uk, Uk+1id. For the quotient space (hU1, . . . , Ukid + hT ′i)/hT ′i we have decomposition (46), that is, (hU1, . . . , Ukid + hT ′i)/hT ′i ∼= V (n) i /(Dp(V (n) i ) + L(V (n) i )), Nk−1Mn=0 Mi∈I ′ n 86 where Nk is the number of non-zero levels of the filtration on hU1, . . . , Ukid. Let {Xj} be all the λ-semicanonical monomials from hU1, . . . , Ukid. Then, obviously, the set of all λ-semicanonical monomials from hU1, . . . , Uk+1id can be represented as {Xj} ∪ {eXj}, i i V (n) i /(Dp(V (n) ) + L(V (n) position (46) for (hU1, . . . , Uk+1id + hT ′i)/hT ′i can be written as in hU1, . . . , Uk, Uk+1id but do not contain in hU1, . . . , Ukid (in particular, the where {eXj} collects the additional λ-semicanonical monomials that contain set {eXj} may be empty). Then from Proposition 5.6, it follows that decom- (hU1, . . . , Uk+1id + hT ′i)/hT ′i ∼= )) Nk−1Mn=0 Mi∈I ′ M )) , Nk+1−1Mn=0 Mh∈I ′′ ) + L(eV (n) n collects the additional spaces eV (n) respond to additional monomials {eXj}. h /(Dp(eV (n) andeV (n) )) )) are non-trivial. We will choose a basis in every direct summand V (n) )) independently. Then from Proposition 5.5, it easily follows that there exists a basis Bk+1 of (hU1, . . . , Uk+1id+hT ′i)/hT ′i such that it expands the previously chosen basis Bk of (hU1, . . . , Ukid + hT ′i)/hT ′i. Therefore, finally, we will construct the increasing sequence of basis )) and eV (n) h /(Dp(eV (n) where Nk+1 is the number of non-zero levels of the filtration on hU1, . . . , Uk, Uk+1id. The set I ′′ , not appearing in I ′ From Corollary 5.4, it follows that the spaces V (n) neV (n) h /(Dp(eV (n) h h ) + L(eV (n) h )+L(eV (n) i /(Dp(V (n) ) + L(V (n) n, that cor- /(Dp(V (n) ) + L(V (n) i i i i n h i h h h B1 ⊆ B2 ⊆ . . . ⊆ Bk ⊆ . . . . Since the ring Z2F /hT ′i is a union of the subspaces (hU1, . . . , Ukid+hT ′i)/hT ′i, k=1 Bk is a basis of Z2F /hT ′i = Z2F /I. From Theorem 5.1 we obtain the claim that served as the main motiva- the unionS∞ tion. Corollary 5.7. The quotient ring Z2F /I is non-trivial. 87 Remark 5.1. It is possible to choose a basis in every direct summand Vi/(Dp(Vi) + L(Vi)), where Vi = hZiid, Zi is a λ-semicanonical monomial, in the following way. In Proposition 4.10, we have proved that Vi/(Dp(Vi) + L(Vi)) is isomorphic to a tensor product of spaces that corresponds to each virtual member of the chart of Zi. Namely, there exist spaces A1, . . . , Aki defined by (32) and the isomorphism µ2[Zi] :A1/(Dp(A1) + L(A1)) ⊗ . . . ⊗ Aki/(Dp(Aki) + L(Aki)) → Vi/(Dp(Vi) + L(Vi)) defined by (36). So, if we choose a basis in all the different spaces Aj/(Dp(Aj)+ L(Aj)), this enables us to construct the corresponding basis in every direct summand Vi/(Dp(Vi) + L(Vi)). From Proposition 5.2 it follows that Aj/(Dp(Aj) + L(Aj)) is generated by cosets of λ-semicanonical generalized fractional powers that belong to Aj. We choose a basis of Aj/(Dp(Aj) + L(Aj)) that consists of such cosets. Let (b(1) l + Dp(A1) + L(A1)) ⊗ . . . ⊗ (b(ki) l + Dp(Aki) + L(Aki)), l ∈ N, be the basis elements, where b(1) fractional powers. Then their images l , . . . , b(ki) l are λ-semicanonical generalized µ2[Zi](b(1) l + Dp(A1) + L(A1)) ⊗ . . . ⊗ (b(ki) l + Dp(Aki) + L(Aki)) = µ[Zi](b(1) l ⊗ . . . ⊗ b(ki) l ) + Dp(Vi) + L(Vi) form a basis of Vi/(Dp(Vi + L(Vi)). We put l = µ[Zi](b(1) Then from Theorem 5.1 it follows that W (i) l ⊗ . . . ⊗ b(ki) l {W (i) l + I l ∈ N} ∞[i=1 ). (49) (50) is a basis of Z2F /I. Notice that, by definition of µ[Zi], W (i) are monomials. From the description given in Section 3.1, it follows that the monomials W (i) are (λ + 2ε)-semicanonical. l l 88 Notice that the constructed basis satisfies the following property. Assume U is a monomial and there is the base decomposition U + I =Xi,l W (i) l + I. in this decomposition is a derived monomial of U, in par- Then every W (i) l ticular, f (W (i) l ) 6 f (U). 6 Geometry of the multiplication The linear span of all λ-semicanonical words is denoted by Sλ. However, it factors. multiplication properties. Namely, the result of the multiplication can be turns out that that it is more convenient to define a subspace eSλ for which Sλ ⊆ eSλ ⊆ Sλ+2ε that contains all the elements of the basis of Z2F /I. The cosets of elements of eSλ linearly generate Z2F /I, and they display nice represented as a sum of monomials from eSλ that form thin triangles with the 6.1 Definition of the space eSλ Let U be a (λ + 2ε)-semicanonical word and a be a virtual member of its chart. Assume l and r are virtual members of the chart of U such that l and r are the left and the right neighbour of a in the chart, respectively. Let al be an overlap between a and l (possibly empty), ar be an overlap between a and r (possibly empty). Then a = alamar. U l U U U l al am ar r al am ar is empty am ar r al is empty both ar and al are empty am 89 l l Definition 6.1 (eSλ-condition). We say that a virtual member of the chart a satisfies eSλ-condition if the subword am of a defined above is λ-semicanonical. We denote by eSλ a linear span of words such that each virtual member of the chart of a word satisfies the eSλ-condition. Clearly, every monomial that belongs to eSλ is (λ + 2ε)-semicanonical. Notice that unlike words from Sλ, words from eSλ may contain subwords that do not belong to eSλ. belong to eSλ. Assume a monomial Uh = LahR ∈ eSλ and let ah be a virtual member of its chart. Let Uh 7→ Pm tary multi-turn ah 7→Pm Recall that in the previous section we constructed a linear basis in Z2F /I, are λ-semicanonical), we obtain (the fixed representatives of the basis elements) fractional powers. Consider monomials Uj = LajR such that aj is a virtual member of the chart of Uj. Then, by Corollary 3.7, we obtain that the images of all virtual members of the chart of U are virtual members of the chart of see (50). By formula (49) (since b(1) that the monomials W (i) Uj be a multi-turn that comes from an elemen- aj such that aj are λ-semicanonical generalized , . . . , b(ki) j=1 j6=h j=1 j6=h l Uj. Uj. Hence, by the definition of the space eSλ, we immediately obtain that these monomials belong to eSλ even if the neighbours of ah are prolonged in Assume a monomial Uh = LahR is a λ-semicanonical word and we per- form a multi-turn of ah as above. If again we consider monomials Uj = LajR such that aj is a virtual member of the chart of Uj, then Uj is not neces- sarily λ-semicanonical, because the neighbours of ah may prolong in Uj and further proofs. become λ-forbidden. That is why the space eSλ is more convenient in our 6.2 Behaviour of eSλ with respect to multi-turns As above, assume Uh = LahR ∈ eSλ, ah is a virtual member of its chart, Uh 7→ Pm ah 7→Pm Let us describe in detail monomials Uj such that aj is not a virtual member of the chart of Uj. We split these monomials into three groups in the same way as we did in Section 3.1. aj such that aj are λ-semicanonical generalized fractional powers. Uj is a multi-turn that comes from an elementary multi-turn j=1 j6=h j=1 j6=h 90 1. LajR, where Λ(aj) > ε; 2. LajR, where aj = 1; 3. LajR, where Λ(aj) 6 ε, aj 6= 1. Being a λ-semicanonical word depends only on the Λ-measure of every virtual member of the chart. Hence, a resulting monomial of a multi-turn of a λ-semicanonical word may become non-λ-semicanonical only if the Λ- measure of some virtual members of the chart increases and λ-forbidden pends on a virtual member of the chart itself and on its neighbours and the notion of virtual member of the chart depends on the whole monomial. subwords appear in this monomial. One can see that the eSλ-condition de- Hence, the eSλ-condition on a virtual member of the chart may not be hold in a resulting monomial of a multi-turn not only when the virtual member of the chart changes itself, but when the virtual member of the chart stay un- changed but some other maximal occurrence becomes shorter. Let us study the possibilities in detail. Consider monomials of type 1. Let us use notations from Section 3.1. We assume that bh is the left neighbour of ah in the chart of Uh, bj is a maximal occurrence of a generalized fractional power corresponding to bh in the chart of Uj. 1. If bh and ah are separated or touch at a point and Λ(bh) < τ , then bj may become not a virtual member of the chart and the eSλ-condition may not hold for its left neighbour. Λ(d) = λ + ε, d is λ-forbidden d Uh bh d Uj bj ah aj 2. If we have configurations 1.3 or 1.4 (page 18) for bj, then bj may become not a virtual member of the chart and the eSλ-condition may not hold for its left neighbour. 91 Λ(d) = λ + ε, d is λ-forbidden d Uh d Uj bj ah bh aj 3. Suppose we have configurations 1.1, 1.2 (page 17) or bj = bh. If bj contains a λ-forbidden subword d that has an overlap with aj, then the eSλ-condition may not hold for bj (because aj is not a virtual member of the chart). Λ(d) = λ + ε, d is λ-forbidden Uh Uj ah d bh aj d bj Obviously, the effects for the right neighbour of ah in the chart are the same. Consider monomials of type 2. First we do cancellations in Uj = LR if j = L′R′. Recall that we suppose that a is possible, after that we obtain U ′ a maximal occurrence of a generalized fractional power that is a terminal subword of L′, b is a maximal occurrence of a generalized fractional power that is an initial subword of R′, a′ and b′ are generalized fractional powers that prolong a and b in L′R′ respectively. 1. In any configuration except 2.1, there may be a maximal occurrence bh, Λ(bh) < τ , fully contained in the subword L′ that becomes not a virtual member of the chart of L′R′. Then the eSλ-condition may not hold for its left neighbour. Λ(d) = λ + ε, d is λ-forbidden d L′ bh 92 R′ The same may happen in the subword R′. 2. Suppose we have configuration 2.2 (page 19). Then a′ or b′ may contain a λ-forbidden subword and do not satisfy the eSλ-condition. We may obtain this situation if the right neighbour of a′ in the chart of L cancels out up to the end of a′ or if the left neighbour of b′ in the chart of R cancels out up to the beginning of b′. Λ(d) = λ + ε, d is λ-forbidden L′ d a′ b′ R′ 3. Suppose we have configuration 2.2 or 2.3 (page 19). In this case, a′ may become too short after cancellations and may not be counted as a virtual member of the chart. If the left neighbour of aj contains a λ-forbidden subword d that has an overlap with a′, then the word L′R′ may not belong to eSλ. Λ(d) = λ + ε, d is λ-forbidden L′ d b′ a′ R′ We may obtain a similar effect for the right neighbour of b′. 4. If we have configuration 2.3 (page 19), then a′ or b′ may contain a λ- forbidden subword that does not satisfy the eSλ-condition. For example, we may obtain this case when b′ is not a virtual member of the chart and a′ contains a λ-forbidden subword d that has an overlap with b′. Λ(d) = λ + ε, d is λ-forbidden L′ d a′ b′ R′ 5. Suppose we have configuration 2.4 (page 20) and ab is not a virtual member of the chart. If the left or the right neighbour of ab contains a λ-forbidden subword d that has an overlap with ab, then the word L′R′ may not belong to eSλ (similar to 3). 93 Λ(d) = λ + ε, d is λ-forbidden L′ d ab R′ 6. If we have configuration 2.4 (page 20), the eSλ-condition may not hold for ab. Consider monomials of type 3. Recall that we suppose that a is a maximal occurrence of a generalized fractional power that is a terminal subword of L, b is a maximal occurrence of a generalized fractional power that is an initial subword of R, a′ and b′ are generalized fractional powers that prolong a and b in LajR, respectively. 1. In any configuration except 3.1 there may be a maximal occurrence bh, Λ(bh) < τ , fully contained in the subword L that becomes not a virtual member of the chart of Uj. Then the eSλ-condition may not hold for its left neighbour. Λ(d) = λ + ε, d is λ-forbidden d Uj bh aj The same may happen in the subword R. 2. If we obtain configuration 3.2 (page 20), then a′ may be too short to be counted as a virtual member of the chart. In this case, the eSλ-condition may not hold for its left neighbour. Λ(d) = λ + ε, d is λ-forbidden L d a′ aj R Clearly, we may obtain a similar effect for the right neighbour of b′. 3. Suppose we have configuration 3.3 or 3.4 (page 21). Then a′ or b′ may contain a λ-forbidden subword d that does not satisfy the eSλ-condition. 94 Λ(d) = λ + ε, d is λ-forbidden L Λ(d) > λ + ε, d is λ-forbidden L d a′ aj d a′ aj R R b′ for the virtual member aajb. 4. If we obtain configuration 3.5 (page 22), the eSλ-condition may not hold 6.3 Multiplication of words from eSλ In Proposition 5.1 we showed one possible way of transformation of an arbi- trary generalized fractional power into a sum of λ-semicanonical generalized fractional powers. But some of the obtained elements may have Λ-measure less than the given threshold τ . In the further argument in Section 6.3, it is more convenient to deal with multi-turns such that all the resulting λ- semicanonical generalized fractional powers are of Λ-measure not less than τ . In the following proposition, we construct an elementary multi-turn with this property. Proposition 6.1. Suppose ah is a generalized fractional power. Then there aj such that aj are λ-semicanonical exists an elementary multi-turn ah 7→Pk j=1 j6=h generalized fractional powers and Λ(aj) > τ , j 6= h, i.e., aj is always counted as a virtual member of the chart (possibly with greater word length and Λ- measure than ah). Moreover, one can choose this multi-turn such that every aj, j 6= h, does not contain subwords of v−1 of Λ-measure greater than ε. Proof. Let us construct this elementary multi-turn directly using Defini- tion 2.4. If we consider v as an element of the field Z2(w) such that v−1 = 1 + w, then in Z2(w) we have 1 = vw + v and v−1 = 1 + w = (vw + v) + (vw + v)w = v + vw2. (51) Therefore, in Z2(w) we have v−m = (v +vw2)m. Hence, the non-commutative Laurent polynomial x−m (8). Hence, the transformation in Z2F 2)m, m > 0, satisfies 1 + (x1 + x1x2 v−m 7→ (v + vw2)m, m > 0, (52) 95 is an elementary multi-turn. Suppose Mh(v, w) is a non-commutative monomial in v, w, that is, Mh(v, w) = wk1vl1 · · · wknvln, where ki, li ∈ Z. Again consider Mh(v, w) as an element of the field Z2(w) such that v−1 = 1 + w. Then, since v and w commute in Z2(w), we have i=1 li. Hence, the non- 2xl 1 satisfies (8). There- wk1vl1 · · · wknvln = wkvl, where k = Pn i=1 ki, l = Pn commutative Laurent polynomial xk1 fore, the transformation in Z2F 1 · · · xkn 1 + xk 2 xln 2 xl1 wk1vl1 · · · wknvln 7→ wkvl, k = ki, l = li, (53) nXi=1 nXi=1 i=1 li < 0, from (52) it follows that is an elementary multi-turn. Since 1 7→ v + vw is an elementary multi-turn, wk1vl1 · · · wknvln 7→ wk(v + vw2)−l is an elementary multi-turn. If l =Pn if l =Pn is an elementary multi-turn. Using (53) -- (55), for every l =Pn wk1vl1 · · · wknvln 7→ wk(v + vw) i=1 li = 0, we obtain that a multi-turn Mh(v, w) 7→ Mj(v, w), (54) (55) (56) i=1 li we obtain sXj=1 j6=h where every Mj(v, w), j 6= h, does not contain negative powers of v and contains at least one positive power of v. By definition, a generalized fractional power is a subword in a non- commutative monomial over the words v and w. Recall that the types of generalized fractional powers are enumerated in (5) -- (7). Clearly, if we con- sider generalized fractional powers with the possibility of cancellations, we obtain that ah before cancellations has one of the following forms: vf Mh(v, w)vi, vmvf Mh(v, w)vivm, wf Mh(v, w)vi, vf Mh(v, w)wi, wf Mh(v, w)wi, wmwf Mh(v, w)wiwm, 96 where Mh(v, w) is a monomial over the words v and w. Suppose ah = vf Mh(v, w)vi, where vf M(v, w)hvi may have cancellations. From (56) it follows that the transformation vf M(v, w)hvi 7→Ps is an elementary multi-turn after all possible cancellations in monomials are done. That is, j=1 j6=h vf Mj(v, w)vi ah 7→ vf Mj(v, w)vi (57) sXj=1 j6=h is an elementary multi-turn. Let us check that (57) satisfies the conditions of Proposition 6.1. Since every Mj(v, w), j 6= h, contains only positive powers of v and v has no cancellations with w and w−1, every monomial in the right hand side of this multi-turn is of Λ-measure not less than 1, that is, it is especially of Λ-measure not less than τ . Since wk can not contain subwords of v and v−1 of Λ-measure greater than ε, every monomial in the right hand side does not contain subwords of v−1 of Λ-measure greater than ε. Thus, multi-turn (57) satisfies the conditions of Proposition 6.1. In a similar way we deal with ah of the other form and we obtain multi- turns vmvf Mj(v, w)vivm, ah 7→ wf Mj(v, w)vi, sXj=1 j6=h vf Mj(v, w)wi, ah 7→ sXj=1 j6=h wmwf Mj(v, w)wiwm, wf Mj(v, w)wi, (58) ah 7→ ah 7→ ah 7→ j6=h sXj=1 sXj=1 sXj=1 j6=h j6=h possibly after cancellations in the right hand sides. Notice that only subwords of w may be involved in cancellations. By the same argument as (57), it is proved that multi-turns (58) satisfy the conditions of Proposition 6.1. Let U1 and U2 be monomials from eSλ. Then their product U1U2 may not belong to eSλ. According to Corollary 5.2, every monomial is equal to a sum of words from Sλ modulo I. Moreover, every monomial is equal to a sum of 97 words from eSλ modulo I, since Sλ ⊆ eSλ. We will represent U1U2 as a sum of monomials from eSλ. In this section, we analyse the process of reduction of the word U1U2 to a sum of words from eSλ modulo I in detail. In parallel we will describe the multiplication diagram that encrypts the history of this process. The process of the reduction of U1U2 is as follows. Step 1 Cancel U1U2 if possible. The multiplication diagram that encrypts cancellations is of the form U ′ U ′−1 U ′ 1 P U ′ 2 1U ′ U1 = U ′ U2 = U ′−1U ′ U = U ′ 1U ′ 2 2 has no further cancellations 1U ′ Let U = U ′ 2. Now we consider the chart of U. As we described in Sec- tion 3.1, we have one of configurations 2.1 -- 2.4 (pages 19 -- 20). If we have configuration 2.4 and the generalized fractional power that is obtained as a result of merging does not satisfy eSλ-condition, then we go to Step 2 of the Denote the generalized fractional power that is obtained as a result of reduction process. So, let us first describe this case. merging by ah, U = LahR. U ′ ah P L R Suppose ah does not satisfy the eSλ-condition. Then the next step of the multiplication process is as follows. Step 2 According to Proposition 6.1, a λ-forbidden virtual member ah can be reduced to a sum of λ-semicanonical generalized fractional powers using aj, Λ(aj) > τ . So, an appropriate elementary multi-turn ah 7→ Pk perform the multi-turn U = LahR 7→ Pk this elementary multi-turn. j=1 j6=h j=1 j6=h LajR, which comes from 98 Since Λ(aj) > τ , the resulting monomials LajR belong to eSλ. The monomial ah corresponds to a unique path in the v-diagram (see Definition 2.3 is Section 2) with the initial point I and the final point F . The monomials aj, j = 1, . . . , k, correspond to certain paths in this v-diagram with the same initial point I and the same final point F . Then we add this v-diagram to the bottom of the multiplication diagram, gluing the point I with the end of L and point F with the beginning of R. The previous part of the multiplication diagram is glued to the added v-diagram at the same point on ah as before. L U ′ P I F R So, one can read the monomials LajR, j = 1, . . . , k, in the low level of the obtained diagram. So far for Step 2. Now consider configurations 2.1 -- 2.3 in U = U ′ 1U ′ 2 after Step 1. Then given in Section 6.2, U contains at most two subwords that do not satisfy the chart of U2 (see analysis of monomials of type 2, page 92; although here we the monomial U also may not belong to eSλ. According to the classification eSλ-condition, one comes from the chart of U1 and the other comes from the deal with a product of two words from eSλ, this analysis is still applicable). does not satisfy the eSλ-condition. To be precise, assume that it comes from the chart of U1, denote it by bh. Then there are the following possible con- figurations depending on the position bh with respect to the point P : First assume that U contains only one virtual member of the chart that U U U bh P bh bh P P LbjR from Proposition 6.1 that reduces bh. Since Λ(bj) > τ , the resulting monomials LbjR belong to As above we perform a multi-turn LbhR 7→Pk eSλ. above and obtain the final multiplication diagram. j=1 j6=h We add a new v-diagram to the multiplication diagram as we described 99 L L L L P P P P R R R R Now assume that U contains two virtual members of the chart that do not satisfy the eSλ-condition, denote them by bh and ch. Then there are the following possible configurations depending on the position of bh and ch with respect to the point P . U U U U U U U U U bh bh P P bh P ch ch ch bh P ch bh bh bh P ch P ch bh P ch P P ch ch bh 100 j=1 j6=h LbjR from Then first, as above, we perform a multi-turn LbhR 7→Pk the eSλ-condition. Denote the image of ch in the monomial LbjR by c(j) Proposition 6.1 that reduces bh. Since Λ(bj) > τ , in every monomial LbjR the image of ch is the only virtual member of the chart that does not satisfy h . So, in every monomial LbjR we perform a multi-turn from Proposition 6.1 that reduces c(j) h . As a result, we obtain a sum of monomials from eSλ. h , we glue a new v-diagram to the multiplication diagram obtained after the reduction of bh and obtain the final multiplica- tion diagram. Let us illustrate the first three configurations of bh and ch. The multiplication diagrams for the remain configurations are constructed similarly. After the reduction of c(j) L L L P P P R R R So, finally we obtain the multiplication diagram as a thin triangle. The result of the multiplication is a sum of words from eSλ that correspond to certain paths in the low level of the obtained thin triangle. Thereby, we proved the following theorem. Theorem 6.1. Assume U1 + I, U2 + I ∈ Z2F /I, where U1 and U2 are of v-diagrams for every i = 1, . . . , k. i=1 Zi + I, and U1, U2 and Zi form a thin triangle consists monomials from eSλ. Then there exist monomials Z1, . . . , Zk from eSλ such that U1U2 + I =Pk but rather an arbitrary elementary multi-turn ah 7→ Pk Remark 6.1. If we use not an elementary multi-turn from Proposition 6.1, aj in order to reduce ah at Step 2, some aj may be of Λ-measure 6 ε. In this case, we may obtain further merging of virtual members of the chart of LajR. Namely, the cases 2.4 and 3.5 (pages 20, 22), in which the neighbours of aj in the chart merge, are possible. For example, for 3.5 we have j=1 j6=h 101 U ′ l aj r merged virtual member of the chart If there exists aj = 1, we may obtain cancellations in the resulting monomial that contain aj = 1. So, when we reduce U1U2, we obtain a sequence of Step 1 and Step 2. We can think of it as of a branching process, starting from the monomial U1U2. After each execution of Step 1 we obtain one monomial, after each execution of Step 2 we obtain a sum of monomials. Each monomial in the sum is treated independently and potentially gives us a new branch of the process. Since we may obtain merging of virtual members of the chart or cancellations at most in one resulting monomial, the process stops in all branches except at most one after each step. From Proposition 4.4, it follows that finally the process stops in all branches. For every branch of the process, we can construct a multiplication diagram in the same way as above. If we do an arbitrary multi-turn in order to reduce the monomials obtained after the end of the sequence of Step 1 and Step 2 in a branch, we may obtain a multiplication diagram that is not a thin triangle, but has a form of tree. Nevertheless, the reduction of these monomials can be done using multi-turns described in Proposition 6.1, and in this case the multiplication diagram for every branch of the process again is a thin triangle but of more complicated form ... So, we obtain the multiplication diagram of the whole multiplication process 102 as a set of thin triangles. The result of the multiplication is a sum of words from eSλ that correspond to certain paths in the low levels of the obtained thin triangles. References [1] G. M. Bergman, The Diamond Lemma for ring theory, Adv. Math. 29 (1978), no. 2, 178-218, [2] M. R. Bridson, A. Haefliger, Metric Spaces of Non-Positive Curvature, Springer-Verlag Berlin Heidelberg New York, 1999. [3] M. Coornaert, T. Delzant, A. Papadopoulos, G´eom´etrie et th´eorie des groupes: les groupes hyperboliques de Gromov, Lecture Notes in Mathe- matics 1441, Springer, 1991. [4] E. Ghys and P. de la Harpe (eds.), Sur les Groupes Hyperboliques d'apr`es Mikhael Gromov, Progress in Math. 83, Birkhauser, 1990. [5] M. Gromov, Hyperbolic groups, Essays in Group Theory, S. Gersten (ed.), MSRI publication no. 8, Springer Berlin Heidelberg New York, 1987, 75 -- 263. [6] S. Lang, Algebra, Springer-Verlag New York, 2002. [7] R. C. Lyndon and P. E. Schupp, Combinatorial Group Theory, Springer, Berlin-Heidelberg-New York, 1977. [8] A. Yu. Ol'shanskii, Geometry of Defining Relations in Groups, Springer Science + Business Media Dordrecht, 1991. [9] E. Rips and Z. Sela, Canonical representatives and equations in hyper- bolic groups, Invent. Math. 120 (1995), no. 1, 489 -- 512. 103
0908.3213	4	0908	2012-06-19T21:26:45	Classification of abelian complex structures on 6-dimensional Lie algebras	[ "math.RA", "math.DG" ]	We classify the 6-dimensional Lie algebras that can be endowed with an abelian complex structure and parameterize, on each of these algebras, the space of such structures up to holomorphic isomorphism.	math.RA	math	CLASSIFICATION OF ABELIAN COMPLEX STRUCTURES ON 6-DIMENSIONAL LIE ALGEBRAS A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI ABSTRACT. We classify the 6-dimensional Lie algebras that can be endowed with an abelian complex structure and parameterize, on each of these algebras, the space of such structures up to holomorphic isomorphism. 1. INTRODUCTION Let g be a Lie algebra, J an endomorphism of g such that J 2 = −I, and let g1,0 be the i-eigenspace of J in gC := g ⊗R C. When g1,0 is a complex subalgebra we say that J is integrable, when g1,0 is abelian we say that J is abelian and when g1,0 is a complex ideal we say that J is bi-invariant. We note that a complex structure on a Lie algebra cannot be both abelian and bi-invariant, unless the Lie bracket is trivial. If G is a connected Lie group with Lie algebra g, by left translating J one obtains a complex manifold (G, J) such that left multiplication is holomorphic and, in the bi-invariant case, also right multiplication is holomorphic, which implies that (G, J) is a complex Lie group. Our concern here will be the case when J is abelian. In this case the Lie algebra has abelian commutator, thus, it is 2-step solvable (see [17]). However, its nilradical need not be abelian (see Remark 4.7). Abelian complex structures have interesting applications in hyper-Kahler with torsion geometry (see [6]). It has been shown in [9] that the Dolbeault cohomology of a nilmanifold with an abelian complex structure can be computed algebraically. Also, deformations of abelian complex structures on nilmanifolds have been studied in [10]. Of importance, when studying complex structures on a Lie algebra g, is the ideal g′ J := g′ + J g′ con- structed from algebraic and complex data. We will say that the complex structure J is proper when g′ J is properly contained in g. Any complex structure on a nilpotent Lie algebra is proper [20]. The 6-dimensional nilpotent Lie algebras carrying complex structures were classified in [20], and those carrying abelian com- plex structures were classified in [12]. There is only one 2-dimensional non-abelian Lie algebra, the Lie algebra of the affine motion group of R, denoted by aff(R). It carries a unique complex structure, up to equivalence, which turns out to be abelian. The 4-dimensional Lie algebras admitting abelian complex structures were classified in [21]. Each of these Lie algebras, with the exception of aff(C), the realification of the Lie algebra of the affine motion group of C, has a unique abelian complex structure up to equivalence. On aff(C) there is a two-sphere of abelian complex structures, but only two equivalence classes distinguished by J being proper or not. Furthermore, aff(C) is equipped with a natural bi-invariant complex structure. In dimension 6 it turns out that, as a consequence of our results, some of the Lie algebras equipped with abelian complex structures are of the form aff(A) where A is a 3-dimensional commutative associative algebra. In this article we obtain the classification of the 6-dimensional Lie algebras that can be endowed with an abelian complex structure. Moreover, on each of the resulting Lie algebras, we parameterize the space of such structures up to holomorphic isomorphism. It turns out that there are three nilpotent Lie algebras carrying curves of non-equivalent structures, while for the remaining Lie algebras the moduli space is finite. The outline of the paper is as follows. In §2 we give the basic definitions, recall known results and study some properties of the Lie algebra aff(C). In addition, we parameterize the space of equivalence classes of complex structures on a Lie algebra with 1-dimensional commutator (see Proposition 2.2). In §3 we 1991 Mathematics Subject Classification. Primary 17B30; Secondary 53C15. Key words and phrases. Abelian complex structures, nilpotent and solvable Lie algebras. 1 2 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI perform the classification in the 6-dimensional nilpotent case. This is done in two steps. Firstly, we obtain the classification of the nilpotent Lie algebras admitting such a structure (see Theorem 3.2), recovering, by a different approach, the results obtained in [12, 20]. In the second step, we parameterize, up to equivalence, the space of abelian complex structures on each of these algebras (see Theorem 3.4). As a consequence, we obtain three nilpotent Lie algebras carrying curves of non-equivalent abelian complex structures, namely, h3× h3, h3(C) and a 3-step nilpotent Lie algebra, where h3 is the 3-dimensional real Heisenberg Lie algebra and h3(C) is the realification of the 3-dimensional complex Heisenberg Lie algebra. In §4 we determine the 6-dimensional non-nilpotent Lie algebras s with an abelian complex structure, studying in Theorems 4.1, 4.2 and 4.4 the case when such a structure is proper. The Lie algebras, a finite number, appearing in Theorems 4.1 and 4.2 are decomposable as a direct product of lower dimensional Lie algebras with the corresponding abelian complex structure preserving this decomposition. In Theorem 4.4, when considering the case s′ J = R4, we obtain a 2-parameter family of solvable Lie algebras carrying both an abelian and a bi-invariant complex structure. Moreover, as a consequence of this result and Theorems 4.1 and 4.2, it follows that the Lie algebra aff(C) × R2 has exactly three abelian complex structures, up to equivalence, distinguished by the ideal s′ J . Finally, in Theorem 4.6 we consider the case when J is not proper and obtain five isomorphism classes of Lie algebras, all of which are of the form aff(A) where A is a commutative associative algebra with identity. Acknowledgement. The authors would like to thank Professor J. Lauret for helpful observations. They are also thankful to E. Rodriguez Valencia, who pointed out a mistake in a previous version of Theorem 3.4. 2. PRELIMINARIES J[x, y] − [Jx, y] − [x, Jy] − J[Jx, Jy] = 0, ∀ x, y ∈ g. 2.1. Complex structures on Lie algebras. A complex structure on a real Lie algebra g is an endomorphism J of g satisfying J 2 = −I and such that (1) It is well known that (1) holds if and only if g1,0, the i-eigenspace of J, is a complex subalgebra of gC := g ⊗R C. A pair (g, J) will denote a Lie algebra g equipped with a complex structure J. (2) In this case, (g, J) is a complex Lie algebra, that is, g is turned into a C-vector space by setting A complex structure J on g is called bi-invariant if the following condition holds: J[x, y] = [Jx, y], x, y ∈ g. and condition (2) ensures that the Lie bracket is now C-bilinear. (a + ib)x = ax + bJx, x ∈ g, a, b ∈ R, Two complex structures J1 and J2 on g are said to be equivalent if there exists an automorphism α of g satisfying J2 α = α J1. Two pairs (g1, J1) and (g2, J2) are holomorphically isomorphic if there exists a Lie algebra isomorphism α : g1 → g2 satisfying J2 α = α J1. Consider a Lie algebra g equipped with a complex structure J. We denote with g′ the commutator ideal [g, g] of g and with g′ J := g′ + J g′. We will say that J is proper if g′ J g. We note that if (g1, J1) and (g2, J2) are holomorphically isomorphic, then J1 is proper if and only if J2 is. J the J-stable ideal of g defined by g′ The next proposition provides sufficient conditions for a complex structure to be proper. We recall that a Lie algebra is called perfect if it coincides with its commutator ideal. Proposition 2.1. (1) Every complex structure on a nilpotent Lie algebra is proper. (2) Every bi-invariant complex structure on a non-perfect Lie algebra is proper. Proof. The first statement was proved in [20, Theorem 1.3] (see also Lemma 2.4 below). The second assertion follows since (2) implies that the commutator ideal is stable by any bi-invariant complex structure. (cid:3) 3 2.2. Abelian complex structures on Lie algebras. An abelian complex structure on g is an endomorphism J of g satisfying (3) It follows that (3) is a particular case of (1); moreover, condition (3) is equivalent to g1,0 being abelian. These structures were first considered in [3]. A class of Lie algebras carrying such structures appears in the next result. [Jx, Jy] = [x, y], ∀x, y ∈ g. J 2 = −I, Let h2n+1 = span{e1, . . . , e2n, z0} denote the (2n + 1)-dimensional Heisenberg algebra with Lie bracket [e2i−1, e2i] = z0, 1 ≤ i ≤ n, and aff(R) the Lie algebra of the group of affine motions of R. Proposition 2.2. If g is an even dimensional real Lie algebra with 1-dimensional commutator g′, then: (1) g is isomorphic to either h2n+1 × R2k+1 or aff(R) × R2k; (2) All these Lie algebras carry abelian complex structures and every complex structure on g is abelian; (3) There are(cid:2) n (4) There is a unique complex structure on aff(R) × R2k up to equivalence. 2(cid:3) + 1 equivalence classes of complex structures on h2n+1 × R2k+1; Proof. (1) follows from the characterization of the Lie algebras with one dimensional commutator (see for example [4, Theorem 4.1]). Je2i−1 = ±e2i, Jz2j = z2j+1, 1 ≤ i ≤ n, 0 ≤ j ≤ k, A proof of (2) can be found in [5, Examples 3.3 and Proposition 3.4]. In order to prove (3), we recall that according to [19, Proposition 3.6] the complex structures on h2n+1 × R2k+1 are given by: (4) where {e1, . . . , e2n, z0} is a basis of h2n+1 such that [e2i−1, e2i] = z0 and {z1, . . . z2k+1} is a basis of R2k+1. Any two complex structures with the same number of minus signs are equivalent, by permuting the elements of the basis. Let Jr, with 0 ≤ r ≤ n, be any complex structure as in (4) such that r is the number of minus signs. By a suitable permutation of the basis, one can show that Jr is equivalent to Jn−r, thus we may assume 0 ≤ r ≤(cid:2) n 2(cid:3), and assume that Jr is equivalent to Js, that is, there exists an automorphism φ of h2n+1 × R2k+1 such that φJr = Jsφ. Since φ preserves the commutator, we have φz0 = cz0 for some c 6= 0. Set v := span{e1, . . . , e2n} and for α = r, s we define a bilinear form Bα on v by [x, Jαy] = Bα(x, y)z0, x, y ∈ v. It turns out that Bα is symmetric (since Jα is abelian) and non-degenerate. Moreover, Bs (φ(x), φ(y)) = cBr(x, y). Therefore, Br and Bs have the same signature if c > 0 or opposite signatures if c < 0. Since r, s ≤(cid:2) n 2(cid:3), and Sign (Bα) = (2α, 2(n − α)), in both cases we The proof of (4) follows by observing that given a complex structure J on aff(R) × R2k, the ideal g′ J = g′ + J g′, isomorphic to aff(R), is J-stable and complementary to the center R2k. It is clear that aff(R) admits a unique complex structure up to equivalence, and the assertion follows. (cid:3) 2(cid:3). Furthermore, let 0 ≤ r, s ≤(cid:2) n must have r = s and (3) follows.1 We include some properties of abelian complex structures in the following lemma, whose proof is straightforward. Lemma 2.3. Let J be an abelian complex structure on a Lie algebra g. Then: (i) the center z(g) of g is J-stable; (ii) for any x ∈ g, adJ x = − adx J; in particular, the family of inner derivations of g is stable under right (iii) for any J-stable ideal u of g, the kernel of the map g → End(u), x 7→ adx u, is J-stable. multiplication by J; Some known obstructions to the existence of abelian complex structures are stated below. (O1) Let g be a real Lie algebra admitting an abelian complex structure. Then g is 2-step solvable [17]. 1 The classification for n = 1, k = 0, was given in [21] and the case n = 2, k = 0 was carried out in [22]. 4 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI (O2) Let g be a solvable Lie algebra such that g′ has codimension 1 in g and dim g > 2. Then g does not admit abelian complex structures [5]. Another obstruction in the case of nilpotent Lie algebras is given in the next lemma. Recall that the J descending central series of g is defined by g0 = g, gi = [g, gi−1], i ≥ 1. Lemma 2.4. Let g be a k-step nilpotent Lie algebra with an abelian complex structure J and set gi J gi−1 gi + J gi. Then gi for some i < k. Then, any x ∈ gi−1 can be Proof. Clearly, {0} = gk written as x = y + Jw, y, w ∈ gi. The fact that J is abelian implies that [g, x] ⊂ gi+1 for all x ∈ gi−1, that is, gi ⊂ gi+1, contradicting that g is k-step nilpotent. 2.3. Dimension less than or equal to four. There are only two 2-dimensional Lie algebras: R2 and aff(R). Both carry a unique abelian complex structure, up to equivalence. for all i ≤ k. In particular, if dim g = 2m, g is at most m-step nilpotent. J gk−1 . Assume that gi J = gi−1 J := (cid:3) J J Let g be a 4-dimensional solvable Lie algebra with basis {e1, e2, e3, e4}. According to [21], if g admits an abelian complex structure, it is isomorphic to one and only one of the following: g1 = R4, g2 = h3 × R, [e1, e2] = e3, g3 = aff(R) × R2, [e1, e2] = e2 g4 = aff(R) × aff(R), [e1, e2] = e2, [e3, e4] = e4 g5 = aff(R) ⋉ad R2, [e1, e2] = e2, [e1, e4] = e4, [e2, e3] = e4. g6 = aff(C), [e1, e3] = e3, [e1, e4] = e4, [e2, e3] = e4, [e2, e4] = −e3, where h3 is the 3-dimensional Heisenberg algebra and aff(C) is the realification of the Lie algebra of the group of affine motions of the complex line. Theorem 2.5 ([21]). Let g be a 4-dimensional solvable Lie algebra with an abelian complex structure J. Then (g, J) is holomorphically isomorphic to one and only one of the following: g1 : Je1 = e2, Je3 = e4, g2 : Je1 = e2, Je3 = e4, g3 : Je1 = e2, Je3 = e4, g4 : Je1 = e2, Je3 = e4, g5 : Je1 = e2, Je3 = −e4, g6 : J1e1 = −e2, J1e3 = e4, g6 : J2e1 = e3, J2e2 = e4. We include below a proof of the fact that, up to equivalence, aff(C) has two abelian complex structures (see also [21]). We recall that aff(C) has a basis {e1, e2, e3, e4} with the following Lie bracket (5) and with automorphism group given by (see for instance [7]): [e2, e4] = −e3, [e2, e3] = e4, [e1, e3] = e3, [e1, e4] = e4, (6) Aut(aff(C)) =    0 ǫ 0 0 1 0 0 0 c −ǫd a −ǫb ǫa d ǫc b   : a, b, c, d,∈ R, ǫ = ±1  . Proposition 2.6. Any abelian complex structure on the Lie algebra aff(C) is equivalent to either J1 or J2, where Moreover, J1 is proper and J2 is not, hence they are not equivalent. J1e1 = −e2, J2e1 = e3, J1e3 = e4, J2e2 = e4. with (7) be the matrix representation of J with respect to the ordered basis {e1, e2, e3, e4}. Since J 2 = −I, we obtain that A2 = −I and C 2 = −I, and therefore J can be written as  α 0 0 β −α 0 0  e a g f −e b J =  c d γ , α2 + βγ = −1, e2 + f g = −1, and certain constraints on a, b, c, d. From the fact that J is abelian, together with (7), we obtain that α = e = 0, β = ±1, f = −g = γ = − 1 β . (a,b), where Using again that J 2 = −I, we arrive at c = −b, d = a, thus, J = J + (a,b) :=   0 0  1 a −1 0 (0,0) = J1. Consider now the automorphisms φ and ψ of aff(C) given by (a,b) or J = J − 0 −1 0 1 a −b b Note that J + 0 0 0 J + 0 0 a 1 . , 1 0 J − (a,b) := 0 0 −1 0  a −b 0 −1 0 b φ =     , ψ =   b 2 a 0 a 2 1 0 0 0 −1 0 0 1 0 2 − b 0 −1 (a,b) = J1ψ, so that all the abelian complex structures J + 1 0 0 1 2 − a − b 2 − b 0 0 0 0 1 0 0 1   2 2 2 . a (a,b) and (a,b) = J1φ and ψJ − It is easy to see that φJ + J − (a,b) are equivalent to J1. This completes this case. • g′ J = g. In this case g = g′ ⊕ J g′. Let f1 = Je3 = ae1 + be2 + v, f2 = Je4 = ce1 + de2 + w, a, b, c, d ∈ R, v, w ∈ span{e3, e4}. Since J is abelian, [f1, f2] = 0 and [f2, e3] = [f1, e4], hence c = −b, d = a. Therefore, we have Proof. Let us assume that J is an abelian complex structure on g = aff(C). Since dim g′ = 2, there are two possibilities for g′ J : • g′ J = g′. In this case, g′ is a J-stable ideal of aff(C). Let 5 J =(cid:18)A 0 B C(cid:19) adf1g′ =(cid:18)a −b a (cid:19) , b adf2g′ =(cid:18)−b −a a −b(cid:19) . Observe that a2 + b2 6= 0, since otherwise g would be abelian. Setting f1 = we obtain 1 a2 + b2 (af1 − bf2), ad f1s′ =(cid:18)1 0 0 1(cid:19) , 1 f2 = a2 + b2 (bf1 + af2), ad f2s′ =(cid:18)0 −1 0 (cid:19) . 1 6 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI The expression of J with respect to the basis { f1, f2, e3, e4} is given by (cid:19) , where B =(cid:18)−a J = J(a,b) =(cid:18) −B−1 B b −b −a(cid:19) . The structure constants of g with respect to the new basis { f1, f2, e3, e4} are given by (5). following automorphism of aff(C): If φ is the 1 0 0 1 φ =  ,   −a b −b −a it turns out that φJ(a,b) = J(1,0)φ, that is, J(a,b) is equivalent to J(1,0) = J2, and the proposition follows. (cid:3) Given x = (x1, x2, x3) ∈ S2, let Jx := x1J1 + x2J2 + x3J1J2 with J1 and J2 as above. Since J1J2 = −J2J1, Jx is an abelian complex structure on aff(C) and Proposition 2.6 implies the following result: Corollary 2.7. There is a two-sphere Jx, x ∈ S2, of abelian complex structures on aff(C). Moreover, Jx is equivalent to J1 if and only if x = (1, 0, 0) or x = (−1, 0, 0). All the other abelian complex structures in this sphere are equivalent to J2. Remark 2.8. It was shown in [2] that aff(C) is the only 4-dimensional Lie algebra carrying an abelian hypercomplex structure, that is, a pair of anticommuting abelian complex structures. Remark 2.9. We point out that aff(C) admits a bi-invariant complex structure J, namely, Je1 = e2, Je3 = e4. Moreover, any bi-invariant complex structure is given by ±J and J is equivalent to −J via an automorphism as in (6) with ǫ = −1. Therefore, aff(C) has a unique bi-invariant complex structure up to equivalence. The next characterization of (aff(C), J1) will be frequently used in §4. Lemma 2.10. Let g be a 4-dimensional Lie algebra spanned by {f1, . . . , f4} with Lie bracket defined by: [f1, f3] = [f2, f4] = cf3 + df4, [f1, f2] = xf3 + yf4, [f1, f4] = −[f2, f3] = −df3 + cf4, with c2 + d2 6= 0, and abelian complex structure J given by Jf1 = f2, Jf3 = f4. Then (g, J) is holomorphically isomorphic to (aff(C), J1). Proof. Consider first f1 = cf1 + df2 c2 + d2 , f2 = J f1 = −df1 + cf2 c2 + d2 . In the new basis { f1, f2, f3, f4}, the Lie bracket of g becomes [ f1, f3] = [ f2, f4] = f3, [ f1, f2] = x′f3 + y′f4, Setting now [ f1, f4] = −[ f2, f3] = f4. e1 := f1 − y′ 2 f3 + x′ 2 f4, e2 := − f2 + x′ 2 f3 + y′ 2 f4, e3 := f3, e4 := f4, we obtain a basis {e1, . . . , e4} of g such that the Lie bracket is given as in (5) and the complex structure J coincides with J1 from Proposition 2.6. (cid:3) 3. DIMENSION SIX: THE NILPOTENT CASE In this section we classify, up to holomorphic isomorphism, the 6-dimensional nilpotent non-abelian Lie algebras n with abelian complex structures. 7 J = 2 then s′ J is abelian. We begin by recalling from Proposition 2.1 that any complex structure J on a nilpotent Lie algebra n is J = n′ + J n′ is a proper nilpotent ideal. In the 6-dimensional case, as a consequence of the proper, that is, n′ following general result, it turns out that n′ Proposition 3.1. If J is an abelian complex structure on a 6-dimensional solvable Lie algebra s such that s′ J is nilpotent, then s′ Proof. If s′ dim s′ J = s then s is nilpotent and Proposition 2.1 implies that s′ J s, a contradiction. Therefore, J is abelian. J is abelian. J = 2 or 4. If dim s′ J = 4 and s′ We show next that if dim s′ J ∼= h3 × R, and there exists a basis e1, . . . , e4 of s′ Je3 = e4. Let span{f1, f2} be a complementary subspace of s′ , it follows from Lemma 2.3(ii) that adf2 s′ J is non-abelian nilpotent, then we get a contradiction. J induces by restriction an abelian complex structure on s′ J . The only 4-dimensional non-abelian nilpotent Lie algebra admitting abelian complex structures is h3 × R and this complex structure is unique up to equivalence (Theorem 2.5). Therefore, s′ J such that [e1, e2] = e3 and J such that f2 = Jf1. Setting Je1 = e2, = −DJ. The Lie bracket [f1, f2] lies in s′ ⊆ s′ J , D = adf1 s′ hence, [f1, f2] = ae1 + be2 + ce3 + de4, with a, b, c, d ∈ R. D =(cid:18)A 0 B C(cid:19) , Since D is a derivation of h3 × R, it is of the form where A, B, C are 2 × 2 real matrices such that c11 = tr A and c21 = 0. Moreover, DJ is also a derivation of h3 × R, and therefore c12 = a12 − a21 and c22 = 0. Observe that s′ J is 2-step solvable and u := span{e3, e4} ⊂ s′ J )′, therefore the Jacobi identity implies2 J is an abelian ideal containing (s′ J J C 2Ju = C(Ju)C, and it follows that C = 0. Therefore, A takes the following form: A =(cid:18)a11 a12 −a11(cid:19) . a12 Since s′ is abelian, we have that rank A ≤ 1, hence a11 = a12 = 0, that is, A = 0. Finally, the Jacobi identity applied to f1, f2, ej, j = 1, 2, implies that a = b = 0, that is, [f1, f2] ∈ u, hence s′ J = u, a contradiction. (cid:3) In the remainder of this section we perform the classification. This will be done in two steps. In the first one, we obtain the classification of the Lie algebras admitting such a structure (see Theorem 3.2), recovering by a different approach the results obtained in [12, 20]. In the second step, we parameterize the space of abelian complex structures up to equivalence on each of these algebras (see Theorem 3.4). Theorem 3.2. Let n be a 6-dimensional nilpotent Lie algebra with an abelian complex structure J. Then n is isomorphic to one and only one of the following Lie algebras: n1 := h3 × R3 : n2 := h5 × R : n3 := h3 × h3 : n4 := h3(C) : n5 : n6 : n7 : [e1, e2] = e5, [e1, e2] = e5, [e1, e2] = e4, [e1, e2] = e6, [e1, e2] = e6 = [e3, e4], [e1, e2] = e5, [e1, e3] = −[e2, e4] = e5, [e3, e4] = e6, [e1, e4] = [e2, e3] = e6, [e1, e4] = [e2, e5] = e6, [e1, e3] = −[e2, e4] = e5, [e1, e4] = [e2, e3] = e6, [e1, e4] = [e2, e3] = e6. 2 Given a 2-step solvable Lie algebra g and an abelian ideal u of g such that g′ ⊆ u, the Jacobi identity implies that {adx u : x ∈ g} is a commutative family of endomorphisms of u. 8 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI Proof. According to Lemma 2.4, n is at most 3-step nilpotent. To carry out the classification, we will consider separately the k-step nilpotent case for k = 2 or 3. • n is 2-step nilpotent: We observe that dim n′ = 1 or 2. In fact, the center z is a J-stable proper ideal, hence dim z = 2 or 4. If dim z = 4, then dim n′ = 1, while if dim z = 2, we have that dim n′ = 1 or 2 since n′ ⊂ z. If dim n′ = 1, according to [4, Theorem 4.1], n is isomorphic to n1 or n2. If dim n′ = 2, since z is J-stable, then n′ = z and we have that dim z = 2. Let v be a J-stable complementary subspace of z. There are two possibilities: It follows immediately that We consider next each case separately. Js,t e1 = e2, Js,t e3 = e4, Js,t e5 = se5 + te6, (i) there exists x ∈ v such that rank adx = 1; (ii) rank adx = 2 for any x ∈ v, x 6= 0. − Case (i): Consider the 3-dimensional subspace W := ker (adx v). dim(W ∩ JW ) = 2. If x /∈ W ∩ JW , there exists e1 ∈ W ∩ JW such that {e1, Je1, x, Jx} is a basis of v. Since J is abelian, it follows that the only non-zero brackets are [e1, Je1], [x, Jx], which must be linearly independent. Setting e3 := x, e4 := Jx, e5 := [e1, Je1], e6 := [x, Jx], we obtain that n ∼= n3 and J = Js,t is given by (8) for some s, t ∈ R, t 6= 0, since the center z = span{e5, e6} is J-stable. If x ∈ W ∩ JW , then {x, Jx} is a basis of W ∩ JW . There exists e1 ∈ W such that Je1 /∈ W and therefore, {e1, Je1, x, Jx} is a basis of v. Since J is abelian, it follows that the only non-zero brackets are [e1, Je1] and [e1, Jx] = −[Je1, x], which must be linearly independent. Setting e2 := Je1, e3 := −x, e4 := Jx, e5 := [e1, Je1], e6 := [e1, Jx], we obtain that n ∼= n5 and J = Js,t is given by (9) for some s, t ∈ R with t 6= 0, since the center z = span{e5, e6} is J-stable. − Case (ii): In this case, there always exists e1 ∈ v such that ker (ade1 v) is not J-stable. Indeed, if ker (adx v) were J-stable for any x ∈ v, then we would have [x, Jx] = 0 for any x ∈ g. Therefore, for any x, y ∈ g, Js,t e1 = e2, Js,t e3 = −e4, Js,t e5 = se5 + te6, 0 = [x + y, J(x + y)] = [x, Jy] + [y, Jx] = 2[x, Jy], and this implies that g is abelian, which is a contradiction. Let us denote W := ker (ade1 v), which is not J-stable. Then v = W ⊕ JW and we fix a basis {e1, y, Je1, Jy} of v with y ∈ W . It follows that [e1, y] = 0 = [Je1, Jy]. A basis of z = n′ is given by {[e1, Je1], [e1, Jy]} since rank ade1 = 2. Moreover, [e1, Jy] = [y, Je1] since J is abelian, and rank ady = 2 implies [y, Jy] 6= 0. Therefore, there exist a, b ∈ R, a2 + b2 6= 0 such that [y, Jy] = a[e1, Je1] + b[e1, Jy]. Since rank adw = 2 for all w ∈ W, w 6= 0, it follows that det(adce1+y JW ) 6= 0 for any c. We have that det(adce1+y JW ) = c2 + cb − a, with respect to the bases {Je1, Jy}, {[e1, Je1], [e1, Jy]} of JW and z, respectively. Therefore, it must be b2 + 4a < 0, and setting e2 := 1 p−(b2 + 4a) it turns out that (−be1 + 2y), e3 := Je1, e4 := Je2, e5 := [e1, e3], e6 := [e1, e4], [e1, e3] = −[e2, e4] = e5, [e1, e4] = [e2, e3] = e6. Hence, n ∼= n4 and J = Js,t is given by (10) for some s, t ∈ R with t 6= 0, since the center z = span{e5, e6} is J-stable. Js,t e1 = e3, Js,t e2 = e4, Js,t e5 = se5 + te6, 9 • n is 3-step nilpotent: In this case, n2 = [n, n′] 6= 0 and n2 ⊆ z. The center z being J-stable, it follows J = 2. J and fix J = n2 + J n2 ⊆ z. Since dim z 6= 4 (otherwise, dim n′ = 1) it follows that n2 J is abelian (Proposition 3.1), we obtain J = 4. Let V0 be a J-stable subspace complementary to n′ J = z and dim n2 J n′ that n2 From Lemma 2.4, n2 a basis {f1, Jf1 = f2} of V0. Using that n′ J , V0 ⊕ n′ n′ = [V0 ⊕ n′ J , thus dim n′ J ] = [V0, V0] + [V0, n′ J ] ⊆ [V0, V0] + n2, J , where V1 is the subspace spanned J and that for any y ∈ V0, y 6= 0, the images of − If dim n2 = 1, we may assume that [f2, [f1, f2]] 6= 0. By making a suitable change of basis of V0, we thus [f1, f2] /∈ z, hence J[f1, f2] /∈ z, which implies that n′ by [f1, f2] and J[f1, f2]. Note that n = V0 ⊕ V1 ⊕ n2 ady : V1 → n2 obtain furthermore that [f1, [f1, f2]] = 0. Setting J coincide with n2. J = V1 ⊕ n2 e1 := f1, e2 := f2, e3 := J[f2, [f1, f2]], e4 := −J[f1, f2], e5 := [f1, f2], e6 := [f2, [f1, f2]], we obtain that n ∼= n6 and J is given by (11) Je1 = e2, Je3 = −e6, Je4 = e5. − If dim n2 = 2 the map adf1 : V1 → n2 J is an isomorphism and setting e1 := f1, e2 := f2, e3 := J[f1, f2], e4 := [f1, f2], e5 := [f1, J[f1, f2]], e6 := [f1, [f1, f2]] one has a basis of n7. Since J is abelian, it leaves z = span{e5, e6} stable, hence it is given by J = Js,t, with (12) for some s, t ∈ R, t 6= 0. Js,te1 = e2, Js,te3 = −e4, Js,te5 = se5 + te6 (cid:3) Remark 3.3. It follows from Theorem 3.2 that if n is a 6-dimensional nilpotent non-abelian Lie algebra and J is any abelian complex structure, then • n′ • n′ J ∼= R2 if and only if n = nk, 1 ≤ k ≤ 5, J ∼= R4 if and only if n = n6 or n7. 3.1. Equivalence classes of complex structures. In this subsection we parameterize the equivalence classes of abelian complex structures on the Lie algebras appearing in Theorem 3.2. Let us begin with n1 and n2. According to [22] (see also Proposition 2.2) it follows that, up to equivalence, n1 has a unique complex structure J, given by Je1 = e2, Je3 = e4, Je5 = e6, whereas n2 has two, given by J± e1 = e2, J± e3 = ± e4, J± e5 = e6. For the Lie algebra n6, it follows from the proof of Theorem 3.2 that it has a unique abelian complex structure, given by (11). We will determine next the equivalence classes of abelian complex structures on the remaining Lie alge- bras. To achieve this, we consider a nilpotent Lie algebra n and the following space: Ca(n) = {J : n → n : J is an abelian complex structure on n}, which we suppose is non-empty. Ca(n) is a closed subset of GL(n). The group Aut(n) acts on Ca(n) by conjugation, and the quotient Ca(n)/ Aut(n) parameterizes the equivalence classes of abelian complex structures on n. 10 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI For n = n3, n4, n5 and n7, it follows from equations (8), (10), (9) and (12) in the proof of Theorem 3.2 that Z := {(s, t) ∈ R2 : t 6= 0} can be considered as a subspace of Ca(n) such that Aut(n)·Z = Ca(n). The induced topology on Z coincides with the usual topology from R2. Therefore, we have a homeomorphism Ca(n)/ Aut(n) ∼= Z/G, Let J be an abelian complex structure on n and v a J-stable complementary subspace of the center z in where G := {φ ∈ Aut(n) : φ(Z) = Z}, a closed Lie subgroup of Aut(n)3. n. It can be shown that Aut0(n) acts transitively on the orbit of J in Ca(n), that is, Aut(n)J = Aut0(n)J, where Aut0(n) = {φ ∈ Aut(n) : φ(v) = v}. Indeed, if J ′ = φJφ−1 with φ = (cid:18)A 0 φ0 =(cid:18)A 0 0 C(cid:19) ∈ Aut0(n) satisfies J ′ = φ0J(φ0)−1. This implies that B C(cid:19) ∈ Aut(n), then Ca(n)/ Aut(n) ∼= Z/G0 for the Lie algebras ni, i = 3, 4, 5, 7, where G0 = G ∩ Aut0(n). Z/G0. We will determine in what follows for each algebra ni, i = 3, 4, 5, 7, the group G0 and the orbit space     a −b a b • n3: In this case, the group G0 is given by a disjoint union G0 = G+ ∪ G−, where , G− =  a2 + b2 6= 0, c2 + d2 6= 0 c −d c d a2 + b2 c2 + d2 G+ = : I I   0 1 1 0 · G+. Let (s, t), (s′, t′) ∈ Z such that φ · (s, t) = (s′, t′), with φ ∈ G0. It can be shown that: (i) if φ ∈ G+, then s′ = s and tt′ > 0; (ii) if φ ∈ G−, then s′ = −s and tt′ < 0. Let us fix now (s, t) ∈ Z. Taking φ+ ∈ G+ with a = pt, c = 1 and b = d = 0, we obtain that t ). Taking now φ− ∈ G− with a = 1, c = √1 + s2 and b = d = 0, we obtain that φ+ · (s, t) = (s, t φ− · (s,−1) = (−s, 1). To sum up, we have that     Js,t is equivalent to either (Js,1 J−s,1 if t > 0, if t < 0. It follows from (i) and (ii) that Js,1 and Js′,1 are equivalent if and only if s = s′, hence Ca(n3)/ Aut(n3) ∼= {Js,1 : s ∈ R}. • n5: In this case, the action of G on Z is transitive, since the automorphism φ of n5 given by 0 −1 0 1 0 − s − s 2 2 0 0 − 1+s2 t φ =   1+s2 t 0 1 −s 0 1+s2 t   3It can be shown that if φ ∈ Aut(n) satisfies φ · (s, t) = (s′, t′) for some (s, t), (s′, t′) ∈ Z, then φ ∈ G. satisfies φ · (s, t) = (0, 1) for any (s, t) ∈ Z. Thus, Ca(n5)/ Aut(n5) = {J(0,1)}. • n7: In this case the subgroup G0 of G is given by a −b a b C× = d d       ad −bd ad bd : a, b ∈ R, d = a2 + b2 6= 0 11 .   We note that (0, 1) and (0,−1) are fixed points by the action of C×. If (s, t), (s′, t′) ∈ Z − {(0,±1)}, (s, t) 6= (s′, t′) and φ · (s, t) = (s′, t′), with φ ∈ C×, then by writing down the equations, it can be shown that tt′ > 0 and both (s, t) and (s′, t′) lie in the circle (13) Sc :=(cid:26)(u, v) ∈ Z : u2 +(cid:16)v − c 2(cid:17)2 =(cid:16) c 2(cid:17)2 − 1(cid:27) , c = t + 1 + s2 t = t′ + 1 + s′2 t′ . Each of these circles intersects the v-axis in points (0, t0) and (0, 1/t0), with 0 < t0 < 1, where t0 = 2(cid:16)c ∓ √c2 − 1(cid:17), depending on the sign of t. We show next that each Sc coincides with an orbit O0,t0 of 1 a point under the action of C×. Since C× ∼= R+ × S1 as Lie groups, and the action of R+ on Z is trivial, we only have to consider the action of S1 on Z. The isotropy group of this action is {±1} and therefore each orbit is homeomorphic to RP 1 ∼= S1. Furthermore, each orbit is contained in one circle Sc for some c, hence, by using topological arguments, we have that each Sc coincides with an orbit O0,t0 (see Figure 3.1), so that, We note that the topology of the quotient space coincides with the relative topology of R, in particular, the moduli space of abelian complex structures on n7 is disconnected. Ca(n7)/ Aut(n7) ∼= {J0,t : 0 < t ≤ 1}. • n4: The group G0 is given by a disjoint union G0 = G+ ∪G−, where   =(cid:16) c 2(cid:17)2 a −ǫb ǫa b a2 − b2 2ab −2ǫab ǫ(a2 − b2) : a, b ∈ R, a2 + b2 6= 0, ǫ = ±1 −I I , G− =    1 0 0 1 · G+ 1 . We note that the G0-orbit of (0, 1) has two points: (0,±1). The orbit Os,t of (s, t) ∈ Z1 − {(0,±1)} is given by the disjoint union of two circles (see Figure 3.1):   a −ǫb ǫa b     G+ = (14) Os,t :=(cid:26)(u, v) ∈ Z1 : u2 +(cid:16)v ± c 2(cid:17)2 − 1(cid:27) , . Each of these orbits intersects the positive v-axis in points (0, t0) and (0, 1/t0), with 1 + s2 t with c = t + 0 < t0 < 1. Thus, Z/G0 is parameterized by (0, 1]. The results in this section are summarized in the next theorem. Theorem 3.4. Let n be a 6-dimensional nilpotent Lie algebra with an abelian complex structure J. Then (n, J) is holomorphically isomorphic to one and only one of the following: (1) (n1, J), with its unique complex structure: Je1 = e2, Je3 = e4, Je5 = e6, (2) (n2, J±), with J± e1 = e2, J± e3 = ±e4, J± e5 = e6, (3) (n3, Js), with Js e1 = e2, Js e3 = e4, Js e5 = se5 + e6, s ∈ R, 12 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI FIGURE 1. Orbits in Ca(n7) u O0,4 O0,3 O0,2 -4 -3 -2 -1 1 2 3 4 v (4) (n4, Jt) with Jt e1 = e3, Jt e2 = e4, Jt e5 = te6, t ∈ (0, 1], (5) (n5, J) with Je1 = e2, Je3 = −e4, Je5 = e6, (6) (n6, J), with Je1 = e2, Je3 = −e6, Je4 = e5, (7) (n7, Jt) with Jte1 = e2, Jte3 = −e4, Jte5 = te6, 0 < t ≤ 1. Remark 3.5. There is another one-parameter family of abelian complex structures on n4, given by J 2 t e5 = te6. J 2 t e1 = e2, J 2 t e3 = −e4, In a previous version of this article, we stated that the abelian complex structures Jt and J 2 t , t ∈ (0, 1], were not equivalent, but this statement was incorrect. E. Rodr´ıguez Valencia provides in [18] an automorphism of n4 which shows the equivalence between these structures. Remark 3.6. In [15] curves of non-equivalent abelian complex structures on the 6-dimensional 2-step nilpo- tent Lie algebras n3 and n4 were obtained by using geometric invariant theory. Remark 3.7. We observe that n4 has a bi-invariant complex structure given by J0e1 = e2, J0e3 = e4, J0e5 = e6. According to [14] J0 is the unique bi-invariant complex structure on n4 compatible with the standard orien- tation (see also [1]). Remark 3.8. It was proved in [22] that every complex structure on n6 is abelian. This fact, together with Theorem 3.4, implies that n6 admits a unique complex structure up to equivalence. The existence of a unique abelian complex structure on n6 was already proved in [8, Section 5.3], where this Lie algebra was denoted h9. Notation. The Lie algebras in Theorem 3.2 appear in the literature with different notations. In [11], the nilpotent Lie algebras equipped with the so-called nilpotent complex structures were denoted by hk, k = 1, . . . , 16. Salamon in [20] associates to each Lie algebra a 6-tuple encoding the expression of the differential of 1-forms. We give next the correspondence among the various notations: n1 ←→ h8 ←→ (0, 0, 0, 0, 0, 12), n2 ←→ h3 ←→ (0, 0, 0, 0, 0, 12 + 34), n3 ←→ h2 ←→ (0, 0, 0, 0, 12, 34), n4 ←→ h5 ←→ (0, 0, 0, 0, 13 + 42, 14 + 23), n5 ←→ h4 ←→ (0, 0, 0, 0, 12, 14 + 23), n6 ←→ h9 ←→ (0, 0, 0, 0, 12, 14 + 25), n7 ←→ h15 ←→ (0, 0, 0, 12, 13 + 42, 14 + 23). 4. DIMENSION SIX: THE GENERAL CASE 13 The classification, up to equivalence, of the 6-dimensional non-nilpotent Lie algebras s admitting abelian complex structures J will be done in several steps, according to the different possibilities for s′ J . When J is proper it follows from §2.3 that the ideal s′ J with its induced abelian complex structure coincides, up to equivalence, with R2, aff(R), R4, h3×R, aff(R)×R2, aff(R)×aff(R), aff(R)⋉ad R2 endowed with their unique abelian complex structure or with (aff(C), J1), (aff(C), J2). Note that according to Proposition 3.1, J ∼= h3 × R does not occur. In order to carry out the classification we will consider three different the case s′ cases: dim s′ J = s. J = 2, dim s′ J = 4 or s′ 4.1. dim s′ J = 2. Theorem 4.1. If s is a 6-dimensional non-nilpotent Lie algebra admitting an abelian complex structure J such that dim s′ J = 2, then (s, J) is holomorphically isomorphic to one of the following: • aff(R) × R4 with its unique complex structure, • aff(C) × R2 equipped with the product (J1 × J), J1 from Theorem 2.5. Proof. Since dim s′ J = 2, we have two possibilities: (1) dim s′ = 1. Since s is non-nilpotent, s is isomorphic to aff(R) × R4 (see, for instance, [4, The- orem 4.1]). According to Proposition 2.2, this Lie algebra has a unique complex structure, up to equivalence. (2) dim s′ = 2, hence s′ = s′ J is J-stable. Let s′ be spanned by e1, e2 and Je1 = e2. Let u be a complementary J-stable subspace of s, s = s′ ⊕ u. From Lemma 2.3(iii), the kernel of the linear map ρ : u → End(s′), u 7→ adu s′ is J-stable. The Jacobi identity implies that Imρ is a commutative family of endomorphisms of s′, therefore dim(Imρ) < 4, thus dim(ker ρ) is 4 or 2. When dim(ker ρ) = 4, s is 2-step nilpotent, contradicting the assumption on s. Indeed, if ρ ≡ 0, then [x, u] = 0 for all x ∈ s′ and u ∈ u. Since s′ is abelian, we obtain that s′ coincides with the center of s and the claim follows. Therefore, we must have dim(ker ρ) = 2. We show next that (s, J) is holomorphically isomorphic to aff(C) × R2 equipped with the product (J1 × J) from Theorem 2.5. Indeed, let f1, f2 = Jf1, f3, f4 = Jf3 be a basis of u with f1, f2 ∈ ker ρ. Using the Jacobi identity one obtains [[f1, f2], f3] = [[f1, f2], Jf3] = 0, therefore, since J is abelian, [f3, [f1, f2]] = [f3, J[f1, f2]] = 0. Thus, [f1, f2] = 0, since otherwise f3 ∈ ker ρ, which is a contradiction. Since ρ(f3)ρ(f4) = ρ(f4)ρ(f3) and ρ(f4) = −ρ(f3)J one has ρ(f3) =(cid:18)x −y x (cid:19) , y with x2 + y2 6= 0. We note that k := span{f3, f4, e1, e2} is a J-stable Lie subalgebra of s, and using Lemma 2.10 we obtain that (k, Jk) is holomorphically isomorphic to (aff(C), J1), so that we may assume [f3, f4] = [e1, e2] = 0, [f3, e1] = −[f4, e2] = e1, [f3, e2] = [f4, e1] = e2. The remaining brackets are given by [f1, f3] = ae1 + be2 = −[f2, f4], [f1, f4] = −be1 + ae2 = [f2, f3], 14 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI where the last line is a consequence of [[f2, f3], f4] + [[f4, f2], f3] = 0 and J abelian. If a = b = 0, then s = aff(C) × R2. If a2 + b2 6= 0, one considers f1 = e1 + 1 a2 + b2 (af1 − bf2), f2 = J f1 = e2 + 1 a2 + b2 (bf1 + af2), and finds that f1, f2 are central elements in s. Therefore s ∼= aff(C)×R2 equipped with the complex structure given in the statement. (cid:3) J = 4. J = 4. Let f1, f2 = Jf1 span a complementary subspace of s′ 4.2. dim s′ Let s be a 6-dimensional solvable Lie algebra equipped with an abelian complex structure J such that dim s′ . It then follows J . The bracket from Lemma 2.3(ii) that adf2 s′ [f1, f2] lies in s′ ⊆ s′ (15) J , and set D := adf1 s′ = −DJ; therefore D and DJ are both derivations of s′ J , we obtain that J , and using the Jacobi identity for f1, f2 and any x ∈ s′ J J ad[f1,f2] s′ J = DJD − D2J. In what follows we analyze separately two different cases, corresponding to s′ J being abelian or non- abelian. Theorem 4.2. Let s be a 6-dimensional Lie algebra with an abelian complex structure J such that s′ 4-dimensional and non-abelian. Then (s, J) is holomorphically isomorphic to one of the following: J is (1) aff(R) × (h3 × R), with the unique complex structure on each factor, (2) aff(R) × aff(C), where the complex structure on aff(C) is given by J1 (Theorem 2.5), (3) aff(R) × aff(R) × R2, where each factor carries its unique complex structure, (4) (aff(R) ⋉ad R2) × R2, where each factor is equipped with its unique abelian complex structure, (5) aff(C) × R2, where the first factor is equipped with J2 from Theorem 2.5. Proof. We will consider several cases, according to the isomorphism class of s′ J . (i) Case s′ J ∼= aff(R) × R2: There exists a basis {e1, . . . , e4} of s′ by Je1 = e2, Je3 = e4. A derivation D of s′ J such that DJ is also a derivation takes the following form: J such that [e1, e2] = e2 and J is given D =  0 0 a b ,   c e d f with a, b, ..., f ∈ R. From (15), we obtain that e = −d, f = c, [f1, f2] = (a2 + b2)e2 + u, u ∈ span {e3, e4}. Since s′ J = aff(R) × R2, it follows that c2 + d2 6= 0 or u 6= 0. We show next that we may assume a = b = 0. Indeed, if a2 + b2 6= 0, consider another complementary subspace of s′ J spanned by f1 = e1 − bf1 + af2 a2 + b2 , f2 = J f1 = e2 + af1 − bf2 a2 + b2 . The only non-vanishing brackets with respect to the basis above are [e1, e2] = e2, [ f1, f2] = u, [ f1, e3] = [ f2, e4] = ce3 + de4, [ f1, e4] = −[ f2, e3] = −de3 + ce4, with u ∈ span{e3, e4}. Note that s can be decomposed as span{e1, e2}⊕ span{ f1, f2, e3, e4}. If c = d = 0, then u 6= 0 and there is a holomorphic isomorphism between span{ f1, f2, e3, e4} and h3 × R, so that s ∼= aff(R) × (h3 × R). If c2 + d2 6= 0, using Lemma 2.10 we obtain that span{ f1, f2, e3, e4} is holomorphically isomorphic to (aff(C), J1), so that, in this case, s ∼= aff(R) × aff(C). 15 (ii) Case s′ J ∼= aff(R)× aff(R): There exists a basis {e1, . . . , e4} of s′ e4 and J is given by Je1 = e2, Je3 = e4. A derivation D of s′ J is of the form J such that [e1, e2] = e2, [e3, e4] = D =  0 0 0 0 a b 0 0 0 0 0 0 0 0 c d   with a, b, c, d ∈ R, and we observe that DJ is also a derivation. From (15), we obtain that [f1, f2] = (a2 + b2)e2 + (c2 + d2)e4. If a2 + b2 = c2 + d2 = 0, we obtain that s = aff(R) × aff(R) × R2. If a2 + b2 6= 0, c2 + d2 = 0, then setting (16) f1 = −e1 + bf1 + af2 a2 + b2 , f2 = J f1 = −e2 + −af1 + bf2 a2 + b2 , we obtain that [ f1, f2] = 0 and f1, f2 commute with span{e1, . . . , e4}. Therefore, s ∼= aff(R)×aff(R)×R2 via a holomorphic isomorphism. The proof for a2 + b2 = 0, c2 + d2 6= 0 is analogous. Finally, when a2 + b2 6= 0, c2 + d2 6= 0, we define f1, f2 as in (16) and this case reduces to the previous one. (iii) Case s′ J ∼= aff(R) ⋉ad R2: There exists a basis {e1, . . . , e4} of s′ e4, [e2, e3] = e4 and Je1 = e2, Je3 = −e4. A derivation D of s′ the form:   0 0 0 d c 0 0 0 0 a b −d c D =  with a, b, c, d ∈ R. Using (15), we obtain that 0 0 0 , [f1, f2] = (c2 + d2)e2 + 2(ad + bc)e4. J such that [e1, e2] = e2, [e1, e4] = J such that DJ is also a derivation is of If a2 + b2 = c2 + d2 = 0, then s = (aff(R) ⋉ad R2) × R2. If c2 + d2 6= 0, a2 + b2 = 0, then setting (17) f1 = −e1 + cf1 + df2 c2 + d2 , f2 = J f1 = −e2 + −df1 + cf2 c2 + d2 , one has that [ f1, f2] = 0 and f1, f2 commute with span{e1, . . . , e4}, and therefore s ∼= (aff(R)⋉ad R2)×R2 via a holomorphic isomorphism. If c2 + d2 = 0, a2 + b2 6= 0, we have again the holomorphic ismorphism s ∼= (aff(R) ⋉ad R2) × R2 by considering the following change of basis: f1 = f1 + be3 + ae4, f2 = J f1 = f2 + ae3 − be4. Finally, when c2 + d2 6= 0, a2 + b2 6= 0, by applying (17) this case reduces to the previous one. (iv) Case s′ J ∼= aff(C): There exists a basis {e1, . . . , e4} of s′ J such that It is known that a derivation D of aff(C) has matrix form [e1, e3] = −[e2, e4] = e3, [e1, e4] = [e2, e3] = e4. D =  0 0 0 0  0 0 0 0  c −d a −b a d c b 16 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI with respect to e1, . . . , e4, for some a, b, c, d ∈ R. From Theorem 2.5, we may assume that the restriction of J to s′ Let us consider first the case when the restriction is J1. It turns out that DJ1 is also a derivation of s′ J1, J is J1 or J2. and using (15), we obtain that [f1, f2] = 2(ad + bc)e3 + 2(bd − ac)e4, for some a, b, c, d ∈ R. It follows that dim s′ If the restriction is J2, we have that DJ2 is also a derivation of s′ J2 J = 2, and therefore, this case cannot occur. , and using (15), we obtain that [f1, f2] = ((a2 − b2) + (c2 − d2))e3 + 2(ab + cd)e4, with a, b, c, d ∈ R. If a2 + b2 = c2 + d2 = 0, then s ∼= aff(C) × R2. If c2 + d2 6= 0, a2 + b2 = 0, then setting f1 = f1 + ce3 + de4, (18) we obtain that [ f1, f2] = 0 and f1, f2 commute with span{e1, . . . , e4}. Therefore, s ∼= aff(C) × R2 via a holomorphic isomorphism. The proof for c2 + d2 = 0, a2 + b2 6= 0 is analogous. Finally, when c2 + d2 6= 0, a2 + b2 6= 0, we define f1, f2 as in (18) and this case reduces to the previous one. f2 = J f1 = f2 − ce1 − de2, (cid:3) J → s′ J , we denote by Dx : s′ J ∼= R4. Given x /∈ s′ We prove next some properties of (s, J) for a 6-dimensional non-nilpotent Lie algebra s with an abelian J the restriction of adx to complex structure J such that s′ s′ J . Lemma 4.3. Let s be a 6-dimensional non-nilpotent Lie algebra with center z (possibly trivial) and denote s2 = [s, s′]. If J is an abelian complex structure on s such that s′ (i) Dx commutes with DxJ for any x /∈ s′ J . (ii) ImDx = s2 for any x /∈ s′ J . (iii) dim s2 = 2 or 4. (iv) ker Dx = z for any x /∈ s′ J . (v) s′ (vi) There exists x /∈ s′ Proof. (i) Since J is abelian, DJ x = −DxJ. Using that s′ follows. J is abelian and the Jacobi identity (see (15)), (i) J such that [x, Jx] ∈ z. J ∼= R4, then J = s2 ⊕ z. (ii) Let x /∈ s′ J and set V := span{x, Jx}. We have that s = V ⊕ s′ J , therefore s2 = [s, s′] = [V, s′] = Dxs′ + DJ xs′ = Dxs′ + DxJ s′ = Dxs′ J . J , then it follows from (ii) that (iii) is equivalent to dim (Im Dx) = 2 or 4. We note first (iii) Let x /∈ s′ that Im Dx = {0} is not possible because this would give dim s′ If dim (Im Dx) = 1, let Im Dx = Ry. There exist a, b ∈ R such that Dxy = ay, DxJy = by. From (i) x = −(DxJ)2, and this implies a = b = 0. Therefore, D2 D2 J x = 0 and s is nilpotent, which is not possible since s is non-nilpotent by assumption. If dim (Im Dx) = 3, then dim (kerDx) = 1, that is, kerDx = Ry. From (i) we have Dx(DxJy) = DxJ(Dxy) = 0, hence DxJy = cy. Using again that D2 xy = −(DxJ)2y = −c2y, hence c = 0. This implies that Jy ∈ kerDx = Ry, a contradiction, and (iii) fol- lows. x = −(DxJ)2, we obtain 0 = D2 J ≤ 2. x = D2 (iv) It follows from (ii) and (iii) that z ⊂ s′ J . Moreover, z = ker Dx ∩ J (ker Dx) , (19) If dim s2 = 4, then using (ii) we obtain that ker Dx = {0} for any x /∈ s′ for any x /∈ s′ J . J , hence z = {0}. 17 If dim s2 = 2, then dim (ker Dx) = 2 for any x /∈ s′ J such that J = ker Dx⊕J (ker Dx). Using (ii) we obtain that Dx : J (ker Dx) → s2 J . Let us assume that there exists x /∈ s′ ker Dx is not J-stable, therefore, s′ is an isomorphism, then Dx(cid:0)s2(cid:1) = DxDxJ (ker Dx) = DxJDx (ker Dx) = 0. x = 0 = D2 This implies s2 = ker Dx, which means that D2 Hence, ker Dx is J-stable for any x /∈ s′ (v) If dim s2 = 4 there is nothing to prove (see proof of (iv)). We consider next the case dim s2 = 2, which implies dim z = 2. We only need to show that s2 ∩ z = {0}. If s2 ∩ z 6= {0}, we fix x /∈ s′ J and J . Using (ii) we obtain that Dx : V → s2 consider a J-stable complementary subspace V of ker Dx = z in s′ is an isomorphism. Let v ∈ V, v 6= 0, such that Dxv ∈ s2 ∩ z, then, using (i) and the fact that z is J-stable, we have J x, contradicting the fact that s is not nilpotent. J and (19) implies (iv). that is, DxJv ∈ ker Dx = z, thus Dx(V ) = z. Therefore, s2 = z and s is nilpotent, a contradiction, and (v) follows. DxDxJv = DxJDxv = 0, (vi) Let y /∈ s′ J , using (v) we have [y, Jy] = Dyv + w with w ∈ z. Setting x = y + 1 2 Jv, we compute [x, Jx] = [y, Jy] − 1 2 Dyv − 1 2 DJ yJv = Dyv + w − Dyv = w ∈ z, which implies (vi). (cid:3) Theorem 4.4. Let s be a 6-dimensional non-nilpotent Lie algebra with an abelian complex structure J such that s′ J ∼= R4. Then (s, J) is holomorphically isomorphic to one and only one of the following: (1) (aff(C) × R2, J) : [f1, e1] = [f2, e2] = e1, [f1, e2] = −[f2, e1] = e2, Jf1 = f2, Je1 = −e2 + e3, Je2 = e1 + e4, Je3 = e4. (2) (s1, J1) : [f1, f2] = e3, [f1, e1] = [f2, e2] = e1, J1f1 = f2, J1e1 = −e2 + e3, J1e2 = e1 + e4, J1e3 = e4. [f1, e2] = −[f2, e1] = e2, (3) (s1, J2) : [f1, f2] = e3, [f1, e1] = [f2, e2] = e1, J2f1 = f2, J2e1 = e2, J2e3 = e4, [f1, e2] = −[f2, e1] = e2, (4) (s2, J) : [f1, e1] = [f2, e2] = e1, [f1, e3] = [f2, e4] = e1 + e3, Jf1 = f2, Je1 = e2, Je3 = e4, [f1, e2] = −[f2, e1] = e2, [f1, e4] = −[f2, e3] = e2 + e4, (5) (s(a,b), J), (a, b) ∈ R (see (23)) : [f1, e1] = [f2, e2] = e1, [f1, e3] = [f2, e4] = ae3 + be4, Jf1 = f2, Je1 = e2, Je3 = e4. [f1, e2] = −[f2, e1] = e2, [f1, e4] = −[f2, e3] = −be3 + ae4, Proof. Let f1 /∈ s′ We consider next the different cases according to dim s2, which can be 2 or 4 (see Lemma 4.3(iii)). J such that [f1, Jf1] ∈ z and denote f2 = Jf1 (see Lemma 4.3(vi)), D := Df1. • If dim s2 = 2, then there are two possibilities, depending on s2 being J-stable or not. (i) Case s2 is J-stable: we note that [f1, f2] 6= 0, since dim s′ J = 4. Let e1, e2 = Je1 be a basis of s2 and e3 = [f1, f2], e4 = Je3 a basis of z. We have D =(cid:18)A 0 0(cid:19) , 0 −DJ =(cid:18)−AJ0 0 0(cid:19) , 0 18 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI where J0 = (cid:18)0 −1 (cid:18)a −b a (cid:19) with a2 + b2 6= 0. Setting 1 b 0 (cid:19) and A is a 2 × 2 non-singular matrix satisfying AJ0 = J0A, therefore, A = (20) f1 = (af1 + bf2), 1 √a2 + b2 f2 = J f1 = 1 √a2 + b2 (−bf1 + af2), we may assume A = I (observe that [ f1, f2] = [f1, f2]), hence the Lie bracket becomes: [f1, f2] = e3, [f1, e1] = [f2, e2] = e1, [f1, e2] = −[f2, e1] = e2, which is the Lie algebra s1 in the statement with the complex structure J2. (ii) Case s2 is not J-stable: fix e1 ∈ s2, e1 6= 0 and set e2 ∈ s2 such that Je1 + e2 ∈ z. It can be shown that {e1, e2} is a basis of s2, using that z is J-stable. Let {e3, e4 = Je3} be a basis of z; in the basis {e1, . . . , e4} of s′ J , D and the restriction of J to s′ J are given by D =(cid:18)A 0 0(cid:19) , 0 J =(cid:18)−J0 Y 0 J0(cid:19) , with J0 as above, and Y 6= 0 satisfies Y J0 = J0Y , which implies that Y is non-singular. Also A is non- singular and commutes with J0, and using (20), we may assume that A = I. By a suitable change of basis in z, Y can be taken equal to the identity I and J is then given by (21) Jf1 = f2, Je1 = −e2 + e3, Je2 = e1 + e4, Je3 = e4. If [f1, f2] = 0, the Lie brackets of s are given by [f1, e1] = [f2, e2] = e1, [f1, e2] = −[f2, e1] = e2, and J is given as in (21). Therefore, s ∼= aff(C) × R2 with a complex structure that preserves the ideal R2 but does not preserve the ideal aff(C). If [f1, f2] = ae3 + be4 ∈ z, with a2 + b2 6= 0, the Lie algebra obtained is isomorphic to s1 from (i) above. Indeed, φ : s1 → s given by φ =   1 0 0 1 a −b a b a −b a b   is a Lie algebra isomorphism. The complex structure J1 on s1 induced by φ is given as in (21). We note that this complex structure is not equivalent to J2 in (i) since J2 preserves s2 while J1 does not. • If dim s2 = 4, then z = 0, hence [f1, f2] = 0, D is an isomorphism, and thus D commutes with J. Therefore, we obtain s = R2 ⋉ R4, where the complex structure on R4 = span{e1, . . . , e4} is given by Je1 = e2, Je3 = e4. Considering D as an element of GL(2, C), the possible normal Jordan forms for D are (cid:18)α 1 0 α(cid:19) , (cid:18)α 0 0 β(cid:19) , with α, β ∈ C − {0}, α ≥ β. Identifying R2 = span{f1, f2} with C = spanC{f1}, we may take α−1f1 and therefore we may assume α = 1, 0 < β ≤ 1. Thinking of D now as an element of GL(4, R), this gives rise to the following possibilities for D, up to complex conjugation: (22) 1 0 1 0 0 1 0 1 1 0 0 1 D′ =    , 1 0 0 1 or D(a,b) =    a −b a b , a, b ∈ R, 0 < a2 + b2 ≤ 1. Let us denote by s2 (resp. s(a,b)) the Lie algebra determined by D′ (resp. D(a,b)). Clearly, s2 is not isomorphic to s(a,b) for any (a, b). Furthermore, it can be shown that the isomorphism classes of the Lie algebras s(a,b) are parameterized by (23) R = {(a, b) ∈ R2 : 0 < a2 + b2 < 1} ∪ {(a, b) ∈ S1 : b ≥ 0}. (cid:3) 19 Remark 4.5. We point out that the Lie algebra s(a,b) from Theorem 4.4 has a bi-invariant complex structure given by Jf1 = −f2, Je1 = e2, Je3 = e4. Indeed, s(a,b) is the realification of the complex Lie algebra r3,λ with λ = a + ib, where r3,λ has a C-basis {w1, w2, w3} with Lie brackets [w1, w2] = w2, [w1, w3] = λw3. The analogous statement holds for s2, which is the realification of the complex Lie algebra r3 with C-basis {w1, w2, w3} and Lie brackets [w1, w2] = w2, [w1, w3] = w2 + w3. J = s. A family of Lie algebras with abelian complex structure can be constructed in the following 4.3. s′ way. Let A be a commutative associative algebra and set aff(A) := A ⊕ A with the following Lie bracket and standard complex structure J: (24) then J is abelian (see [5]). Moreover, J is non-proper if and only if A2 = A. [(x, y), (x′, y′)] = (0, xy′ − x′y), J(x, y) = (−y, x), We show next that any 6-dimensional Lie algebra with a non-proper abelian complex structure is obtained using this procedure. Theorem 4.6. Let s be a 6-dimensional Lie algebra with a non-proper abelian complex structure J. Then (i) dim s′ = 3 and (s, J) is holomorphically isomorphic to aff(A) with its standard complex structure, where A is a 3-dimensional commutative associative algebra such that A2 = A; (ii) (s, J) is holomorphically isomorphic to one and only one of the following: (1) aff(R) × aff(R) × aff(R), where each factor carries its unique complex structure, (2) aff(R) × aff(C), where the second factor is equipped with J2 from Theorem 2.5, (3) aff(R)×(aff(R)⋉ad R2), where each factor is equipped with its unique abelian complex structure, (4) (s3, J) : [e2, f1] = f2, [e2, f2] = f3, [e3, f1] = f3, (5) (s4, J) : [e2, f1] = f2, [e3, f1] = f3, [e1, fi] = fi, i = 1, 2, 3, Jei = fi, i = 1, 2, 3, [e1, fi] = fi, i = 1, 2, 3, Jei = fi, i = 1, 2, 3. Proof. (i) Since dim s = 6 and s′ J = s, it follows from (O2) in §2.1 that dim s′ = 3 or 4. We show first that dim s′ = 4 is not possible. If dim s′ = 4 then dim(s′ ∩ J s′) = 2. We note that J abelian implies s′ ∩ J s′ ⊂ z and s has a basis of the form {e1, e2, e3, Je1, Je2, Je3} with e1, e2 ∈ s′, e3 ∈ s′ ∩ J s′, therefore s′ = span{[e1, Je1], [e2, Je2], [e1, Je2] = [e2, Je1]}, contradicting that dim s′ = 4. Thus dim s′ = 3 and s = s′ ⊕ J s′. Note that (25) The Lie bracket on s induces a structure of commutative associative algebra on s′ in the following way s′ = [s′, J s′]. Setting A := (s′,·) it turns out that x · y = [x, Jy], for x, y ∈ s′. φ : aff(A) → s, φ(x, y) = y − Jx, 20 A. ANDRADA, M. L. BARBERIS, AND I. G. DOTTI is a holomorphic isomorphism, where aff(A) is equipped with its standard complex structure (24). More- over, (25) implies that A2 = A and (i) follows. (ii) According to (i), in order to obtain the classification of the non-proper abelian complex structures in dimension 6, we must consider aff(A) with its standard complex structure, where A is a 3-dimensional commutative associative algebra satisfying A2 = A. It was shown in [13] that there are five isomorphism classes of such associative algebras. These algebras can be realized as matrix algebras as follows: b a A1 =    ,  c   A4 =    a b a A2 =     a   c a b , ,  b  b −c  c  A5 =    b a   , c b    A3 =   a   , a 0 c a b with a, b, c ∈ R. The Lie brackets on aff(Ai), i = 1, . . . , 5, together with its standard complex structure are those given in the statement. It is easily verified that these Lie algebras are pairwise non-isomorphic, and this completes the proof of the theorem. (cid:3) Remark 4.7. It is straightforward from Theorems 4.1, 4.2 and 4.4 that among the non-nilpotent 6-dimensional Lie algebras carrying proper abelian complex structures, (aff(R) ⋉ad R2) × R2 from Theorem 4.2 is the unique Lie algebra whose nilradical is non-abelian. Indeed, its nilradical is h3 × R2. In the non-proper case, only aff(R) × aff(R) × aff(R) and aff(R) × aff(C) have abelian nilradical (see Theorem 4.6). Corollary 4.8. Let s be a 6-dimensional unimodular solvable Lie algebra equipped with an abelian complex structure. Then s is isomorphic to nk, k = 1, . . . , 7, or to s(−1,0). Moreover, the simply connected Lie groups corresponding to these Lie algebras admit compact quotients. Proof. The first assertion follows easily from Theorems 3.2, 4.1, 4.2, 4.4 and 4.6. The simply connected Lie groups associated to nk, k = 1, . . . , 7, admit compact quotients since their structure constants are rational (see [16]). The existence of compact quotients of the simply connected solvable Lie group with Lie algebra s(−1,0) was proved in [23]. (cid:3) REFERENCES [1] E. ABBENA, S. GARBIERO AND S. SALAMON, Hermitian geometry on the Iwasawa manifold, Boll. Un. Mat. Ital. bf 11-B (1997), 231 -- 249. [2] M. L. BARBERIS, Hypercomplex structures on four dimensional Lie groups, Proc. Amer. Math. Soc. 125 (4) (1997), 1043 -- 1054. [3] M. L. BARBERIS, I. G. DOTTI AND R. J. MIATELLO, On certain locally homogeneous Clifford manifolds, Ann. Glob. Anal. Geom. 13 (1995), 289 -- 301. [4] M. L. BARBERIS AND I. DOTTI, Hypercomplex structures on a class of solvable Lie groups, Quart. J. Math. Oxford (2), 47 (1996), 389 -- 404. [5] M. L. BARBERIS AND I. DOTTI, Abelian complex structures on solvable Lie algebras, J. Lie Theory 14 (2004), 25 -- 34. [6] M. L. BARBERIS, I. DOTTI AND M. VERBITSKY, Canonical bundles of complex nilmanifolds, with applications to hyper- complex geometry, Math. Research Letters 16(2) (2009), 331 -- 347. [7] T. CHRISTODOULAKIS, G. O. PAPADOPOULOS AND A. DIMAKIS, Automorphisms of real 4-dimensional Lie algebras and the invariant characterization of homogeneous 4-spaces, J. Phys. A: Mathematical & General 36 (2003), 427 -- 442. [8] R. CLEYTON AND Y. S. POON, Differential Gerstenhaber Algebras Associated to Nilpotent Algebras, Asian J. Math. 12 (2) (2008), 225 -- 250. [9] S. CONSOLE AND A. FINO, Dolbeault cohomology of compact nilmanifolds, Transformation Groups 6(2) (2001), 111 -- 124. [10] S. CONSOLE, A. FINO AND Y. S. POON, Stability of abelian complex structures, Internat. J. Math. 17(4) (2006), 401 -- 416. 21 [11] L. A. CORDERO, M. FERN ´ANDEZ, A. GRAY AND L. UGARTE, Nilpotent complex structures on compact nilmanifolds, Rend. Circolo Mat. Palermo 49 (Suppl.) (1997), 83 -- 100. [12] L. A. CORDERO, M. FERN ´ANDEZ AND L. UGARTE, Abelian complex structures on 6-dimensional compact nilmanifolds, Comment. Math. Univ. Carolinae 43(2) (2002), 215 -- 229 . [13] M. GOZE AND E. REMM, Affine structures on abelian Lie groups, Linear Algebra Appl. 360 (2003), 215 -- 230. [14] G. KETSETZIS AND S. SALAMON, Complex structures on the Iwasawa manifold, Adv. Geom. 4 (2004), 165 -- 179. [15] J. LAURET, Minimal metrics on nilmanifolds, in Differential geometry and its applications, Matfyzpress, Prague, 2005, 79 -- 97. [16] A. MAL' CEV, On a class of homogeneous spaces, reprinted in Amer. Math. Soc. Trans. Ser. 1 9 (1962), 276 -- 307. [17] A. P. PETRAVCHUK, Lie algebras decomposable into a sum of an abelian and a nilpotent subalgebra, Ukr. Math. J. 40 (3) (1988), 331 -- 334. [18] E. RODR´IGUEZ VALENCIA, Invariants of complex structures on nilmanifolds, in progress. [19] S. ROLLENSKE, Geometry of nilmanifolds with left-invariant complex structure and deformations in the large, Proc. London Math. Soc. 99 (2) (2009), 425 -- 460. [20] S. SALAMON, Complex structures on nilpotent Lie algebras, J. Pure Appl. Algebra 157 (2001), 311 -- 333. [21] J. E. SNOW, Invariant complex structures on four dimensional solvable real Lie groups, Manuscripta Math. 66, (1990), 397 -- 412. [22] L. UGARTE, Hermitian structures on six-dimensional nilmanifolds, Transformation Groups 12 (1), (2007), 175 -- 202. [23] T. YAMADA, A pseudo-Kahler structure on a nontoral compact complex parallelizable solvmanifold, Geom. Dedicata 112 (2005), 115 -- 122. FAMAF-CIEM, UNIVERSIDAD NACIONAL DE C ´ORDOBA, CIUDAD UNIVERSITARIA, 5000 C ´ORDOBA, ARGENTINA E-mail address: [email protected] E-mail address: [email protected] E-mail address: [email protected]
1708.01430	2	1708	2017-10-20T15:14:48	A Koszul sign map	[ "math.RA", "math.RT" ]	We define a Koszul sign map encoding the Koszul sign convention. A cohomological interpretation is given.	math.RA	math	A Koszul sign map Roland Berger Abstract We define a Koszul sign map encoding the Koszul sign convention. We view the Koszul sign map as a 2-cochain with respect to the cohomology of the permutation group. 1 Introduction The Koszul sign convention is a rule which plays a fundamental role in graded algebra. This convention was used by Koszul in his thesis [3]. Actually, whenever the algebraic framework is based on graded categories as in homology theory [4], homotopy theory [2], algebraic operads theory [5], this convention is systematically used in the definitions of the new concepts naturally appearing in the theory, and in the statements of the new results as well. In general – in particular in the cited books – the convention is just stated, without further explanation. The aim of this paper is to propose a tool encoding precisely the Koszul sign convention. In 1966, without any reference to Koszul, Boardman introduced a Principle of Signs [1], in order to make precise the way of inserting the right signs in the identities frequently obtained in algebraic topology. His approach lies on a class of various n-ary operations and on an involution, subjected to some axioms, allowing him to characterize the identities written in a standard form. Then he proved that the set of these identities is stable under natural algebraic transformations. Our approach is different, and it can be viewed as a preliminary step of Boardman's one. Although we fix a set of graded elements as in Boardman, we do not need to define the operations and the identities acting on these elements. For us, only the order of the elements inside of the result of an operation is important to produce a sign in front of the algebraic expression. 2 A Koszul sign map The Koszul sign convention is used in various graded contexts. The objects on which the convention is applied are homogeneous, and the nature of the objects – graded elements, graded maps – depends on the context. However the convention does not depend on the nature of the objects, but just on their degrees. In our general setting, the homogeneous objects will be called symbols. At each symbol, we associate a degree in Z. Roughly speaking, the Koszul sign convention is the following: if in a manipulation of a monomial algebraic expression concerning on symbols naturally written from the left to the right, a symbol fi jumps over a symbol fj situated on the left or on the right of fi, then the 1 sign (−1)fifj appears in front of the expression, where fi denotes the degree of fi. If the algebraic expression is a sum of monomial algebraic expressions, the convention is applied to each term. Let us note that the convention is independent of the algebraic operations included in the monomial algebraic expression. Although the manipulation passes from a monomial algebraic expression to another one, only the order of the objects in the initial and final expressions are significant. For example, in the definition of a tensor product of graded linear maps (f ⊗ g)(a ⊗ b) = (−1)gaf (a) ⊗ g(b), the ordered symbols in the initial – final – expression are f , g, a, b – f , a, g, b. In our setting, we are led to permute arbitrarly the symbols from the order of the initial expression. The result of a permutation might be organized in a final monomial algebraic expression respecting the final order, but such an expression would be irrelevant for us. Throughout the paper, we fix an integer n ≥ 2, a sequence f = (f1, . . . , fn) of symbols, and a sequence of degrees called the degree of f . f = (f1, . . . , fn) ∈ Zn We denote by Ef the set of the permutations of f . If Sn is the group of permutations of {1, . . . , n}, we have Ef = {g = (g1, . . . , gn) ; ∃ρ ∈ Sn, gi = fρ−1(i) f or i = 1, . . . , n}, where g1, . . . , gn are still seen as symbols. The degree of g is defined by g = (g1, . . . , gn) with gi = fρ−1(i). The group Sn acts on the left on Ef by defining σ(g) = (gσ−1(1), . . . , gσ−1(n)), for g ∈ Ef and σ ∈ Sn. We want to define now a Koszul sign map κ : Sn × Ef → {±1} (σ, g) 7→ κ(σ, g), which respects the Koszul sign convention. For the moment, we state this convention in a rather heuristic form as follows. If in the permutation ρ : (g1, . . . , gn) → (f1, . . . , fn) correcting the permutated sequence (g1, . . . , gn) into the initial sequence (f1, . . . , fn), gi jumps over gj, then the sign (−1)gigj appears in κ. We begin to remark that it is not clear how to define a map κ(−, f ) : Sn → {±1} respecting the convention when n = 3 – if n = 2, κ(−, f ) : S2 → {±1} is well-defined and is a group morphism. 2 In fact, after acting the transpositions (1, 2) and (2, 3) on f = (f1, f2, f3), we obtain κ((1, 2), f ) = (−1)f1f2, κ((2, 3), f ) = (−1)f2f3. Since (2, 3)(1, 2)f = (f3, f1, f2) is corrected into (f1, f2, f3) by jumping f3 over f1 and f2, we obtain κ((2, 3)(1, 2), f ) = (−1)f3(f1+f2), so that κ((2, 3)(1, 2), f ) 6= κ((2, 3), f ) k((1, 2), f ) for a certain choice of the degrees. There- fore it is not possible to define a group morphism κ(−, f ) : Sn → {±1} in great generality. However, if we set g = (1, 2)f = (f2, f1, f3), we have κ((2, 3), g) = (−1)g2g3 = (−1)f1f3, Thus we are led to the right formula κ((2, 3)(1, 2), f ) = κ((2, 3), (1, 2)(f )) κ((1, 2), f ). In order to define κ from this formula and its generalizations, we want to be sure that κ(σ, g) does not depend on the way to correct g into f by using transpositions. Consequently, we first define κ on the free group Fn−1 generated by the transpositions si = (i, i + 1) for i = 1, . . . n − 1. Let us recall that the group Sn is defined by these generators and the following relations in Fn−1 s2 i = e, sisj = sjsi if i − j > 1, sisi+1si = si+1sisi+1, (2.1) where e denotes the unit of the group Fn−1. For g in Ef and 1 ≤ i ≤ n − 1, we set si(g) = s−1 i (g) = (g1, . . . gi−1, gi+1, gi, gi+2, . . . , gn), (2.2) which defines an action x(g) of the elements x of the group Fn−1 on the set Ef . This action induces naturally the action of Sn on Ef defined above. We define the map κ : Fn−1 × Ef → {±1} (x, g) 7→ κ(x, g) as follows. For any g in Ef , we set κ(e, g) = 1, and for 1 ≤ i ≤ n − 1, Moreover, for any x in Fn−1 decomposed in a reduced form κ(si, g) = κ(s−1 i , g) = (−1)gigi+1. x = ti1 . . . tim (2.3) (2.4) where tij = sij or tij = s−1 ij , and i1, . . . , im are in {1, . . . , n − 1}, we set κ(x, g) = κ(ti1 , ti2 . . . tim (g)) κ(ti2 , ti3 . . . tim(g)) . . . κ(tim , g). (2.5) Reduced form means that two consecutive factors tij and tij+1 are never inverse to each other. A reduced form being unique, the map (2.3) is well-defined. 3 Lemma 2.1 For g in Ef and 1 ≤ i ≤ n − 1, we have κ(si, s−1 i (g)) κ(s−1 i , g) = 1 = κ(s−1 i , si(g)) κ(si, g). (2.6) Proof. Since κ(si, g) = κ(s−1 From (2.2), we draw κ(si, s−1 i , g) and si(g) = s−1 (g)) = (−1)gi+1gi = κ(s−1 i (g), it suffices to verify the first equality. i , g). i The lemma shows that the formula (2.5) extends to any decomposition x = ti1 . . . tim reduced or not. Therefore, one has κ(x y, g) = κ(x, y(g)) κ(y, g) for g in Ef , x, y in Fn−1, and consequently κ(x−1, g) = κ(x, x−1(g)). (2.7) (2.8) Proposition 2.2 Passing through the relations (2.1), the map κ : Fn−1 × Ef → {±1} induces a map κ : Sn × Ef → {±1}. Proof. We will prove that, for each relation in (2.1) and for each fixed g, κ(−, g) gives the same result on the left-hand side and on the right-hand side of the relation. i , g) = κ(si, si(g)) κ(si, g) according to (2.7), hence κ(s2 i , g) = 1 by the Fisrstly, κ(s2 lemma. Secondly, let us suppose that j > i + 1, so that sj(g) = (g1, . . . gi, gi+1, . . . gj+1, gj, . . . , gn), thus κ(si, sj(g)) = (−1)gigi+1 and κ(sj, g) = (−1)gjgj+1. Using (2.7), we obtain κ(sisj, g) = (−1)gigi+1+gjgj+1. The same if j < i − 1. So, if i − j > 1, we obtain the expected equality Thirdly, from (2.7), we draw κ(sisj, g) = κ(sjsi, g). κ(si si+1 si, g) = κ(si, si+1 si(g)) κ(si+1, si(g)) κ(si, g). (2.9) Using si(g) = (g1, . . . gi+1, gi, gi+2, . . . , gn) and si+1si(g) = (g1, . . . gi+1, gi+2, gi, . . . , gn), the formula (2.9) implies κ(si si+1 si, g) = (−1)gi+1gi+2+gigi+2+gigi+1. Using si+1(g) = (g1, . . . gi, gi+2, gi+1, . . . , gn) and sisi+1(g) = (g1, . . . gi+2, gi, gi+1, . . . , gn), the formula (2.9) in which i and i + 1 are exchanged gives κ(si+1 si si+1, g) = (−1)gigi+1+gigi+2+gi+1gi+2. Then we arrive to κ(si si+1 si, g) = κ(si+1 si si+1, g). 4 An equivalent way to say what we have obtained is the following. Writing each relation (2.1) as an equality r = e for a certain element r in Fn−1, we have κ(r, g) = 1 for any g – it suffices to apply the formula (2.7). Now for any x in Fn−1, we have κ(xrx−1, g) = κ(x, r(x−1(g)) κ(r, x−1(g)) κ(x−1, g) = κ(x, x−1(g)) κ(x−1, g) since r(g′) = g′ for any g′ in Ef . Therefore (2.8) implies that κ(xrx−1, g) = 1 for any g in Ef . Using again (2.7), we deduce κ(xrx−1x′r′x′−1, g) = 1 for any g in Ef , and any elements r, r′ of Fn−1 generating the relations. Inductively, we obtain that κ(a, g) = 1 for any g in Ef and any a in the normal subgroup of Fn−1 generated by the elements r. More generally, κ(ax, g) = κ(x, g) for any a as previously and any x in Fn−1. Our construction of the map κ : Sn × Ef → {±1} shows immediately the following proposition. This proposition could be used as a definition. Proposition 2.3 The map κ : Sn × Ef → {±1} is the unique map such that 1. ∀σ ∈ Sn, ∀τ ∈ Sn, ∀g ∈ Ef , κ(σ τ, g) = κ(σ, τ (g)) κ(τ, g), 2. ∀i ∈ {1, . . . , n − 1}, ∀g ∈ Ef , κ(si, g) = (−1)gigi+1. From f and its degree f , we have constructed the map κ which should be rather denoted by κf . For any f ′ in Ef , one has Ef ′ = Ef . The degree of f ′ is obviously defined from the degree of f . Then Proposition 2.3 shows that the map κf ′ : Sn × Ef ′ → {±1} coincides with κf . However, it is possible that κf (−, f ) 6= κf ′ (−, f ′), as in the example f = (f1, f2, f3) and f ′ = (f1, f3, f2), with f1 and f2 odd, f3 even. Example 2.4 Take n = 5 and g = (f4, f1, f3, f5, f2), so that g = ρ(f ) where ρ = (cid:18) 4 1 1 2 3 5 3 4 2 5 (cid:19) . Using a bubble sort, we find that κ(ρ, f ) = (−1)Z with zi = fi and Z = z1z4 + z3z4 + z2z5 + z2z4 + z2z3. 5 3 When κ(−, g) is a group morphism An integer n ≥ 2, a sequence f = (f1, . . . , fn) of symbols, and a sequence of degrees f = (f1, . . . , fn) ∈ Zn being given, we have defined the map κ = κf in the previous section. If n = 2, then κ(−, (f1, f2)) = κ(−, (f2, f1)) is always a group morphism from S2 to {±1}. Let us suppose that n ≥ 3. We want to know when is a group morphism. From 1. in Proposition 2.3, it is the case if and only if for any σ and τ in Sn, one has κ(−, g) : Sn → {±1} that is, if and only if, for any σ in Sn and h in Ef , κ(σ, τ (g)) = κ(σ, g), κ(σ, h) = κ(σ, g), which implies that κ(−, h) is a group morphism as well. So, it suffices to examine when κ(−, f ) is a group morphism, and we have seen that it is the case if and only if ∀σ ∈ Sn, ∀τ ∈ Sn, ∀g ∈ Ef , κ(σ, τ (g)) = κ(σ, g). (3.1) From 1. in Proposition 2.3, it is equivalent to ∀i ∈ {1, . . . , n − 1}, ∀j ∈ {1, . . . , n − 1}, ∀g ∈ Ef , κ(si, sj(g)) = κ(si, g). (3.2) It is clear that (3.2) holds whenever i = j or i − j > 1. Thus (3.2) is equivalent to ∀i ∈ {1, . . . , n − 2}, ∀g ∈ Ef , κ(si, si+1(g)) = κ(si, g) and κ(si+1, si(g)) = κ(si+1, g). (3.3) Some calculations included in the proof of Proposition 2.2 show that it is equivalent to ∀i ∈ {1, . . . , n − 2}, ∀g ∈ Ef , (−1)gigi+2 = (−1)gigi+1 = (−1)gi+1gi+2, (3.4) which, in turn, is equivalent to say that among the parities of the triplets (gi, gi+1, gi+2), only the cases of one even parity and two odd parities are forbidden. The last condition is satisfied if all the degrees f1, . . . , fn have the same parities, otherwise if only one is odd. Conversely, if the last condition is satisfied and if f1, . . . , fn do not have the same parities, it is not possible to find one even parity and two odd parities among them – a suitable permutation of f then providing a forbidden triplet. We have obtained the following. Proposition 3.1 Suppose that n ≥ 2, f = (f1, . . . , fn), and f = (f1, . . . , fn) ∈ Zn are given. 1. The map κ(−, g) : Sn → {±1} is a group morphism for an element g of Ef if and only if κ(−, f ) is a group morphism, and in this case, all the group morphisms κ(−, g) are equal. 2. The map κ(−, f ) is a group morphism if and only if either all the integers f1, . . . , fn have the same parities or only one is odd among them. 3. The map κ(−, f ) is constant equal to 1 if and only if all the integers f1, . . . , fn are even, with possibly one exception. 6 4 A cohomological interpretation of κf An integer n ≥ 2, a sequence f = (f1, . . . , fn) of symbols, and a sequence of degrees f = (f1, . . . , fn) ∈ Zn are given. For σ and ρ in Sn, we put cf (σ, ρ) = κf (σ, ρ(f )). So we define a map cf : Sn × Sn → {±1}. This is the unique map such that 1. ∀σ ∈ Sn, ∀τ ∈ Sn, ∀ρ ∈ Sn, cf (σ τ, ρ) = cf (σ, τ ρ) cf (τ, ρ), 2. ∀i ∈ {1, . . . , n − 1}, ∀ρ ∈ Sn, cf (si, ρ) = (−1)fρ−1(i)fρ−1 (i+1). We want to regard the map cf as a 2-cochain for the cohomology of the group Sn with coefficients in the multiplicative group {±1}. The automorphism group of {±1} is formed of +Id and −Id, and it is identified to the group {±1}. Then a structure of Sn-module on the group {±1} is equivalent to the datum of a group morphism u : Sn → {±1}, the action of σ ∈ Sn on +1 or −1 being given by the product by u(σ). It is well-known that there are only two such morphisms u: the constant morphim u = 1 and the signature u = sgn. Let us choose a structure u of Sn-module on {±1}. Let us calculate the coboundary operator δ of the 2-cochain cf . For σ, τ and ρ in Sn, one has δ(cf )(σ, τ, ρ) = (σ.cf (τ, ρ)) cf (στ, ρ)−1 cf (σ, τ ρ) cf (σ, τ )−1. Since σ.cf (τ, ρ) = u(σ) cf (τ, ρ), the relation 1. just above shows that δ(cf )(σ, τ, ρ) = u(σ) cf (σ, τ )−1. Thus cf is a 2-cocycle if and only if cf (σ, τ ) = u(σ) for any σ and τ . This condition implies that cf (−, τ ) : Sn → {±1} is a group morphism for any τ , thus either all the integers f1, . . . , fn have the same parities or only one is odd among them (Proposition 3.1). Conversely, if either all the integers f1, . . . , fn have the same parities or only one is odd among them, then the numbers ui = (−1)fifi+1 are all equal. By u(si) = ui for i = 1, . . . , n− 1, we define a group morphism u : Sn → {±1}, hence a structure of Sn-module on {±1} for which cf is a 2-cocycle. Moreover we have u = cf (−, e), so that δ(u)(σ, τ ) = u(σ)cf (τ, e) cf (στ, e) cf (σ, e). Then the relation 1. obtained. just above implies that δ(u) = cf . Let us sum up what we have Proposition 4.1 Let us suppose that n ≥ 2, f = (f1, . . . , fn), and f = (f1, . . . , fn) ∈ Zn are given. Let us endow the group {±1} with the Sn-module structure defined by a group morphism u : Sn → {±1}. Then the map cf : Sn × Sn → {±1} 7 is a 2-cocycle with coefficients in the group {±1} if and only if either all the integers f1, . . . , fn have the same parities or only one is odd among them. When this condition holds, u = 1 if and only if all the integers f1, . . . , fn are even, with possibly one exception, and u = sgn if and only if all the integers f1, . . . , fn are odd. Moreover, in both cases, cf is equal to the coboundary of u. The 2-cochain cf is not symmetric in general. In fact, cf (e, τ ) = 1 for any τ , while cf (si, e) = (−1)fifi+1. Question. Find another cohomological interpretation for which cf is always a 2-cocycle, and is a 2-coboundary if and only if cf (−, e) is a group morphism. References [1] J. M. Boardman, The principle of signs, Enseignement Math. 12 (1966), 191-194. [2] Y. Felix, S. Halperin, J.-C. Thomas, Rational Homotopy Theory, Graduate Texts in Mathematics, 205, Springer, 2001. [3] J.-L. Koszul, Homologie et cohomologie des alg`ebres de Lie (French), Bull. Soc. Math. France 78 (1950), 65-127. [4] J.-L. Loday, Cyclic homology, Grundlehren der mathematischen Wissenschaften 301, Second Edition, Springer, 1998. [5] J.-L. Loday, B. Vallette, Algebraic operads, Grundlehren der mathematischen Wis- senschaften 346, Springer, 2012. Roland Berger: Univ Lyon, UJM-Saint-´Etienne, CNRS UMR 5208, Institut Camille Jordan, F-42023, Saint-´Etienne, France [email protected] 8
1309.0409	4	1309	2013-11-15T15:02:32	The endomorphism ring of an injective square-free module	[ "math.RA" ]	Let $M_R$ be an injective right module over a ring $R$. The goal of this paper to prove that the endomorphism ring $S=(M_R)$ of $M$ is quasi-duo if and only if $M_R$ is square-free.	math.RA	math	THE ENDOMORPHISM RING OF A SQUARE-FREE INJECTIVE MODULE MAI HOANG BIEN Dedicated to Professor Alberto Facchini for his 60th birthday Abstract. Let MR be an injective right module over a ring R. The goal of this paper to prove that the endomorphism ring S = End(MR) of MR is quasi-duo if and only if MR is square-free. ] . A R h t a m [ 4 v 9 0 4 0 . 9 0 3 1 : v i X r a 1. Introduction In this paper, rings are meant to be associative with identity and modules are understood to be right modules. It is well known that an injective module is indecomposable if and only if its endomorphism ring is a local ring [4]. The aim of this paper is to generalize this result from indecomposable modules to square-free ones. Recall that an injective R-module MR is said to be square-free if it contains no direct summand isomorphic to X ⊕ X for some direct summand 0 6= X of MR. A ring S is called left quasi-duo (resp. right quasi-duo) if every maximal left ideal (resp. maximal right ideal) of S is a two-sided ideal. If S is both left and right quasi-duo then S is said to be quasi-duo. The class of quasi-duo rings is not "small". For example, commutative rings and local rings are quasi-duo. Factors of a quasi- duo ring are also quasi-duo. In [5], the author proved that S is quasi-duo if and only if S/J(S) is quasi-duo. Here, J(S) is the Jacobson radical of S. The direct product of a family of rings is quasi-duo if and only if each ring is quasi-duo. The matrices ring Mn(D) of degree n over a division ring D is quasi-duo if and only if n = 1. Hence, a direct product of finitely many division rings is quasi-duo. Therefore, rings of finite type are quasi-duo [1]. In particular, the endomorphism rings of univerial modules, couniform projective modules, cyclically presented modules over a local ring, and the kernels of morphisms between indecomposable injective modules are quasi-duo (see [1]). Examples of right quasi-duo rings which are not left quasi-duo are unknown [3, Question 7.7]. Of course, if there exists such ring R then the opositive ring Rop of R is left quasi-duo and not right quasi-duo. Let MR be an injective module over R. In this paper, we will show that MR is square-free if the endomorphism ring S := End(MR) of MR is either left quasi- duo or right quasi-duo (Theorem 2.1). Conversely, if MR is square-free then S is quasi-duo. In fact, we prove that if MR is square-free then every one-sided ideal of S containing J(S) is two-sided (Theorem 2.6 and 2.8). The symbols and notations we use in this paper are familiar to module and ring theory. ⊆e will denote an essential submodule, and E(N ) will denote an injective Key words and phrases. Injective module, endomorphism ring, quasi-duo, square-free. 2010 Mathematics Subject Classification. 16D25, 16D50, 16D70, 16D80. 1 2 MAI HOANG BIEN envelope of a module N . For any direct summand N of a R-module MR, ιN and πN , respectively, are the embedding of N into MR and the projection of MR onto N . 2. Results Theorem 2.1. Let MR be an injective R-module. If the endomorphism ring S = End(MR) of MR is either left quasi-duo or right quasi-duo then MR is square-free. Proof. Assume that S is left quasi-duo and there exist direct summands A, B, C of MR such that A ⊕ B ⊕ C = MR and A is isomorphic to B. Call I a maximal left ideal of S containing s = ιB⊕C πB⊕C . Then I is a two-sided ideal of S by hypothesis. Let α be an isomorphism from A to B. Define f := ιBαπA and g := ιAα−1πB + ιC πC . Then sf + gs ∈ I and ker(sf + gs) = { a + b + c ∈ A ⊕ B ⊕ C (sf + gs)(a + b + c) = 0 } = { a + b + c ∈ A ⊕ B ⊕ C α(a) + α−1(b) + c = 0 } = 0. Hence, sf + gs is injective. Consider the diagram 0 / A ⊕ B ⊕ C = MR sf +gs MR 1 tMR h Since MR is injective, there exists morphism h of S such that h(sf + gs) = 1, which imlies I = S. Contradiction. Therefore, if S is left quasi-duo then MR is square-free. Similarly to the right side. Lemma 2.2. Let MR be a square-free injective R-module. If N1, N2 are two iso- morphic direct summands of MR then N1 and N2 are two injective envelopes of N1 ∩ N2. Proof. Let A1, A2 be injective envelopes of N1 ∩ N2 respectively in N1, N2 and B1, B2 be respectively direct summands of N1, N2 such that N1 = A1 ⊕ B1, N2 = A2 ⊕ B2. Since N1 ∼= N2 and B1 ∩ A2 = 0, B1 is isomorphic to a direct summand C of N2 and since B1 ∩ N2 = 0, B1 + N2 = B1 ⊕ N2 is injective, which implies that MR contains B1 ⊕ C, with B1 ∼= C, as a direct summand. Therefore B1 = 0. Similarly, B2 = 0. Thus, N1, N2 are two injective envelopes of N1 ∩ N2. Lemma 2.3. Let MR be a square-free injective R-module with MR = M1⊕M2. If N is a direct summand of MR then N is an injective envelope of (N ∩M1)⊕ (N ∩M2). Proof. Let A be an injective envelope of N ∩ M1 in N and B be a direct summand of N such that N = A ⊕ B. Consider π := πM2 , the projection of M1 ⊕ M2 onto M2, and the morphism π B : B → M2 restricted on B of π. Because B ∩M1 = 0, πB is injective. Hence, B is isomorphic to a direct summand C of M2. By Lemma 2.2, B is an injective envelope of B ∩ C. Therefore, B is an injective / / / t THE ENDOMORPHISM RING OF A SQUARE-FREE INJECTIVE MODULE 3 envelope of N ∩ M2 = B ∩ M2 ⊇ B ∩ C. This implies N is an injective envelope of (N ∩ M1) ⊕ (N ∩ M2). Lemma 2.4. Let MR be an injective R-module, S = End(MR) be the endomor- phism ring of MR and J(S) be the Jacobson radical of S. For any element f ∈ S, there exist e1, e2, g1, g2, h1, h2 ∈ S and i1, i2, j1, j2 ∈ J(S) such that e1, e2 are idem- potents and e1 = f g1 + i1, f = e1h1 + j1, e2 = g2f + i2, f = h2e2 + j2. Proof. It is well-known that S/J(S) is a Von Neumann regular ring and any idempotent of S/J(S) can be lifted to an idempotent of S (see [2, Theorem 13.1]). Hence, for any f ∈ S, there exists idempotent e1 ∈ S such that (f S + J(S))/J(S) = (e1S + J(S))/J(S). Therefore, e1 = f g1 + i1, f = e1h1 + j1 for some g1, h1 ∈ S and i1, j1 ∈ J(S). Similarly to e2 = g2f + i2, f = h2e2 + j2. Lemma 2.5. Let MR be an injective R-module, S = End(MR) be the endomor- phism ring of MR and e be an idempotent of S. For any element f ∈ S, f ∈ eS if and only if f (MR) ⊆ e(MR). Proof. Assume that f = eh for some h ∈ S. Then f (MR) = eh(MR) ⊆ e(MR). Conversely, assume that f (MR) ⊆ e(MR). Then for any x ∈ MR, f (x) = e(y) for some y ∈ MR. Hence, Therefore, f = ef ∈ eS. f (x) = e(y) = e(e(y)) = e(f (x)). Theorem 2.6. Let MR be an injective R-module and S = End(MR) be the endo- morphism ring of MR. If MR is square-free then every right ideal of S containing J(S) of S is a two-sided ideal. In particular, if MR is square-free then S is right quasi-duo. Proof. Let I be a right ideal of S containing J(S). Let f ∈ I and φ ∈ S. We must show that φf ∈ I. By Lemma 2.4, there exist e, g, h ∈ S and i, j ∈ J(S) such that e = f g + i is an idempotent and f = eh + j. We have e belongs to I and φf = φeh + φj, so that it suffices to show that φe ∈ I. Indeed, let N ′ be a direct summand of MR such that MR = N ⊕ N ′ with N = e(MR). If set M1 is an injective envelope of ker φ, then there exists a direct summand M2 of MR such that MR = M1 ⊕ M2. Let N1, N2 be respectively injective envelopes of N ∩ M1, N ∩ M2 in N . By Lemma 2.3, we may assume that MR = N ⊕ N ′ = N1 ⊕ N2 ⊕ N ′. Consider the morphism φ N2 : N2 → N1 ⊕ N2 ⊕ N ′ restricted on N2 of φ. It is easy to check that φ N2 is injective. Hence, N2 ∼= φ N2(N2) and by Lemma 2.2, A = N2 ∩φ N2 (N2) is an essential submodule of N2. Set B = φ−1 A (A) with φA : A → N2 is the morphism restricted on A of φ. Then B is also an essential submodule of N2. Write ψ : N2 → N2 for a morphism extending of φ A and let ψ′ = ιN2ψπN2 . One has ψ′(MR) ⊆ N2 ⊆ N = e(MR), which implies from Lemma 2.5 that ψ′ ∈ eS ⊆ I. Moreover, since for any a + b + c ∈ (ker φ ∩ N1) ⊕ B ⊕ N ′, (φe − ψ′)(a + b + c) = 0 4 MAI HOANG BIEN and (ker φ ∩ N1) ⊕ B ⊕ N ′ is essential in MR, φe − ψ′ belongs to J(S). This shows that φe ∈ I. Lemma 2.7. Let MR be an injective R-module, S = End(MR) be the endomor- phism ring of MR. Then for any two elements f, g of S, g = hf + j for some h ∈ S, j ∈ J(R) if and only if ker f ∩ A ⊆ ker g for some essential submodule A of MR. Proof. Assume that g = hf + j for some h ∈ S and j ∈ J(S). Then ker g ⊇ ker(hf ) ∩ ker j ⊇ ker f ∩ A with A = ker(j) ⊆e MR. Conversely, assume that ker f ∩ A ⊆ ker g for a submodule A ⊆e MR. Let M1 be an injective envelope of ker g in MR, and N1 be an injective envelope of ker f ∩ A in M1. Assume that MR = N1 ⊕ N2 ⊕ M2 for some direct summands N2, M2 respectively of M1 and MR. Consider the diagram 0 / N2 ⊕ M2 f N2⊕M2 MR g N2⊕M2 h vMR Here, f N2⊕M2 , g N2⊕M2 are the morphisms respectively restricted on N2 ⊕ M2 of f, g. Since MR is injective, there exists h : MR → MR such that g N2⊕M2 = h.f N2⊕M2 . Because (g − hf )(a + b + c) = 0 for any a + b + c ∈ (ker f ∩ A) ⊕ N2 ⊕ M2 which is essential in MR, j = g − hf ∈ J(S). Theorem 2.8. Let MR be an injective R-module and S = End(MR) be the endo- morphism ring of MR. If MR is square-free then every left ideal of S containing the Jacobson radical J(S) of S is a two-sided ideal. In particular, if MR is square-free then S is left quasi-duo. Proof. Let I be a left ideal of S containing J(S). Let f, φ be elements of S with f ∈ I. We must show that f φ ∈ I. By Lemma 2.4, there exist e, g, h ∈ S and i, j ∈ J(S) such that e = gf + i is an idempotent and f = he + j. Hence, e ∈ I and f φ = heφ + jφ. It suffices to show that eφ ∈ I. Indeed, set N1 = ker e and let N2 be a direct summand of MR such that MR = N1 ⊕ N2. Consider If set N ′ ψ := πN2 φιN1 : N1 → N2. 1 is an injective envelope of A := ker φ in N1 then exists a direct summand N ′′ 1 of MR such that N1 = N ′ 1 . Since ker ψ N ′′ 1 is isomorphic to a direct summand C of N2. Hence, MR contains a direct summand isomorphic to C ⊕C. By hypothesis, C = 0. In other words, A is essential in N1. Now, one has that φ can be 1 is injective. This implies N ′′ 1 ∩ A = 0, ψ N ′′ 1 = N ′′ 1 ⊕ N ′′ written as the matrix φ = (cid:18) πN1 φιN1 πN1φιN2 πN2φιN2 (cid:19). Then ker(eφ) = φ−1(ker e) = πN2 φι1 φ−1(N1) ⊇ ker(πN2 φι1) ⊕ ker(πN2 φιN2 ) ⊇ A = (A ⊕ N2) ∩ N1 = (A ⊕ N2) ∩ ker e. Since A ⊕ N2 is essential in MR, by Lemma 2.7, there exist h ∈ S, j ∈ J(S) such that eφ = he + j. Thus eφ ∈ I. / / / v THE ENDOMORPHISM RING OF A SQUARE-FREE INJECTIVE MODULE 5 References [1] Facchini, A.: Direct-sum decompositions of modules with semilocal endomorphism rings. Bull. Math. Sci. 2 (2), , 225 -- 279, (2013). [2] Lam, T.Y.: Lectures on modules and rings. Graduate Texts in Mathematics, Vol. 189, Springer, Berlin, Heidelberg, New York (1998). [3] Lam, T.Y, Dugas, A.S.: Quasi-duo rings and stable range descent. J. Pure Appl. Algebra. 195 (3), 243 -- 259 (2005). [4] Matlis, E.: Injective modules over Noetherian rings. Pacific J. Math. 8, 51 -- 528 ( 1958). [5] Yu, H.-P.: On quasi-duo rings. Glasgow Math. J. 1 37 (1), 21 -- 31 (1995). Mathematisch Instituut, Leiden Universiteit, Niels Bohrweg 1,2333 CA Leiden,The Netherlands. Current address: Dipartimento di Matematica, Universit`a degli Studi di Padova, Via Trieste 63, 35121 Padova, Italy. E-mail address: [email protected]
1804.08441	1	1804	2018-04-20T03:22:02	Fractal nil graded Lie, associative, Poisson, and Jordan superalgebras	[ "math.RA" ]	We construct a just infinite fractal 3-generated Lie superalgebra $\mathbf Q$ over arbitrary field, which gives rise to an associative hull $\mathbf A$, a Poisson superalgebra $\mathbf P$, and two Jordan superalgebras $\mathbf J$, $\mathbf K$. One has a natural filtration for $\mathbf A$ which associated graded algebra has a structure of a Poisson superalgebra and $\mathrm{gr} \mathbf A\cong\mathbf P$, also $\mathbf P$ admits an algebraic quantization. The Lie superalgebra $\mathbf Q$ is finely $\mathbb Z^3$-graded by multidegree in the generators, $\mathbf A$, $\mathbf P$ are $\mathbb Z^3$-graded, while $\mathbf J$, $\mathbf K$ are $\mathbb{Z}^4$-graded. These five superalgebras have clear monomial bases and slow polynomial growth. We describe multihomogeneous coordinates of bases of $\mathbf Q$, $\mathbf A$, $\mathbf P$ in space as bounded by "almost cubic paraboloids". A similar hypersurface in $\mathbb R^4$ bounds monomials of $\mathbf J$, $\mathbf K$. Constructions of the paper can be applied to Lie superalgebras studied before and get Poisson and Jordan superalgebras as well. The algebras ${\mathbf Q}$, ${\mathbf A}$, and the algebras without unit $\mathbf P^o$, $\mathbf J^o$, $\mathbf K^o$ are direct sums of two locally nilpotent subalgebras and there are continuum such decompositions. Also, $\mathbf Q=\mathbf Q_{\bar 0}\oplus \mathbf Q_{\bar 1}$ is a nil graded Lie superalgebra. In case $\mathrm{char}\, K=2$, $\mathbf Q$ has a structure of a restricted Lie algebra with a nil $p$-mapping. The Jordan superalgebra $\mathbf K$ is just infinite nil finely $\mathbb Z^4$-graded, while such examples (say, analogues of the Grigorchuk group) of Lie and Jordan algebras in characteristic zero do not exist. We call $\mathbf Q$, $\mathbf A$, $\mathbf P$, $\mathbf J$, $\mathbf K$ fractal because they contain infinitely many copies of themselves.	math.RA	math	FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS VICTOR PETROGRADSKY AND I.P. SHESTAKOV Abstract. The Grigorchuk and Gupta-Sidki groups play fundamental role in modern group theory. They are natural examples of self-similar finitely generated periodic groups. The first author constructed their analogue in case of restricted Lie algebras of characteristic 2 [50], Shestakov and Zelmanov extended this con- struction to an arbitrary positive characteristic [66]. Thus, we have examples of finitely generated restricted Lie algebras with a nil p-mapping. In characteristic zero, similar examples of Lie and Jordan algebras do not exist by results of Martinez and Zelmanov [43] and [76]. The first author constructed analogues of the Grigorchuk and Gupta-Sidki groups in the world of Lie superalgebras of arbitrary characteristic, the virtue of that construction is that Lie superalgebras have clear monomial bases [51], they have slow polynomial growth. As an analogue of periodicity, Z2-homogeneous elements are ad-nilpotent. A recent example of a Lie superalgebra is of linear growth, of finite width 4, just infinite but not hereditary just infinite [13]. By that examples, an extension of the result of Martinez and Zelmanov [43] for Lie superalgebras of characteristic zero is not valid. Now, we construct a just infinite fractal 3-generated Lie superalgebra Q over arbitrary field, which gives rise to an associative hull A, a Poisson superalgebra P, and two Jordan superalgebras J and K, the latter being a factor algebra of J. In case char K 6= 2, A has a natural filtration, which associated graded algebra has a structure of a Poisson superalgebra such that gr A ∼= P, also P admits an algebraic quantization using a deformed superalgebra A(t). The Lie superalgebra Q is finely Z3-graded by multidegree in the generators, A, P are also Z3-graded, while J and K are Z4-graded by multidegree in four generators. By virtue of our construction, these five superalgebras have clear monomial bases and slow polynomial growth. We describe multihomogeneous coordinates of bases of Q, A, P in space as bounded by "almost cubic paraboloids". We determine a similar hypersurface in R4 that bounds monomials of J and K. Constructions of the paper can be applied to Lie (super)algebras studied before to obtain Poisson and Jordan superalgebras as well. The algebras Q, A, and the algebras without unit Po, Jo, Ko are direct sums of two locally nilpotent subalgebras and there are continuum such decompositions. Also, Q = Q¯0 ⊕ Q¯1 is a nil graded Lie superal- gebra, so, Q again shows that an extension of the result of Martinez and Zelmanov for Lie superalgebras of characteristic zero is not valid. In case char K = 2, Q has a structure of a restricted Lie algebra with a nil p-mapping. The Jordan superalgebra K is nil finely Z4-graded, in contrast with non-existence of such ex- amples (roughly speaking, analogues of the Grigorchuk group) of Jordan algebras in characteristic zero [76]. Also, K is of slow polynomial growth, just infinite, but not hereditary just infinite. We call the superalgebras Q, A, P, J, K fractal because they contain infinitely many copies of themselves. In Section 1 we survey known results, Section 2 supplies basic definitions. In Section 3 we briefly describe constructions and formulate main properties of our five main objects: a Lie superalgebra Q, its associative hull A, a related Poisson superalgebra P, and two Jordan superalgebras J, K. The present research is a continuation of a series of papers on fractal (self-similar) (restricted) Lie (super)algebras, the main feature is that we extend the results to the classes of Poisson and Jordan superalgebras. 1. Introduction: Self-similar groups and algebras 1.1. Golod-Shafarevich algebras and groups. The General Burnside Problem puts the question whether a finitely generated periodic group is finite. The first negative answer was given by Golod and Shafarevich, who proved that, for each prime p, there exists a finitely generated infinite p-group [22]. The construction is based on a famous construction of a family of finitely generated infinite dimensional associative nil- algebras [22]. This construction also yields examples of infinite dimensional finitely generated Lie algebras L such that (ad x)n(x,y)(y) = 0, for all x, y ∈ L, the field being arbitrary [23]. The field being of positive 2000 Mathematics Subject Classification. 16P90, 16N40, 16S32, 17B50, 17B65, 17B66, 17B70, 17A70, 17B63, 17C10, 17C50, 17C70 . Key words and phrases. restricted Lie algebras, p-groups, growth, self-similar algebras, nil-algebras, graded algebras, Lie superalgebra, Lie algebras of differential operators, Poisson superalgebras, Jordan superalgebras, algebraic quantization. The first author was partially supported by grants CNPq 309542/2016-2, FAPESP 2016/18068-9. The second author was partially supported by grants FAPESP 2014/09310-5, CNPq 2014/09310-5 . 1 2 VICTOR PETROGRADSKY AND I.P. SHESTAKOV characteristic p, one obtains an infinite dimensional finitely generated restricted Lie algebra L such that the p-mapping is nil, namely, x[pn(x)] = 0, for all x ∈ L. This gives a negative answer to a question of Jacobson whether a finitely generated restricted Lie algebra L is finite dimensional provided that each element x ∈ L is algebraic, i.e. satisfies some p-polynomial fp,x(x) = 0 ([29, Ch. 5, ex. 17]). It is known that the construction of Golod yields associative nil-algebras of exponential growth. Using specially chosen relations, Lenagan and Smoktunowicz constructed associative nil-algebras of polynomial growth [38]. On further developments concerning Golod-Shafarevich algebras and groups see [74], [17]. A close by spirit but different construction was motivated by respective group-theoretic results. A re- stricted Lie algebra G is called large if there is a subalgebra H ⊂ G of finite codimension such that H admits a surjective homomorphism on a nonabelian free restricted Lie algebra. Let K be a perfect at most count- able field of positive characteristic. Then there exist infinite-dimensional finitely generated nil restricted Lie algebras over K that are residually finite dimensional and direct limits of large restricted Lie algebras [2]. 1.2. Grigorchuk and Gupta-Sidki groups. The construction of Golod is rather undirect, Grigorchuk gave a direct and elegant construction of an infinite 2-group generated by three elements of order 2 [24]. This group was defined as a group of transformations of the interval [0, 1] from which rational points of the form {k/2n 0 ≤ k ≤ 2n, n ≥ 0} are removed. For each prime p ≥ 3, Gupta and Sidki gave a direct construction of an infinite p-group on two generators, each of order p [27]. This group was constructed as a subgroup of an automorphism group of an infinite regular tree of degree p. The Grigorchuk and Gupta-Sidki groups are counterexamples to the General Burnside Problem. More- over, they gave answers to important problems in group theory. So, the Grigorchuk group and its further generalizations are first examples of groups of intermediate growth [25], thus answering in negative to a conjecture of Milnor that groups of intermediate growth do not exist. The construction of Gupta-Sidki also yields groups of subexponential growth [18]. The Grigorchuk and Gupta-Sidki groups are self-similar. Now self-similar, and so called branch groups, form a well-established area in group theory [26, 47]. Below we discuss existence of analogues of the Grigorchuk and Gupta-Sidki groups for other algebraic structures. 1.3. Self-similar nil graded associative algebras. The study of these groups lead to investigation of group rings and other related associative algebras [68]. In particular, there appeared self-similar associative algebras defined by matrices in a recurrent way [5]. Sidki suggested two examples of self-similar associative matrix algebras [69]. A more general family of associative algebras was introduced in [55], this family gen- eralizes the second example of Sidki [69], also it yields a realization of a Fibonacci restricted Lie algebras (see below) in terms of self-similar matrices [55]. Another important feature of some associative algebras A constructed in [55] is that they are sums of two locally nilpotent subalgebras A = A+ ⊕ A− (see similar de- compositions (1) below). Recall that an algebra is said locally nilpotent if every finitely generated subalgebra is nilpotent. But the desired analogues of the Grigorchuk and Gupta-Sidki groups should be (self-similar) associative nil-algebras, in a standard way yielding new examples of finitely generated periodic groups. But such examples are not known yet. On similar open problems in theory of infinite dimensional algebras see review [75]. 1.4. Self-similar nil restricted Lie algebras, Fibonacci Lie algebra. Unlike associative algebras, for restricted Lie algebras, natural analogues of the Grigorchuk and Gupta-Sidki groups are known. Namely, over a field of characteristic 2, the first author constructed an example of an infinite dimensional restricted Lie algebra L generated by two elements, called a Fibonacci restricted Lie algebra [50]. Let char K = p = 2 and R = K[tii ≥ 0]/(tp , i ≥ 0. Define the following two derivations of R: i i ≥ 0) a truncated polynomial ring. Put ∂i = ∂ ∂ti v1 = ∂1 + t0(∂2 + t1(∂3 + t2(∂4 + t3(∂5 + t4(∂6 + ··· ))))); ∂2 + t1(∂3 + t2(∂4 + t3(∂5 + t4(∂6 + ··· )))). v2 = These two derivations generate a restricted Lie algebra L = Liep(v1, v2) ⊂ Der R and an associative algebra A = Alg(v1, v2) ⊂ End R. The Fibonacci restricted Lie algebra has a slow polynomial growth with Gelfand- Kirillov dimension GKdim L = log(√5+1)/2 2 ≈ 1.44 [50]. Further properties of the Fibonacci restricted Lie algebra and its generalizations are studied in [54, 56]. Probably, the most interesting property of L is that it has a nil p-mapping [50], which is an analog of the periodicity of the Grigorchuk and Gupta-Sidki groups. We do not know whether the associative hull A is a FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 3 nil-algebra. We have a weaker statement. The algebras L, A, and the augmentation ideal of the restricted enveloping algebra u = ωu(L) are direct sums of two locally nilpotent subalgebras [54]: L = L+ ⊕ L−, A = A+ ⊕ A−, u = u+ ⊕ u−. (1) There are examples of infinite dimensional associative algebras which are direct sums of two locally nilpotent subalgebras [34, 15]. Infinite dimensional restricted Lie algebras can have different decompositions into a direct sum of two locally nilpotent subalgebras [57]. In case of arbitrary prime characteristic, Shestakov and Zelmanov suggested an example of a finitely gen- erated restricted Lie algebra with a nil p-mapping [66]. That example yields the same decompositions (1) for some primes [37, 55]. An example of a p-generated nil restricted Lie algebra L, characteristic p being arbi- trary, was studied in [57]. The virtue of that example is that for all primes p we have decompositions (1) into direct sums of two locally nilpotent subalgebras. But computations for that example are rather complicated. Observe that only the original example has a clear monomial basis [50, 54]. In other examples, elements of a Lie algebra are linear combinations of monomials, to work with such linear combinations is sometimes an essential technical difficulty, see e.g. [66, 57]. A family of nil restricted Lie algebras of slow growth having good monomial bases is constructed in [52], these algebras are close relatives of a two-generated Lie superalgebra of [51]. 1.5. Narrow groups and Lie algebras. Let G be a group and G = G1 ⊇ G2 ⊇ ··· its lower central series. One constructs a related N-graded Lie algebra LK(G) = ⊕i≥1Li, where Li = Gi/Gi+1 ⊗Z K, i ≥ 1. A product is given by [aGi+1, bGj+1] = (a, b)Gi+j+1, where a ∈ Gi, b ∈ Gj, and (a, b) = a−1b−1ab the group commutator. A residually p-group G is said of finite width if all factors Gi/Gi+1 are finite groups with uniformly bounded orders. The Grigorchuk group G is of finite width, namely, dimF2 Gi/Gi+1 ∈ {1, 2} for i ≥ 2 [62, 7]. In particular, the respective Lie algebra L = LK(G) = ⊕i≥1Li has a linear growth. Bartholdi presented LK(G) as a self-similar restricted Lie algebra and proved that the restricted Lie algebra LF2(G) is nil while LF4(G) is not nil [6]. Also, LK(G) is nil graded, namely, for any homogeneous element x ∈ Li, i ≥ 1, the mapping ad x is nilpotent, because the group G is periodic. A Lie algebra L is called of maximal class (or filiform), if the associated graded algebra with respect to (2) the lower central series gr L = ∞⊕n=1 dim gr L1 = 2, gr Ln, where gr Ln = Ln/Ln+1, n ≥ 1, satisfies dim gr Ln ≤ 1, n ≥ 2, gr Ln+1 = [gr L1, gr Ln], n ≥ 1, in particular, gr L is generated by gr L1. An infinite dimensional filiform Lie algebra L has the smallest nontrivial growth function: γL(n) = n + 1, n ≥ 1. In case of positive characteristic, there are uncountably many such algebras [11]. Nevertheless, in case p > 2, they were classified in [12]. There are generalizations of filiform Lie algebras. Naturally N-graded Lie algebras over R and C satisfying the condition dim Ln + dim Ln+1 ≤ 3, n ≥ 1, are classified recently by Millionschikov [45]. More generally, an N-graded Lie algebra L = ∞⊕n=1 Ln is said of finite width d in the case that dim Ln ≤ d, n ≥ 1, the integer d being minimal. Pro-p-groups and N-graded Lie algebras cannot be simple. Instead, appears an important notion of being just infinite, namely, not having non-trivial normal subgroups (ideals) of infinite index (codimension). A group (algebra) is said hereditary just infinite if and only if any normal subgroup (ideal) of finite index (codimension) is just infinite. The Gupta-Sidki groups were the first in the class of periodic groups to be shown to be just infinite [28]. The Grigorchuk group is also just infinite but not hereditary just infinite [26]. 1.6. Lie algebras in characteristic zero. Since the Grigorchuk group is of finite width, a right analogue of it should be a Lie algebra of finite width having ad-nil elements, in the next result the components are of bounded dimension and consist of ad-nil elements. Informally speaking, there are no "natural analogues" of the Grigorchuk and Gupta-Sidki groups in the world of Lie algebras of characteristic zero, strictly in terms of the following result. Theorem 1.1 (Martinez and Zelmanov [43]). Let L = ⊕α∈ΓLα be a Lie algebra over a field K of charac- teristic zero graded by an abelian group Γ. Suppose that i) there exists d > 0 such that dimK Lα ≤ d for all α ∈ Γ, ii) every homogeneous element a ∈ Lα, α ∈ Γ, is ad-nilpotent. Then the Lie algebra L is locally nilpotent. 4 VICTOR PETROGRADSKY AND I.P. SHESTAKOV 1.7. Fractal nil graded Lie superalgebras. In the world of Lie superalgebras of an arbitrary character- istic, the first author constructed analogues of the Grigorchuk and Gupta-Sidki groups [51]. Namely, two Lie superalgebras R, Q were constructed, which are also analogues of the Fibonacci restricted Lie algebra and other (restricted) Lie algebras mentioned above. Constructions of both Lie superalgebras R, Q are similar, computations for R are simpler, but Q enjoys some more specific interesting properties. The virtue of both examples is that they have clear monomial bases. They have slow polynomial growth, namely, GKdim R = log3 4 ≈ 1.26 and GKdim Q = log3 8 ≈ 1.89. Thus, both Lie superalgebras are of infinite width. In both examples, ad a is nilpotent, a being an even or odd element with respect to the Z2-gradings as Lie superalgebras. This property is an analogue of the periodicity of the Grigorchuk and Gupta-Sidki groups. The Lie superalgebra R is Z2-graded, while Q has a natural fine Z3-gradation with at most one-dimensional components (See on importance of fine gradins for Lie and associative algebras [3, 16]). In particular, Q is a nil finely graded Lie superalgebra, which shows that an extension of Theorem 1.1 (Martinez and Zel- manov [43]) for the Lie superalgebras of characteristic zero is not valid. Also, Q has a Z2-gradation which yields a continuum of different decompositions into sums of two locally nilpotent subalgebras Q = Q+⊕ Q−. Both Lie superalgebras are self-similar, they also contain infinitely many copies of itself, we call them fractal due to the last property. (Except this paragraph, Q denotes another Lie superalgebra, one of the main object of this paper). In [13], we construct a similar but simpler and "smaller" example. Namely, we construct a 2-generated fractal Lie superalgebra R (the same notation as above but this is a different algebra) over arbitrary field. This Lie superalgebra R is Z2-graded by multidegree in the generators and the Z2-components are at most one-dimensional. As an analogue of periodicity, we establish that homogeneous elements of the Z2-grading R = R¯0 ⊕ R¯1 are ad-nilpotent. In case of N-graded algebras, a close analogue to being simple is being just infinite. Unlike previous examples of Lie superalgebras [51], we are able to prove that R is just infinite, but not hereditary just infinite. This example is close to the smallest possible one, because R has a linear growth with a growth function γR(m) ≈ 3m, as m → ∞. Moreover, its degree N-gradation is of finite width 4 (char K 6= 2). In case char K = 2, we obtain a Lie algebra of width 2 that is not thin. 1.8. Poisson and Jordan (super)algebras. Poisson algebras naturally appear in different areas of alge- bra, topology and physics. Probably, Poisson algebras were first introduced in 1976 by Berezin [8], see also Vergne [73] (1969). The free Poisson (super)algebras were introduced by Shestakov [64]. Applying Poisson algebras, Shestakov and Umirbaev managed to solve a long-standing problem: they proved that the Nagata automorphism of the polynomial ring in three variables C[x, y, z] is wild [67]. Related algebraic properties of free Poisson algebras were studied by Makar-Limanov, Shestakov and Umirbaev [39, 40]. A basic theory of identical relations for Poisson algebras was developed by Farkas [19, 20]. See further developments on the theory of identical relations of Poisson algebras, in particular, the theory of so called codimension growth in characteristic zero by Mishchenko, Petrogradsky, Regev [46], and Ratseev [60]. Simple finite dimensional nontrivial Jordan superalgebras over an algebraically closed field of characteristic zero were classified [31, 33]. Infinite-dimensional Z-graded simple Jordan superalgebras with a unit element over an algebraically closed field of characteristic zero which components are uniformly bounded are classified in [32]. Recently, just infinite Jordan superalgebras were studied in [77]. Theorem 1.2 (Zelmanov, private communication [76]). Jordan algebras in characteristic zero satisfy a verbatim analogue of Theorem 1.1. Strictly in terms of this result, we say again that there are no natural analogues of the Grigorchuk and Gupta-Sidki groups in the class of Jordan algebras too. On the other hand, the Jordan superalgebra K constructed in the present paper shows that an extension of this result to the Jordan superalgebras is not valid. These facts resemble those for Lie algebras and superalgebras mentioned above. We continue this research and construct a similar but "smaller" example, namely, a fractal nil Jordan superalgebra of finite width in [58]. 2. Basic definitions: superalgebras, growth 2.1. Associative and Lie Superalgebras. Denote N0 = {0, 1, 2, . . .}. By K denote the ground field, hSiK a linear span of a subset S in a K-vector space. Superalgebras appear naturally in physics and mathematics [30, 63, 1]. Put Z2 = {¯0, ¯1}, the group of order 2. A superalgebra A is a Z2-graded algebra A = A¯0 ⊕ A¯1. The elements a ∈ Aα are called homogeneous FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 5 of degree a = α ∈ Z2. The elements of A¯0 are even, those of A¯1 odd. In what follows, if a enters an expression, then it is assumed that a is homogeneous of degree a ∈ Z2, and the expression extends to the other elements by linearity. Let A, B be superalgebras, a tensor product A ⊗ B is the superalgebra whose space is the tensor product of the spaces A and B with the induced Z2-grading and the product: (a1 ⊗ b1)(a2 ⊗ b2) = (−1)b1·a2a1a2 ⊗ b1b2, ai ∈ A, bi ∈ B. An associative superalgebra A is a Z2-graded associative algebra A = A¯0 ⊕ A¯1. A Lie superalgebra is a Z2-graded algebra L = L¯0 ⊕ L¯1 with an operation [ , ] satisfying the axioms (char K 6= 2, 3): • [x, y] = −(−1)x·y[y, x], • [x, [y, z]] = [[x, y], z] + (−1)x·y[y, [x, z]], (super-anticommutativity); (Jacobi identity). Assume that A = A¯0 ⊕ A¯1 is an associative superalgebra. One obtains a Lie superalgebra A(−) by All commutators in the present paper are supercommutators. Long commutators are right-normed: [x, y, z] = [x, [y, z]]. We use a standard notation ad x(y) = [x, y], where x, y ∈ L. supplying the same vector space A with a supercommutator: [x, y] = xy − (−1)x·yyx, If A(−) is abelian, then A is called supercommutative. Let L be a Lie superalgebra, one defines a universal enveloping algebra U (L) = T (L)/(x⊗ y− (−1)x·yy ⊗ x− [x, y] x, y ∈ L), where T (L) is the tensor algebra of the vector space L. Now, the product in L coincides with the supercommutator in U (L)(−). A basis of U (L) is given by PBW-theorem [1, 63]. x, y ∈ A. Let V = V¯0 ⊕ V¯1 be a vector space, we say that it is Z2-graded. The associative algebra of all vector space endomorphisms End V is an associative superalgebra: End V = End¯0 V ⊕ End¯1 V , where Endα V = {φ ∈ End V φ(Vβ ) ⊂ Vα+β, β ∈ Z2}, α ∈ Z2. Thus, End(−) V is a Lie superalgebra, called the general linear superalgebra gl(V ). Let A = A¯0 ⊕ A¯1 be a Z2-graded algebra of arbitrary signature. A linear mapping φ ∈ Endβ A, β ∈ Z2, is a superderivative of degree β if it satisfies φ(a · b) = φ(a) · b + (−1)βaa · φ(b), a, b ∈ A. Denote by Derα A ⊂ Endα A the space of all superderivatives of degree α ∈ Z2. One checks that Der A = Der¯0 A ⊕ Der¯1 A is a subalgebra of the Lie superalgebra End(−) A. All superderivations of the Grassmann algebra Λ(n) = Λ(x1, . . . , xn) is a simple Lie superalgebra W(n) for n ≥ 2. In this paper by a derivation we always mean a superderivation. 2.2. Lie superalgebras in small characteristics. In case char K = 2, 3 the axioms of the Lie superalgebra have to be augmented ([1, section 1.10], [10], [51]). (in case char K = 3). Substituting x = y ∈ L¯1 in the Jacobi identity, we get 2(ad x)2z = [[x, x], z]. In case char K 6= 2 we get an identity • [z, [z, z]] = 0, z ∈ L¯1 (ad x)2z = 1 2 [[x, x], z], x ∈ L¯1, z ∈ L. In the present paper we study Lie superalgebras of the form A(−), they have squares for odd elements: [x, x] = 2x2, x ∈ A(−) . One obtains an identity which is also valid for algebras A(−) in case char K = 2: ¯1 (ad x)2z = [x2, z], x ∈ A(−) ¯1 , z ∈ A(−). (3) So, in case char K = 2, we add more axioms for the Lie superalgebras: • there exists a quadratic mapping (a formal square): ( )[2] : L¯1 → L¯0, x 7→ x[2], x ∈ L¯1, satisfying: (λx)[2] = λ2x[2], x ∈ L¯1, λ ∈ K; (x + y)[2] = x[2] + [x, y] + y[2], (ad x)2z = [x[2], z], x, y ∈ L¯1; (4) x ∈ L¯1, z ∈ L, (a formal substitute of (3)); • [x, x] = 0, x ∈ L¯0. By putting y = x in the second relation above, we get [y, y] = 0, y ∈ L¯1. Thus, a Lie superalgebra in case char K = 2 is just a Z2-graded Lie algebra supplied with a quadratic mapping L¯1 → L¯0, which is similar to the p-mapping (see below). In case p = 2, to get the universal enveloping algebra, we additionally factor out {y ⊗ y − y[2] y ∈ L¯1}. 6 VICTOR PETROGRADSKY AND I.P. SHESTAKOV 2.3. Restricted Lie (super)algebras. Let char K = p > 0. A Lie algebra L is a restricted Lie algebra (or Lie p-algebra), if it is supplied with a unary operation x 7→ x[p], x ∈ L, that satisfies the following axioms [29, 71, 72]: • (λx)[p] = λpx[p], for λ ∈ K, x ∈ L; • ad(x[p]) = (ad x)p, x ∈ L; • (x + y)[p] = x[p] + y[p] + Pp−1 polynomial ad(tx + y)p−1(x) ∈ L[t]. i=1 si(x, y), x, y ∈ L, where isi(x, y) is the coefficient of ti−1 in the This notion is motivated by the following observation. Let A be an associative algebra over a field K, char K = p > 0. Then the mapping x 7→ xp, x ∈ A(−), satisfies these conditions considered in the Lie algebra A(−). A restricted Lie superalgebra L = L¯0 ⊕ L¯1 is a Lie superalgebra such that the even component L¯0 is a restricted Lie algebra and L¯0-module L¯1 is restricted, i.e. ad(x[p])y = (ad x)py, for all x ∈ L¯0, y ∈ L¯1 (see. e.g. [44, 1]). Remark that in case char K = 2, the restricted Lie superalgebras and Z2-graded restricted Lie algebras are the same objects. (Let L = L¯0 ⊕ L¯1 be a restricted Lie superalgebra, it has the p-mapping on the even part: L¯0 → L¯0 and the formal square on the odd part: L¯1 → L¯0. We obtain the p-mapping on the whole of algebra by setting (x + y)[2] = x[2] + y[2] + [x, y], x ∈ L¯0, y ∈ L¯1). Let L be a restricted Lie (super)algebra, and J the ideal of the universal enveloping algebra U (L) generated by {x[p] − xp x ∈ L¯0}. Then u(L) = U (L)/J is the restricted enveloping algebra. In this algebra, the formal operation x[p] coincides with the ordinary power xp for all x ∈ L¯0. One has an analogue of PBW-theorem describing a basis of u(L) [29, p. 213], [1]. Let L be a Lie (super)algebra. One defines the lower central series as L1 = L and Ln = [L, Ln−1], n ≥ 2. In case of a restricted Lie In case char K = 2 the terms above are augmented by hx2x ∈ (L[n/2])¯1iK. (super)algebra, we also add hxpx ∈ (L[n/p])¯0iK. 2.4. Poisson superalgebras. A Z2-graded vector space A = A¯0 ⊕ A¯1 is called a Poisson superalgebra provided that, beside the addition, A has two K-bilinear operations as follows: where a, b ∈ A. We assume that A is supercommutative, i.e. a · b = (−1)a·bb · a, for all a, b ∈ A. {a, b}, where a, b ∈ A. • A = A¯0 ⊕ A¯1 is an associative superalgebra with unit whose multiplication is denoted by a· b (or ab), • A = A¯0 ⊕ A¯1 is a Lie superalgebra whose product is traditionally denoted by the Poisson bracket • these two operations are related by the super Leibnitz rule: {a · b, c} = a · {b, c} + (−1)b·c{a, c} · b, a, b, c ∈ A. Let L be a Lie superalgebra, {Unn ≥ 0} the natural filtration of its universal enveloping algebra U (L). Consider the symmetric algebra S(L) = gr U (L) = ∞⊕n=0 Un/Un+1 (see [14]). Recall that S(L) is identified with a supercommutative algebra K[vi i ∈ I] ⊗ Λ(wj, j ∈ J), where {vi i ∈ I}, {wj j ∈ J}, are bases of L¯0, L¯1, respectively. Define a Poisson bracket by setting {v, w} = [v, w], v, w ∈ L, and extending to the whole of S(L) by linearity and using the Leibnitz rule. Thus, S(L) is turned into a Poisson superalgebra, called the symmetric algebra of L. Let L(X) be the free Lie superalgebra generated by a graded set X, then S(L(X)) is a free Poisson superalgebra [64]. Let char K = 2, the axioms of a Lie superalgebra require existence of a formal square y 7→ y[2] for all odd y. Consider a free Poisson superalgebra A = A¯0 ⊕ A¯1 over Q, let a ∈ A¯0, b ∈ A¯1, then (ab)[2] = 2{ab, ab} = 1 2 ({a, a}bb + aa{b, b} + 2ab{a, b}) = a2b[2] + ab{a, b}. Thus, we add additional axioms for a Poisson superalgebra in case char K = 2: 1 • (ab)[2] = a2b[2] + ab{a, b} for all a ∈ A¯0, b ∈ A¯1; • b2 = 0, for all b ∈ A¯1. One checks that validity of these axioms on any basis imply them for all elements (the second axiom is needed here). Also, the computation above yields an additional axiom for a restricted Poisson algebra A in case char K = 2: • (ab)[2] = a2b[2] + a[2]b2 + ab{a, b} for all a ∈ A. Again, one checks that it is sufficient to verify validity of this axiom on any basis. Observe that the case p = 2 was not considered in a definition of a restricted Poisson algebra given for all p > 2 in [9, 4]. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 7 Let A, P be Poisson superalgebras, their tensor product A⊗ P is a Poisson superalgebra with operations: (a⊗ v) · (b ⊗ w) = (−1)vbab⊗ vw and {a⊗ v, b ⊗ w} = (−1)vb({a, b}⊗ vw + ab⊗ {v, w}), where a, b ∈ A, v, w ∈ P . Let Λ(n) = Λ(x1, . . . , xn) be the Grassmann algebra in n variables. It is an associative superalgebra, where the Z2-grading Λ(n) = Λ¯0(n)⊕ Λ¯1(n) is given by parity of monomials in the generators. One supplies Λ(n) with a bracket: {f, g} = (−1)f−1 ∂f ∂xi ∂g ∂xi , f, g ∈ Λ(n). n Xi=1 This bracket is induced by relations {xi, xj} = δi,j, 1 ≤ i, j ≤ n. Then Λ(n) is a simple Poisson superalgebra. Consider a modification of this construction. Let Hn = Λ(x1, . . . , xn, y1, . . . , yn) be the Grassmann superalgebra supplied with a bracket determined by: {xi, yj} = δi,j, {xi, xj} = {yi, yj} = 0 for 1 ≤ i, j ≤ n. We obtain a simple Hamiltonian Poisson superalgebra with a bracket: {f, g} = (−1)f−1 n Xi=1(cid:18) ∂f ∂xi ∂g ∂yi + ∂f ∂yi ∂g ∂xi(cid:19), f, g ∈ Hn. Let P = P¯0⊕P¯1 be a Poisson superalgebra with products · and { , }. Recall that an algebraic quantization of P is a polynomial extension P [t] supplied with an associative product ∗ that agrees with the grading P [t] = P¯0[t] ⊕ P¯1[t] and such that (see e.g. [64]): • a ∗ b = a · b (mod t), • a ∗ b − (−1)abb ∗ a = t{a, b} (mod t2), • f ∗ t = t ∗ f = f t, f ∈ P [t]. a, b ∈ P ; a, b ∈ P ; 2.5. Jordan superalgebras. While studying Jordan (super)algebras we always assume that char K 6= 2. A Jordan algebra is an algebra J satisfying the identities • ab = ba; • a2(ca) = (a2c)a. A Jordan superalgebra is a Z2-graded algebra J = J¯0 ⊕ J¯1 satisfying the graded identities: • ab = (−1)abba; • (ab)(cd) + (−1)bc(ac)(bd) + (−1)(b+c)d(ad)(bc) = ((ab)c)d + (−1)b(c+d)+cd((ad)c)b + (−1)a(b+c+d)+cd((bd)c)a. 1 Let A = A¯0 ⊕ A¯1 be an associative superalgebra. The same space supplied with the product a ◦ b = 2 (ab + (−1)abba) is a Jordan superalgebra A(+). A Jordan superalgebra J is called special if it can be embedded into a Jordan superalgebra of the type A(+). Also, J is called i-special (or weakly special) if it is a homomorphic image of a special one. I.L. Kantor suggested the following doubling process, which is applied to a Poisson (super)algebra A and the result is a Jordan superalgebra Kan(A) [33]. The K-module Kan(A) is the direct sum A ⊕ ¯A, where ¯A is a copy of A, let a ∈ A then ¯a denotes the respective element in ¯A. Also, ¯A is supplied with the opposite Z2-grading, i.e., ¯a = 1 − a for a Z2-homogeneous a ∈ A. The multiplication • on Kan(A) is defined by: a • b = ab, ¯a • b = (−1)bab, a • ¯b = ab, ¯a • ¯b = (−1)b{a, b}, a, b ∈ A. This construction is important because it yielded a new series of finite dimensional simple Jordan superal- gebras Kan(Λ(n)), n ≥ 2 [33, 35]. 2.6. Growth. We recall the notion of growth. Let A be an associative (or Lie) algebra generated by a finite set X. Denote by A(X,n) the subspace of A spanned by all monomials in X of length not exceeding n, n ≥ 0. In case of a Lie superalgebra of char K = 2 we also consider formal squares of odd monomials of length at most n/2. If A is a restricted Lie algebra, put A(X,n) = h [x1, . . . , xs]pk xi ∈ X, spk ≤ niK [48]. Similarly, 8 VICTOR PETROGRADSKY AND I.P. SHESTAKOV one defines the growth for restricted Lie superalgebras. In either situation, one defines an (ordinary) growth function: γA(n) = γA(X, n) = dimK A(X,n), n ≥ 0. Let f, g : N → R+ be eventually increasing and positive valued functions. Write f (n) 4 g(n) if and only if there exist positive constants N, C such that f (n) ≤ g(Cn) for all n ≥ N . Introduce equivalence f (n) ∼ g(n) if and only if f (n) 4 g(n) and g(n) 4 f (n). Different generating sets of an algebra yield equivalent growth functions [36]. It is well known that the exponential growth is the highest possible growth for finitely generated Lie and associative algebras. A growth function γA(n) is compared with polynomial functions nα, α ∈ R+, by computing the upper and lower Gelfand-Kirillov dimensions [36]: GKdim A = lim n→∞ GKdim A = lim n→∞ ln γA(n) ln n ln γA(n) ln n = inf{α > 0 γA(n) 4 nα}; = sup{α > 0 γA(n) < nα}. By Bergman's theorem, the Gelfand-Kirillov dimension of an associative algebra cannot belong to the interval (1, 2) [36]. Similarly, there are no finitely generated Jordan algebras with Gelfand-Kirillov dimension strictly between 1 and 2 [42]. Such a gap for Lie algebras does not exist, the Gelfand-Kirillov dimension of a finitely generated Lie algebra can be arbitrary number {0} ∪ [1, +∞) [49]. a weight growth function: Assume that generators X = {x1, . . . , xk} are assigned positive weights wt(xi) = λi, i = 1, . . . , k. Define γA(n) = dimKhxi1 ··· xim wt(xi1 ) + ··· + wt(xim ) ≤ n, xij ∈ XiK, n ≥ 0. Set C1 = min{λi i = 1, . . . , k}, C2 = max{λi i = 1, . . . , k}, then γA(C1n) ≤ γA(n) ≤ γA(C2n) for n ≥ 1. Thus, we obtain an equivalent growth function γA(n) ∼ γA(n). Therefore, we can use the weight growth function γA(n) in order to compute the Gelfand-Kirillov dimensions. By f (n) ≈ g(n), n → ∞, denote that lim n→∞ f (n)/g(n) = 1. Similarly, one studies the growth for Poisson and Jordan superalgebras. Suppose that L is a Lie (super)algebra and X ⊂ L. By Lie(X) denote the subalgebra of L generated by X, (including application of the quadratic mapping in case char K = 2). Let L be a restricted Lie (super)algebra, by Liep(X) denote the restricted subalgebra of L generated by X. Assume that X is a subset of an associative algebra A. Write Alg(X) ⊂ A to denote the associative subalgebra (without unit) generated by X. In case of Poisson and Jordan superalgebras we use notations Poisson(X) and Jord(X). A grading of an algebra is called fine if it cannot be splitted by taking a bigger grading group (see definitions in [3, 16]). 2.7. Lie superalgebra W(ΛI ) of special superderivations. Assume that I is a well-ordered set of arbitrary cardinality. Put Z2 = {0, 1}. Let ZI 2 = {α : I → Z2} be a set of functions with finitely many nonzero values. Suppose that α ∈ ZI 2 has nonzero values at {i1, . . . , it} ⊂ I, where i1 < ··· < it, put xα = xi1 xi2 ··· xit and α = t. Now {xα α ∈ ZI 2} is a basis of the Grassmann algebra ΛI = Λ(xi i ∈ I), which is an associative superalgebra ΛI = Λ¯0 ⊕ Λ¯1, all xi, i ∈ I, being odd. Let ∂i, i ∈ I, denote the superderivatives of Λ, which are determined by the values ∂i(xj) = δij, i, j ∈ I. We identify xi, i ∈ I, with the operator of the left multiplication on ΛI , thus we get odd elements xi ∈ End¯1(ΛI ), i ∈ I. Consider a space of all formal sums W(ΛI ) =(cid:26) Xα∈ZI 2 xα m(α) Xj=1 λα,ij ∂ij (cid:12)(cid:12)(cid:12)(cid:12) λα,ij ∈ K, ij ∈ I(cid:27). (5) It is essential that the sum at each xα, α ∈ ZI 2, is finite. This construction is similar to the Lie algebra of special derivations, see [59], [61], [53]. It is similarly verified that the product in W(ΛI ) is well defined and W(ΛI ) acts on ΛI by superderivations. 3. Main results: superalgebras Q, A, P, J, K, and their properties In this paper, we study the following five objects. A core of our constructions is a Lie superalgebra Q. Next we construct the associative hull A, a related Poisson superalgebra P, and two Jordan superalgebras J and K. We call these superalgebras fractal because they contain infinitely many copies of themselves. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 9 Let us briefly describe their constructions, the next picture shows relations between constructions. ✏✶ ✏ ✏ Q P P Pq A ❄ P ✲ J K ✲ ✏✏✶ ✏ ✏ Consider the Grassmann algebra in infinitely many variables Λ = Λ(xi i ≥ 0). Let ∂i be its superderiva- tive defined by ∂i(xj ) = δi,j, i, j ≥ 0. Observe that {xi, ∂i i ≥ 0} are odd elements of the associative superalgebra End Λ, where xi is identified with the left miltiplication on Λ. These elements anticommute except for nontrivial relations: Now we define pivot elements: [∂i, xi] = ∂ixi + xi∂i = 1; x2 i = 0, ∂2 i = 0, i ≥ 0. vi = ∂i + xixi+1(∂i+3 + xi+3xi+4(∂i+6 + xi+6xi+7(∂i+9 + ··· ))), i ≥ 0. (6) The action of the pivot elements on the Grassmann letters is well defined and produces letters with smaller indices: 0, 1, xnxn+1 xn+2xn+3xn+4 xn+5 ··· xk−4xk−3xk−2, 0, k < n; k = n; k = n + 3l, l ≥ 1; k − n 6= 0 (mod 3); (7) vn(xk) =  where xi denote omitted variables. Thus, we obtain a sequence of superderivatives {vi i ≥ 0} ⊂ Der Λ, moreover, they belong to W(Λ). First, we define a Lie superalgebra Q = Lie(v0, v1, v2) ⊂ W(Λ) ⊂ Der Λ generated by {v0, v1, v2}. Second, we take its associative hull, namely, we consider the associative superal- gebra A = Alg(v0, v1, v2) ⊂ End Λ generated by {v0, v1, v2}. (We warn that another Lie superalgebra was denoted by Q in [51], see also subsection 1.7). We start the present paper with a study of properties of the algebras Q and A. Next, in Section 9 we consider the Grassmann algebra H∞ = Λ(xi, yii ≥ 0) which is turned into a Poisson superalgebra by a bracket determined by relations: In its completion H∞, the next elements will be referred to as the pivot elements as well: {yi, xj} = δi,j, {xi, xj} = {yi, yj} = 0, i, j ≥ 0. Vi = yi + xixi+1(yi+3 + xi+3xi+4(yi+6 + xi+6xi+7(yi+9 + ··· ))) ∈ H∞, We actually obtain the same Lie superalgebra: Q = Lie(v0, v1, v2) ∼= Lie(V0, V1, V2). Third, we define a Poisson subalgebra P = Poisson(V0, V1, V2) ⊂ H∞ generated by {V0, V1, V2}. Using the Kantor double, we construct the forth object, a Jordan superalgebra J = Kan(P(V0, V1, V2)) = P ⊕ ¯P and prove that J = Jord(V0, V1, V2, ¯1) (Section 12). Finally, a Jordan superalgebra K is a factor algebra of J, it also can be constructed directly as a double i ≥ 0. K = J or(Q), namely, as a vector space supplied with an operation as follows (Section 13): K = h1i ⊕ Q ⊕ h¯1i ⊕ ¯Q, ¯x • ¯y = [x, y], x • ¯1 = (−1)x¯1 • x = ¯x, x, y ∈ Q; 1 the unit. Now we formulate main properties of these five superalgebras established in the paper. i) Section 4 yields multiplication rules of the Lie superalgebra Q. ii) Q has a clear monomial basis consisting of standard monomials of two types (char K 6= 2, Theo- rem 5.1). In case char K = 2, a basis of Q consists of monomials of the first type and squares of the pivot elements (Corollary 5.2), and Q coincides with the restricted Lie (super)algebra Liep(v0, v1, v2). iii) In Section 9 we define the Poisson superalgebra P(V0, V1, V2) = Poisson(V0, V1, V2), determined (actually, generated) by the Lie superalgebra Q. iv) We describe monomial bases of the Poisson superalgebra P and associative hull A. In case char K 6= 2, we prove that for a filtration of A, the associated graded algebra has a structure of a Poisson su- peralgebra such that gr A ∼= P, in particular, both algebras have "the same" bases. Also, the Poisson superalgebra P admits an algebraic quantization using a deformed superalgebra A(t) (Section 10). 10 VICTOR PETROGRADSKY AND I.P. SHESTAKOV v) We essentially use weight functions additive on products of monomials. We prove that Q, A, P, J, and K are N3 0-graded by multidegree in three generators (Theorem 6.2, Lemma 11.1, Lemma 12.4, but the Jordan superalgebras have one more generator). This allows us to introduce coordinate systems in space: multidegree coordinates (X1, X2, X3), and twisted weight coordinates (Y1, Y2, Y3) (Section 6). vi) Components of the N3 0-gradation of Q by multidegree in the generators are at most one-dimensional (Theorem 7.2), so the N3 0-grading of Q is fine. vii) Q is just infinite but nor hereditary just infinite (Section 7). viii) We compute initial coefficients of generating functions of Q (Section 7). The results and proofs on basis monomials of Q are illustrated by Figure 1. ix) We find bounds on weights of the basis monomials of Q, P, and A (Sections 8, 11) and prove that images of their monomials in space are inside "almost cubic paraboloids" (Theorem 8.5, see Figure 1, and Theorem 11.4). Asymptotically, a nonzero share of lattice points inside the first paraboloid corresponds to monomials of Q (Corollary 8.7). x) We conjecture that the superalgebras Q, A, and P are not self-similar. We discuss the notion of self-similarity for Jordan superalgebras in [58]. xi) The Jordan superalgebras J, K are N4 0-graded by multidegree in the generators (Corollary 12.7), we determine a hypersurface in R4 that bounds monomials of J and K (Theorems 12.11, 13.4). xii) Q, A, P, J, K have slow polynomial growth: GKdim Q = GKdim K = logλ 2 ≈ 1.6518 and xiii) J, K are weakly special, but not special (Corollary 12.3, Theorem 13.4). xiv) Q, A, and the algebras without unit Po, Jo, Ko are direct sums of two locally nilpotent subalgebras GKdim A = GKdim P = GKdim J = 2 logλ 2 ≈ 3.3036 (Theorems 8.4, 11.3, 12.9, 13.4). and there are continuum such different decompositions (Theorem 14.2). xv) Q = Q¯0 ⊕ Q¯1 is a nil graded Lie superalgebra (Theorem 14.3). Thus, Q again shows that an extension of Theorem 1.1 (Martinez and Zelmanov [43]) for Lie superalgebras of characteristic zero is not valid. Such a counterexample of a nil finely Z3-graded Lie superalgebra of slow polynomial growth Q was suggested before [51]. There is also a recent counterexample of a nil finely Z2-graded Lie superalgebra of linear growth and of finite width 4 [13]. xvi) In case char K = 2, Q has a structure of a restricted Lie algebra Q = Liep(v0, v1, v2) with a nil p-mapping (Theorem 14.5). xvii) Components of the N4 0-gradation of K by multidegree in the generators are at most one-dimensional (Theorem 13.4), so the N4 0-grading of K is fine. xviii) K is just infinite but nor hereditary just infinite (Theorem 13.4). xix) An extension of Theorem 1.2 to Jordan superalgebras of characteristic zero is not valid. Indeed, K is 0-graded Jordan superalgebra with at most one-dimensional components, where the subalgebra a N4 without unit Ko is nil of bounded degree. xx) The constructions of the paper can be applied to Lie (super)algebras studied before to obtain Poisson and Jordan superalgebras as well. Remark 1. Indeed, one can apply constructions of the paper to two Lie superalgebras of [51] and one more Lie superalgebra of [13] and obtain respective associative, Poisson, and Jordan superalgebras. But these new superalgebras shall enjoy only triangular decompositions (1) as sums of three subalgebras, e.g. Jo = J− ⊕ J0 ⊕ J+, because the roots of that characteristic polynomials are integers. In the present paper we get decompositions into sums of two locally nilpotent subalgebras because of nonintegral roots of the characteristic polynomial. Remark 2. In particular, recall that the Lie superalgebra R constructed in [13] is just infinite, two-generated, nil Z2-graded, with at most one-dimensional Z2-components, of linear growth, moreover, of finite width 4. Namely, its N-gradation by degree in the generators has non-periodic components of dimensions {2, 3, 4}. The arguments of the present paper yield the following. Consider the related Jordan superalgebra K = J or(R). Then K is just infinite, three-generated, Z3-graded with at most one-dimensional components, the ideal without unit Ko is nil of bounded degree. Also, K is of linear growth, moreover, of finite width 4, namely, its N-gradation by degree in the generators has components of dimensions {0, 2, 3, 4}, their sequence is non- periodic [58]. That example also shows that just infinite Z-graded Jordan superalgebras of finite width can have a fractal complicated structure unlike the classification of such simple algebras over an algebraically closed field of characteristic zero [32]. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 11 Remark 3. We continue this research in [58], were in particular we discuss self-similarity of different types of superalgebras. Despite that all our superalgebras look very "self-similar", we conjecture that Q is not self-similar in terms of the definition of Bartholdi [6]. 4. Multiplication rules of Lie superalgebra Q Since {xi, ∂i i ≥ 0} are odd, the pivot elements (6) are also odd. Write them recursively: Recall that we consider the Lie superalgebra Q = Lie(v0, v1, v2) ⊂ W(Λ) ⊂ Der Λ and the associative algebra A = Alg(v0, v1, v2) ⊂ End Λ, where vi = ∂i + xixi+1vi+3, i ≥ 0. (8) v0 = ∂0 + x0x1v3, v1 = ∂1 + x1x2v4, v2 = ∂2 + x2x3v5, i ≥ 0. (9) Define a shift mapping τ : Λ → Λ, τ : W(Λ) → W(Λ) by τ (xi) = xi+1, τ (∂i) = ∂i+1, i ≥ 0. Clearly, we We shall use the following basic commutation relations without special mentioning. get endomorphisms such that τ (vi) = vi+1 for all i ≥ 0. Lemma 4.1. For all i ≥ 0 we have: i = xi+1vi+3; i = 2xi+1vi+3; i) v2 ii) [vi, vi] = 2v2 iii) [vi, vi+1] = −xivi+3; iv) [v2 i , vi+1] = −vi+3; v) [vi, vi+2] = −xixi+1xi+2vi+5. Proof. We check the first claim v2 i = (∂i + xixi+1vi+3)2 = [∂i, xixi+1vi+3] = xi+1vi+3. Now, the second claim is evident. We check claims (iii) and (iv): [vi, vi+1] = [∂i + xixi+1vi+3, ∂i+1 + xi+1xi+2vi+4] = [xixi+1vi+3, ∂i+1] = −xivi+3; [v2 i , vi+1] = [vi, [vi, vi+1]] = [vi,−xivi+3] = −vi+3. Finally, let us check claim (v): [vi, vi+2] = [∂i + xixi+1∂i+3 + xixi+1xi+3xi+4vi+6, ∂i+2 + xi+2xi+3vi+5] = [xixi+1∂i+3, xi+2xi+3vi+5] = −xixi+1xi+2vi+5. (cid:3) Lemma 4.2. General multiplication rules for the pivot elements are as follows. Let i, k ≥ 0. [vi, vi+3k] = 2(cid:18) k−1 xi+3nxi+3n+1(cid:19)xi+3k+1vi+3k+3; Yn=0 [vi, vi+3k+1] = −(cid:18) k−1 xi+3nxi+3n+1(cid:19)xi+3kvi+3k+3; Yn=0 [vi, vi+3k+2] = −(cid:18) k xi+3nxi+3n+1(cid:19)xi+3k+2vi+3k+5. Yn=0 Proof. Iterating (8), we get another presentation: vi = ∂i + xixi+1∂i+3 + . . . + xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−6xi+3k−5∂i+3k−3 + xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−3xi+3k−2vi+3k, i ≥ 0, k ≥ 1. (10) 12 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Using presentation (10) and Lemma 4.1, we obtain [vi, vi+3k] = xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−3xi+3k−2[vi+3k, vi+3k] = 2xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−3xi+3k−2 · xi+3k+1vi+3k+3; [vi, vi+3k+1] = xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−3xi+3k−2[vi+3k, vi+3k+1] = −xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−3xi+3k−2 · xi+3kvi+3k+3; [vi, vi+3k+2] = xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−3xi+3k−2[vi+3k, vi+3k+2] = −xixi+1 xi+2xi+3xi+4 xi+5 ··· xi+3k−3xi+3k−2 · xi+3kxi+3k+1xi+3k+2vi+3k+5. (cid:3) Consider Lie superalgebras Li = Lie(vi, vi+1, vi+2) for all i ≥ 0, so L0 = Q. Corollary 4.3. Let Q = Lie(v0, v1, v2). Then i) vi ∈ Q, i ≥ 0 (we get these elements using Lie bracket only in case of an arbitrary K); ii) τ i : Q → Li is an isomorphism for any i ≥ 1; iii) we get a proper chain of isomorphic subalgebras: Q = L0 % L1 % ··· % Li % Li+1 % ··· , ∞∩n=0 Li = {0}. iv) Q is infinite dimensional. Proof. We have v0, v1, v2 ∈ Q. By Lemma, [v2 0, v1] = −v3 ∈ Q. Similarly, by induction we conclude that vi ∈ Q for all i ≥ 0. Claim (ii) follows because we have an isomorphism τ : W(Λ) → W(Λ) such that τ (vi) = vi+1, i ≥ 0. The intersection of Li is trivial by a description of a basis of Q (Theorem 5.1). (cid:3) 5. Monomial basis of Lie superalgebra Q By rn denote a tail monomial: rn = xξ0 0 ··· xξn n = x∗0 ··· x∗n ∈ Λ, ξi ∈ {0, 1}; n ≥ 0, (11) where x∗i denote any power {0, 1}. If n < 0, we consider that rn = 1. Another monomials of type (11) will be denoted by r′n, rn, etc. Below, xi denote the missing variable in a product. We call rn−3vn, where n ≥ 0, a quasi-standard monomial of the first type, and rn−5xn−2vn, where n ≥ 2, a quasi-standard monomial of the second type. Among them, we exclude 24 false monomials, see below, the remaining monomials are standard monomials, we prove that they constitute a basis of Q in case char K 6= 2. Let us call n the length, vn the head, rn−3 (or rn−5) the tail, and xn−2 the neck of a (quasi)standard monomial. Theorem 5.1. Let char K 6= 2. A basis of the Lie superalgebra Q = Lie(v0, v1, v2) is given by the following standard monomials of two types (where rn are tail monomials (11)) i) monomials of the first type: {rn−3vn n ≥ 0} \ {x0x∗1v4, x0x∗1x∗2x3x∗4v7}, (i.e. in case of length 4 we exclude monomials containing x0, and in case of length 7 we exclude monomials containing both {x0, x3}). We shall refer to the excluded monomials as false monomials of the first type); ii) monomials of the second type: {x1v3, x2v4, x3v5} ∪ {rn−5xn−2vn n ≥ 6} \ {x0x∗1x∗2 x5v7, x0x∗1x∗2x3x∗4x5 x8v10}, (i.e. in case of length 7 we exclude monomials containing x0, and in case of length 10 we ex- clude monomials containing all three letters {x0, x3, x5}). We refer to the excluded monomials and {x0v2, x0x3v5} as false monomials of the second type. Proof. A) We prove that all standard monomials belong to Q. A1) We start with monomials of the first type. By Corollary 4.3, {vi i ≥ 0} ⊂ Q. Using Lemma 4.1, [v0, v1] = −x0v3 and [v1, v2] = −x1v4 belong to Q. Thus, all non-false monomials of the first type of length at most 4 belong to Q. This is the base of induction. Let n ≥ 5 and assume that the standard monomials of the first type of length less than n belong to Q. Using claim (v) of Lemma 4.1, we get [rn−6vn−3, vn−5] = rn−6[vn−5, vn−3] = −rn−6xn−5xn−4xn−3vn ∈ Q. (12) FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 13 Multiplying by vn−5 and (or) vn−4, vn−3 we can delete any subset of letters {xn−5, xn−4, xn−3} in (12) and obtain all monomials of the first type of length n. But this argument fails when rn−6vn−3 was a false monomial. We have two cases. a) Consider that rn−6vn−3 above is a false monomial of the first type of length 4, so n = 7. By setting rn−6vn−3 = x∗1v4 in (12), we get all standard monomials of the first type of degree 7 without x0. Using [r2v5, v4] = −r2 x3x4v7 and deleting x4 (if necessary), we obtain all standard monomials of the first type of degree 7 without x3. b) Let rn−6vn−3 be a false monomial of the first type of length 7, so n = 10. Using [x∗1x∗2x∗3x∗4v7, x∗0v5] = ±x∗0x∗1x∗2x∗3x∗4x5x6x7v10 ∈ Q, and deleting (if necessary) letters x5, x6, x7 we get all standard monomials of the first type of length 10. A2) Next, we deal with monomials of the second type. Using (formal) squares, we get v2 n−3 = xn−2vn ∈ Q for all n ≥ 3. In particular, we obtain all non-false standard monomials of the second type of length at most 5. Let n ≥ 6. We commute monomials of the first type with the pivot elements or their squares: [rn−6vn−3, x∗n−5vn−3] = ±rn−6x∗n−5[vn−3, vn−3] = ±2rn−6x∗n−5xn−2vn ∈ Q, n ≥ 6. As a rule, we get all required monomials of the second type. The arguments fail in case rn−6vn−3 is a false monomial (of the first type). a) The case of a false monomial of the first type of length 4. Nevertheless, using rn−6vn−3 = x∗1v4 above, we obtain [x∗1v4, x∗2v4] = ±2x∗1x∗2x5v7 ∈ Q, the required standard monomials of the second type of length 7, i.e. those without x0. b) Consider that rn−6vn−3 is a false monomial of the first type of length 7. Nevertheless, we can get the following monomials: [x∗1x∗2x∗3x∗4v7, x∗5v7] = ±2x0x∗1x∗2x∗3x∗4x∗5 x8v10 ∈ Q; [x∗0x∗1x∗2 x3x∗4v7, x∗5v7] = ±2x∗0x∗1x∗2 x3x∗4x∗5 x8v10 ∈ Q; [x∗1x∗2x∗3x∗4v7, x∗0v7] = ±2x∗0x∗1x∗2x∗3x∗4 x5 x8v10 ∈ Q. (13) Thus, we can obtain all monomials of the second type of length 10, i.e. those that contain at most two of the letters {x0, x3, x5}, as required. B) We prove that products of the standard monomials are expressed via the standard monomials. We write two standard monomials as a = rn−2vn, b = rm−2vm and assume that their lengths satisfy 0 ≤ n ≤ m. B1). Let m ≡ n ( mod 3). Using presentation (10), we have a = rn−2∂n + rn+1∂n+3 + ··· + rm−5∂m−3 + rm−2vm; [a, b] =(cid:16)rn−2∂n(rm−2) + rn+1∂n+3(rm−2) + ··· + rm−5∂m−3(rm−2)(cid:17)vm + r′′m−2[vm, vm]. (14) The last term (14) is of the second type because r′′m−2[vm, vm] = 2r′′m−2vm+1vm+3. If b was of the first type, namely, b = rm−3vm, then all terms (13) remain of the first type. Assume that b was of the second type b = rm−5xm−2vm, then all terms (13) remain to be of the second type. We need to check that (14) cannot yield a false monomial of the second type. Suppose the contrary and it is false of length 10, then m = 7. The second factor b is one of three types: b = x0x∗1x∗2x∗3x∗4v7, or b = x∗0x∗1x∗2 x3x∗4v7, or b = x0x∗1x∗2x5v7. Consider different possibilities for the first factor a. a) Let n = 7, then the first factor a is of the same three types. Their mutual product does not contain one of the letters {x0, x3, x5}. b) Let n = 4. Then the first factor in (14) comes from the last term in a = r2v4 = x∗1(∂4+x4x5v7) or a = r2v4 = x2(∂4 + x4x5v7). The product does not contain one of {x0, x3}. c) Let n = 1. Then the first factor in (14) comes from the last term in a = v1 = ∂1 + x1x2∂3 + x1x2 x3x4x5v7. Again, the product does not contain one of {x0, x3}. Now, let us check that (14) cannot be a false monomial of the second type of length 7. Otherwise, either b = x∗1v4 or b = x2v4. The first factor a is either of the same type or the last term in a = ∂1 + x1x2v4. Their products lack x0, as required. Also, the false monomial x0x3v5 cannot appear in (14) because in this case m = 2 but we have only the product [v2, v2] = x3v5. Moreover, we cannot obtain the false monomial x0v2. Similarly, one needs a special check that the action on tails (13) cannot produce false monomials. Recall that we cannot change the type, i.e. a neck remains the same. The case of a standard monomial of length 4 is trivial. Next, consider a standard monomial of length 7. Let it does not contain x0. (e.g. b = rm−2vm = x∗1x∗2x∗3x∗4v7.) We are acting by monomials of length at most 4. Observe that all standard 14 VICTOR PETROGRADSKY AND I.P. SHESTAKOV monomials of length 4 do not contain x0, thus, the action by them cannot help. The only possibility to obtain x0 is to use either x0v3 = x0(∂3 + x3x4v6) or v0 = ∂0 + x0x1∂3 + x0x1x3x4v6. Thus, we can obtain x0 at price of loosing x3 and the resulting monomial is not false. If a standard monomial of length 7 lacked x3, then the cation cannot produce x3, because we act "at most" by + ··· x3∂4 + ··· . Now, consider a standard monomial of the second type of length 10, namely b = rm−2vm = x∗0x∗1x∗2x∗3x∗4x∗5 x8v10. If it is lacking x5, then the result is lacking it as well, because for this we need to kill a senior absent letter x7, (recall that the neck x8 is untouchable). Next, assume that b does not contain x3, we can produce it only by using ··· x2x3∂5 + ··· or x3v5 = x3(∂5 + ··· ), thus loosing x5. Finally, assume that b lacks x0. The action by a monomial of length 5 (i.e. a = rn−2v5 = rn−2(∂5 + x5x6v8)) deletes x5. Recall that all standard monomials of length 4 do not contain x0 and all their terms do not as well. Consider a monomial of length 3: a = x0v3 = x0(∂3 + x3x4∂5 + x3x4x6x7v8), it can yield x0 but we loose either x3 or x5. Again, the standard monomials of lengths 1,2 do not contain x0 and all their terms do not as well. It remains to consider the monomial of length 0: v0 = ∂0 + x0x1∂3 + x0x1x3x4∂5 + x0x1x3x4x6x7v9. Again, we can get x0 but loose either x3 or x5. All these considerations also apply to the actions in the brackets of cases B2), B3) below. B2). Let m − n ≡ 1 ( mod 3). Using presentation (10), a = rn−2∂n + rn+1∂n+3 + ··· + rm−6∂m−4 + rm−3vm−1; [a, b] =(cid:16)rn−2∂n(rm−2) + rn+1∂n+3(rm−2) + ··· + rm−6∂m−4(rm−2)(cid:17)vm + r′′m−2[vm−1, vm]. The last term is r′′m−2[vm−1, vm] = −r′′m−2xm−1vm+2, which is of the first type, one again needs to check that it cannot be false. Consider length 4, then m = 2 and we have only [v1, v2] = −x1v4. Consider length 7, then m = 5 and either x∗1v4 or x2v4 is multiplied by either x∗0x∗1x∗2v5 or x3v5. The product does not contain either x0 or x3. B3). Let m − n ≡ 2 ( mod 3). Using presentation (10), a = rn−2∂n + rn+1∂n+3 + ··· + rm−7∂m−5 + rm−4vm−2; [a, b] =(cid:16)rn−2∂n(rm−2) + rn+1∂n+3(rm−2) + ··· + rm−7∂m−5(rm−2)(cid:17)vm + r′′m−2[vm−2, vm]. The last term is r′′m−2[vm−2, vm] = −r′′m−2xm−2xm−1xmvm+3, which is of the first type. We check that it cannot be false. Consider length 4, then m = 1 and there are no such products. Consider length 7, then m = 4 and we have either [v2, x∗1v4] = ±x∗1x2x3x4v7 or [v2, x2v4] = v4. (cid:3) Corollary 5.2. Let char K = p = 2. Then i) a basis of the Lie algebra Q = Lie(v0, v1, v2) is given by the standard monomials of the first type; ii) a basis of the Lie superalgebra Q = Lie(v0, v1, v2), as well as a basis of the restricted Lie (su- per)algebra Liep(v0, v1, v2), is given by (a) the standard monomials of the first type; (b) squares of the pivot elements: {xn−2vn n ≥ 3}. 6. Weight functions, N3 0-gradation, and three coordinate systems In this section we introduce different weight functions. Using theses functions we prove that our algebras 0-graded my multidegree in the generators and derive further corollaries. We introduce three coordinate are N3 systems that allow to put monomials in space and determine their positions. We start with the Lie superalgebra W(ΛI ) of special superderivations of the Grassmann algebra ΛI = Λ(xi i ∈ I) and consider a subalgebra spanned by pure Lie monomials: Wfin(ΛI ) = hxi1 ··· xim ∂j ik, j ∈ IiK ⊂ W(ΛI ). Define a weight function on the Grassmann variables and respective superderivatives related as: wt(∂i) = −wt(xi) = αi ∈ C, i ≥ 0, and extend it to pure Lie monomials as wt(xi1 ··· xim ∂j) = −αi1 − ··· − αim + αj, ik, j ∈ I. One checks that the weight function is additive, namely, wt([w1, w2]) = wt(w1) + wt(w2), where w1, w2 are pure Lie monomials. The weight function is also extended to an associative hull Alg(Wfin(ΛI )) and it is additive on associative products of its monomials. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 15 Now we return to our algebras Q = Lie(v0, v1, v2) and A = Alg(v0, v1, v2). We want to extend a weight function on the pivot elements so that all terms in (8) have the same weight. Namely, we additionally assume that the weight function satisfies the equalities: We get a recurrence relation wt(vi) = wt(∂i) = αi = −αi − αi+1 + αi+3, i ≥ 0. It has the characteristic polynomial t3 − t − 2 = 0. Using Cardano's formula, denote αi+3 = αi+1 + 2αi, i ≥ 0. θ1 = 3q1 +p26/27 ≈ 1.255, θ2 = 3q1 −p26/27 ≈ 0.265. ǫ = e2/3πi = −1 + √3i , 2 Observe that θ1θ2 = 1/3. One has three different roots: tk = ǫkθ1 + ǫ−kθ2, k = 0, 1, 2. Denote these roots as (we keep these notations for the whole of the paper): By Viet's formulas, one has λ = t0 = θ1 + θ2 ≈ 1.5214, µ = t1 = ǫθ1 + ǫ2θ2 ≈ −0.761 + 0.858i, ¯µ = t2 = ǫ2θ1 + ǫθ2 ≈ −0.761 − 0.858i. λ + µ + ¯µ = 0; λµ + λ¯µ + µ¯µ = −1; λµ¯µ = 2. Thus, µ =p2/λ ≈ 1.147. The characteristic equation also yields 2 λ = λ2 − 1, 2 µ = µ2 − 1, 2 ¯µ = ¯µ2 − 1. (15) (16) Thus, a weight function wt(∗) satisfies wt(∂n) = wt(vn) = −wt(xn), n ≥ 0. Moreover, by construction, all pure Lie monomials of the expansion of a pivot element (6) have the same weight as the pivot element. Below, a monomial is any (Lie or associative) product of the letters {xi, ∂i, vi i ≥ 0} ⊂ End Λ. Lemma 6.1. We identify weight functions with the space of solutions of recurrence equation (15), then i) A basis of the space of weight functions given by: wt(vn) = λn, n ≥ 0, swt(vn) = µn, n ≥ 0, swt(vn) = ¯µn, n ≥ 0, (weight); (superweight); (conjugate superweight). ii) We replace the superweight functions by two real functions: wt1(vn) = Re(µn) = wt2(vn) = Im(µn) = µn + ¯µn 2 µn − ¯µn 2i , n ≥ 0; , n ≥ 0. iii) We combine these functions together into two vector weight functions: Wt(vn) =(cid:0)wt(vn), swt(vn), swt(vn)(cid:1) = (λn, µn, ¯µn), n ≥ 0, WtR(vn) = (wt(vn), wt1(vn), wt2(vn)) = (λn, Re(µn), Im(µn)), (vector weight); n ≥ 0, (twisted vector weight). iv) The weight functions are well defined on monomials. They are additive on (Lie or associative) products of monomials, e.g., Wt(a · b) = Wt(a) + Wt(b), where a, b are monomials of A. v) Let w be a monomial, then WtR(w) = (wt w, Re(swt w), Im(swt w)). Proof. Let us check the last claim. Let w be a pivot element, the equality follows by definition. Now, the relation extends to all monomials by additivity. (cid:3) As a first application, we establish N3 0-gradations. 16 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Theorem 6.2. The Lie superalgebra Q = Lie(v0, v1, v2) and its associative hull A = Alg(v0, v1, v2) are N3 0-graded by multidegree in the generators {v0, v1, v2}: Q = ⊕n1,n2,n3≥0 Qn1,n2,n3, A = ⊕n1,n2,n3≥0 An1,n2,n3 . Proof. By Lemma 6.1, the generators have the following vector weights: Wt(v0) = (1, 1, 1), Wt(v1) = (λ, µ, ¯µ), Wt(v2) = (λ2, µ2, ¯µ2). For any n1, n2, n3 ≥ 0, let Qn1n2n3 ⊂ Q be the subspace spanned by all Lie products of multidegree (n1, n2, n3) in {v0, v1, v2}. By Lemma 6.1, all v ∈ Qn1n2n3 have the same vector weight: Wt(v) = n1 Wt(v0) + n2 Wt(v1) + n3 Wt(v2). Elements of Qn1,n2,n3 ⊂ W(ΛI ) are infinite linear combinations of pure Lie monomials having the same vector weight. Since Wt(v0), Wt(v1), Wt(v2) are linearly independent, different components Qn1,n2,n3 and Qn′ have different vector weights, hence their elements are expressed via different sets of pure Lie monomials. Hence, the sum of the components is direct. The N3 0-gradation follows by definition of these components. (cid:3) 2,n′ 3 1,n′ Given a nonzero homogeneous element v ∈ An1n2n3 , n1, n2, n3 ≥ 0, we define its multidegree (vector) and a (total) degree: Gr(v) = (n1, n2, n3) ∈ N3 0 ⊂ R3, deg(v) = n1 + n2 + n3. We put it in space using standard coordinates (X1, X2, X3) ∈ R3, which we also call multidegree coordi- nates. Thus, we write Gr(v) = (n1, n2, n3) = (X1, X2, X3). We also introduce complex weight coordinates (Z1, Z2, Z3) = Wt(v) ∈ C3 and real twisted (weight) coordinates (Y1, Y2, Y3) = WtR(v) ∈ R3. Using Lemma 6.1, we introduce transition matrices: B =(cid:16) WtT (v0), WtT (v1), WtT (v2)(cid:17) =  C =(cid:16) WtR T (v0), WtR T (v1), WtR T (v2)(cid:17) =  1 1 0 1 λ λ2 1 µ µ2 1 ¯µ ¯µ2  ; λ µ+¯µ 2 µ−¯µ 2i λ2 µ2+¯µ2 µ2−¯µ2 2 2i Lemma 6.3. B−1 =  λ 2/λ 3λ2−1 3λ2−1 3λ2−1 1 µ 2/µ 3µ2−1 3µ2−1 3µ2−1 1 ¯µ 2/¯µ 3¯µ2−1 3¯µ2−1 3¯µ2−1 1   (17) 1.521 1 2.313 1 −0.761 −0.157 0 0.858 −1.306  .   ≈  . Proof. Using the formula of the inverse matrix, one computes the inverse of Vandermonde's matrix (alter- natively, a direct check shows that B · (the matrix below) = I): (µ−λ)(µ−¯µ) (µ−λ)(µ−¯µ) (µ−λ)(µ−¯µ) (¯µ−λ)(¯µ−µ) (¯µ−λ)(¯µ−µ) (¯µ−λ)(¯µ−µ) (λ−µ)(λ−¯µ) (λ−µ)(λ−¯µ) (λ−µ)(λ−¯µ) −µ−¯µ −λ−µ −λ−¯µ µ¯µ λ¯µ λµ 1 1 1 . Using Viet's formulas and (16), we treat the denominators in the columns above as follows: (λ− µ)(λ− ¯µ) = λ2 − (µ + ¯µ)λ + µ¯µ = λ2 − (−λ)λ + 2/λ = 3λ2 − 1. (cid:3) B−1 =    One also has C−1 ≈  0.221 0.779 0.256 −0.256 0.168 −0.168 −0.447  . 0.298 0.484 Lemma 6.4. Let v ∈ A be a monomial with the multidegree coordinates Gr(v) = (X1, X2, X3) ∈ N3 0. Let WtT (v) = (Z1, Z2, Z3) and WtR T (v) = (Y1, Y2, Y3) be the respective weight and twisted coordinates. Then i) (Y1, Y2, Y3) = (Z1, Re Z2, Im Z2); ii) Z1 ∈ R and Z3 = ¯Z2; iii) WtT (v) = B · GrT (v); FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 17 iv) WtR T (v) = C · GrT (v). Proof. The first two claims follow from Lemma 6.1. By assumption, v is a product that involves X1 factors v0, X2 factors v1, and X3 factors v2. We check the last two claims using additivity and (17) WtT (v) = X1 WtT (v0) + X2 WtT (v1) + X3 WtT (v2) = B ·  X1 X2 X3   . Corollary 6.5. Let (X1, X2, X3) ∈ R3 be a point of space in standard coordinates. We introduce its weight coordinates (Z1, Z2, Z3) and twisted coordinates (Y1, Y2, Y3) using formulas of Lemma. Consider the axis OY1 ⊂ R3 which is determined by Y2 = Y3 = 0 in terms of the twisted coordinates. Lemma 6.6. The axis OY1 is determined by the vector (2/λ, λ, 1) in terms of the standard coordinates. Proof. Since, Z2 = Y2 + iY3, the condition Y2 = Y3 = 0 is equivalent to Z2 = ¯Z3 = 0. We take Wt(v) = (1, 0, 0) and use claim (iii) of Lemma 6.4 and Lemma 6.3. The axis OY1 is determined by the vector: GrT (v) = B−1  1 0 0  = 1 3λ2 − 1  2/λ λ 1   . Lemma 6.7. The axis OY1 does not contain the lattice points Z3 ⊂ R3 in terms of the standard coordinates (X1, X2, X3), except the origin O = (0, 0, 0). Proof. Consider a lattice point O 6= (n1, n2, n3) = A ∈ Z3 ⊂ R3. Assume that A belongs to OY1. Then (n1, n2, n3) = r(2/λ, λ, 1) for some r ∈ R. Hence λ = n2/n3 ∈ Q, a contradiction with irrationality of λ. (cid:3) Lemma 6.8. Let σ = logµ λ ≈ 3.068. The pivot elements {vn n ≥ 0} belong to a paraboloid-like surface with equation in twisted coordinates: Y1 = (Y 2 2 + Y 2 3 )σ/2. Proof. By Lemma 6.1, Wt(vn) = (Z1, Z2, Z3) = (λn, µn, ¯µn) and WtR(vn) = (Y1, Y2, Y3) = (Z1, Re Z2, Im Z2), n ≥ 0. Then 3 = Z22 = µ2n; Y 2 2 + Y 2 Y1 = Z1 = λn = λ1/2 logµ(Y 2 2 +Y 2 3 ) = (Y 2 2 + Y 2 3 )1/2 logµ λ. Lemma 6.9. The multidegree coordinates of the pivot elements Gr(vn) = (X1, X2, X3) are as follows: (cid:3) (cid:3) (cid:3) (cid:3) Proof. We use Lemma 6.4, Lemma 6.1, and Lemma 6.3: X1 X2 X3   λ λn µn 2/λ 2/µ 1   + 3λ2 − 1  3µ2 − 1   =  GrT (vn) = B−1 WtT (vn) = B−1 ¯µn   3µ2 − 1  3λ2 − 1  1   + 2/λ λ λn µn µn λn = µ 1   + ¯µn 3¯µ2 − 1  2/¯µ ¯µ 1   , n ≥ 0. 2/µ µ 1   + ¯µn 3¯µ2 − 1  2/¯µ ¯µ 1   . ¯µn, n ≥ 0. Corollary 6.10. The total degrees of the pivot elements in the generators {v0, v1, v2} are as follows deg(vn) = 4λ2 + λ + 6 4µ2 + µ + 6 λn + 26 26 4¯µ2 + ¯µ + 6 µn + 26 Proof. By definition of the degree, deg(vn) = X1 + X2 + X3, the sum of the multidegree coordinates, the latter are computed in Theorem. By (16), 2/λ + λ + 1 = λ2 + λ. A direct check in the field Q[λ] shows that The same computations are valid for the remaining roots µ, ¯µ. λ2 + λ 3λ2 − 1 = 4λ2 + λ + 6 26 . (cid:3) 18 VICTOR PETROGRADSKY AND I.P. SHESTAKOV 7. Q is finely N3 0-graded and just infinite, its generating functions The second example in [51] yields a Z3-graded Lie superalgebra with at most one-dimensional components. Similarly, the Lie superalgebra constructed in [13] is Z2-graded with at most one-dimensional components. Now, we establish a similar fact, that components of the multidegree Z3-grading of the Lie superalgebra Q are at most one-dimensional (Theorem 7.2). This implies that the Z3-grading of Q is fine (see definitions in [16]). We also prove that Q is just infinite but nor hereditary just infinite, the same properties were established for the example [13]. At the end of this section we supply computations of generating functions for Q. Figure 1 below gives a geometric illustration of the results and proofs of the paper. Lemma 7.1. Let τ : A → A be the shift endomorphism. Consider a multihomogeneous element 0 6= v ∈ A with Gr(v) = (n1, n2, n3), n1, n2, n3 ≥ 0. Then Gr(τ (v)) = (2n3, n1 + n3, n2). Proof. The relation [v2 0, v1] = −v3 implies that Gr(v3) = (2, 1, 0). By assumption, v is a linear combination of products involving n1, n2, n3 factors v0, v1, v2, respectively. Since τ is an endomorphism, τ (v) is a linear combination of products involving n1 factors τ (v0) = v1, n2 factors τ (v1) = v2, and n3 factors τ (v2) = v3. Using additivity of the multidegree function, we get Gr(τ (v)) = n1 Gr(v1) + n2 Gr(v2) + n3 Gr(v3) = n1(0, 1, 0) + n2(0, 0, 1) + n3(2, 1, 0) = (2n3, n1 + n3, n2). (cid:3) ⊕n1,n2,n3≥0 Qn1,n2,n3 by multidegree in the generators We make an observation. Let v be a quasi-standard monomial. We can present it as v = xα Theorem 7.2. Components of the N3 0-gradation Q = {v0, v1, v2} (Theorem 6.2) are at most one-dimensional. Proof. Recall that the standard monomials and the false monomials are the quasi-standard monomials. Let us list all quasi-standard monomials of length at most 4, of the first type: {v0, v1, v2, x∗0v3, x∗0x∗1v4}, and of the second type: {x0v2, x1v3, x2v4}. We shall prove a more general fact, namely, that different quasi-standard monomials have different multidegrees. 0 τ (v′), where α ∈ {0, 1}, τ the shift endomorphism, and v′ is a quasi-standard monomial of length less by one (and of the same type as a rule). There is one exception: v = v0, let us treat it now. One has the multidegree Gr(v0) = (1, 0, 0). So, the standard monomials with the same multidegree must contain the only factor v0. Thus, the only standard monomial of the same multidegree is v0. It remains to compare with multidegrees of the false monomials. (First, consider the false monomials of small length. We have Gr(x0v2) = (−1, 0, 2); using [v2 2 = x3v5 we get Gr(x0x3v5) = (−1, 0, 2). Let v be a false monomial of length n ≥ 5. Being of the same multidegree implies that it has the same weight. But by Corollary 8.2, wt(v) > λn−5 ≥ 1 = wt v0.) By way of contradiction, assume that u 6= v are quasi-standard monomials of the same multidegree, i.e. Gr(u) = Gr(v). Also, assume that in this counterexample the minimum of lengths of u, v is the minimum possible. By the observation above, u = xα 0 τ (v′), where u′, v′ are quasi-standard monomials of the same types and of lengths less by one and α, β ∈ {0, 1}. Let Gr(u′) = (n1, n2, n3) and Gr(v′) = (m1, m2, m3). Since Gr(x0) = (−1, 0, 0), using Lemma 7.1, we have 1, v2] = −v4 we get Gr(x0v4) = (−1, 2, 1) and Gr(x0x1v4) = (−1, 3, 1); since v2 0 τ (u′) and v = xβ Gr(u) = (2n3 − α, n1 + n3, n2) = (2m3 − β, m1 + m3, m2) = Gr(v). Since α, β ∈ {0, 1} we conclude that n3 = m3 and α = β, then also n1 = m1 and n2 = m2. Hence, Gr(u′) = Gr(v′). By minimality of the example, u′ = v′. Therefore, u = v, a contradiction. (cid:3) Corollary 7.3. Let u, w be standard monomials of Q such that wt u = wt w. Then u = w. Proof. Consider respective multidegrees and assume that Gr u = (n1, n2, n3) 6= Gr w = (m1, m2, m3). Then wt u = n1 + n2λ + n3λ2 = wt w = m1 + m2λ + m3λ2 and (m3 − n3)λ2 + (m2 − n2)λ + (m1 − n1) = 0, a contradiction with the fact that λ satisfies an irreducible polynomial of degree 3. Hence, Gr u = Gr w. By Theorem, u = w. (cid:3) Theorem 7.4. The Lie superalgebra Q is just infinite. Proof. Let I be a nonzero ideal of Q and 0 6= a ∈ I. By Corollary 7.3, a = ν1w1 + ··· + νmwm ∈ I, 0 6= νj ∈ K, wj are standard monomials, wt w1 < ··· < wt wm. (18) FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 19 Let us prove by induction on m that some pivot element belongs to I. We shall multiply (18) by monomials, the senior term wm will be transformed into a senior term, we shall keep its coefficient nonzero. By The- orem 7.2, the terms move to different at most one-dimensional multihomogeneous components. Hence, we get a similar decomposition (18) with the same (or smaller) number of terms m. Consider the senior term wm = xi1 ··· xik vn, i1 < ··· < ik. Then [vik , . . . , vi1 , wm] = vn is a senior term of [vik , . . . , vi1 , a]. Thus, we get a pivot element w′m′ = vn in (18). If m′ = 1, the base of induction is proved. By our arguments, we can assume that the senior term in (18) is wm = vn. Using [vn−1, vn−1, vn] = −vn+2, we can make n arbitrary big. Since we always multiply by homogeneous monomials, we either keep the following difference or it is even getting smaller in case the smallest term disappear: wt w′m′ − wt w′1 ≤ wt wm − wt w1 = C. Thus, wt w′1 ≥ wt v′m′ − C = λn − C, and the last number exceeds λn−1 = wt vn−1 for sufficiently large n. Hence, we can consider that all standard monomials in (18) are of length at least n. On the other hand, using wt wj ≤ wt wm = wt vn = λn and the lower estimates of Lemma 8.1, we can have only standard monomials of the first type of length at most n + 4 and of the second type of length at most n + 3. Take a standard monomial w = rk−2vk, where n ≤ k ≤ n + 5, of our decomposition (18) and assume that it has a factor xi where i < n. Then xivn+2 ∈ Q is a standard monomial of the first type and we get a new senior term [xivn+2, vn] = −xixnxn+1xn+2vn+5 6= 0 (Lemma 4.1) while [xivn+2, w] = ±[xivn+2, xir′k−2vk] = 0, thus reducing the number of monomials, and we apply the inductive assumption. It remains to consider a few standard monomials with restrictions on lengths above having no factors n) = 2λn, xi, i < n. We compute their weights, monomials of the second type: wt(xn+1vn+3) = wt(v2 wt(xnvn+2) = 2λn−1 and of the first type: wt(x∗nvn+3) ≥ λn(λ3 − 1) = λn(λ + 1), wt(x∗nx∗n+1vn+4) ≥ λn(λ4 − λ − 1) = λn(λ2 + λ − 1) > λn, and {vn+2, vn+1, vn}. These monomials except vn cannot appear because their weights exceed the weight of the senior term wt wm = wt vn = λn. Hence, our decomposition consists of a unique pivot element vn. Thus, we have vN ∈ Q for a large integer N . By Lemma 4.1, [v2 N−1, vN ] = −vN +2 ∈ I and b = [vN−2, vN ] = −xN−2xN−1xN vN +3 ∈ I, [vN , vN−1, vN−2, b] = −vN +3 ∈ I. By induction, we derive that vk ∈ I for k ≥ N + 2. Fix k ≥ N + 2, using claim (v) of Lemma 4.1, we get [rk−1vk+2, vk] = rk−1[vk+2, vk] = −rk−1xkxk+1xk+2vk+5 ∈ I. Multiplying by vk and (or) vk+1, vk+2 we get all standard monomials of the first type of length k + 5 ≥ N + 7. Using (formal) squares, we get v2 n−3 = xn−2vn ∈ I for all n ≥ N + 5. In case char K 6= 2 we also get [rn−6vn−3, x∗n−5vn−3] = ±rn−6x∗n−5[vn−3, vn−3] = ±2rn−6x∗n−5xn−2vn ∈ I, n ≥ N + 10. We proved that I contains all basis monomials of lengths n ≥ N + 10. Therefore, dim Q/I is bounded by a finite number of basis monomials of length at most N + 9. (cid:3) Lemma 7.5. The Lie superalgebra Q is not hereditary just infinite. Proof. Fix m ≥ 1. Let Q(m) ⊂ Q be the linear span of its basis monomials of length at least m, so v0 /∈ Q(m). By multiplication rules (see Section 4), Q(m) is an ideal of Q. Observe that the ideal Q(m) ⊂ Q has a finite codimension. In particular, Q = hv0i ⊕ Q(1) and dim Q/Q(1) = 1. Let J = x0Q(m) be the subspace of Q(m) spanned by its basis monomials involving x0. Since x0 can be deleted only by v0 that does not belong to Q(m), we see that J is an abelian ideal of Q(m). Since vi ∈ Q(m)\J for all i ≥ m, we conclude that dim Q(m)/J = ∞ and the ideal Q(m) is not just infinite. (cid:3) Let A = ⊕n,m,kAnmk be a Z3-graded algebra, one has an induced Z-gradation: A = ⊕n An, where Al = ⊕n+m+k=l Anmk. Define respective generating functions: H(A, t1, t2, t3) = Xn,m,k H(A, t) =Xn dim Anmktn 1 tm 2 tk 3; dim Antn = H(A, t, t, t). 20 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Using somewhat recursive structure of the basis of Q (Theorem 5.1), computer calculations yield the following series: 2 + t2t3 + t2 3 2t3 + t2t2 3 + t2 + t3 + t4 H(Q, t1, t2, t3) = t1 + t2 + t3 + t2 1t2 + t2 1t3 + t1t2 1t2 1t2 + t2 1t2 + t3 1t2 + t2 1t3 3 + t1t3 1t2 2 + t4 + t2 1t2t3 2 + t2 2 + t3 2t3 + t1t2 1t3 3 + t1t4 1 + t1t2 + t1t3 + t2 2 + t1t2t3 + t1t2 3 + t2 2 + t1t2 3 + t1t3 1t2 1t3 2 + t2 1t2 2t3 + t2 3 + t1t2t3 3 + t1t4 1t2 2t3 + t3 1t2t2 2t2 3 + t1t2 1t2t3 + t2 1t2t3 + t2 2t2 2 + t3 2t3 + t1t3 1t2t3 + t3 3 + t2 1t4 + t4 3 + t1t3 2t3 + t1t2t2 1t2t2 3 3 + t4 2t3 + t3 2t2 3 + t2 3 + t2 1t3 2t3 + t2 1t2 2t2 3 2t3 3 + t1t2t4 3 + t4 2t2 2t3 3 3 + t3 2t3 + t2 2t2 3 + t2t3 3 H(Q, t) = 3t + 6t2 + 7t3 + 11t4 + 14t5 + 15t6 + 17t7 + 18t8 + 21t9 + 25t10 + 25t11 + 26t12 + 30t13 + 32t14 + 33t15 + 35t16 + 35t17 + 35t18 + 38t19 + 39t20 + 38t21 + 38t22 + 39t23 + 43t24 + 44t25 + 42t26 + 47t27 + 51t28 + 50t29 + 53t30 + ··· 3 + t3 2t3 3 + ··· Figure 1. Three small read vectors at origin are generators v0, v1, v3. Dots show standard monomials of Q (first type – green, second – blue). Pivot elements are red, marked by red dashed arrows, and belong to small "paraboloid". Two "paraboloids" are cut by plane of fixed weight: X3 16 14 12 10 9 8 7 6 5 4 3 2 1 O Y1 =wt= X1 + λX2 + λ2X3 ≤ λ9 ≈ 43.6 Y1 X2 2 + Y 2 3 < 3Y 0.326 1 pY 2 1 2 3 4 5 6 7 8 9 10 Y3 Y2 12 14 16 18 20 22 24 26 X1 8. Bounds on weights, growth, and paraboloid for Lie superalgebra Q In this section, we establish estimates on weights and superweights of standard monomials of the Lie superalgebra Q. Using these estimates we specify the growth of Q (Theorem 8.4) and prove that the standard monomials are situated in a region of space restricted by a surface of rotation close to a cubic paraboloid (Theorem 8.5, see Fig. 1). Below, λ, µ are the roots of the characteristic polynomial (Section 6). FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 21 Lemma 8.1. We have estimates for weights of the quasi-standard monomials of the first and second type: 1.3λn−5 < wt(rn−3vn) ≤ λn, n ≥ 0; 1.1λn−4 < wt(rn−5xn−2vn) ≤ 2λn−3, n ≥ 2. Proof. One checks that (λ − 1)−1 = (λ2 + λ)/2. The upper bound wt(rn−3vn) ≤ λn, n ≥ 0, is trivial. First, consider a tail rm = xξ0 m , ξi ∈ {0, 1}, and find a bound on its weight: 0 ··· xξm wt(rm) ≥ −(λ0 + λ1 + ··· + λm) > − λm+1 λ − 1 = − λm+1(λ2 + λ) λm+3 + λm+2 2 = − 2 , m ≥ 0. (19) This bound is formally valid for m = −1,−2,−3. Using (19), we get: λn + λn−1 λn−1(λ − 1) n ≥ 0, because λ4(λ − 1)/2 ≈ 1.39. For monomials of the second type, one has an upper bound wt(rn−3vn) > − 2 2 wt(rn−5xn−2vn) ≤ λn − λn−2 = λn−3(λ3 − λ) = λn−3(λ + 2 − λ) = 2λn−3, n ≥ 2. Using (19) and λ(3 − λ)/2 ≈ 1.12, we check the lower bound for monomials of the second type: > 1.3λn−5, + λn = wt(rn−5xn−2vn) > − 2 − λn−2 + λn = λn−3(−3/2λ − 1/2 + λ3) λn−2 + λn−3 = λn−3(−3/2λ − 1/2 + λ + 2) = λn−3 3 − λ 2 > 1.1λn−4, n ≥ 2. Corollary 8.2. Let w be a quasi-standard monomial of length n ≥ 0. Then λn−5 < wt w ≤ λn. Lemma 8.3. Let w be a quasi-standard monomial of length n ≥ 0. Then swt w < 7µn. Proof. Write monomials of both types as w = rn−2vn, n ≥ 0. Below, we use that µ ≈ 1.14656: swt w = swt(rn−2vn) ≤ µn + n−2 Xi=0 µi < µn + µn−2 1 − 1/µ = µn(cid:18)1 + 1 µ2 − µ(cid:19) < 7µn. Theorem 8.4. Consider the Lie superalgebra Q = Lie(v0, v1, v2) over an arbitrary field K. Then GKdim Q = GKdim Q = logλ 2 ≈ 1.6518. Proof. Let us find an upper bound on the weight growth function γQ(m) which counts standard monomials w such that wt w ≤ m, where m ≥ 1. Consider such a monomial w of length n. By Corollary 8.2, λn−5 < wt w ≤ m, hence n ≤ n0 = [logλ m] + 5. Counting standard monomials of both types of length at most n0, we get a desired upper bound n0 n0 γQ(m) ≤ 3 + 2n−2 + 2n−4 < 3 + 2n0−1 + 2n0−3 < 2n0 ≤ 2logλ m+5 = 32mlogλ2. Xn=2 Xn=4 Fix m ≥ 1. Set n = [logλ m]. Consider all monomials w = rn−3vn of the first type of length n. By Corol- lary 8.2, wt w ≤ λn ≤ m. Counting all such monomials we get a lower bound in case of any characteristic: γQ(m) ≥ 2n−2 ≥ 2logλ m−3 = 2−3mlogλ 2. Theorem 8.5. Put σ = logµ λ ≈ 3.068. The points of space depicting the (quasi)standard monomials of the Lie superalgebra Q are inside an "almost cubic paraboloid", which equation is written in terms of the twisted coordinates WtR(w) = (Y1, Y2, Y3): Proof. Let w be a standard monomial of Q of length n ≥ 0 and the weight coordinates Wt(w) = (Z1, Z2, Z3) = (wt w, swt w, swt w). By Corollary 8.2, λn−5 < wt w = Z1, thus n < logλ Z1 + 5. By Lemma 8.3, we get 2 + Y 2 qY 2 3 < 14 σpY1. Z2 = swt w < 7µn < 7µlogλ Z1+5 = 7µ5Z logλ µ 1 < 14Z 1/σ 1 , using 7µ5 ≈ 13.89 < 14. Applying Lemma 6.4, we get a transition to the twisted coordinates Y 2 2 + Y 2 3 = (Re Z2)2 + (Im Z2)2 = Z22 < 196Z 2/σ 1 = 196Y 2/σ 1 . (cid:3) (cid:3) (cid:3) (cid:3) 22 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Figure 1 shows a paraboloid but with a smaller constant 3. We have a weaker bound. Corollary 8.6. The monomials of Q are inside of a cubic paraboloid: Proof. We use that σ > 3 and Y1 = Z1 = wt w ≥ 1 for any standard monomial w. Corollary 8.7. Consider the "almost cubic paraboloid" of Theorem. 2 + Y 2 qY 2 3 < 14 3pY1. (cid:3) i) The volume of a part of the paraboloid cut by plane Y1 ≤ m is equal to Volume(m) = Const · mlogλ 2, m ≥ 1. ii) Asymptotically, a nonzero share of lattice points inside the paraboloid corresponds to monomials of Q. Proof. We have a figure of rotation: 0 ≤ Y1 ≤ m, pY 2 Volume(Y1,Y2,Y3)(m) =Z m 0 πR2(y)dy =Z m 0 2 + Y 2 3 ≤ R(Y1) = 14Y 1/σ 1 with a volume: π196y2/σ = 196π 1 + 2/σ m1+2/σ, where 1 + 2/σ = 1 + 2 logλ µ = logλ(λµ2) = logλ 2, by Viet's formulas. Compute volume of the same figure in terms of the standard coordinates: Volume(X1,X2,X3)(m) = Volume(Y1,Y2,Y3)(m)/ det C, because C makes a transition between these coordinates. A number of lattice points Z3 (in the standard coordinates) inside the figure is asymptotically equal to Volume(X1,X2,X3)(m). On the other hand, by the proof of Theorem 8.4 on the growth of Q, we have a lower polynomial bound with the same degree. Recall that the multihomogeneous components of Q are at most one-dimensional (Theorem 7.2). Thus, a nonzero share of the lattice points inside the paraboloid correspond to monomials of Q. (cid:3) 9. Poisson superalgebra P In this section, we define a Poisson superalgebra P(V0, V1, V2), determined (actually, generated) by the Lie superalgebra Q = Lie(v0, v1, v2). Recall our basic construction. We take the Grassmann algebra Λ = Λ(xii ≥ 0) and consider its generators and respective superderivatives {xi, ∂i i ≥ 0} ⊂ End¯1 Λ. They satisfy the commutation relations: [∂i, xj] = δij, [xi, xj] = [∂i, ∂j] = 0, i, j ≥ 0. (20) Next, we defined the pivot elements: vi = ∂i + xixi+1(∂i+3 + xi+3xi+4(∂i+6 + xi+6xi+7(∂i+9 + ··· ))) ∈ Der Λ, (21) Now, consider the Grassmann superalgebra H∞ = Λ(xi, yii ≥ 0) which is turned into a Poisson superal- i ≥ 0. H∞ =(cid:26)X¯α< ¯β λα,βxαyβ(cid:12)(cid:12)(cid:12)(cid:12) λα,β ∈ K(cid:27). gebra by a bracket determined by relations: {yi, xj} = δi,j, We obtain the bracket: {f, g} = (−1)f−1 {xi, xj} = {yi, yj} = 0, i, j ≥ 0. (22) ∞ Xi=1(cid:18) ∂f ∂xi ∂g ∂yi + ∂f ∂yi ∂g ∂xi(cid:19), f, g ∈ H∞. Next, we define a completion of H∞. Denote by Ξ the set of all tuples α = (αiαi ∈ {0, 1}, i ≥ 0) with finitely many nonzero entrees. Denote by ǫi ∈ Ξ the tuple with unique 1 on the ithe place, i ≥ 0. Let α ∈ Ξ, then put α =Pi≥0 αi, ¯α = max{i ≥ 0 αi 6= 0}, and xα =Qi≥0 xαi i ∈ H∞, products being taken in increasing order. Let α ∈ Ξ and for some i ≥ 0 we have αi = 0, then we consider that xα−ǫi = 0. Below we assume that all degree tuples α, β belong to Ξ. Consider the following completion of H∞ that consists of all infinite formal sums: i ∈ H∞, yα =Qi≥0 yαi FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 23 Since below yis will be substituted by derivatives, we define differential operators of finite order k: H k ∞ =(cid:26) X¯α< ¯β, β=k H = ∞⊕k=0 H k ∞. λα,β ∈ K(cid:27), k ≥ 0; λα,βxαyβ(cid:12)(cid:12)(cid:12)(cid:12) Lemma 9.1. We formally extend the products of H∞ onto H∞. Then i) H∞ is a Poisson superalgebra; ii) H ⊂ H∞ is its subalgebra. Proof. Clearly, the associative product is well defined. We check that the Poisson bracket is also well defined: λα′β′xα′ (cid:26) X¯α′< ¯β′ yβ′ = X¯α′< ¯β′ ¯α′′< ¯β′′ µα′′β′′xα′′ yβ′′(cid:27) , X¯α′′< ¯β′′ λα′β′µα′′β′′ Xi≤ ¯α′< ¯β′ ±xα′−ǫiyβ′ i≤ ¯β′′ xα′′ yβ′′−ǫi + Xi≤ ¯α′′< ¯β′′ i≤ ¯β′ ±xα′ yβ′−ǫixα′′−ǫiyβ′′! =Xα,β Xα′+α′′=α ±λα′,β′+ǫiµα′′+ǫi,β′′!xαyβ, where the signs ± are uniquely determined. While deleting yi above, we have either i < ¯β′ or i < ¯β′′, the latter yield a factor yj with i < j, inherited by yβ. Hence, ¯α < ¯β and the product belongs to H∞. ∞ and {f, g} ∈ H k+m−1 . Thus, H (cid:3) ∞, k, m ≥ 0. By computations above, f · g ∈ H k+m ±λα′+ǫi,β′µα′′,β′′+ǫi + Xi< ¯β′′ Let f ∈ H k is a subalgebra. Xi< ¯β′ ∞, g ∈ H m β′+β′′=β ∞ The next elements will be referred to as the pivot elements as well: Vi = yi + xixi+1(yi+3 + xi+3xi+4(yi+6 + xi+6xi+7(yi+9 + ··· ))) ∈ H 1 (23) Let π : Λ → H∞ be the natural embedding. Namely, consider a monomial xα ∈ Λ, α ∈ Ξ. Then π maps xα ∈ Λ on the same xα ∈ H∞. So, we identify a tail rm ∈ Λ with the respective element rm ∈ H∞. Also, we define the mapping on pure derivatives π(xα∂i) = xαyi ∈ H 1 ∞ ⊂ H, for all i ≥ 0, α ∈ Ξ, ¯α < i. We extend the mapping onto infinite sums. In particular, we get π(vi) = Vi for all i ≥ 0. We have images of the standard monomials: ∞ ⊂ H, i ≥ 0. π(rn−3vn) = rn−3Vn, n ≥ 0; π(rn−5xn−2vn) = rn−5xn−2Vn, n ≥ 3. Lemma 9.2. The mapping π : Q = Lie(v0, v1, v2) → Lie(V0, V1, V2) ⊂ H 1 onto a Lie subsuperalgebra of the Poisson superalgebra H. ∞ ⊂ H is an isomorphic embedding Proof. Observe that the Lie brackets (20) and (22) are "the same". We conclude that the Lie brackets on the pivot elements (21) and their images (23) are "the same", thus π([vi, vj]) = {Vi, Vj} for all i, j ≥ 0. The same observation applies to the standard monomials and their images. (cid:3) relation (8) is rewritten as: Now we define a Poisson subalgebra P = Poisson(V0, V1, V2) ⊂ H generated by {V0, V1, V2}. Recursive (24) i ≥ 0} ⊂ H freely generate a Lemma 9.3. Using the associative product only, the elements {xi, Vi Grassmann algebra in the same variables. Vi = yi + xixi+1Vi+3, i ≥ 0. Proof. Observe that both terms in (24) are odd, they anticommute, and their squares are equal to zero. Thus, we get V 2 (cid:3) i = 0, i ≥ 0. Lemma 9.4. Let char K = 2. Then P is Poisson superalgebra, namely, it has a formal square on the odd part and satisfy the additional axioms for char K = 2. 24 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Proof. Let us discuss a formal square that should be defined on the odd part of P. First, define a formal square on the odd part of H∞. Since (ad xi)2 = (ad yi)2 = 0, we put x[2] i = 0, i ≥ 0. By the additional axiom (Subsection 2.4), w[2] = 0, where w is any monomial in {xi, yii ≥ 0} of odd length, on the other hand, one checks that (ad w)2 = 0. Similar to the restricted Lie algebras [29], this leads to a formal square on the whole of the odd components of H∞ and H∞. One checks that it satisfies the additional axiom, as was remarked above, it is sufficient to verify it on a basis consisting of words in {xi, yii ≥ 0}. Next, we restrict the formal square to P and see that it coincides with the regular square on Q. Finally, by the additional axiom, a formal square on the whole of the odd part of P does not lead to new monomials, i.e. P is spanned by products of the basis of Q. (cid:3) i = y[2] Define Poisson superalgebras Pi = Poisson(Vi, Vi+1, Vi+2) ⊂ H, i ≥ 0, so P0 = P. We extend the shift endomorphism τ : Q → Q onto P by τ (1) = 1 and τ (w1 ··· wm) = τ (w1)··· τ (wm), where wj ∈ Q. Corollary 9.5. Let P = Poisson(V0, V1, V2). Then i) Vi ∈ P, i ≥ 0; ii) τ i : P → Pi is an isomorphism for any i ≥ 0; iii) we get a proper chain of isomorphic Poisson superalgebras: P = P0 % P1 % ··· % Pi % Pi+1 % ··· , ∞∩n=0 Pi = h1iK. iv) P is infinite dimensional. 10. Bases of Poisson superalgebra P and associative hull A gr A ∼= P, in particular, both algebras have "the same" bases. In this section, we find bases for P and A. In case char K 6= 2, we prove that for a filtration of A one has For a series of previous examples of (self-similar) (restricted) Lie (super)algebras, bases for respective asso- ciative hulls were not found [54, 57, 56, 51, 52]. Instead, we considered bigger (restricted) Lie (super)algebras R ⊃ R whose bases were given by quasi-standard monomials and we determined and used bases of their associative hulls A = Alg( R) ⊃ A. The virtue of the example of a Lie superalgebra of linear growth [13] is that for the first time, we were able to describe explicitly a basis of the associative hull. Now, we are also able to describe bases of A and P. Consider a filtration {Am m ≥ 0} of A, where Am is spanned by all at most m-fold products of standard monomials of the Lie superalgebra Q, m ≥ 0. Define the associated graded algebra gr A = ∞⊕m=0 Am, where Am = Am/Am−1, m ≥ 0, A−1 = {0}. Similarly, let Pm ⊂ P denote the linear span of all m-fold products of the standard monomials of Q, where m ≥ 0. We get a direct sum P = ∞⊕m=0 Pm, which is not a grading of a Poisson superalgebra because one has {Pn, Pm} ⊂ Pn+m−1, n, m ≥ 1. Theorem 10.1. Let char K 6= 2. A basis of the Poisson superalgebra P = Poisson(V0, V1, V2) is given by the unit and the following monomials: 0 xα1 xα0 1 ··· xαn−2 n−2 V β0 0 V β1 1 ··· V βn−1 n−1 Vn, αi, βi ∈ {0, 1}, n ≥ 0, (25) (n will be referred to as the length) where αis satisfy the following restrictions: i) let βn−1 = βn−2 = 1, then α0, . . . , αn−2 take all combinations; ii) let βn−1 = 1, βn−2 = 0, then at least one of {αn−4, αn−3, αn−2} is zero; iii) let βn−1 = 0, βn−2 = 1, then at least one of {αn−3, αn−2} is zero; iv) let βn−1 = βn−2 = 0, then either αn−2 = 0 or αn−3 = αn−4 = 0; v) let βn−1 = ··· = β0 = 0 then we have the standard monomials of Theorem 5.1; vi) we exclude finitely many monomials (of degree at most 10) that are products involving series of standard monomials related with false monomials, see an algorithm below. Proof. Using the basis of the free Poisson superalgebra [64], we conclude that P is spanned by all products of the standard monomials of Q (Theorem 5.1). Now, we consider all possible at most 3-fold products of the standard monomials, the first monomial being of lengths n, and two optional monomials being of lengths FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 25 n − 1 and n − 2. There are technical considerations because the monomials are of two types, we omit this arguments. One obtains restrictions (i–iv). If a product involves only one standard monomial, we get (v). We need to exclude products that involve false monomials. A series of standard monomials is the set of the standard monomials with a head Vn (i.e. the length n) and a neck xαn−2 n−2 fixed (so, the type is also fixed) while the tail takes all allowed values so that we do not get a false monomial. We have the series of the standard monomials related to false monomials: {x0x∗1V4}, {x0x∗1x∗2 x5V7}, x0 x3V5, (26) x0V2, {x0x∗1x∗2 x3x∗4V7}, {x0x∗1x∗2 x3x∗4 x5 x8V10}, where denotes that the series cannot contain all the letters with this sign, ∗ denotes that all powers are possible. Above, the first line contains all the series, that are simply described as not containing x0. There are some more series, actually consisting of one element, of the standard monomials not containing x0: V0, V1, x1V3, x2V4. (27) We consider a basis of P as obtained by products of different series of the standard monomials. The series of the standard monomials except (26) and (27) have arbitrary powers of x0. Observe that, multiplying by them remove all restrictions of (26). Thus, restrictions arise for products of the series, that include at least one (26) and optionally some (27). Of course, we take only products without squares of any letters. One obtains finitely many families of monomials (25) with restrictions on powers of the xis. This leads to a finite list of monomials excluded from (25). (cid:3) Remark 4. Consider char K = 2. A basis of Q consists of the standard monomials of the first type and squares of the pivot elements (Corollary 5.2), the latter give a specific influence on a basis of P. Recall that by the additional axiom (Subsection 2.4), a formal square does not lead to new monomials, i.e. P is spanned by products of the basis of Q. For our purposes, we give only the following rough description of a basis of P. Corollary 10.2. Let char K = 2, and P = Poisson(V0, V1, V2) ⊂ H. Then i) P is contained in a span of monomials (25); ii) monomials (25) with n ≥ 8, αn−1 = αn−2 = 0, and arbitrary α0, . . . , αn−3, β0, . . . , βn−1 ∈ {0, 1} are linearly independent and belong to P. Proof. We take the standard monomials of the first type xα0 by arbitrary powers of V0, . . . , Vn−1. Theorem 10.3. Let char K 6= 2, consider the associative hull A = Alg(v0, v1, v2) ⊂ End(Λ). Then n−3 Vn ∈ Q, where n ≥ 8, and multiply 0 ··· xαn−3 (cid:3) i) a basis of A consists of the unit and the replica of monomials (25): xα0 0 xα1 1 ··· xαn−2 n−2 vβ0 0 vβ1 1 ··· vβn−1 n−1 vn, αi, βi ∈ {0, 1}, n ≥ 0, (28) that obey to all restrictions of Theorem 10.1 (n will be referred to as the length); decreasing lengths, where m ≥ 1; ii) Am modulo Am−1 is spanned by products w1 ··· wm of standard monomials wi ∈ Q of strictly iii) one has a natural isomorphisms of vector spaces Am ∼= Pm, m ≥ 0; iv) gr A has a natural structure of a Poisson superalgebra and gr A ∼= P. Proof. Let us prove (ii) by induction on m. The cases m = 0, 1 are clear. Let m ≥ 2. Fix a total order ≺ on the standard monomials that obeys to their lengths. Consider a product w1w2 ··· wm ∈ Am, where wi are standard monomials. Since the commutator of two different monomials [wi, wi+1] ∈ Q is expressed via standard monomials, we can superpermute these monomials modulo Am−1. Thus, we assume that w1 (cid:23) w2 (cid:23) ··· (cid:23) wm. Suppose that we obtain two elements of the same length n, we treat such a product: wiwi+1 = rn−1vn · r′n−1vn = ±rn−1r′n−1v2 rn−1r′n−1[vn, vn] = 1 2 [rn−1vn, r′n−1vn] = 1 2 1 n = ± 2 [wi, wi+1] ∈ Q. Thus, products containing such pairs belong to Am−1 and we apply the inductive assumption. As a result, we get products of standard monomials with strictly decreasing lengths, (ii) is proved. 26 VICTOR PETROGRADSKY AND I.P. SHESTAKOV By (ii), Am modulo Am−1 is spanned by m-fold products of the standard monomials as follows: rn1−1vn1 · rn2−1vn2 ··· rnm−1vnm, n = n1 > n2 > ··· > nm ≥ 0, m ≥ 1. (29) Now, we move all Grassmann letters in (29) to the left. We proceed as follows. Let xi be a Grassmann variable in a standard monomial rnj−1vnj , j ≥ 2, then i < nj. The standard monomials before it in (29) have lengths greater than nj, thus, greater than i. By (7), xi supercommutes with the preceding heads {vnk 1 ≤ k < j}, and while moving all Grassmann letters to the left we obtain no additional terms. Since the associative algebra P is supercommutative, Pm is spanned by ordered m-fold products of standard monomials the same as (29) (one only needs to replace vis by Vis). Both products are reordered (both yield zeros provided that a Grassmann letter appears twice) to obtain respective bases in the same way, one of them being given by the list (25) under the specified restrictions. We get isomorphisms of vector spaces ρm : Am = Am/Am−1 ∼= Pm for all m ≥ 0. We get an isomorphism ρ : gr A ∼= P, a check shows that this is an isomorphism of associative superalgebras. Applying ρ−1 to monomials (25), we get Claim (i). Since gr A is supercommutative, we supply it with a bracket as follows. Let a = w1 ··· wn ∈ An\An−1 and b = w′1 ··· w′m ∈ Am\Am−1, where wis, w′js are standard monomials of Q, n, m ≥ 1. Observe that the order in such products influences the sign only. Denote by ¯a, ¯b the respective images in An = An/An−1 and Am = Am/Am−1. Put {¯a, ¯b} = [a, b] (mod An+m−2) =Xp,q ±(cid:18)Yi6=p wiYj6=q w′j(cid:19)[wi, w′j] ∈ An+m−1(mod An+m−2). This bracket satisfies the Leibnitz rule because it came from a supercommutator of an associative algebra that satisfies the Leibnitz rule. We get an isomorphism of Poisson superalgebras gr A ∼= P because the brackets coincide on Q that generate both algebras as associative algebras, thus yielding (iv). (cid:3) Corollary 10.4. Let char K = 2. i) The associative algebra A = Alg(v0, v1, v2) has a basis the same as in other chractristics (28). ii) We have a proper inclusion of Poisson superalgebras P $ gr A. Proof. Let us show that all standard monomials of the second type belong to A by repeating the arguments of the proof of Theorem 5.1. Recall that xn−2vn ∈ Q ⊂ A for all n ≥ 3, thus yielding all standard monomials of the second type of length at most 5. Let n ≥ 6, then rn−6vn−3 · x∗n−5vn−3 = rn−6x∗n−5v2 n−3 = rn−6x∗n−5xn−2vn ∈ A, n ≥ 6. We obtain all monomials of the second type except the cases when rn−6vn−3 is false (of the first type). a) The case of a false monomial of the first type of length 4, we get the required standard monomials of the second type of length 7 by x∗1v4 · x∗2v4 = x∗1x∗2x5v7 ∈ A. b) Consider that rn−6vn−3 is a false monomial of the first type of length 7. We get all monomials of the second type of length 10, i.e. those that contain at most two of the letters {x0, x3, x5} by: x∗1x∗2x∗3x∗4v7 · x∗5v7 = x0x∗1x∗2x∗3x∗4x∗5 x8v10 ∈ A; x∗0x∗1x∗2 x3x∗4v7 · x∗5v7 = x∗0x∗1x∗2 x3x∗4x∗5 x8v10 ∈ A; x∗1x∗2x∗3x∗4v7 · x∗0v7 = x∗0x∗1x∗2x∗3x∗4 x5 x8v10 ∈ A. Now, the arguments on products of standard monomials of both types (29) above yield the same basis of A as that in case char K 6= 2. To prove the second claim recall that respective products of P similar to (29) contain only standard monomials of the first type and squares of the pivot elements. As a result, in the case m = 1 we get P1 $ A1 = A1/A0. (cid:3) Lemma 10.5. Define Ai = Alg(vi, vi+1, vi+2) ⊂ End Λ for i ≥ 0. Then i) τ i : A → Ai is an isomorphism for any i ≥ 0; ii) we get a proper chain of isomorphic associative superalgebras: A = A0 % A1 % ··· % Ai % Ai+1 % ··· , ∞∩n=0 Ai = {0}. Theorem 10.6. Let char K 6= 2. The Poisson superalgebra P admits an algebraic quantization. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 27 Proof. Consider a polynomial extension Λ(t) = K[t] ⊗K Λ(xii ≥ 0), where t commutes with the Grassmann variables. As above ∂i denote the superderivative ∂i(xj ) = δi,j, i, j ≥ 0. Let x(t) i be the operator of the left multiplication by txi on Λ(t), i ≥ 0. These operators anticommute except for nontrivial relations: [∂i, x(t) j ](t) = ∂ix(t) j + x(t) (30) Below we omit the indices x(t) i = xi, i ≥ 0. Let elements of the Lie superalgebra Q act on Λ(t) using relations above. Their respective Lie products are sums of commutators of pure Lie monomials, the latter involving one commutator of type (30). Thus, Q(t) = K[t] ⊗K Q is supplied with a deformed Lie superbracket: j ∂i = tδij, i ≥ 0. (x(t) i )2 = 0, ∂2 i = 0, [f (t) ⊗ a, g(t) ⊗ b](t) = t · f (t)g(t) ⊗ [a, b], f (t), g(t) ∈ K[t], a, b ∈ Q. The actions of Q(t) generate an associative superalgebra A(t) = Alg(Q(t)) ⊂ End Λ(t). One checks that A(t) = K[t] ⊗K A, where elements of A commute using the deformed superbracket. superbracket is uniquely determined by relations: ∞ = K[t] ⊗K Λ(xi, yii ≥ 0) with the deformed Similarly, we define the deformed Poisson superalgebra H (t) i, j ≥ 0. {yi, xj}(t) = tδi,j, {xi, xj}(t) = {yi, yj}(t) = 0, (31) We continue our considerations above and construct the deformed Poisson superalgebra P(t) = K[t] ⊗K P, the bracket { , }(t) obeying to (31). Let {Amm ≥ 0} be the filtration discussed in Theorem 10.3. By its arguments {K[t] ⊗K Amm ≥ 0} is a filtration of A(t). By construction, A(t) and P(t) are free left K[t]-modules with "the same" bases (28). Repeating arguments of Theorem 10.3 we get an isomorphism of associative superalgebras gr A(t) ∼= Pt. We identify the vector spaces A(t) = P(t), this will be our algebraic quantization. Let ∗ be the asso- ciative product of A(t) and · the associative product of P(t). Consider a = w1 ··· wn ∈ An\An−1 and b = w′1 ··· w′m ∈ Am\Am−1, where wis, w′j s are standard monomials of Q, n, m ≥ 1. Denote respective images ¯a ∈ An/An−1 ∼= Pn, and ¯b ∈ Am/Am−1 ∼= Pm. Permuting two basis elements yields a factor [wi, wj](t) = t[wi, wj ] ∈ tA(t)t, we simply write O(t). We have a ∗ b = ¯a · ¯b (mod t). Similarly, products of A(t) that involve either two commutators e.g. wi, w′j yield a factor t2. Thus, such products belong to t2A(t), we simply write O(t2). We have [wi, wj](t), or a triple commutator in a ∗ b − (−1)abb ∗ a = [a, b](t) =Xp,q ±(cid:18)Yi6=p wiYj6=q w′j(cid:19)[wi, w′j] (mod O(t2)) = tXp,q = t[a, b] (mod O(t2)) = t{¯a, ¯b} (mod O(t2)). ±(cid:18)Yi6=p wiYj6=q w′j(cid:19)[wi, w′j](t) (mod O(t2)) (cid:3) 11. Weights, growth, and paraboloid for superalgebras P and A In this section, we establish bounds on weights of algebras P and A, prove that both algebras have a polynomial growth, and determine positions of their multihomogeneous N3 0-components in space. In Section 6 we defined different weight functions on the Lie superalgebra Q. Since theses functions are determined by the weights of the letters {xi, yi i ≥ 0}, these functions are extended onto P by additivity. A Poisson monomial is a product in the letters {xi, Vi i ≥ 0}, they are either even or odd with respect to Z2 superalgebra grading. The next result is proved as Theorem 6.2. Lemma 11.1. The Poisson superalgebra P = Poisson(V0, V1, V2) is N3 ators {V0, V1, V2}: 0-graded by multidegree in the gener- P = ⊕n1,n2,n3≥0 Pn1,n2,n3. Below, λ, µ are the roots of the characteristic polynomial (Section 6). Since P and A have the same bases (they differ only in case char K = 2), the proofs below are given only in case of P. Lemma 11.2. Let w be a monomial (25) of P (or a monomial (28) of A) of length n, n ≥ 0. Then λn−5 < wt w < 2λn+1, swt w < 8µn, n ≥ 0. 28 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Proof. Recall that w arises from a product of standard monomials, one of them of length n, each monomial being of positive weight. Thus, the lower bound on the weight function follows from the lower bound of Corollary 8.2. We compute the upper bound, using that (λ − 1)−1 ≈ 1.92 < 2. wt(xα0 n 0 n−2 V β0 ··· V βn−1 0 ··· xαn−2 λn+1 λ − 1 i ) = µi(βi − αi) ∈ {0,±µi} for all i ≥ 0. Then n−1 Vn) ≤ Xi=0 λi < < 2λn+1, n ≥ 0. Observe that swt(xαi i V βi swt(xα0 0 ··· xαn−2 n−2 V β0 0 ··· V βn−1 n−1 Vn) ≤ n Xi=0 µi < µn 1 − 1/µ < 8µn, where we used that (1 − 1/µ)−1 ≈ 7.8 < 8. Theorem 11.3. Consider the Poisson superalgebra P and associative hull A over an arbitrary field. Then (cid:3) GKdim P = GKdim P = GKdim A = GKdim A = 2 logλ 2 ≈ 3.3036. Proof. Let us find an upper bound on the weight growth function γP(m) which counts basis monomials w such that wt w ≤ m, where m ≥ 1. Consider such a monomial w of length n. By Lemma 11.2, λn−5 < wt w ≤ m, hence n ≤ n0 = [logλ m] + 5. We get an upper bound by counting the number of all monomials (25) of length at most n0 γP(m) ≤ 1 + n0 Xn=1 22n−1 < 1 + 2 3 4n0 < 4n0 ≤ 4logλ m+5 = 210m2logλ2. Fix m and set n = [logλ(m/2)] − 1, we may assume that n ≥ 8. By Corollary 10.2, monomials (25) of length n with αn−1 = αn−2 = 0 belong to a basis of P in case of any characteristic. By Lemma 11.2, wt w < 2λn+1 ≤ m. Our monomials w contain 2n − 2 arbitrary powers and their number yields a lower bound: γP(m) ≥ 22n−2 ≥ 22 logλ(m/2)−6 = 2−6−2 logλ 2m2 logλ 2. (cid:3) Theorem 11.4. Put σ = logµ λ ≈ 3.068. The lattice points of space corresponding to basis monomials of the Poisson superalgebra P (or the associative hull A) in terms of the standard (i.e. multidegree) coordinates are inside an "almost cubic paraboloid", given by an equation in terms of the twisted coordinates WtR(w) = (Y1, Y2, Y3): Proof. Let w be a monomial (25) of P of length n ≥ 0 with the weight coordinates Wt(w) = (Z1, Z2, Z3) = (wt w, swt w, swt w). By Lemma 11.2, λn−5 < wt w = Z1, thus n < logλ Z1 + 5. The second inequality of Lemma 11.2 yields 2 + Y 2 qY 2 3 < 16 σpY1. Z2 = swt w < 8µn < 8µlogλ Z1+5 = 8µ5Z logλ µ 1 < 16Z 1/σ 1 , using 8µ5 ≈ 15.86 < 16. By Lemma 6.4, we have relations Z1 = Y1 and Z2 = Y2 + iY3. We obtain: 2 + Y 2 3 = Z2 < 16Z 1/σ 1 = 16Y 1/σ 1 qY 2 . (cid:3) 12. Jordan superalgebra J, its Z4-grading and properties Assume that char K 6= 2. Now we consider the Poisson superalgebra P = Poisson(V0, V1, V2), its Kantor double yields a Jordan superalgebra J = Kan(P) = P ⊕ ¯P. In this section, we determine its properties. Namely, we establish a Z4-grading of the Jordan superalgebra J, determine its growth, and determine positions of its basis monomials in R4. First, let us determine its generators. Lemma 12.1. The Jordan superalgebra J = Kan(P(V0, V1, V2)) is generated by {V0, V1, V2, ¯1}. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 29 Proof. Let J = Jord(V0, V1, V2, ¯1) ⊂ J be a Jordan superalgebra generated by {V0, V1, V2, ¯1}. We identify Q with Lie(V0, V1, V2) (Lemma 9.2). Let us prove by induction on n that Vn, ¯Vn ∈ J for all n ≥ 0. Using Lemma 4.1, we get Vn • ¯1 = ¯Vn; ¯Vn−1 • ¯Vn = {Vn−1, Vn} = −xn−1Vn+2; xn−1Vn+2 • ¯1 = xn−1Vn+2; ¯Vn−1 • xn−1Vn+2 = {Vn−1, xn−1Vn+2} = Vn+2. Similarly, for any standard monomial w ∈ Q we show that w, ¯w ∈ J. Indeed, consider standard monomials w1, w2 ∈ Q and suppose that w1, w2 ∈ J ∩ P. Then w = (w1 • 1) • (w2 • 1) = ¯w1 • ¯w2 = {w1, w2} ∈ J ∩ P. Recall that P is spanned by products of standard monomials (proof of Theorem 10.1). Let w1, . . . , wm ∈ Q be standard monomials. Then w = w1 ··· wm = w1 • ··· • wm ∈ J and w = w • ¯1 ∈ ¯J. Therefore, J = P ⊕ ¯P = J. Corollary 12.2. (Vn−1 • ¯1) • (((Vn−1 • ¯1) • (Vn • ¯1)) • ¯1) = −Vn+2, n ≥ 1. (cid:3) We extend the shift endomorphism τ : P → P onto J by τ (¯v) = τ (v), v ∈ P. We get τ (¯1) = ¯1, and τ (Vi) = Vi+1, for all i ≥ 0. Define Jordan superalgebras Ji = Jord(Vi, Vi+1, Vi+2, ¯1) for all i ≥ 0, so J0 = J. Corollary 12.3. Let J = Jord(V0, V1, V2, ¯1). Then i) {Vi i ≥ 0} ⊂ J; ii) τ i : J → Ji is an isomorphism for any i ≥ 0; iii) we get a proper chain of isomorphic subalgebras: J = J0 % J1 % ··· % Ji % Ji+1 % ··· , ∞∩n=0 Ji = h1, ¯1i. iv) J is infinite dimensional; v) J is weakly special but not special. Proof. The last claim follows from the known fact that the Kantor double of a Poisson superalgebra is weakly special [64, 70] (a more general similar fact for arbitrary Poisson brackets is established in [41]). On the other hand, the Kantor double is special if and only if the Poisson superalgebra is Lie nilpotent of class 2, namely, it satisfies the identity {X,{Y, Z}} = 0 [64], which is not true in our case. (cid:3) We extend the weight functions of Q and P onto J by setting wt(¯1) = swt(¯1) = 0. Using Lemma 11.1, we get. Lemma 12.4. The Jordan superalgebra J = Jord(V0, V1, V2, ¯1) is N3 {V0, V1, V2}: J = ⊕n1,n2,n3≥0 Jn1,n2,n3. 0-graded by a partial multidegree in Remark 5. Consider the case char K = 2. The Kantor double of the Poisson superalgebra P yields an algebra with a binary operation J = Kan(P). Similarly, below we can define an algebra with a binary operation K = J or(Q) as well. Probably, these superalgebras can be supplied with appropriate ternary operations and be considered as Jordan superalgebras in characteristic 2. Now let us study J in more details. A monomial of the Jordan superalgebra J = P ⊕ ¯P is either w ∈ P or ¯w ∈ P, where w is a product in the letters {xi, Vi i ≥ 0}, such monomials are either even or odd with respect to the Z2-grading of the superalgebra. Below, formulas involving J are written for Z2-homogeneous elements either of P or ¯P. Let w ∈ Jn1,n2,n3 we keep the notation deg(w) = n1 + n2 + n3, the total degree in the set {V0, V1, V2}. Now we are going to introduce functions specific to the Jordan algebra J. Consider a monomial (32) We count a multiplicity of the pivot elements in this record of u ∈ P (or in its copy ¯u = u• ¯1 ∈ P) by setting: u = xi1 ··· xik Vj1 ··· Vjm ∈ P, m ≥ 0. i1 < ··· < ik, 0 ≤ j1 < ··· < jm, multV (u) = multV (¯u) = m. 30 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Let w ∈ J = P ⊕ ¯P, put ǫ(w) =(0, 1, w ∈ P; w ∈ P, where using ǫ(w) we assume that either w ∈ P or w ∈ P. Define a specific Jordan weight function jwt(∗): (33) jwt(w) = 2 multV (w) − ǫ(w), w ∈ J. Lemma 12.5. The Jordan weight jwt(∗) has the following properties. Let a, b be monomials of J. Then i) jwt(1) = 0; ii) jwt(¯1) = −1; iii) jwt(Vj ) = 2, j ≥ 0; iv) jwt(V j) = 1, j ≥ 0; v) 1 ≤ jwt(a) for a 6= 1, ¯1; vi) jwt(a • b) = jwt(a) + jwt(b) (i.e. the function is additive); vii) −1 ≤ jwt(a) < 12 + 2 logλ wt(a). Proof. Items (i–iv) follow by definition. Consider (v), we observe that multV (a) ≥ 1, hence jwt(a) ≥ 1. Let us prove the additivity. The cases a, b ∈ P and a ∈ P, b ∈ P are trivial. Consider the case a, b ∈ P. Then we can consider that a = a′(x)Vi1 ··· Vik , i1 < ··· < ik and b = b′(x)Vj1 ··· Vjm , j1 < ··· < jm, where a′(x), b′(x) are monomials in {xi i ≥ 0}. The product a • b is a linear combination of products, where the original factors are being kept except those of either {Vip , Vjq} = Pl cl(x)Vl, cl(x) ∈ Λ(xi i ≥ 0) or {Vip , xjq} = c(x) ∈ Λ(xi i ≥ 0). In both cases, we lose one pivot letter Vi, thus jwt(a • b) = 2(k + m − 1) = 2k − 1 + 2m − 1 = jwt(a) + jwt(b). Let us check bounds (vii). The lower bound follows from the definition (33). Let u ∈ P be a monomial (32) of length n ≥ 0, i.e. jm = n. Then multV (u) = m ≤ n + 1. By Lemma 11.2, λn−5 < wt(u). Hence, multV (u) < 6 + logλ wt(u). Finally, either a = u or a = ¯u and we apply (33). Lemma 12.6. Consider the Jordan superalgebra J as generated by the set {V0, V1, V2, ¯1}. Let w ∈ J be a monomial with the multidegree coordinates Gr(w) = (X1, X2, X3) and a (partial) degree deg w = X1+X2+X3 (see Lemma 12.4). (cid:3) i) there exists a well-defined degree deg¯1(w) with respect to ¯1 for a monomial w ∈ J. ii) deg¯1(w) = 2 deg w − jwt w, w ∈ J; iii) deg¯1(w) = 2(deg w − multV (w)) + ǫ(w), w ∈ J; iv) deg¯1(∗) is additive on J. Proof. Let w ∈ J be a Jordan monomial, which involves X1, X2, X3 factors V0, V1, V2, respectively, and deg¯1(w) factors ¯1. Using additivity of jwt(∗) and its basic values (Lemma 12.5), we get jwt(w) = X1 jwt(V0) + X2 jwt(V1) + X3 jwt(V2) + deg¯1(w) jwt(¯1) = 2 deg w − deg¯1(w); deg¯1(w) = 2 deg w − jwt w, thus proving (i), (ii). Using (33) we get (iii). Additivity of deg(∗) and jwt(∗) yields additivity of deg¯1(∗). (cid:3) Consider a monomial w ∈ J with the multidegree coordinates Gr(w) = (X1, X2, X3). We introduce one more coordinate X4 = deg¯1(w). Define an extended multidegree with respect to the generators {V0, V1, V2, ¯1} and an extended degree: Gr♯(w) = (X1, X2, X3, X4) ∈ N4 0; deg♯(w) = X1 + X2 + X3 + X4 = deg w + deg¯1(w), w ∈ J; in particular, deg♯(1) = 0, deg♯(¯1) = 1. We draw monomials w ∈ J using the extended multidegree coordinates, thus putting monomials at lattice points Gr♯(w) ∈ Z4 ⊂ R4. Corollary 12.7. Consider the Jordan superalgebra J. i) The functions Gr♯(∗), deg♯(∗) are additive on J; ii) J is N4 0-graded using the extended multidegree Gr♯(w) = (X1, X2, X3, X4) ∈ N4 0 in the generators {V0, V1, V2, ¯1}. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS Proof. Follow from Lemma and the definitions. Corollary 12.8. Consider a monomial w ∈ J. Then i) deg♯ w = 3 deg w − jwt w; ii) deg w ≤ deg♯ w < 3 deg w, where w 6= ¯1, 1; iii) deg♯ Vn = 3 deg Vn − 2, n ≥ 0. 31 (cid:3) Proof. We use item (ii) of Lemma, definition of the extended degree, and items (iii), (v) of Lemma 12.5. (cid:3) Theorem 12.9. Consider the Jordan superalgebra J = Jord(V0, V1, V2, ¯1). Then GKdim J = GKdim J = 2 logλ 2 ≈ 3.3036. Proof. Fix m ≥ 0. The ordinary growth function γJ(m,{V0, V1, V2, ¯1}) counts basis monomials w ∈ J such that deg♯(w) ≤ m, by the lower inequality of Corollary 12.8, we have deg w ≤ deg♯(w) ≤ m. Thus, the above set of monomials is contained in {u, ¯u u basis monomial of P, deg u ≤ m}. Since γP(m,{V0, V1, V2}) counts basis monomials u ∈ P such that deg u ≤ m, we obtain the upper bound below 2(γP(m/3,{V0, V1, V2}) − 1) ≤ γJ(m,{V0, V1, V2, ¯1}) ≤ 2γP(m,{V0, V1, V2}), m ≥ 1. (34) Similarly, let u ∈ P be a basis monomial with deg u ≤ m/3 and u 6= 1. Then w = u and w = ¯u are basis elements of J with deg♯ w ≤ 3 deg w ≤ m by the upper bound of Corollary 12.8, thus we prove the claimed lower bound. Now, it remains to use bounds of Theorem 11.3. (cid:3) Consider a monomial w ∈ J, then either w = u ∈ P or w = ¯u ∈ P. By our constructions above, this monomial has the twisted coordinates WtR(w) = (Y1, Y2, Y3) ∈ R3. We add one more coordinate Y4 = jwt w. Now we define extended twisted coordinates: WtR♯(w) = (Y1, Y2, Y3, Y4) ∈ R4. Lemma 12.10. Let w ∈ J be a monomial. i) the function WtR♯(∗) is additive on J; ii) the first three components of Gr♯(w) = (X1, X2, X3, X4) and WtR♯(w) = (Y1, Y2, Y3, Y4) are related by (iv) of Lemma 6.4. The forth coordinates are related by Y4 = 2(X1 + X2 + X3) − X4; iii) −1 ≤ Y4 < 12 + 2 logλ Y1; iv) The axis OY1 in terms of the standard coordinates is given by (2/λ, λ, 1, 2λ2 + 2λ). Proof. The additivity of WtR♯(∗) follows from that for jwt(∗). By (ii) of Lemma 12.6, X4 = deg¯1(w) = 2 deg w − jwt w = 2(X1 + X2 + X3) − Y4, thus yielding the second claim. Recall that Y1 = wt w and Y4 = jwt w. Using estimates (vii) of Lemma 12.5, we have −1 ≤ Y4 < 12 + 2 logλ wt(w) = 12 + 2 logλ Y1. Let us prove (iv). By Lemma 6.6, let (X1, X2, X3) = (2/λ, λ, 1). The condition Y4 = 0, (ii), and (16) (cid:3) yield X4 = 2(X1 + X2 + X3) = 2(2/λ + λ + 1) = 2(λ2 − 1 + λ + 1) = 2λ2 + 2λ. Theorem 12.11. Let monomials w of the Jordan superalgebra J be drawn in R4 using the extended multi- degrees Gr♯(w) = (X1, X2, X3, X4) ∈ N4 0 ⊂ R4. In terms of the extended twisted coordinates (Y1, Y2, Y3, Y4), the respective points are inside a figure determined by inequalities: qY 2 2 + Y 2 3 < 16 σpY1, −1 ≤ Y4 < 12 + 2 logλ Y1, (where σ = logµ λ ≈ 3.068); (1 ≤ Y4 < 12 + 2 logλ Y1, if w 6= 1, ¯1). Proof. The inequalities are established in Theorem 11.4 and Lemma 12.10. (cid:3) Corollary 12.12. Consider the N4 {V0, V1, V2, ¯1}: 0-grading of the Jordan superalgebra J by multidegree in the generators ⊕ The numbers {dim Jn1n2n3n4 (n1, n2, n3, n4) ∈ N4 0} are not bounded. n1,n2,n3,n4≥0 Jn1n2n3n4 . J = 32 VICTOR PETROGRADSKY AND I.P. SHESTAKOV Proof. By way of contradiction, suppose that the dimensions are bounded by a constant C. The ordinary growth function γJ(m,{V0, V1, V2, ¯1}) counts basis monomials of J with deg♯(w) ≤ m, where m ≥ 0. By Corollary 12.8, Y1 = deg w ≤ deg♯(w) ≤ m. So, we introduce a bigger function g(m) = dimhw ∈ J deg w ≤ mi. We cut the figure of Theorem by the hyperplane Y1 ≤ m, consider a larger cylinder, and evaluate volume of the latter (in the extended twisted coordinates): {(Y1, Y2, Y3, Y4) 0 ≤ Y1 ≤ m, qY 2 Volume(m) = m · π256m2/σ · (13 + 2 logλ m) ≤ C1m5/3, m ≫ 1, 3 < 16 σ√m, −1 ≤ Y4 < 12 + 2 logλ m}; 2 + Y 2 (35) (36) because σ > 3. The volume of the cylinder in the extended standard coordinates and the number of lattice points in it (in terms of the extended standard coordinates) have the same asymptotic, with a constant C2. Thus, γJ(m) ≤ g(m) ≤ CC2m5/3, m ≫ 1, a contradiction with Theorem 12.9. (cid:3) 13. Jordan superalgebra K and its properties Now we introduce our last object, the Jordan superalgebra K and study its properties. We show that K is a factor algebra of the Jordan superalgebra J constructed above, thus we can apply all the machinery developed for J. Let L be an arbitrary Lie superalgebra. Its symmetric algebra S(L) has the structure of a Poisson superalgebra. Observe, that the subspace H ⊂ S(L) spanned by all tensors of length at least two is its ideal. Thus, one obtains a (rather trivial) Poisson superalgebra P (L) = S(L)/H, which equivalently can be obtained as a vector space endowed with Poisson products which are nontrivial in the following cases only: P (L) = h1i ⊕ L, 1 · x = x, {x, y} = [x, y], x, y ∈ L. Using Kantor double, define a Jordan superalgebra J or(L) = Kan(P (L)). Equivalently, one can just take a vector space supplied with a product • which is nontrivial in the following cases (see an example at the end [65]): J or(L) = h1i ⊕ L ⊕ h¯1i ⊕ ¯L, ¯x • ¯y = [x, y], x • ¯1 = (−1)x¯1 • x = ¯x, x, y ∈ L; 1 the unit. Now we define the Jordan superalgebra K = J or(Q). Lemma 13.1. Let K = J or(Q). Then i) We have generators: K = Jord(v0, v1, v2, ¯1); ii) define Jordan superalgebras Ki = Jord(vi, vi+1, vi+2, ¯1) ⊂ K for all i ≥ 0, so K0 = K. We get a proper chain of isomorphic subalgebras: K = K0 % K1 % ··· % Ki % Ki+1 % ··· , ∞∩n=0 Ki = h1, ¯1i; Proof. Define the subalgebra K′ = Jord(v0, v1, v2, ¯1) ⊂ K. Computations of Lemma 12.1 yield that vi ∈ K′ for all i ≥ 0, moreover all basis elements of Q belong to K′. Thus, K′ = K. The second claim follows by applying the endomorphism τ . (cid:3) If an associative superalgebra A is just infinite then the related Jordan superalgebra A(+) is just infinite as well [77]. We establish a similar fact. Lemma 13.2. Let L be a Lie superalgebra, consider the Jordan superalgebra J or(L). i) J or(L) is just infinite if and only if L is just infinite. ii) The ideal without unit J oro(L) = L ⊕ h¯1i ⊕ ¯L is solvable of length 3. iii) This is a nil-ideal of bounded degree: a6 = 0 for a ∈ J oro(L). Proof. Let L be not just infinite. Then there exists an ideal of infinite codimension I ⊳ L and I ⊕ ¯I is an ideal of infinite codimension in J or(L). Therefore, J or(L) is not just infinite. Conversely, suppose that L is just infinite. By way of contradiction, assume that H ⊂ J or(L) is an ideal of infinite codimension. Then H = H ∩ (L⊕ ¯L) ⊂ J or(L) is also an ideal of infinite codimension. Denote by H0 and ¯H1 the projections of H onto L, ¯L, respectively ( ¯H1 being the copy of a subspace H1 ⊂ L). Since H is an ideal, ¯1 • H = ¯H0 ⊂ ¯H1 and ¯L • H = [L, H1] ⊂ H0 and we get [L, H1] ⊂ H0 ⊂ H1 ⊂ L. Hence H0 ⊂ L is an ideal, which must be either zero or of finite codimension by our assumption. Let H0 ⊂ L be of finite codimension then H ⊂ J or(L) is of finite codimension, a contradiction. Now assume that H0 = 0. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 33 Then [L, H1] = 0 and H1 is central. By taking 0 6= z ∈ H1, we get an ideal hzi ⊂ L of infinite codimension, a contradiction. Thus, J or(L) is just infinite. We repeat the arguments of [65]. Denote J = J oro(L). Then J 2 ⊂ L ⊕ ¯L, (J 2)2 ⊂ L, and ((J 2)2)2 = 0. Thus, J is solvable of length 3. (cid:3) Let Ko be the ideal of the Jordan superalgebra K = J or(Q) without unit. We have a basis K = h1, ¯1, w, ¯w w are standard monomials of Qi. In particular, all the pivot elements {vii ≥ 0}, as well as their copies {¯vii ≥ 0} belong to K. Lemma 13.3. One has a canonical isomorphism of Jordan superalgebras K ∼= J/I, where I is the ideal of J spanned by all its monomials containing two pivot letters Vi or two their copies ¯Vi. Proof. Consider the Jordan superalgebra J = Kan(P) = P ⊕ ¯P with the product •. Fix m ≥ 0, as above, denote by Pm ⊂ P a linear span of all m-fold products of standard monomials of Q, equivalently, Pm is spanned by the basis monomials containing exactly m letters Vi. We get vector space decompositions P = ∞⊕m=0 Pm and ¯P = ∞⊕m=0 Pn • Pm ⊂ Pn+m, Pn • ¯Pm = ¯Pm • Pn ⊂ ¯Pn+m, ¯Pm. Observe that ¯Pn • ¯Pm ⊂ Pn+m−1, n, m ≥ 0. (Pn ⊕ ¯Pn). The multiplication rules above imply that I is an ideal in J. Indeed, one needs to Let I = ⊕n≥2 check the last product, where we use that ¯P0 = h¯1i and ¯1 • ¯L = 0. We get J/I ∼= P0 ⊕ P1 ⊕ ¯P0 ⊕ ¯P1 = h1i ⊕ Q ⊕ h¯1i ⊕ ¯Q ∼= J or(Q) = K. (cid:3) By (33), jwt(Pn) = 2n and jwt( ¯Pn) = 2n − 1 for all n ≥ 0. We have another description of the ideal above, namely, I is spanned by monomials u = w or u = ¯w, where w are basis monomial of P such that jwt(u) ≥ 3. Theorem 13.4. Consider the Jordan superalgebra K = J or(Q). Then i) K is generated by {v0, v1, v2, ¯1}; ii) K is N3 dimensional and are inside the "almost cubic paraboloid" of Theorem 8.5 (see also Figure 1); 0-graded by multidegree in {v0, v1, v2}, the respective components are either trivial or two- 0-graded by multidegree in the generators {v0, v1, v2, ¯1}: iii) K is N4 K = ⊕ n1,n2,n3,n4≥0 Kn1n2n3n4, the components are at most one-dimensional, so, the N4 0-grading is fine. The points for nontrivial N4 0- components in R4 satisfy the following inequalities using the extended twisted coordinates WtR♯(w) = (Y1, Y2, Y3, Y4): σ = logµ λ ≈ 3.068; (1 ≤ Y4 ≤ 2 if w 6= 1, ¯1). Kn by degree in the generators; except the initial components qY 2 2 + Y 2 3 < 14 σpY1, −1 ≤ Y4 ≤ 2; iv) consider a (natural) gradation K = ⊕n≥0 K0 = h1i, K1 = {v0, v1, v2, ¯1}, we have: K3n−2 = Qn, K3n−1 = ¯Qn, K3n = 0, n ≥ 1; v) GKdim K = GKdim K = logλ 2 ≈ 1.6518; vi) K is just infinite but not hereditary just infinite; vii) the ideal without unit Ko is solvable of length 3; viii) elements of Ko are nil of degree at most 6; ix) K is weakly special but not special. Proof. Since K is a factor algebra of J by a homogeneous ideal, all the weight functions Gr, Gr♯, Wt, WtR, WtR♯ as well as the N3 0-gradings are inherited. We get the almost cubic paraboloid by Theorem 8.5. Let w = rn−2vn ∈ Qn1n2n3 be a standard monomial of Q. We get a two-dimensional component Kn1n2n3 = hw, ¯wi. Also, K000 = h1, ¯1i. By (33), jwt w = 2 and jwt ¯w = 1, also jwt 1 = 0 and 0 and N4 34 VICTOR PETROGRADSKY AND I.P. SHESTAKOV jwt ¯1 = −1. Hence, due to the forth different coordinates, the components of the N4 0-grading of K are at most one dimensional. Also, we get Y4 = jwt u ∈ {1, 2}, where u ∈ K is a basis element distinct from 1, ¯1. Consider the gradation of K be degree in all generators, which was called the extended degree. Let u ∈ K be a basis monomial, distinct from 1, ¯1. We have two cases. First, assume that u is a standard monomial of Q, which has a unique letter Vi. By (33) and (i) of Corollary 12.8, we have jwt u = 2 and deg♯ u = 3 deg u − jwt u = 3 deg u − 2. Second, u = ¯w, w being a standard monomial. Then jwt ¯w = 1 and deg♯ u = 3 deg u − 1. Recall that the condition deg w = n is equivalent to w ∈ Qn. We obtain the desired correspondence between components. To evaluate the growth we use estimates (34) and Theorem 8.4. By Lemma 13.2, K is just infinite. Let us prove that K is not hereditary just infinite. We use notations of Lemma 7.5. Fix m ≥ 1. Let Q(m) ⊂ Q be the linear span of the standard monomials of length at least m. By multiplication rules, H = Q(m) ⊕ Q(m) ⊂ K is an ideal of finite codimension. Let J = x0Q(m) ⊕ x0Q(m) ⊂ H be the subspace spanned by the monomials involving x0. We see that J is an abelian ideal of H. Since vi ∈ H\J, where i ≥ m, we conclude that dim H/J = ∞ and the ideal H is not just infinite. Consider the last claim. As an image of J, K is weakly special as well. Also, K is a Kantor double of the Poisson superalgebra P (Q) = h1i ⊕ Q which is not Lie nilpotent of class 2, hence, K is not special by [64]. (cid:3) In particular, we get a Jordan superalgebra K which Gelfand-Kirillov dimension belongs to (1, 2), that is not possible for associative and Jordan algebras [36, 42]. A more general fact that the gap (1, 2) does not exist for Jordan superalgebras is proved in [58]. 14. Nillity of superalgebras Q, A, P, J, and K In this section, we establish different statements on nillity of our five superalgebras. First, we prove that Q, A, and superalgebras without unit Po, Jo, Ko are direct sums of two locally nilpotent subalgebras and there are continuum such different decompositions (Theorem 14.2). Second, Q is ad-nil for Z2-homogeneous elements (Theorem 14.3). Third, in case char K = 2, the restricted Lie algebra Q = Liep(v0, v1, v2) has a nil p-mapping. Proofs of the last two facts are omitted because they are the same as that in supplied references. We start with a technical fact. Lemma 14.1. Let λ, µ, ¯µ be the real and complex roots of the polynomial t3 − t − 2. Then i) µn /∈ R for any n ≥ 1. ii) The set {arg(µn) n ≥ 1} is dense on [0, 2π]. Proof. Consider the field extension Q ⊂ Q[λ, µ]. Since the Galois group has the conjugation, this is an extension of degree 6 and the Galois group is S3. Assume that µn ∈ R for some n ≥ 1. By Viet's formulas, λµ¯µ = 2. Denote ξ = µ2λ/2, then ξ = 1. We obtain ξn = µ2nλn/2n ∈ R+ and ξn = 1. Hence, ξn = 1, we have a root of unity such that ξ ∈ Q[λ, µ]. Moreover, we can assume that ξ is primitive of degree n. Let n = Qp pnp , then by Euler's formula, Q[ξ] : Q = φ(n) = Qp(p − 1)pnp−1. Since the Galois group of a cyclotomic extension is abelian, φ(n) properly divides 6. Clearly, p ∈ {2, 3} and n ∈ {2, 3, 4}. We have µ3 = µ + 2 /∈ Q. For two remaining cases observe that R ∩ Q[λ, µ] = Q[λ]. We have either µ2 ∈ Q[λ], or µ4 = µµ3 = µ(µ + 2) = µ2 + 2µ ∈ Q[λ], in both cases, µ satisfies a polynomial of degree 2 over Q[λ]. On the other hand, µ satisfies the following irreducible polynomial of degree 2: h(t) = (t − µ)(t − ¯µ) = t2 − (µ + ¯µ)t + µ¯µ = t2 + λt + 2/λ ∈ Q[λ][t]. A contradiction proves the first claim. By the first claim, 2π/ arg(µ) /∈ Q, we obtain an irrational rotation of the unit circle, the classical example of ergodic theory. Ergodic theory says that an orbit of an irrational rotation of a circle is dense. (cid:3) Let Po ⊂ P, Jo ⊂ J, and Ko ⊂ K be the respective Poisson and Jordan superalgebras without unit. Theorem 14.2. Consider the Lie superalgebra Q = Lie(v0, v1, v2), its associative hull A = Alg(v0, v1, v2), the Poisson superalgebra without unit Po, and the Jordan superalgebras without unit Jo and Ko. i) there exist decompositions into direct sums of two locally nilpotent subalgebras: Q = Q+ ⊕ Q−, ii) there are continuum such different decompositions. Po = P+ ⊕ P−, A = A+ ⊕ A−, Jo = J+ ⊕ J−, Ko = K+ ⊕ K−. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 35 Proof. First, we consider the Lie superalgebra Q. Consider a plane Π passing through the axis OY1, it is determined by an equation αY2 + βY3 = 0 in the twisted coordinates, where α, β ∈ R are some constants. By Lemma 6.7, the axis OY1 does not contain lattice points Z3, except the origin. By rotation of the plane Π around OY1, we obtain a continuum of planes that intersect Z3 only at origin, because the number of points of the lattice is countable. Fix such a plane Π. Let Q+, Q− be sums of homogeneous components of Q that lie on different sides of Π. By construction, we get a vector space decomposition Q = Q+ ⊕ Q−. Additivity of the multidegree implies that Q+ and Q− are subalgebras. The plane Π splits the "paraboloid" (Theorem 8.5) into two halves, see Figure 1. Now the same geometric arguments as in [57] prove that the subalgebras Q+, Q− are locally nilpotent. Two such different planes yield different decompositions. Indeed, consider all pivot elements {vk k ≥ 0}, and their weight and twisted coordinates swt(vk) = µk = Z2(k) = Y2(k) + iY3(k), k ≥ 0. Since the set of their arguments is dense on [0, 2π] (Lemma 14.1), the decompositions determined by two different planes differ by (infinitely many) pivot elements. Similarly, we get decompositions for A and Po because their monomials are inside another paraboloid (Theorem 11.4). Finally, consider the Jordan superalgebra without unit Jo. We use Z3-grading of J by multidegree in {V0, V1, V2} only. Let J be a span of all monomials u, ¯u, where u is a basis monomial of P such that u 6= 1. All such monomials belong to lattice points Z3 distinct from the origin. As above, using continuum different appropriate planes passing through OY1, we split monomials into two parts and get decompositions into direct sums of two locally nilpotent subalgebras J = J+ ⊕ J−. Since ¯1 is at the origin, a multiplication by ¯1 keeps the lattice points, thus, ¯1 • J+ ⊂ J+ and ¯1 • J− ⊂ J−. By our construction, Jo = h¯1iK ⊕ J. Put J− = J− and J+ = J+ ⊕ h¯1iK. Then Jo = J+ ⊕ J−. We have (J+)2 ⊂ J+, and J+ is a locally nilpotent subalgebra as well. Theorem 14.3. Consider the Lie superalgebra Q = Lie(v0, v1, v2) = Q¯0 ⊕ Q¯1. For any a ∈ Q¯n, ¯n ∈ {¯0, ¯1}, the operator ad(a) is nilpotent. (cid:3) Proof. The same as [51, Theorem 10.1] or [13, Theorem 12.1]. (cid:3) Corollary 14.4. For any a ∈ Qn1,n2,n3, where n1, n2, n3 ≥ 0, the the operator ad(a) is nilpotent. Recall that in case char K = 2 the Lie superalgebra Q = Lie(v0, v1, v2) coincides with the restricted Lie algebra generated by the same elements, i.e. Q = Liep(v0, v1, v2) (Corollary 5.2). Theorem 14.5. Let char K = 2. The restricted Lie algebra Q = Liep(v0, v1, v2) has a nil p-mapping. Proof. The same as in [66, Proposition 1]. The ideas of that proof were further developed in [6, Corollary 2.9] and [52, Theorem 8.6]. (cid:3) References [1] Bahturin Yu.A., Mikhalev A.A., Petrogradsky V.M., and Zaicev M. V., Infinite dimensional Lie superalgebras, de Gruyter Exp. Math. vol. 7, de Gruyter, Berlin, 1992. [2] Bahturin, Yu.A.; Olshanskii, A., Large restricted Lie algebras, J. Algebra (2007) 310, No. 1, 413–427. [3] Bahturin, Yu.A.; Sehgal, S.K.; Zaicev, M.V., Group gradings on associative algebras, J. Algebra (2001) 241, No. 2, 677–698. [4] Bao, Y., Ye Y.,; Zhang, J., Restricted Poisson algebras. Pac. J. Math. 289, No. 1, 1-34 (2017). [5] Bartholdi L., Branch rings, thinned rings, tree enveloping rings. Israel J. Math. 154 (2006), 93–139. [6] Bartholdi L., Self-similar Lie algebras. J. Eur. Math. Soc. (JEMS) 17 (2015), no. 12, 3113–3151. [7] Bartholdi L., Grigorchuk R.I., Lie methods in growth of groups and groups of finite width. Computational and geometric aspects of modern algebra,1–27, London Math. Soc. Lecture Note Ser., 275, Cambridge Univ. Press, Cambridge, 2000. [8] Berezin, F. A., Several remarks on the associative hull of a Lie algebra. (Russian) Funkcional. Anal. i Prilozen 1 (1967), no. 2, 1–14. [9] Bezrukavnikov, R.; Kaledin, D. Fedosov quantization in positive characteristic. J. Am. Math. Soc. 21 (2008). No. 2, 409–438. [10] Bouarroudj, S., Grozman, P., Leites, D. Classification of finite dimensional modular Lie superalgebras with indecomposable Cartan matrix, SIGMA 5 (2009), 060, 1–63. [11] Caranti, A.; Mattarei, S.; Newman, M.F. Graded Lie algebras of maximal class. Trans. Am. Math. Soc. 349 (1997) No. 10, 4021–4051. [12] Caranti, A.; Newman, M.F. Graded Lie algebras of maximal class. II. J. Algebra 229 (2000), No. 2, 750–784. [13] de Morais Costa O.A., Petrogradsky V., Fractal just infinite nil Lie superalgebra of finite width, J. Algebra, 504, (2018), 291–335. [14] Dixmier J., Enveloping algebras. AMS, Rhode Island, 1996. 36 VICTOR PETROGRADSKY AND I.P. SHESTAKOV [15] Drensky V. and Hammoudi L., Combinatorics of words and semigroup algebras which are sums of locally nilpotent subal- gebras. Canad. Math. Bull. 47 (2004), no. 3, 343–353. [16] Elduque, A. Fine gradings on simple classical Lie algebras. J. Algebra 324 (2010), No. 12, 3532–3571. [17] Ershov M., Golod-Shafarevich groups: a survey, Int. J. Algebra Comput. 22, (2012) No. 5, Article ID 1230001. [18] Fabrykowski, J., Gupta, N., On groups with sub-exponential growth functions. J. Indian Math. Soc. (N.S.) 49 (1985), no. 3-4, 249–256. [19] Farkas D. R., Poisson polynomial identities. Comm. Algebra 26 (1998), no. 2, 401–416. [20] Farkas D. R., Poisson polynomial identities. II. Arch. Math. (Basel) 72 (1999), no. 4, 252–260. [21] Futorny F., Kochloukova D.H., Sidki S.N., On Self-Similar Lie Algebras and Virtual Endomorphisms, arXiv:1801.03005. [22] Golod, E.S. On nil-algebras and finitely approximable p-groups. Am. Math. Soc., Translat., II. Ser. 48, 103–106 (1965); translation from Izv. Akad. Nauk SSSR, Ser. Mat. 28, 273–276 (1964). [23] Golod, E.S. On some problems of Burnside type. Am. Math. Soc., Translat., II. Ser. 84, (1969) 83–88; translation from Tr. Mezdunarod. Kongr. Mat., Moskva 1966, 284–289 (1968). [24] Grigorchuk, R.I., On the Burnside problem for periodic groups., Funktsional. Anal. i Prilozhen. 14 (1980), no. 1, 53–54. [25] Grigorchuk, R.I. Degrees of growth of finitely generated groups, and the theory of invariant means. Math. USSR, Izv. 25 (1985), 259–300; translation from Izv. Akad. Nauk SSSR, Ser. Mat. 48 (1984), No.5, 939–985. [26] Grigorchuk, R.I., Just infinite branch groups. New horizons in pro-p groups, 121–179, Progr. Math., 184, Birkhauser Boston, Boston, MA, 2000. [27] Gupta N., and Sidki S., On the Burnside problem for periodic groups., Math. Z. 182 (1983), no. 3, 385–388. [28] Gupta, N.; Sidki, S. Some infinite p-groups. Algebra Logic 22, (1983) 421–424; translation from: Algebra Logika 22 (1983) No. 5, 584–589. [29] Jacobson N., Lie algebras, Interscience, New York. 1962. [30] Kac, V.G. Lie superalgebras. Adv. Math. 26, (1977), 8–96. [31] Kac, V.G. Classification of simple Z-graded Lie superalgebras and simple Jordan superalgebras. Comm. Algebra 5, (1977) 1375–1400. [32] Kac, V.G., Martinez, C.; Zelmanov, E. Graded simple Jordan superalgebras of growth one. Mem. Am. Math. Soc. 711, 140p. (2001). [33] Kantor I.L., Jordan and Lie superalgebras determined by a Poisson algebra. Aleksandrov, I. A. (ed.) et al., Second Siberian winter school Algebra and Analysis. Proceedings of the second Siberian school, Tomsk State University, Tomsk, Russia, 1989. Transl. AMS Transl., Ser. 2, Am. Math. Soc. 151, 55–80 (1992). [34] Kelarev A.V., A sum of two locally nilpotent rings may be not nil. Arch. Math. 60 (1993), no. 5, 431–435. [35] King D.; McCrimmon K. The Kantor construction of Jordan superalgebras. Commun. Algebra 20 (1992)., No.1, 109–126. [36] Krause G.R. and Lenagan T.H., Growth of algebras and Gelfand-Kirillov dimension, AMS, Providence, R.I., 2000. [37] Krylyouk Ia., The enveloping algebra of the Petrogradsky-Shestakov-Zelmanov algebra is not graded-nil in the critical characteristics, J. Lie Theory, 21, (2011), No. 3, 703–709. [38] Lenagan, T.H., Smoktunowicz Agata., An infinite dimensional affine nil algebra with finite Gelfand-Kirillov dimension, J. Am. Math. Soc. 20, (2007) No. 4, 989–1001. [39] Makar-Limanov, L., Shestakov, I., Polynomial and Poisson dependence in free Poisson algebras and free Poisson fields. J. Algebra 349, (2012) No. 1, 372–379. [40] Makar-Limanov, L., Umirbaev, U., The Freiheitssatz for Poisson algebras. J. Algebra 328, (2011) No. 1, 495–503. [41] Martinez, C., Shestakov, I., Zelmanov, E., Jordan superalgebras defined by brackets. J. Lond. Math. Soc., II. Ser. 64, (2001) No. 2, 357–368. [42] Martinez C., Zelmanov E., Jordan algebras of Gelfand-Kirillov dimension one. J. Algebra 180,(1996). No.1, 211–238. [43] Martinez C., Zelmanov E., Nil algebras and unipotent groups of finite width. Adv. Math. 147, (1999) No.2, 328–344. [44] Mikhalev A.A., Subalgebras of free Lie p-superalgebras. Math. Notes 43, (1988) No.2, 99–106; translation from Mat. Zametki 43 (1988) No.2, 178–191. [45] Millionschikov D.V., Naturally graded Lie algebras (Carnot algebras) of slow growth. arXiv:1705.07494. [46] Mishchenko S.P., Petrogradsky V.M. and Regev A., Poisson PI algebras, Trans. Amer. Math. Soc., 359, (2007), no. 10, 4669–4694. [47] Nekrashevych, V., Self-similar groups. Mathematical Surveys and Monographs 117. Providence, RI: American Mathemat- ical Society (AMS) (2005). [48] Passman D.S. and Petrogradsky V.M., Polycyclic restricted Lie algebras, Comm. Algebra, 29 (2001), no. 9, 3829–3838. [49] Petrogradsky V.M., On Lie algebras with nonintegral q-dimensions. Proc. Amer. Math. Soc. 125 (1997), no. 3, 649–656. [50] Petrogradsky V.M., Examples of self-iterating Lie algebras, J. Algebra, 302 (2006), no. 2, 881–886. [51] Petrogradsky V., Fractal nil graded Lie superalgebras, J. Algebra, 466 (2016), 229–283. [52] Petrogradsky V., Nil Lie p-algebras of slow growth, Comm. Algebra. 45, (2017), no. 7, 2912–2941. [53] Petrogradsky V. M., Yu. P. Razmyslov, and E. O. Shishkin, Wreath products and Kaluzhnin-Krasner embedding for Lie algebras, Proc. Amer. Math. Soc., 135, (2007), 625–636. [54] Petrogradsky V.M. and Shestakov I.P. Examples of self-iterating Lie algebras, 2, J. Lie Theory, 19 (2009), no. 4, 697–724. [55] Petrogradsky V.M. and Shestakov I.P. Self-similar associative algebras, J. Algebra, 390 (2013), 100–125. [56] Petrogradsky V.M. and Shestakov I.P. On properties of Fibonacci restricted Lie algebra, J. Lie Theory, 23 (2013), no. 2, 407–431. [57] Petrogradsky V.M., Shestakov I.P., and Zelmanov E., Nil graded self-similar algebras, Groups Geom. Dyn., 4 (2010), no. 4, 873–900. [58] Petrogradsky V., and Shestakov I.P., On Jordan doubles of slow growth of Lie superalgebras, preprint. FRACTAL NIL GRADED LIE, ASSOCIATIVE, POISSON, AND JORDAN SUPERALGEBRAS 37 [59] Radford D. E., Divided power structures on Hopf algebras and embedding Lie algebras into special-derivation algebras, J. Algebra, 98 (1986), 143–170. [60] Ratseev, S.M., Poisson algebras of polynomial growth, Sib. Math. J. 54, (2013) No. 3, 555–565, translation from Sib. Mat. Zh. 54, (2013) No. 3, 700–711. [61] Razmyslov Yu.P., Identities of algebras and their representations. AMS, Providence RI 1994. [62] Rozhkov, A.V. Lower central series of a group of tree automorphisms, Math. Notes 60, No.2, 165–174 (1996); translation from Mat. Zametki 60, No.2, 225–237 (1996). [63] Scheunert, M. The theory of Lie superalgebras. Lecture Notes in Mathematics. 716. Berlin. Springer-Verlag. 1979. [64] Shestakov I.P., Quantization of Poisson superalgebras and speciality of Jordan Poisson superalgebras. Algebra i Logika 32 (1993), no. 5, 571–584; English translation: Algebra and Logic 32 (1993), no. 5, 309–317. [65] Shestakov, I.P. Alternative and Jordan superalgebras. Sib. Adv. Math. 9, (1999). No.2, 83–99. [66] Shestakov I.P. and Zelmanov E., Some examples of nil Lie algebras. J. Eur. Math. Soc. (JEMS) 10 (2008), no. 2, 391–398. [67] Shestakov, I.; Umirbaev, U.U., The tame and the wild automorphisms of polynomial rings in three variables. J. Am. Math. Soc. 17, (2004) No. 1, 197–227. [68] Sidki S.N., A primitive ring associated to a Burnside 3-group. J. London Math. Soc. (2) 55 (1997), no. 1, 55–64. [69] Sidki S.N., Functionally recursive rings of matrices - Two examples. J.Algebra 322 (2009), no. 12, 4408–4429. [70] Skosyrskij, V.G. Prime Jordan algebras and the Kantor construction. Algebra Logic 33, (1994) No.3, 169–179; translation from Algebra Logika 33, (1994) No.3, 301–316. [71] Strade H., Simple Lie algebras over fields of positive characteristic. I: Structure theory. Berlin: de Gruyter. 2004. [72] Strade H. and Farnsteiner R., Modular Lie algebras and their representations, New York etc.: Marcel Dekker, 1988. [73] Vergne M., La structure de Poisson sur l'alg`ebre sym´etrique d'une alg`ebre de Lie nilpotente. C. R. Acad. Sc. Paris 269 (1969), S´er. A-B, 950–952. [74] Voden T., Subalgebras of Golod-Shafarevich algebras, Int. J. Algebra Comput. 19, (2009) No. 3, 423–442. [75] Zelmanov E., Some open problems in the theory of infinite dimensional algebras., J. Korean Math. Soc. 44 (2007), no. 5, 1185–1195. [76] Zelmanov E., A private communication. [77] Zhelyabin, V.N., Panasenko, A.S., Hersteins construction for just infinite superalgebras, Siberian El. Math. Reports (2017) 14, 1317–1323. Department of Mathematics, University of Brasilia, 70910-900 Brasilia DF, Brazil E-mail address: [email protected] Instituto de Mathem´atica e Estat´ıstica, Universidade de Sao Paulo, Caixa postal 66281, 05315-970, Sao Paulo, Brazil E-mail address: [email protected]
1410.8212	3	1410	2015-03-06T00:18:04	PBW deformations of skew polynomial rings and their group extensions	[ "math.RA", "math.QA" ]	We examine PBW deformations of finite group extensions of skew polynomial rings, in particular the quantum Drinfeld orbifold algebras defined by the first author. We give a homological interpretation, in terms of Gerstenhaber brackets, of the necessary and sufficient conditions on parameter functions to define a quantum Drinfeld orbifold algebra, thus clarifying the conditions. In case the acting group is trivial, we determine conditions under which such a PBW deformation is a generalized enveloping algebra of a color Lie algebra; our PBW deformations include these algebras as a special case.	math.RA	math	PBW DEFORMATIONS OF QUANTUM SYMMETRIC ALGEBRAS AND THEIR GROUP EXTENSIONS PIYUSH SHROFF AND SARAH WITHERSPOON Abstract. We examine PBW deformations of finite group extensions of quantum symmetric al- gebras, in particular the quantum Drinfeld orbifold algebras defined by the first author. We give a homological interpretation, in terms of Gerstenhaber brackets, of the necessary and sufficient con- ditions on parameter functions to define a quantum Drinfeld orbifold algebra, thus clarifying the conditions. In case the acting group is trivial, we determine conditions under which such a PBW de- formation is a generalized enveloping algebra of a color Lie algebra; our PBW deformations include these algebras as a special case. 1. Introduction Poincar´e-Birkhoff-Witt (PBW) deformations of quantum symmetric algebras were studied by Berger [2] and include important classes of examples such as the generalized enveloping algebras of color Lie algebras. PBW deformations of group extensions of (quantum) symmetric algebras include many other algebras such as rational Cherednik algebras and their generalizations studied by a number of mathematicians (see, e.g., [3, 4, 5, 10, 15]). The first author [21] gave necessary and sufficient conditions on parameter functions to define such PBW deformations in this general context. In this paper we clarify these conditions by connecting them to homological information contained in the Gerstenhaber algebra structure of Hochschild cohomology. We show explicitly how color Lie algebras are related to these PBW deformations. We begin with the quantum symmetric algebra (or skew polynomial ring), Sq(V ) := khv1, . . . , vn vivj = qijvjvi for all 1 ≤ i, j ≤ ni, where k is a field of characteristic 0, V is a finite dimensional vector space over k with basis v1, v2, . . . , vn and q := (qij)1≤i,j≤n is a tuple of nonzero scalars for which qii = 1 and qji = q−1 for ij all i, j. Let G be a finite group acting linearly on V in such a way that there is an induced action on Sq(V ) by algebra automorphisms. For example, if G acts diagonally on the chosen basis of V , this will be the case. There are other possible actions as well; see, for example, [1, 9] for actions leading to interesting deformations. We denote the action of G by left superscript, that is, gv is the element of V that results from the action of g ∈ G on v ∈ V . We may form the corresponding skew group algebra: In general for any algebra S with action of G by automorphisms, the skew group algebra S ⋊ G is S ⊗k kG as a left S-module, and has the following multiplicative structure. Write S ⋊ G = ⊕g∈GSg, where Sg = S ⊗k kg, and for each s ∈ S and g ∈ G, denote by s#g the element s ⊗ g in this g-component Sg. Multiplication on S ⋊ G is determined by (r#g)(s#h) := r(gs)#gh Date: March 3, 2014. 2010 Mathematics Subject Classification. 16E40, 16S35, 16S80, 17B35, 17B75. Key words and phrases. Quantum Drinfeld orbifold algebra, Hochschild cohomology, skew group algebra, color Lie algebra, quantum symmetric algebra, Gerstenhaber bracket. The second author was supported by NSF grants DMS-1101399 and DMS-1401016. 1 for all r, s ∈ S and g, h ∈ G. Then S ⋊ G is a graded algebra, where elements of V have degree 1 and elements of G have degree 0. Let κ : V × V → (k ⊕ V ) ⊗k kG be a bilinear map for which κ(vi, vj) = −qijκ(vj , vi) for all 1 ≤ i, j ≤ n. Let T (V ) denote the tensor algebra on V over k, in which we suppress tensor symbols denoting multiplication. Identify the target space of κ with the subspace of T (V ) ⋊ G consisting of all elements of degree less than or equal to 1. Define (1.1) Hq,κ := (T (V ) ⋊ G)/(vivj − qijvjvi − κ(vi, vj) 1 ≤ i, j ≤ n), a quotient of the skew group algebra T (V ) ⋊ G by the ideal generated by all elements of the form vivj − qijvjvi − κ(vi, vj). Note that Hq,κ is a filtered algebra. We call Hq,κ a quantum Drinfeld orbifold algebra if it is a PBW deformation of Sq(V ) ⋊ G, that is, if its associated graded algebra is isomorphic to Sq(V ) ⋊ G. Equivalently, the set {vm1 n #g mi ≥ 0, g ∈ G} is a k-basis for Hq,κ. 1 vm2 2 · · · vmn Quantum Drinfeld orbifold algebras include as special cases many algebras of interest, from rational Cherednik algebras and generalizations (see [4, 5, 7, 10, 15]), to generalized enveloping algebras of color Lie algebras and quantum Lie algebras in case G is the trivial group (see [8, 13, 14, 24]). Our analysis in this paper of the necessary and sufficient conditions on the parameter function κ to define a quantum Drinfeld orbifold algebra applies to all of these algebras as special cases. Organization. This paper is organized as follows. In Section 2, we first recall from [21] the necessary and sufficient conditions (called "PBW conditions") for Hq,κ to be a quantum Drinfeld orbifold algebra, and show that these are all the PBW deformations of Sq(V ) ⋊ G in which the action of G on V is preserved. In Section 3, we give a precise relationship between color Lie algebras and quantum Drinfeld orbifold algebras. In Section 4, we interpret the PBW conditions in terms of Gerstenhaber brackets on Hochschild cohomology. Throughout this paper, k denotes a field of characteristic 0, and tensor products and exterior powers are taken over k. Some results are valid more generally in other characteristics, however some of the homological techniques of Section 4 require the characteristic of k to be coprime to the order of G, so we stick with characteristic 0 throughout for efficiency of presentation. 2. Necessary and sufficient conditions We decompose κ into its constant and linear parts: κ = κC + κL where κC : V × V → kG and κL : V × V → V ⊗ kG. For each g ∈ G, let κg : V × V → k ⊕ V be the function determined by the equation κ(v, w) = Xg∈G κg(v, w)#g, where κg also decomposes into its constant and linear parts, i.e., κg = κC g : V ×V → k and κL g : V × V → V . We recall the following theorem from [21] which gives necessary and sufficient conditions for Hq,κ to be a quantum Drinfeld orbifold algebra, that is to be a PBW deformation of Sq(V ) ⋊ G. g where κC g +κL We will need some notation to state the conditions. For each g ∈ G and basis vector vj ∈ V , write gj i ∈ k for the scalars given by gvj = gj i vi. n Xi=1 2 The quantum (i, j, k, l)-minor determinant of g is detijkl(g) := gj l gi k − qjigi l gj k. The following theorem is a simultaneous generalization of main results in [9, 17]. Theorem 2.1. [21, Theorem 2.2] The algebra Hq,κ, defined in (1.1), is a quantum Drinfeld orbifold algebra if and only if the following conditions hold: (1) For all g, h ∈ G and 1 ≤ i < j ≤ n, κC g (vj, vi) =Xk<l detijkl(h)κC hgh−1(vl, vk) and For all distinct i, j, k and for all g ∈ G, h(cid:0)κL g (vj, vi)(cid:1) =Xk<l detijkl(h)κL hgh−1(vl, vk). (2) qjiqkiviκL g (vk, vj) − κL g (vk, vj)gvi − qkjvjκL + qjiκL g (vk, vi)gvj + vkκL g (vk, vi) g (vj, vi) − qkiqkjκL g (vj, vi)gvk = 0 , gh−1(κL h (vj, vk), hvi) − κL gh−1(vi, κL h (vj, vk)) + qikqjkκL gh−1(κL h (vk, vi), hvj) −qijqikκL gh−1(vj, κL h (vk, vi)) + κL gh−1(κL h (vi, vj), hvk) − qikqjkκL gh−1(vk, κL (3) Xh∈G(cid:16)qijqikκL (4) Xh∈G(cid:16)qijqikκC g (vj, vk)(vi − qijqik = 2(cid:0)κC gvi) + κC g (vk, vi)(qijqikvj − qikqjk gvj) + κC g (vi, vj)(qikqjkvk − gvk)(cid:1), gh−1(κL h (vj, vk), hvi) − κC gh−1(vi, κL h (vj, vk)) + qikqjkκC gh−1(κL h (vk, vi), hvj) −qijqikκC gh−1(vj, κL h (vk, vi)) + κC gh−1(κL h (vi, vj), hvk) − qikqjkκC gh−1(vk, κL h (vi, vj))(cid:17) h (vi, vj))(cid:17) = 0 . We note that condition (1) above is equivalent to G-invariance of κ, that is, hκ(v, w) = κ(hv, hw) for all h ∈ G and v, w ∈ V . Example 2.2. By modifying Example 5.5 of [12], we obtain a quantum Drinfeld orbifold algebra for which some qij 6= 1 and κL 6≡ 0: Let G be a cyclic group of order 3 generated by g. Let q be a primitive third root of 1 in C. Let V = C3 with basis v1, v2, v3 and q21 = q, q32 = q, q13 = q. Take the following diagonal action of G on V with respect to this basis: gv3 = v3. gv2 = q2v2, gv1 = qv1, Let κ(v2, v1) = v3, κ(v3, v2) = 0, κ(v1, v3) = 0. We check the conditions of Theorem 2.1: Condition (1) is G-invariance, and we may check that indeed κ(gv2, gv1) = q3κ(v2, v1) = v3 = gκ(v2, v1), and similarly for other triples consisting of one group element and two basis vectors. Condition (2) holds: q13q23v3κL 1 (v2, v1) − κL 1 (v2, v1)v3 + 0 + 0 = qq−1v3v3 − v3v3 = 0. 3 Finally, Conditions (3) and (4) hold as all terms are equal to 0. The resulting quantum Drinfeld orbifold algebra is Hq,κ = (T (V ) ⋊ G)/(v2v1 − qv1v2 − v3, v3v2 − qv2v3, v1v3 − qv3v1). Example 2.3. Another example has trivial group (G = 1): Let V = k3 with basis v1, v2, v3 and qij = −1 whenever i 6= j. Let One may check that the conditions of Theorem 2.1 hold, and consequently κ(v2, v1) = v1, κ(v3, v2) = v3, κ(v1, v3) = 0. Hq,κ = T (V )/(v2v1 + v1v2 − v1, v3v2 + v2v3 − v3, v1v3 + v3v1) is a quantum Drinfeld orbifold algebra. Now consider any PBW deformation U of Sq(V ) ⋊ G in which the action of G on V is preserved, that is, the relations gv = gvg (g ∈ G, v ∈ V ) hold in U . Then U is a Z-filtered algebra for which gr U ∼= Sq(V ) ⋊ G. We will show next that U ∼= Hq,κ for some κ. Theorem 2.4. Let U be a PBW deformation of Sq(V ) ⋊ G in which the action of G on V is preserved. Then U ∼= Hq,κ, a quantum Drinfeld orbifold algebra as defined in (1.1), for some κ. Proof. Let F denote the filtration on U . Since the associated graded algebra of U is Sq(V ) ⋊ G, we may identify F1U with (V ⊗ kG) ⊕ kG. By hypothesis, vivj − qijvjvi ∈ F1U for each i, j, and we set this element equal to κL(vi, vj) + κC(vi, vj), where κL(vi, vj) ∈ V ⊗ kG and κC (vi, vj) ∈ kG, thus defining κL and κC on pairs of basis elements. Extend bilinearly to V × V , and set κ = κL + κC . By definition, κ(vi, vj) = −qijκ(vj, vi). We will show that U is isomorphic to Hq,κ. Let σ : T (V ) ⋊ G → U be the algebra homomorphism determined by σ(vi#1) = vi and σ(1#g) = g for all vi, g. There is indeed such a (uniquely determined) algebra homomorphism since the action of G on V is preserved in U by hypothesis. By its definition, σ is surjective, since U is generated by the vi, g. We will show that the kernel of σ is precisely the ideal generated by all elements of the form vivj − qijvjvi − κ(vi, vj). Let I be this ideal. Then Hq,κ = T (V ) ⋊ G/I by definition of Hq,κ. By the definition of κ, the kernel of σ contains I, and so σ factors through Hq,κ, inducing a surjective homomorphism σ : Hq,κ → U . Now in each degree, Hq,κ and U have the same dimension, as each has associated graded algebra Sq(V ) ⋊ G. This forces σ to be injective as well. (cid:3) 3. Color Lie algebras We first recall the definition of a color Lie algebra and of its generalized enveloping algebras. For more details, see, for example, Petit and Van Oystaeyen [14]. Let A be an abelian group and let ε : A × A → k× be an antisymmetric bicharacter, where k× is the group of units in k, that is, (3.1) (3.2) (3.3) for all a, b, c ∈ A. ε(a, b)ε(b, a) = 1, ε(a, bc) = ε(a, b)ε(a, c), ε(ab, c) = ε(a, c)ε(b, c), An (A, ε)-color Lie algebra is an A-graded vector space L = ⊕a∈ALa equipped with a bilinear bracket [−, −] for which (3.4) [La, Lb] ⊆ Lab, 4 (3.5) (3.6) [x, y] = −ε(x, y)[y, x], ε(z, x)[x, [y, z]] + ε(x, y)[y, [z, x]] + ε(y, z)[z, [x, y]] = 0, whenever a, b ∈ A, and x, y, z ∈ L are homogeneous elements (any element x ∈ La is called homogeneous of degree a, and we write x = a). Now let L be a color Lie algebra and let ω : L × L → k for which (3.7) ε(z, x)ω(x, [y, z]) + ε(x, y)ω(y, [z, x]) + ε(y, z)ω(z, [x, y]) = 0 whenever x, y, z ∈ L are homogeneous elements. The generalized enveloping algebra of L associated with ω is Uω(L) := T (L)/(vivj − ε(vi, vj )vjvi − [vi, vj] − ω(vi, vj)), where vi, vj ∈ L range over a basis of homogeneous elements. If L is a Lie algebra, that is if ε takes only the value 1, the generalized enveloping algebras are precisely the Sridharan enveloping algebras [20]. If ω ≡ 0, the generalized enveloping algebras are sometimes called universal enveloping algebras, as they have a universal property generalizing that of a universal enveloping algebra of a Lie algebra [16]: U0(L) is universal with respect to linear maps f : L → S for associative algebras S that take the bracket in L to the ε-commutator [−, −]ε on the image of L in S ( [f (x), f (y)]ε := f (x)f (y) − ε(x, y)f (y)f (x) for all homogeneous x, y ∈ L). Example 3.8. Let U be the associative algebra generated by x, y, and z subject to the relations xy + yx = z, xz + zx = 0, yz + zy = 0. This is the universal enveloping algebra of a color Lie algebra analogous to the Heisenberg Lie algebra. For the definition of the color Lie algebra itself, one may take A = Z3 2 with ε(a, b) = (−1)a1b1+a2b2+a3b3 where a = (a1, a2, a3), b = (b1, b2, b3). Take L to be of dimension 3, with basis x, y, z of degrees (1, 1, 0), (1, 0, 1), (0, 1, 1), respectively. The next theorem describes a relationship between generalized enveloping algebras Uω(L) and quantum Drinfeld orbifold algebras Hq,κ. It states that the generalized enveloping algebras of color Lie algebras are precisely those quantum Drinfeld orbifold algebras with G = 1 that satisfy two technical conditions on the parameter function κ, as detailed in the theorem. We will use Theorem 2.1 to prove part (a) of the next theorem; alternatively Berger's quantum PBW Theorem [2] may be used. Part (b) largely follows from Petit and Van Oystaeyen's work on generalized enveloping algebras of color Lie algebras. We will need some notation: Letting Hq,κ be a quantum Drinfeld orbifold algebra in which G = 1, scalars C i,j l ∈ k are defined by κL(vi, vj) =Xl C i,j l vl. Theorem 3.9. (a) Let U = Hq,κ be a quantum Drinfeld orbifold algebra with G = 1. Assume that for each triple i, j, l of indices (i 6= j), if C i,j 6= 0, then qimqjm = qlm for all m, and that the left and right sides of the equation in Theorem 2.1(3) are each equal to 0 for all triples of vectors. Then U ∼= Uω(L), a generalized enveloping algebra for some color Lie algebra L and ω satisfying (3.7). (b) Let U = Uω(L) be a generalized enveloping algebra of a color Lie algebra L. Then U ∼= Hq,κ, a quantum Drinfeld orbifold algebra as defined in (1.1), for some q, κ and G = 1. Moreover, for each triple of i, j, l indices (i 6= j), if C i,j 6= 0, then qimqjm = qlm for all m, and the left and right sides of the equation in Theorem 2.1(3) are each equal to 0 for all triples of vectors. l l 5 Remark 3.10. One may check that Example 2.3 fails the first condition (corresponding to nonzero scalars C i,j l ) of Theorem 3.9(a); it is not a generalized enveloping algebra of a color Lie algebra since it cannot satisfy (3.4). By contrast, Example 3.8 satisfies both conditions (with ω ≡ 0). Proof. (a) Let U = Hq,κ be a quantum Drinfeld orbifold algebra as defined in (1.1), with G = 1, under the stated assumptions. Let A = Zn, a free abelian group on a choice of generators a1, . . . , an (where n is the dimension of the vector space V ). Let ε(ai, aj) := qij for each i, j ∈ {1, . . . , n}. Then ε(ai, aj)ε(aj , ai) = qijqji = 1 for all i, j, that is, (3.1) holds for the generators of A. Since A is a free abelian group, we may extend ε uniquely to an antisymmetric bicharacter on all of A via the relations (3.2), (3.3). Set L = V . We will show that L is a color Lie algebra with respect to a quotient group of A. Let [vi, vj] := κL(vi, vj). Then the condition [x, y] = −ε(x, y)[y, x] holds for all homogeneous x, y ∈ L, as a result of the condition κ(vi, vj) = −qijκ(vj, vi). Note that by the hypothesis on κ, (3.6) holds as a consequence of Theorem 2.1(3): The left side of (3) in Theorem 2.1 is assumed equal to 0, and this condition may be rewritten (with G = 1, g = 1, h = 1) as qijqik[[vj, vk], vi] − [vi, [vj, vk]] + qikqjk[[vk, vi], vj ] − qijqik[vj, [vk, vi]] + [[vi, vj], vk] − qikqjk[vk, [vi, vj]] = 0. We wish to rewrite half of these terms in order to compare with (3.6). By hypothesis, if C i,j for some i, j, l, then qlk = qikqjk for all k, so l 6= 0 n [[vi, vj], vk] = C i,j l [vl, vk] n Xl=1 Xl=1 Xl=1 n = − = − C i,j l qlk[vk, vl] C i,j l qikqjk[vk, vl] = −qikqjk[vk, C i,j l vl] = −qikqjk[vk, [vi, vj]]. n Xl=1 Similarly we have [[vk, vi], vj] = −qijqkj[vj, [vk, vi]] and [[vj, vk], vi] = −qjiqki[vi, [vj , vk]]. Substitut- ing into the earlier equation, it now becomes [vi, [vj, vk]] + qijqik[vj, [vk, vi]] + qikqjk[vk, [vi, vj]] = 0. Multiplying by qki, we obtain (3.6). We will need to pass next to a quotient of A to obtain the required relation between the bracket and a grading on L: Let N = rad(ε) = {a ∈ A ε(a, b) = 1 for all b ∈ A}. Let A = A/N and Lai := Spank{vi} for each i, where ai := aiN , and La := 0 for all other elements a of A. It only remains to show that (cid:2)La, Lb(cid:3) ⊆ Lab 6 for all a, b ∈ A. By hypothesis, if C i,j for all m. This implies that l 6= 0 in the expression [vi, vj] =Pl C i,j 1 = qimqjmq−1 lm = ε(ai, am)ε(aj, am)ε(a−1 l l , am) = ε(aiaja−1 , am). l vl, then qimqjm = qlm It follows that aiaja−1 color Lie algebra. l ∈ N , so aiaj = al. Thus [La, Lb] ⊆ Lab for all a, b ∈ A, implying that L is a Now, for all i, j, let ω(vi, vj) := κC(vi, vj). Then (3.7) is a consequence of Theorem 2.1(4) by a similar computation to that above for (3.6). Hence U ∼= Uω(L), a generalized enveloping algebra of the color Lie algebra L. (b) Let U = Uω(L) be a generalized enveloping algebra of a color Lie algebra L. Let V = L. Choose a basis v1, . . . , vn of V consisting of homogeneous elements and for each i, j, let Let G = 1. Set κL(vi, vj) := [vi, vj] and κC (vi, vj) := ω(vi, vj). By [14, Theorem 3.1], the associated graded algebra of U is Sq(V ). So the conditions of Theorem 2.1 must hold, and Hq,κ is a quantum Drinfeld orbifold algebra. By their definitions, Hq,κ = Uω(L). qij = ε(vi, vj ). l One may check that (3.4) implies that if C i,j 6= 0, then qimqjm = qlm for all m (similarly to computations in the proof of part (a)). From this and (3.6) it now follows that the left side of the equation in Theorem 2.1(3) is equal to 0 (similarly to computations in the proof of part (a)), and therefore the right side is also 0. (cid:3) Remark 3.11. The hypothesis on the scalars C i,j in Theorem 3.9(a) is not as restrictive as it appears. If we assume that κ is a Hochschild 2-cocycle written in the canonical form given in [11, Theorem 4.1], this condition holds automatically as a consequence of the relations defining the space C1 there (see [11, (12)]). l 4. Homological conditions We first recall the definition of Hochschild cohomology and some resolutions that we will need. For more details, see, e.g., [6]. Let R be an algebra over k, and let M be an R-bimodule. Identify M with a (left) Re-module, where Re = R ⊗ Rop; here, Rop denotes the algebra R with the opposite multiplication. The Hochschild cohomology of R with coefficients in M is HH q(R, M ) := Ext q Re(R, M ), where R is itself considered to be an Re-module under left and right multiplication. Let R = S ⋊ G, where S is a k-algebra with an action of a group G by automorphisms. Since the characteristic of k is 0, we have HH q(S ⋊ G) ∼= HH q(S, S ⋊ G)G, where the superscript G denotes invariants under the induced action of G (see, e.g., [22]). As a graded vector space, HH q(S, S ⋊ G) = Ext q Ext q Se(S, Sg), Se(S, S ⋊ G) ∼=Mg∈G where, as before, Sg denotes the g-component S ⊗ kg. Letting S = Sq(V ), each summand above can be explicitly determined using the following free Se-resolution of S = Sq(V ), called its Koszul resolution (see [23, Proposition 4.1(c)]): (4.1) · · · −→ Se ⊗V2(V ) d2−→ Se ⊗V1(V ) 7 d1−→ Se mult−−−→ S −→ 0, with differentials for 1 ≤ p ≤ n: dp(1 ⊗ 1 ⊗ vj1 ∧ · · · ∧ vjp) = p (−1)i+1" i Ys=1 Xi=1 qjs,ji! vji ⊗ 1 − p Ys=i qji,js! ⊗ vji# ⊗ vj1 ∧ · · · ∧ vji ∧ · · · ∧ vjp whenever 1 ≤ j1 < · · · < jp ≤ n. Applying HomSe(−, Sg), dropping the term HomSe(S, Sg), and identifying HomSe(Se ⊗Vp(V ), Sg) with Homk(Vp(V ), Sg), we obtain d∗ d∗ (4.2) where V ∗ denotes the vector space dual to V . Thus the space of cochains is 0 −→ Sg 1−→ Sg ⊗V1(V ∗) C q =Mg∈G C q g, where C p 2−→ Sg ⊗V2(V ∗) −→ · · · , g = Sg ⊗Vp(V ∗), for each degree p and g ∈ G. For convenience in notation, we define vj ∧ vi := −qjivi ∧ vj whenever i < j (in contrast to the standard exterior product). We view the function κ, in the definition (1.1) of quantum Drinfeld orbifold algebra, as an element of C 2 by setting κ(vi ∧ vj) = κ(vi, vj) for all i, j. The bar resolution of any k-algebra R is: · · · (4.3) δ3−→ R⊗4 δ2−→ R⊗3 δ1−→ Re mult i=0(−1)ir0 ⊗ · · · ⊗ riri+1 ⊗ · · · ⊗ rm+1 for all r0, . . . , rm+1 ∈ R, and −−−→ R −→ 0 the action of Re is by multiplication on the leftmost and rightmost factors. where δm(r0 ⊗ · · · ⊗ rm+1) =Pm From [23] (see also [11]), maps φp : Se ⊗Vp(V ) → S⊗(p+2) defining an embedding from the Koszul resolution to the bar resolution of S = Sq(V ) are given by (4.4) φp(1 ⊗ 1 ⊗ vj1 ∧ · · · ∧ vjp) = Xπ∈Sp (sgn π)qj1,...,jp π ⊗ vjπ(1) ⊗ · · · ⊗ vjπ(p) ⊗ 1 where the scalars qj1,...,jp π are determined by the equation qj1,...,jp π vjπ(1) · · · vjπ(p) = vj1 · · · vjp. We wish to use maps ψp : S⊗(p+2) → Se ⊗Vp(V ) defining a chain map from the bar resolution to the Koszul resolution. For our purposes, we need only define these maps for particular arguments in low degrees: Let ψ0 be the identity map, and ψ1(1 ⊗ vi ⊗ 1) = 1 ⊗ 1 ⊗ vi. One checks directly that ψ0δ1 and d1ψ1 take the same values on elements of the form 1 ⊗ vi ⊗ 1. We define ψ1(1 ⊗ vivj ⊗ 1) = 1 2 (qij ⊗ vi + vi ⊗ 1) ⊗ vj + 1 2 (qijvj ⊗ 1 + 1 ⊗ vj) ⊗ vi for all i, j. (This is a different, more symmetric, choice than that made in [12], and it will better suit our purposes.) Again we may check that ψ0δ1 takes the same values as d1ψ1 on elements of the form 1 ⊗ vivj ⊗ 1. The map ψ1 may be extended to elements of degrees higher than 2 in Sq(V )⊗3, but we will not need these further values, and they will not affect our calculations in the next step. Our choices allow us to define ψ2(1 ⊗ vi ⊗ vj ⊗ 1) = 1 2 ⊗ 1 ⊗ vi ∧ vj whenever i 6= j, and we may check that ψ1δ2 and d2ψ2 take the same values on elements of the form 1 ⊗ vi ⊗ vj ⊗ 1. (We may take ψ2(1 ⊗ vi ⊗ vi ⊗ 1) = 0.) As a consequence, ψ2(vi ⊗ vj − qijvj ⊗ vi) = vi ∧ vj for i < j 8 (here we have dropped extra tensor factors of 1), and thus ψ2φ2 is the identity map on input of this form, as is ψ1φ1 on the input considered above. If α and β are elements of HomRe(R⊗4, R) ∼= Homk(R⊗2, R), for any algebra R, then their circle product α ◦ β ∈ Homk(R⊗3, R) is defined by α ◦ β(r1 ⊗ r2 ⊗ r3) := α(β(r1 ⊗ r2) ⊗ r3) − α(r1 ⊗ β(r2 ⊗ r3)) for all r1, r2, r3 ∈ R. The Gerstenhaber bracket in degree 2 is then [α, β] := α ◦ β + β ◦ α. In our setting, R = Sq(V ) ⋊ G, whose Hochschild cohomology we identify with the G-invariant subalgebra of HH q(S, S ⋊ G). We may use either the bar resolution or the Koszul resolution of S = Sq(V ) to gain information about this Hochschild cohomology. If α and β are given as cocycles on the Koszul resolution (4.1) instead of on the bar resolution (4.3), we apply the chain map ψ to convert α and β to functions on the bar resolution, compute the Gerstenhaber bracket of these functions, and then apply φ to convert back to a function on the Koszul resolution. Thus in this case, [α, β] := φ∗(ψ∗(α) ◦ ψ∗(β) + ψ∗(β) ◦ ψ∗(α)). Of course, the images of α and β may involve elements in S ⋊ G \ S, in which case we employ a standard technique to manage the group elements that appear in such a computation: Lemma 4.5. Let µ : Sq(V ) ⊗ Sq(V ) → Sq(V ) ⋊ G be a Hochschild 2-cocycle representing an element of HH2(Sq(V ), Sq ⋊ G). Then µ may be extended to a Hochschild 2-cocycle for Sq(V ) ⋊ G by defining µ(r#g, s#h) = µ(r, gs)gh for all r, s ∈ Sq(V ) and g, h ∈ G. Proof. This is standard; see, e.g. [17, Lemma 6.2]. Since the characteristic of k is 0, a bimodule resolution for Sq(V ) ⋊ G is given by tensoring (over k) a bimodule resolution for Sq(V ) with kG on one side (say the right). The action of G on the left is taken to be the semidirect product action. (cid:3) We are now ready to express the PBW conditions of Theorem 2.1 in terms of the Gerstenhaber algebra structure of Hochschild cohomology. The following theorem is similar to [17, Theorem 7.2]. It gives necessary and sufficient conditions for κL and κC to define a quantum Drinfeld orbifold algebra, in terms of their Gerstenhaber brackets. Theorem 4.6. The algebra Hq,κ, defined in (1.1), is a quantum Drinfeld orbifold algebra if and only if the following conditions hold: • κ is G-invariant. • κL is a cocycle, that is d∗κL = 0. • [κL, κL] = 2d∗κC as cochains. • [κC , κL] = 0 as cochains. Proof. We will show that conditions (1) -- (4) of Theorem 2.1 are equivalent to the four conditions stated in the theorem, respectively. We have already discussed the equivalence of the G-invariance condition with Theorem 2.1(1). Next note that d∗κL = κL ◦ d = 0 exactly when κL ◦ d3(vi ∧ vj ∧ vk) = 0 for all i, j, k, where d3(vi ∧ vj ∧ vk) =(vi ⊗ 1 − qijqik ⊗ vi) ⊗ vj ∧ vk − (qijvj ⊗ 1 − qjk ⊗ vj) ⊗ vi ∧ vk + (qikqjkvk ⊗ 1 − 1 ⊗ vk) ⊗ vi ∧ vj. 9 In other words, d∗κL = 0 when viκL(vj, vk) − qijqikκL(vj, vk)vi − qijvjκL(vi, vk) + qjkκL(vi, vk)vj + qikqjkvkκL(vi, vj) − κL(vi, vj)vk = 0 for all i < j < k. Multiply by qjiqkiqkj to obtain qjiqkiqkjviκL(vj, vk) − qkjκL(vj, vk)vi − qkjqkivjκL(vi, vk) + qjiqkiκL(vi, vk)vj + qjivkκL(vi, vj) − qjiqkiqkjκL(vi, vj)vk = 0. Using the relation κL(vl, vm) = −qlmκL(vm, vl), we may rewrite the equation as − qjiqkiviκL(vk, vj) + κL(vk, vj)vi − qkjqkivjκL(vi, vk) + qjiqkiκL(vi, vk)vj − vkκL(vj, vi) + qkiqkjκL(vj, vi)vk = 0. elements to the right, and apply the relation κL(vi, vk) = −qikκL(vk, vi). This is precisely Theorem 2.1(2), once we substitute κL(−, −) = Pg∈G κL Identify HomRe (R⊗(p+2), −) with Homk(R⊗p, −) and HomSe(Se⊗Vp(V ), −) with Homk(Vp(V ), −). If α, β ∈ HomSe(Se ⊗V2(V ), S ⋊ G)G, then by definition, g (−, −)#g, move group (α ◦ β)(vi ∧ vj ∧ vk) = (ψ∗(α) ◦ ψ∗(β))φ(vi ∧ vj ∧ vk). Applying (4.4), we thus have (α ◦ β)(vi ∧ vj ∧ vk) = (ψ∗(α) ◦ ψ∗(β))(vi ⊗ vj ⊗ vk − qijvj ⊗ vi ⊗ vk − qjkvi ⊗ vk ⊗ vj − qikqijqjkvk ⊗ vj ⊗ vi +qijqjkvj ⊗ vk ⊗ vi + qikqjkvk ⊗ vi ⊗ vj) = ψ∗(α)(β(vi ⊗ vj) ⊗ vk − vi ⊗ β(vj ⊗ vk)) − qijψ∗(α)(β(vj ⊗ vi) ⊗ vk − vj ⊗ β(vi ⊗ vk)) −qjkψ∗(α)(β(vi ⊗ vk) ⊗ vj − vi ⊗ β(vk ⊗ vj)) − qikqijqjkψ∗(α)(β(vk ⊗ vj) ⊗ vi − vk ⊗ β(vj ⊗ vi)) +qijqikψ∗(α)(β(vj ⊗ vk) ⊗ vi − vj ⊗ β(vk ⊗ vi)) + qikqjkψ∗(α)(β(vk ⊗ vi) ⊗ vj − vk ⊗ β(vi ⊗ vj)) = ψ∗(α)(β(vi ⊗ vj) ⊗ vk − qijβ(vj ⊗ vi) ⊗ vk) + ψ∗(α)(qij qikβ(vj ⊗ vk) ⊗ vi − qikqijqjkβ(vk ⊗ vj) ⊗ vi) +ψ∗(α)(qikqjkβ(vk ⊗ vi) ⊗ vj − qjkβ(vi ⊗ vk) ⊗ vj) − ψ∗(α)(vi ⊗ β(vj ⊗ vk) − qjkvi ⊗ β(vk ⊗ vj)) −ψ∗(α)(qij qikvj ⊗ β(vk ⊗ vi) − qijvj ⊗ β(vi ⊗ vk)) − ψ∗(α)(qikqjkvk ⊗ β(vi ⊗ vj) − qikqijqjkvk ⊗ β(vj ⊗ vi)) = ψ∗(α)(β(vi ∧ vj) ⊗ vk) + qijqikψ∗(α)(β(vj ∧ vk) ⊗ vi) + qikqjkψ∗(α)(β(vk ∧ vi) ⊗ vj) −ψ∗(α)(vi ⊗ β(vj ∧ vk)) − qijqikψ∗(α)(vj ⊗ β(vk ∧ vi)) − qikqjkψ∗(α)(vk ⊗ β(vi ∧ vj)) = ψ∗(α)(β(vi ∧ vj) ⊗ vk − qikqjkvk ⊗ β(vi ∧ vj)) +qijqikψ∗(α)(β(vj ∧ vk) ⊗ vi − qjiqkivi ⊗ β(vj ∧ vk)) +qikqjkψ∗(α)(β(vk ∧ vi) ⊗ vj − qijqkjvj ⊗ β(vk ∧ vi)). Now assume that κL is a G-invariant cocycle (in C 2) representing an element of HH q(A, A ⋊ G)G. Replacing α and β by κL in the above formula, we will get the left side of Theorem 2.1(3) for each 10 g ∈ G: 1 2 [κL, κL](vi ∧ vj ∧ vk) = (κL ◦ κL)(vi ∧ vj ∧ vk) = ψ∗(κL)(κL(vi, vj) ⊗ vk − qikqjkvk ⊗ κL(vi, vj)) +qijqikψ∗(κL)(κL(vj, vk) ⊗ vi − qjiqkivi ⊗ κL(vj, vk)) +qikqjkψ∗(κL)(κL(vk, vi) ⊗ vj − qijqkjvj ⊗ κL(vk, vi)) = ψ∗(κL)(Xh∈G κL h (vi, vj) ⊗ hvk#h − qikqjkvk ⊗Xh∈G κL h (vi, vj)#h) κL h (vj, vk) ⊗ hvi#h − qjiqkivi ⊗Xh∈G h (vk, vi) ⊗ hvj#h − qijqkjvj ⊗Xh∈G κL κL h (vj, vk)#h) κL h (vk, vi)#h) = h (vi, vj), hvk) − qikqjkκL(vk, κL h (vi, vj)) +qijqikψ∗(κL)(Xh∈G +qikqjkψ∗(κL)(Xh∈G 2 Xh∈G(cid:0)κL(κL 1 +qijqikκL(κL +qikqjkκL(κL h (vj, vk), hvi) − κL(vi, κL h (vk, vi), hvj) − qijqikκL(vj, κL h (vj, vk)) h (vk, vi))(cid:1)#h. Indeed, this agrees with half of the left side of Theorem 2.1(3), after rewriting κL as a sum, over g ∈ G, of κL gh−1 (for each h), and then considering separately each expression involving a fixed g = (gh−1)h. A similar calculation yields 2d∗κC equal to the right side of Theorem 2.1(3). Hence, Theo- rem 2.1(3) is equivalent to [κL, κL] = 2d∗κC . By again comparing coefficients of fixed g ∈ G, we see that Theorem 2.1(4) is equivalent to (cid:3) [κC , κL] = 0. References [1] Y. Bazlov and A. Berenstein, "Noncommutative Dunkl operators and braided Cherednik algebras," Selecta Math. 14 (2009), no. 3 -- 4, 325 -- 372. [2] R. Berger, "The quantum Poincar´e-Birkhoff-Witt Theorem," Commun. Math. Phys. 143 (1992), 215 -- 234. [3] I. Cherednik, "Double affine Hecke algebras and Macdonald's Conjectures," Ann. of Math. (2) 141 (1995), no. 1, 191 -- 216. [4] V. G. Drinfeld, "Degenerate affine Hecke algebras and Yangians," Funct. Anal. Appl. 20 (1986), 58 -- 60. [5] P. Etingof and V. Ginzburg, "Symplectic reflection algebras, Calogero-Moser space, and deformed Harish- Chandra homomorphism," Invent. Math. 147 (2002), no. 2, 243 -- 348. [6] M. Gerstenhaber, "The cohomology structure of an associative ring," Ann. Math. 78 (1963), 267 -- 288. [7] I. Gordon, "Rational Cherednik algebras," Proceedings of the International Congress of Mathematicians, Volume III, 1209 -- 1225, Hindustan Book Agency, New Delhi, 2010. [8] V. K. Kharchenko, "An algebra of skew primitive elements," Algebra and Logic 37 (1998), 101 -- 126. [9] V. Levandovskyy and A.V. Shepler, "Quantum Drinfeld Hecke algebras," Canad. J. Math. 66 (2014), no. 4, 874 -- 901. [10] G. Lusztig, "Affine Hecke algebras and their graded version," J. Amer. Math. Soc. 2 (1989), no. 3, 599-635. [11] D. Naidu, P. Shroff and S. Witherspoon, "Hochschild cohomology of group extensions of quantum symmetric algebras," Proc. Amer. Math. Soc. 139 (2011), 1553 -- 1567. [12] D. Naidu and S. Witherspoon, "Hochschild cohomology and quantum Drinfeld Hecke algebras," to appear in Selecta Mathematica. [13] B. Pareigis, "On Lie algebras in braided categories," in Quantum Groups and Quantum Spaces, Vol. 40, Banach Cent. Publ. (1997), 139 -- 158. 11 [14] T. Petit and F. Van Oystaeyen, "On the generalized enveloping algebra of a color Lie algebra," Algebr. Represent. Theor. 10 (2007), 367 -- 378. [15] A. Ram and A.V. Shepler, "Classification of graded Hecke algebras for complex reflection groups," Comment. Math. Helv. 78 (2003), 308 -- 334. [16] M. Scheunert, "Generalized Lie algebras," J. Math. Phys. 20 (1979), no. 4, 712 -- 720. [17] A.V. Shepler and S. Witherspoon, "Drinfeld orbifold algebras," Pacific J. Math. (2012) 259-1:161 -- 193. [18] A.V. Shepler and S. Witherspoon, "A Poincare-Birkhoff-Witt Theorem for quadratic algebras with group actions," Trans. Amer. Math. Soc. 366 (2014), no. 12, 6483 -- 6506. [19] A.V. Shepler and S. Witherspoon, "PBW deformations of skew group algebras in positive characteristic," to appear in Algebras and Representation Theory. [20] R. Sridharan, "Filtered algebras and representations of Lie algebras," Trans. Amer. Math. Soc. 100 (1961), 530 -- 550. [21] P. Shroff, "Quantum Drinfeld orbifold algebras," Comm. Algebra 43 (2015), no. 4, 1563 -- 1570. [22] D. S¸tefan, "Hochschild cohomology on Hopf Galois extensions," J. Pure Appl. Algebra 103 (1995), 221 -- 233. [23] M. Wambst, "Complexes de Koszul quantiques," Ann. Fourier 43 (1993), no. 4, 1089 -- 1156. [24] S. L. Woronowicz, "Differential calculus on compact matrix pseudogroups (quantum groups)," Comm. Math. Phys. 122 (1989), no. 1, 125 -- 170. E-mail address: [email protected] Department of Mathematics, Texas State University, San Marcos, Texas 78666, USA E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, Texas 77843, USA 12
1801.09020	2	1801	2019-05-18T05:43:27	Auslander's Theorem for group coactions on noetherian graded down-up algebras	[ "math.RA" ]	We prove a version of a theorem of Auslander for finite group coactions on noetherian graded down-up algebras.	math.RA	math	AUSLANDER'S THEOREM FOR GROUP COACTIONS ON NOETHERIAN GRADED DOWN-UP ALGEBRAS J. CHEN, E. KIRKMAN AND J.J. ZHANG Abstract. We prove a version of a theorem of Auslander for finite group coactions on noetherian graded down-up algebras. 0. Introduction Maurice Auslander [3] proved that if G is a finite subgroup of GLn(k), con- taining no pseudo-reflections (e.g. subgroups of SLn(k)), acting linearly on the commutative polynomial ring A = k[x1, . . . , xn], with fixed subring AG, then the natural map from the skew group algebra A ∗ G to EndAG (A) is an isomorphism of graded algebras. This theorem is the main ingredient in the McKay correspon- dence, relating representations of G and AG-modules. Noncommutative versions of this theorem of Auslander [4, 5] are an important ingredient in establishing a noncommutative McKay correspondence. One of the main open questions concern- ing a noncommutative version of Auslander's theorem is the following conjecture that was stated in [4, Conjecture 0.4] and [9, Conjecture 0.2], where the condition that the homological determinant of the H-action is trivial generalizes the result for group actions by subgroups of SLn(k): Let A be a connected graded noetherian Artin-Schelter regular algebra [1] and H be a semisimple Hopf algebra acting on A inner-faithfully and homogeneously. If the homological determinant of the H-action on A is trivial, then there is a natural graded algebra isomorphism A#H ∼= EndAH (A). By [9, Theorem 0.3] the above conjecture holds when A has global dimension two, which is one of the main results in [9]. It is natural to search for a proof of this conjecture for global dimension three (or higher). The paper [5] started this program by showing that the above conjecture holds for certain finite group actions on noetherian graded down-up algebras, which are Artin-Schelter regular algebras of global dimension three [5, Theorem 0.6]. Some interesting partial results concerning Auslander's theorem have been proven in [4, 5, 12, 13, 21]. The goal of this paper is to verify the conjecture for finite group coactions on Artin-Schelter regular down-up algebras [Theorem 0.1]. The idea of the proof is to use the pertinency introduced in [4] that has been one major tool for proving the noncommutative Auslander theorem. 2010 Mathematics Subject Classification. 16E10, 16E65, 16S40, 16T05, 16W22. Key words and phrases. Noetherian graded down-up algebra, group coaction, Auslander's theorem, trivial homological (co)determinant, pertinency. 1 2 J. CHEN, E. KIRKMAN AND J.J. ZHANG Throughout the paper, let k be a base field of characteristic zero, and all objects are over k. Down-up algebras were introduced in 1998 by Benkart-Roby in [6], and, since then, these algebras have been studied extensively. Noetherian graded down-up algebras are Artin-Schelter regular algebras of global dimension three with two generators by a result of [20]. Let α and β be two scalars in k. The graded down- up algebra, denoted by D(α, β), is generated by two elements d and u and subject to two relations (E0.0.1) (E0.0.2) d2u = αdud + βud2, du2 = αudu + βu2d. This algebra is noetherian if and only if β 6= 0, and in this paper we always assume that β 6= 0. When α = 0, we use Dβ instead of D(0, β). The groups of graded algebra automorphisms of the down-up algebras were computed in [15]. Recently, the invariant theory of graded down-up algebras under finite group actions and coactions has been studied in [17, 11, 13]. In a general setting, let H be a semisimple Hopf algebra and let K be its k- linear dual. Then K is also a semisimple Hopf algebra. It is well-known that a left H-action on an algebra A is equivalent to a right K-coaction on A. Suppose H is a semisimple Hopf algebra with integral R , and A is an algebra with GKdim A < ∞. Here GKdim A denotes the Gelfand-Kirillov dimension of A. If H acts on A, by [4, Definition 0.1], the pertinency of the H-action on A is defined to be (E0.0.3) p(A, H) = GKdim A − GKdim((A#H)/I) where I is the 2-sided ideal of A#H generated by 1#R . Define the fixed subring of the H-action to be AH = {a ∈ A h · (a) = ǫ(h)a, ∀h ∈ H} where ǫ is the counit of H. For any algebra A with H-action, there is a natu- ral algebra homomorphism φ : A#H → EndAH (A) which sends a#h to an AH - endomorphism of A: φ(a#h) : x 7→ a(h · (x)), ∀ x ∈ A. By [4, Theorem 0.3], if A is a noetherian, connected graded, Artin-Schelter regular and Cohen-Macaulay domain of GKdim ≥ 2, then p(A, H) ≥ 2 if and only if the canonical map (E0.0.4) φ : A#H −→ EndAH (A) is an isomorphism. For simplicity, if φ is an isomorphism, we say that (A, H) has the isom-property. In this paper we are interested in the case when H is kG := Homk(kG, k), or equivalently, K is the group algebra kG for some finite group G, and when A is the noetherian graded down-up algebra D(α, β). Our main result is Theorem 0.1. Let H := kG act on A := D(α, β) homogeneously and inner faith- fully, where β 6= 0. If the action has trivial homological determinant, then the pertinency p(A, H) ≥ 2. As a consequence, Auslander's theorem holds, namely, there is a natural isomorphism of graded algebras φ : A#H ∼= EndAH (A). AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 3 Theorem 0.1 fails without the hypothesis of "trivial homological determinant", see Remark 1.6(2). Theorem 0.1 suggests there is a McKay correspondence for down-up algebras D(α, β); it follows from [9, Theorem A] that when Auslander's theorem holds, there are bijections between several categories of modules, e.g. sim- ple left H-modules and indecomposable direct summands of A as a left AH -modules. The paper [22] shows that whenever Auslander's theorem holds one can view A#H as a generalized noncommutative crepant resolution (NCCR) of AH , and when AH is a central subalgebra of A#H, A#H is an NCCR of AH . The paper is organized as follows: Section 1 contains some preliminary results, Section 2 contains the proof of Theorem 0.1, and Section 3 contains some examples. 1. Preliminaries In this section we recall some basic definitions and make some comments. We will omit the definition of Artin-Schelter Gorensteinness and Artin-Schelter regularity [1] since these can be found in many other papers and we will not need these in the proof of the main result. As mentioned in the introduction, noetherian graded down-up algebras are Artin-Schelter regular of global dimension three. We introduce a temporary concept. For a graded module C and an integer w, the wth shift of C, denoted by C(w), is defined by C(w)m = Cw+m for all m ∈ Z. Definition 1.1. Let H be a semisimple Hopf algebra acting on a connected graded algebra A homogeneously and inner faithfully. Decompose A into (E1.1.1) ( sMi=1 AH (−wi))M B as a right AH -module for some integer s ≥ 1, where B has no direct summand that is isomorphic to AH (w) for some integer w. If B = 0, H is called a reflection Hopf algebra with respect to A. If s ≥ 2 (but B 6= 0), we say H is a fractional- reflection Hopf algebra with respect to A, since part (but not all) of A is a graded free AH -module. Lemma 1.2. Suppose H acts on a connected graded algebra A as a reflection (or fractional-reflection) Hopf algebra. Then: (1) (A, H) does not have the isom-property. (2) If A is a noetherian Artin-Schelter Gorenstein algebra, then the H-action on A does not have trivial homological determinant. Proof. (1) Since H is a fractional-reflection Hopf algebra, s ≥ 2 in (E1.1.1). We write A = AH ⊕ AH (−w2) ⊕ C where C is a right AH -module. Note that w2 is necessarily positive since A is connected graded. There is a homogeneous AH - module map of degree −w2: pr −−−−−−−→ AH (−w2) AH (−w2 ) A shif t by degree w2 −−−−−−−−−−−−→ AH inclusion −−−−−−→ A. Then EndAH (A) has a nonzero element of negative degree. On the other hand, every nonzero homogeneous element in A#H has nonnegative degree. Therefore A#H 6∼= EndAH (A). (2) We now assume that A is noetherian and Artin-Schelter Gorenstein. If the H-action on A has trivial homological determinant, then, by [16, Theorem 3.6] and the proof of [16, Lemma 3.5(d)], we have 4 J. CHEN, E. KIRKMAN AND J.J. ZHANG (a) AH is noetherian and Artin-Schelter Gorenstein, (b) injdim A = injdim AH =: d, and (c) the AS indices of A and AH are the same, denoted by ℓ. Let m be the graded maximal ideal of AH . We consider the local cohomology RdΓm(A)∗ as in [2, 16]. Since AH (−w2) is a direct summand of A (as a right AH - module), RdΓm(AH (−w2))∗ is a direct summand of RdΓm(A)∗. If both A and AH are Artin-Schelter Gorenstein, by [23, Lemma 3.5], RdΓm(A)∗ ∼= A(−ℓ) and RdΓm(AH (−w2))∗ ∼= AH (−ℓ + w2). The lowest degree of nonzero element in RdΓm(AH (−w2))∗ is ℓ − w2 and the lowest degree of nonzero element in RdΓm(A)∗ is ℓ. Since w2 is positive, this is impossible. Therefore the H-action on A does not have trivial homological determinant. (cid:3) Remark 1.3. Lemma 1.2(2) is a generalization of [8, Theorem 2.3]. The definition of maximal Cohen-Macaulay modules was extended to this context in [10, Definition 3.5]. Proposition 1.4. Let A be connected graded and suppose that (A, kG) has the isom-property. Write A =Lg∈G Ag. If g 6= h, then Ag is not isomorphic to Ah(w) for any w ∈ Z. As a consequence, if A is noetherian Artin-Schelter Gorenstein, there are at least G non-isomorphic graded maximal Cohen-Macaulay modules over Aco G, up to degree shift. Proof. Let B = Aco G. Suppose to the contrary that Ag ∼= Ah(w) for some g 6= h. If w 6= 0, then EndB(A) has an element of negative degree. So A#kG 6∼= EndB(A), a contradiction. If w = 0, then the degree zero part of EndB(A) contains a 2 × 2 matrix algebra which is not commutative. However the degree 0 part of A#kG is kG, which is commutative. Therefore A#kG 6∼= EndB(A), a contradiction. Therefore Ag is not isomorphic to Ah(w) if g 6= h. The consequence is clear. (cid:3) The homological (co)determinant is defined in [16]. We need some facts about the homological (co)determinant of group coactions on down-up algebras. Suppose that Dβ is G-graded with degG d = g1 and degG u = g2 (or equivalently, G coacts on Dβ). Assume that the G-coaction on Dβ is inner-faithful, which is equivalent to the condition that G is generated by g1 and g2, in this case. Lemma 1.5. Retain the above notation. The homological (co)determinant of the kG-action (or G-coaction) on Dβ is g2 2 = 1, where 1 is the unit of G. 2, and is trivial if and only if g2 1g2 1g2 Proof. Let A = Dβ. Since G coacts on A homogeneously, A is a Z × G-graded algebra. Recall that Dβ is generated by d and u subject to relations d2u = βud2, du2 = βu2d. By using the generators and relations of A, one checks that the G-graded resolution of the trivial A-module k is 2 ) → A(g−1 2 ) → A → k → 0. 2 ) → A(g−1 1 ) ⊕ A(g−1 0 → A(g−2 1 g−2 1 g−2 2 ) ⊕ A(g−2 1 g−1 Using this resolution to compute the Ext-group, one sees that Ext3 1g2 2) as a G-graded vector space. Hence the G-coaction maps a basis element e ∈ A(k, k) ∼= k(g2 AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 5 A(k, k) to e ⊗ g2 Ext3 coaction is g2 1g2 2. By definition, the homological codeterminant of the G- (cid:3) 1g2 2. The assertion follows. Next we make some comments about [11, Example 2.1]. Remark 1.6. Consider the algebra D := D1 as in [11, Example 2.1]. (1) By [5, Theorem 0.6], if H = kG for any finite group G acting on D, then p(D, G) ≥ 2 and D ∗ G ∼= EndDG (D), so that Auslander's theorem holds for group actions on D; this result was expected because all finite groups acting on D are "small", since they have no reflections, in a sense made precise in [17]. But, when H = kG as in [11, Example 2.1 and Lemma 2.2], H is a fractional-reflection Hopf algebra with respect to D, so by Lemma 1.2(1), (D, H) does not have the isom-property, namely, Auslander's theorem fails. By [4, Theorem 0.3], p(D, kG) ≤ 1, and by Lemma 1.2(2), the kG-action does not have trivial homological determinant. Hence group actions behave differently from group coactions. (2) By [16, Corollary 4.11] or [17, Proposition 0.2(2)], if H = kG for some finite group G, then DG is Gorenstein if and only if the G-action on D has trivial homological determinant. This result was expected because, again, these groups contain no reflections of D. However, [11, Example 2.1] shows that when H = kG, this statement fails, namely, Dco G is Gorenstein, but the kG-action does not have trivial homological determinant. This result is surprising, and there might be a relationship between the facts in parts (1) and (2). (3) Theorem 0.1 implies that if the kG-action on D has trivial homological determinant, then p(D, kG) ≥ 2 and D#kG ∼= EndDco G (D). (4) In the commutative case, when a semisimple Hopf algebra acts on a poly- nomial ring A := k[x1, · · · , xn], Auslander's theorem fails if and only if there is a nontrivial Hopf subalgebra H0 ⊆ H (in this case, H and H0 are group algebras kG and kG0 respectively for some G0 ⊆ G) such that AH0 is Artin-Schelter regular; this happens if and only if G contains a reflection, or equivalently, G is not small. Recall that a finite subgroup G of GLn(k) is small if it does not contain any reflections. Hence one might conjecture that Auslander's holds for a semisimple Hopf algebra if and only if there is no such Hopf subalgebra, and that this definition is the generalization for Hopf algebras of the notion of a "small group". However, [11, Example 2.1], where H = kG, shows that this definition of an analogue of a "small subgroup" does not work, since in this example Auslander's Theorem fails, but as one can easily check, or use [17, Proposition 0.2(2)], that there is NO nontrivial Hopf subalgebra H0 ⊆ H such that DH0 is Artin-Schelter regu- lar. So it is not clear how to generalize Auslander's theorem beyond our noncommutative analogue of subgroups of SLn(k) (namely H-actions with trivial homological determinant) to a noncommutative analogue for Hopf algebras of the notion of "small" groups (groups containing no reflections). Question 1.7. For actions (and coactions) by semisimple Hopf algebras H on Artin-Schelter regular algebras A, is there an analogue of the action on k[x1, . . . , xn] by a finite "small" subgroup of GLn(k) (a condition for Hopf actions with non-trivial homological determinant for which Auslander's Theorem holds)? 6 J. CHEN, E. KIRKMAN AND J.J. ZHANG To prove Theorem 0.1, we only need to show that p(D(α, β), kG) ≥ 2. The pertinency p(A, H) is defined in (E0.0.3). Let R be the integral of a semisimple Hopf algebra H, and I be the two-sided ideal of A#H generated by 1#R . Recall that a kG-action on an algebra A is equivalent to a G-grading on A. We recall the following result from [4] that will be used in the pertinency com- putation. Lemma 1.8. Let H := (kG)◦ act on A inner faithfully, and write A = ⊕g∈GAg. (1) [4, Lemma 5.1 (3)] If f ∈ ∩g∈GAAg then f #1 ∈ I. (2) [4, Lemma 5.1 (6)] p(R, H) ≥ d − GKdim A/(∩g∈GAAg) ≥ d − max{GKdim A/AAgg ∈ G}. The following is a modification of [4, Lemma 5.1(4)]. Lemma 1.9. Let G be a finite group and kG act on A inner-faithfully and homo- geneously. Let z ∈ A. (1) Suppose that, for each g ∈ G, there is an x ∈ A of G-degree g and y ∈ A such that z = yx. Then z#1 is in the ideal of A#kG generated by e := 1#R . (2) Suppose z = fn · · · f1 is such that the collection (with possible repetitions) {1, degG(f1), degG(f2f1), · · · , degG(fn−1 · · · f2f1), degG(z)} includes all elements in G. Then z#1 is in the ideal of A#kG generated by e := 1#R . Proof. (1) Since z = yx ∈ AAg for each g, we have z ∈ Tg∈G AAg. The assertion follows from Lemma 1.8. (2) This is a special case of part (1). (cid:3) In the next lemma we use some arguments from Bergman's Diamond Lemma [7]. Recall that D(α, β) is generated by d and u. We use the ordering d < u in this paper. Two relations of D(α, β), namely, (E0.0.1)-(E0.0.2) can be written as ud2 = lower terms, u2d = lower terms where "lower terms" stands for a linear combination of monomials that have lower degree (in the lexicographic order) than the terms explicitly appearing in the same equation. Lemma 1.10. Retain the above notation. (1) Let W be an ideal of D(α, β) such that, in the factor ring D(α, β)/W , there are relations ds(ud)i = lower terms, ut = lower terms for some i, s, t ≥ 0. Then GKdim D(α, β)/W ≤ 1. (2) Let W be an ideal of Dβ such that, in the factor ring Dβ/W , there are relations d2s(du)i = lower terms, (ud)j u2t = lower terms for some i, j, s, t ≥ 0. Then GKdim Dβ/W ≤ 1. AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 7 Proof. (1) Together with (E0.0.1)-(E0.0.2), we have at least four relations ud2 = lower terms, u2d = lower terms, ds(ud)i = lower terms, ut = lower terms in the factor ring D(α, β)/W . By the Diamond Lemma [7] and using the first two relations, D(α, β)/W has a k-linear basis consisting of monomials of the form da(ud)buc, a, b, c ≥ 0 with some constraints. (A similar statement is [11, Lemma 1.1(3)] where we use the order u < d.) Two of the constraints are (i) either a < s or b < i and (ii) c < t, which follows from the last two relations of D(α, β)/W . Therefore, for each N-degree d, the k-dimension of (D(α, β)/W )d is uniformly bounded. As a consequence of a Gelfand-Kirillov dimension computation [5, (E1.1.6)], GKdim D(α, β)/W ≤ 1. (2) The proof is similar to the one of part (1) and uses the fact that d2 and u2 are normal elements of Dβ. Without loss of generality, we can assume that s = t = i = j =: a > 0 and re-use the letters i and j. Let d2a(du)a = lower terms, (ud)au2a = lower terms in Dβ/W . Then d4au4a = λ(d2a(du)a)((ud)au2a) = lower terms in Dβ/W , for some λ ∈ k. Then d4au4a = d4a+1f in Dβ/W for some f . Since u2 skew-commutes with d and u, we obtain that or Therefore d4a(ud)ju4a = d4a+1f ′ d4a(ud)j u4a = lower terms. di(ud)juk = lower terms in Dβ/W when at least two of indices i, j, k are larger than 4a. By the Diamond Lemma argument as in the proof of part (1), for each N-degree d, the k-dimension of (Dβ/W )d is uniformly bounded. By [5, (E1.1.6)], GKdim Dβ/W ≤ 1. (cid:3) 2. Proof of Theorem 0.1 In this section we prove the main result, the theorem of Auslander for group coactions on down-up algebras. First we recall a result from [11]. Let F be the algebra generated by x and y, subject to two relations (E2.0.1) x3 = yxy and y3 = xyx. As a graded algebra, F is isomorphic to D−1 [11, Lemma 1.5(1)]. Let H be the algebra generated by x and y, subject to two relations (E2.0.2) x2y + yx2 − 2y3 = 0 and − 2x3 + xy2 + y2x = 0. Then, as a graded algebra, H is isomorphic to D(−2, −1) [11, Lemma 1.9(1)]. 8 J. CHEN, E. KIRKMAN AND J.J. ZHANG Lemma 2.1. [11, Proposition 1.12] Suppose G is a finite non-cyclic group coacting on A := D(α, β) homogeneously and inner-faithfully. Then one of the following occurs. (1) α = 0 and u and d are G-homogeneous after a change of variables. (2) A is isomorphic to F and using the generators of F, both x and y are G- homogeneous. (3) A is isomorphic to H and using the generators of H, both x and y are G-homogeneous. (4) G is abelian and there are linearly independent elements x and y of D(α, −1) of degree one such that αx2y + (−2 − α)xyx + αyx2 + (2 − α)y3 = 0, (2 − α)x3 + αxy2 + (−2 − α)yxy + αy2x = 0 and x and y are G-homogeneous. (5) G is abelian and u and d are G-homogeneous after a change of variables. The above lemma shows that there are plenty of interesting examples of finite group coactions on noetherian down-up algebras. Note that the hypothesis of "G being non-cyclic" is needed in the above lemma which was proved in [11]. In the present paper we will also consider cyclic cases. In particular, our main theorem does not need the hypothesis of "G being non-cyclic". We separate the proof of Theorem 0.1 into subcases according to the above lemma. In Cases 1 and 2 we assume that G is not cyclic; the cyclic cases will be included in Case 3. 2.1. Case 1: α = 0, u and d are G-homogeneous. In this subsection, as α = 0, A is the down-up algebra Dβ := khd, ui/(d2u − βud2, du2 − βu2d), β ∈ k×. Suppose that Dβ is G-graded with degG d = g1 and degG u = g2. Since Dβ is generated by d and u, G is generated by g1 and g2. Let (E2.1.1) and (E2.1.2) X1 := {(g2g1)i i ≥ 0} ∪ {g1(g2g1)i i ≥ 0} ⊆ G X2 := {(g1g2)i i ≥ 0} ∪ {g2(g1g2)i i ≥ 0} ⊆ G. As in [4, Lemma 5.1] let (E2.1.3) J := the ideal of A generated by \g∈G AAg when a group G coacts on A. Lemma 2.2. Suppose that hg1iX1 = G = hg2iX2. Then p(Dβ, kG) ≥ 2. Proof. Let A = Dβ. By Lemma 1.8 (2) it suffices to show that GKdim A/J ≤ 1 where J is defined as in (E2.1.3). By Lemma 1.10(2), it suffices to show that v := d2a(du)a+1 and w := u2a(ud)a are in the ideal J, where a = G. By symmetry, we show only that v is in J. Since v = dd2a(ud)au, it suffices to show that f := d2a(ud)a is in J. By hypoth- 1(g2g1)j for some a ≥ i, j ≥ 0. esis hg1iX1 = G, every element g in G is of the form gi AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 9 Since d2 is normal, we can write f as c(ud)a−jd2a−i · di(ud)j, for some c ∈ k× with degG di(du)j = gi 1(g2g1)j = g. Then f ∈ AAg for all g, which implies that f ∈ J as required. (cid:3) Lemma 2.3. Suppose G is generated by g1, g2 and hg2 G = hg2iX2. 1i = hg2 2i. Then hg1iX1 = Proof. Let N be the normal subgroup of G generated by g2 2. Then G/N is a dihedral group D2n. In this case the image of X1 in G/N consists of all elements in G/N . Then G = N X1. Under the hypothesis, we have N = hg2 1i. Hence hg1iX1 = G. By symmetry, G = hg2iX2. (cid:3) 1 and g2 Now we are ready to prove a part of Theorem 0.1. Proposition 2.4. Retain the notation as in Theorem 0.1. Suppose further that α = 0 and u and d are G-homogeneous. Then p(A, H) ≥ 2. Proof. By Lemma 1.5, when the kG-action on Dβ has trivial homological determi- nant, g2 2i. By Lemma 2.3, hg1iX1 = G = hg2iX2. Now the main assertion follows from Lemma 2.2. (cid:3) 2 . Hence hg2 1 = g−2 1i = hg2 2.2. Case 2: A = H, x and y are G-homogeneous. In this subsection we have that A = H and that x and y in H are G-homogeneous. Let g1 = degG x and g2 = degG y. By the relations of H, one sees that g2 2. Two relations of H can be written as 1 = g2 x(x2 − y2) = −(x2 − y2)x and y(x2 − y2) = −(x2 − y2)y. Define a filtration F on H by FiH = (kx + ky + kz)i, i ≥ 0 where z = x2 − y2. It is easy to see that the G-coaction preserves this filtration. Let B be grF H. Then B ∼= (khx, yi/(x2 − y2))[z, σ] where σ maps x → −x and y → −y. Then G coacts on B by degG x = g1, degG y = g2 and degG z = g2 1. The following lemma follows from [4, Lemma 3.6]. Lemma 2.5. Retain the above notation. Then p(H, kG) ≥ p(B, kG). By the above lemma, it suffices to show that p(B, kG) ≥ 2. For the rest of the proof we follow the proof in Case 1. Lemma 2.6. Let J be an ideal of B containing both x2s(yx)i and (xy)jzt for some i, j, s, t ≥ 0. Then GKdim B/J ≤ 1. Proof. Without loss of generality, we can assume that s = t = i = j =: a > 0 and re-use letters i and j. Let f1 = x2a(yx)a and f2 = (xy)aza in J. Then x6aza ∈ kf1f2 ⊆ J. Note that B has a k-linear basis {xi(yx)j zk i, j, k ≥ 0} ∪ {xi(yx)j zky i, j, k ≥ 0}. Since x2(= y2) and z are skew-commuting with x, y, z, every element is of the form xi(yx)jzk or xi(yx)jzky is 0 in B/J when at least two of indices i, j, k are larger than 6a. An elementary counting argument shows that GKdim B/J ≤ 1. (cid:3) 10 J. CHEN, E. KIRKMAN AND J.J. ZHANG Use the notation introduced in (E2.1.1) and (E2.1.2): X1 := {(g2g1)i i ≥ 0} ∪ {g1(g2g1)i i ≥ 0} ⊆ G and X2 := {(g1g2)i i ≥ 0} ∪ {g2(g1g2)i i ≥ 0} ⊆ G. Lemma 2.7. Retain the above notation. 1iX1 = G = hg2 (1) hg2 (2) p(B, kG) ≥ 2. (3) p(H, kG) ≥ 2. 2iX2. Proof. (1) Let N be the normal subgroup of G generated by g2 2). Then G/N is a dihedral group D2n. In this case the image of X1 in G/N consists of all elements in G/N . Then G = N X1 = hg2 1iX1. Similarly, G = hg2 1 (or by g2 2iX2. (2) By Lemma 1.8(2) it suffices to show that GKdim B/J ≤ 1 where J is the ideal of B generated by Tg∈G BBg. By Lemma 2.6, it suffices to show that f1 := x2a(yx)a and f2 := (xy)aza are in the ideal J, where a = G. 1(g2g1)j for some By part (1), hg1iX1 = G, every element in G is of the form gi 0 ≤ i, j ≤ a. By the fact that x2 commutes with y, we obtain that f1 = f ′ 1(xi(yx)j ) for some f ′ 1 ∈ B. Then f1 ∈ BBg for all g, which implies that f1 ∈ J. Since z skew-commutes with x and y, a similar argument shows that f2 ∈ J. Now the assertion follows by Lemma 2.6. (3) This follows from part (2) and Lemma 2.5. (cid:3) Part (3) of the above lemma says that Auslander's theorem holds in this case, even without the hypothesis that the homological determinant of the H-action is trivial in this special case. For the sake of completeness we calculate the homological (co)determinants of the G-coactions easily in the next lemma. Lemma 2.8. Suppose a finite group G coacts on A. (1) If A = H and x and y are G-homogeneous with G-degree g1 and g2 respec- 1, which is tively, then the homological codeterminant of the G-coaction is g4 also g4 2. (2) If A = F and x and y are G-homogeneous with G-degree g1 and g2 respec- 1, which is tively, then the homological codeterminant of the G-coaction is g4 also g4 2. (3) Let A = D(α, −1), for α 6= 2, and x = 1 2 (d + u) and y = 1 2 (d − u). By [11, Proposition 1.12(4)], A is generated by x and y and subject to relations (E2.8.1) (E2.8.2) αx2y + (−2 − α)xyx + αyx2 + (2 − α)y3 = 0, (2 − α)x3 + αxy2 + (−2 − α)yxy + αy2x = 0. Suppose that G is abelian and that x and y are G-homogeneous with G- degree g1 and g2 respectively. Then the homological codeterminant of the G-coaction is g4 1, which is also g4 2. Proof. Since the proofs are similar to the proof of Lemma 1.5, the details are omitted. (cid:3) AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 11 isomorphic to kG as a Hopf algebra. character group Homgroups(G, k×). Since k is algebraically closed of characteristic 2.3. Case 3: G is abelian. Let G be a finite abelian group and let bG be the zero, bG is isomorphic to G as an abstract group. As a consequence, (kG)∗ is Let A be a down-up algebra D(α, β) generated by d and u. Every graded al- gebra automorphism g of A can be written as a 2 × 2-matrix with respect to the basis {d, u}. We say g is diagonal (respectively, non-diagonal) if its matrix pre- sentation with respect to {d, u} is diagonal (respectively, non-diagonal). When the basis {d, u} is replaced by {d′, u′} = {c1d, c2u} for some c1, c2 ∈ k×, the matrix presentation of g could change accordingly, but the diagonal property of g will not change. We call this kind of change of basis a scalar base change which we use in the proof of Lemma 2.9. Let G be a finite abelian group that coacts on A inner-faithfully and homoge- neously. This G-coaction on A is equivalent to a bG-action on A preserving the N-grading. Therefore we can consider the bG-action instead of the G-coaction. The theorem of Auslander was proved for finite group actions on graded noetherian down-up algebras in [5, Theorem 0.6] except for the case A = D(α, −1) for α 6= 2. In fact their proof [5, Proof of Theorem 0.6] works for any diagonal automorphisms of D(α, −1), too, and [13, Proposition 4.6] handles another special class of groups acting on A = D(α, −1) for α 6= 2. In this subsection we prove Auslander's theo- rem only for a finite abelian group bG of graded automorphisms of D(α, −1) with α 6= 2 that is not all diagonal. Combining with the results in [5], we take care of all abelian groups (including cyclic ones). Throughout the rest of this subsection let A be D(α, −1) for some α 6= 2. The next lemma classifies all possible finite abelian groups that are not diagonal having trivial homological determinant. Lemma 2.9. Consider the following subgroup of GL2(k) T =(cid:26)(cid:18)a 0 0 a(cid:19) ,(cid:18)0 b b 0(cid:19) : a, b ∈ {±1, ±i}(cid:27) . The following hold. (1) T is an abelian group acting naturally on A, with respect to the basis {d, u}, inner-faithfully and homogeneously with trivial homological determinant. following. (2) Let bG be a finite abelian group acting on A inner-faithfully and homoge- neously with trivial homological determinant. If bG contains a non-diagonal matrix, then bG is a subgroup of T after a scalar base change. (3) Let bG be as in part (2). Then, up to a scalar base change, bG is one of the (cid:26)(cid:18)1 0 0 1(cid:19) ,(cid:18)0 1 0(cid:19) : a, b ∈ {±1}(cid:27) , (cid:26)(cid:18)a 0 0 a(cid:19) ,(cid:18)0 0(cid:19) : a ∈ {±1}, b ∈ {±i}(cid:27) , 1 0(cid:19)(cid:27) , (cid:26)(cid:18)a 0 0 a(cid:19) ,(cid:18)0 b b b b or T. Proof. (1) This follows by a direct computation. 12 J. CHEN, E. KIRKMAN AND J.J. ZHANG b c 0 d(cid:19) and g :=(cid:18)0 (2) Suppose that f :=(cid:18)a 0 0(cid:19) are in bG. The commutativity of G forces a = d. By [15, Theorem 1.5], the homological determinant of (cid:18)a 0 0 a(cid:19) is a4. Thus a ∈ {±1, ±i} as bG has trivial homological determinant. In other words, Then b ∈ {±1, ±i} and g ∈ T . Now assume that bG contains another non-diagonal automorphism h := (cid:18)0 c′ 0(cid:19). Then the equation gh = hg implies that c′ = c. So h ∈ T and bG is a subgroup of T . f ∈ T . After a scalar base change, we may assume that b = c in the matrix g. By [15, Theorem 1.5], the homological determinant of g (with b = c) is b4. (3) This follows by a direct computation. (cid:3) c Using the classification in Lemma 2.9, we can work out the corresponding coac- 2 (d − u), or equivalently, d = x + y and u = x − y. tions. Let x = 1 By the proof of [11, Proposition 1.12(4)], we have the following. 2 (d + u) and y = 1 Lemma 2.10. Suppose G is a finite abelian group coacting on A := D(α, −1), for α 6= 2, such that (a) the G-coaction has trivial homological codeterminant, and (b) the corresponding bG-action contains a non-diagonal matrix with respect to the basis {d, u}. Then the following hold. (1) There are linearly independent elements x and y of D(α, −1) of degree one such that αx2y + (−2 − α)xyx + αyx2 + (2 − α)y3 = 0, (2 − α)x3 + αxy2 + (−2 − α)yxy + αy2x = 0 and x and y are G-homogeneous. (2) Let degG x = g1, degG y = g2. Then g2 1 = g2 2 and g4 1 = g4 2 = 1 in G. Proof. (1) A part of the proof appeared in the proof of [11, Proposition 1.12], so we give only a sketch of the argument here. the forms given there, let x = 1 First, the bG-action on A has the special forms as listed in Lemma 2.9(3). Using u = x − y. Then both x and y are bG-eigenvectors. This means that both x and y are G-homogeneous in the corresponding G-coaction. The two relations are obtained in the proof of [11, Proposition 1.12] by direct computation, which we will not repeat here. 2 (d − u), or d = x + y and 2 (d + u) and y = 1 (2) By the relations and the hypothesis that α 6= 2, one sees that g2 1 = g2 2. The (cid:3) second assertion is Lemma 2.8(3). Ueyama [24] introduced the notion of a graded isolated singularity, and we recall his definition here. For a graded algebra A, let grmod A denote the category of finitely generated graded left A-modules. For a graded finitely generated A-module an element x ∈ M is called torsion if there exists a positive integer n such that A≥nx = 0. The module M is called a torsion module if every element of M is torsion. Let tors A denote the full subcategory of grmod A consisting of torsion modules. We can then define the quotient category tails A = grmod A/ tors A. AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 13 Following [24], we say that AG has a graded isolated singularity if gldim(tails AG) < ∞. Mori and Ueyama prove that if the Auslander map is an isomorphism, then AG has a graded isolated singularity if and only if A#G/I is finite-dimensional [21, Theorem 3.10]. Examples of graded isolated singularities are of particular interest, since when AG has a graded isolated singularity, the category of graded CM AG-modules has several nice properties (see [25]). Next we compute the pertinency for G-coactions. Lemma 2.11. Retain the hypothesis of Lemma 2.10. (1) If g2 = 1 and g1 6= 1 then p(A, kG) = 3. As a consequence, Aco G has a graded isolated singularity. (2) If g2 6= 1, g1 6= 1, g1 6= g2, and g2 (3) If g1 6= 1, g2 1 = 1 and g2 = g1, then p(A, kG) = 3. As a consequence, Aco G 1 = g2 2 = 1, then p(A, kG) ≥ 2. has a graded isolated singularity. (4) If g2 1 6= 1, g2 = g1, then p(A, kG) = 3. As a consequence, Aco G has a graded isolated singularity. 1 6= 1, and g2 = g−1 (5) If g2 (6) If G = T , then p(A, kG) ≥ 2. 1 , then p(A, kG) ≥ 2. Proof. Let J be the ideal generated by Tg∈G AAg as defined in (E2.1.3). (1) In this case degG x = g1 6= 1 and degG y = 1. Then degG x2 = degG xyx = degG xy2x = 1. It is easy to see that x2, xyx, xy2x ∈ J. By the first relation of A, y3 ∈ J. Thus A/J is finite dimensional, or GKdim A/J = 0. This means that p(A, kG) = 3, and by [4, Corollary 3.8], Aco G has a graded isolated singularity. (2) It is easy to check that xyx, yxy ∈ AAg for all g ∈ G. So xyx, yxy ∈ J. Using relations of A, we have, in A/J, y3 = ax2y + byx2, y2x = xy2 + cx3 for some a, b, c ∈ k. By using Bergman's Diamond Lemma [7] with degree lexico- graphic monomial order with y > x, A/J has a monomial basis and each of the monomials does not contain subwords y3, y2x, xyx, yxy. This implies that A/J is spanned by {xi : i ≥ 0} ∪ {yxi, xiy, xiy2 : i ≥ 0} ∪ {yxiy, yxiy2 : i ≥ 0}. Thus GKdim A/J ≤ 1, and hence p(A, kG) ≥ 2. (3,4) In these cases, every monomial of degree 4 is in J. So GKdim A/J = 0 as required. (5) In this case, one can show that x3 ∈ J as degG x generates the group G. Similarly, we have y3 ∈ J. Using the relations in A, one sees that, in A/J, x3 = 0, y3 = 0, yx2 = −x2y + axyx, y2x = −xy2 + ayxy for some a ∈ k. By Bergman's Diamond Lemma [7], A/J is spanned by {xi(yx)j yk : 0 ≤ i, k ≤ 2, j ≥ 0}. 14 J. CHEN, E. KIRKMAN AND J.J. ZHANG Therefore GKdim A/J ≤ 1 and p(A, kG) ≥ 2. G such that the set {Q7 (6) Let {d7, · · · , d1} be an ordered set of elements (possibly with repetitions) in s=1 ds, · · · , d2d1, d1} is equal to G \ {1}. Suppose fs ∈ A are homogeneous of degree ds for all s = 1, · · · , 7. By Lemma 1.9(2), the product f7f6 · · · f1 is in J. Using this observation one sees that the following elements are in J: s=1 ds,Q6 y2xy3x, xyx3yx, yx3yx2, x3yx3, x2yx3y, yxy3xy, xy3xy2, y3xy3. (The reason for verifying a product of 7 letters is that any subword of these mono- mials does not have G-degree 1. This list is all degree 7 monomials in J.) Using the fact that x = 1 2 (d − u), we obtain the following relation in A/J: 2 (d + u) and y = 1 or equivalently, 0 = 27(x3yx3 − y3xy3) = (−2)u7 + lower terms. u7 = lower terms In other words, we can write u7 in terms of terms in lower degree in the lexicographic order. Similarly, by using x = 1 2 (d − u), we calculate the following in A/J: 2 (d + u) and y = 1 0 = 27(x3yx3 + y3xy3) = (−2a5 − 2a4 + 6a3 + 8a2 − 4a − 6)udu5 + (−2a6 + 2a5 + 6a4 − 6a3 − 6a2 + 4a + 2)udududu + lower terms; 0 = 27(y2xy3x + x2yx3y) = (−2a5 + 2a4 + 10a3 − 4a2 − 12a + 2)udu5 + (−2a6 − 2a5 + 6a4 + 6a3 − 6a2 − 4a + 2)udududu + lower terms; 0 = 27(xyx3yx + yxy3xy) = (2a5 − 2a4 − 6a3 + 8a2 + 4a − 6)udu5 + (−2a6 − 2a5 + 6a4 + 6a3 − 6a2 − 4a + 2)udududu + lower terms; 0 = 27(yx3yx2 + xy3xy2) = (2a5 + 2a4 − 10a3 − 4a2 + 12a + 2)udu5 + (−2a6 + 2a5 + 6a4 − 6a3 − 6a2 + 4a + 2)udududu + lower terms; where "lower term" means a linear combination of monomials of degree 7 that have lower degrees than terms appearing the expression (in this case, udududu) with respect to lexicographic order. Recall that a is the scalar that appeared in one of the relations of A, y3 = ax2y + byx2, AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 15 see the proof of part (2). If a2 6= 1 and a2 6= 2, then by a linear algebra computation, both udu5 and udududu can be expressed as "lower terms": udu5 = lower terms, udududu = lower terms. Since u7 and udududu (and then (ud)4) are equal to lower terms in A/J, Lemma 1.10(1) implies that GKdim A/J ≤ 1, as required. If a = 1, then we have 0 = 27(x3yx3 + y3xy3) = −6du6 + 8d3u4 − 8d4udu + 8d5u2 + 2d7; 0 = 27(y2xy3x + x2yx3y) = −4udu5 + 2du6 − 4d2udu3 + 4d3u4 − 4d3udud + 4d4udu − 8d5u2 + 2d7; 0 = 27(xyx3yx + yxy3xy) = −2du6 − 4d3u4 + 2d7; 0 = 27(yx3yx2 + xy3xy2) = 4udu5 − 2du6 + 4d2udu3 − 4d3udud + 4d4udu − 8d5u2 + 2d7. By a linear algebra computation, we have d3(ud)2 = lower terms. Since u7 and d3(ud)2 are equal to lower terms in A/J, Lemma 1.10(1) implies that GKdim A/J ≤ 1, as required. If a = −1, then we have 0 = 27(x3yx3 + y3xy3) = 2du6 − 4d3u4 + 2d7; 0 = 27(y2xy3x + x2yx3y) = 4udu5 + 2du6 − 4d2udu3 − 4d3udud + 4d4udu + 8d5u2 + 2d7; 0 = 27(xyx3yx + yxy3xy) = 6du6 + 8d3u4 − 8d4udu − 8d5u2 + 2d7; 0 = 27(yx3yx2 + xy3xy2) = −4udu5 − 2du6 + 4d2udu3 + 4d3u4 − 4d3udud + 4d4udu + 8d5u2 + 2d7. By a linear algebra computation, we have d3(ud)2 = lower terms. Since u7 and d3(ud)2 are equal to lower terms in A/J, Lemma 1.10(1) implies that GKdim A/J ≤ 1, as required. If a2 = 2, we need to use a different set of elements in J. By a similar argument as before and by Lemma 1.9(2), the following elements are in J: x3y3x3, y3x3y3, y2xy3x3, x2yx3y3, y3x3yx2, x3y3xy2, yxy3xyx2, xyx3yxy2, y2xy3xy2, x2yx3yx2, yxy3xy3, xyx3yx3, yx4yx3, xy4xy3, xy4xyx2, yx4yxy2. 16 J. CHEN, E. KIRKMAN AND J.J. ZHANG By a Sage computation, we have, in A/J, 0 = 29(x3y3x3 − y3x3y3) = (−2)u9 + (8 + 4a)ududu5 + (18 + 12a)ududududu + lower terms 0 = 29(y2xy3x3 − x2yx3y3) = (−2)u9 + (−4 − 4a)ududu5 + (−2 − 4a)ududududu + lower terms 0 = 29(y3x3yx2 − x3y3xy2) = 2u9 + (−4)ududu5 + 2ududududu + lower terms 0 = 29(y2xy3xy2 − x2yx3yx2) = (−2)u9 + 2ududududu + lower terms An easy linear algebra computation shows that u9 = lower terms (ud)2u5 = lower terms (ud)4u = lower terms Since u9 and (ud)4u (and then (ud)5) are equal to lower terms in A/J, Lemma 1.10(1) implies that GKdim A/J ≤ 1, as required. This finishes the proof. (cid:3) 2.4. Case 4: A = F, x and y are G-homogeneous. The final case is when A = F and the proof is also quite tricky. We start with a result of [11]. Lemma 2.12. [11, Lemma 1.6] Let F be generated by x and y and subject to two relations (E2.0.1). (1) Define an order on monomials by extending x < y lexicographically. Then we have a complete set of five relations that is the reduction system in the sense of [7, p.180]. y3 = xyx, yxy = x3, y2x3 = xyx2y, yx2yx = x3y2, yx4 = x4y. (2) We also have the other relations: y4 = x4, yxyx = x4, xyxy = y4. (3) There is a k-linear basis consisting of the monomials of the form xi(yx3)j(yx2)ǫ(y2x2)kyaxb where i, j, k ≥ 0, ǫ is either 0 or 1, and (a, b) = (0, 0), (1, 0), (1, 1), (1, 2), (1, 3), (2, 0), (2, 1), (2, 2) AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 17 if j + ǫ + k = 0, (a, b) = (1, 0), (1, 1), (1, 2), (1, 3), (2, 0), (2, 1), (2, 2) if j > 0 and ǫ + k = 0, and (a, b) = (1, 0), (2, 0), (2, 1), (2, 2) if ǫ + k > 0. Let G be a finite group coacting on A := F such that x and y are G-homogeneous. Let J be defined as in (E2.1.3). Lemma 2.13. Suppose there are α, β, α′, β′ ≥ 0 such that (yx2)(y2x2)β ′ (yx3)α(y2x2)β and (yx3)α′ ∈ J, then GKdim A/J ≤ 1. Proof. We again use the Diamond Lemma [7]. By the fact that x4 = y4 is central in A [Lemma 2.12(2)], there is a monomial f in x and y such that f (yx3)i(y2x2)k = x4w for some w. So we have the following equations in A/J: x4w = 0, (yx3)α(y2x2)β = 0, (yx2)(y2x2)β ′ = 0. (yx3)α′ Therefore, if there are i, j, k with xi(yx3)j(yx2)ǫ(y2x2)kyaxb 6= 0 in A/J for (a, b) as in Lemma 2.12(b), then there is a uniform bound on at least two of {i, j, k}. Then GKdim A/J ≤ 1 by [5, (E1.1.6)] and Lemma 2.12(3). (cid:3) The goal of the rest of this subsection is to find some monomials (yx3)i(y2x2)k and (yx3)i′ (yx2)(y2x2)k′ in J. We introduce the following notation. Let We will use Xi for i ∈ Z/(2). Let and X0 = x, X1 = y. V0 = yx3, V1 = xyx2, V2 = x2yx, V3 = x3y, W0 = y2x2, W1 = xy2x, W2 = x2y2, W3 = yx2y. We will use Vi and Wi for i ∈ Z/(4). The following lemma follows from a direct computation and the relations given in Lemma 2.12. Lemma 2.14. Retain the above notation. (1) XiVj = Vj+1Xi and XiWj = Wj+1Xi for all i ∈ Z/(2) and j ∈ Z/(4). (2) Elements {V0, · · · , V3, W0, · · · , W3} are pairwise commutative. 18 J. CHEN, E. KIRKMAN AND J.J. ZHANG (3) The following relations hold (a) V0V2 = x8 = V1V3. (b) W0W2 = x8 = W1W3. (c) V0V1 = x4W0. Now let G be a finite group coacting on F such that x and y are G-homogeneous. We also assume that the G-coaction has trivial homological codeterminant, namely, degG(x4) = 1. Let xi = degG Xi, vj = degG Vj , wk = degG Wk for i ∈ Z/(2) and j, k ∈ Z/(4). Let N be the subgroup generated by {vi}3 {wk}3 k=0. By the lemma above, we have i=0 ∪ Lemma 2.15. Retain the above notation. (1) G is generated by x0(= g1) and x1(= g2). (2) N is an abelian subgroup of G. (3) N is a normal subgroup of G. (4) N is generated by v0 and v1 and G/N is generated by the image x0 of x0; and x0 = x1 in G/N . (5) N is also generated by {vi, wj } for any pair (i, j). (6) G = N ∪ N xi1 ∪ N xj1 xj2 ∪ N xk1 xk2 xk3 for any fixed is, js, ks ∈ Z/(2). (7) Let n be the order of v0. For any fixed is ∈ Z/(2) and js, ks ∈ Z/(4) for s = 1, 2, 3, 4, any element in G is a right subword of j4 wn vn k4 xi3 vn j3 wn k3 xi2 vn j2 wn k2 xi1 vn j1 wn k1 . Proof. (1) Since G-coaction is inner-faithful, G is generated by degG x and degG y. (2) This follows from Lemma 2.14(2). (3) This follows from Lemma 2.14(1) and part (1). (4) Since degG x4 = 1, x4 1 = 1. By Lemma 2.14(3), v2 = v−1 w0 = v0v1. It is easy to check from Lemma 2.14(1) that w1 = v−1 and w3 = v0v−1 x0 = x1. So G/N is generated by x0. 1 , and 0 v−1 1 . Therefore N is generated by v0 and v1. It is clear that in G/N , 0 , v3 = v−1 0 v1, w2 = v−1 0 = x4 1 (5) By the proof of part (4), we have 0 , v−1 1 }, 0 v1, v−1 0 v−1 Therefore N is generated by {vi, wj} for any pair (i, j). {v0, v1, v2, v3} = {v0, v1, v−1 {w0, w1, w2, w3} = {v0v1, v−1 1 , v0v−1 1 }. (6) This follows from the fact that x0 = x1 in the quotient group G/N and that G/N ∼= Z/(4). (7) This follows from parts (5,6). (cid:3) We have a version of Lemma 2.15 for monomials in F. Lemma 2.16. Retain the above notation. (1) For any fixed is ∈ Z/(2) and js, ks ∈ Z/(4) for s = 1, 2, 3, 4, any element in G is the degree of some right subword of j2 W n k4 Xi3 V n k3 Xi2 V n Φ := V n j3 W n j4 W n k2 Xi1 V n j1 W n k1 . AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 19 As a consequence, Φ ∈ J . (2) (yx3)4n(yx2)(y2x2)4n and (yx3)4n(y2x2)4n+1 are elements in J . Proof. (1) This is a slightly stronger version of Lemma 2.15(7). For example, if an element g is of the form j2 wb va k2 x0vn j1 wn k1 then we take a right subword of the form f := V a X0V n Wt all commute, f is a subword of Φ. Clearly, degG f = va consequence follows from Lemma 1.9(1). (2) By using Lemma 2.14(1), Φ equals j2 W b k2 j1 W n k1 j2 wb k2 . Since Vs and x0vn . The j1 wn k1 j4 V n V n j3+1V n j2 +2V n j1+3Xi3 Xi2 Xi1 Wk4−3Wk3−2Wk2−1Wk1 . By taking j4 = j3 + 1 = j2 + 2 = j1 + 3 = 0 and k4 − 3 = k3 − 2 = k2 − 1 = k1 = 0 and i3 = 1, i2 = i1 = 0, we have that (yx3)4n(yx2)(y2x2)4n ∈ J. By taking j4 = j3 + 1 = j2 + 2 = j1 + 3 = 0 and k4 − 3 = k3 − 2 = k2 − 1 = k1 = 1 and i3 = i2 = 1 and i1 = 0, we have that (yx3)4n(y2x)(xy2x)4n ∈ J. Then (yx3)4n(y2x2)4n+1 ∈ J. (cid:3) Now we can prove the result of this subsection. Proposition 2.17. Retain the notation as in Theorem 0.1. Suppose that A = F and x and y are G-homogeneous. Then p(A, kG) ≥ 2. Proof. Combining Lemma 2.16(2) with Lemma 2.13, GKdim A/J ≤ 1. This is equivalent to p(A, kG) ≥ 2. (cid:3) Putting all these pieces together we have a proof of Theorem 0.1. Proof of Theorem 0.1. First we assume that G is abelian (that could be cyclic). If A is not D(α, −1) for some α 6= 2, the assertion follows from [5, Theorem 0.6]. Now we assume that A = D(α, −1) for some α 6= 2. Using notations introduced in with respect to the basis {d, u}, then the assertion follows from [5, Proof of Theorem subsection 2.3, the character group of G is denoted by bG. If bG acts on A diagonally 0.6]. Otherwise, bG contains non-diagonal matrices with respect to the basis {d, u}. In this case the assertion follows from Lemmas 2.9, 2.10 and 2.11. This takes care of the case when G is abelian. Next we assume that G is not abelian. As a consequence, G is not cyclic, so we can apply Lemma 2.1. Using the classification given in Lemma 2.1, we only need to show the assertion for the first three cases as in the last two cases G is abelian. In case Lemma 2.1(1), this is Proposition 2.4. In case Lemma 2.1(2), this is Proposition 2.17. In case Lemma 2.1(3), this is Lemma 2.7(3). This finishes the proof. (cid:3) 20 J. CHEN, E. KIRKMAN AND J.J. ZHANG 3. Examples We conclude this paper with several examples that indicate the variety of covari- ant subrings that can be obtained from coactions on down-up algebras, and give some concluding comments. Example 3.1. A group G = ha, bi coacts on Dβ, where degG d = a and degG u = b, with trivial homological codeterminant if a2b2 = 1 [Lemma 1.5]. One such family of groups is the dihedral groups, for n ≥ 2, D2n = ha, b a2 = b2 = (ba)n = 1i. Each element of Dβ can be written as a linear combination of monomials of the form di(ud)juk, and such a monomial is in the identity component of Dβ exactly when ai(ba)jbk = 1. Clearly d2 and u2 are covariants, so it suffices to consider the four cases (i, k) = (0, 0), (1, 0), (0, 1), (1, 1), and it is easy to check that the subring of covariants is generated as a k-algebra by d2, u2, (du)n, (ud)n, i.e. Dco D2n β = khd2, u2, (du)n, (ud)ni. When β = ±1, Dco D2n ±1 is isomorphic to the commutative algebra Dco D2n ±1 ∼= k[X, Y, Z, W ] (X nY n − ZW ) , a hypersurface in A4. When β 6= ±1, the ring of covariants is a hypersurface in a noncommutative skew polynomial ring in the sense of [18, Definition 1.3(c)]. For any β 6= 0, Dβ#kD2n is a noncommutative quasi-resolution or NQR (a generalization of NCCR) of Dco D2n in the sense of [22]. When n = 4, this example should be compared to [11, Example 2.1], where a different coaction of D8 (without trivial homological codeterminant) on D1 is given; in that example the ring of covariants is a commutative hypersurface in A4, but Auslander's Theorem fails [Remark 1.6(1)]. β Next we consider a second coaction on Dβ, where the ring of covariants is quite different. Example 3.2. The quaternion group G = Q8 of order 8 coacts on Dβ by degG d = i and degG u = k with trivial homological determinant [Lemma 1.5]. A monomial de1 (ud)e2 ue3 has group grade the identity of Q8 exactly when ie1 je2 ke3 = 1 holds in Q8. following 9 monomials; It is not hard to check that the covariants are generated by the d4, u4, d2u2, d2(ud)2, (ud)2u2, d(ud)u3 = (du)2u2, d3(ud)u = d2(du)2, d(ud)3u = (du)4, (ud)4. When β 6= 0, Dβ#kQ8 is a noncommutative quasi-resolution (NQR) of the covariant subring Dco Q8 in the sense of [22]. β Finally we consider the down-up algebra H. AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 21 Example 3.3. The dihedral groups D2n coact on H homogeneously with trivial homological codeterminant, although our proof of Auslander's theorem holds for any group coaction in this case (see Lemma 2.7 and the comments after that). Suppose that G := D2n = ha, b a2 = b2 = 1 = (ab)ni and that degG x = a and degG y = b. The relations in H (E2.0.2) can be written as (x2 − y2)y = −y(x2 − y2) x(x2 − y2) = −(x2 − y2)x, and hence x2 − y2 is a normal element of H, and, moreover, xy and yx commute with y2 − x2. It is clear that x2 and y2 are covariants under this action, and that H/(x2 − y2) ∼= khx, yi/(x2 − y2). Since x2 − y2 is also a normal element of Hco D2n , we obtain that Hco D2n (x2 − y2) ∼=(cid:18) khx, yi (x2 − y2)(cid:19)co D2n . It follows that the generators of Hco D2n are the 4 elements x2, y2, (yx)n, x(yx)n−1y = (xy)n. Next we show that Hco D2n is a hypersurface [18, Definition 1.3(c)] in an iterated Ore extension that is an AS regular algebra of dimension 4. Multiplying the relation x2y + yx2 − 2y3 = 0 by y on each side and subtracting gives the relation x2y2 − y2x2 = 0. We next give a number of relations in H; note that the defining relations of H are symmetric in x and y, so the relations with x and y interchanged also hold. It is easy to check the following relation x2y − yx2 = 2y(y2 − x2); multiplying by x on left gives: [x2, (xy)] = 2xy(y2 − x2) = (y2 − x2)(2xy). Similarly Inductively one can show that [y2, (xy)] = 2xy(y2 − x2). Further [y2, (xy)n] = 2n(xy)n(y2 − x2). (yx)(xy) = y2(y2 + (y2 − x2)), and inductively we get (yx)n(xy)n = y2(y2 + (y2 − x2))(y2 + 2(y2 − x2))(y2 + 3(y2 − x2)) · · · (y2 + (2n − 1)(y2 − x2)). Let the right side of the above equation be denoted by w(x2, y2). We claim that the subalgebra generated by x2, y2, (yx)n, (xy)n is a generalized Weyl algebra (or ambiskew polynomial ring, as in [14]), i.e. it is an iterated Ore algebra modulo one relation. To simplify notation, let X = x2, Y = y2, Z + = (yx)n, Z − = (xy)n. From the relations above we have the following relations: XY = Y X, Z +X = (X + 2n(Y − X))Z +, Z +Y = (Y + 2n(Y − X))Z +. 22 J. CHEN, E. KIRKMAN AND J.J. ZHANG In this notation Z +Z − = w(X, Y ) and Z −Z + = w(Y, X). Let B be the algebra generated by X, Y, Z + defined by the first three relations above. Then B is the Ore extension k[X, Y ][Z +; σ], where σ is the automorphism of k[X, Y ] given by σ(X) = X + 2n(Y − X) and σ(Y ) = Y + 2n(Y − X). Adjoining Z − to B adds the following three relations: Z −X = (X − 2n(Y − X))Z −, Z −Y = (Y − 2n(Y − X))Z −, Z −Z + = Z +Z − − f (X, Y ). where f (X, Y ) = w(X, Y ) − w(Y, X) ∈ k[X, Y ]. One checks that σ−1(X) = X − 2n(Y − X), σ−1(Y ) = Y − 2n(Y − X) and σ−1(Z +) = Z + defines an algebra automorphism of B. Define the map δ on B, by δ(X) = δ(Y ) = 0 and δ(Z +) = −f (X, Y ). One checks that δ extends to a σ−1-skew derivation of B, preserving the three relations defining B. Hence we have Hco D2n ∼= k[X, Y ][Z +; σ][Z −; σ−1, δ] (Z +Z − − w(X, Y )) , and Hco D2n is a hypersurface [18, Definition 1.3(c)] in an AS regular algebra of dimension 4, with Hilbert series 1 − t4n (1 − t2)2(1 − t2n)2 . Down-up algebras have no reflections [19, Proposition 6.4], so we would expect Auslander's theorem to hold for all finite group actions; this has been proved for almost all finite group actions, except for some finite groups acting on D(α, −1) for α 6= 2 (abelian groups with trivial homological determinant are covered by Theo- rem 0.1). Theorem 0.1 also covers finite group coactions with trivial homological determinant, and we have shown Theorem 0.1 does not hold for all group coactions [Remark 1.6(1)]. It remains to examine actions by other Hopf algebras on down-up algebras. Acknowledgments. The authors thank Kenneth Chan and Zhibin Gao for sev- eral conversations on the subject and their help with some computations in this paper and thank the referees for their very careful reading and extremely valu- able comments. J. Chen was partially supported by the National Natural Science Foundation of China (Grant No. 11571286) and the Natural Science Foundation of Fujian Province of China (Grant No. 2016J01031). E. Kirkman was partially supported by grant #208314 from the Simons Foundation. J.J. Zhang was partially supported by the US National Science Foundation (Grant Nos. DMS-1402863 and DMS-1700825). References [1] M. Artin and W. F. Schelter, Graded algebras of global dimension 3, Adv. in Math. 66 (1987), no. 2, 171 -- 216. [2] M. Artin and J. J. Zhang, Noncommutative projective schemes, Adv. Math. 109 (1994), 228-287. [3] M. Auslander, On the purity of the branch locus, Amer. J. Math. 84 (1962), 116 -- 125. [4] Y.-H. Bao, J.-W. He and J.J. Zhang, Pertinency of Hopf actions and quotient cate- gories of Cohen-Macaulay algebras, preprint (2016) to appear J. of Noncommut. Geom., arXiv:1603.02346. AUSLANDER'S THEOREM FOR DOWN-UP ALGEBRAS 23 [5] Y.-H. Bao, J.-W. He and J.J. Zhang, Noncommutative Auslander theorem, Trans. Amer. Math. Soc. 370 (2018), no. 12, 86138638. [6] G. Benkart and T. Roby, Down-up algebras, J. Algebra 209 (1998), 305 -- 344. Addendum, J. Algebra 213 (1999), no. 1, 378. [7] G.M. Bergman, The diamond lemma for ring theory, Adv. in Math. 29 (1978), no. 2, 178-218. [8] K. Chan, E. Kirkman, C. Walton and J.J. Zhang, Quantum binary polyhedral groups and their actions on quantum planes, J. Reine Angew. Math. 719 (2016), 211 -- 252. [9] K. Chan, E. Kirkman, C. Walton and J.J. Zhang, McKay Correspondence for semisimple Hopf actions on regular graded algebras I, J. Algebra 508 (2018), 512 -- 538. [10] K. Chan, E. Kirkman, C. Walton and J.J. Zhang, McKay Correspondence for semisimple Hopf actions on regular graded algebras II, J. Noncommut. Geom. 13 (2019), no. 1, 87 -- 114. [11] J. Chen, E. Kirkman, J.J. Zhang, Rigidity of down-up algebras with respect to finite group coactions, J. Pure Applied Algebra 221 (2017), 3089 -- 3103. [12] J. Gaddis, E. Kirkman, W.F. Moore and R. Won, Auslander's Theorem for permutation actions on noncommutative algebras, Proc. Amer. Math. Soc. 147 (2019), no. 5, 1881 -- 1896. [13] J.-W. He and Y. Zhang, Local cohomology associated to the radical of a group action on a noetherian algebra, preprint, (2017) arXiv: 1712.00657. [14] D. A. Jordan, Down-up algebras and ambiskew polynomial rings, J. Algebra 228 (2000), no. 1, 311 -- 346. [15] E. Kirkman and J. Kuzmanovich, Fixed subrings of Noetherian graded regular rings, J. Al- gebra 288 (2005), no. 2, 463 -- 484. [16] E. Kirkman, J. Kuzmanovich and J.J. Zhang, Gorenstein subrings of invariants under Hopf algebra actions, J. Algebra 322 (2009), no. 10, 3640 -- 3669. [17] E.Kirkman, J. Kuzmanovich, and J.J. Zhang, Invariant theory of finite group actions on down-up algebras, Transform. Groups 20 (2015), no. 1, 113 -- 165. [18] E. Kirkman, J. Kuzmanovich, and J.J. Zhang, Noncommutative complete intersections, J. Algebra 429 (2015), 253 -- 286. [19] E. Kirkman, J. Kuzmanovich, and J.J. Zhang, Rigidity of graded regular algebras, Trans. Amer. Math. Soc. 360 (2008), no. 12, 6331 -- 6369. [20] E. Kirkman, I. Musson, D. Passman,Noetherian down-up algebras, Proc. Amer. Math. Soc. 127 (1999), 3161-3167. [21] I. Mori and K. Ueyama, Ample Group Action on AS-regular Algebras and Noncommutative Graded Isolated Singularities, Trans. Amer. Math. Soc. 368 (2016), no. 10, 7359 -- 7383. [22] X.-S. Qin, Y.-H. Wang and J.J. Zhang, Noncommutative quasi-resolutions, preprint (2018), arXiv:1802.09092. [23] M. Reyes, D. Rogalski and J.J. Zhang, Skew Calabi-Yau algebras and homological identities, Adv. Math. 264 (2014), 308 -- 354. [24] K. Ueyama, Graded maximal Cohen-Macaulay modules over noncommutative graded Goren- stein isolated singularities, J. Algebra 383 (2013), 85 -- 103. [25] K. Ueyama, Cluster tilting modules and noncommutative projective schemes, Pacific J. Math. 289 (2017), 449 -- 468. Chen: School of Mathematical Sciences, Xiamen University, Xiamen 361005, Fujian, China E-mail address: [email protected] Kirkman: Department of Mathematics, P. O. Box 7388, Wake Forest University, Winston-Salem, NC 27109 E-mail address: [email protected] Zhang: Department of Mathematics, Box 354350, University of Washington, Seattle, Washington 98195, USA E-mail address: [email protected]
1703.04055	1	1703	2017-03-12T01:46:05	Dynkin Diagrams and Short Peirce Gradings of Kantor Pairs	[ "math.RA" ]	In a recent article with Oleg Smirnov, we defined short Peirce (SP) graded Kantor pairs. For any such pair P, we defined a family, parameterized by the Weyl group of type BC_2, consisting of SP-graded Kantor pairs called Weyl images of P. In this article, we classify finite dimensional simple SP-graded Kantor pairs over an algebraically closed field of characteristic 0 in terms of marked Dynkin diagrams, and we show how to compute Weyl images using these diagrams. The theory is particularly attractive for close-to-Jordan Kantor pairs (which are variations of Freudenthal triple systems), and we construct the reflections of such pairs (with nontrivial gradings) starting from Jordan matrix pairs.	math.RA	math	DYNKIN DIAGRAMS AND SHORT PEIRCE GRADINGS OF KANTOR PAIRS BRUCE ALLISON AND JOHN FAULKNER Abstract. In a recent article with Oleg Smirnov, we defined short Peirce (SP) graded Kantor pairs. For any such pair P , we defined a family, parameterized by the Weyl group of type BC2, consisting of SP-graded Kantor pairs called Weyl images of P . In this article, we classify finite dimensional simple SP- graded Kantor pairs over an algebraically closed field of characteristic 0 in terms of marked Dynkin diagrams, and we show how to compute Weyl images using these diagrams. The theory is particularly attractive for close-to-Jordan Kantor pairs (which are variations of Freudenthal triple systems), and we construct the reflections of such pairs (with nontrivial gradings) starting from Jordan matrix pairs. Suppose in this introduction that K is a field of characteristic 6= 2 or 3. A Kantor pair is a pair P = (P −, P +) of K-modules together with trilinear products { , , }σ : P σ × P −σ × P σ → P σ, σ = ±, satisfying two 5-linear identities (see Section 4.2) which were first written down by Isai Kantor in [K1] in the special case of Kantor triple systems (when P − = P + and { , , }− = { , , }+). Examples of Kantor pairs, or structures that give rise to Kantor pairs, have arisen in the work of many different authors (see Section 4.3 and [AFS, §3.1] for some references). The motivation for the study of Kantor pairs is their relationship with 5-graded Lie algebras. We now recall that relationship in the special case of primary interest to us when the structures involved are simple, in which case the relationship is a 1-1 correspondence. (Our convention, as described in Subsection 2.1, is that the term simple for a graded structure is interpreted in the ungraded sense.) If L = L−2 ⊕ L−1 ⊕ L0 ⊕ L1 ⊕ L2 is a simple 5-graded Lie algebra, then (L−1, L1) is a simple Kantor pair with products given by {x, y, z}σ = [[x, y], z] for σ = ±, in which case we say that the pair (L−1, L1) is enveloped by L. Conversely, any simple Kantor pair P is enveloped by a simple 5-graded Lie algebra L, which is unique up to graded isomorphism [AFS] and which has the property that Lσ2 ≃ K σ(P σ, P σ) (as vector spaces) for σ = ±, where K σ(x, z) ∈ Hom(P −σ, P σ) is defined by K σ(x, z)w = {x, w, z}σ − {z, w, x}σ for x, z ∈ P σ. A Kantor pair P satisfying K σ(P σ, P σ) = 0 for σ = ± is called a (linear) Jor- dan pair [L], and the relationship between Kantor pairs and 5-graded Lie algebras, generalizes the well-known relationship between Jordan pairs and 3-graded Lie al- gebras. With this important special case in mind, we define a close-to-Jordan pair to be a Kantor pair P with dim(K σ(P σ, P σ)) = 1 for σ = ±. Date: February 10, 2017. 2010 Mathematics Subject Classification. Primary 17B60, 17B70; Secondary 17C99. Key words and phrases. Kantor pairs, graded Lie algebras, Jordan pairs, Freudenthal triple systems, homomorphisms of root systems. 1 DYNKIN DIAGRAMS AND KANTOR PAIRS 2 If ∆ is a finite root system (possibly not reduced), we define a ∆-grading of a Lie algebra L, to be a grading of L by the root lattice of ∆, with support con- tained in ∆ ∪ {0} (see Section 3). With this terminology we see that 5-graded Lie algebras can be thought of as BC1-graded Lie algebras, by which we mean a Lie algebra graded by the root system {−2α1, −α1, α1, 2α1} of type BC1. Hence the correspondence described above can be viewed as a 1-1 correspondence between simple Kantor pairs (up to isomorphism) and simple BC1-graded Lie algebras (up to graded isomorphism). We next recall how BC2-graded Lie algebras also play a role in the study of Kantor pairs. A short Peirce grading (SP-grading) of a Kantor pair P is a Z- grading P = P0 ⊕ P1 of P with Pi = 0 for i 6= 0, 1. These gradings were introduced in [AFS], where it was shown that there is a 1-1 correspondence between simple SP- graded Kantor pairs and simple BC2-graded Lie algebras that is analogous to the one just described. Given a simple SP-graded Kantor pair P , this correspondence was used to construct a simple SP-graded Kantor pair uP , called the u-image of P , for each u in the Weyl group of the root system of type BC2. In this way one obtains eight Weyl images of a simple SP-graded Kantor pair P , no two of which are (in general) graded-isomorphic. Of particular interest is the s1-image of P , which we simply call the reflection of P , where s1 is the reflection corresponding to the short basic root. As an application of Weyl images, it was shown in [AFS] that reflection applied to Jordan pairs of skew-transformations yields a new class of (in general) infinite dimensional simple SP-graded Kantor pairs that are not themselves Jordan. In this article, we use subsets of Dynkin diagrams to classify simple SP-graded Kantor pairs and to compute their Weyl images in the special case of finite dimen- sional pairs over an algebraically closed field K of characteristic 0. For the rest of this introduction, we make these additional assumptions and briefly outline our main results. In these results, a simple Kantor pair is said to be of type Xn, if the root system of its enveloping simple 5-graded Lie algebra is of type Xn. With the exception of Section 5, which we discuss below, Sections 1 to 6 are devoted to laying the necessary groundwork on homomorphisms of root systems, root graded Lie algebras and Kantor pairs. Of particular interest is Theorem 3.2.2, which uses homomorphisms of root systems to classify all ∆-gradings (with ∆ arbitrary) of a finite dimensional semi-simple Lie algebra. In Section 7, we assume that Π is the Dynkin diagram of type Xn, where Xn is the type of an irreducible reduced finite root system. We first define a set SPA(Π) of pairs of subsets of Π, whose elements are said to be SP-admissible. In Proposition 7.1.5. we show how to easily write down the elements of SPA(Π) using the extended diagram of Π. Then, in Theorem 7.2.4, we give a classification of simple SP-graded Kantor pairs of type Xn by showing that they are (up to graded isomorphism) in 1-1 correspondence with the orbits in SPA(Π) under the right action of Aut(Π). The proof uses Theorem 3.2.2 when ∆ is irreducible of type BC2. If we specialize Theorem 7.2.4 to the case when the SP-grading is the zero grading (i.e. P = P 0), we obtain a classification of simple (ungraded) Kantor pairs of type Xn, which states that they are (up to isomorphism) in 1-1 correspondence with the orbits in the set KA(Π) of Kantor admissible subsets of Π under the right action of Aut(Π). For the sake of readability, we actually state this classification theorem earlier as Theorem 5.2.4 in Section 5. We note that Theorem 5.2.4 is equivalent to Kantor's classification of non-polarized simple Kantor triple systems [K1] (see DYNKIN DIAGRAMS AND KANTOR PAIRS 3 Remark 5.2.5). Nevertheless we include a complete treatment of the topic here, since we are working in the context of pairs rather than triple systems, and since the article [K1] is unavailable to some readers and does not include all details. Both Theorems 5.2.4 and 7.2.4 take particularly simple and attractive forms in the case of close-to-Jordan pairs, and for this reason we highlight this example throughout the paper. For example, we see in Theorems 5.4.1 and 5.5.2 that if n ≥ 2 there is exactly one simple close-to-Jordan pair of type Xn and that this Kantor pair can be viewed as the signed double of a Freudenthal triple system [M1] with a modified product. If Section 8, we introduce a left action, denoted by ∗, of the Weyl group W∆ of the root system ∆ of type BC2 on the set SPA(Π). In Theorem 8.3.1, we show that, if u ∈ W∆, the u-image of the SP-graded Kantor pair corresponding to (S, T ) ∈ SPA(Π) is the SP-graded Kantor pair corresponding to u∗(S, T ) ∈ SPA(Π). Then in Theorem 8.3.2, we show how to compute u ∗ (S, T ) using only information about the Dynkin diagram Π. The proof of this result is rather intricate, involving the longest element of the Weyl group of certain subsets of Π. Combining Theorems 8.3.1 and 8.3.2 with our classification Theorem 7.2.4, one can easily compute the Weyl images of any simple SP-graded Kantor pair using only information about Π. As an application of our results, we obtain in Section 9 a construction of the reflections of all simple nontrivially SP-graded close-to-Jordan pairs in the form J ⊗U , where J is a Jordan pair of matrices and U is a pair of two-dimensional spaces. This generalizes Kantor's construction of his remarkable Kantor triple system C2 55, since the double of that triple system is a reflection of the close-to-Jordan pair of type E6 [AFS, §7]. Acknowledgments: Isai Kantor's methods and ideas are used in many places in this article, and we are pleased to acknowledge his significant influence. Also, we have greatly benefited from many conversations with Oleg Smirnov about this work, and about Kantor pairs and graded Lie algebras in general. 1. Some sets of homomorphisms between root systems In this section we investigate some sets of homomorphisms between roots sys- tems. 1.1. Root systems. In this paper, a root system will mean a finite root system ∆ in a finite dimensional real Euclidean space E∆ as described for example in [B, VI, §1.1]. ∆ is said to be reduced if (2∆)∩∆ = ∅. The rank n∆ of ∆ is the dimension of E∆. If ∆ is irreducible and n = n∆, the type of ∆ is one of An(n ≥ 1)), Bn(n ≥ 2), Cn(n ≥ 3)), Dn(n ≥ 4)), En(n = 6, 7, 8), F4, G2, or BCn(n ≥ 1), where the last case occurs if and only if ∆ is not reduced [B, VI, §1.4, Prop. 14]. We use the notation Q∆ := spanZ(∆) for the root lattice of ∆, which is a free abelian group of rank n∆. The automorphism group of ∆, denoted by Aut(∆), is the stabilizer of ∆ in GL(E∆). Using the restriction map, we often identify Aut(∆) with the stabilizer of ∆ in Aut(Q∆). The Weyl group of ∆ (contained in Aut(∆)), will be denoted by W∆. We denote the Euclidean product on E∆ by ( , ), and write hα, βi = 2(α, β)/(β, β) for α, β ∈ ∆. If ∆ is irreducible, we denote the set of roots of minimum length (short roots) in ∆ by ∆sh. If a base Π∆ for the root system ∆ is fixed, we let ∆+ be the set of positive nγγ ∈ Q∆, where nγ ∈ Z, we define the height roots relative to Π∆. If α =Pγ∈Π∆ of α to be htΠ∆(α) :=Pγ∈Π∆ nγ and the support of α in Π∆ to be suppΠ∆ (α) := DYNKIN DIAGRAMS AND KANTOR PAIRS 4 {γ ∈ Π∆ : nγ 6= 0}. We denote the stabilizer of Π∆ in Aut(∆) by Aut(Π∆). If ∆ is reduced, the set Π∆ has the additional structure of a Dynkin diagram, in which case the restriction map identifies Aut(Π∆) with the automorphism group of the Dynkin diagram Π∆ [B, VI, §4.2]. If Π∆ = {αi : i ∈ I} is indexed by a finite set I, the I × I-matrix C(Π∆) = (hαi, αji)i,j∈I is called the Cartan matrix of Π∆. If G is a finite dimensional simple Lie algebra over an algebraically closed field of characteristic 0 with Cartan subalgebra H, the classical theory for such Lie algebras tells us that the root system Σ(G, H) of G relative to H is an irreducible reduced root system in the Euclidean space EΣ(G,H) = R ⊗Q spanQ Σ(G, H) [H, §8.5]. In that case, the type of G is defined to be the type of Σ(G, H). 1.2. The sets Hom(Σ, ∆) and Homp(Σ, ∆). For the rest of the section, we assume that ∆ is a root system of rank n∆ with base Π∆, and Σ is a root system of rank nΣ with base ΠΣ. Elements of Q∆ will be normally denoted by α, β, . . . , while elements of QΣ will normally be denoted by µ, ν, . . . . Let Hom(Σ, ∆) := {ρ ∈ Hom(QΣ, Q∆) : ρ(Σ) ⊆ ∆ ∪ {0}}. We call elements of Hom(Σ, ∆) homomorphisms of Σ into ∆. We say that ρ ∈ Hom(Σ, ∆) is positive (relative to ΠΣ and Π∆) if ρ(Σ+) ⊆ ∆+ ∪{0} (or equivalently ρ(ΠΣ) ⊆ ∆+ ∪ {0}). Let Homp(Σ, ∆) := {ρ ∈ Hom(Σ, ∆) : ρ is positive}. 1.3. Some actions by composition. Recall that if G and G′ are abelian groups (written additively) then there is a right action of Aut(G) by composition on the group Hom(G, G′), as well as a left action by composition of Aut(G′) on Hom(G, G′). These are defined respectively by ρ · ϕ := ρ ◦ ϕ and θ · ρ := θ ◦ ρ for ϕ ∈ Aut(G), ρ ∈ Hom(G, G′) and θ ∈ Aut(G′). Since these two actions com- mute, we can write expressions like θ · ρ · ϕ for ϕ ∈ Aut(G), ρ ∈ Hom(G, G′) and θ ∈ Aut(G′). In particular, Aut(QΣ) acts on the right by composition and Aut(Q∆) acts on the left by composition on Hom(QΣ, Q∆). Moreover, clearly Aut(∆) · Hom(Σ, ∆) · Aut(Σ) ⊆ Hom(Σ, ∆). Hence Aut(Σ) (resp. Aut(∆)) acts on the right (resp. left) on Hom(Σ, ∆) by com- position. Lemma 1.3.1. Each orbit in Hom(Σ, ∆) under the right action of WΣ by compo- sition contains a unique element of the set Homp(Σ, ∆). Proof. We will need some notation for the proof. First let C(ΠΣ) = {µ ∈ EΣ : (µ, ν) ≥ 0 for ν ∈ ΠΣ} denote the closure of the fundamental Weyl chamber in EΣ determined by ΠΣ. Also, if τ ∈ Hom(QΣ, Z), then (since QΣ is a lattice in EΣ) there exists a unique τ ∗ ∈ EΣ such that (τ ∗, µ) = τ (µ) for µ ∈ QΣ. Since WΣ preserves the form ( , ) we see that (τ · w)∗ = w−1τ ∗ for τ ∈ Hom(QΣ, Z) and w ∈ WΣ. Moreover, if τ ∈ Hom(QΣ, Z), then τ ∗ ∈ C(ΠΣ) if and only if τ (µ) ≥ 0 for µ ∈ Σ+. DYNKIN DIAGRAMS AND KANTOR PAIRS 5 ρi(µ)ρj(µ) ≥ 0 for µ ∈ Σ and 1 ≤ i 6= j ≤ n∆, (1) for µ ∈ QΣ, where Π∆ = {α1, . . . , αn∆}. Then if µ ∈ Σ, we have To prove existence, suppose that ρ ∈ Hom(Σ, ∆). Since ρ ∈ Hom(QΣ, Q∆), there i=1 ρi(µ)αi exists unique elements ρ1, . . . , ρn∆ ∈ Hom(QΣ, Z) such that ρ(µ) =Pn∆ since ρ(µ) ∈ Σ ∪ {0}. Let ρ := Pn∆ Therefore,Pn∆ i=1 ρi ∈ Hom(QΣ, Z). Then, since C(ΠΣ) is a fundamental domain for WΣ [B, V, § 3.2, Thm. 1], we can choose w ∈ WΣ such that w−1 ρ∗ ∈ C(ΠΣ). Hence (ρ · w)∗ ∈ C(ΠΣ), so (ρ · w)(µ) ≥ 0 for µ ∈ Σ+. i=1 ρi(wµ) ≥ 0 for µ ∈ Σ+, and hence, by (1), ρi(wµ) ≥ 0 for µ ∈ Σ+, 1 ≤ i ≤ n∆. So ρ · w is positive. For uniqueness suppose that ρ ∈ Hom(Σ, ∆), w ∈ WΣ and both ρ and ρ · w are positive. Choose ρ1, . . . , ρn∆ ∈ Hom(QΣ, Z) as in the previous paragraph. Then ρi(µ) ≥ 0 and (ρi · w)(µ) ≥ 0 for µ ∈ Σ+ and 1 ≤ i ≤ n∆. Thus ρ∗ i and w−1ρ∗ i are in C(ΠΣ). So, again since C(ΠΣ) is a fundamental domain for WΣ, we have ρ∗ i = w−1ρ∗ (cid:3) i and hence ρi = ρi · w. Thus ρ = ρ · w. Finally, it is clear that Aut(Π∆) · Homp(Σ, ∆) · Aut(ΠΣ) ⊆ Homp(Σ, ∆), so Aut(ΠΣ) (resp. Aut(Π∆)) acts on the right (resp. left) on Homp(Σ, ∆) by com- position. 1.4. The left action ∗ of Aut(∆) on Homp(Σ, ∆). If θ ∈ Aut(∆) and ρ ∈ Homp(Σ, ∆), we let θ ∗ ρ denote the unique element of θ · ρ · WΣ that is contained in Homp(Σ, ∆) (see Lemma 1.3.1). That is {θ ∗ ρ} = Homp(Σ, ∆) ∩ (θ · ρ · WΣ). One checks easily that ∗ : (θ, ρ) 7→ θ ∗ ρ is a left action of Aut(∆) on Homp(Σ, ∆), and that this action commutes with the right action · of Aut(ΠΣ) on the same set. (The latter fact uses only the observation that ϕ−1WΣϕ ⊆ WΣ for ϕ ∈ Aut(ΠΣ).) 1.5. The sets Homsh(Σ, ∆) and Hompsh(Σ, ∆). Suppose ∆ is irreducible. Let Homsh(Σ, ∆) := {ρ ∈ Hom(Σ, ∆) : ρ(Σ) ∩ ∆sh 6= ∅}. Note that Aut(∆)·Homsh(Σ, ∆)·Aut(Σ) ⊆ Homsh(Σ, ∆), so Aut(Σ) (resp. Aut(∆)) acts on the right (resp. left) on Homsh(Σ, ∆) by composition. Also let Hompsh(Σ, ∆) := Homp(Σ, ∆) ∩ Homsh(Σ, ∆). Again Aut(Π∆) · Hompsh(Σ, ∆) · Aut(ΠΣ) ⊆ Hompsh(Σ, ∆), and hence Aut(ΠΣ) (resp. Aut(Π∆)) acts on the right (resp. left) on Hompsh(Σ, ∆) by composition. Finally, it is clear that Aut(∆) ∗ Hompsh(Σ, ∆) ⊆ Hompsh(Σ, ∆), so we have a left action ∗ of Aut(∆) on Hompsh(Σ, ∆). 2. Root graded Lie algebras We will assume for the rest of the article that K is a commutative associative 6 . (We will add the additional assumption that K is an ring of scalars containing 1 algebraically closed field of characteristic 0 in our classification results.) All modules, algebras, trilinear pairs and triple systems will be assumed to be over K. DYNKIN DIAGRAMS AND KANTOR PAIRS 6 2.1. Terminology for graded structures. For graded algebras and graded trilin- ear pairs, we will use the unmodified terms simple and isomorphic in the ungraded sense. More specifically, suppose G be an abelian group. A G-graded Lie alge- bra L will be said to be simple (resp. graded simple) if the only non-trivial proper ideals (resp. graded-ideals) of L are 0 and L. If L and L′ are G-graded algebras, we will say that L is isomorphic (resp. graded-isomorphic) to L′, written L ≃ L′ (resp. L ≃gr L′), if there is an isomorphism (resp. graded-isomorphism) of L onto L′. We will use similar (and evident) terminology and notation for graded trilinear pairs (see Subsection 6.1). If L is an algebra, two G-gradings of L are said to be isomorphic if the cor- responding graded algebras are graded-isomorphic. (The modifier graded in not needed in this term since there is no ambiguity.) Again we will use similar termi- nology for gradings of trilinear pairs. If L is a G-graded Lie algebra, an automorphism ω of L is said to be grade reversing if ω(Lγ) = L−γ for γ ∈ G. If L is a G-graded algebra and θ ∈ Aut(G), we let θL be the G-graded algebra such that θL = L as algebras and for α ∈ G. We call θL the θ-image of L. Clearly 1L = L and (θL)α = Lθ−1α θ1(θ2 L) = θ1θ2L for θ1, θ2 ∈ Aut(G). (2) 2.2. Root graded Lie algebras. Let ∆ be a root system. As in [AFS], a ∆-grading of a Lie algebra L will mean a Q∆-grading of L such that suppQ∆ (L) ⊆ ∆ ∪ {0}, where suppQ∆ (L) denotes the support of L in Q∆. In that case we call L together with the ∆-grading a ∆-graded Lie algebra. (We do not assume the existence of a grading subalgebra as in [ABG] and [BS], or equivalently a family of sl2-triples as in [N1].) Finally, if ∆ is irreducible of type Xn, where n = n∆, we sometimes call a ∆-graded Lie algebra an Xn-graded Lie algebra. If Π∆ is a base for ∆, we often use the basis Π∆ for Q∆ to identify Q∆ with Zn∆ , and in this way ∆-graded Lie algebras are Zn∆ -graded Lie algebras. If L is a ∆-graded Lie algebra and θ ∈ Aut(∆), then the θ-image θL of L is a ∆-graded Lie algebra. If θ ∈ W∆, we call θL a Weyl image of L. 3. Root gradings of finite dimensional semi-simple Lie algebras Suppose in this section that K is an algebraically closed field of characteristic 0, and that G is a finite dimensional semisimple Lie algebra over K. Fix a Cartan subalgebra H of G, let G = ⊕µ∈H∗Gµ be the root space decomposition of G relative to H, and let Σ = Σ(G, H) be the root system of G relative to H (see Subsection 1.1). We fix a base ΠΣ for this root system. 3.1. Gradings of G by a finitely generated free abelian group. We first describe the G-gradings of G up to isomorphism, where G is a free abelian group of finite rank written additively. If ρ ∈ Hom(QΣ, G), let G(ρ) be the G-graded Lie algebra whose underlying Lie algebra is G and whose grading is defined by for γ ∈ G. We call this grading the ρ-grading of G. G(ρ)γ =Pµ∈QΣ, ρ(µ)=γ Gµ (3) DYNKIN DIAGRAMS AND KANTOR PAIRS It is easily checked that θG(ρ) = G(θ · ρ) as G-graded Lie algebras for θ ∈ Aut(G) and ρ ∈ Hom(QΣ, G). 7 (4) Remark 3.1.1. The gradings on θL and G(ρ) defined above are each examples of gradings induced by homomorphisms as defined in [EK, §1.3]; and (2) and (4) follow from an evident general fact about induced gradings. Nevertheless, we use different notations for θL and G(ρ) to emphasize the different roles they play in our theory. The first statement of the next proposition is a corollary of a more general result on graded algebras proved in [EK] using methods from algebraic groups (see [EK, Prop. 1.34 and Cor. 1.35]). It seems likely that the second and third statements are also known, although we are not aware of a reference. Nevertheless, for the reader's convenience, we give a proof of all three statements using Lie algebra methods. Proposition 3.1.2. Suppose that G is a finitely generated free abelian group. (i) Any G-grading of G is isomorphic to a ρ-grading for some ρ ∈ Hom(QΣ, G). (ii) If ρ, ρ′ ∈ Hom(QΣ, G), then the ρ-grading and the ρ′-grading are isomorphic lie in the same orbit in Hom(QΣ, G) under the right if and only if ρ, ρ′ action of Aut(Σ) by composition. (iii) If G =Lγ∈G Gγ is a G-grading of G, there exists a period 2 grade reversing automorphism of G. So dim(G−γ) = dim(Gγ) for γ ∈ G. Proof. If G = 0 the statements are trivial, so we can assume that G = Zℓ, where ℓ ≥ 1. (i) Suppose that G is G-graded. Define derivations d1, . . . , dℓ of G by di(x) = kix for x ∈ G(k1,...,kℓ). Since G is semisimple, every derivation is inner [H, Thm. 5.3], so we can write di = ad(hi) with hi ∈ G for 1 ≤ i ≤ n. Since the di are semisimple and commute, the hi span an abelian ad-diagonalizable subalgebra of G, and hence they are contained in a Cartan subalgebra H′ of G [H, Cor. 15.3]. By [H, Cor. 16.4], we may choose η ∈ Aut(G) with η(H′) = H. Using η to transfer the grading, we can assume that H′ = H. Define ρi ∈ Hom(QΣ, Z) by ρi(µ) = µ(hi) for µ ∈ QΣ. Then, di(x) = [hi, x] = ρi(µ)x for x ∈ Gµ. Thus, the grading is the ρ-grading of G, where ρ(α) = (ρ1(α), . . . , ρℓ(α)) for α ∈ QΣ. (ii) Suppose first that ρ = ρ′ · ϕ, where ρ, ρ′ ∈ Hom(QΣ, G), ϕ ∈ Aut(Σ). Then [H, Thm. 14.2] tells us that there is η ∈ Aut(G) with η(Gµ) = Gϕ(µ) for µ ∈ QΣ. Hence for γ ∈ G we have η(G(ρ)γ ) =Pµ∈QΣ, ρ(µ)=γ Gϕ(µ) =Pν∈QΣ, ρ′(ν)=γ Gν = G(ρ′)γ. so the ρ and ρ′-gradings are isomorphic. Conversely, suppose that ρ, ρ′ ∈ Hom(QΣ, G) and η : G(ρ) → G(ρ′) is a G-graded- isomorphism. Clearly H is a Cartan subalgebra of G(ρ)0, so η(H) and H are Cartan subalgebras of G(ρ′)0. By Corollary 16.4 of [H], there exists η′ ∈ Aut(G) such that η′η(H) = H and η′ is a product of automorphisms of G the form exp(ad(x)), where x ∈ G(ρ′)0 and the transformation ad(x) of G is nilpotent. Clearly, η′ is a graded automorphism of G(ρ′), so we can replace η by η′η to assume that η(H) = H. Let H∗ be the dual space of H and let η♯ ∈ GL(H∗) be the inverse dual of ηH ∈ GL(H), so (η♯(µ))(η(h)) = µ(h) for h ∈ H, µ ∈ H∗. Then η(Gµ) = Gη♯(µ) for µ ∈ H∗, so η♯(Σ) = Σ. Hence ϕ := η♯QΣ ∈ Aut(Σ). Thus for µ ∈ Σ, we have Gϕ(µ) = η(Gµ) ⊆ η(G(ρ)ρ(µ)) = G(ρ′)ρ(µ). So ρ′(ϕ(µ)) = ρ(µ) for µ ∈ Σ, and hence ρ = ρ′ · ϕ. DYNKIN DIAGRAMS AND KANTOR PAIRS 8 (iii) By (i), we can assume that the given grading of G is the ρ-grading where ρ ∈ Hom(QΣ, G). Choose ω ∈ Aut(G) of period 2 of G such that ω(Gµ) = G−µ for µ ∈ QΣ [B, VIII, §4.4, Prop. 5]. Then, by (3), we have ω(G(ρ)γ) = G(ρ)−γ for γ ∈ G. (cid:3) 3.2. Classification of the root gradings of G. The following lemma is clear from the definitions involved. If ∆ is a root system and ρ ∈ Hom(QΣ, Q∆), then the ρ-grading Lemma 3.2.1. of G is a ∆-grading if and only if ρ ∈ Hom(Σ, ∆). Moreover, if ∆ is irreducible and ρ ∈ Hom(Σ, ∆), then G(ρ)α 6= 0 for some α ∈ ∆sh if and only if ρ ∈ Homsh(Σ, ∆). We now classify the ∆-gradings of G up to isomorphism in terms of orbits in Homp(Σ, ∆). Theorem 3.2.2. Suppose ∆ is a root system with base Π∆. Then (i) Any ∆-grading of G is isomorphic to the ρ-grading of G for some ρ ∈ Homp(Σ, ∆), where Homp(Σ, ∆) is the space of positive homomorphisms of Σ into ∆ relative to ΠΣ and Π∆. (ii) If ρ and τ are in Homp(Σ, ∆), then the ρ and τ -gradings of G are isomorphic if and only if ρ and τ are in the same orbit in Homp(Σ, ∆) under the right action of Aut(ΠΣ) by composition. Proof. (i): By Proposition 3.1.2(i), we can assume the grading is a ρ-grading for some ρ ∈ Hom(QΣ, Q∆). Then, by Lemma 3.2.1, ρ ∈ Hom(Σ, ∆). Thus, by Lemma 1.3.1 and Proposition 3.1.2(ii), we can assume that ρ ∈ Homp(Σ, ∆). (ii): The implication "⇐" follows from Proposition 3.1.2(ii). For the converse, suppose that the ρ and τ -gradings are isomorphic. Then, by Proposition 3.1.2(ii), there exists ϕ ∈ Aut(Σ) such that τ = ρ · ϕ. By [B, VI, § 1.5, Prop. 16], we may write ϕ = πw, where π ∈ Aut(ΠΣ) and w ∈ WΣ. So τ · w−1 = ρ · π is positive, and hence by uniqueness in Lemma 1.3.1, τ = τ · w−1. Thus τ = ρ · π as desired. (cid:3) 4. Kantor pairs We assume again that K is a commutative associative ring containing 1 6 . 4.1. Trilinear pairs and triple systems. A trilinear pair is by definition a triple (P, { , , }−, { , , }+), consisting of a pair P = (P −, P +) of K-modules together with two trilinear products { , , }σ : P σ × P −σ × P σ → P σ, σ = ±. The map { , , }σ is called the σ-product for P , σ = ±. We usually abbreviate (P, { , , }−, { , , }+) as P or (P −, P +). If P = (P −, P +) is a trilinear pair, the opposite of P is the trilinear pair P op, with (P op)σ = P −σ and with σ-product equal to the −σ-product of P for σ = ±. A triple system (sometimes also called a ternary algebra) is a pair (X, { , , }) consisting of a K-module X together with a trilinear product { , , } : X × X × X → X. Again we usually write (X, { , , }) simply as X. We have evident notions of ideal, simplicity and isomorphism for trilinear pairs [AFS, §2.1] and for triple systems. There are two natural ways to obtain a trilinear pair from a triple system X. Indeed, if ξ = ±1, the ξ-double of a triple system X is the trilinear pair (X, X) with σ-products defined by { , , }+ = { , , } and { , , }− = ξ{ , , }. We call the 1-double (resp. the (−1)-double) of X the double of X (resp. the signed double) of X. DYNKIN DIAGRAMS AND KANTOR PAIRS 9 A polarization of a triple system X is a module decomposition X = X − ⊕ X + such that {X σ, X −σ, X σ} ⊆ X σ, {X σ, X σ, X} = 0 and {X, X σ, X σ} = 0 for σ = ±. We say that X is non-polarized if X does not have a polarization. As noted in [AFS, §2.1], the following fact is easy to verify: Lemma 4.1.1. Suppose X is a triple system and ξ = ±1. Then the ξ-double of X is a simple pair if and only if X is non-polarized and simple. We say that two triple systems X and X ′ are isotopic if there exist linear iso- morphisms ϕσ : X → X ′ such that ϕσ{x, y, z} = {ϕσx, ϕ−σy, ϕσz} for σ = ±. (In [K1, K2], Kantor uses the term weakly isomorphic rather than isotopic.) Note that if X and X ′ are triple systems, then X and X ′ are isotopic if and only if their doubles (resp. their signed doubles) are isomorphic. 4.2. Kantor pairs. If P is a trilinear pair, σ = ±, x ∈ P σ, y ∈ P −σ and z ∈ P σ, we define Dσ(x, y) ∈ End(P σ) and K σ(x, z) ∈ Hom(P −σ, P σ) by Dσ(x, y)u = {x, y, u}σ and K σ(x, z)w = {x, w, z}σ − {z, w, x}σ. for u ∈ P σ, w ∈ P −σ, σ = ±. We then say that P is a Kantor pair [AF1, AFS] if the following identities hold: [Dσ(x, y), Dσ(z, w)] = Dσ(Dσ(x, y)z, w) − Dσ(z, D−σ(y, x)w), K σ(x, z)D−σ(w, u) + Dσ(u, w)K σ(x, z) = K σ(K σ(x, z)w, u) for x, z, u ∈ P σ, y, w ∈ P −σ, σ = ±, Clearly, the opposite of a Kantor pair is a Kantor pair. A (linear) Jordan pair is a Kantor pair P such that K σ(P σ, P σ) = 0 for σ = ±1. These pairs make up the best understood and most studied class of Kantor pairs [L, N1]. The reader can consult [AFS] (and its references) to see many other examples of Kantor pairs. 4.3. Freudenthal-Kantor triple systems. If ξ = ±1, a (−ξ, 1)-Freudenthal- Kantor triple system (abbreviated as (−ξ, 1)-FKTS), is a triple system X whose ξ-double is a Kantor pair [YO]. The defining identities for such a triple system are: [L(x, y), L(z, w)] = L(L(x, y)z, w) − ξL(z, L(y, x)w), ξK(x, z)L(w, u) + L(u, w)K(x, z) = K(K(x, z)w, u) for x, y, z, w, u ∈ X, where L(x, y), K(x, z) ∈ End(X) are defined by L(x, y)z = {x, y, z}, K(x, z)y = {x, y, z} − {z, y, x}. Example 4.3.1. We now mention two special cases which together with Remark 4.3.2 explain the names used in the term Freudenthal-Kantor triple system. (i): We call a (−1, 1)-FKTS a Kantor triple system, since these triple systems were studied in depth by Kantor in [K1, K2] (where they were called generalized Jordan triple systems of second order). (ii): Suppose that K is a field. A (1, 1)-FKTS X is said to be balanced if K(x, y) = hx, yi idX for some bilinear form h , i : X × X → K [EKO, §1]. In that case h , i is a uniquely determined skew-symmetric form that we call the skew form of X. If h , i is non-degenerate, we see in the following remark that X is a Freudenthal triple system [M1] with a slightly modified product. DYNKIN DIAGRAMS AND KANTOR PAIRS 10 Remark 4.3.2. There are three other classes of triple systems which have been studied in the literature, whose definitions involve a skew form, and which are none other than balanced (1, 1)-FKTS's with slightly modified products. To be more precise, suppose K is a field, X is a finite dimensional vector space, h , , i : X × X × X → X is a trilinear product and h , i : X × X → K is a non-degenerate skew-symmetric form. Then (X, h , , i, h , i) is a balanced symplectic ternary algebra [FF], a symplectic triple system [YA], or a Freudenthal triple system if and only if (X, { , , }) is a balanced (1, 1)-FKTS with skew form h , i, where {x, y, z} = hz, x, yi, {x, y, z} = − 1 2 hx, y, zi + 1 2 hx, yiz or {x, y, z} = − 1 2 hx, y, zi + 1 2 hx, yiz + 1 2 hz, xiy + 1 2 hz, yix respectively. (This is easy to check in the first case. For the other two cases see [E1, Thm. 2.18] and [E2, Thm. 4.7].) Here for the definition of a Freudenthal triple system we use the one given in [M1] except that we allow the quartic form h , h , , ii to be trivial; in other words we allow the product h , , i in a Freudenthal triple system to be trivial. 4.4. BC1-graded Lie algebras and Kantor pairs. The motivation for the study of Kantor pairs is their relationship with BC1-graded Lie algebras. To recall this from [AFS] suppose that ∆ = ∆BC1 := {−2α1, −α1, α1, 2α1} is the irreducible root system of type BC1 with base Π∆ = {α1}. We identify Q∆ = Z using the Z-basis Π∆ for Q∆. Then a BC1-graded Lie algebra is the same thing as a 5-graded Lie algebra L = L−2 ⊕ L−1 ⊕ L0 ⊕ L1 ⊕ L2. If L = L−2 ⊕ L−1 ⊕ L0 ⊕ L1 ⊕ L2 is a 5-graded Lie algebra, then P = (L−1, L1) is a Kantor pair with products defined by {x, y, z}σ = [[x, y], z] for x, z ∈ Lσ1, y ∈ L−σ1, σ = ± [AF1, Thm. 7]. We call P the Kantor pair enveloped by the 5-graded Lie algebra L, and we say that the 5-graded Lie algebra L envelops P . Note that ∆sh = {±α1}. So if L is a 5-graded Lie algebra and P is the Kantor Lα in L. pair enveloped by L, then P − ⊕ P + =Lα∈∆sh Every Kantor pair is enveloped by some 5-graded Lie algebra. Indeed, given a Kantor pair P , there exists a 5-graded Lie algebra K(P ), which is unique up to graded-isomorphism, such that K(P ) envelops P , K(P ) is generated as an algebra by T , where T = K(P )−1 + K(P )1, and the centre of K(P ) intersects trivially with [T, T ] (see [AFS, Cor. 3.5.2]). The 5-graded Lie algebra K(P ) is called the Kantor Lie algebra of P , and it is constructed explicitly in [AF1, §3 -- 4] (see also [AFS, §3.3]) with K(P )σ1 ≃K-mod P σ, K(P )σ2 ≃K-mod K σ(P σ, P σ), K(P )0 = [K(P )−1, K(P )1] (5) for σ = ±, where ≃K-mod indicates isomorphism as K-modules. Note that isomorphisms of Kantor pairs induce graded isomorphisms of their Kantor Lie algebras (see [AFS, §3.5]), and hence two Kantor pairs are isomorphic if and only if their Kantor Lie algebras are graded-isomorphic. DYNKIN DIAGRAMS AND KANTOR PAIRS 11 Lemma 4.4.1. Suppose that P is a Kantor pair and ξ = ±1. If P is isomorphic to the ξ-double of a trilinear pair, then there is grade reversing automorphism ω of K(P ) such that ω2 P −+P + = ξ idP −+P + . Conversely if L is a 5-graded Lie algebra that envelops P and there is grade reversing automorphism ω of L satisfying ω2 P −+P + = ξ idP −+P + , then P is isomorphic to the ξ-double of the triple system P − with product {x, y, z} = {x, ωy, z}−. Proof. (See also [L, pp. 5 -- 6] and [At, Thm. 1] for related observations.) For the first statement we can assume that P equals the ξ-double of a trilinear pair X. Then (idX , ξ idX ) is an isomorphism of P onto P op, which therefore induces an iso- morphism ω of K(P ) onto K(P op). Since K(P op) is K(P ) with the grading reversed, we have the first statement. For the second statement, an easy calculation shows that (idP − , ωP + ) is an isomorphism of P onto the ξ-double of the triple system P − with the indicated product. (cid:3) If P is a Kantor pair, then [GLN, Prop. 2.7(iii)] P is simple if and only if K(P ) is simple. (6) Hence, any simple Kantor pair is enveloped by a simple 5-graded Lie algebra. Con- versely we have the following proposition: Proposition 4.4.2. [AFS, Thm. 3.5.5 and Lemma 3.2.3] Let P be a nonzero Kan- tor pair and suppose that L is a simple 5-graded Lie algebra that envelops P . Then P is simple, L ≃gr K(P ) and L = [L−1, L−1] ⊕ L−1 ⊕ [L−1, L1] ⊕ L1 ⊕ [L1, L1] (7) 4.5. Balanced dimension and close-to-Jordan pairs. Suppose in this subsec- tion that K is a field. We say that a Kantor pair P is finite dimensional if each P σ is finite dimensional. In view of (5), P is finite dimensional if and only if K(P ) is finite dimensional. As in [AFS, §2.1], we say that a Kantor pair P has balanced dimension, if dim(P +) = dim(P −) = d, where d is a non-negative integer, in which case we call d the balanced dimension of P . Similarly, as in [AFS, §3.8], we say P has balanced 2-dimension, if dim(K +(P +, P +)) = dim(K −(P −, P −)) = e, where e is a non-negative integer, in which case we call e the balanced 2-dimension of P . Note that by (5), the balanced dimension of P (resp. the balanced 2-dimension of P ), is given, when it is defined, by d = dim(K(P )σ1) (resp. e = dim(K(P )σ2)) for σ = ±. Lemma 4.5.1. If P is finite dimensional and P op ≃ P , then P has balanced dimension and balanced 2-dimension. Proof. We have dim(P +) = dim((P op)−) = dim(P −) and dim(K +(P +, P +)) = dim(K −((P op)−, (P op)−)) = dim(K −(P −, P −)). (cid:3) Note that a Kantor pair P is Jordan if and only P has balanced 2-dimension 0. With this in mind, we view balanced 2-dimension (when it is defined) as a measure of distance from Jordan theory. In this spirit, we make the following definition. Definition 4.5.2. A close-to-Jordan Kantor pair, or simply a close-to-Jordan pair, is a Kantor pair of balanced 2-dimension 1. DYNKIN DIAGRAMS AND KANTOR PAIRS 12 4.6. Finite dimensional simple Kantor pairs. Suppose in this subsection that K is an algebraically closed field of characteristic 0. Proposition 4.6.1. A trilinear pair P is a finite dimensional simple Kantor pair if and only it is isomorphic to the double of a finite dimensional non-polarized simple Kantor triple system. Moreover in that case P op ≃ P , and hence P has balanced dimension and balanced 2-dimension. Proof. By Lemma 4.1.1, we have the implication "⇐" in the first statement. Sup- pose conversely that P is a finite dimensional simple Kantor pair. By (6) and Proposition 3.1.2(iii), we can choose a period 2 grade reversing ω ∈ Aut(K(P )) . Then ω exchanges P + and P −, so (ω P − , ω P + ) is an isomorphism of P onto P op. The rest now follows from Lemmas 4.4.1 and 4.5.1 (cid:3) Remark 4.6.2. We see from Proposition 4.6.1 that the problem of classifying finite dimensional simple Kantor pairs up to isomorphism is equivalent to the problem of classifying finite dimensional non-polarized simple Kantor triple systems up to isotopy. If P is a finite dimensional simple Kantor pair over K, we define the type of P to be the type of the simple Lie algebra K(P ). Proposition 4.6.3. Suppose P is a nonzero Kantor pair. If L is any finite di- mensional simple 5-graded Lie algebra that envelops P , then P is finite dimensional and simple, L ≃gr K(P ), the type of P is the same as the type of L, the balanced dimension of P equals dim(Lσ1) for σ = ±, and the balanced 2-dimension of P equals dim(Lσ2) for σ = ±. Proof. This follows from Propositions 4.4.2 and 4.6.1. (cid:3) 5. Classification of finite dimensional simple Kantor pairs Suppose in this section that K is an algebraically closed field of characteristic 0 and that Π is the connected Dynkin diagram of type Xn. We will use certain admissible subsets of Π, or equivalently certain markings of Π, to classify the finite dimensional simple Kantor pairs over K. For this purpose, we let G be a finite dimensional simple Lie algebra over K of type Xn with Cartan subalgebra H and root system Σ = Σ(G, H) relative to H (see Subsection 1.1); and let ΠΣ be a base for this root system. Then there exists a diagram isomorphism ι : Π → ΠΣ. For simplicity we use ι to identify Π = ΠΣ. Let µ+ be the highest root of Σ relative to Π, µ− := −µ+, and Π := Π ∪ {µ−}. The Dynkin diagram for Π (or simply Π) is called the extended Dynkin diagram (or completed Dynkin graph in [B]) for Π. We also assume as in Subsection 4.4 that ∆ = ∆BC1 := {−2α1, −α1, α1, 2α1} is the irreducible root system of type BC1 with base Π∆ = {α1}, and we identify Q∆ = Z using the Z-basis Π∆ for Q∆. Note that Aut(∆) = W∆ = {±1}. DYNKIN DIAGRAMS AND KANTOR PAIRS 13 5.1. Kantor-admissible subsets of Π. To emphasize the role of BC1 here, we write the sets Hom(Σ, ∆), Homsh(Σ, ∆), Homp(Σ, ∆), and Hompsh(Σ, ∆) respec- tively as Hom(Σ, BC1), Homsh(Σ, BC1), Homp(Σ, BC1) and Hompsh(Σ, BC1). Then we have Hom(Σ, BC1) = {ρ ∈ Hom(QΣ, Z) : ρ(Σ) ⊆ {0, ±1, ±2}}, Homsh(Σ, BC1) = {ρ ∈ Hom(Σ, BC1) : 1 ∈ ρ(Σ)}. If S ⊆ Π, we define χS ∈ Hom(QΣ, Z) by χS(λ) =(cid:26) 1 0 if λ ∈ S if λ ∈ Π \ S, (8) (9) and we set χλ = χ{λ} for λ ∈ Π. Note that χ∅ = 0. A subset S of Π is said to be Kantor-admissible if χS ∈ Homsh(Σ, BC1) (or equivalently χS ∈ Hompsh(Σ, BC1)). Let KA(Π) = the set of all Kantor-admissible subsets of Π. Remark 5.1.1. The set KA(Π) (or equivalently the term Kantor-admissible) has been defined using our choice of G, H, ΠΣ and ι : Π → ΠΣ. However, it is easy to see (since Dynkin diagram isomorphisms induce root system isomorphisms), that a different choice of these objects leads to the same set KA(Π). In short, KA(Π) is well-defined. Proposition 5.1.2. The map S 7→ χS is a bijection of the set KA(Π) onto the set Hompsh(Σ, BC1). Proof. All but surjectivity is clear. For surjectivity, suppose ρ ∈ Hompsh(Σ, BC1). Then 1 ∈ ρ(Σ+), so 1 ∈ ρ(Π). Note also thatPλ∈Π λ ∈ Σ+, since Σ is irreducible. So 0 ≤Pλ∈Π ρ(λ) ≤ 2. At least one term in this sum is 1, so all are 0 or 1. Let S = {λ ∈ Π : ρ(λ) = 1}. Then ρ and χS agree on Π, so ρ = χS. Finally S ∈ KA(Π) by definition. (cid:3) We have the following characterization of Kantor-admissible subsets of Π. Proposition 5.1.3. If S is a subset of Π, then the following are equivalent: (a) S ∈ KA(Π). (b) S 6= ∅ and χS ∈ Hom(Σ, BC1). (c) χS(µ+) ∈ {1, 2}. (d) S = {λ} with χλ(µ+) ∈ {1, 2}; or S = {λ, λ′} with λ 6= λ′ and χλ(µ+) = χλ′ (µ+) = 1. Proof. The implications in cyclic order are clear using (8). (cid:3) We define a right action of Aut(Π) on the set of subsets of Π by S · ϕ := ϕ−1(S). One checks that for S ⊆ Π and ϕ ∈ Aut(Π). Hence the right action of Aut(Π) stabilizes the set KA(Π), so we have a right action · of Aut(Π) on KA(Π). χS · ϕ = χS·ϕ. (10) DYNKIN DIAGRAMS AND KANTOR PAIRS 14 5.2. Classification. We next use Kantor's approach for constructing Kantor triple systems [K1] to construct Kantor pairs. Construction 5.2.1. (The Kantor pair P(Π; S)) Let S ∈ KA(Π). Then χS ∈ Homsh(Σ, BC1), so G(χS) is a 5-graded Lie algebra with G(χS)−1 6= 0 or G(χS)1 6= 0 by Lemma 3.2.1. Let P(Π; S) be the Kantor pair enveloped by G(χS) (see Section 4.4). Then P(Π; S) = (P(Π; S)−, P(Π; S)+) is nonzero and given explicitly by for σ = ± with products given by {x, y, z}σ = [[x, y], z] in G. P(Π; S)σ = G(χS)σ1 =Pµ∈Σ, χS (µ)=σ1 Gµ Remark 5.2.2. The definition of P(Π; S) uses our choice of G, H, ΠΣ and ι : Π → ΠΣ. It is easy to see (since Dynkin diagram isomorphisms induce Lie algebra isomorphisms) that a different choice of these objects leads to a Kantor pair that is isomorphic to P(Π; S). That is to say, P(Π; S) is well-defined up to isomorphism. Since Π is fixed in our discussion, we will usually abbreviate our notation and write P(Π; S) simply as P(S) Remark 5.2.3. Let S ∈ KA(Π). (i) By Proposition 4.6.3, P(S) is a finite dimensional simple Kantor pair of type Xn and G(χS) ≃gr K(P(S)). (ii) Again using Proposition 4.6.3, we see that the balanced dimension and the balanced 2-dimension of P(S) equal respectively dim(G(χS)1) and dim(G(χS)2). Hence these quantities are given respectively by {µ ∈ Σ : χS(µ) = 1} and {µ ∈ Σ : χS(µ) = 2}. (iii) P(S) is Jordan if and only if it has balanced 2-dimension 0, which by (ii) holds if and only if χS(µ+) = 1. We now state a classification theorem for finite dimensional simple Kantor pairs in terms of Kantor admissible subsets of Π. Theorem 5.2.4 (Kantor). Suppose Π is the connected Dynkin diagram of type Xn. (i) If S ∈ KA(Π), then P(S) is a finite dimensional simple Kantor pair of type Xn. (ii) If P is a finite dimensional simple Kantor pair of type Xn, then P is iso- morphic to P(S) for some S ∈ KA(Π). (iii) If S and S′ are in KA(Π), then P(S) and P(S′) are isomorphic if and only S and S′ are in the same orbit in KA(Π) under the right action of Aut(Π). We have seen (i) above; and (ii) and (iii) follow easily using Theorem 3.2.2, Proposition 4.6.3 and Proposition 5.1.2. We omit the details since we will obtain the theorem as a special case of our classification of finite dimensional simple SP- graded Kantor pairs (see Theorem 7.2.4 and Remark 7.2.5). Remark 5.2.5. We have attributed Theorem 5.2.4 to Isai Kantor, since in [K1, §4] he proved the equivalent classification result (see Remark 4.6.2) for finite di- mensional non-polarized simple Kantor triple systems up to isotopy, although he omitted some details both in his statements and proofs. We note that Kantor also gave models of each of the Kantor triple systems that he considered ([K1, §5 -- 6], [K2]). We will not look at these models in general, but rather focus on models for the reflections of simple close-to-Jordan pairs later in Section 9. DYNKIN DIAGRAMS AND KANTOR PAIRS 15 5.3. Marked Dynkin diagrams for simple Kantor pairs. We represent S ∈ KA(Π) by the Dynkin diagram for Π with the nodes in S marked with a circle; and we use the same marked Dynkin diagram to represent the Kantor pair P(S). Example 5.3.1. (Type E6). Suppose that Π = {µ1, . . . , µ6} is of type E6 with Dynkin diagram µ6 Now µ+ = µ1 + 2µ2 + 3µ3 + 2µ4 + µ5 + 2µ6, and the extended Dynkin diagram Π is µ1 µ2 µ3 µ4 µ5 µ− µ6 µ1 µ2 µ3 µ4 µ5 (11) [B, VI, §4.12, (IV)]. So by Proposition 5.1.3)(d) there are, up to the right action of Aut(Π), four Kantor-admissible subsets S of Π: (i) : {µ1}, (ii) : {µ6}, (iii) : {µ2}, (iv) : {µ1, µ5}. These are represented respectively by the following marked diagrams (i) : E6(16, 0) (ii) : E6(20, 1) (iii) : E6(20, 5) (iv) : E6(16, 8) . So by Theorem 5.2.4 there are four simple Kantor pairs of type E6 up to isomor- phism, which are represented by these marked diagrams. For each S above, we have labelled the marked diagram representing S and P(S) by E6(d, e), where d is the balanced dimension of P(S) and e is the balanced 2-dimension of P(S). (d and e were computed using Remark 5.2.3(ii) and a list of roots in Σ [B, Plate V].) We will sometimes also use the label E6(d, e) for the Kantor pair P(S) itself. Note that the unique pair in the list that is Jordan is E6(16, 0), and the unique pair in the list that is close-to-Jordan is E6(20, 1). For each of the other types Xn, it is easy using the same method to write down the marked Dynkin diagrams representing the simple Kantor pairs of type Xn up to isomorphism. 5.4. The close-to-Jordan case. We next consider the classification in the special case of simple close-to-Jordan pairs of type Xn. By Theorem 5.2.4, there is only one simple finite dimensional Kantor pair of type A1, and it is Jordan so not close- to-Jordan. For the other types, we have the following: Theorem 5.4.1 (Kantor and Skopec). Suppose that Xn 6= A1, and let S be the set of nodes of Π that are adjacent to µ− in Π. (i) χS(µ) = hµ, µ+i for µ ∈ QΣ; so χS(µ+) = 2. (ii) S is the unique element of KA(Π) such that P(S) is close-to-Jordan. (iii) P(S) is, up to isomorphism, the unique finite dimensional simple close-to- Jordan pair of type Xn. DYNKIN DIAGRAMS AND KANTOR PAIRS 16 Proof. Although this result is not stated in this form in [KS], it is implicit in the statement and proof of Theorem 4 in [KS]. For the reader's convenience we give a different proof. (i): We can assume that µ ∈ Π. So µ /∈ Qµ+ (since Xn 6= A1). Hence hµ, µ+i ∈ {0, 1} by [B, VI, §1.8, Prop. 25(iv)]; whereas χS(µ) ∈ {0, 1} by definition of χS. Finally hµ, µ+i 6= 0 iff µ ∈ S iff χS(µ) 6= 0. (ii): If S′ ∈ KA(Π), recall that by Remark 5.2.3(ii), P(S′) is close-to-Jordan if and only if {µ ∈ Σ : χS ′ (µ) = 2} = {µ+}. χS(µ) ≤ 2 − 1 = 1. Now S ∈ KA(Π) by Proposition 5.1.3(c). To see that P(S) is close-to-Jordan, it is enough to show that χS(µ) 6= 2 for µ ∈ Σ+ \ {µ+}. But if µ ∈ Σ+ \ {µ+}, there exists κ ∈ Π such that µ ∈ µ+ − κ −Pλ∈Π Z≥0λ and µ+ − κ ∈ Σ. Thus, κ ∈ S, so Finally, suppose that S′ ∈ KA(Π) and P(S′) is close-to-Jordan. So χS ′(µ+) = 2. If there exists κ ∈ S such that κ /∈ S′, then µ+ −κ ∈ Σ+ \{µ+} and χS ′(µ+ −κ) = 2, a contraction. So S ⊆ S′. Moreover, this inclusion is not proper, since otherwise we would have 2 = χS(µ+) < χS ′ (µ+). So S′ = S (iii) follows from (ii) and Theorem 5.2.4. (cid:3) Example 5.3.1 provides an illustration of Theorem 5.4.1(iii) in type E6. Remark 5.4.2. Suppose we have the assumptions of Theorem 5.4.1. (i) One sees checking types case-by-case that card(S) = 1 if Xn 6= An. (ii) The 5-grading of G(χS) induces a 5-grading of the set Σ ∪ {0} (which is viewed as a root system in [LN]). Since χS(λ) = hλ, µ+i for λ ∈ QΣ, one sees that this 5-grading of Σ ∪ {0} is the one described previously in [LN, §17.10]. We won't need this fact, so we omit the details. 5.5. Close-to-Jordan pairs and Freudenthal-Kantor triple systems. Sup- pose in this subsection that S ∈ KA(Π) and P = P(S) is close-to-Jordan. Recall by Theorem 5.4.1(ii) that Xn 6= A1 and S is the set of nodes of Π that are adjacent to µ− in Π. Choose nonzero eσ ∈ Gµσ for σ = ± such that [h+, eσ] = σ2eσ, where h+ = [e+, e−]. Then S := Ke− ⊕ Kh+ ⊕ e+ ≃ sl2(K). Also, since P is close-to-Jordan, we have G(χS)σ2 = Keσ (12) for σ = ±. We set ω := exp(ad e+) exp(− ad e−) exp(ad e+) ∈ Aut(G), in which case (13) We now use sl2-theory to verify some properties of the triple (e−, h+, e+) and ω(eσ) = −e−σ, σ = ±, and ω(h+) = −h+. the automorphism ω. Lemma 5.5.1. We have (i) µ(h+) = hµ, µ+i = χS(µ) for µ ∈ QΣ. (ii) G(χS)i = {g ∈ G : [h+, g] = ig} for i ∈ Z. (iii) ω reverses the 5-grading of G(χS), and hence ω(P σ) = P −σ for σ = ±. (iv) ω2 P −+P + = − idP −+P + . (v) ω(x) = −σ[e−σ, x] for x ∈ P σ, σ = ±. DYNKIN DIAGRAMS AND KANTOR PAIRS 17 (vi) If σ = ±, there exists a unique bilinear form ζ σ : P σ × P −σ → K such that [[x, a], eσ] = ζ σ(x, a)eσ for x ∈ P σ, a ∈ P −σ. Further [x, ωa] = −σζ σ(x, a)eσ, ζ σ(x, a) = ζ−σ(a, x) for x ∈ P σ, a ∈ P −σ; and ζ σ is nondegenerate. (14) (15) (16) (vii) If x, y ∈ P σ and a ∈ P −σ, then [[ωx, a], y] = ζ σ(x, a)ωy (viii) If x, y, z ∈ P σ, then [[x, ωy], z] − [[z, ωy], x] = −ζ σ(x, ωz)y. and ω(cid:0)[[x, a], y](cid:1) = ζ σ(x, a)ωy + [[x, a], ωy]. Proof. (i): The first equality is standard [H, Prop. 8.2(g)] and the second was seen in Theorem 5.4.1(i). (ii) and (iii): (ii) follows from (i), and (iii) follows from (ii) and (13). (iv) and (v): Let T = P − ⊕ P + in G. Then T is an S-submodule of G under the adjoint action. Also, by (ii), P − and P + are respectively the −1 and 1 eigenspaces for ad(h+) in T . So, by sl2-theory [B, VIII, §1.2 -- 1.3], the S-module T is the direct sum of copies of the 2-dimensional irreducible S-module. Then (iv) and (v) follow by well known 2 × 2-matrix calculations (see for example [B, VIII, §1.5] with Xσ = σeσ). (vi): The first statement follows from (12). Then [x, ωa] = σ[x, [eσ, a]] = −σζ σ(x, a)eσ; and (16) follows by applying ω and using (13) and (iv). For nonde- generacy, it is enough to show that if x ∈ P σ and [x, P σ] = 0, then x = 0. For this, let κ be the Killing form. Then 0 = κ([x, P σ], e−σ) = κ(x, [P σ, e−σ]) = κ(x, P −σ) by (iii) and (v). But G(χS)1 and G(χS)−1 are paired by κ, so x = 0. (vii): Using (15), (16) and (v), we have [[ωx, a], y] = −σζ−σ(a, x)[e−σ, y] = ζ σ(x, a)ωy. Then, since ad(e−σ) is a derivation of G which kills P −σ, we have ω(cid:0)[[x, a], y](cid:1) = [[ωx, a], y] + [[x, a], ωy] = ζ σ(x, a)ωy + [[x, a], ωy]. (viii): Using (iv), the first identity of (vii) and (16), we have [[x, ωy], z] − (cid:3) [[z, ωy], x] = [[x, z], ωy] = [[ω2z, x], ωy] = ζ−σ(ωz, x)ω2y = −ζ σ(x, ωz)y. Theorem 5.5.2. A trilinear pair is a finite dimensional simple close-to-Jordan pair if and only if it is isomorphic to the signed double of a nonzero finite dimensional balanced (1, 1)-FKTS with non-degenerate skew-form (see Example 4.3.1(ii)). Proof. "⇒" By Theorem 5.4.1(iii) we can assume that the trilinear pair is P = P(S). Then by Lemma 4.4.1 and Lemma 5.5.1 (parts (iii) and (iv)), P is isomor- phic to the signed double of the triple system X = P − with product {a, b, c} := {a, ωb, c}−. So by definition X is a (1, 1)-FKTS. Moreover X is balanced with skew form ha, bi = −ζ−(a, ωb) by Lemma 5.5.1(viii); h , i is nondegenerate by Lemma 5.5.1(vi); and X 6= 0 since P is simple. "⇐" Conversely suppose that X is a non-zero balanced (1, 1)-FKTS with non- degenerate skew form. Then, we can use the construction from [F, Thm. 1] of a simple 5-graded Lie algebra S = S(X, R(X)) with dim(S)σ2 = 1. (Actually, the construction starts with a balanced symplectic ternary algebra, but we have seen the translation in Remark 4.3.2.) One can easily check that S envelops the signed-double of X. Hence the signed-double of X is a simple close-to-Jordan pair by Proposition 4.6.3. (cid:3) DYNKIN DIAGRAMS AND KANTOR PAIRS 18 In [FF, Cor. 2], the authors used the work of Meyberg [M1] on Freudenthal triple systems, to give constructions (mainly as 2 × 2 matrix systems with Jordan entries) of all balanced symplectic algebras with nondegenerate skew forms. (Note however that there is a missing term (a1 × a2) × b3 in the expression for c in [FF, (3.4)].) In view of Remark 4.3.2, this gives constructions of all finite dimensional balanced (1, 1)-FKTS's with nondegenerate skew forms. Hence, by Theorem 5.5.2, we obtain constructions of all finite dimensional simple close-to-Jordan pairs. 6. SP-graded Kantor pairs Assume again that K is a commutative associative ring containing 1 6 . 6.1. SP-graded Kantor pairs. Suppose P is a trilinear pair. If G is an abelian group and Pg = (P − g is an K-submodule of P σ for g ∈ G, g ) for g ∈ G, where P σ g , P + σ = ±, we write P =Lg∈G Pg to mean that P σ =Lg∈G P σ We say that P =Lg∈G Pg is a G-grading of P if g′′ } ⊆ P σ , P σ g−g′+g′′ {P σ g , P −σ g′ g for σ = ±. for g, g′, g′′ ∈ G, σ = ±. In that case, each Pg is a subpair of P , and we say that P is G-graded. If P is G-graded, we endow P op with the G-grading given by (P op)σ g = P −σ g . We recall from Subsection 2.1 that the unmodified terms simple and isomorphic for G-graded pairs will be used in the ungraded sense, and that we have a notion of isomorphism for G-gradings on a trilinear pair. P = Li∈Z Pi such that P σ If P is a Kantor pair, a short Peirce grading (or SP-grading) of P is a Z-grading In that case we have P = P0 ⊕ P1, and we call the graded pair P a short Peirce graded (or SP-graded ) Kantor pair. i = 0 for σ = ± and i 6= 0, 1. Any Kantor pair P has at least two SP-gradings, the zero SP-grading P = P0 with P1 = 0, and the one SP-grading P = P1 with P0 = 0. We call these two SP-gradings trivial. Clearly, the opposite of an SP-graded Kantor pair is an SP-graded Kantor pair. 6.2. BC2-graded Lie algebras and SP-graded Kantor pairs. We now recall from [AFS, §4] the relationship between SP-graded Kantor pairs and BC2-graded Lie algebras. For this purpose, suppose for the rest of this section that ∆ = ∆BC2 := {±α1, ±α2, ±(α1 + α2), ±2α1, ±(2α1 + α2), ±(2α1 + 2α2)}. is the irreducible root system of type BC2 with base Π∆ = {α1, α2}. We identify Q∆ = Z2 using the Z-basis Π∆ for Q∆, so any BC2-graded Lie algebra is a Z2- graded Lie algebra. for brevity. Then the first component grading of L is defined to be the Z-grading If L = L(i,j)∈Z2 L(i,j) is a Z2-graded Lie algebra, we often write L(i,j) as Li,j L =Li∈Z Li,∗, where Li,∗ =Lj∈Z Li,j. Suppose that L is a BC2-graded Lie algebra. Then, L with its first component grading is a 5-graded Lie algebra, which therefore envelops a Kantor pair P = (L−1,∗, L1,∗) = (L−1,0 ⊕ L−1,−1, L1,0 ⊕ L1,1). DYNKIN DIAGRAMS AND KANTOR PAIRS 19 Moreover, P = P0 ⊕ P1, where P σ i = Lσ1,σi for σ = ±, i ∈ Z, is an SP-grading of P [AFS, §4.3]. We call P with this grading the SP-graded Kantor pair enveloped by the BC2-graded Lie algebra L, and we say that the BC2-graded Lie algebra L envelops the SP-graded Kantor pair P . Observe that ∆sh = {±α1, ±(α1 + α2)}. So if L is a BC2-graded Lie algebra and P is the SP-graded Kantor pair enveloped by L, then (as in the BC1 case) P − ⊕ P + =Lα∈∆sh Lα in L. (17) Every SP-graded Kantor pair is enveloped by some BC2-graded Lie algebra. Indeed, if P is an SP-graded Kantor pair, then K(P ) has a unique BC2-grading, called its standard BC2-grading, such that the K(P ) with this grading envelops P [AFS, Prop. 4.4.1]. Note that two SP -graded Kantor pairs are graded-isomorphic if and only if the corresponding Kantor Lie algebras with their standard BC2-gradings are graded- isomorphic [AFS, Prop. 4.4.2]. We know that any simple SP-graded Kantor pair is enveloped by a simple BC2- graded Lie algebra, namely K(P ) with its standard BC2-grading. Conversely, we have the following proposition: Proposition 6.2.1. [AFS, Prop. 4.4.2(iii)] Let P be a nonzero SP-graded Kantor pair and suppose that L is a simple BC2-graded Lie algebra that envelops P . Then L is graded-isomorphic to K(P ) with its standard BC2-grading. Proposition 6.2.2. If K is an algebraically closed field of characteristic 0 and P is a finite dimensional simple SP-graded Kantor pair over K, then P op ≃gr P . Consequently Pi ≃ P op , so the Kantor pair Pi has balanced dimension and balanced 2-dimension for i = 0, 1. i Proof. By Proposition 3.1.2(iii), we can choose ω ∈ Aut(K(P )) of period 2 such that ω(K(P )k,i) = K(P )−k,−i for k, i ∈ Z. Then ω exchanges P + for i = 0, 1, so (ω P −, ω P + ) is a graded-isomorphism of P onto P op. Hence Pi ≃ P op for i = 0, 1, and the proof is complete by Lemma 4.5.1. (cid:3) i and P − i i 6.3. Weyl images of SP-graded Kantor pairs. Let sα ∈ W∆ be the reflection through the hyperplane orthogonal to α for α ∈ ∆, and put si = sαi for i = 1, 2. The generators s1 and s2 of W∆ satisfy s1s2s1s2 = s2s1s2s1 = −1, and Aut(∆) = W∆ = {1, s1, s2, s2s1, −1, −s1, −s2, −s2s1} is the dihedral group of order 8. Let P be an SP-graded Kantor pair and let u ∈ W∆. If we choose a BC2-graded Lie algebra L that envelops P , then uL is also a BC2-graded Lie algebra that therefore envelops an SP-graded Kantor pair which we denote by uP . It turns out that uP is independent of the choice of L [AFS, Lemma 5.1.2(iv)], and we call uP the u-image (or a Weyl image) of P . It is clear that Weyl images respect graded uQ. Moreover, 1P = P , and, by (2), isomorphisms; that is P ≃gr Q =⇒ uP ≃gr u1 (u2 P ) = u1u2 P for u1, u2 ∈ W∆. Since s1 and s2 generate W∆, the SP-graded Kantor pairs s1 P and s2 P are of particular importance. For convenience, we use the notation P := s1 P DYNKIN DIAGRAMS AND KANTOR PAIRS 20 and call this SP-graded Kantor pair the reflection of P . It is easy to check that for σ = ±, i = 0, 1, where π(0) = − and π(1) = + [AFS, Prop. 5.2.1]. Moreover the σ-product {, , }σ on P is given in terms of the products on P by i = P π(i)σ P σ i , (18) {xσ i , y−σ i {xσ , zσ i , y−σ i }σ = {xσ 1−i, zσ i , y−σ i }σ = 0, i i }π(i)σ, {xσ , zσ i , y−σ and {xσ 1−i, y−σ 1−i, zσ i }σ = −{y−σ 1−i, zσ 1−i}σ = K π(i)σ(xσ 1−i, xσ i , y−σ i }π(i)σ, 1−i, zσ 1−i)zσ 1−i. j , yτ j , zτ for σ = ±, i = 0, 1, where xτ j in each case [AFS, Prop. 5.2.1]. However, we will not use these expressions in this article, but rather directly use the definition of P given above. It turns out that in general P is not isomorphic to P as an ungraded Kantor pair (as we saw in [AFS] and will see again in Example 8.3.7). In contrast, the SP-graded Kantor pair s2 P has an easy description. We have j ∈ P τ s2 P = ¯P , where ¯P is the SP-graded Kantor pair, called the shift of P that equals P as a Kantor pair and has Z-grading given by ¯Pi = P1−i for i ∈ Z [AFS, (21)]. In particular, ¯P is isomorphic as an ungraded pair to P . Finally, it is clear that the SP-graded pair −1P is simply P op. Proposition 6.3.1. Suppose P is a simple SP-graded Kantor pair and u ∈ W∆. Then uP is simple. Moreover, if K is an algebraically closed field of characteristic 0 and P is finite dimensional, then uP has the same balanced dimension as P . Proof. The first statement is seen in [AFS, Prop. 4.1.4]. For the second statement, it suffices to prove that ¯P and P have the same balanced dimension as P (since W∆ = hs1, s2i). This is clear for ¯P ; while for P it follows from (18) and the fact that P0 has balanced dimension by Proposition 6.2.2. (cid:3) Remark 6.3.2. Under the assumptions of the second statement of the proposition, it is clear that the shift ¯P of an SP-graded P has the same balanced 2-dimension as P . However that is not true for the reflection P of P (see Example 8.3.7 below). 7. Classification of finite dimensional simple SP-graded Kantor pairs For the rest of this article we assume that K is an algebraically closed field of characteristic 0, and Π is the connected Dynkin diagram of type Xn. We also assume G is a finite dimensional simple Lie algebra over K of type Xn with Cartan subalgebra H and root system Σ = Σ(G, H) relative to H. Let ΠΣ be a base for this root system, in which case there exists a diagram isomorphism ι : Π → ΠΣ, which we use to identify Π = ΠΣ. Also, as in Section 5, µ+ is the highest root of Σ relative to Π, µ− = −µ+ and Π = Π ∪ {µ−} is the extended Dynkin diagram for Π. If Y is a non-empty subset of Π, then Y is the disjoint union of its connected components; and if λ ∈ Y , we use the notation comp(Y, λ) for the connected component of Y containing λ. We further assume that ∆ = ∆BC2 := {±α1, ±α2, ±(α1 + α2), ±2α1, ±(2α1 + α2), ±(2α1 + 2α2)} is the irreducible root system of type BC2 with base Π∆ = {α1, α2}. DYNKIN DIAGRAMS AND KANTOR PAIRS 21 We identify Q∆ = Z2 using the Z-basis Π∆ for Q∆. We then have the identifi- cation Hom(QΣ, Q∆) = Hom(QΣ, Z2) = Hom(QΣ, Z)2, where in the last equality (ρ1, ρ2) in Hom(QΣ, Z)2 is identified with the element of Hom(QΣ, Z2) given by µ 7→ (ρ1(µ), ρ2(µ)). 7.1. SP-admissible pairs of subsets of Π. To emphasize the role of BC2 here, we will write Hom(Σ, ∆), Homsh(Σ, ∆), Homp(Σ, ∆) and Hompsh(Σ, ∆) respectively as Hom(Σ, BC2), Homsh(Σ, BC2), Homp(Σ, BC2) and Hompsh(Σ, BC2). So, setting (cid:3)2 := {0, 1, 2} × {0, 1, 2}, we have Hom(Σ, BC2) = {ρ ∈ Hom(QΣ, Z2) : ρ(Σ) ⊆ (cid:3)2 ∪ (−(cid:3)2), (1, 2) /∈ ρ(Σ), (0, 2) /∈ ρ(Σ)}, Homsh(Σ, BC2) = {(ρ1, ρ2) ∈ Hom(Σ, BC2) : 1 ∈ ρ1(Σ)}. (19) (20) If (S, T ) is a pair of subsets of Π, we use the notation χ(S,T ) := (χS, χT ) ∈ Hom(QΣ, Z2) and say (S, T ) is SP-admissible if χ(S,T ) ∈ Homsh(Σ, BC2) (or equivalently χ(S,T ) ∈ Hompsh(Σ, BC2)). We let SPA(Π) = the set of all SP-admissible pairs of subsets of Π. Remark 7.1.1. Just as in Remark 5.1.1, the set SPA(Π) is well-defined. Proposition 7.1.2. The map (S, T ) 7→ χ(S,T ) is a bijection of the set SPA(Π) onto the set Hompsh(Σ, BC2). Proof. All but surjectivity is clear. For surjectivity, suppose that ρ = (ρ1, ρ2) ∈ Hompsh(Σ, BC2). Then, by (19) and (20), ρ1(Σ) ⊆ {0, ±1, ±2} and 1 ∈ ρ1(Σ). Hence, since ρ1(α) ≥ 0 for α ∈ Σ+, we have ρ1 = χS for some S ⊆ Π by Proposition 5.1.2. Also, if 1 ∈ ρ2(Σ), we have ρ2 = χT for some T ⊆ Π by the same argument. So we can assume that 1 /∈ ρ2(Σ). Hence, by (19), ρ(Σ) ⊆ {(0, 0), ±(1, 0), ±(2, 0), ±(2, 2)}. Therefore, G(ρ)−1,∗ + G(ρ)1,∗ ⊆ G(ρ)∗,0 (using the notation of Section 6.2). But the left hand side of this inclusion is nonzero by Lemma 3.2.1, so this space generates G as an algebra by (7). Therefore G(ρ) = G(ρ)∗,0, so ρ2(Σ) = 0. Thus ρ2 = 0 = χ∅. (cid:3) Lemma 7.1.3. Let ρ = (ρ1, ρ2) ∈ Hom(QΣ, Z2). Then ρ ∈ Homsh(Σ, BC2) if and only if ρ(Σ) ⊆ (cid:3)2 ∪ (−(cid:3)2), (1, 2) /∈ ρ(Σ) and 1 ∈ ρ1(Σ). Proof. In view of (19), we only need to prove "⇐". Let L = G(ρ), which is a Z2-graded Lie algebra with suppZ2(L) = ρ(Σ ∪ {0}). By our assumptions, the first component grading L =Li∈Z Li,∗ of L is a 5-grading with L−1,∗ +L1,∗ 6= 0. Hence, by (7), each element in L has the form x +P[yj, zj], where x, yj, zj ∈ L−1,∗ + L1,∗. Therefore each element of suppZ2(L) is either an element in suppZ2(L−1,∗ + L1,∗) or it is a sum of two such elements. But by our assumptions suppZ2(L−1,∗ + L1,∗) ⊆ {±(1, 0), ±(1, 1)}. Hence (0, 2) /∈ suppZ2 (L) as needed. (cid:3) Lemma 7.1.4. If µ ∈ Σ, then {µ−} ∪ suppΠ{µ+ − µ} is a connected subset of Π. DYNKIN DIAGRAMS AND KANTOR PAIRS 22 Proof. Let S(µ) = {µ−} ∪ suppΠ{µ+ − µ} for µ ∈ Σ. We prove that S(µ) is connected by induction on the non-negative integer htΠ(µ+ − µ). First, if htΠ(µ+ − µ) = 0, then µ = µ+ and S(µ) = {µ−} is connected. Suppose that htΠ(µ+ −µ) > 0. Then µ 6= µ+. Furthermore, we can assume that µ /∈ −Π, since otherwise S(µ) = Π is connected. So there exists λ ∈ Π with ν = µ + λ ∈ Σ. Then, by the induction hypothesis, S(ν) is connected. Further, S(µ) = S(ν) ∪ {λ}. So it remains to show that hS(ν), λi 6= 0. For this we can assume that λ /∈ S(ν), so χλ(ν) = χλ(µ+). Hence ν + λ /∈ Σ. Thus, since ν − λ ∈ Σ, we have hν, λi 6= 0. But then, since ν lies in the group generated by S(ν), we have hS(ν), λi 6= 0. (cid:3) We have the following characterization of SP-admissible pairs of subsets of Π. Proposition 7.1.5. Suppose that S, T ⊆ Π. Then the following are equivalent: (a) (S, T ) ∈ SPA(Π). (b) S 6= ∅ and χ(S,T ) ∈ Hom(Σ, BC2) (c) χS(µ+) ∈ {1, 2}; χT (µ+) ∈ {0, 1, 2}; and if χT (µ+) = 2, then χS(µ+) = 2 and comp( Π \ T, µ−) ∩ S = ∅. Proof. The equivalence of (a) and (b) is clear, so we only consider the equivalence of (b) and (c). For this we can assume that χS(µ+) ∈ {1, 2} and χT (µ+) ∈ {0, 1, 2} (since these statements hold if either (b) or (c) is assumed). Also, if χT (µ+) = 0 or 1, then χT (µ) 6= ±2 for µ ∈ Σ, so (b) and (c) are each true. Thus we can suppose χT (µ+) = 2. Then if χS(µ+) = 1, statements (b) and (c) are each false. So we can suppose χS(µ+) = 2. It remains to show that (S, T ) ∈ SPA(Π) if and only if comp( Π \ T, µ−) ∩ S = ∅. We establish the contrapositives of these implications. Suppose that comp( Π \ T, µ−) ∩ S 6= ∅. Then there is a path λ0, λ1, . . . , λr in Π which does not pass through T with r ≥ 1, λ0 = µ−, λr ∈ S. We can shorten this path if necessary to assume that the λi's are distinct and that λi ∈ Π \ S for 1 ≤ i ≤ r − 1. Then P := {λ0, λ1, . . . , λr} is a non-empty, proper and connected subset of Π, so the diagram for P is the diagram of an irreducible reduced finite root system [Kac, Prop. 4.7(c)]. Hence µP :=Pr −µP = µ+ − λ1 − · · · − λr ∈ Σ+. i=0 λi ∈ Σ. So Since λi /∈ T for 1 ≤ i ≤ r, we have χT (−µP ) = χT (µ+) = 2. Also, since λi 6∈ S for 1 ≤ i ≤ r − 1, we have χS(−µP ) = χS(µ+) − χS(λr) = 2 − 1 = 1. Thus (S, T ) /∈ SPA(Π). Conversely, suppose that (S, T ) /∈ SPA(Π). Then, by Lemma 7.1.3, there exists µ ∈ Σ+ with (χS(µ), χT (µ)) = (1, 2). Let S(µ) = {µ−} ∪ suppΠ{µ+ − µ}. Now χT (µ+ − µ) = 2 − 2 = 0, so T ∩ S(µ) = ∅. Also, χS(µ+ − µ) = 2 − 1 = 1, so S ∩ S(µ) 6= ∅. Since S(µ) is connected by Lemma 7.1.4, S(µ) ⊆ comp( Π \ T, µ−), so comp( Π \ T, µ−) ∩ S 6= ∅. (cid:3) It follows from Propositions 7.1.5(c) and 5.1.3(c) that if (S, T ) ∈ SPA(Π), then S is Kantor-admissible and T is either empty or Kantor-admissible. We define a right action of Aut(Π) on the set of pairs of subsets of Π by (S, T ) · ϕ := (S · ϕ, T · ϕ) = (ϕ−1(S), ϕ−1(T )). Now, by (10), we have χ(S,T ) · ϕ = χ(S,T )·ϕ (21) DYNKIN DIAGRAMS AND KANTOR PAIRS 23 for S, T ⊆ Π, ϕ ∈ Aut(Π). Using this and the fact that Homsh(Σ, BC2) is stabilized by the right action of Aut(Π), we see that the SPA(Π) is stabilized by the right action of Aut(Π). So we have a right action · of Aut(Π) on SPA(Π). 7.2. Classification. Construction 7.2.1. (The SP-graded Kantor pair P(Π; S, T )) Suppose(S, T ) ∈ SPA(Π), in which case χ(S,T ) ∈ Homsh(Σ, BC2). Then, by Lemma 3.2.1, G(χ(S,T )) is a BC2-graded Lie algebra with G(χ(S,T ))α 6= 0 for some α ∈ ∆sh. Let P(Π; S, T ) be the SP-graded Kantor pair enveloped by G(χ(S,T )) (see Section 6.2). Note that P(Π; S, T ) is nonzero by (17). Explicitly P(Π; S, T ) is the Kantor pair P(Π; S) (see Construction 5.2.1) with the SP-grading given by σi = G(χ(S,T ))σ1,σi =Pµ∈QΣ, χS (µ)=σ1, χT (µ)=σi Gµ for σ = ± and i = 0, 1. We call this grading the SP-grading of P(Π; S) determined by T . P(Π; S)σ (22) Remark 7.2.2. Just as in Remark 5.2.2, P(Π; S, T ) is well-defined up to graded- isomorphism. Again since Π is fixed in our discussion we will usually write P(Π; S, T ) as P(S, T ) Remark 7.2.3. Let (S, T ) ∈ SPA(Π). (i) By Proposition 6.2.1, G(χ(S,T )) is graded-isomorphic to K(P(S, T )) with its standard BC2-grading. (ii) If i = 0, 1, P(S, T )i has balanced dimension by Proposition 6.2.2. Moreover, this balanced dimension equals dim(G(χ(S,T ))1,i) by (i), which equals {µ ∈ Σ : χS(µ) = 1, χT (µ) = i}. We have the following classification of simple SP-graded Kantor pairs of type Xn. Theorem 7.2.4. Suppose that Π is the connected Dynkin diagram of type Xn. (i) If (S, T ) ∈ SPA(Π), then P(S, T ) is a finite dimensional simple SP-graded Kantor pair of type Xn. (ii) If P is a finite dimensional simple SP-graded Kantor pair of type Xn, then P is graded-isomorphic to P(S, T ) for some (S, T ) ∈ SPA(Π). (iii) If (S, T ), (S′, T ′) ∈ SPA(Π), then the Kantor pairs P(S, T ) and P(S′, T ′) are graded-isomorphic if and only (S, T ) and (S′, T ′) are in the same orbit in SPA(Π) under the right action of Aut(Π). Proof. (i): This follows from Proposition 4.6.3. (ii): K(P ) is simple of type Xn, so there is an isomorphism η : K(P ) → G. We use η to transport the standard BC2-grading of K(P ) to G, so η is a BC2- graded isomorphism. Next by Theorem 3.2.2(i) we can assume that the BC2- grading of G is the ρ-grading for some ρ ∈ Homp(Σ, BC2). Now K(P )α 6= 0 for some α ∈ ∆sh by (17), and thus G(ρ)α 6= 0 for some α ∈ ∆sh. So by Lemma 3.2.1(ii), ρ ∈ Hompsh(Σ, BC2), and hence by Proposition 7.1.2, ρ = χ(S,T ) for some (S, T ) ∈ SPA(Π). Thus, under the restriction of η, P ≃gr P(S, T ). (iii): Let P = P(S, T ) and P ′ = P(S′, T ′); and let ρ = χ(S,T ) and ρ′ = χ(S ′,T ′) in Hompsh(Σ, BC2). Then by Theorem 3.2.2(ii) and (21), we know that G(ρ) ≃gr G(ρ′) if and only if (S, T ) and (S,′ T ′) are in the same orbit in SPA(Π) under the right DYNKIN DIAGRAMS AND KANTOR PAIRS 24 action of Aut(Π). So it remains to show that P ≃gr P ′ if and only if G(ρ) ≃gr G(ρ′). The implication "⇐" in this statement is clear. To prove the converse, suppose that P ≃gr P ′. Then, as noted in Subsection 6.2, K(P ) ≃gr K(P ′); so, by Remark 7.2.3(i), G(ρ) ≃gr G(ρ′). (cid:3) Remark 7.2.5. We note that if we take T = ∅ everywhere in Theorem 7.2.4, we obtain the classification Theorem 5.2.4. As a corollary, we obtain the following classification up to isomorphism of the SP-gradings on a fixed simple Kantor pair. Corollary 7.2.6. Suppose S ∈ KA(Π). Any SP-grading on P(S) is isomorphic to the SP-grading determined by T for some subset T of Π such that (S, T ) ∈ SPA(Π). Also, for two such subsets T and T ′ of Π, the SP-gradings of P(S) determined by T and T ′ are isomorphic if and only if (S, T ) and (S, T ′) are in the same orbit in SPA(Π) under the right action of Aut(Π). If S ∈ KA(Π), then, by Proposition 7.1.5(c), (S, ∅) and (S, S) are in SPA(Π). Moreover, the SP-gradings of P(S) determined by ∅ and S are respectively the zero SP-grading and the one SP-grading of P(S). Of course our main interest is in non-trivial gradings of P(S), which occur in Corollary 7.2.6 when T is not equal to ∅ or S. 7.3. The close-to-Jordan case. The classification of SP-gradings has a particu- larly simple description for close-to-Jordan pairs. Theorem 7.3.1. Suppose S ∈ KA(Π) and P = P(S) is close-to-Jordan. If T ⊆ Π, then (S, T ) ∈ SPA(Π) if and only if T = ∅, T = S or T = {λ} for some λ ∈ Π with χλ(µ+) = 1. (23) Hence the SP-gradings of P(S) are, up to isomorphism, precisely the SP-gradings determined by subsets of Π of the form (23). Finally, if T and T ′ are subsets of Π of the form (23), then T and T ′ determine isomorphic SP-gradings of P(S) if and only if there exists ϕ ∈ Aut(Π) such that T ′ = T · ϕ. Proof. Recall that Xn 6= A1, S is the set of nodes of Π that are adjacent to µ− in Π and χS(µ+) = 2 (see Theorem 5.4.1). If T of the form (23), then (S, T ) ∈ SPA(Π) by Proposition 7.1.5(c). For the converse, suppose T ⊆ Π with (S, T ) ∈ SPA(Π), and suppose T 6= ∅ and T 6= S. Then it suffices to show that χT (µ+) = 1, so we suppose the contrary. Hence µT (µ+) = 2. But if ν ∈ S \ T , then µ−, ν is a path in Π \ T , so ν ∈ comp( Π, µ−) ∩ S, which is empty by Proposition 7.1.5(c). Therefore S ⊆ T . So since χT (µ+) = 2 = χS(µ+), we have T = S. With this contradiction we have proved the first statement. The second and third statements now follow by Corollary 7.2.6 and the fact that each ϕ ∈ Aut(Π) fixes µ−. (cid:3) Looking at the explicit expression for µ+ for each type, we see the following: Corollary 7.3.2. Suppose that P is the simple close-to-Jordan Kantor pair of type Xn 6= A1. If Xn = G2, F4 or E8, then P does not have a non-trivial SP-grading. For the other types, the number of isomorphism classes of non-trivial SP-gradings of P is ⌊ n+1 2 ⌋ if Xn = An(n ≥ 2), 2 if Xn = Dn(n ≥ 5) and 1 if Xn = Bn(n ≥ 2), Cn(n ≥ 3), D4, E6 or E7. See Example 7.4.1(ii) below for an illustration of Theorem 7.3.1 and Corollary 7.3.2 in type E6. DYNKIN DIAGRAMS AND KANTOR PAIRS 25 7.4. Marked Dynkin diagrams for SP-gradings. We represent an element (S, T ) ∈ SPA(Π) by the Dynkin diagram for Π with the nodes in S marked with a circle and the nodes in T marked with an asterisk; and we use the same marked Dynkin diagram to represent the SP-graded Kantor pair P(S, T ) as well as the SP-grading of P(S) determined by T . Example 7.4.1. (Type E6) Suppose that Π = {µ1, . . . , µ6} is of type E6 with Dynkin diagram as in Example 5.3.1. Recall that in Example 5.3.1 we saw that (up to the right action of Aut(Π)) there are four Kantor-admissible subsets S of Π: (i) {µ1}, (ii) {µ6}, (iii) {µ2}, and (iv) {µ1, µ5}. We now use Proposition 7.1.5(c) and the extended diagram (11) for E6 to list for each choice of S the marked diagrams that represent, up to the right action of Aut(Π), the SP-admissible pairs of the form (S, T ) with T 6= ∅ and T 6= S: (i) : E6(16, 0, 8) (ii) : E6(20, 1, 10) ∗ (iii) : E6(20, 5, 10) ∗ E6(20, 5, 12) (iv) : E6(16, 8, 8a) ∗ ∗ ∗ E6(20, 5, 8) E6(16, 8, 8b) ∗ ∗ Hence, for each S, Corollary 7.2.6 tells us that the non-trivial SP-gradings on P(S) are represented up to isomorphism by the listed marked diagrams. Thus, up to graded-isomorphism, each simple SP-graded Kantor pair of type E6 with non-trivial grading is represented by exactly one of the above seven marked diagrams. Note that each marked diagram representing (S, T ) and P(S, T ) in (i), (ii) and (iii) is labelled as E6(d, e, f), where d is the balanced dimension of P(S, T ) and e is the balanced 2-dimension of P(S, T ) as in Example 5.3.1, and where f is the balanced dimension of P(S, T )1. (f was computed using Remark 7.2.3(ii) and a list of roots in Σ [B, Plate V].) In (iv), we have used the notations E6(16, 8, 8a) and E6(16, 8, 8b) because there are two graded pairs with parameters 16,8,8 that are not graded-isomorphic. In any of the cases, we will sometimes use the label for the SP-graded Kantor pair P(S, T ) itself. For each of the other types Xn, it is not difficult using the same method to write down marked Dynkin diagrams representing up to graded-isomorphism the simple SP-graded Kantor pairs of type Xn with non-trivial gradings. We leave this to the interested reader as an exercise. 8. Weyl images of finite dimensional simple SP-graded Kantor pairs We continue with the assumptions and notation of Section 7. In this section (see Theorems 8.3.1 and 8.3.2), we compute the marked Dynkin diagram that represents the SP-graded Kantor pair uP(S, T ) for each (S, T ) ∈ SPA(Π) and each u ∈ W∆. DYNKIN DIAGRAMS AND KANTOR PAIRS 26 In view of Theorem 7.2.4, this computes all Weyl images of all finite dimensional simple SP-graded Kantor pairs. 8.1. The maps wX , σX and wX . Suppose that X ⊆ Π. Let EX = spanR(X), ΣX = Σ ∩ EX and Σ+ X = Σ+ ∩ EX . If X 6= ∅, then ΣX is a root system with base X in the Euclidean space EX , and Σ+ X is the set of positive roots in ΣX relative to X. We then use the simplified notation QX = QΣX and WX = WΣX respectively for the root lattice and Weyl group of ΣX . If X = ∅, then EX = {0}, ΣX = ∅, Σ+ X = ∅, and we set QX = {0} and WX = {1}. If X 6= ∅, we let wX be the unique element of WX such that wX (Σ+ X ) = −Σ+ X . Equivalently, wX is the longest element of WX relative to the base X [B, VI, §1.6, Cor. 3]. Clearly −wX stabilizes X and we set σX = −wX ∈ Aut(X), where recall that Aut(X) is the group of diagram automorphisms of X. If X = ∅, we adopt the conventions that wX = 1 and σX = 1. Evidently in all cases w2 X = 1 and σ2 X = 1. If X 6= ∅, it is well known that the map σX can be read from the Dynkin diagram for X. Indeed, if X is connected, then σX is the nontrivial diagram automorphism of X if X has type An with n ≥ 2, Dn with n odd ≥ 5, or E6; and σX = 1 for all other types [B, VI, §4.5 -- §4.13]. Furthermore, if X has connected components Xi, 1 ≤ i ≤ r, then σX stabilizes each Xi and σX Xi = σXi . If w ∈ WX , then w extends uniquely to an isometry w of E with w(τ ) = τ for (τ, EX ) = 0. It is clear that the map ϕ 7→ ϕ is a monomorphism of WX into WΣ. We will abuse notation and writegwX as wX . Note that if λ ∈ E, we have wX (λ) ∈ λ + QX. 8.2. The left action ∗ of W∆ on SPA(Π). Recall from Section 1.4 that we have left action ∗ of W∆ = Aut(∆) on Hompsh(Σ, BC2). We use the bijection (S, T ) 7→ χ(S,T ) in Proposition 7.1.2 to transfer this action to a left action ∗ of W∆ on SPA(Π). So by definition we have for u ∈ W∆ and (S, T ) ∈ SPA(Π) that (24) Using (21) and (25), we see that the left action ∗ of W∆ on SPA(Π) commutes with the right action · of Aut(Π) on the same set. χu∗(S,T ) = u ∗ χ(S,T ). (25) 8.3. Weyl images of P(S, T ). Theorem 8.3.1. Suppose that (S, T ) ∈ SPA(Π). Then uP(S, T ) ≃gr P(u ∗ (S, T )) for u ∈ W∆. Proof. By definition of ∗, we have u ∗ χ(S,T ) = u · χ(S,T ) · w for some w ∈ WΣ. Then uG(χ(S,T )) = G(u · χ(S,T )) by (4) ≃gr G(u · χ(S,T ) · w) = G(u ∗ χ(S,T )) = G(χu∗(S,T )) by Proposition 3.1.2(ii) by (25). DYNKIN DIAGRAMS AND KANTOR PAIRS 27 Since uG(χ(S,T )) and G(χu∗(S,T )) determine uP(S, T ) and P(u∗(S, T )) respectively, we have our conclusion. (cid:3) Because of Theorem 8.3.1. it is important to be able to compute u ∗ (S, T ) for u ∈ W∆, (S, T ) ∈ SPA(Π). We do this in the next theorem, where there is a case that requires special treatment, namely the case when: Π has type An, n ≥ 3, and there exists distinct λ, λ′, µ ∈ Π such that S = {λ, λ′}, T = {µ}, and λ′ lies between λ and µ on the Dynkin diagram for Π. (ex) Recall from Section 6.3 that W∆ = {1, s1, s2, s2s1, −1, −s1, −s2, −s2s1}. Theorem 8.3.2. Suppose (S, T ) ∈ SPA(Π). Then the elements u ∗ (S, T ) for u ∈ W∆ are given in the following table u 1 s1 s2 s2s1 −1 −s1 −s2 −s2s1 u ∗ (S, T ) (S, T ) ( S, T ) (S, ¯T ) ( S, ¯T · σΠ) , (26) (S · σΠ, T · σΠ) ( S · σΠ, T · σΠ) (S · σΠ, ¯T · σΠ) ( S · σΠ, ¯T ) S = σΠ\T (S \ T ) [ {λ ∈ T : χS\T ( wΠ\T (λ)) + χS∩T (λ) = 1} ¯T = σΠ\S(T \ S) [ {λ ∈ S \ T : suppΠ( wΠ\S(λ)) ∩ (T \ S) = ∅}. (See Remark 8.3.4 below about the notation S and ¯T .) Moreover, we have the following expressions, which allow us to read S and ¯T directly from the marked Dynkin diagram representing (S, T ) together with the expression for the highest root µ+ in Σ: where and and (27) (28) (29) (30) S σΠ\T (S \ T ) ∪ (T \ S) σΠ\T (S \ T ) if χS\T (µ+) = 0, if χS\T (µ+) = 1, if χS\T (µ+) = 2, S \ T {σΠ\S(µ), λ} σΠ\S(T \ S) if T ⊆ S, if (ex) holds, otherwise. S := ¯T := Proof. We postpone the proofs of (29) and (30) until Section 8.4, since these proofs require some additional information about the Weyl group elements wΠ\S and wΠ\T . We prove the first statement here. DYNKIN DIAGRAMS AND KANTOR PAIRS 28 If λ ∈ Π, we have ((−1) · χ(S,T ) · wΠ)(λ) = −(χS(wΠ(λ)), χT (wΠ(λ))) = (χS(σΠ(λ)), χT (σΠ(λ)), which has non-negative entries. Hence, (−1) ∗ χ(S,T ) = (−1) · χ(S,T ) · wΠ = χ(S,T ) · σΠ = χ(S,T )·σΠ using (21). So using (25), (−1)∗(S, T ) = (S, T )·σΠ = (S ·σΠ, T ·σΠ). This establishes row 5 (not counting the header row) of (26), and we see that rows 6, 7 and 8 follow respectively from rows 2, 3 and 4. Thus it is sufficient to establish rows 2, 3 and 4. Row 2: Let ρ = (ρ1, ρ2) = s1 · χ(S,T ) · wΠ\T ∈ Homsh(Σ, BC2). Recall that we are identifying Q∆ = Z2 using the Z-basis Π∆ = {α1, α2} for Q∆. So s1 · χ(S,T )(λ) = s1(χS(λ)α1 + χT (λ)α2) = χS(λ)(−α1) + χT (λ)(2α1 + α2) = (2χT (λ) − χS(λ), χT (λ)). for λ ∈ Σ. Also, if λ ∈ Π, we have χT ( wΠ\T (λ)) = χT (λ) by (24). Thus ρ1(λ) = 2χT (λ) − χS( wΠ\T (λ)) and ρ2(λ) = χT (λ) (31) for λ ∈ Π. We check next using (31) that ρ is positive; that is ρ1(λ) ≥ 0 and ρ2(λ) ≥ 0 for λ ∈ Π. First if λ ∈ T , we have ρ2(λ) = 1. But since ρ ∈ Hom(Σ, BC2), we have ρ1(µ)ρ2(µ) ≥ 0 for µ ∈ Σ. So ρ1(λ) ≥ 0 for λ ∈ T . Next, if λ ∈ Π \ T , we have ρ2(λ) = 0 and ρ1(λ) = −χS( wΠ\T (λ)) = χS(σΠ\T (λ)) ≥ 0. So ρ is positive as desired. Consequently we have s1 ∗ χ(S,T ) = ρ, so χs1∗(S,T ) = ρ. Thus s1 ∗ (S, T ) = ( S, T ), where S = {λ ∈ Π : ρ1(λ) = 1} and T = {λ ∈ Π : ρ2(λ) = 1}. (32) It follows now from (31) that T = T , so it remains establish (27). Now S = ( S \ T ) ∪ ( S ∩ T ). Moreover, using (31) and (32), S \ T = {λ ∈ Π \ T : χS(σΠ\T (λ)) = 1} = σΠ\T (S \ T ) and S ∩ T = {λ ∈ T : 2 − χS( wΠ\T (λ)) = 1} = {λ ∈ T : χS( wΠ\T (λ)) = 1}. But if λ ∈ T , then χS( wΠ\T (λ)) = χS\T ( wΠ\T (λ)) + χS∩T ( wΠ\T (λ)) = χS\T ( wΠ\T (λ)) + χS∩T (λ) using (24). So we have (27). Row 3: Let τ = (τ1, τ2) = s2 · χ(S,T ) · wΠ\S ∈ Homsh(Σ, BC2). Now s2 · χ(S,T )(λ) = s2(χS(λ)α1 + χT (λ)α2) = χS(λ)(α1 + α2) + χT (λ)(−α2) = (χS(λ), χS(λ) − χT (λ)). for λ ∈ Σ. Also if λ ∈ Π, χS( wΠ\S(λ)) = χS(λ) by (24). Thus τ1(λ) = χS(λ) and τ2(λ) = χS(λ) − χT ( wΠ\S(λ)) (33) for λ ∈ Π. Arguing as in Row 2 we now easily see that τ is positive, χs2∗(S,T ) = τ and s2 ∗ (S, T ) = ( ¯S, ¯T ), where ¯S = {λ ∈ Π : τ1(λ) = 1} and ¯T = {λ ∈ Π : τ2(λ) = 1}. DYNKIN DIAGRAMS AND KANTOR PAIRS 29 It follows then from (33) that ¯S = S, so it remains to prove (28). Again arguing as in Row 2, we easily see that ¯T = σΠ\S(T \ S) [ {λ ∈ S : χT \S( wΠ\S(λ)) + χT ∩S(λ) = 0}. Finally, let λ ∈ S. Then, since wΠ\S(λ) ∈ λ+ QΠ\S, we see that wΠ\S(λ) is positive and hence χT \S( wΠ\S(λ)) ≥ 0. Thus χT \S( wΠ\S(λ)) + χT ∩S(λ) = 0 if and only if λ ∈ S \ T and χT \S( wΠ\S(λ)) = 0, which holds if and only if λ ∈ S \ T and suppΠ( wΠ\S(λ)) ∩ (T \ S) = ∅. Row 4: Using Row 2 and Row 3 (applied to ( S, T )), we have (s2s1) ∗ (S, T ) = s2 ∗ (s1 ∗ (S, T )) = s2 ∗ ( S, T ) = ( S, T ′) for some subset T ′ of Π. On the other hand, s2s1 = −s1s2. So, using Row 3, Row 2 (applied to (S, ¯T )) and Row 5 (applied to (S′, ¯T )), we have (s2s1) ∗ (S, T ) = (−1) ∗ (s1 ∗ (s2 ∗ (S, T ))) = (−1) ∗ (s1 ∗ (S, ¯T )) = (−1) ∗ (S′, ¯T ) = (S′ · σΠ, ¯T · σΠ) for some subset S′ of Π. Combining these equalities gives our conclusion. (cid:3) If (S, T ) ∈ SPA(Π), then we see using Theorems 8.3.1 and 8.3.2 that P(S, T )op = −1P(S, T ) ≃gr P((−1) ∗ (S, T )) = P(S · σΠ, T · σΠ) ≃gr P(S, T ). This together with Theorem 7.2.4(ii) gives another proof of Proposition 6.2.2. The following corollary, which we state for emphasis and convenience of reference, follows taking u = s1 and u = s2 in Theorems 8.3.1 and 8.3.2. Corollary 8.3.3. If (S, T ) ∈ SPA(Π) and S and ¯T are given by (29) and (30) (or (27) and (28)), then P(S, T ) ≃gr P( S, T ) and P(S, T ) ≃gr P(S, ¯T ). Remark 8.3.4. Corollary 8.3.3 explains our choice of notation for S and ¯T . It should be noted however that this is a (convenient) abuse of notation, since S and ¯T each depend on both S and T . Example 8.3.5 (The trivial SP-gradings). Suppose that S ∈ KA(Π). Recall from Section 7.2 that the zero and one SP-gradings on P(S) are determined by ∅ and S respectively. For these SP-gradings one can check easily using Corollary 8.3.3, (29) and (30) that P(S, ∅) ≃gr P(S, ∅), P(S, S) ≃gr P(S, S), and P(S, ∅) ≃gr P(S, S), P(S, S) ≃gr P(S, ∅). The computation of Weyl images is particularly simple in the close-to-Jordan case. Corollary 8.3.6. Suppose that (S, T ) ∈ SPA(Π) and P = P(S, T ) is close-to- Jordan with non-trivial SP-grading. Then P(S, T ) ≃gr P(σΠ\T (S \ T ), T ) and P(S, T ) ≃gr P(S, T ). DYNKIN DIAGRAMS AND KANTOR PAIRS 30 Proof. Recall that Xn 6= A1, S is the set of nodes of Π that are adjacent to µ− in Π, and T = {λ} with λ ∈ Π and χλ(µ+) = 1 (see Theorems 5.4.1(ii) and 7.3.1). We assume that λ is not an end node of Π if Xn = An, leaving the excluded case for the reader to check. Now χS(µ+) = 2 by Theorem 5.4.1(i). Also S ∩ T = ∅. Indeed this holds by assumption if Xn = An, whereas it holds when Xn 6= An since in that case card(S) = 1 by Remark 5.4.2(i). Thus, we see from Corollary 8.3.3, (29) and (30), that P(S, T ) ≃gr P(σΠ\T (S \ T ), T ) and P(S, T ) ≃gr P(S, σΠ\S(T \ S)). Finally, by uniqueness in Corollary 7.3.2, we have P(S, T ) ≃gr P(S, T ) if Xn 6= An and Xn 6= Dn. But if Xn = An or Xn = Dn, σΠ\S extends to an automorphism of Π, so P(S, T ) ≃gr P(S, T ) by Theorem 7.3.1. (cid:3) Example 8.3.7. (Type E6). Recall that in Example 7.4.1 we saw that there are, up to graded-isomorphism, seven simple SP-graded Kantor pairs of type E6 whose gradings are non-trivial. It is straightforward to apply Corollary 8.3.3, together with (29) and (30), to calculate the reflection and shift of each of these SP-graded Kantor pairs up to graded-isomorphism. One sees that shifting exchanges E6(20, 5, 8) and E6(20, 5, 12); and that shifting fixes the other five graded Kantor pairs. One also sees that reflection exchanges E6(16, 0, 8) and E6(16, 8, 8a); reflection exchanges E6(20, 1, 10) and E6(20, 5, 10); and that reflection fixes E6(20, 5, 8), E6(20, 5, 12) and E6(16, 8, 8b). In particular, reflection does not preserve balanced 2-dimension. In fact, we see that the reflection of the Jordan pair E6(16, 0, 8) is not Jordan, and that the reflection of the close-to- Jordan pair E6(20, 1, 10) is not close-to-Jordan. 8.4. The proofs of (29) and (30). We now return to the proofs of (29) and (30) that we postponed earlier. The reader may elect to further postpone reading these arguments, as they will not be used in the final section. The information that we need about the Weyl group elements wΠ\T and wΠ\S is provided by the next two lemmas: Lemma 8.4.1. If (S, T ) ∈ SPA(Π), then χS\T ( wΠ\T (λ)) = χS\T (µ+) for λ ∈ T . Proof. Let λ ∈ T . We can assume ν0 := wΠ\T (λ) 6= µ+. Now χT ( wΠ\T (λ)) = χT (λ) by (24), so χT (ν0) = 1. Hence ν0 ∈ Σ+, so there are λ1, . . . , λr in Π with r ≥ 1 such that νi := νi−1 + λi ∈ Σ+ for 1 ≤ i ≤ r and νr = µ+. We next claim that λ1 ∈ T . Indeed, otherwise λ1 ∈ Π \ T = σΠ\T (Π \ T ), so λ1 = σΠ\T (ν) for some ν ∈ Π \ T . Hence wΠ\T (λ − ν) = wΠ\T (λ) − wΠ\T (ν) = ν0 + λ1 = ν1 ∈ Σ and therefore λ − ν ∈ Σ. This is a contradiction since λ ∈ T and ν ∈ Π \ T . So we have our claim. Next, for 1 ≤ p ≤ r, we have χT (νp) = χT (ν0 + λ1 + · · · + λp) ≥ χT (ν0) + χT (λ1) = 1 + 1 = 2. DYNKIN DIAGRAMS AND KANTOR PAIRS 31 Hence χT (νp) = 2 for 1 ≤ p ≤ r. Thus, since (S, T ) ∈ SPA(Π), we have χS(νp) 6= 1 for 1 ≤ p ≤ r. So, for 2 ≤ p ≤ r, we have χS(λp) = χS(νp) − χS(νp−1) ∈ 2Z and hence λp /∈ S. If follows from the previous two paragraphs that λp /∈ S \ T for 1 ≤ p ≤ r. So (cid:3) χS\T (ν0) = χS\T (νr) = χS\T (µ+). Lemma 8.4.2. If X ⊆ Π and λ ∈ Π \ X, then suppΠ( wX (λ)) = comp(X ∪ {λ}, λ). Proof. Let Y1, . . . , Yr be the connected components of X ∪ {λ} with λ ∈ Y1, and let Z1, . . . , Zs be the connected components of Y1 \ {λ}. Then Z1, . . . , Zs, Y2, . . . , Yr are the connected components of X. We must show that V = Y1, where V = suppΠ( wX (λ)). Now it is clear that wX = wZ1 . . . wZr wY2 . . . wYr , so wX (λ) = wZ1 . . . wZr (λ) ∈ λ + QZ1∪···∪Zs ⊆ QY1 . Thus λ ∈ V ⊆ Y1. Next, to show that Y1 ⊆ V , it suffices to show that any µ ∈ X ∪ {λ} that is adjacent to some ν ∈ V lies in V . For this we can assume that µ 6= λ, so µ ∈ X. Then µ = σX (µ′) for some µ′ ∈ X, so wX (λ) + µ = wX (λ − µ′) /∈ Σ. Hence h wX (λ), µi ≥ 0. Since hν, µi < 0, this forces hν′, µi > 0 for some ν′ ∈ V . So, since µ, ν′ ∈ Π, we have µ = ν′, and hence µ ∈ V . (cid:3) Proof of (29). By (27) and Lemma 8.4.1, S = σΠ\T (S \ T ) ∪ ( S ∩ T ) and S ∩ T = {λ ∈ T : χS\T (µ+) + χS∩T (λ) = 1} Furthermore, by Proposition 7.1.5(c), χS(µ+) ∈ {1, 2}, so χS\T (µ+) ∈ {0, 1, 2}. If χS\T (µ+) = 0 (or equivalently S ⊆ T ), then S ∩ T = {λ ∈ T : χS(λ) = 1} = S. Also, if χS\T (µ+) = 1, then S ∩ T = {λ ∈ T : χS∩T (λ) = 0} = T \ S. Finally, if χS\T (µ+) = 2, then S ∩ T = {λ ∈ T : χS∩T (λ) = −1} = ∅. (cid:3) Proof of (30). By (28) and Lemma 8.4.2 (with X = Π \ S), we have ¯T = σΠ\S(T \ S) [ {λ ∈ S \ T : comp(cid:0)(Π \ S) ∪ {λ}, λ(cid:1) ∩ (T \ S) = ∅}. If T ⊆ S, then T \ S = ∅, so ¯T = S \ T by (34). Next if (ex) holds, then it is (34) easy to check using (34) that ¯T = {σΠ\S(µ), λ}. We leave this to the reader. Finally suppose that T 6⊆ S and (ex) does not holds. We suppose for contradic- tion (which will complete the proof of (30)) that (35) comp(cid:0)(Π \ S) ∪ {λ}, λ(cid:1) ∩ (T \ S) = ∅, for some λ ∈ S \ T . If S = {λ}, then (Π \ S) ∪ {λ} = Π which is connected, so, by (35), we have T \ S = ∅, giving a contradiction. Hence we can assume that S contains an element λ′ 6= λ. Thus, by Proposition 7.1.5(c), we have S = {λ, λ′}, with (36) Then (Π \ S) ∪ {λ} = Π \ {λ′} and T \ S = T \ {λ′} since λ /∈ T . So, by (35), we have χλ(µ+) = χλ′ (µ+) = 1. (37) Now if Π \ {λ′} is connected, then (Π \ {λ′}) ∩ T = ∅ implies that T ⊆ {λ′} ⊆ S, a contradiction. So comp(Π \ {λ′}, λ) ∩ T = ∅. Π \ {λ′} has at least 2 connected components. (38) DYNKIN DIAGRAMS AND KANTOR PAIRS 32 Now a check of µ+ for each type shows that the existence of distinct elements λ, λ′ ∈ Π satisfying both (36) and (38) implies that Π has type An, where n ≥ 3. Then, by (37), comp(Π \ {λ′}, λ) ∪ {µ−} is a connected subset of Π \ T , so λ ∈ comp( Π \ T, µ−) ∩ S. Therefore, by Proposition 7.1.5(c), we have χT (µ+) = 0 or 1. But certainly χT (µ+) 6= 0 since T 6= ∅. Hence χT (µ+) = 1, so T = {µ} for some µ ∈ Π, with µ 6= λ′ since T 6⊆ S. It is now clear from (37) that we have (ex), giving a contradiction. (cid:3) 9. Reflections of simple close-to-Jordan pairs We continue with the assumptions and notation of Section 7. In this section we use results from Sections 5 -- 8 to give a construction of the reflection of each simple SP-graded close-to-Jordan pair of type Xn whose grading is non-trivial. We do this using the description of these graded pairs given in Theorem 7.3.1. 9.1. The trilinear pair T(J, τ ). In order to construct some trilinear pairs, we fix a pair U = (U −, U +) of 2-dimensional vector spaces and a nondegenerate bilinear map ( , ) : U − × U + → K; and we set (u+, u−) = (u−, u+) for uσ ∈ U σ. In order to construct gradings on our trilinear pairs, we fix bases {uσ 1 } for j ) = δij (so these bases are dual with respect to ( , )). U σ, σ = ±, such that (u− i , u+ 0 , uσ Construction 9.1.1. (The graded trilinear pair T(J, τ )) Let J be a trilinear pair with products { , , }σ J (x, a)y = J. Also, let τ = (τ −, τ +), where τ σ : J σ × J −σ → K is a bilinear map for {x, a, y}σ σ = ±. Then J , σ = ±, and define Dσ J (x, a) ∈ End(J σ) by Dσ T(J, τ ) = J ⊗ U := (J − ⊗ U −, J + ⊗ U +) is a trilinear pair (but not in general a Kantor pair) with products { , , }σ given by {x ⊗ r, a ⊗ ℓ, y ⊗ s}σ = {x, a, y}σ J ⊗ (r, ℓ)s − τ σ(x, a)y ⊗ (r, ℓ, s) (39) for x, y ∈ J σ, a ∈ J −σ, r, s ∈ U σ, ℓ ∈ U −σ, where (r, ℓ, s) = (r, ℓ)s − (s, ℓ)r for r, s ∈ U σ, ℓ ∈ U −σ. Moreover, one checks easily that T(J, τ ) =Li∈Z T(J, τ )i = T(J, τ )0 ⊕ T(J, τ )1 is a Z-graded trilinear pair with T(J, τ )i = J σ ⊗ uσ i for i = 0, 1 and T(J, τ )i = 0 otherwise. It is clear that the trilinear pair T(J, τ ) does not depend up to isomorphism on the choice of U −, U + or ( , ), and that the grading T(J, τ ) = T(J, τ )0 ⊕ T(J, τ )1 does not depend up to isomorphism on the choice of the dual bases {uσ 0 , uσ 1 }. Remark 9.1.2. The above construction of trilinear pairs is a pair version, with basis free products, of a construction of triple systems given by Kantor in [K1]. More precisely suppose that J = (X, X) is the double of a triple system X and τ − = τ + : X × X → K is a bilinear map. Then one can easily check that the pair T(J, τ ) is isomorphic to the double of the Kronecker product X ⊗ K2 defined in [K1, §6. Defn. 6] with k = 2, λ = µ = 1 and f (x, y) = τ σ(y, x), σ = ±. (It appears however that there is a typo in Kantor's definition: the defining expression α=1 ((yαxαzi) + λf (xα, yi)zα − µf (xα, yα)zi).) for (Y XZ)i should read asPk DYNKIN DIAGRAMS AND KANTOR PAIRS 33 9.2. Constructing the reflection of P(Π; S, T ). Suppose for the rest of the ar- ticle that (S, T ) ∈ SPA(Π) and that P(Π; S, T ) is close-to Jordan with nontrivial SP-grading. Our goal is to construct the reflection of P(Π; S, T ) in the form T(J, τ ) for a specified choice of J and τ . As noted in Corollary 7.3.2, our assumptions imply that Xn = An (n ≥ 2), Bn (n ≥ 2), Cn (n ≥ 3), Dn (n ≥ 4), E6 or E7. In particular, n ≥ 2. We write the distinct elements of Π as µ1, . . . , µn. By Theorems 5.4.1 and 7.3.1, we know that S is the set of nodes of Π that are adjacent to µ− in Π, χS(µ+) = 2 and T = {µt} for some 1 ≤ t ≤ n with χµt (µ+) = 1. We let Π′ = Π \ T and S′ = S \ T. In the remainder if this section we will for convenience exclude from consideration the case when Xn = An and µt is an interior (not an end) node of Π. We will return to consider this excluded case in the last subsection. Lemma 9.2.1. If Xn = An, suppose that the element µt of T is an end node of Π. Then S =({µs, µt} where µs is the other end node in Π, {µs} where µs ∈ Π with χµs(µ+) = 2, if Xn = An; otherwise. Also there exists a unique µt′ in Π′ that is adjacent to µt in Π, so Π′ is connected. Proof. The first statement follows from Remark 5.4.2(i), and the second is easily checked considering types case-by-case. (cid:3) With the assumptions and notation of Lemma 9.2.1, we define a constant θΠ,T ∈ Q by θΠ,T :=(cid:26) p + 1 if Xn = An (n ≥ 2) p + 2 otherwise, where p is the product of the (t, t′)-entry of the Cartan matrix C(Π) and the (t′, s)-entry of the inverse of C(Π′). (Note that C(Π) is an I × I matrix, where I = {1, . . . , n}, whereas C(Π′) is an I ′ × I ′ matrix with I ′ = I \ {t}. See Subsection 1.1.) Theorem 9.2.2. Suppose that (S, T ) ∈ SPA(Π), P = P(Π; S, T ) is close-to Jordan with nontrivial SP-grading, and, if Xn = An, the element µt of T is an end node of Π. Then (i) P0 is a simple Jordan pair that is isomorphic to P(Π′; S′). (ii) Suppose that J is any simple Jordan pair that is isomorphic to P(Π′; S′), and let τ = τJ,Π,T := (τ −, τ +), where τ σ : J σ × J −σ → K is given by τ σ(x, a) = θΠ,T dim(J σ) tr(Dσ J (x, a)) (40) for x ∈ J σ, x ∈ J −σ, σ = ±. Then τ σ(x, a) = τ −σ(a, x); τ σ is non- degenerate for σ = ±; T(J, τ ) is a simple SP-graded Kantor pair; and P ≃gr T(J, τ ). DYNKIN DIAGRAMS AND KANTOR PAIRS 34 Proof. We first set some notation. As usual, choose hi ∈ [Gµi , G−µi ] so that µi(hi) = 2 for 1 ≤ i ≤ n, in which case {hi}n i=1 is a basis for H. Also, choose nonzero eσ ∈ Gµσ for σ = ± such that [h+, eσ] = σ2eσ, where h+ = [e+, e−]. Then since χS(µ+) = 2 and χT (µ+) = 1, we see by (12) that G(χ(S,T ))σ2,∗ = G(χ(S,T ))σ2,σ1 = Gµσ = Keσ. (41) (i): Let Σ′ = {µ ∈ Σ : χT (µ) = 0}. Then, since Π′ is connected, Σ′ is an irreducible root system of rank n − 1 (in its real span) with base Π′. Let G′ be the subalgebra of G generated by {Gµi + G−µi : i 6= t}, and let H′ = H ∩ G′. Then G′ = H′ ⊕ (Lµ∈Σ′ Gµ(cid:1); H′ =Pi6=t Kh; G′ is a simple Lie algebra with Cartan subalgebra H′; and Σ(G′, H′) = Σ′ (identi- fying elements of Σ′ with their restrictions to H′). We now use G′, H′, Π′ and S′ ∈ KA(Π′) in Construction 5.2.1 to obtain a 5- graded Lie algebra G′(χS ′ ) = ⊕i∈ZG′(χS ′ )i and hence the Kantor pair P(Π′; S′) enveloped by G′(χS ′ ) (see Construction 5.2.1). Note that for i 6= 0 we have G′(χS ′ )i =Pµ∈Σ′, χS ′ (µ)=i G′ µ =Pµ∈Σ, χS (µ)=i, χT (µ)=0 Gµ = G(χ(S,T ))i,0. Hence P(Π′; S′)σ = G′(χS ′)σ1 = G(χ(S,T ))σ1,0 = P σ 0 , (42) so P0 = P(Π′; S′). Moreover, G′(χS ′ )σ2 = G(χ(S,T ))σ2,0 = 0 by (41), so G′(χS ′ ) is 3-graded and thus P0 is Jordan by Remark 5.2.3(ii). (ii): To prove (ii), we can assume that J = P0 and use the notation and conclu- sions in the proof of (i). Let ω := exp(ad e+) exp(− ad e−) exp(ad e+) ∈ Aut(G). We will use Lemma 5.5.1, which gives us detailed information about ω. First, by Lemma 5.5.1(v), ω(x) = −σ[e−σ, x] for x ∈ P σ. Thus, since e−σ ∈ G(χ(S,T ))−σ2,−σ1 by (41), we see that ω(G(χ(S,T ))σ1,σi) = G(χ(S,T ))−σ1,−σ(1−i) for i = 0, 1. That is ω exchanges P σ i and P −σ 1−i. Next let τ σ = ζ σJ σ ×J −σ : J σ ×J −σ → K, with ζ σ as defined in Lemma 5.5.1(vi), so (43) for x ∈ J σ, a ∈ J −σ. At this point we will take this as the definition of τ σ, and then at the end of the proof we will prove (40). [[x, a], eσ] = τ σ(x, a)eσ Note that τ σ(x, a) = τ −σ(a, x) by (16). Also, for σ = ±, i = 0, 1, we have 1−i], eσ] ⊆ [[G(χ(S,T ))σ1,σi, G(χ(S,T ))−σ1,−σ(1−i)], G(χ(S,T ))σ2,σ1], i , P −σ [[P σ which is contained in G(χ(S,T ))σ2,σ2i = 0. Thus, by (14), ζ σ(P σ non-degeneracy of ζ σ implies that of τ σ. i , P −σ 1−i) = 0. So the In the rest of the proof, it is more convenient to work with Q := P op, rather than P itself. We next show that the Z-graded trilinear pairs Q and T(J, τ ) are graded-isomorphic. This will show that T(J, τ ) is a simple SP-graded Kantor pair and, by Proposition 6.2.2, that P ≃gr T(J, τ ) (leaving only (40) to prove). Now Q = Q0 ⊕ Q1, where Qσ 0 and Qσ 1 = P −σ 1 = P −σ 1 = ω(P σ 0 ), so 0 = P −σ 0 = P σ i = ωi(J σ) Qσ for σ = ±, i = 0, 1. Moreover, by the definition of Weyl images (see Section 6.3), the products in Q are given by {x, a, y}σ = [[x, a], y] in G. On the other hand, DYNKIN DIAGRAMS AND KANTOR PAIRS 35 T(J, τ ) = T(J, τ )0 ⊕ T(J, τ )1 with T(J, τ )i = J σ ⊗ uσ i and products given by (39). With this in mind, we define ϕ = (ϕ−, ϕ+), where ϕσ : T(J, τ )σ → Qσ is the linear isomorphism such that ϕσ(x ⊗ uσ i ) = ωix for x ∈ J σ, σ = ± and i = 0, 1. In order to prove that ϕ is an isomorphism of trilinear pairs, we must show that ϕσ(cid:0){x ⊗ uσ i , a ⊗ u−σ j , y ⊗ uσ k }(cid:1) = [[ωix, ωja], ωky] i , u−σ for σ = ±, x, y ∈ J σ, a ∈ J −σ and i, j, k = 0, 1. But (uσ (uσ i . So we must prove that k ) = δij uσ k − δjkuσ i , u−σ , uσ j δijωk([[x, a], y]) − δijτ σ(x, a)ωky + δjkτ σ(x, a)ωiy = [[ωix, ωja], ωky]. )uσ j k = δij uσ k and (44) Now if (i, j, k) = (0, 0, 0), (44) is trivial; whereas if (i, j, k) = (1, 1, 1), (44) holds since ω is an automorphism. If (i, j, k) = (1, 0, 0) or (0, 0, 1), then (44) follows from Lemma 5.5.1(vii); whereas if (i, j, k) = (0, 1, 0), (44) holds since its right hand side lies in G(χ(S,T ))σ3, ∗, which is 0. Finally the cases (1, 1, 0), (1, 0, 1) and (0, 1, 1) follow by applying ω to the cases (0, 0, 1), (0, 1, 0) and (1, 0, 0) respectively. It remains to prove (40). For this let {ℓ′ i}i6=t be the basis for H′ that is dual to {µi}i6=t. Recall from the proof of (i) that G′(χS ′) = ⊕i∈ZG′(χS ′)i is 3-graded and that G′(χS ′ )σ1 = P σ 0 = J σ. To simplify notation, we set f := G′(χS ′ )0. Then f = H′ ⊕Pµ∈Σ′, χS ′ (µ)=0 G′ µ = H′ ⊕Pµ∈Σ, χS (µ)=0, χT (µ)=0 Gµ = Khs ⊕ k as vector spaces, where k is the subalgebra of G generated by {Gµi + G−µi : i 6= s, t}. Further, k is semi-simple and ℓ′ s is in the centre of f. So ℓ′ s /∈ k and hence (45) f = Kℓ′ s ⊕ k. (46) as algebras. We next construct some elements of the one-dimensional space F := {λ ∈ Hom(f, K) : λ(k) = 0}. First note that f ⊆ G(χS)0,0 by (45). So by (41), [f, eσ] ⊆ Keσ for σ = ±. Thus for σ = ±, there exists a unique νσ ∈ Hom(f, K) such that [z, eσ] = νσ(z)eσ for z ∈ f. Then since k = [k, k], we have νσ ∈ F . Also νσ(ℓ′ s) = µσ(ℓ′ s) = σµ+(ℓ′ s). Next, since G′(χS ′ ) is 3-graded, we have [f, J σ] ⊆ J σ for σ = ±. So we can define ξσ ∈ Hom(f, K) by ξσ(z) = tr(ad(z) J σ ) for z ∈ f. Once again we see that ξσ ∈ F . Moreover, since J σ = G′(χS ′ )σ1 by (42), we have ad(ℓ′ s) J σ = σ idJ σ , so ξσ(ℓ′ s) = σ dim(J σ) 6= 0. Now, since F is one dimensional, we have νσ = rσξσ for some rσ ∈ K, in which case (evaluating at ℓ′ dim(J σ ) for σ = ±. So for z ∈ f we have s) we have rσ = µ+(ℓ′ s) µ+(ℓ′ s) dim(J σ) νσ(z) = tr(ad(z) J σ ). But if x ∈ J σ, a ∈ J −σ, we have [x, a] ∈ f since G′(χS ′ ) is 3-graded. Also τ σ(x, a) = νσ([x, a]) by (43), so τ σ(x, a) = µ+(ℓ′ s) dim(J σ) tr(Dσ J (x, a)). DYNKIN DIAGRAMS AND KANTOR PAIRS 36 Finally, we compute µ+(ℓ′ s). Let C = C(Π) with (i, j)-entry cij := hµi, µji, and let {ℓi} be the basis for H that is dual to {µi}. Then for 1 ≤ j ≤ n. Similarly, since C(Π′) has (i, j)-entry cij for i, j 6= t, we have ij for i, j 6= t, i for j 6= t. So letting D′ = C(Π′)−1 with (i, j)- entry d′ hj =Pi cijℓi hj =Pi6=t cij ℓ′ we have ℓ′ ℓ′ j =Pi6=t d′ s =Pi6=t d′ =(cid:16)Pi6=t ctid′ ijhi for j 6= t. Therefore ishi =Pi6=t d′ is(cid:17) ℓt +Pk6=t(cid:16)Pi6=t ckid′ isPk ckiℓk =Pk(cid:16)Pi6=t ckid′ is(cid:17) ℓk = ctt′ d′ is(cid:17) ℓk t′sℓt + ℓs, since cti = 0 for i 6= t′. Thus µ+(ℓ′ s) = ctt′ d′ t′sµ+(ℓt) + µ+(ℓs) = θΠ,T . (cid:3) Theorem 9.2.2 shows that the reflection P of any Kantor pair P satisfying the given assumptions can be constructed in the form T(J, τ ), where J and τ are de- scribed in terms of marked Dynkin diagrams. In the next corollary, we describe J and τ using classical matrix constructions. Our notation in Columns 3 and 4 of Table 1 follows [L, §17.4]. Indeed, Ip,q is the Jordan pair (Mp,q(K), Mp,q(K)) with products {x, a, y}σ = xaty + yatx; IIn is the subpair (An(K), An(K)) of In,n, where An(K) is the space of alternating n × n- matrices; IVn is the Jordan pair (Kn, Kn) with products {x, a, y}σ = q(x, a)y + q(y, a)x − q(x, y)a, where q : Kn × Kn → K is the non-degenerate symmetric bilinear form on Kn; and V is the Jordan pair (M1,2(C), M1,2(C)) with products {x, a, y}σ = x(¯aty) + y(¯atx), where C is the (split) Cayley algebra with standard involution c 7→ ¯c and trace form tC : C → K given by tC(c) = c + ¯c [L, §12.10]. Corollary 9.2.3. Suppose we have the assumptions and notation of Theorem 9.2.2. The first column of Table 1 lists the possibilities, up to diagram automorphism, for the marked Dynkin diagrams representing P = P(Π; S, T ) (with the restriction on the rank n of Π also indicated). The second column lists the corresponding marked Dynkin diagram representing P ≃gr P(Π; σΠ′ (S′), T ) (see Corollary 8.3.6). Finally, in each row, we have P ≃gr T(J, τ ), where J and τ are listed in Columns 3 and 4 of the table respectively. Proof. Constructing Column 1 is an easy exercise considering types case-by-case using the discussion at the beginning of this subsection. To complete row i, where 1 ≤ i ≤ 7, we chose S and Π as listed in Column 1. To get the entry in Column 2, we just calculate σΠ′ (S′). To get the entries in Columns 3 and 4, we do the following: (a) Find a Jordan pair J of matrices that is isomorphic to P(Π′; S′); and then J (x, a)) for x ∈ J σ, a ∈ J −σ (which allows us to calculate (b) Calculate tr(Dσ τ σ using (40)). Fortunately, (a) and (b) can be accomplished using well known facts from the theory of Jordan pairs. Indeed for (a), Loos and Neher have written down a table which lists the finite dimensional simple Jordan pairs (as pairs of matrices) together with their representing marked Dynkin diagrams. (See [LN, §19.9], which refers to [N1] DYNKIN DIAGRAMS AND KANTOR PAIRS 37 P = P(Π; S, T ) n P ≃gr P(Π; σΠ′ (S′), T ) J τ σ(x, a) ∗ ∗ . . . . . . . . . ∗ ≥ 2 ≥ 3 ≥ 2 ∗ ∗ . . . . . . . . . I1,n−1 xat IV2n−3 q(x, a) ∗ I1,n−1 2xat ∗ . . . ≥ 4 ∗ . . . IV2n−4 q(x, a) . . . ∗ ≥ 5 . . . I2,n−2 tr(xat) ∗ ∗ ∗ 6 7 ∗ II5 1 2 tr(xa) ∗ V tC(x¯at) Table 1. Reflections of simple close-to-Jordan pairs (with a case excluded) and [N2] for the necessary arguments.) For (b), Meyberg observes in [M2, (2.20)] (using Theorem 17.3 in [L]) that tr(D+ J (x, a)) = gJ mJ (x, a) (47) for x ∈ J +, a ∈ J −, where gJ is the genus of J and mJ : J + × J − → K is the generic trace of J. We carry out steps (a) and (b) in detail for the second last row of the table, leaving the other rows to the reader. In that row the marked diagram representing P(Π′; S′) is isomorphic to . So from the table in [LN, §19.9], we see that we may take J = II5 = (A5(K), A5(K)). Then from [L, §17.2] we know that gJ = 8 and mJ (x, a) = 1 2 tr(xa), which using (47) gives us the equality tr(D+(x, a)) = 4 tr(xa). So since J = J op, we have tr(Dσ(x, a)) = 4 tr(xa) for σ = ±. Thus, since dim(J σ) = 10, we have by (40) that τ σ(x, a) = 2 5 θΠ,T tr(xa). Now, labeling the roots in Π as in Example 5.3.1, we have t = 1, t′ = 2 and s = 6. Also, the (1, 2)-entry of C(Π) is −1, and the (2, 6)-entry of C(Π′)−1 is 3 4 [H, Table 1,§13.2], so θΠ,T = − 3 (cid:3) 4 . Hence τ σ(x, a) = 1 4 + 2 = 5 2 tr(xa). Note that if we ignore the grading, the ungraded Kantor pair P in the second last row of Table 1 is the pair labelled E6(20, 5) in Example 5.3.1. Our construction of E6(20, 5) in the form T(P, τ ) is a basis-free pair version, with full proofs, of the construction given by Kantor in [K1, §6.6] (see also [K2, §4]) of the Kantor triple system C2 55. More precisely, T(J, τ ) shown in the table is the double of C2 55. The pair E6(20, 5) is of particular interest since it is one of only two finite dimensional simple Kantor pairs of exceptional type that does not arise by doubling a structurable algebra. (See [AFS, §7.9], where another construction of E6(20, 5) DYNKIN DIAGRAMS AND KANTOR PAIRS 38 is given as the reflection of an SP-graded Kantor pair that is constructed using exterior algebras.) 9.3. The excluded case. In this final subsection, we consider, without proofs, the case that was excluded in Theorem 9.2.2 and Corollary 9.2.3. We make the assumptions and use the notation of the first three paragraphs of Subsection 9.2. To treat the excluded case, we assume that Π = {µ1, . . . , µn} is of type An with roots labelled in order on the diagram from left to right, n ≥ 3, S = {µ1, µn} and T = {µt} with 1 < t < n. Let P = P(Π; S, T ), in which case the marked Dynkin diagrams representing P and P ≃gr P(Π; σT ′ (S′), T ) are respectively: . . . ∗ . . . and . . . ∗ . . . . Note that Π′ is not connected (which is the reason we are treating this case 1 = {µ1, . . . , µt−1} 1; {µ1}) and 2 = {µt+1, . . . , µn}. One sees as in Theorem 9.2.2(i) that P(Π′ In fact, the connected components of Π′ are Π′ separately). and Π′ P(Π′ 2; {µn}) are Jordan pairs with P0 = P(Π′ 1; {µ1}) ⊕ P(Π′ 2; {µn}). Next let J = J1 ⊕ J2, where J1 ≃ P(Π′ 1; {µ1}) and J2 ≃ P(Π′ 2; {µn}) are simple Jordan pairs. We obtain as in Theorem 9.2.2(ii) that P ≃gr T(J, τ ), where J2 (x2, a2)) J1 (x1, a1)) + 1 n−t+1 tr(Dσ t tr(Dσ (48) τ σ(x1 + x2, a1 + a2) = 1 i , ai ∈ J −σ . i for xi ∈ J σ Finally, we may choose J = J1 ⊕ J2, where J1 = I1,t−1 and J2 = I1,n−t ; (49) and we see as in Corollary 9.2.3 that 1 + x2at 2 (In this last equation, the superscript t denotes the transpose map.) P ≃gr T(J, τ ) with τ σ(x1 + x2, a1 + a2) = x1at Remark 9.3.1. The proof of the above facts is obtained by modifying the argu- ments in Subsection 9.2. The main difference is in the proof of (48), where we consider two different spaces of homomorphisms F1 and F2 obtained from simple 3-graded Lie algebras which envelop the Jordan pairs P(Π′ 2; {µn}) respectively (just as F is obtained from the simple 3-graded Lie algebra G′(χS ′) in the proof of (40)). We leave the details to the reader. 1; {µ1}) and P(Π′ Remark 9.3.2. The reader may have noticed that the Jordan pairs J that appear in either Column 3 of Table 1 or in (49) consist of all finite dimensional semi-simple Jordan pairs of rank 1 or 2 [L, §17]; and that τ is almost uniquely determined by J. The intrinsic reason for these mysterious facts will be explained in [AF2], where we will use Jordan techniques to study the construction T(J, τ ) over an arbitrary field of characteristic 6= 2 or 3. References [ABG] B. Allison, G. Benkart and Y. Gao, Lie algebras graded by the root systems BCr, r ≥ 2, Mem. Amer. Math. Soc. 158 (2002), no. 751. [AF1] B. Allison and J. Faulkner, Elementary groups and invertibility for Kantor pairs, Comm. Algebra 27 (1999), 519-556. DYNKIN DIAGRAMS AND KANTOR PAIRS 39 [AF2] B. Allison and J. Faulkner, Constructing Kantor pairs as tensor products: A Jordan the- oretic approach, in preparation. [AFS] B. Allison, J. Faulkner and O. Smirnov, Weyl images of Kantor pairs, to appear in the Canad. J. of Math., arXiv:1404.3339v2. [At] K. Atsuyama, On the algebraic structures of graded Lie algebras of second order, Kodai Math. J. 5 (1982), 225-229. [BS] G. Benkart and O. Smirnov, Lie algebras graded by the root system BC1, J. Lie Theory 13 [B] (2003), 91 -- 132. N. Bourbaki, Lie groups and Lie algebras, Chapters 4 -- 6 and 7 -- 9, Translated from the French, Elements of Mathematics, Springer-Verlag, Berlin, 2002 and 2005. [E1] A. Elduque, New simple Lie superalgebras in characteristic 3, J. Algebra 296 (2006), 196- 233. [E2] A. Elduque, The magic square and symmetric compositions II, Rev. Mat. Iberoam. 23 (2007), 57-84. [EKO] A. Elduque, N. Kamiya, A. Okubo, (−1, −1)-balanced Freudenthal Kantor triple systems and noncommutative Jordan algebras, J. Algebra 294 (2005), 19-40. [EK] A. Elduque and M. Kochetov, Gradings on simple Lie algebras, Mathematical Surveys and Monographs, 189, American Mathematical Society, Providence, RI; Atlantic Association for Research in the Mathematical Sciences (AARMS), Halifax, NS, 2013. J. Faulkner, A construction of Lie algebras from a class of ternary algebras. Trans. Amer. Math. Soc. 155 (1971), 397-408. J. Faulkner and J. Ferrar, On the structure of symplectic ternary algebras, Nederl. Akad. Wetensch. Proc. Ser. A 75=Indag. Math. 34 (1972), 247-256. [FF] [F] [GLN] E. Garc´ıa, M. G´omez Lozano and E. Neher, Nondegeneracy for Lie triple systems and [H] Kantor pairs, Canad. Math. Bull. 54 (2011), 442-455. J. E. Humphreys, Introduction to Lie algebras and representation theory, Springer-Verlag, New York, 1972. [Kac] V. Kac, Infinite dimensional Lie algebras, third edition, Cambridge University Press, Cam- [K1] [K2] [KS] bridge, 1990. I.L. Kantor, Some generalizations of Jordan algebras (Russian), Trudy Sem. Vektor. Ten- zor. Anal. 13 (1972), 407 -- 499. I.L. Kantor, Models of exceptional Lie algebras, Soviet Math. Dokl. 14 (1973), 254 -- 258. I.L. Kantor and I.M. Skopec, Freudenthal trilinear operations (Russian), Trudy. Sem. Vek- tor. Tenzor. Anal. 18 (1978), 250 -- 263. O. Loos, Jordan pairs, Lecture Notes in Mathematics 460, Springer-Verlag, Berlin, 1975. [L] [LN] O. Loos and E. Neher, Locally finite root systems, Mem. Amer. Math. Soc. 171 (2004). [Mc] K. McCrimmon, A taste of Jordan algebras, Universitext, Springer-Verlag, New York, 2004. [M1] K. Meyberg, Eine Theorie der Freudenthalschen Tripelsysteme I, II, Nederl. Akad. Weten- sch. Proc. Ser. A 71=Indag. Math. 30 (1968) 162-174, 175-190. [M2] K. Meyberg, Trace formulas and derivations in simple Jordan pairs, Comm. Algebra 12 (1984), 1311-1326. [N1] E. Neher, Lie algebras graded by 3-graded root systems and Jordan pairs covered by grids, Amer. J. Math. 118 (1996), 439-491. [N2] E. Neher, Quadratic Jordan superpairs covered by grids, J. Algebra 269 (2003), 28-73 [YA] K. Yamaguti and H. Asano, On the Freudenthal's construction of exceptional Lie algebras, Proc. Japan Acad. 51 (1975), 253-258. [YO] K. Yamaguti and A. Ono, On representations of Freudenthal-Kantor triple systems U (ǫ, δ), Bull. Fac. School Ed. Hiroshima Univ. Part II 7 (1984), 43-51. (Bruce Allison) Department of Mathematical and Statistical Sciences, University of Alberta E-mail address: [email protected] (John Faulkner) Department of Mathematics, University of Virginia, Kerchof Hall, P.O. Box 400137, Charlottesville VA 22904-4137 USA E-mail address: [email protected]
1203.5939	1	1203	2012-03-27T11:31:42	Full Descripion of ring varieties whose finite rings are uniquely determined by their zero-divisor graphs	[ "math.RA" ]	The zero-divisor graph $\Gamma(R)$ of an associative ring $R$ is the graph whose vertices are all nonzero zero-divisors (one-sided and two-sided) of $R$, and two distinct vertices $x$ and $y$ are joined by an edge iff either $xy=0$ or $yx=0$. In the present paper, we give a full description of ring varieties where every finite ring is uniquely determined by its zero-divisor graph.	math.RA	math	Е.В. Журавлев, А.С. Кузьмина, Ю.Н. Мальцев 1 ОПИСАНИЕ МНОГООБРАЗИЙ КОЛЕЦ, В КОТОРЫХ КОНЕЧНЫЕ КОЛЬЦА ОДНОЗНАЧНО ОПРЕДЕЛЯЮТСЯ СВОИМИ ГРАФАМИ ДЕЛИТЕЛЕЙ НУЛЯ В данной работе рассматриваются ассоциативные кольца (не обязательно коммута- тивные и не обязательно имеющие единицу). Определение. Графом делителей нуля кольца R называется граф, вершинами ко- торого являются все ненулевые делители нуля кольца (односторонние и двусторонние), причем две различные вершины x, y соединяются ребром тогда и только тогда, когда xy = 0 или yx = 0. Обычно граф делителей нуля кольца R обозначается через Γ(R). Мы также будем использовать это обозначение. Понятие графа делителей нуля было введено в работе [6]. И. Бек ввел это понятие для коммутативного кольца и вершинами графа делителей нуля считал все элементы кольца. В статье [5] определение было изменено: в качестве вершин графа делителей нуля коммутативного кольца авторы этой работы рассматривали лишь ненулевые де- лители нуля. Затем понятие графа делителей нуля было распространено и на неком- мутативный случай (см., например, [4]). Нетрудно привести примеры неизоморфных колец, графы делителей нуля которых равны. Например, если A (cid:22) счетномерная алгебра с нулевым умножением над полем Zp, а B (cid:22) счетномерная алгебра с нулевым умножением над полем Zq, где p, q (cid:22) это различ- ные простые числа, то Γ(A) ∼= Γ(B), но A 6∼= B. Другими словами, даже в многообразии var hxy = 0i существуют примеры бесконечных неизоморфных колец, графы делителей нуля которых имеют одинаковое строение. В связи с этим интерес представляет такой вопрос: при каких условиях из равенства графов делителей нуля следует изоморфизм колец? Некоторые результаты, дающие ответ на этот вопрос для коммутативных ко- лец, были получены в работе [3]. В настоящей работе данная проблема исследуется на языке многообразий, а именно: исследуются многообразия ассоциативных колец, в которых каждое конечное кольцо однозначно определяется своим графом делителей нуля. Другими словами, изучаются свойства многообразия колец M, для которого из равенства Γ(R) = Γ(S) для конечных колец R, S ∈ M, следует, что R ∼= S. Ранее такие многообразия исследовались в работах [1, 7]. Однако полного описания получено не бы- ло. В настоящей же работе многообразия, в которых все конечные кольца однозначно определяются своими графами делителей нуля, полностью описаны. Введем обозначения и понятия, используемые в настоящей работе. 2 Полным n -- вершинным графом Kn называется граф (без петель и кратных ребер), все n вершин которого смежны между собой. Пусть аддитивная группа кольца R разлагается в прямую сумму своих ненулевых аддитивных подгрупп Ai, где i = 1, . . . , n и n ≥ 2, т.е. R = A1 подгруппы Ai являются двусторонними идеалами кольца R, то кольцо R называется разложимым ( в обозначении R = A1 ⊕ . . . ⊕ An). . + . . . . + An. Если все Порядок конечного кольца R мы будем обозначать через R . Для любого элемен- та a ∈ R, где R -- произвольное кольцо, будем использовать следующее обозначе- ние: ann(a) = {x ∈ R; xa = ax = 0}. Для любых элементов x, y кольца R положим [x, y] = xy − yx и x ◦ y = xy + yx. Через Zn мы будем обозначать кольцо классов вычетов по модулю n. Для простого числа p будем полагать, что N0,p = hai , pa = 0, a2 = 0. Пусть Z hXi = Z hx1, x2, . . .i -- свободное ассоциативное кольцо от счетного числа переменных X = {x1, x2, . . .} и f (x1, . . . , xd) ∈ Z hXi. Многочлен f (x1, . . . , xd) суще- ственно зависит от x1, x2, . . . , xd, если f (0, x2, . . . , xd) = . . . = f (x1, . . . , xd−1, 0) = 0. Минимальная из степеней одночленов, входящих в запись f (x1, . . . , xd) с ненулевым коэффициентом, называется нижней степенью многочлена f (x1, . . . , xd). Пусть M -- многообразие колец. Через T (M) будем обозначать множество всех мно- гочленов из Z hXi, являющихся тождествами на всех кольцах из M. Назовем множество T (M) идеалом тождеств многообразия M. Если идеал тождеств T (M) порождается (как вполне характеристический идеал) многочленами fi, i ∈ I, то будем использовать i ∈ I}T . Через M ∨ N обозначается объединение следующее обозначение: T (M) = {fi многообразий M и N. Нетрудно заметить, что T (M ∨ N) = T (M) ∩ T (N) . Пусть M1,p = var hxyz = 0, x2 = 0, px = 0i и M2,p = var hxyz = 0, [x, y] = 0, px = 0i, где p (cid:22) произвольное простое число. Заметим, что M1,2 ⊆ M2,2, а при нечетном p мы имеем M1,p∩M2,p = var hxy = 0, px = 0i. Далее, пусть Fi (cid:22) приведенно свободное кольцо с шестью порождающими {x1, . . . , x6} многообразия Mi,p, i = 1, 2. Рассмотрим кольца Ai,p = Fi/ hx3x4 − x1x2i , Bi,p = Fi/ hx5x6 − x1x2 − x3x4i , где ha, b, . . .i (cid:22) это идеал кольца Fi, порожденный элементами {a, b, . . .}, i = 1, 2. Наша ближайшая цель (cid:22) доказать, что Γ(A1,p) ∼= Γ(B1,p) для любого простого числа p и Γ(A2,p) ∼= Γ(B2,p) при нечетном простом p. По теореме Тарского любое ненулевое многообразие колец содержит одно из ми- нимальных многообразий: var Zp или var N0,p, где p -- некоторое простое число [8]. Оказывается, что в минимальных многообразиях var Zp и var N0,p, где p -- любое про- стое число, все конечные кольца однозначно определяются своими графами делите- лей нуля так же, как и в многообразии var N0,p1 ∨ . . . ∨ var N0,ps [1]. В многообразии var N0,p1 ∨ . . . ∨ var N0,ps ∨ var Zp, где p1, . . . , ps -- попарно различные простые числа, p -- 3 любое простое число (возможно, совпадающее с одним из чисел pi), каждое конечное кольцо однозначно определяется своим графом делителей нуля тогда и только тогда, когда (pi, p) 6= (3, 2) при i ≤ s [1]. Наша цель (cid:22) показать, что любое многообразие, в ко- тором все конечные кольца однозначно определяются своими графами делителей нуля, является подмногообразием многообразия вида var N0,p1 ∨ . . . ∨ var N0,ps ∨ var Zp, где p1, . . . , ps -- попарно различные простые числа, p -- любое простое число и (pi, p) 6= (3, 2) при всех i ≤ s. Для этого нам понадобятся некоторые вспомогательные утверждения. Покажем сначала, что Γ(A1,p) ∼= Γ(B1,p), в то время, как A1,p 6∼= B1,p для любого простого числа p. Замечание. Порядок алгебр Ai,p, Bi,p, i = 1, 2, равен p14. Возникает вопрос о существовании в указанных многообразиях примеров неизоморфных конечных колец небольшого порядка с одинаковыми графами делителей нуля. Однако нами было до- казано, что в многообразии M1,2 все конечные кольца порядка ≤ 64 однозначно опре- деляются своими графами делителей нуля (фактически были полностью описаны все конечные кольца в многообразии M1,2, порядок которых не превышает 64). Лемма 1. Множество C1 = {xixj; (i, j) 6= (3, 4), 1 ≤ i < j ≤ 6} является базисом алгебры A2 1,p, где p (cid:22) простое число.) Доказательство. Множество C1 является системой образующих векторного простран- ства A2 1,p. Докажем, что это множество линейно независимо. Если множество C1 ли- нейно зависимо, то существуют элементы α, δij ∈ Zp, не все равные нулю, такие, что в алгебре F1 справедливо равенство δijxixj = α(x3x4 − x1x2). (1) Xi < j (i, j) 6= (3, 4) Положим в этом равенстве x3 = x4 = x5 = x6 = 0. Тогда δ12 = −α, другими словами, δijxixj = αx3x4. Положим x1 = x2 = x5 = x6 = 0. Получим, что α = 0 и δij = 0 Pi < j (i, j) 6= (3, 4) (i, j) 6= (1, 2) для всех i, j, таких, что i < j и (i, j) 6= (3, 4), (1, 2). Противоречие доказывает лемму. Предложение 1. Пусть a = αixi + u, b = βixi + v (cid:22) произвольные элементы из кольца A1,p, причем u, v ∈ A2 (β1, . . . , β6) = λ(α1, . . . , α6) для некоторого ненулевого элемента λ ∈ Zp. 1,p. Если a, b /∈ A2 1,p, то ab = 0 тогда и только тогда, когда 6 Pi=1 6 Pi=1 Доказательство. Пусть (β1, . . . , β6) = λ(α1, . . . , α6), где 0 6= λ ∈ Zp. Тогда a = c + u, b = λc + v и ab = λc2 = 0, т.к. x2 = 0 (cid:22) тождество в алгебре A1,p. Докажем обратное утверждение. Пусть ab = 0. Тогда 4 ab = (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α2 β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α3 α4 β3 β4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ! x1x2 + Xi < j (i, j) 6= (1, 2) (i, j) 6= (3, 4) xixj = 0. (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) αi αj βi βj (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) По лемме 1 множество C1 = {xixj; (i, j) 6= (3, 4), i < j} (cid:22) базис векторного пространства A2 1,p. Поэтому получаем, что (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α2 β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α3 α4 β3 β4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0 и (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) αi αj βi βj (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0, где i < j, (i, j) 6= (1, 2) и (i, j) 6= (3, 4). Рассмотрим следующие случаи. Случай 1. Пусть α1 6= 0. Поскольку (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α3 β1 β3 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0,(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α4 β1 β4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0, то вторые строки этих определителей линейно выражаются через первые, т.е. β1 = λα1, β3 = λα3 и β1 = µα1, β4 = µα4 для некоторых элементов λ, µ ∈ Zp. Поскольку α1 6= 0 и (λ − µ)α1 = β1 − β1 = 0, то λ = µ и β4 = λα4. Аналогично доказывается, что β5 = λα5 и β6 = λα6. = 0, β3 = λα3 и β4 = λα4 следует, что α1 α2 Далее, из равенств (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α2 β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α3 α4 β3 β4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0 и, значит, β2 = λα2. Таким образом, (β1, . . . , β6) = λ(α1, . . . , α6). Случай 2. Пусть α1 = 0. Если хотя бы один из элементов α2, α3, α4 не равен нулю, то рассуждаем так же, как при рассмотрении случая 1. Поэтому можем полагать, что α1 = α2 = α3 = α4 = 0 и, например, α5 6= 0. (Заметим, что поскольку a /∈ A2 1,p, то один из элементов α5 или α6 отличен от нуля.) Аналогично мы можем считать, что β1 = β2 = β3 = β4 = 0. Следовательно, ab =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) x5x6 = 0, или (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0, т.е. β5 = λα5, β6 = λα6 для некоторого ненулевого элемента λ ∈ Zp. Таким образом, (β1, . . . , β6) = λ(α1, . . . , α6). Из предложения 1 получаем, что для каждого элемента a = αixi ∈ A1,p \ A2 1,p мно- жество {αa + u; 0 6= α ∈ Zp, u ∈ A2 1,p} образует полный подграф Γa графа Γ(A1,p), причем каждая вершина подграфа Γa смежна только с вершинами этого подграфа и элементами из множества A2 1,p. Покажем, что такое же строение имеет граф Γ(B1,p). 6 Pi=1 5 Лемма 2. Множество D1 = {xixj; (i, j) 6= (5, 6), 1 ≤ i < j ≤ 6} является базисом алгебры B2 1,p для любого простого числа p. Доказательство аналогично доказательству леммы 1. Предложение 2. Пусть a = αixi + u, b = βixi + v (cid:22) произвольные элементы из кольца B1,p, где u, v ∈ B2 (β1, . . . , β6) = λ(α1, . . . , α6) для некоторого ненулевого элемента λ ∈ Zp. 1,p. Если a, b /∈ B2 1,p, то ab = 0 тогда и только тогда, когда 6 Pi=1 6 Pi=1 Доказательство. Пусть (β1, . . . , β6) = λ(α1, . . . , α6), где 0 6= λ ∈ Zp. Тогда a = c + u, b = λc + v и ab = λc2 = 0, т.к. x2 = 0 для любого элемента x ∈ B1,p. Докажем обратное утверждение. Пусть ab = 0. Тогда ab = (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α2 β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α3 α4 β3 β4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ! x1x2+ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) + Xi < j (i, j) 6= (3, 4) (i, j) 6= (1, 2) (i, j) 6= (5, 6) αi αj α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) βi βj (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) xixj = 0. ! x3x4+ По лемме 2 множество D1 = {xixj; (i, j) 6= (5, 6), i < j} (cid:22) базис векторного пространства B2 1,p. Поэтому (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α2 β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0,(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α3 α4 β3 β4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0 и (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) αi αj βi βj (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0, если i < j, (i, j) 6= (1, 2), (i, j) 6= (3, 4) и (i, j) 6= (5, 6). Так же, как при доказательстве предложения 1, рассмотрим два случая. Случай 1. Пусть α1 6= 0. Аналогично тому, как это было сделано в доказательстве предложения 1, получа- ем, что βi = λαi для любого i 6= 2. Поэтому из равенства (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) вытекает равенство (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 6= 0 мы имеем, что (β1, . . . , β6) = λ(α1, . . . , α6). β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α1 α2 α1 α2 β1 β2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0 = 0. Отсюда следует, что β2 = λα2. Таким образом, при Случай 2. Пусть α1 = 0. Если хотя бы один из элементов α2, α3, α4, β1, β2, β3, β4 не равен нулю, то рассуждаем так же, как в случае 1. Поэтому можем считать, что αi = βj = 0 при i, j ∈ {1, 2, 3, 4}. Значит, ab = (α5x5 + α6x6)(β5x5 + β6x6) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α5 α6 β5 β6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Поскольку a /∈ B2 1,p, то, например, α5 6= 0. Таким образом, β5 = λα5, β6 = λα6 для некоторого ненулевого элемента λ ∈ Zp, т.е. (β1, . . . , β6) = λ(α1, . . . , α6). (x1x2 + x3x4) = 0. Таким образом, справедливо Следствие 1. Γ(A1,p) ∼= Γ(B1,p) для любого простого числа p. 6 Предложение 3. Алгебра A1,p не изоморфна алгебре B1,p для любого простого числа p. Доказательство. Предположим противное: существует изоморфизм ϕ алгебры B1,p на алгебру A1,p. Введем такие обозначения: xi = xi + hx1x2 + x3x4 − x5x6i ∈ B1,p и i = xi + hx1x2 − x3x4i ∈ A1,p, i = 1, 6. Заметим что ϕ(B2 1,p, а значит, отобра- x′ жение ϕ : B1,p/B2 1,p, a ∈ B1,p, также является изоморфизмом. Отсюда следует, что существует невырожден- ная матрица P = (pij)6×6, pij ∈ Zp, такая, что 1,p, определенное по правилу ϕ(a + B2 1,p) = = ϕ(a) + A2 1,p → A1,p/A2 1,p) ⊆ A2 ϕ(x1) = p11x′ ϕ(x2) = p12x′ ϕ(x3) = p13x′ ϕ(x4) = p14x′ ϕ(x5) = p15x′ ϕ(x6) = p16x′ 1 + p21x′ 1 + p22x′ 1 + p23x′ 1 + p24x′ 1 + p25x′ 1 + p26x′ 2 + p31x′ 2 + p32x′ 2 + p33x′ 2 + p34x′ 2 + p35x′ 2 + p36x′ 3 + p41x′ 3 + p42x′ 3 + p43x′ 3 + p44x′ 3 + p45x′ 3 + p46x′ 4 + p51x′ 4 + p52x′ 4 + p53x′ 4 + p54x′ 4 + p55x′ 4 + p56x′ 5 + p61x′ 5 + p62x′ 5 + p63x′ 5 + p64x′ 5 + p65x′ 5 + p66x′ 6 + b1; 6 + b2; 6 + b3; 6 + b4; 6 + b5; 6 + b6,   где b1, b2, b3, b4, b5, b6 -- некоторые элементы из A2 1,p. Учитывая свойства изоморфизма, получаем ϕ(x1x2 + x3x4 − x5x6) = = (p11x′ 1 + p21x′ 2 + p31x′ 3 + p41x′ 4 + p51x′ 5 + p61x′ 6)(p12x′ 1 + p22x′ 2 + p32x′ 3 + p42x′ 4 + p52x′ 5 + p62x′ 6)+ +(p13x′ 1 + p23x′ 2 + p33x′ 3 + p43x′ 4 + p53x′ 5 + p63x′ 6)(p14x′ 1 + p24x′ 2 + p34x′ 3 + p44x′ 4 + p54x′ 5 + p64x′ 6)− −(p15x′ 1 + p25x′ 2 + p35x′ 3 + p45x′ 4 + p55x′ 5 + p65x′ 6)(p16x′ 1 + p26x′ 2 + p36x′ 3 + p46x′ 4 + p56x′ 5 + p66x′ 6) = = (p11p62 − p12p61 + p13p64 − p14p63 − p15p66 + p16p65)x′ 1x′ 6+ +(p21p62 − p22p61 + p23p64 − p24p63 − p25p66 + p26p65)x′ 2x′ 6+ +(p31p62 − p32p61 + p33p64 − p34p63 − p35p66 + p36p65)x′ 3x′ 6+ +(p41p62 − p42p61 + p43p64 − p44p63 − p45p66 + p46p65)x′ 4x′ 6+ +(p51p62 − p52p61 + p53p64 − p54p63 − p55p66 + p56p65)x′ 5x′ 6+ +f (x′ 1, x′ 2, x′ 3, x′ 4, x′ 5) = 0, где f (x′ 1, x′ 2, x′ 3, x′ 4, x′ держащий переменную x′ 5) -- некоторый многочлен от переменных x′ 5x′ 6. Так как x′ 6, x′ 6, x′ 6, x′ 6, x′ 3, x′ 2, x′ 5, не со- 1, x′ 6, линейно независимы, 4, x′ 1x′ 3x′ 4x′ 2x′ то 7 p11p62 − p12p61 + p13p64 − p14p63 − p15p66 + p16p65 = 0; p21p62 − p22p61 + p23p64 − p24p63 − p25p66 + p26p65 = 0; p31p62 − p32p61 + p33p64 − p34p63 − p35p66 + p36p65 = 0; p41p62 − p42p61 + p43p64 − p44p63 − p45p66 + p46p65 = 0; p51p62 − p52p61 + p53p64 − p54p63 − p55p66 + p56p65 = 0. Добавив очевидное равенство, получим p11p62 − p12p61 + p13p64 − p14p63 − p15p66 + p16p65 = 0; p21p62 − p22p61 + p23p64 − p24p63 − p25p66 + p26p65 = 0; p31p62 − p32p61 + p33p64 − p34p63 − p35p66 + p36p65 = 0; p41p62 − p42p61 + p43p64 − p44p63 − p45p66 + p46p65 = 0; p51p62 − p52p61 + p53p64 − p54p63 − p55p66 + p56p65 = 0; p61p62 − p62p61 + p63p64 − p64p63 − p65p66 + p66p65 = 0. Отсюда (p62, −p61, p64, −p63, −p66, p65) · P T = 0. Матрица P невырожденная и, в частности, не имеет нулевых строк, а значит, получен- ное равенство невозможно. Противоречие доказывает предложение. Из следствия 1 и предложения 3 вытекает справедливость следующего утверждения. Следствие 2. Если в многообразии колец M все конечные кольца однозначно опреде- ляются своими графами делителей нуля, то для любого простого числа p многообра- зие M не содержит многообразия M1,p. Далее, покажем, что алгебры A2,p и B2,p не изоморфны для любого простого нечетно- го p, однако Γ(A2,p) ∼= Γ(B2,p). Для этого нам понадобятся некоторые вспомогательные леммы. Лемма 3. Множество C2 = {x 2 является базисом алгебры A2 2,p (p > 2.) k , xixj; (i, j) 6= (3, 4), 1 ≤ i < j ≤ 6, 1 ≤ k ≤ 6} Доказательство. Множество C2 является системой образующих векторного простран- ства A2 2,p. Если оно линейно зависимо, то существуют элементы α, γ1, . . . , γ6, δij ∈ Zp, не все равные нулю, такие, что в алгебре F2 справедливо равенство 6 Xi=1 γix2 i + Xi < j (i, j) 6= (3, 4) δijxixj = α(x3x4 − x1x2). (2) 8 Полагая в равенстве (1) x1 = . . . = xi−1 = xi+1 = . . . = x6, где 1 ≤ i ≤ 6, получим, что γix2 i = 0, т.е. γ1 = . . . = γ6 = 0. Положим теперь в равенстве (1) x3 = x4 = x5 = x6 = 0. Тогда δ12 = −α, другими словами, δijxixj = αx3x4. Положим, наконец, Pi < j (i, j) 6= (3, 4) (i, j) 6= (1, 2) x1 = x2 = x5 = x6 = 0. Получим, что α = 0 и δij = 0 для всех i, j, таких, что i < j и (i, j) 6= (3, 4), (1, 2). Лемма 4. Если a = αixi + u ∈ A2,p, где p > 2, αi ∈ Zp для всех чисел i ∈ {1, . . . , 6}, u ∈ A2 2,p, a /∈ A2 2,p, то ann(a) = A2 2,p. 6 Pi=1 Доказательство. Пусть b = βixi + v ∈ ann(a), где βi ∈ Zp, i ∈ {1, . . . , 6}, v ∈ A2 2,p. 6 Pi=1 βixi! = αixi! 6 Xi=1 Тогда 0 = ab = 6 Xi=1 Xi=1 = 6 αiβix 2 i + (α1β2 + α2β1 + α3β4 + α4β3) x1x2 + Xi < j (i, j) 6= (1, 2) (αiβj + αjβi) xixj. Из леммы 3 следует, что (i, j) 6= (3, 4) α1β1 = . . . = α6β6 = 0, α1β2 + α2β1 + α3β4 + α4β3 = 0 и αiβj + αjβi = 0 для всех i, j ∈ {1, . . . , 6}, таких, что i < j, (i, j) 6= (1, 2) и (i, j) 6= (3, 4). Пусть α1 6= 0. Тогда из равенств α1β1 = 0 и α1βi+αiβ1 = 0, где i = 3, 4, 5, 6, получаем, что β1 = 0 и β3 = β4 = β5 = β6 = 0. Далее, из равенства α1β2 + α2β1 + α3β4 + α4β3 = 0 следует, что α1β2 = 0, т.е. β2 = 0. Другими словами, мы показали, что в этом случае b = v ∈ A2 2,p. Расссмотрим теперь случай, когда α1 = 0. Ввиду доказанного выше, можем считать, что α2 = α3 = α4 = 0. Значит, имеем равенства α5β5 = α6β6 = 0, α5β6 + α6β5 = 0. Поскольку a /∈ A2 2,p, то α6 6= 0 или α5 6= 0. Пусть α5 6= 0. Тогда из равенства α5β5 = 0 получаем, что β5 = 0. Поэтому α5β6 = 0. Отсюда β6 = 0. Другими словами, b ∈ A2 2,p. Случай, когда α6 6= 0 рассматривается аналогично. Таким образом, граф Γ(A2,p) имеет следующее строение: множество ненулевых эле- 2,p образуют полный подграф, а любая вершина a ∈ A2,p \ A2 2,p ментов из множества A2 смежна со всеми вершинами из этого подграфа и только с ними. Для алгебры B2,p (p > 2) справедливы аналоги лемм 3 -- 4. Лемма 5. Множество D2 = {x 2 является базисом алгебры B2 2,p (p > 2.) k , xixj; (i, j) 6= (5, 6), 1 ≤ i < j ≤ 6, 1 ≤ k ≤ 6} Доказательство аналогично доказательству леммы 3. 9 Лемма 6. Пусть a = αixi + u ∈ B2,p, где p > 2, αi ∈ Zp для всех чисел i ∈ {1, . . . , 6}, u ∈ B2 2,p и a /∈ B2 2,p. Тогда ann(a) = B2 2,p. 6 Pi=1 Доказательство. Возьмем элемент b = i ∈ {1, . . . , 6} и v ∈ A2 2,p. Получаем 6 Pi=1 βixi + v ∈ ann(a), где βi ∈ Zp для всех чисел 0 = ab = 6 Xi=1 αiβix 2 i + (α1β2 + α2β1 + α5β6 + α6β5) x1x2+ + (α3β4 + α4β3 + α5β6 + α6β5) x3x4 + Xi < j (i, j) 6= (1, 2) (αiβj + αjβi) xixj. (i, j) 6= (3, 4) (i, j) 6= (5, 6) Из леммы 5 следует, что α1β1 = . . . = α6β6 = 0,α1β2 + α2β1 + α5β6 + α6β5 = 0, α3β4 + α4β3 + α5β6 + α6β5 = 0 и αiβj + αjβi = 0 для всех i, j ∈ {1, . . . , 6}, таких, что i < j и (i, j) 6= (1, 2), (3, 4), (5, 6). Предположим, что α1 6= 0. Из равенства α1β1 = 0 получаем, что β1 = 0. Далее, из равенств α1βi + αiβ1 = 0, i = 3, 4, 5, 6, следует, что β3 = β4 = β5 = β6 = 0. Наконец, из равенства α1β2+α2β1+α5β6+α6β5 = 0 вытекает, что β2 = 0. Таким образом, b = v ∈ B2 2,p. Аналогично рассматриваются случаи, когда отличен от нуля один из коэффициентов α2, α3, α4. Итак, мы можем полагать, что αi = 0 для i = 1, 2, 3, 4. Так как a /∈ A2 2,p, то α6 6= 0 или α5 6= 0. Предположим, что α5 6= 0. Из равенства α5β5 = 0 получаем, что β5 = 0. Следовательно, α5β6 = 0, т.е. и β6 = 0. Из равенств αiβ5 + α5βi = 0, где i = 1, 2, 3, 4, следует, что βi = 0 для i = 1, 2, 3, 4. Значит, b = v ∈ B2 2,p. Из лемм 4 и 6 получаем Следствие 3. Γ(A2,p) ∼= Γ(B2,p) для любого простого нечетного числа p. Предложение 4. Алгебра A2,p не изоморфна алгебре B2,p для любого простого нечет- ного числа p. Доказательство. Предположим противное. Пусть ϕ -- изоморфизм B2,p на A2,p. Обо- i = xi + hx1x2 − x3x4i ∈ A2,p, i = 1, 6. значим xi = xi + hx1x2 + x3x4 − x5x6i ∈ B2,p и x′ Заметим что ϕ(B2 2,p, опреде- ленное по правилу ϕ(a + B2 2,p, a ∈ B2,p, также является изоморфизмом. 2,p, а значит, отображение ϕ : B2,p/B2 2,p) = ϕ(a) + A2 2,p → A2,p/A2 2,p) ⊆ A2 10 Отсюда следует, что существует невырожденная матрица P = (pij)6×6, pij ∈ Zp, такая, что ϕ(x1) = p11x′ ϕ(x2) = p12x′ ϕ(x3) = p13x′ ϕ(x4) = p14x′ ϕ(x5) = p15x′ ϕ(x6) = p16x′ 1 + p21x′ 1 + p22x′ 1 + p23x′ 1 + p24x′ 1 + p25x′ 1 + p26x′ 2 + p31x′ 2 + p32x′ 2 + p33x′ 2 + p34x′ 2 + p35x′ 2 + p36x′ 3 + p41x′ 3 + p42x′ 3 + p43x′ 3 + p44x′ 3 + p45x′ 3 + p46x′ 4 + p51x′ 4 + p52x′ 4 + p53x′ 4 + p54x′ 4 + p55x′ 4 + p56x′ 5 + p61x′ 5 + p62x′ 5 + p63x′ 5 + p64x′ 5 + p65x′ 5 + p66x′ 6 + b1; 6 + b2; 6 + b3; 6 + b4; 6 + b5; 6 + b6,   где b1, b2, b3, b4, b5, b6 -- некоторые элементы из A2 2,p. Учитывая свойства изоморфизма, имеем ϕ(x1x2 + x3x4 − x5x6) = = (p11x′ 1 + p21x′ 2 + p31x′ 3 + p41x′ 4 + p51x′ 5 + p61x′ 6)(p12x′ 1 + p22x′ 2 + p32x′ 3 + p42x′ 4 + p52x′ 5 + p62x′ 6)+ +(p13x′ 1 + p23x′ 2 + p33x′ 3 + p43x′ 4 + p53x′ 5 + p63x′ 6)(p14x′ 1 + p24x′ 2 + p34x′ 3 + p44x′ 4 + p54x′ 5 + p64x′ 6)− −(p15x′ 1 + p25x′ 2 + p35x′ 3 + p45x′ 4 + p55x′ 5 + p65x′ 6)(p16x′ 1 + p26x′ 2 + p36x′ 3 + p46x′ 4 + p56x′ 5 + p66x′ 6) = = (p11p62 + p12p61 + p13p64 + p14p63 − p15p66 − p16p65)x′ 1x′ 6+ +(p21p62 + p22p61 + p23p64 + p24p63 − p25p66 − p26p65)x′ 2x′ 6+ +(p31p62 + p32p61 + p33p64 + p34p63 − p35p66 − p36p65)x′ 3x′ 6+ +(p41p62 + p42p61 + p43p64 + p44p63 − p45p66 − p46p65)x′ 4x′ 6+ +(p51p62 + p52p61 + p53p64 + p54p63 − p55p66 − p56p65)x′ 5x′ 6+ +(p61p62 + p63p64 − p65p66)x′ 6x′ 6 + f (x′ 1, x′ 2, x′ 3, x′ 4, x′ 5) = 0, где f (x′ 1, x′ жащий x′ 4, x′ 3, x′ 2, x′ 6. Поскольку элементы x′ 5) -- некоторый многочлен от переменных x′ 6, x′ 6, x′ 6, x′ 6, x′ 6, x′ 2x′ 3x′ 1x′ 5x′ 4x′ 2, x′ 3, x′ 1, x′ 6x′ 5, не содер- 6 линейно независимы, 4, x′ то p11p62 + p12p61 + p13p64 + p14p63 − p15p66 − p16p65 = 0; p21p62 + p22p61 + p23p64 + p24p63 − p25p66 − p26p65 = 0; p31p62 + p32p61 + p33p64 + p34p63 − p35p66 − p36p65 = 0; p41p62 + p42p61 + p43p64 + p44p63 − p45p66 − p46p65 = 0; p51p62 + p52p61 + p53p64 + p54p63 − p55p66 − p56p65 = 0; p61p62 + p63p64 − p65p66 = 0. Изменив последнее уравнение, получим 11 p11p62 + p12p61 + p13p64 + p14p63 − p15p66 − p16p65 = 0; p21p62 + p22p61 + p23p64 + p24p63 − p25p66 − p26p65 = 0; p31p62 + p32p61 + p33p64 + p34p63 − p35p66 − p36p65 = 0; p41p62 + p42p61 + p43p64 + p44p63 − p45p66 − p46p65 = 0; p51p62 + p52p61 + p53p64 + p54p63 − p55p66 − p56p65 = 0; p61p62 + p62p61 + p63p64 + p64p63 − p65p66 − p66p65 = 0. Отсюда (p62, p61, p64, p63, −p66, −p65) · P T = 0. Так как матрица P невырожденная, то полученное равенство невозможно. Противоре- чие. Из следствия 3 и предложения 4 вытекает справедливость следующего утверждения. Следствие 4. Если в многообразии колец M все конечные кольца однозначно опреде- ляются своими графами делителей нуля, то для любого нечетного простого числа p многообразие M не содержит многообразия M2,p. Пусть M (cid:22) многообразие, в котором все конечные кольца однозначно определяются своими графами делителей нуля. В работе [1] доказано, что T (M) содержит многочлены вида mx, dx + x2g(x), причем m = pα1 s , αi ≤ 3 для всех i ≤ s, g(x) ∈ Z[x], d = 1 1 . . . pαs или d = pi1 . . . pik , где pi1, . . . , pik -- попарно различные простые делители числа m. Далее, пусть Ni = var hpαi i x = 0i ∩ M, где 1 ≤ i ≤ s. Тогда M = N1 ∨ . . . ∨ Ns и T (M) = T (N1) ∩ . . . ∩ T (Ns). Лемма 7. Если кольцо A удовлетворяет тождеству x(y − yn) = 0, n ≥ 2, а кольцо B удовлетворяет тождеству x(y − ym) = 0, m ≥ 2, то в кольце A ⊕ B выполняется тождество x(y − y(n−1)(m−1)+1) = 0. Доказательство. Пусть a, b (cid:22) произвольные элементы из кольца A. Тогда ab = abn = ab · bn−1 = (ab · bn−1)bn−1 = ab · b2(n−1) = . . . = ab1+(n−1)t, где t (cid:22) произвольное неотрицательное целое число. Другими словами, в кольце A вы- полняются тождества xy = xy1+(n−1)t, t ≥ 1. Аналогично доказывается, что кольцо B удовлетворяет тождествам xy = xy1+(m−1)s, где s (cid:22) произвольное целое неотрицатель- ное число. Значит, xy = xy1+(n−1)(m−1) (cid:22) тождество в кольце A ⊕ B. Предложение 5. T (Ni) содержит многочлен x(y − ypi), i = 1, . . . , s. 12 Доказательство. Из работы [7] следует, что T (Ni) содержит многочлены вида pαi i x и dx + x2f (x), где d = 1 или d = pi, f (x) ∈ Z[x], αi ≤ 3. Если d = 1, то идеал тождеств T (Ni) содержит многочлен x+x2f (x), и из [2] следует, что Ni порождается полем Zpi или Ni = var hx = 0i . В каждом из этих случае получаем, что T (Ni) содержит многочлен x(y − ypi). Рассмотрим случай, когда d = pi, т.е. pix + x2f (x) ∈ T (Ni). Из следствия 2 имеем, что многообразие Ni не содержит многообразия M1,pi. Это означает, что существует многочлен g(x1, . . . , xN ) ∈ T (Ni), существенно зависящий от переменных x1, . . . , xN , такой, что g(x1, . . . , xN ) /∈ {xyz, x2, pix}T . Ясно, что N ≤ 2. Пусть N = 1. В этом случае многочлен g можно записать в виде: g(x) = bx + apix + x2h(x), где h(x) ∈ Z[x], a, b ∈ Z, причем число b не делится на pi, По лемме о НОД существуют целые числа u, v, такие, что bu + piv = 1. Поскольку pix + x2f (x) ∈ T (Ni), то T (Ni) содержит многочлен вида g1(x) = x + cpix + x2h1(x) для некоторых c ∈ Z и h1(x) ∈ Z[x]. Отсюда, наконец, получаем, что (cid:0)x + cpix + x2h1(x)(cid:1) − c(cid:0)pix + x2f (x)(cid:1) = x + x2(h1(x) − cf (x)) ∈ Ni. Из [7] следует, что T (Ni) содержит многочлен x(y − ypi). Пусть теперь N = 2. Тогда можем записать g(x, y) = αxy + β(x ◦ y) + γpixy + ϕ(x, y), где ϕ(x, y) ∈ Z[x, y], α, β, γ ∈ Z, причем число α не делится на pi, а нижняя степень многочлена ϕ(x, y) больше 2. По лемме о НОД существуют целые числа u, v, такие, что αu + piv = 1. Поэтому T (Ni) содержит многочлен g1(x, y) =ug(x, y) = uαxy + uβ(x ◦ y) + uγpixy + uϕ(x, y) = =xy + uβ(x ◦ y) + (uγ − v)pixy + uϕ(x, y) = =xy + uβ(x ◦ y) + (uγ − v)(cid:0)−x2y2f (xy)(cid:1) + uϕ(x, y) = =xy + β1(x ◦ y) + ϕ1(x, y), где β1 = uβ и ϕ1(x, y) (cid:22) некоторый многочлен с целыми коэффициентами, нижняя степень которого больше 2. Итак, T (Ni) содержит многочлен g1(x, y) = xy + β1(x ◦ y) + ϕ1(x, y), где β1 ∈ Z и ϕ1(x, y) (cid:22) многочлен с целыми коэффициентами, нижняя степень которого больше 2. Рассмотрим случай, когда pi = 2. Тогда 13 g1(x, x) = x2 + 2β1x2 + ϕ1(x, x) = x2 + β1(cid:0)−x4f (x2)(cid:1) + ϕ1(x, x) = x2 + x3ϕ2(x) ∈ T (Ni) для некоторого многочлена ϕ2(x) ∈ Z[x]. Отсюда следует, что многообразие Ni удовле- творяет тождеству вида x ◦ y + ψ(x, y) = 0, где ψ(x, y) ∈ Z[x, y], причем нижняя степень многочлена ψ(x, y) больше 2. Следовательно, g1(x, y) = xy + β1 (−ψ(x, y)) + ϕ1(x, y) = xy + ψ1(x, y), где ψ1(x, y) = −β1ψ(x, y) + ϕ1(x, y) (cid:22) многочлен, нижняя степень которого больше 2. Из теоремы 1 работы [7] имеем, что Ni ⊆ var N0,2 ⊕ Z2. Следовательно, многообразие Ni удовлетворяет тождеству x(y − y2) = 0. Рассмотрим теперь случай, когда pi (cid:22) нечетное число. Из следствия 4 имеем, что многообразие Ni не содержит многообразия M2,pi. Значит, существует многочлен p(x1, . . . , xM ) ∈ T (Ni), существенно зависящий от переменных x1, . . . , xM , такой, что p(x1, . . . , xM ) /∈ {xyz, [x, y], pix}T . Ясно, что M ≤ 2. Положим сначала M = 1. Тогда можем записать p(x) = λx + µpix + νx2 + δpix2 + x3σ(x), где σ(x) ∈ Z[x], λ, µ, ν, δ ∈ Z, причем либо λ (cid:22) ненулевое число, не делящееся на pi, либо ν (cid:22) ненулевое число, не делящееся на pi. Пусть λ не равно нулю и взаимно просто с числом pi. Тогда лемме о НОД существуют целые числа u, v, такие, что λu + piv = 1. Отсюда следует, что up(x) =x + pi(µu − v)x + νux2 + δupix2 + ux3σ(x) = =x + (µu − v)(−x2f (x)) + νux2 + δupix2 + ux3σ(x), т.е. Ni удовлетворяет тождеству вида x+x2q(x) = 0, где q(x) ∈ Z[x]. Ранее мы отмечали, что в этом случае x(y − ypi) ∈ T (Ni). Пусть теперь λ = 0. Тогда число ν не равно нулю и не делится на pi. Отсюда следует, что лемме о НОД существуют целые числа u, v, такие, что νu + piv = 1. Значит, имеем up(x) =µupix + x2 + pi(δu − v)x2 + ux3σ(x) = =µupix + x2 + (δu − v)(−x4f (x2)) + ux3σ(x) = µupix + x2 + x3σ1(x) для некоторого многочлена σ1(x) ∈ Z[x]. Линеаризуя многочлен µupix + x2 + x3σ1(x), мы получим, что идеал тождеств T (Ni) содержит многочлен вида x ◦ y + w(x, y), где w(x, y) (cid:22) некоторый многочлен с целыми коэффициентами, нижняя степень которого больше 2. Из тождеств x ◦ y + w(x, y) = 0 и g1(x, y) = 0 получаем, что T (Ni) содержит многочлен вида xy+w1(x, y), где w1(x, y) ∈ Z[x, y] и нижняя степень многочлена w1(x, y) 14 больше 2. По теореме 1 статьи [7] следует, что многообразие Ni удовлетворяет тождеству x(y − ypi) = 0. Пусть, наконец, M = 2. Тогда можем записать p(x, y) = λ1xy + µ1[x, y] + ν1pixy + ω(x, y), где λ1, µ1, ν1 ∈ Z, причем λ1 не делится на pi, и ω(x, y) (cid:22) многочлен с целыми коэф- фициентами, нижняя степень которого больше 2. Используя лемму о НОД так же, как это было сделано выше, мы можем считать, что λ1 = 1, т.е. p(x, y) = xy + µ1[x, y] + ν1pixy + ω(x, y). Пользуясь тождеством pix + x2f (x) = 0, получаем, что p(x, y) = xy + µ1[x, y] + ν1(−x2y2f (xy)) + ω(x, y) = xy + µ1[x, y] + ω1(x, y) для некоторого многочлена ω1(x, y) ∈ Z[x, y], нижняя степень которого больше 2. Тогда p(x, x) = x2 + x3ω2(x), где ω2 ∈ Z[x]. Линеаризуя тождество p(x, x) = 0, получим, что многообразие Ni удовлетворяет тождеству вида x ◦ y + w(x, y) = 0, где w(x, y) (cid:22) некоторый многочлен с целыми коэффициентами, нижняя степень которого больше 2. Из тождеств x ◦ y + w(x, y) = 0 и g1(x, y) = 0 следует тождество вида xy + w1(x, y) = 0, где w1(x, y) ∈ Z[x, y] и нижняя степень многочлена w1(x, y) больше 2. По теореме 1 статьи [7] получаем, что многообразие Ni удовлетворяет тождеству x(y − ypi) = 0. Из предложения 5 и леммы 7 получаем следующее утверждение. Следствие 5. Пусть M (cid:22) многообразие, в котором все конечные кольца однозначно определяются своими графами делителей нуля. Тогда T (M) содержит многочлен вида x(y − yN ), где N ≥ 2. Теперь мы можем доказать основной результат настоящей работы. Теорема 1. Для любого многообразия M ассоциативных колец следующие условия эквивалентны: (1) Произвольное конечное кольцо из M однозначно определяется своим графом де- лителей нуля; (2) M ⊆ var hN0,p1 ⊕ . . . ⊕ N0,ps ⊕ Zpi , где s ∈ N и (pi, p) 6= (3, 2) для любого числа i ∈ {1, . . . , s}. Доказательство. Импликация (2) ⇒ (1) следует из предложения 4 работы [1]. Дока- жем импликацию (1) ⇒ (2). Пусть в многообразии M все конечные кольца однозначно определяются своими графами делителей нуля. Тогда по следствию 5 идеал тождеств T (M) содержит многочлен вида x(y − yN ), где N ≥ 2. По теореме 1.1 из статьи [7] имеем, что M ⊆ var hN0,p1 ⊕ . . . ⊕ N0,ps ⊕ Zpi , где s ∈ N и (pi, p) 6= (3, 2) для любого числа i ∈ {1, . . . , s}. Список литературы 15 [1] Кузьмина А.С. О некоторых свойствах многообразий колец, в которых конечные кольца однозначно определяются своими графами делителей нуля [Электронный ресурс]// Сибирские электронные математические известия. -- 2011. -- №8. -- С. 179 -- 190. [2] Джекобсон Н. Строение колец. -- М.: Изд-во иностр. литературы, 1961. -- 392 с. [3] Akbari S., Mohammadian A. On the zero-divisor graph of a commutative ring// Journal of Algebra. -- 2004. -- 274. -- p.847 -- 855. [4] Akbari S., Mohammadian A. On zero-divisor graphs of finite rings// Journal of Algebra. -- 2007. -- 314. -- p.168 -- 184. [5] Anderson D.F., Livingston P.S. The Zero-Divisor Graph of a Commutative Ring // Journal of Algebra. -- 1999. -- 217. -- № 2. -- p. 434 -- 447. [6] Beck I. Coloring of Commutative Rings // Journal of Algebra. -- 1988. -- 116. -- p.208 -- 226. [7] Kuzmina A.S., Maltsev Y.N. On varieties of rings whose finite rings are determined by their zero-divisor graphs, http:// arxiv.org/abs/1201.3441, to appear in Asian- European J. Math. [8] Tarski A. Equationally complete rings and relation algebras, Indag. Math., 18 (1956), 39 -- 46.
1107.3056	1	1107	2011-07-15T12:29:25	Multiple Commutator Formulas	[ "math.RA" ]	Let A be a quasi-finite R-algebra (i.e., a direct limit of module finite algebras) with identity. Let I_i, i=0,...,m, be two-sided ideals of A, \GL_n(A,I_i) the principal congruence subgroup of level I_i in GL_n(A) and E_n(A,I_i) be the relative elementary subgroup of level I_i. We prove a multiple commutator formula [E_n(A,I_0),\GL_n(A,I_1),& \GL_n(A, I_2),..., \GL_n(A, I_m)] = [E_n(A,I_0),E_n(A,I_1),E_n(A, I_2),..., E_n(A, I_m)], which is a broad generalization of the standard commutator formulas.	math.RA	math	MULTIPLE COMMUTATOR FORMULAS R. HAZRAT AND Z. ZHANG Abstract. Let A be a quasi-finite R-algebra (i.e., a direct limit of module finite algebras) with identity. Let Ii, i = 0, ..., m, be two-sided ideals of A, GLn(A, Ii) the principal congruence subgroup of level Ii in GLn(A) and En(A, Ii) be the relative elementary subgroup of level Ii. We prove a multiple commutator formula (cid:2)En(A, I0), GLn(A, I1), GLn(A, I2), . . . , GLn(A, Im)(cid:3) = (cid:2)En(A, I0), En(A, I1), En(A, I2), . . . , En(A, Im)(cid:3), which is a broad generalization of the standard commutator formulas. This result contains all the published results of commutator formulas over commutative rings and answers a problem posed by A. Stepanov and N. Vavilov (cf. Problem 4 in [24]). Introduction Let A be an associative ring with 1, GLn(A) the general linear group of degree n over A, and let En(A) be its elementary subgroup. For a two-sided ideal I of A, we denote the principal congruence subgroup of level I by GLn(A, I) and the relative elementary subgroup of level I by En(A, I) (see §1.4). One of the major contributions towards non-stable K-theory of rings is the work of Suslin [21, 23] who proved that if A is a module finite ring namely, a ring that is finitely generated as module over its center, and n ≥ 3 then En(A) is a normal subgroup of GLn(A). Thus the non-stable K1, i.e., GLn(A)/En(A), can be defined. Later Borevich and Vavilov [6] and Vaserstein [26], independently, building on Suslin's method established the standard commutator formula: Theorem 1 (Suslin, Borevich-Vavilov, Vaserstein). Let A be a module finite ring, I a two-sided ideal of A and n ≥ 3. Then En(A, I) is normal in GLn(A), i.e., Furthermore (cid:2)En(A, I), GLn(A)(cid:3) = En(A, I). (cid:2)En(A), GLn(A, I)(cid:3) = En(A, I). One natural question raised here is whether one has a "finer" mixed commutater formulas involving two ideals. In fact this had already been established by Bass for general linear groups of degrees sufficiently larger than the stable rank when he proved his celebrated classification of subgroups of GLn normalized by En (see [5, Theorem 4.2]). Theorem 2 (Bass). Let A be a ring, I, J two-sided ideals of A and n ≥ max(sr(R) + 1, 3). Then Later Mason and Stothers building on Bass' result prove ([18, Theorem 3.6, Corollary 3.9], and [16, (cid:2)En(A, I), GLn(A, J)(cid:3) = (cid:2)En(A, I), En(A, J)(cid:3). Theorem 1.3]): Theorem 3 (Mason-Stothers). Let A be a ring, I, J two-sided ideals of A and n ≥ max(sr(R) + 1, 3). Then (cid:2) GLn(A, I), GLn(A, J)(cid:3) = (cid:2)En(A, I), En(A, J)(cid:3). The first author acknowledges the support of EPSRC (Grant EP/I007784/1). The second author acknowledges the support of NSFC (Grant 10971011). The authors thank Nikolai Vavilov for suggesting the topic of the paper to them, and Anthony Bak for very useful discussions. 1 2 R. HAZRAT AND Z. ZHANG There are (counter)examples that the Mason-Strothers Theorem does not hold for general module finite rings [1]. However recently Stepanov and Vavilov [24, 25] proved Bass' Theorem 2 for any commutative ring and n ≥ 3 and the authors using Bak's localization and patching method extend it to all module finite rings [13]. We refer to this as the generalized commutator formula. In [24] it is asked whether one can establish a multiple commutator formula, namely for a commutative ring R, Ii, i = 0, ..., m, ideals of R and n ≥ 3, whether (cid:2)En(R, I0), GLn(R, I1), GLn(R, I2), . . . , GLn(R, Im)(cid:3) = (cid:2)En(R, I0), En(R, I1), En(R, I2), . . . , En(R, Im)(cid:3), (1) is valid which is a broad generalization of the standard/generalized commutator formulas. Here for sim- plicity we write [A1, A2, A3, . . . , An] for (cid:2) . . .(cid:2)[A1, A2], A3(cid:3), . . . , An(cid:3) (see §1.3). Questions of this type arise from the study of subnormal subgroups of GLn from one hand and the nilpotent structure of nonstable K1 from the other hand (see [12, §10 and §12] for a survey on these topics). In this paper we prove Formula (1) for quasi finite rings (which include module finite and commutative rings) (see Corollary 15). In particular this result contains all the published results of commutator formulas over commutative rings. In fact in Theorem 17 we show that the multiple commutator formulas are valid for any meaningful way of the distribution of commutators. To establish these results, we use the general "yoga of commutators" which are developed in [13] and [14] based on the work of Bak on the localization and patching in general linear groups (see [1, 15] and [12, §13]). In order to utilize this method, one needs to overcome two problems. First to devise an appropriate conjugation calculus to approach the identity (1) and then perform the actual calculations. Both of these are equally challenging as the nature of conjugation calculus depends on the problem in hand. In fact the term yoga of commutators is chosen to stress the overwhelming feeling of technical strain and exertion. However once this is done for general linear groups, one can adapt the approach to more complex settings, such as general quadratic groups and Chevalley groups. These shall be established in a sequel to this paper. 1. Preliminaries In this section we fix some notations. At the same time, we list some preliminary results concerning the localization and patching method without proofs. We refer to Bak's original paper [1] or a survey version in [12, §13] for details. 1.1. Let R be a commutative ring with 1, S a multiplicative closed system in R and A an R-algebra. Then S−1R and S−1A denote the corresponding localization. In the current paper, we mostly use localization with respect to the following two types of multiplicative systems. 1.) For any s ∈ R, the multiplicative system generated by s is defined as hsi = {1, s, s2, . . .}. The localization with respect to multiplicative system hsi is usually denoted by Rs and As. Note that, for any α ∈ Rs, there exists an integer n and an element a ∈ R such that α = a/sn. 2.) If m is a maximal ideal of R, and S = R\m a multiplicative system, then we denote the localization with respect to S by Rm and Am. For a multiplicative system S, the canonical localization map with respect to S is denoted by θS : R → S−1R. For the special cases mentioned above, we write θs : R → Rs and θM : R → RM , respectively. 1.2. An R-algebra A is called module finite over R, if A is finitely generated as an R-module. An R- algebra A is called quasi-finite over R if there is a direct system of module finite R-subalgebras Ai of A such that lim −→ Ai = A. Proposition 4. An R-algebra A is quasi-finite over R if and only if it satisfies the following equivalent conditions: (1) There is a direct system of subalgebras Ai/Ri of A such that each Ai is module finite over Ri and MULTIPLE COMMUTATOR FORMULAS 3 such that lim −→ Ri = R and lim −→ Ai = A. (2) There is a direct system of subalgebras Ai/Ri of A such that each Ai is module finite over Ri and each Ri is finitely generated as a Z-algebra and such that lim −→ Ri = R and lim −→ Ai = A. 1.3. Let G be a group. For any x, y ∈ G, xy = xyx−1 denotes the left x-conjugate of y. Let [x, y] = xyx−1y−1 denote the commutator of x and y. Sometimes the double commutator [[x, y], z] will be denoted simply by [x, y, z] and Thus we write [A1, A2, A3, . . . , An] for (cid:2) . . .(cid:2)[A1, A2], A3(cid:3), . . . , An(cid:3) and call it the standard form of the multiple commutator formulas. (cid:2)[A, B], C(cid:3) = [A, B, C]. The following formulas will be used frequently (sometimes without giving a reference to them), (C1) [x, yz] = [x, y](y[x, z]); (C1+) An easy induction, using identity (C1), shows that (cid:2)x, j=1 uj = 1. where by convention Q0 (C2) [xy, z] = (x[y, z])[x, z]; (C2+) As in (C1+), we have k Yi=1 ui] = k Yi=1 Qi−1 j=1 uj [x, ui], k Yi=1 (cid:2) ui, x(cid:3) = k Yi=1 Qk−i j=1 uj [uk−i+1, x]. (C3) (the Hall-Witt identity): x(cid:2)[x−1, y], z(cid:3) z(cid:2)[z−1, x], y(cid:3) y(cid:2)[y−1, z], x(cid:3) = 1; (C4) [x,y z] =y [y−1 (C5) [yx, z] =y [x,y−1 (C6) If H and K are subgroups of G, then [H, K] = [K, H]. x, z]; z]. 1.4. For any associative ring A, GLn(A) denotes the general linear group of A, and En(A) denotes the elementary subgroup of GLn(A). Let I be any two-sided ideal of A. If ρI denotes the natural ring homomorphism A → A/I, then ρI induces a group homomorphism, denoted also by ρI , ρI : GLn(A) → GLn(A/I). The congruence subgroup of level I is defined as GLn(A, I) = ker(ρI : GLn(A) → GLn(A/I)). The elementary subgroup of level I is, by definition, the subgroup generated by all elementary matrices ei,j(α) with α ∈ I. The normal closure of En(I) in En(A), the relative elementary subgroup of level I, is denoted by En(A, I). We use EL n (I) to denote the subset of En(I), which can be represented as the product L elementary matrices. EL n (I) is not necessarily a group. We have the following relations among elementary matrices which will be used in the paper: (E1) ei,j(a)ei,j(b) = ei,j(a + b). (E2) [ei,j(a), ek,l(b)] = 1 if i 6= l, j 6= k. (E3) [ei,j(a), ej,k(b)] = ei,k(ab) if i 6= k. 1.5. GLn and En define two functors from the category of associative rings to the category of groups. These functors commute with direct limits. In another words, let Ai be an inductive system of rings, and A = lim −→ Ai. Then GLn(A, J) = GLn(lim −→ Ai, lim −→ Ji) ∼= lim −→ GLn(Ai, Ji). GLn(A) = GLn(lim −→ Ai) ∼= lim −→ GLn(Ai) and En(lim −→ Also, if J is an ideal of A, then there are ideals Ji of Ai such that J = lim −→ Ai) ∼= lim −→ Ji and En(Ai). 4 R. HAZRAT AND Z. ZHANG By Proposition 4 and the above observation, we may reduce some of our problems to the case of the Noetherian rings. Let S be a multiplicative system in R, Rs with s ∈ S is an inductive system with respect to the localization map : θt : Rs → Rst. If F is a functor commuting with direct limits (here GLn and En), then This allows us to reduce our problems in any localization to the localization in one element. Starting from Section 2, we will be working in the ring At. However, eventually we need to return to the ring A. The following Lemma provides a way to "pull back" elements from GLn(At) to GLn(A). F(S−1R) = lim −→ F (Rs). Lemma 5. [1, Lemma 4.10] Let A be a module finite R-algebra, where R is a commutative Noetherian ring. Then for any t ∈ R, there exists a positive integer l such that the homomorphism θt : GLn(A, tlA) −→ GLn(At) is injective. Definition 6. Let A be an R-algebra, I a two-sided ideal of A, t ∈ R, and l a positive integer. Define En(tlA, tlI) to be a subgroup of En(A, tlI) generated by eei,j(tlα) for all α ∈ I, e ∈ En(tlA) and 1 ≤ i 6= j ≤ n. Here by tlI, we are considering the image of t ∈ R in A under the algebra structure homomorphism. It is clear that tlI is also an ideal of A. For any element α ∈ A, we use En(tlA, tlα) to denote the subgroup generated by eei,j(tlα) for all e ∈ En(tlA) and 1 ≤ i, j ≤ n. From the definition, it is clear that En(tlA, tlI) is normalized by En(tlA). This will be used throughout out the calculations. Also, by Lemma 5, both En(tlA, tlI) and En(tlA, tlα) are embedded in GLn(At) for a sufficiently large integer l. This fact will be used in Theorem 14. 1.6. Finally we need the following elementary conjugation calculus, Lemmas 7, 8 and 11 from [13], re- spectively. Note that in Equations 2, 3 and 4 the calculations take place in the group En(At). Lemma 7 (cf. [13]). Let A be a module finite R-algebra, I, J two-sided ideals of A, a, b, c ∈ A and t ∈ R. If m, l are given, there is an integer p such that there is an integer p such that E 1 n( c tm )En(tpA, tphai) ⊆ En(tlA, tlhai), and there is an integer p such that E 1 n( c tm )(cid:2)En(tpA, tphai), En(tpA, tphbi)(cid:3) ⊆ (cid:2)En(tlA, tlhai), En(tlA, tlhbi)(cid:3), hEn(tpA, tpI), E1 n(cid:0) J tm(cid:1)i ⊆ (cid:2)En(tlA, tlI), En(tlA, tlJ)(cid:3). By Lemma 7, one obtains the following result easily. The proof is left to the reader. (2) (3) (4) Lemma 8. Let A be a module finite R-algebra, I, J two-sided ideals of A, a, b, c ∈ A and t ∈ R. If m, l, L are given, there is an integer p such that hEn(tpA, tpI), EL tm(cid:1)E1 n(cid:0) A n(cid:0) J tm(cid:1)i ⊆ (cid:2)En(tlA, tlI), En(tlA, tlJ)(cid:3). 2. Commutator subgroups (5) In this section we study the relations between multiple commutator subgroups over a quasi-finite algebra. The proofs are heavily depend on the computation in [13] (see Lemma 7). Throughout the section ideals are two sided and we assume n ≥ 3 for any general linear group GLn. We record the following well-known lemma originally established by Suslin and Vaserstein (cf. [1, Lemma 4.8]) which is needed in computations. Lemma 9. Let A be a ring and I a two-ideal of A. Then En(A, I) is generated as a group by the elements MULTIPLE COMMUTATOR FORMULAS 5 where i 6= j, a ∈ A and α ∈ I. ei,j(a)ej,i(α), Using Lemma 9 it is not hard to prove that En(A, I 2) ⊆ En(I) (see [1, Corollary 4.9] and [22, Proposi- tion 2]). This containment can be slightly generalized to the case of two ideals. The following Lemma will be used throughout our calculations. Lemma 10. Let A be a ring and I, J be two-ideals of A. Then En(A, IJ + JI) ⊆ (cid:2)En(I), En(J)(cid:3) ⊆ (cid:2)En(A, I), En(A, J)(cid:3) ⊆ GLn(A, IJ + JI). Proof. The proof is routine by using Lemma 9 and is left to the reader. (cid:3) Lemma 11. Let A be a ring and I, J be two-ideals of A. Then (cid:2)En(A, I), En(A, J)(cid:3) is generated as a group by the elements of the form where 1 ≤ i 6= j ≤ n, α ∈ I, β ∈ J , a ∈ A and c ∈ En(A). c(cid:2)ej,i(α), ei,j (a)ej,i(β)(cid:3), c(cid:2)ej,i(α), ei,j (β)(cid:3), cei,j(αβ), and cei,j(βα), (6) Proof. A typical generator of (cid:2)En(A, I), En(A, J)(cid:3) is of the form [e, f ], where e ∈ En(A, I) and f ∈ En(A, J). Thanks to Lemma 9, we may assume that e and f are products of elements the form where a, b ∈ A, α ∈ I and β ∈ J, respectively. Applying (C1+) and then (C2+), one gets that ei =ep′,q′ (a) eq′,p′(α) and fj =ep,q(b) eq,p(β), (cid:2)En(A, I), En(A, J)(cid:3) is generated by the elements of the form c(cid:2)ei′,j′ (a)ej ′,i′(α), ei,j (b)ej,i(β)(cid:3), where c ∈ En(A). Furthermore, c(cid:2)ei′ ,j′ (a)ej ′,i′(α), ei,j (b)ej,i(β)(cid:3) =cei′,j′ (a) (cid:2)ej ′,i′(α), ei′ ,j′ (−a)ei,j (b)ej,i(β)(cid:3). The normality of En(A, J) implies that ei′,j′ (−a)ei,j (b)ej,i(β) ∈ En(A, J), which is a product of ep,q(a)eq,p(β), a ∈ A and β ∈ J by Lemma 9. Again by (C1+), one reduces the proof to the case of showing that is a product of the generators listed in (6). We need to consider following cases: (cid:2)ei′,j ′(α), ei,j (a)ej,i(β)(cid:3) • If i′ = j, j′ = i: Then there is nothing to proof. • if i′ = j, j′ 6= i: (cid:2)ej,j ′(α), ei,j (a)ej,i(β)(cid:3) = ei,j(a)(cid:2)ei,j (−a)ej,j ′(α), ej,i(β)(cid:3) = ei,j(a)(cid:2)[ei,j(−a), ej,j ′(α)]ej,j ′(α), ej,i(β)(cid:3) = ei,j(a)(cid:2)ei,j ′(−aα)ej,j ′(α), ej,i(β)(cid:3). Applying now (C2), [ei,j ′(−aα)ej,j ′(α), ej,i(β)] = (cid:0)ei,j′ (−aα)[ej,j ′(α), ej,i(β)](cid:1)[ei,j ′(−aα), ej,i(β)] = [ei,j ′(−aα), ej,i(β)] = [ej,i(β), ei,j ′(−aα)]−1 = ej,j ′(−βaα)−1 = ej,j ′(βaα) Thus which satisfies the lemma. (cid:2)ej,j ′(α), ei,j (a)ej,i(β)(cid:3) = ei,j (a)ej,j ′(βaα) 6 R. HAZRAT AND Z. ZHANG • if i′ 6= j, j′ = i: The argument is similar to the previous case. • if i′ 6= j, j′ 6= i: We consider four cases: -- if i′ = i, j′ = j: -- if i′ = i, j′ 6= j: -- if i′ 6= i, j′ = j: -- if i′ 6= i, j′ 6= j: (cid:2)ei,j(α), ei,j (a)ej,i(β)(cid:3) = ei,j(a)(cid:2)ei,j(α), ej,i(β)(cid:3). (cid:2)ei,j ′(α), ei,j (a)ej,i(β)(cid:3) = ei,j (a)(cid:2)ei,j ′(α), ej,i(β)(cid:3) = ei,j (a)ej,j ′(−βα). (cid:2)ei′,j(α), ei,j (a)ej,i(β)(cid:3) = ei,j (a)(cid:2)ei′,j(α), ej,i(β)(cid:3) = ei,j (a)ei,i′(αβ). This finishes the proof. (cid:2)ei′,j ′(α), ei,j (a)ej,i(β)(cid:3) = 1. (cid:3) Denote by EL tm , K n(cid:0) A tm(cid:1) the product of L elements (or fewer) of the form E 1 In the following two Lemmas, as in Lemma 7, all the calculations take place in the fraction ring At (see §1.6). All the subgroups used in the Lemmas, such as En(A, I) or GLn(A, J) are in fact the images of these groups in GLn(At) under the ring homomorphisms A → At. This allows us to use Lemmas such as Lemma 10 and the generalized commutator formula on these subgroups, precisely because these are homomorphic images of the similar subgroups in GLn(A) which Lemma 10, etc. hold. n( A tm )E1 n(cid:0) K tm(cid:1) (see also §1.4). Lemma 12. Let A be a module finite R-algebra, I, J two-sided ideals of A, and t ∈ R. If e ∈ GLn(At, Jt), there is an integer p such that for any g ∈ GLn(A, tpI) [e, g] ∈ GLn(cid:0)A, tl(IJ + JI)(cid:1). Proof. Note that all the entries of g − 1 and g−1 − 1 are in tpI (to emphasize our convention, they are in the image of tpI under the homomorphism θ : A → At) and all the entries of e − 1 and e−1 − 1 are in Jt. Choose k ∈ N such that one can write all the entries of e − 1 and e−1 − 1 of the form j/tk, j ∈ J. Let g = 1 + ε e = 1 + δ and and g−1 = 1 + ε′ e−1 = 1 + δ′. A straightforward computation shows that ε + ε′ + εε′ = ε + ε′ + ε′ε = 0 δ + δ′ + δδ′ = δ + δ′ + δ′δ = 0. By the equalities above, one has [e, g] = [1 + δ, 1 + ε] = 1 + δ′ε′ + εδ′ + εδ′ε′ + δδ′ε′ + δεδ′ + δεδ′ε′. So the entries of [e, g] − 1 belong to tp−2k(IJ + JI). We finish the proof by choosing p ≥ l + 2k. (cid:3) The following lemma is crucial for proving the main result, i.e., Theorem 14 of this paper. Lemma 13. Let A be a module finite R-algebra, I, J, K two-sided ideals of A and t ∈ R. For any given e2 ∈ En(At, Kt) and an integer l, there is a sufficiently large integer p, such that where e1 ∈ [En(tpI), En(A, J)]. [e1, e2] ∈ h(cid:2)En(A, tlI), En(A, tlJ)(cid:3), En(A, tlK)i. (7) Proof. For any given e2 ∈ En(At, Kt), one may find some positive integers m and L, such that MULTIPLE COMMUTATOR FORMULAS 7 e2 ∈ EL n(cid:0) A tm , K tm(cid:1). Applying the identity (C1+) and repeated application of (2) in Lemma 7, we reduce the problem to show that γ tm )i ⊆ h(cid:2)En(A, tlI), En(A, tlJ)(cid:3), En(A, tlK)i, ), ek,j ′( tm+p′ )] for some integer γ We use a variant of the Hall-Witt identity (see (C3)) where c ∈ E1 p′. Then n( A to obtain h[En(tpI), En(A, J)], cei′,j ′( tm ) and γ ∈ K. We further decompose ei′,j ′( γ tm )i = he1,(cid:2)cei′,k(tp′ he1, cei′,j ′( γ γ ), cek,j ′( tm ) = [ei′,k(tp′ tm+p′ )(cid:3)i. (cid:2)x, [y−1, z](cid:3) = y−1x(cid:2)[x−1, y], z(cid:3) y−1z(cid:2)[z−1, x], y(cid:3), he1,(cid:2)cei′,k(tp′ =y−1x(cid:20)he−1 y−1z(cid:20)hcek,j ′( tm+p′ )(cid:21)× )(cid:21), tm+p′ ), e1i, cei′,k(−tp′ n( A tm+p′ ) and as before c ∈ E1 tm+p′ )(cid:3)i = )i, cek,j ′( 1 , cei′,k(−tp′ ), cek,j ′( −γ γ γ γ (8) tm ) ⊆ E1 n( A tm+p′ ). We will look at where x = e1, y = cei′,k(−tp′ each of the two factors of (8) separately. ), z = cek,j ′( By (2) in Lemma 7, for any given p′′, one may find a sufficiently large p′ such that y = cei′,k(−tp′ ) ∈ En(tp′′ A, tp′′ A) ⊆ En(A). (9) Then he−1 1 , cei′,k(−tp′ )i ∈ (cid:2)[En(tpI), En(A, J)], En(A)(cid:3) ⊆ (cid:2) GLn(A, tp(IJ + JI)), En(A)(cid:3) ⊆ En(A, tp(IJ + JI)). Set p1 = p. Thanks to Lemma 10, Hence we obtain that En(cid:0)A, tp1(IJ + JI)(cid:1) ⊆ hEn(t⌊ p1 y−1x(cid:20)he−1 )i, cek,j ′( 1 , cei′,k(−tp′ 2 ⌋A), En(cid:0)t⌊ p1 tm+p′ )(cid:21) ∈ y−1x(cid:20)En(t⌊ p1 γ 2 ⌋(IJ + JI)(cid:1)i ⊆ En(cid:0)t⌊ p1 2 ⌋A, t⌊ p1 2 ⌋A, t⌊ p1 2 ⌋(IJ + JI)), cek,j ′( 2 ⌋(IJ + JI)(cid:1). tm+p′ )(cid:21), γ where x ∈ [En(tp1I), En(A, J)], y ∈ En(tp′′ sufficiently large p1, such that A, tp′′ A). By Lemma 8, for any given integer l′ we may find a y−1x(cid:20)En(t⌊ p1 2 ⌋A, t⌊ p1 2 ⌋(IJ + JI)),cek,j ′( γ tm+p′ )(cid:21) ∈ y−1xhEn(cid:0)t2l′ A, t2l′ A, tl′ I), En(tl′ A, tl′ A, tl′ I), En(tl′ A, tl′ A, tl′ I), y−1xEn(tl′ ⊆ y−1xh(cid:2)En(tl′ ⊆ y−1xh(cid:2)En(tl′ = h(cid:2)y−1xEn(tl′ A, t2l′ K)i (IJ + JI)(cid:1), En(t2l′ K)i A, t2l′ J)(cid:3), En(t2l′ K)i J)(cid:3), En(tl′ J)(cid:3), y−1xEn(tl′ A, tl′ A, tl′ A, tl′ K)i, 8 R. HAZRAT AND Z. ZHANG where by definition y−1x ∈ En( A large l′, such that t0 , A t0 ). By (2) in Lemma 7, for any given integer l, we may find a sufficiently y−1xh(cid:2)En(tl′ A, tl′ I), En(tl′ A, tl′ J)(cid:3), En(tl′ A, tl′ K)i ⊆ h(cid:2)En(tlA, tlI), En(tlA, tlJ)(cid:3), En(tlA, tlK)i. This shows that for any given l, one may find a sufficiently large p1 such that the first factor of (8) y−1x(cid:20)he−1 1 , cei′,k(−tp′ )i, cek,j ′( γ tm+p′ )(cid:21) ∈ h(cid:2)En(tlA, tlI), En(tlA, tlJ)(cid:3), En(tlA, tlK)i. Next we consider the second factor of (8), y−1z(cid:20)hcek,j ′( −γ tm+p′ ), e1i, cei′,k(−tp′ )(cid:21). Set p2 = p. Note that and e1 ∈ (cid:2)En(tp2I), En(A, J)(cid:3) ⊆ GLn(cid:0)A, tp2(IJ + JI)(cid:1) cek,j ′( γ tm+p′ ) ∈ E 1 n( A tm+p′ ) E1 n( K tm+p′ ), where p′ is given by (9) from the first part of the proof. We may apply Lemma 12 to find a sufficiently large p2 such that hcek,j ′( −γ tm+p′ ), e1i ∈ GLn(cid:16)A, tp′′ (K(IJ + JI) + (IJ + JI)K)(cid:17) (10) for any given p′′. Using the commutator formula together with (9), one gets y−1z(cid:20)hcek,j ′( −γ tm+p′ ), e1i, cei′,k(−tp′ )(cid:21) ∈ y−1zEn(cid:16)A, tp′′(cid:0)K(IJ + JI) + (IJ + JI)K(cid:1)(cid:17) Applying Lemma 10 twice, one gets En(cid:16)A, tp′′(cid:0)K(IJ + JI) + (IJ + JI)K(cid:1)(cid:17) ⊆ (cid:20)En(cid:16)t⌊ 2p′′ ⊆ (cid:20)hEn(t⌊ p′′ 3 ⌋(cid:0)(IJ + JI) + (IJ + JI)(cid:1)(cid:17), En(cid:16)t⌊ p′′ 3 ⌋I), En(t⌊ p′′ 3 ⌋J)i, En(t⌊ p′′ 3 ⌋K)(cid:21). 3 ⌋K(cid:17)(cid:21) Hence, we have y−1z(cid:20)hcek,j ′( −γ tm+p′ ), e1i, cei′,k(−tp′ )(cid:21) ⊆ y−1z(cid:20)hEn(t⌊ p′′ = (cid:20)hy−1zEn(t⌊ p′′ 3 ⌋J)i, En(t⌊ p′′ 3 ⌋K)(cid:21) 3 ⌋J)i, y−1zEn(t⌊ p′′ 3 ⌋K)(cid:21). 3 ⌋I), En(t⌊ p′′ 3 ⌋I), y−1zEn(t⌊ p′′ Now applying (2) in Lemma 7 to every components of the commutator above, we may find a sufficiently large p′′ such that for any given l, (cid:20)hy−1zEn(t⌊ p′′ 3 ⌋I), y−1zEn(t⌊ p′′ 3 ⌋J)i, y−1zEn(t⌊ p′′ 3 ⌋K)(cid:21) ⊆ h(cid:2)En(tlA, tlI), En(tlA, tlJ)(cid:3), En(tlA, tlK)i. Choose p2 in (10) according to this p′′ and then consider p to be the largest among p1 and p2. This finishes the Lemma. (cid:3) MULTIPLE COMMUTATOR FORMULAS 9 3. Main result Now we are in a position to prove the main result of the paper, namely the multiple commutator formulas. We first study the 3-folded commutator formula. The n-folded commutator formula is an easy application of the following theorem. Note that so far most of the conjugation calculus has been performed in At. Using the fact that for a suitable positive integer l, by Lemma 5, the restriction of θt to GLn(A, tlA) induces an injective homomorphism θt : GLn(A, tlA) → GLn(At), we are able to "pull back" the elements into the group GLn(A). This will be used in the Theorem 14. Theorem 14. Let A be a quasi-finite R-algebra and I, J, K be two-sided ideals of A. Then for n ≥ 3, h(cid:2)En(A, I), GLn(A, J)(cid:3), GLn(A, K)i = h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i (11) Proof. The functors En and GLn commute with direct limits. By Proposition 4 and §1.5, one reduces the proof to the case A is finite over R and R is Noetherian. First by the generalized commutator formula, we have Thus it suffices to prove the following equation (cid:2)En(A, I), GLn(A, J)(cid:3) = (cid:2)En(A, I), En(A, J)(cid:3). (12) h(cid:2)En(A, I), En(A, J)(cid:3), GLn(A, K)i = h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i. By Lemma 11, (cid:2)En(A, I), En(A, J)(cid:3) is generated by the conjugates of the following four types of elements e = ei,j(αβ), and e = ei,j(βα), e = hej,i(α), ei,j (r)ej,i(β)i, e = (cid:2)ej,i(α), ei,j (β)(cid:3), where i 6= j, α ∈ I, β ∈ J. We claim that for any g ∈ En(A, K), (cid:2)e, g] ∈ h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i. For any maximal ideal ml ⊳ R, choose a tl ∈ R\ml and an arbitrary positive integer pl. (We will later is not contained in any maximal ideal, choose pk according to Lemma 13.) Since the collection of all tpl l we may find a finite number of tl and xl ∈ R, l = 1, . . . , k (relabeling if necessary) such that tpl l xl = 1. Xl First we take the generators of the first kind, namely the conjugates of e = hej,i(α), ei,j (r)ej,i(β)i. Consider e = hej,i(α), ei,j (r)ej,i(β)i = hej,i(cid:16)(Xl tpl l xl)α(cid:17), ei,j (r)ej,i(β)i = hYl ej,i(tpl l xlα), ei,j (r)ej,i(β)i. By (C2+) identity, e = hYl e = (cid:16)e1hej,i(tp1 ej,i(tpl l xlα), ei,j (r)ej,i(β)i can be written as a product of the following form: k xkα), ei,j (r)ej,i(β)i(cid:17), (13) 2 x2α), ei,j (r)ej,i(β)i(cid:17) · · ·(cid:16)ekhej,i(tpk 1 x1α), ei,j (r)ej,i(β)i(cid:17)(cid:16)e2hej,i(tp2 where e1, e2, . . . em ∈ En(A). Note that all ei's are products of elementary matrices of the form ej,i(A). Thus el = ej,i(al), l = 1, . . . , k, which clearly commutes with ej,i(a) for any a ∈ A. So the commutator (13) equals to e = (cid:16)hej,i(tp1 1 x1α), e1 ei,j (r)ej,i(β)i(cid:17)(cid:16)hej,i(tp2 2 x2α), e2 ei,j(r)ej,i(β)i(cid:17) · · ·(cid:16)hej,i(tpk k xkα), ek ei,j(r)ej,i(β)i(cid:17). (14) Thus [e, g] = (cid:20)hej,i(α), ei,j (r)ej,i(β)i, g(cid:21) = (cid:20)hYl ej,i(tpl l xlα), ei,j (r)ej,i(β)i, g(cid:21). At. Note that all hej,i(tpi′ hej,i(tpi′ 10 R. HAZRAT AND Z. ZHANG Using (C2+) and in view of (14) we obtain that [e, g] is a product of the conjugates in En(A) of (cid:20)hej,i(tpi′ i′ xi′α), ej,i(al)ei,j (r)ej,i(β)i, g(cid:21), where al ∈ A and l = 1, . . . , k. For any maximal ideal m of R, the ring Am contains Km as an ideal. Consider the natural homomorphism θm : A → Am which induces a homomorphism (call it again θm) on the level of general linear groups, θm : GLn(A) → GLn(Am). Therefore, for g ∈ GLn(A, K), θm(g) ∈ GLn(Am, Km). Since Am is module finite over the local ring Rm, Am is semilocal [4, III(2.5), (2.11)], therefore its stable rank is 1. It follows that GLn(Am, Km) = En(Am, Km) GL1(Am, Km) (see [11, Th. 4.2.5]). So θm(g) can be decomposed as θm(g) = εh, where ε ∈ En(Am, Km) and h is a diagonal matrix all of whose diagonal coefficients are 1, except possibly the k-th diagonal coefficient, and k can be chosen arbitrarily. By (§1.5), we may reduce the problem to the case At with t ∈ R\m. Namely θt(g) is a product of ε and h, where ε ∈ En(At, Kt), and h is a diagonal matrix with only one non-trivial diagonal entry which lies in the i, j rows and the i, j columns. By the assumption n > 2, we may choose h so that it commutes with i′ xi′α), ej,i(al)ei,j (r)ej,i(β)i with i′ = 1, . . . , k differ from the identity matrix at only i′ xi′α), ej,i(al)ei,j (r)ej,i(β)i. This allows us to reduce our consideration to the case (cid:20)hej,i(tpi′ i′ xkα), ej,i(al)ei,j (r)ej,i(β)i, ε(cid:21). By Lemma 13, one gets that for any given li′, there is a sufficiently large pi′ for every i′ = 1, . . . k, such that (cid:20)hej,i(tpi′ i′ xkα), ej,i(al)ei,j (r)ej,i(β)i, ε(cid:21) ∈ h(cid:2)En(A, tli′ I), En(A, tli′ J)(cid:3), En(A, tli′ K)i. Let's choose every li to be large enough so that the restriction of θt : GLn(tli′ A) → GLn(At) is injective. Then for every i′, we have (cid:20)hej,i(tpi′ i′ xkα), ej,i(al)ei,j (r)ej,i(β)i, ε(cid:21) ∈ h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i. Hence [e, g] ∈ h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i. When the generator is of the second kind, e = [ei,j(α), ej,i(β)], a similar argument goes through, which is left to the reader. Now consider the generators of the 3rd and 4th kind, namely, the conjugates of the following two types of elements, e = ei,j(αβ), or e = ei,j(βα). By the normality of En(A, IJ + JI), the conjugates of e are in En(A, IJ + JI). Then By the generalized commutator formula, one obtains [e, g] ∈ (cid:2)En(A, IJ + JI), GLn(A, K)(cid:3). Now applying Lemma 10, we finally get (cid:2)En(A, IJ + JI), GLn(A, K)(cid:3) = (cid:2)En(A, IJ + JI), En(A, K)(cid:3). [En(A, IJ + JI), En(A, K)] ⊆ h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i. Therefore [e, g] ∈ h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i. This proves our claim. Let e ∈ (cid:2)En(A, I), GLn(A, J)(cid:3), and g ∈ GLn(A, K). Then by Lemma 11, e = c1e1 × c2e2 × · · · × ckek MULTIPLE COMMUTATOR FORMULAS with ci ∈ En(A) and ei takes the form in (6). Thanks to (C2+) identity, it suffices to show that The normality of En and GLn groups reduces the problem to show that [c1e1, g] ∈ h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i. But this exactly what has been shown above. This completes the proof. [ei, g] ∈ h(cid:2)En(A, I), En(A, J)(cid:3), En(A, K)i, i = 1, . . . , k. 11 (cid:3) Corollary 15. Let A be a quasi-finite ring with identity and Ii, i = 0, ..., m, be two-sided ideals of A. Then hEn(A, I0), GLn(A, I1), GLn(A, I2), . . . , GLn(A, Im)i = hEn(A, I0), En(A, I1), En(A, I2), . . . , En(A, Im)i. (15) Proof. We prove the statement by induction. For i = 1 this is the generalized commutator formula (cid:2)En(A, I0), GLn(A, I1)(cid:3) = (cid:2)En(A, I0), En(A, I1)(cid:3) which was proved in [13]. For i = 2, this is proved in Theorem 14 which will be the first step of induction. Suppose the statement is valid for i = m − 1 (i.e., there are m ideals in the commutator formula). To prove (15), using Theorem 14, we have (cid:20)h(cid:2)En(A, I0), GLn(A, I1)(cid:3), GLn(A, I2)i, GLn(A, I3), . . . , GLn(A, Im)(cid:21) = (cid:20)h(cid:2)En(A, I0), En(A, I1)(cid:3), En(A, I2)i, GLn(A, I3), . . . , GLn(A, Im)(cid:21). By Lemma 10, [En(A, I0), En(A, I1)] ⊆ GLn(A, I0I1 + I1I0). Thus (cid:20)h(cid:2)En(A, I0), En(A, I1)(cid:3), En(A, I2)i, GLn(A, I3), . . . , GLn(A, Im)(cid:21) ⊆ (cid:20)h GLn(A, I0I1 + I1I0), En(A, I2)i, GLn(A, I3), . . . , GLn(A, Im)(cid:21). Since there are m groups involved in the commutator subgroups in the right hand side, by induction we get (cid:20)h GLn(A, I0I1 + I1I0), En(A, I2)i, GLn(A, I3), . . . , GLn(A, Im)(cid:21) = (cid:20)hEn(A, I0I1 + I1I0), En(A, I2)i, En(A, I3), . . . , En(A, Im)(cid:21). Finally again by Lemma 10, En(A, I0I1 + I1I0) ⊆ (cid:2)En(A, I0), En(A, I1)(cid:3). Replacing this in the above equation we obtain that the left hand side of (15) is contained in the right hand side. The opposite inclusion is obvious. This completed the proof. (cid:3) The following corollary shows in fact it doesn't matter where the elementary subgroup appears in the multiple commutator formula. Corollary 16. Let A be a quasi-finite ring with identity and Ii, i = 0, ..., m, be two-sided ideals of A. Let Gi be subgroups of GLn(A) such that If there is an index j such that Gj = En(A, Ij), then En(A, Ii) ⊆ Gi ⊆ GLn(A, Ii), for i = 0, . . . , m. (cid:2)G0, G1, . . . , Gm(cid:3) = hEn(A, I0), En(A, I1), En(A, I2), . . . , En(A, Im)i. 12 R. HAZRAT AND Z. ZHANG Proof. Define two-sided ideals of A inductively as follows I0 = I0 Ik = IkIk−1 + Ik−1Ik, where k = 1, . . . , m. Now the proof of the lemma divides into several cases: If j = 0, the proof follows directly from Corollary 15. If j = 1, by a basic property of 2-fold commutator subgroups, (cid:2)G0, En(A, I1)(cid:3) = (cid:2)En(A, I1), G0(cid:3), so we reduce the problem to the case of j = 0. When j = k with k ≥ 2, we have (cid:2)G0, G1, . . . , Gk, Gk+1, . . . , Gm(cid:3) = h(cid:2)G0, G1, . . . Gk(cid:3), Gk+1, . . . , Gmi = h(cid:2)G0, G1, . . . , Gk−1, En(A, Ik)(cid:3), Gk+1, . . . , Gmi. Furthermore, (cid:2)G0, G1, . . . , En(A, Ik)(cid:3) = h(cid:2)G0, G1, . . . , Gk−1(cid:3), En(A, Ik)i, and it follows from Lemma 12, by putting t = 1, that (cid:2)G0, G1, . . . , Gk−1(cid:3) ⊆ GLn(A, Ik−1). By the generalized commutator formula and h(cid:2)En(A, I0), En(A, I1), . . . , En(A, Ik−1)(cid:3), En(A, Ik)i ⊆ h(cid:2)G0, G1, . . . Gk−1(cid:3), En(A, Ik)i Lemma 10 we have ⊆ (cid:2) GLn(A, Ik−1), En(A, Ik)(cid:3) = (cid:2)En(A, Ik−1), En(A, Ik)(cid:3) ⊆ h(cid:2)En(A, I0), En(A, I1), . . . , En(A, Ik−1)(cid:3), En(A, Ik)i. So and therefore (cid:2)G0, G1, . . . , Gk−1, En(A, Ik)(cid:3) = h(cid:2)En(A, I0), En(A, I1), . . . , En(A, Ik−1)(cid:3), En(A, Ik)i, (cid:2)G0, G1, . . . , Gm(cid:3) = (cid:2)En(A, I0), En(A, I1), . . . , En(A, Ik−1), En(A, Ik), Gk+1, . . . , Gm(cid:3). Finally, we finish the proof by applying Corollary 15. (cid:3) We finish the paper by a most general multiple commutator formula. Note that taking a commutator is a binary operation, and for G1, . . . , Gn, n ≥ 3, there any many ways to insert the commutator brackets [ , ] to make the sequence into a meaningful multi-commutator expression. For example for n = 4, we can have the following two arrangements h(cid:2)G0, [G1, G2](cid:3), G3i and h(cid:2)G0, G1(cid:3),(cid:2)G2, G3(cid:3)i among many others. We denote by qG0, G1, . . . , Gmy a "meaningful" multi-commutator formula. Theorem 17. Let A be a quasi-finite ring with identity and Ii, i = 0, ..., m, be two-sided ideals of A. Let Gi be subgroups of GLn(A) such that If there is an index j such that Gj = En(A, Ij), then En(A, Ii) ⊆ Gi ⊆ GLn(A, Ii), for i = 0, . . . , m. qG0, G1, . . . , Gmy = qEn(A, I0), En(A, I1), . . . , En(A, Im)y. Proof. For simplicity denote En(A, Ii) by Ei. The proof is by induction on m. For m = 0 and m = 1 there is nothing to prove. For m=2, the commutator qG0, G1, G2y can take one of the the forms (1) h(cid:2)G0, G1(cid:3), E2i, (2) hE0,(cid:2)G1, G2(cid:3)i, (3) h(cid:2)E0, G1(cid:3), G2i, (4) h(cid:2)G0, E1(cid:3), G2i, MULTIPLE COMMUTATOR FORMULAS 13 (5) hG0,(cid:2)E1, G2(cid:3)i, (6) hG0,(cid:2)G1, E2(cid:3)i. Since by (C(6)), for two subgroups H and K, we have [H, K] = [K, H], we can reduce the cases (1) and (2) and (3) -- (6) to each other, respectively, and therefore it is enough to prove the theorem for the cases (1) and (3). For the first arrangement (1), using Lemma 12, for t = 1, Lemma 10 and the generalized commutator formula we have qE0, E1, E2y = h(cid:2)E0, E1(cid:3), E2i ⊆ h(cid:2)G0, G1(cid:3), E2i ⊆ h(cid:2) GLn(A, I0), GLn(A, I1)(cid:3), En(A, I2)i ⊆ h GLn(A, I0I1 + I1I0), En(A, I2)i = hEn(A, I0I1 + I1I0), En(A, I2)i ⊆ h(cid:2)En(A, I0), En(A, I1)(cid:3), En(A, I2)i = h(cid:2)E0, E1(cid:3), E2i = qE0, E1, E2y. This shows that qG0, G1, G2y = qE0, E1, E2y. The arrangement (3) (and therefore (4) -- (6)) follows imme- diately from Theorem 14. For the main step of induction, we consider two cases. Suppose first there is a mixed commutator [Gi, Gi+1] in qG0, G1, . . . , Gmy, where neither Gi nor Gi+1 is the fixed elementary subgroup Ej. Then qG0, G1, . . . , Gmy = qG0, G1, . . . , [Gi, Gi+1], . . . , Gmy ⊆ qG0, G1, . . . ,(cid:2) GLn(A, Ii), GLn(A, Ii+1)(cid:3), . . . , Gmy ⊆ qG0, G1, . . . , GLn(A, IiIi+1 + Ii+1Ii), . . . , Gmy. (16) Note that there are one fewer ideal involved in the last commutator formula (i.e., m − 1 ideals) which also contains an elementary subgroup, and so by induction qG0, G1, . . . , GLn(A, IiIi+1 + Ii+1Ii), . . . , Gmy = qE0, E1, . . . , En(A, IiIi+1 + Ii+1Ii), . . . , Emy ⊆ qE0, E1, . . . ,(cid:2)En(A, Ii), En(A, Ii+1)(cid:3), . . . , Emy = qE0, E1, . . . , Emy. (17) Putting 16 and 17 together, we get qG0, G1, . . . , Gmy = qE0, E1, . . . , Emy. For the remaining case, suppose now that if there is a mixed commutator of the form [Gi, Gi+1], in qG0, G1, . . . , Gmy, then one of Gi or Gi+1 is our fixed elementary subgroup Ej. Write qG0, G1, . . . , Gmy = hqG0, G1, . . . , Gky, qGk+1, . . . , Gmyi. Since the fixed elementary subgroup Ej is in one of the factors, one of qG0, G1, . . . , Gky or qGk+1, . . . , Gmy has to have a mixed commutator of the form [Gi′ , Gi′+1] with neither Gi′ nor Gi′+1 the fixed Ej, which has been excluded from the outset. This forces k = 0 or k = m − 1, i.e., qG0, G1, . . . , Gmy = hG0, qG1, . . . , Gmyi, or qG0, G1, . . . , Gmy = hqG0, G1, . . . , Gm−1y, Gmi. Repeating this argument, by an easy induction and (C(6)) one can see that because of absence of [Gi, Gi+1], the multiple commutator qG0, G1, . . . , Gmy has the standard form (see §1.3) qG0, G1, . . . , Gmy = (cid:2)Gi0 , Gi1, . . . , Gim(cid:3). 14 R. HAZRAT AND Z. ZHANG By Corollary 16 Now using (C(6)) again and re-arranging Ei in the reverse order we get (cid:2)Gi0, Gi1 , . . . , Gim(cid:3) = (cid:2)Ei0, Ei1 , . . . , Eim(cid:3). (cid:2)Ei0, Ei1, . . . , Eim(cid:3) = qE0, E1, . . . , Emy. This finishes the proof. (cid:3) References [1] A. Bak, Nonabelian K-theory: the nilpotent class of K1 and general stability, K-Theory 4 (1991), 363 -- 397. 2, 4, 5 [2] A. Bak, R. Hazrat, N.A. Vavilov, Localization completion strikes again: relative K1 is nilpotent by abelian. J. Pure Appl. Algebra 213 (2009), 1075 -- 1085. [3] A. Bak, N.A. Vavilov, Normality for elementary subgroup functors, Math. Proc. Camb. Philos. Soc. 118(1) (1995), 35 -- 47. [4] H. Bass, Algebraic K-theory. Benjamin, New York, 1968. 10 [5] H. Bass, K-theory and stable algebra. Inst. Hautes Etudes Sci., Publ. Math. 22 (1964), 5 -- 60. 1 [6] Z. Borevic, N.A. Vavilov, The distribution of subgroups in the full linear group over a commutative ring, Proc. Steklov Institute Math 3 (1985), 27 -- 46. 1 [7] V.N. Gerasimov, The group of units of the free product of rings. Mat. Sbornik 134 (1987)(1), 42 -- 65. [8] I.Z. Golubchik, On the general linear group over weakly Noetherian associative rings. Fundam. Appl. Math. 1 (1995)(3), 661 -- 668. [9] I.Z. Golubchik, A. V. Mikhalev, On the group of elementary matrices over PI-rings. in Investigations in Algebra (Iad. Tbil. Gos. Univ., Tbilisi, 1985), 20 -- 24. [10] S.G. Khlebutin, Some properties of the elementary subgroup. in Algebra, Logic, and Number Theory (Izd. Mosk. Gos. Univ., Moscow, 1986), 86 -- 90. [11] A.J. Hahn and O.T. O'Meara. The Classical groups and K-Theory, Springer, 1989. 10 [12] R. Hazrat, N. Vavilov, Bak's work on the K-theory of rings, with an appendix by Max Karoubi, J. K-Theory 4 (2009), 1 -- 65. 2 [13] R. Hazrat, Z. Zhang, Generalized commutator formulas, Comm. in Algebra, 39 (2011), 1441 -- 1454. 2, 4, 11 [14] R. Hazrat, N.A. Vavilov, Z. Zhang, Relative unitary commutator calculus and applications, J. Algebra, to appear, arXiv:0911.5510 2 [15] R. Hazrat, A.V. Stepanov, N.A. Vavilov, Z. Zhang, The yoga of commutators, J. Math. Sci., to appear. 2 [16] A.W. Mason, On subgroup of GLn(n, A) which are generated by commutators, II. J. reine angew. Math. 322 (1981), 118 -- 135. 1 [17] A.W. Mason, A further note on subgroups of GLn(n, A) which are generated by commutators. Arch. Math. 37 (1981)(5) 401 -- 405. [18] A.W. Mason, W.W. Stothers, On subgroup of GLn(n, A) which are gnerated by commutators. Invent. Math. 23 (1974), 327 -- 346. 1 [19] A.V. Stepanov, On the normal structure of the general linear group over a ring, Zap. Nauch. Sem. POMI 236 (1997), 162 -- 169. [20] A.V. Stepanov, N.A. Vavilov, Decomposition of transvections: A theme with variations. K-theory 19 (2000), 109 -- 153. [21] A.A. Suslin, On the structure of the special linear group over the ring of polynomials, Izv. Akad. Nauk SSSR, Ser. Mat. 141 (1977)(2), 235 -- 253. 1 [22] J. Tits, Syst`emes g´en´erateurs de groupes de congruence, C. R. Acad. Sci. Paris, S´er A, 283 (1976), 693 -- 695. 5 [23] M.S. Tulenbaev, The Schur multiplier of the group of elementary matrices of finite order. Zap. Nauch. Sem LOMI 86 (1979), 162 -- 169. 1 [24] N.A. Vavilov, A.V. Stepanov, Standard commutator formula. Vestnik St. Petersburg State Univ., ser.1 41 No. 1(2008), 5 -- 8. 1, 2 [25] N.A. Vavilov, A.V. Stepanov, Standard commutator formula, revisited. Vestnik St. Petersburg State Univ., ser.1, 43 No. 1 (2010), 12 -- 17. 2 [26] L.N. Vaserstein, On the normal subgroups of GLn over a ring. Lecture Notes in Math. 854 (1981), 456 -- 465. 1 Department of Pure Mathematics, Queen's University Belfast, Belfast BT7 1NN, Northern Ireland, United Kingdom E-mail address: [email protected] Department of Mathematics, Beijing Institute of Technology, Beijing, China E-mail address: [email protected]
1504.04431	2	1504	2015-05-07T17:42:52	Classification of Real Solvable Lie Algebras Whose Simply Connected Lie Groups Have Only Zero or Maximal Dimensional Coadjoint Orbits	[ "math.RA" ]	In this paper we study a special subclass of real solvable Lie algebras having small dimensional or small codimensional derived ideal. It is well-known that the derived ideal of any Heisenberg Lie algebra is 1-dimensional and the derived ideal of the 4-dimensional real Diamond algebra is 1-codimensional. Moreover, all the coadjoint orbits of any Heisenberg Lie group as well as 4-dimensional real Diamond group are orbits of dimension zero or maximal dimension. In general, a (finite dimensional) real solvable Lie group is called an $MD$-group if its coadjoint orbits are zero-dimensional or maximal dimensional. The Lie algebra of an $MD$-group is called an $MD$-algebra and the class of all $MD$-algebras is called $MD$-class. Simulating the mentioned above characteristic of Heisenberg Lie algebras and 4-dimensional real Diamond algebra, we give a complete classification of $MD$-algebras having 1-dimensional or 1-codimensional derived ideals.	math.RA	math	CLASSIFICATION OF REAL SOLVABLE LIE ALGEBRAS WHOSE SIMPLY CONNECTED LIE GROUPS HAVE ONLY ZERO OR MAXIMAL DIMENSIONAL COADJOINT ORBITS ANH VU LE, VAN HIEU HA, ANH TUAN NGUYEN, TRAN TU HAI CAO, AND THI MONG TUYEN NGUYEN ABSTRACT. In this paper we study a special subclass of real solvable Lie alge- bras having small dimensional or small codimensional derived ideal. It is well- known that the derived ideal of any Heisenberg Lie algebra is 1-dimensional and the derived ideal of the 4-dimensional real Diamond algebra is 1-codimensional. Moreover, all the coadjoint orbits of any Heisenberg Lie group as well as 4- dimensional real Diamond group are orbits of dimension zero or maximal di- mension. In general, a (finite dimensional) real solvable Lie group is called an M D-group if its coadjoint orbits are zero-dimensional or maximal dimensional. The Lie algebra of an M D-group is called an M D-algebra and the class of all M D-algebras is called M D-class. Simulating the mentioned above char- acteristic of Heisenberg Lie algebras and 4-dimensional real Diamond algebra, we give a complete classification of M D-algebras having 1-dimensional or 1- codimensional derived ideals. 1. CLASSIFICATION OF SOLVABLE LIE ALGEBRAS: A QUICK INTRODUCTION Classifying all Lie algebras of dimension less than 4 is an elementary exercise. However, when considering Lie algebras of dimension n (n > 4), complete classi- fications are much harder. As it has long been well known, there exist three differ- ent types of Lie algebras: the semisimple, the solvable and those which are neither semi-simple nor solvable. By the Levi-Maltsev Theorem [Mal45] in 1945, any finite-dimensional Lie algebra over a field of characteristic zero can be expressed as a semidirect sum of a semi-simple subalgebra and its maximal solvable ideal. It reduces the task of classifying all finite-dimensional Lie algebras to obtaining the classification of semi-simple and of solvable Lie algebras. The problem of the classification of semi-simple Lie algebras over the complex field has been completely classified by Killing, E. Cartan [Car94] in 1894, over the real field by F. R. Gantmakher [Gan39] in 1939. Although several classifications of solvable Lie algebras of small dimension are known, but the problem of the complete classification of the (real or complex) solvable Lie algebras is still open up to now. There are two ways of proceeding in the classification of solvable Lie algebras: by dimension or by structure. Date: October 10, 2018. 2000 Mathematics Subject Classification. Primary 17B, 22E60, Secondary 20G05. Key words and phrases. K-orbit, M D-algebra, M D(∗, 1)-algebra, M D(∗, ∗ − 1)-algebra. 1 2 ANH VU LE ET AL. First, we list some results about the classification of solvable Lie algebras in the dimensional approach. • All solvable Lie algebras up to dimension 6 over the complex field C and the real field R were classified by G. M. Mubarakzyanov [Mub63] in 1963 and by P. Turkowski [Tur90] in 1990. • All solvable Lie algebras up to dimension 4 over any perfect field were classified by J. Patera and H. Zassenhaus [PZ90] in 1990. • Some incomplete classifications of solvable Lie algebras in dimension 7 and nilpotent algebras up to dimension 8 were given by G. Tsagas [Tsa99] in 1999. It seems to be very difficult to proceed by dimension in the classification of Lie algebras of dimension greater than 6. However, it is possible to proceed by structure, i.e. to classify solvable Lie algebras with a specific given property. Now, we list some results about the classification of solvable Lie algebras in the structural approach. • In 1973, M. A. Gauger [Gau73] gave a complete classification of metabelian Lie algebras of dimension no more than 7 and nearly complete results for dimension 8. • In 1995, D. Arnal, M. Cahen and J. Ludwig [ACL95] gave the list of all solvable Lie algebras such that the coadjoint orbits of the connected Lie groups corresponding to them are of dimension zero or two. But they have not classified them yet, up to isomorphism. • In 1999, L. Yu. Galitski and D. A. Timashev [GT99] completely classified all of metabelian Lie algebras of dimension 9. • In 2007, R. Campoamor-Stursberg [CS07] gave a complete classification of nine-dimensional Lie algebras with nontrivial Levi decomposition. • In 2007, I. Kath [Kat07] classified the class of nilpotent quadratic Lie al- gebras of dimension no more than 10. • In 2010, another class of Lie algebras relating to the nilradicals has been being classified by L. Snobl [Sno10] • In 2012, M. T. Duong, G. Pinczon, and R. Ushirobira [DPU12] gave a classification of Solvable singular quadratic Lie algebras. • In 2012, L. Chen [Che12] classified a class of solvable Lie algebras with triangular decompositions. In an attempt to classify solvable Lie algebras by structure, we study in this paper a special subclass of real solvable Lie algebras having small dimensional or small codimensional derived ideals. This idea comes from an investigation of Kirillov's Orbit Method on the (2m + 1)-dimensional Heisenberg Lie algebras (0 < m ∈ N) and the 4-dimensional real Diamond Lie algebra. Recall that, in 1962, A. A. Kirillov [Kir76] introduced the Orbit Method which quickly became the most important method in the theory of representations of Lie groups and Lie algebras. The key of Kirillov's Orbit Method is the coadjoint orbits or K-orbits (i.e., orbits in the coadjoint representation) of Lie groups. We emphasize that any K-orbit of the (2m + 1)-dimensional Heisenberg Lie group and the 4-dimensional LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 3 real Diamond Lie group has dimension zero or maximal. Hence, it is reasonable to consider the class of solvable Lie groups (and corresponding algebras) having the similar property. A (finite dimensional) real solvable Lie group is called an M D-group (in term of N. D. Do [Do99]) if its K-orbits are orbits of dimension zero or maximal dimension. The Lie algebra of an M D-group is called an M D- algebra and the class of all M D-algebras is called M D-class. In particular, if the maximal dimension of the K-orbits of some M D-group G is equal to dim G then G is called an SM D-group and its algebra is called an SM D-algebra. The class of all SM D-algebras is called SM D-class. It is clear that SM D-class is a subclass of M D-class. The investigation of M D-class was first time suggested by N. D. Do [Do99] in 1982. Now, we list main results about M D-class. • In 1984, H. V. Ho [VH84] completely classified all of SM D-algebras (of arbitrary dimension). • In 1990, A. V. Le [Le90, Le90b, Le93] gave a complete classification of all 4-dimensional M D-algebras. • In 1995, D. Arnal, M. Cahen and J. Ludwig [ACL95] gave the list of all M D-algebras such that the maximal dimension of K-orbits of correspond- ing M D-groups is just 2, but they have not yet classified them up to iso- morphism. • Up to 2012, A. V. Le et al. [LS08, LHT11] had classified (up to isomor- phism) all of M D-algebras of dimension 5. • In 2013, the M D-class was listed as a specific attention in classification of Lie Algebras by L. Boza, E. M. Fedriani, J. Nunez and A. F. Tenorio [BFNT13]. The investigation of general properties of M D-class, in particular, the complete classification of M D-class is still open up to now. As we say above, the (2m + 1)-dimensional real Heisenberg Lie algebra and the 4-dimensional real Diamond Lie algebra are M D-algebras. The real Lie Heisen- berg algebras and their extensions are investigated by a lot of mathematicians be- cause of their physical origin and applications. Moreover, the first derived ideal of the Heisenberg Lie algebra is 1-dimensional and the first derived ideal of the 4-dimensional Diamond Lie algebra is 1-codimensional. We will generalize these properties to consider M D-algebras having the first derived ideal of dimension 1 or codimension 1. For convenience, we shall denote by M D(∗, 1) or M D(∗, ∗ − 1) the subclasses of M D-algebras having 1-dimensional or 1-codimensional de- rived ideals, respectively. If G belongs to M D(∗, 1) or M D(∗, ∗ − 1) then it is called an M D(∗, 1)-algebra or M D(∗, ∗ − 1)-algebra, respectively. In particu- lar, every M D(∗, 1)-algebra or M D(∗, ∗ − 1)-algebra of dimension n is called an M D(n, 1)-algebra or M D(n, n − 1)-algebra, respectively. Of course, the (2m + 1)-dimensional Heisenberg Lie algebra belongs to M D(2m + 1, 1) and the 4-dimensional real Diamond Lie algebra belongs to M D(4, 3). The main pur- pose of this paper is to completely classify, up to isomorphism, M D(∗, 1)-class and M D(∗, ∗ − 1)-class. We also prove that any real solvable Lie algebra having 4 ANH VU LE ET AL. 1-dimensional derived ideal belongs to M D(∗, 1) and give a sufficient and nec- essary condition in order that a n-dimensional real solvable Lie algebra having 1-codimensional derived ideal belongs to M D(n, n − 1) with n > 4. The next part of the paper will be organized as follows: Section 2 gives some ba- sic concepts, especially we recall the Lie algebra of the group of the affine transfor- mations of real straight line, the real Heisenberg Lie algebras and the real Diamond Lie algebras. Section 3 deals with some well-known remarkable classifications of some subclasses of M D-algebras. The main results about the complete classifica- tions of M D(∗, 1)-class and M D(∗, ∗ − 1)-class, are given in Section 4. The last section is devoted the discussion of some open problems. 2. SOME BASIC CONCEPTS We first recall in this section some preliminary results and notations which will be used later. For details we refer the reader to the book [Kir76] of A. A. Kirillov and the book [Do99] of N. D. Do. 2.1. The coadjoint representation and coadjoint orbits. Let G be a Lie group, G = Lie(G) be the corresponding Lie algebra of G and G ∗ be the dual space of G. For every g ∈ G, we denote the internal automorphism associated with g by A(g), and hence, A(g) : G → G can be defined as follows A(g) := g.x.g−1, ∀x ∈ G. This automorphism induces the following map A(g)∗ : G → G which is defined as follows A(g)∗ (X) := d This map is called tangent map of A(g). Definition 2.1. The action dt(cid:2)g. exp (tX) .g−1(cid:3) t=0 ; ∀X ∈ G. Ad : G → Aut (G) g 7→ A(g)∗ is called the adjoint representation of G in G. The coadjoint representation is the dual of the adjoint representation. Namely, we have the following definition. Definition 2.2. The coadjoint representation or K-representation K : G → Aut (G ∗) g 7→ K(g) of G in G ∗ is defined by (cid:10)K(g)F, X(cid:11) :=(cid:10)F, Ad(cid:0)g−1(cid:1) X(cid:11) ; (F ∈ G ∗, X ∈ G) , where hF, Y i denotes the value of a linear functional F on an arbitrary vector Y ∈ G. A geometrical interpretation of the coadjoint representation of G is as the action by left-translation on the space of right-invariant 1-form on G. LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 5 Definition 2.3. Each orbit of the coadjoint representation of G is called a K-orbit of G. We denote the K-orbit containing F by ΩF . For every F ∈ G ∗, the K-orbit containing F can be defined by ΩF := (cid:8)K(g)F g ∈ G(cid:9). The dimension of every K-orbit of an arbitrary Lie group G is always even. In order to define the dimension of the K-orbits ΩF for each F from the dual space G ∗ of the Lie algebra G = Lie(G), it is useful to consider the following (skew-symmetric bilinear) Kirillov form BF on G corresponding to F : BF (X, Y ) = hF, [X, Y ]i for all X, Y ∈ G. Denote the stabilizer of F under the co-adjoint representation of G in G ∗ by GF and GF := Lie(GF ). We shall need in the sequel of the following result. Proposition 2.4 (see [Kir76, Section 15.1]). ker BF = GF and dim ΩF = dim G − dim GF = rankBF . 2.2. M D-groups and M D-algebras and some their properties. Definition 2.5. An n-dimensional M D-group or, for brevity, an M Dn-group is an n-dimensional real solvable Lie group such that its K-orbits are orbits of di- mension zero or maximal dimension. The Lie algebra of an M Dn-group is called an M Dn-algebra. M D-class and M Dn-class are the sets of all M D-algebras (of arbitrary dimension) and M Dn-algebras, respectively. Definition 2.6. An M D(n, m)-algebra is an M Dn-algebra whose the first derived ideal is m-dimensional with m, n ∈ N and 0 < m < n. M D(n, m)-class is the set of all M D(n, m)-algebras. In particular, M D(∗, 1)-class and M D(∗, ∗ − 1)-class are the sets of all M D-algebras (of arbitrary dimension) having the first derived ideal of dimension 1 and codimension 1, respectively. Remark 2.7. Note that all the Lie algebras of dimension n (n 6 3) are M D- algebras, and moreover they can be listed easily. So we only take interest in M Dn- algebras for n > 4. For any real Lie algebra G, as usual, we denote the first and second derived ideals of G by G1 := [G, G] and G2 := [G1, G1], respectively. Now, we introduce some well-known properties of M D-algebras. First, the following proposition gives a necessary condition for a Lie algebra belonging to M D-class. Proposition 2.8 (see [VH84, Theorem 4]). Let G be an M D-algebra. Then its second derived ideal G2 is commutative. We point out here that the converse of the above result is in general not true. In other words, the above necessary condition is not a sufficient condition. Proposition 2.9 (see [Do99, Chapter 2, Proposition 2.1]). Let G be an M D- there exists U ∈ G1 algebra. such that hF, U i 6= 0, then the K-orbit ΩF has maximal dimension. If F ∈ G ∗ is not vanishing perfectly in G1, i.e. 6 ANH VU LE ET AL. Proposition 2.10 (see [LHT11]). There is no M D-algebra G such that its second derived ideal G2 is not trivial and dim G2 = dim G1 − 1. In other words, if 0 < dim G2 = dim G1 − 1 then G is not an M D-algebra. To illustrate and show the role of the M D-class, in the rest of this section, we will introduce some typical examples and counter-examples of M D-algebras. 2.3. The Lie algebra of the group of affine transformations of real straight line. The Lie algebra aff(R) of the group Aff(R) of affine transformations of real straight line R is the unique non-commutative real Lie algebra of dimension 2 and it is defined as follows: aff(R) := Span(X, Y ); [X, Y ] = Y. Remark 2.11. Clearly, every real Lie algebra of dimension n 6 3 is an M D- algebra. In particular aff(R) is an M D(2, 1)-algebra. 2.4. The real Heisenberg Lie algebras. The (2m + 1)-dimensional real Heisen- berg Lie algebra (0 < m ∈ N) is the following real Lie algebra: h2m+1 := Span (Xi, Yi, Z i = 1, 2, . . . , m) ; [Xi, Yi] = Z; i = 1, 2, . . . , m; the other Lie brackets are trivial. Remark 2.12. The first derived ideal h1 dimensional and coincides with the center of h2m+1. 2m+1 = R.Z = Span(Z) of h2m+1 is 1- Let (X ∗ 1 , Y ∗ 1 , . . . , X ∗ m, Y ∗ m, Z ∗) be the dual basis of (X1, Y1, . . . , Xm, Ym, Z) in the dual space h∗ 2m+1 of h2m+1, and 1 + . . . + amX ∗ 1 + b1Y ∗ F = a1X ∗ m + bmY ∗ m + cZ ≡ (a1, b1, . . . , am, bm, c) be an arbitrary element in h∗ following matrix 2m+1. Then the Kirillov form BF is given by the BF = = diag (Λ, · · · , Λ, 0)   0 0 0 0 0 0 · · · 0 c 0 0 0 · · · −c 0 0 0 0 0 · · · c 0 0 0 −c 0 · · · 0 ... ... ... ... ... .. . 0 · · · 0 0 c 0 0 · · · −c 0 0 0 0 0 0 0 · · · 0 0 0 0 0 0 ... 0 ... 0 0 0 0   with m blocks Λ =(cid:20) 0 −c 0(cid:21). c In view of Proposition 2.4, it is a simple matter to get the following proposition. Proposition 2.13. h2m+1 is one M D(2m + 1, 1)-algebra and the maximal dimen- 2m+1 is 2m. Moreover, for F = (a1, b1, . . . , am, bm, c) ∈ sion of K-orbits in h∗ 2m+1, we have h∗ (i) K-orbits containing F is of dimension 0 if and only if c = 0. (ii) K-orbits containing F is of dimension 2m if and only if c 6= 0. LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 7 2.5. The real Diamond Lie algebras. The (2m + 2)-dimensional real Diamond Lie algebra (0 < m ∈ N) is one semi-direct extension of the (2m+1)-dimensional Heisenberg algebra by R, namely it is the following real Lie algebra: R.h2m+1 := Span(Xi, Yi, Z, T i = 1, 2, . . . , m) where the Lie structure is given by [Xi, Yi] = Z, [T, Xi] = −Xi, [T, Yi] = Yi; i = 1, 2, . . . , m; the other Lie brackets are trivial. Remark 2.14. The (2m+1)-dimensional real Heisenberg algebra is the first derived ideal of the (2m + 2)-dimensional real Diamond Lie algebra. In particular, the first derived ideal of R.h2m+1 is of codimension 1. Let (X ∗ 1 , Y ∗ m, Y ∗ 1 , . . . , X ∗ m, Z ∗, T ∗) be the dual basis of (X1, Y1, . . . , Xm, Ym, 1 + . . . + m + cZ + dT ≡ (a1, b1, . . . , am, bm, c, d) be an arbitrary element Z, T ) in the dual space (R.h2m+1)∗ of R.h2m+1 and F = a1X ∗ +amX ∗ in (R.h2m+1)∗. Then we get the Kirillov form BF as follows m + bmY ∗ 1 + b1Y ∗ 0 −c 0 0 ... 0 0 0 −a1 0 c 0 0 0 0 0 −c ... ... 0 0 0 0 0 0 b1 −a2 0 0 c 0 ... 0 0 0 b2 0 0 · · · a1 0 −b1 0 · · · 0 0 · · · a2 0 −b2 0 · · · ... ... ... . . . 0 · · · 0 am 0 −bm · · · −c · · · 0 0 · · · −am bm 0 0 0 0 0 ... c 0 0 0 0   .   BF = By virtue of Proposition 2.4, one can verify the following proposition. Proposition 2.15. The (2m + 2)-dimensional real Diamond Lie algebra R.h2m+1 is an M D(2m + 2, 2m + 1)-algebra if and only if m = 1. That means the 4- dimensional real Diamond Lie algebra is an M D(4, 3)-algebra and the (2m + 2)- dimensional real Diamond Lie algebra is not an M D-algebra for every natural number m > 1. 3. SOME SUBCLASSES OF M D-CLASS In this section, we would like to introduce some well-known remarkable results of classification of M D-class. First, recall that all of the M D-algebras of dimen- sion 4 or 5 were classified, up to isomorphism, by A. V. Le et al. [Le90, Le93, LS08]. However, to illustrate the general results which will be given in the last section of the paper, we will introduce here the classification of M D(n, 1)-class and M D(n, n − 1)-class for small n, namely n = 4 or n = 5. 8 ANH VU LE ET AL. 3.1. Classification of M D(4, 1)-class and M D(4, 3)-class. Proposition 3.1 (Classification of M D(4, 1)-algebras, see [Le93]). Let G be an M D(4, 1)-algebra. Then G is decomposable and we can choose a suitable basis (X, Y, Z, T ) of G such that G1 = Span(Z) = R.Z, and G is isomorphic to one of the following Lie algebras. 1.1. G4,1,1 := h3 ⊕ R.T, [X, Y ] = Z; the others Lie brackets are trivial. 1.2. G4,1,2 := aff(R) ⊕ R.Z ⊕ R.T, [X, Y ] = Y ; the others Lie brackets are trivial. Proposition 3.2 (Classification of MD (4 , 3 )-algebras, see [Le93]). Let G be an M D(4, 3)-algebra. Then G must be indecomposable and we can choose a suit- able basis (X, Y, Z, T ) of G such that G is isomorphic to one of the following Lie algebras. λ1 0 0 0 0 λ2 0 0 λ 1 0 0 λ 0 1 1 0 0 1 1 1. G1 = Span(X, Y, Z) ≡ R3, adT ∈ AutR(cid:0)G1(cid:1) ≡ GL3(R) 1.1. G4,3,1(λ1,λ2) : adT = 1  ; λ1, λ2 ∈ R \ {0}.  1.2. G4,3,2(λ) : adT = 0 0 1  ; λ ∈ R \ {0}.  1.3. G4,3,3 : adT = 0 0 1  .  1.4. G4,3,4(λ,ϕ) : adX1 =  2. G1 = Span(X, Y, Z) = h3, adT ∈ EndR(cid:0)G1(cid:1) ≡ M at3(R) 2.1. G4,4,1 : adT =  adT =  0 0  . 0 0  . 2.2. G4,4,2 = R.h3 (the 4-dimensional Diamond Lie algebra) : cos ϕ − sin ϕ 0 0 sin ϕ cos ϕ 0 1 0 −1 0 0 0 −1 0 0 0 1 0 0 0 0 λ  ; λ ∈ R\{0}, ϕ ∈ (0, π). 3.2. Classification of M D(5, 1)-class and M D(5, 4)-class. Proposition 3.3 (Classification of M D(5, 1)-algebras, see [LS08, LHT11]). Let G be an M D(5, 1)-algebra, Then we can choose a suitable basis (X1, X2, X3, X4, X5) of G such that G1 = Span(X5) = R.X5 and G is isomorphic to one of the following Lie algebras. 1. G5,1,1 = h5 (the 5-dimensional real Heisenberg Lie algebra): [X1, X2] = [X3, X4] = X5; the other Lie brackets are trivial. In this case, G is inde- composable. LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 9 2. G5,1,2 = aff(R) ⊕ R.X1 ⊕ R.X2 ⊕ R.X3 : [X4, X5] = X5; the other Lie brackets are trivial. In this case, G is decomposable. Proposition 3.4 (Classification of M D(5, 4)-algebras, see [LS08]). Let G be an M D(5, 4)-algebra. Then G must be indecomposable, and G1 is commutative. Moreover, we can choose a suitable basis (X1, X2, X3, X4, X5) of G such that G1 = Span(X2, X3, X4, X5) ≡ R4, adX1 ∈ Aut(G1) ≡ GL4(R) and G is iso- morphic to one of the following Lie algebras. λ1 0 0 0 0 λ2 0 0 0 0 0 0 λ3 0 0 1 ; λ1, λ2, λ3 ∈ R \ {0, 1};   ; λ1, λ2 ∈ R \ {0, 1}, λ1 6= λ2. 0 0 0 λ2 0 0 1 0 0 0 0 1   ; λ ∈ R \ {0, 1}. ; λ ∈ R \ {0, 1}. 4.1. G5,4,1(λ1,λ2,λ3) : adX1 =   λ1 6= λ2 6= λ3 6= λ1. 4.2. G5,4,2(λ1,λ2) : adX1 =  λ1 0 0 0 λ 0 0 0 0 λ 0 0 0 0 1 0 0 0 0 1 λ 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1     4.3. G5,4,3(λ) : adX1 =  4.4. G5,4,4(λ) : adX1 =  4.5. G5,4,5 : adX1 =  .   1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 λ2 0 0 0 1 1 0 1 0 λ1 0 0 0   4.6. G5,4,6(λ1,λ2) : adX1 =  4.7. G5,4,7(λ) : adX1 =  4.8. G5,4,8(λ) : adX1 =  λ 0 0 0 0 λ 0 0 0 0 1 1 0 0 0 1 λ 1 0 0 0 λ 0 0 0 0 1 1 0 0 0 1     ; λ1, λ2 ∈ R \ {0, 1}, λ1 6= λ2. ; λ ∈ R \ {0, 1}. ; λ ∈ R \ {0, 1}. 0 0 4.12. G5,4,12(λ,ϕ) : adX1 =  adX1 =  adX1 =  4.13. G5,4,13(λ,ϕ) : 0 0 4.14. G5,4,14(λ,µ,ϕ) : cos ϕ cos ϕ − sin ϕ 0 0 sin ϕ 0 0 λ 0 0 λ 0 0 0 0 cos ϕ cos ϕ − sin ϕ 0 0 0 0 sin ϕ λ 1 0 λ 0 0       ; λ ∈ R \ {0}, ϕ ∈ (0, π). ; λ ∈ R \ {0}, ϕ ∈ (0, π). 10 ANH VU LE ET AL. ; λ ∈ R \ {0, 1}. 4.9. G5,4,9(λ) : adX1 =  4.10. G5,4,10 : adX1 =  4.11. G5,4,11(λ1,λ2,ϕ) : λ 0 0 0 0 1 1 0 0 0 1 1 0 0 0 1 1 1 0 0 0 1 1 0 0 0 1 1 0 0 0 1     . cos ϕ cos ϕ − sin ϕ 0 0 sin ϕ λ1 0 0 0 ; λ1, λ2 ∈ R \ {0}, λ1 6= λ2, ϕ ∈ (0, π). 0 0 0 λ2   adX1 =  cos ϕ − sin ϕ 0 0 sin ϕ 0 0 λ −µ µ λ cos ϕ 0 0 0 0 ; λ, µ ∈ R, µ > 0, ϕ ∈ (0, π). In the next subsection, we introduce one noticeable result of D. Arnal, M. Cahen and J. Ludwig [ACL95] in 1995. 3.3. List of M D-algebras whose simply connected M D-groups have only coad- joint orbits of dimension zero or two. In an attempt to classify solvable Lie al- gebras by structure, in 1995, D. Arnal, M. Cahen and J. Ludwig [ACL95] have listed, up to a direct central factor, all Lie algebras (solvable or not) such that the maximal dimension of K-orbits of corresponding connected and simply connected Lie groups is just two. However, they have not yet classified, up to isomorphism, these algebras. Proposition 3.5 (see [ACL95]). Let G be a connected, simply connected solvable Lie group whose coadjoint orbits have dimension smaller or equal to two. Let G be the Lie algebra of G. Then, up to a direct central factor, G belongs to the following list of algebras: LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 11 (i) R.T ⊕ a where a is an abelian ideal and adT ∈ End(a). (ii) R.T ⊕ h3 where h3 is the 3-dimensional Heisenberg algebra spanned by (X, Y, Z) with [X, Y ] = Z and • either [T, X] = X, [T, Y ] = −Y, [T, Z] = 0 (the 4-dimensional Diamond algebra). • or [T, X] = Y, [T, Y ] = −X, [T, Z] = 0. (iii) G is 5-dimensional with basis (X1, X2, X3, Y1, Y2) and the multiplicative law reads [X1, X2] = X3, [X1, X3] = Y1, [X2, X3] = Y2. (iv) G is 6-dimensional with basis (X1, X2, X3, Y1, Y2, Y3) and the nonvanish- ing brackets are [X1, X2] = Y3, [X2, X3] = Y1, [X3, X1] = Y2. Remark 3.6. Clearly we have two following remarks. (i) There is an infinite family of non-isomorphic M D-algebras in part (i) of Proposition 3.5. Namely, the part (i) of Proposition 3.5 includes all M D(4, 1)-algebras, M D(5, 1)-algebras, M D(5, 4)-algebras except the 5- dimensional Heisenberg Lie algebra. Furthermore, the last two M D(4, 3)- algebras in Proposition 3.2 coincide with two algebras in the part (ii) of Proposition 3.5, but remaining four M D(4, 3)-algebras in Proposition 3.2 are included in the part (i) of Proposition 3.1. (ii) However, it should be noted that the indecomposable M D(2, 1)-algebras aff(R) and h3; the decomposable M D(4, 1)-algebras h3 ⊕ R.T (in Propo- sition 3.1), aff(R) ⊕ R.Z ⊕ R.T and the decomposable M D(5, 1)-algebra aff(R) ⊕ R.X1 ⊕ R.X2 ⊕ R.X3 are not included in the list of Proposition 3.5, although it is obvious that all coadjoint orbits of the Lie groups corre- sponding to h3 ⊕ R.T, aff(R)⊕ R.Z ⊕ R.T, aff(R)⊕ R.X1 ⊕ R.X2 ⊕ R.X3 have dimension zero or two. So that was one shortcoming in Arnal's list. 4. CLASSIFICATION OF M D(∗, 1)-CLASS AND M D(∗, ∗ − 1)-CLASS Now we will introduce the complete classification, up to an isomorphism, of all M D-algebras (of arbitrary dimension) having the first derived ideal of dimension one or codimension one. These results are generalizations of Propositions 3.1, 3.2, 3.3 and 3.4 in Section 3. 4.1. The main results. Theorem 4.1 (The Complete Classification of M D(∗, 1)-class). M D(∗, 1)-class coincides with the class of all real solvable Lie algebras whose the first derived ideal is 1-dimensional, moreover M D(∗, 1) includes only the Lie algebra of the group of affine transformations of the real straight line, the real Heisenberg Lie algebras and their direct extensions by the real commutative Lie algebras. In other words, if G is a n-dimensional real solvable Lie algebra whose the first ideal G1 := [G, G] is 1-dimensional (2 6 n ∈ N) then G is an M D(n, 1)-algebra and G is isomorphic to one and only one of the following Lie algebras. 12 ANH VU LE ET AL. (i) The Lie algebra aff(R) of the group Aff(R) of all affine transformations on R; n = 2. (ii) aff(R) ⊕ Rn−2; 3 6 n. (iii) The real Heisenberg Lie algebra h2m+1; 3 6 n = 2m + 1. (iv) h2m+1 ⊕ Rn−2m−1; 3 6 2m + 1 < n. It is clear that Theorem 4.1 can be formulated by another way in the following consequence which gives a new character of the real Heisenberg Lie algebras. Corollary 4.2 (A New Character of the Real Heisenberg Lie Algebras). Let G be a real Lie algebra of dimension n (3 6 n ∈ N). Then the following conditions are equivalent. (i) G is indecomposable and has the first derived ideal G1 = [G, G] ∼= R. (ii) G is an indecomposable M D(n, 1)-algebra. (iii) G is the n-dimensional Heisenberg Lie algebra (in particular, n is odd). The next theorem gives one necessary and sufficient condition to recognize one M D(n, n − 1)-algebra (4 6 n ∈ N). Theorem 4.3 (Necessary and Sufficient Conditions to identify M D(∗, ∗−1)-alge bras). Let G be a real solvable Lie algebra of dimension n (3 6 n ∈ N) such that its first derived ideal G1 is (n − 1)-dimensional. (i) If G1 is commutative then G is an M D(n, n − 1)-algebra, moreover G is indecomposable. (ii) If n > 4 and G is an M D(n, n − 1)-algebra then G1 is commutative. Remark 4.4. When n 6 4, assertion (ii) is not true. Namely, if n < 4, all the n-dimensional Lie algebras are M D-algebras, and moreover, they can be listed easily. If n = 4, as previously indicated, the derived ideal of the 4-dimensional real Diamond Lie algebra is the 3-dimensional Heisenberg algebra which is non commutative and 1-codimensional. In fact, all M D4-algebras were completely classified in 1990 by A. V. Le [Le90b, Le93], and the classification of M D(4, 1)- class and M D(4, 3)-class were recalled in Propositions 3.1 and 3.2. The last theorem will characterize every M D(n, n − 1)-algebra by an invertible real (n − 1)-square matrix and reduces the task of classifying M D(n, n − 1)- class to obtaining the well-known classification of equivalent of proportional sim- ilar matrices. Let G be an M D(n, n − 1)-algebra (3 6 n ∈ N) generated by a basis (X1, X2, . . . , Xn) such that the first derived ideal G1 is 1-codimensional and spanned by (X1, X2, . . . , Xn−1). It is obviously that the Lie structure of G is well understood by the invertible real (n − 1)-square matrix of map adXn considering as an automorphism of G1 for the basis (X1, X2, . . . , Xn−1). Theorem 4.5. Let G be a real vector space of dimension n (3 6 n ∈ N) gen- erated by a basis (X1, X2, . . . , Xn) and G1 := Span(X1, X2, . . . , Xn−1) is the 1-codimensional subspace of G. Then we have the following assertions. (i) Each invertible real (n − 1)-square matrix A always defines one Lie struc- ture on G such that G is an M D(n, n − 1)-algebra with the first derived LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 13 ideal is commutative, exactly equals to G1 and A is exactly the matrix of adjoint map adXn on G1 in the chosen basis (X1, X2, . . . , Xn−1). (ii) Two invertible real (n − 1)-square matrices A, B define two Lie structures on G which are isomorphic if and only if there exist a non-zero real number c and an invertible real (n − 1)-square matrix C so that cA = CBC −1. Remark 4.6. In view of Theorem 4.5, we have the following remarks. (i) Two invertible real square matrices A, B of the same order are called pro- portional similar if (and only if) there exist a non-zero real number c and an invertible real square matrix C of the same order as of A, B so that cA = CBC −1. In fact, assertion (ii) of Theorem 4.5 gives the classifica- tion of M D(n, n−1)-algebras (n > 4) by using the well-known classifica- tion of invertible real matrices in proportional similar equivalent relation. (ii) The classification of indecomposable M D(5, 4)-algebras in Proposition 3.4 of this paper gives one concrete illustration of Theorem 4.5 when n = 5. On principle, it is not hard to list all non-isomorphic indecom- posable M D(n, n − 1)-algebras by applying Theorem 4.5 for n is small, for example n = 6, 7, ... . 4.2. Proof of Theorem 4.1. In this section, we always consider G as a real solv- able Lie algebra of dimension n > 3 whose the derived ideal G1 = [G, G] is 1-dimensional, in particular G1 is commutative. Without loss of generality, we can choose a suitable basis such that G = Span (X1, X2, . . . , Xn) , G1 = Span (Xn) = R.Xn. Let G be the connected, simply connected Lie group corresponding to G. When n = 2, it is obvious that the part (i) in Theorem 4.1 holds because aff(R) is an M D(2, 1)-algebra (see Remark 2.11) and it is the unique non-commutative real Lie algebra of dimension 2. Therefore, in the rest of this subsection, we can suppose that n > 3. Denote [Xi, Xn] = aiXn; [Xi, Xj] = aij Xn (ai, aij ∈ R); i, j = 1, 2, . . . , n − 1. Evidently, the Lie structure on G is well understood by the vector a := (a1, a2, . . . , an−1) and the skew-symmetric real (n − 1)-square matrix A := (aij)i,j=1,n−1. There are two cases to consider for the values of the vector a: a = 0 or a 6= 0. The first case: a 6= 0. First, we consider the case a 6= 0, i.e. ∃i ∈ {1, . . . , n − 1} such that ai 6= 0, that means Xn is not in the center Z(G) of G. Renumber the chosen basis, if necessary, we can always suppose that an−1 6= 0. Then [Xn−1, Xn] = an−1Xn 6= 0. In this case, we will show that G is a trivial ex- tension of the Lie algebra aff(R). Namely, we have the following lemma. Lemma 4.7. If a 6= 0 then G is an M D(n, 1)-algebra which is isomorphic to aff(R) ⊕ Rn−2 when n > 3. 14 ANH VU LE ET AL. Proof. Using the following change of basis Yi = Xi − ai an−1 we get Xn−1, i = 1, 2, . . . , n − 2; Yn−1 = 1 an−1 Xn−1, Yn = Xn [Yi, Yn] = 0; i = 1, 2, . . . , n − 2; [Yn−1, Yn] = Yn. Hence, without loss of generality, we can now assume [Xi, Xn] = 0; i = 1, 2, . . . , n − 2; [Xn−1, Xn] = Xn. Using the Jacobi identities for triples (Xi, Xj , Xn−1), 1 6 i < j 6 n − 2, we get [[Xi, Xj], Xn−1] + [[Xn−1, Xi], Xj] + [[Xj , Xn−1], Xi] = 0 ⇒ aij[Xn, Xn−1] = 0 ⇒ −aij Xn = 0 ⇒ aij = 0 ⇒ [Xi, Xj] = aijXn = 0; i, j = 1, 2, . . . , n − 2. Now, using the change of basis as follows Zi = Xi + ai,n−1Xn; i = 1, 2, . . . , n − 2; Zn−1 = Xn−1, Zn = Xn we get [Zi, Zn−1] = 0; i = 1, 2, . . . , n − 2. So we can suppose now that Hence, in this case, G is isomorphic to the following Lie algebra [Xi, Xn−1] = 0, i = 1, 2, . . . , n − 2. aff(R) ⊕ Rn−2 = Span (X1, X2, . . . , Xn) , [Xn−1, Xn] = Xn, where the other Lie brackets are trivial. Obviously, the K-orbits of G is of dimen- sion 0 or 2. This means that G is an M D(n, 1)-algebra. (cid:3) The second case: a = 0. Now, we consider the second case a = 0, i.e. [Xi, Xn] = 0 for all i = 1, 2, . . . , n − 1, in particular Xn ∈ Z(G). Then the Lie structure of G is uniquely defined by the skew-symmetric real (n − 1)-square matrix A = (aij)i,j=1,n−1, which is called the structure matrix of G. Since G1 = Span(Xn) is 1-dimensional, A is non-trivial and 0 < rankA is even. We have the following lemma. Lemma 4.8. If a = 0, i.e. [Xi, Xn] = 0 for all i = 1, 2, . . . , n − 1, then the Lie algebra G is an M D(n, 1)-algebra and the maximal dimension of the K-orbits of G is the rank of the structure matrix A. Proof. Let G ∗ ≡ Rn be the dual space of G with dual basis (X ∗ F = f1X ∗ G ∗. The Kirillov form BF is given as follows n) and n ≡ (f1, f2, . . . , fn) be an arbitrary element of 2 + . . . + fnX ∗ 2 , . . . , X ∗ 1 + f2X ∗ 1 , X ∗ BF := (hF, [Xi, Xj]i)i,j=1,n = fn  a11 ... an−1,1 0 ... . . . a1,n−1 0 ... . . . . . . an−1,n−1 0 0 . . . 0   = fn(cid:20)A 0 0(cid:21) 0 and rankBF ∈ {0, 2k} where 2k = rankA is the rank of the structural matrix. More precisely • rankBF = 0 if and only if fn = 0, i.e. F = (f1, f2, . . . , fn−1, 0). LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 15 • rankBF = rankA = 2k > 0 if and only if fn 6= 0. Hence, in view of Proposition 2.4, G is an M D(n, 1)-algebra and the maximal dimension of K-orbits of G is the rank of the structure matrix. (cid:3) Now, we will consider whether G is decomposable in the second case. Lemma 4.9. If a = 0, i.e. [Xi, Xn] = 0 for all i = 1, 2, . . . , n − 1, then the Lie algebra G is decomposable if and only if the dimension of the center of G is greater than 1. Proof. Denote Z(G) to be the center of G. Obviously, Xn is in Z(G) because of [Xi, Xn] = 0 for all i = 1, 2, . . . , n − 1, i.e. dim Z(G) > 0. (=⇒) Suppose that G is decomposable, i.e. G = A ⊕ B in which A, B are non- trivial proper Lie subalgebras of G. Put Xn = Xa+Xb ∈ Z(G) with some Xa ∈ A and some Xb ∈ B. Let us consider an arbitrary element Y = Ya + Yb ∈ G with Ya ∈ A, Yb ∈ B. We have 0 = [Xn, Y ] = [Xa + Xb, Ya + Yb] = [Xa, Ya] + [Xb, Yb] ⇒ [Xa, Ya] = [Xb, Yb] = 0 ⇒ Xa, Xb ∈ Z(G). • If Xa 6= 0 6= Xb then they are of course linear independent and dim Z(G) > 1. • If Xa or Xb is 0. Without loss of generality, we can suppose that Xa = 0, i.e. Xn = Xb ∈ B. In particular G1 = Span(Xn) ⊆ B. Let X 6= 0 ∈ A be an arbitrary element. Obviously, [X, Z] = 0 for every Z ∈ B. On the other hand, we have [X, T ] ∈ A ∩ G1 ⊆ A ∩ B = 0 ⇒ [X, T ] = 0, ∀T ∈ A. This means that X commutes with any element of G = A ⊕ B, i.e. X ∈ Z(G). Because of X ∈ A, Xn ∈ B so X, Xn are linear independent and dim Z(G) > 1. Hence, dim Z(G) > 1 in any case. (⇐=) Suppose dim Z(G) > 1. There exists X ∈ Z(G) such that X, Xn are independent. We can add T1, . . . , Tn−2 in (X, Xn) to get a new basis of G. Then we have G = Span(X) ⊕ Span(Xn, T1, . . . , Tn−2). Therefore G is decomposable. (cid:3) Remark 4.10. The center of the Heisenberg Lie algebra is 1-dimensional, so its indecomposableness is unsurprised. Recall that each M D(n, 1)-algebra G in the second case is always defined uniquely by an (n − 1)-square (skew-symmetric) structure matrix A. Now we will consider whether two structure A and B give us isomorphic Lie algebras. Lemma 4.11. Let A = (aij)i,j=1,n−1, B = (bij)i,j=1,n−1 be skew-symmetric real (n − 1)-square matrices and GA, GB be M D(n, 1)-algebras which are defined by A, B respectively. Then (GA ∼= GB) ⇔ (∃c ∈ R∗, ∃C ∈ GLn−1(R) such that cA = C T BC), 16 ANH VU LE ET AL. where C T is the transpose of C. Proof. (=⇒) Let f : GA → GB be an isomorphism. Since f(cid:0)G1 B, there is a non-zero real number c so that f (Xn) = cXn. Clearly the matrix of f in the basis (X1, X2, . . . , Xn−1, Xn) is given as follows A(cid:1) = G1 M =  c11 ... cn−1,1 cn1 · · · · · · · · · · · · ... c1,n−1 0 ... cn−1,n−1 0 cn,n−1 c   =(cid:20) C 0 c (cid:21) ∗ in which C = (cij)i,j=1,n−1 is a real (n − 1)-square matrix, and ∗ is the vector (cn1, . . . , cn,n−1). Because f is an isomorphism, M is invertible and so is C. Hence, the linear map f is a Lie isomorphism if and only if f ([Xi, Xj ]A) = [f (Xi), f (Xj)]B; ∀ i, j = 1, 2, . . . , n − 1 n−1 n−1 ckiXk + cniXn, ⇔ caij Xn = ⇔ f (aijXn) =(cid:20)n−1 Pk=1 Pl=1 Pk=1 ⇔ caij Xn = n−1 cki.bkl.clj! Xn; Pk,l=1 cki. clj[Xk, Xl]B; n−1 Pk,l=1 ⇔ cA = C T AC. ⇔ caij = cki.bkl.clj; ∀ i, j = 1, 2, . . . , n − 1 n−1 Pl=1 cljXl + cnjXn(cid:21)B ; ∀ i, j = 1, 2, . . . , n − 1 ∀ i, j == 1, 2, . . . , n − 1 ∀ i, j = 1, 2, . . . , n − 1 (⇐=) Conversely, suppose that there exist a non-zero real number c and an invert- ible real (n − 1)-square matrix C = (cij )i,j=1,n−1 such that cA = C T BC. Let f : GA → GB be a linear map which is defined, in the basis (X1, X2, . . . , Xn), by the matrix M ′ as follows c11 ... cn−1,1 0 M ′ =  · · · · · · · · · · · · ... c1,n−1 0 ... cn−1,n−1 0 c 0   =(cid:20)C 0 c(cid:21) . 0 Since C is invertible and c 6= 0, f is a linear isomorphism. Moreover, it is easy to check that f is also a Lie homomorphism. Therefore, f is a Lie isomorphism. (cid:3) Remark 4.12. Recall that two real (n − 1)-square matrices A, B are said to be congruent if there exists an invertible (n − 1)-square matrix C such that B = C T AC. Furthermore, any non-zero skew-symmetric real square matrix can be always transformed into the canonical form. More precisely, for any non-zero skew-symmetric real (n − 1)-square matrix A, there exists a real orthogonal matrix C such that C T AC = diag(Λ1, Λ2, . . . , Λm, 0, . . . , 0) LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 17 where Λj := (cid:20) 0 −λi λi 0(cid:21) and {±iλ1, . . . , ±iλm} (i is the imaginary unit in the complex field C) is the set of all multiple eigenvalues of A. For example, the real Heisenberg Lie algebra h2m+1 := hXi, Yi, Z : i = 1, 2, . . . , mi; [Xi, Yi] = Z, i = 1, 2, . . . , m has the structure matrix H = diag(I, . . . , I) including n blocks I = (cid:20) 0 This matrix H has exactly two m-multiple eingenvalues ±i and H has no eigen- value 0. −1 0(cid:21). 1 Proof of Theorem 4.1. Recall that we need only show the part (ii), (iii), (iv) of Theorem 4.1. Lemmas 4.7, 4.8 and 4.9 show that the considered Lie algebra G belongs to M D(n, 1)-class. Moreover, the part (ii) of Theorem 4.1 is implied directly from Lemma 4.7. We only need to prove the part (iii) and (iv). In the basis (X1, X2, . . . , Xn), the structure matrix of G is A. We will choose a new basis to get the standard form B = C T AC. By Lemma 4.11, the Lie algebra defined by the matrix B is isomorphic to G. If B has no zero eigenvalue, i.e. B = diag(Λ1, Λ2, . . . , Λm), 2m = n − 1. Put D := diag(cid:18)1, 1 λ1 , 1, 1 λ2 , . . . , 1, 1 λm(cid:19) . Then we get the structure matrix H = DT BD of the (2m + 1)-dimensional real Heisenberg algebra h2m+1, 2m + 1 = n. By Lemma 4.11, G is isomorphic to h2m+1. So the part (iii) is proved. If B has eigenvalues 0, i.e. 0 < 2m < n − 1, then Z(G) is generated by the basis X2m+1, . . . , Xn whose dimension is greater than 1. By Lemma 4.9, G is decomposable, namely G is isomorphic to h2m+1⊕Rk where k = n−(2m+1) > 0. Actually, the direct summand Rk is the commutative Lie subalgebra of G generated by (X2m+2, . . . , Xn) and h2m+1 is generated by (X1, X2, . . . , X2m; X2m+1). So the part (iv) is proved and the proof of Theorem 4.1 is complete. (cid:3) 4.3. Proof of Theorem 4.3. In this section, we always consider G as a real solv- able Lie algebra of dimension n > 3 with the 1-codimensional first derived ideal G1 = [G, G]. Assume that dim G2 = dim([G1, G1]) = k 6 n − 2. Without loss of generality, we can choose a suitable basis such that G = Span (X1, X2, . . . , Xn) , G1 = [G, G] = Span (X1, X2, . . . , Xn−1) , G2 =(cid:2)G1, G1(cid:3) = Span (X1, . . . , Xk) , k 6 n − 2. ij (1 6 i < j 6 n, 1 6 l 6 n) be the structure constants of G. Then the Lie Let cl brackets of G are given by [Xi, Xj] = n−1 Xl=1 cl ij Xl; 1 6 i < j 6 n. 18 ANH VU LE ET AL. In view of the Proposition 2.8, if G is an M D-algebra then G2 is commutative and we get n−1 [Xi, Xj ] = cl ij Xl = 0 ⇔ cl ij = 0; 1 ≤ i < j 6 k, 1 6 l 6 n − 1. Xl=1 In order to prove Theorem 4.3, we need some lemmas. Lemma 4.13. If G is an M Dn-algebra (n > 3) whose the first derived ideal G1 is commutative and 1-codimensional then dim ΩF ∈ {0, 2}, for every F ∈ G ∗. Proof. Because G1 = Span (X1, X2, . . . , Xn−1) is (n − 1)-dimensional commu- tative and G is non-commutative, so cl ij = 0; 1 6 i < j 6 n−1, 1 6 l 6 n−1 and the adjoint adXn is an isomorphism on G1. Therefore the matrix(cid:16)−ci In particular, the (1 6 j 6 n − 1) are not concomitantly vanish. Choose n−1 ∈ G ∗. It is easily seen that the matrix of the Kirillov form BF in the of adXnin the basis (X1, X2, . . . , Xn−1) of G1 is invertible. structure constants cn−1 F = X ∗ basis (X1, X2, . . . , Xn) as follows jn(cid:17)i,j=1,n−1 jn BF = 0 0 ... 0 0 0 ... 0 −cn−1 1n −cn−1 2n   0 0 ... 0 · · · · · · . . . · · · · · · −cn−1 n−1,n 1n , . . . , cn−1 cn−1 1n cn−1 2n ... cn−1 n−1,n 0 .   It is obvious that rankBF = 2 because cn−1 n−1,n are not concomitantly vanish. Since G is an M D-algebra, we get dim ΩF = rankBF ∈ {0, 2} for any F ∈ G ∗. (cid:3) Lemma 4.14. The following (n − k − 1)-square matrix A is invertible ck+1 k+1,n ... cn−1 k+1,n A =  . . . . . . · · · ck+1 n−1,n ... cn−1 n−1,n .   Proof. Since G1 = [G, G], there exist real numbers αij (1 6 i < j 6 n) such that = n−1 Xk+1 = P16i<j6n Pj=k+1 Pj=k+1 Pj=k+1 n−1 n−1 = = αij [Xi, Xj] k αjn [Xj , Xn] + Pj=1 αjn [Xj , Xn] + LC1(cid:0)G2(cid:1) αjn k Pl=k+1 Pl=1 cl jnXl + n−1 αjn [Xj , Xn] + P16i<j6n−1 jnXl! + LC1(cid:0)G2(cid:1) . cl αij [Xi, Xj ] LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 19 Hence, we get Xk+1 = n−1 Pl=k+1 n−1 Pj=k+1 cl jnαjn! Xl + LC2(cid:0)G2(cid:1) ; where LC1(cid:0)G2(cid:1) and LG2(cid:0)G2(cid:1) are a linear combinatory of the vectors in the basis (X1, . . . , Xk) of G2. Because of the independence of chosen basis (X1, X2, . . . , Xn), these assertions imply that there exists one row-vector Yk+1 ∈ Rn−k−1 such that Yk+1A = (1, 0, . . . , 0). Similarly, there exist Yk+2, . . . , Yn−1 ∈ Rn−k−1 such that Yk+2A = (0, 1, . . . , 0), . . . , Yn−1A = (0, . . . , 0, 1). So there exist a real matrix P such that P A = I, where I is the unit matrix of M atn−k−1 (R). Therefore A is an invertible matrix. (cid:3) The following lemma is the well-known result of Linear Algebra for any skew- symmetric real 4-square matrix and it can be easily verified by simple computation. Lemma 4.15. For any skew-symmetric real 4-square matrix (aij)i,j=1,4, its deter- minant is zero if and only if a12.a34 − a13.a24 + a14.a23 = 0. (cid:3) Proof of Theorem 4.3. We now prove Theorem 4.3. Proof of the part (i). Let G be a real solvable Lie algebra of dimension n whose the first derived ideal G1 ∼= Rn−1 is 1-codimensional and commutative. Recall that, with notations as above, G1 ≡ R.X1 ⊕ R.X2 ⊕ · · · ⊕ R.Xn−1. Let F be an arbitrary element in G ∗ ≡ Rn. Put hF, [Xi, Xn]i = ai; 1 6 i 6 n − 1. Then, by simple computation, we can see that the matrix BF of the Kirillov form BF is given as follows. BF =   0 0 ... 0 0 0 ... 0 −a1 −a2 0 0 ... 0 · · · · · · .. . · · · · · · −an−1 a1 a2 ... an−1 0 .   It is clear that rankBF ∈ {0, 2} and, for every F ∈ G ∗, rankBF is not concomi- tantly vanish. Hence, by virtue of Proposition 2.4, G is an M D(n, n − 1)-algebra. Proof of the part (ii). We will show that if G is an real solvable Lie algebra of dimension n > 4 whose the first derived ideal G1 is 1-codimensional and non- commutative then G is not be an M D-algebra. Recall that, we always choose one basis (X1, X2, . . . , Xn) of G such that G1 = Span (X1, X2, . . . , Xn−1) and G2 = Span (X1, . . . , Xk) , k 6 n − 2. There are some cases which contradict each other for the values of k as follows. The first case: k = n−2. Then, dim G2 = dim G1 −1. According to Proposition 2.10, G is not an M D-algebra. 20 ANH VU LE ET AL. The second case: k 6 n−3. It is sufficient to prove for just k = n−3 because the proof for each k in this case is similar. That means G2 = Span (X1, X2, . . . , Xn−3). [Xn−1, Xn] = n−1 Xl=1 cl n−1,nXl; [Xn−2, Xn] = n−3 cl n−2,nXl; n−1 Xl=1 [Xn−2, Xn−1] = n−3 cl n−2,n−1Xl; Xl=1 Xl=1 n−3 Xl=1 [Xi, Xn] = cl inXl; [Xi, Xj ] = cl ij Xl; 1 6 i < j 6 n − 3. According to the Lemma 4.14, matrix P = (cid:20) cn−2 F be an arbitrary element of G ∗. Put cn−2 n−2,n n−1,n C n−1 n−2,n C n−1 n−1,n(cid:21) is invertible. Let hF, [Xn−2, Xn−1]i = a; hF, [Xn−2, Xn]i = b; hF, [Xn−1, Xn]i = c. Then the matrix of the Kirillov form BF in the basis (X1, X2, . . . , Xn) is given by BF = ... 0 0 · · · 0 0 · · · ... .. . 0 0 · · · ∗ ∗ · · · ∗ ∗ · · · ∗ ∗ · · ·   ∗ ∗ 0 ∗ ∗ 0 ... ... ... ∗ ∗ 0 0 ∗ b ∗ −a c ∗ −b −c 0 ∗ ∗ ... ∗ a 0 ,   in which the asterisks denote the undetermined numbers. Let us consider the 4-square submatrices of BF established by the elements which are on the rows and the columns of the same ordinal numbers i, n − 2, n − 1, n (i 6 n − 3). According to Lemma 4.13, rankBF ∈ {0, 2} and this implies that the determinants of these considered 4-square submatrices are zero for any F ∈ G ∗. In view of Lemma 4.15, the following structure constants are vanished: cl i,n−2 = cl i,n−1 = 0; 1 6 i, l 6 n − 3. This implies [Xi, Xn−2] = [Xi, Xn−1] = 0, 1 6 i 6 n − 3. Therefore, we get Span (X1, . . . , Xn−3) = G2 =(cid:2)G1, G1(cid:3) = Span ([Xi, Xj] 1 6 i, j 6 n − 1) = Span ([Xn−2, Xn−1]) . So n − 3 = dim G2 6 1, i.e. n 6 4, which conflicts with the assumption that n > 4. The proof is complete. (cid:3) LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 21 4.4. Proof of Theorem 4.5. As vector spaces (without the Lie structures), we have G = Span (X1, X2, . . . , Xn) ≡ Rn; G1 = Span (X1, X2, . . . , Xn−1) ≡ Rn−1. Let A = (aij)i,j=1,n−1 be a some invertible real (n − 1)-square matrix. Proof. Proof of part (i). We define a Lie structure on G such that G1 is commu- tative and A is exactly the matrix of adjoint map adXn on G1 in the chosen basis (X1, X2, . . . , Xn). Namely, the Lie brackets [·, ·]A on G are given as follows. (4.1) [Xn, Xj]A :=Xi<n aij Xi ; j = 1, 2, . . . , n − 1; the others are trivial. With such Lie structure, the derived ideal of G is commutative and exactly equals to G1. Hence, G is an M D(n, n − 1)-algebra. Conversely, suppose there is a Lie structure on G whose the Lie brackets [·, ·] satisfies the above property (4.1). Because the first derived ideal of G is commu- tative and equals to G1, one has [Xi, Xj] = 0 for all i, j = 1, 2, . . . , n − 1. On the other hand, A is the matrix of adjoint map adXn on G1. Therefore, we get aijXi for all j = 1, 2, . . . , n − 1. That means [·, ·] ≡ [·, ·]A and the part (i) is proved. Proof of part (ii). Let B = (bij)i,j=1,n−1 be an other invertible real (n − 1)- square matrix and [·, ·]B the Lie brackets on G which is defined by B. Then, we have bij Xi ; j = 1, 2, . . . , n − 1 (the other Lie brackets are trivial). [Xn, Xj] = Pi<n (=⇒) Suppose that A and B define two Lie structures on G which are isomorphic. We will show that there exist a real number c 6= 0 and an invertible real (n − 1)- square matrix C such that cA = CBC −1. Denote by f : (G, [·, ·]B) → (G, [·, ·]A) the isomorphism between two Lie structures on G defined by B and A, respec- tively. This means that f is a linear isomorphism and f preserves the Lie brack- ets. Let M = (cij )i,j=1,n be the invertible n-square matrix of f in the basis (X1, X2, . . . , Xn), i.e. f (Xj) = cijXi for al j = 1, 2, . . . , n. Since f is an n isomorphism, f(cid:0)G1, [·, ·]B(cid:1) =(cid:0)G1, [·, ·]A(cid:1), i.e. cnj = 0 for all j = 1, 2, . . . , n − 1. That means f (Xj) = Pi<n and cnn = c, we get f (Xn) = Pi<n cij Xi for all j = 1, 2, . . . , n − 1. Put C = (cij)i,j=1,n−1 cinXi + cXn. Then the matrix of f in the basis (X1, X2, . . . , Xn) is given by Pi=1 [Xn, Xj]B = Pi<n M =  c11 ... cn−1,1 0 · · · · · · · · · · · · c1,n−1 ... c1n ... cn−1,n−1 cn−1,n 0 c =(cid:20)C ∗ c(cid:21) , 0   where the asterisk denotes the column vector (c1n, c2n, . . . , cn,n−1)T . Since f is an isomorphism, 0 6= det M = c det C. Therefore, det C 6= 0 and C is an invertible clj Xl(cid:21)A ; ∀j = 1, . . . , n 22 ANH VU LE ET AL. (n − 1)-square matrix. Furthermore, we have f ([Xn, Xj]B) = [f (Xn), f (Xj)]A; ∀j = 1, . . . , n cln[Xn, Xl]A; ∀j = 1, . . . , n cknXk + cXn + cniXn,Pl<n aklXk(cid:19) ; aklcln(cid:19)Xk; ckiXk(cid:19) = cPl<n cln(cid:18)Pk<n ckibij(cid:19)Xk = c Pk<n(cid:18)Pl<n aklcln(cid:19) Xk = 0; ckibij − cPl<n aklcln = 0; bij Xi(cid:19) =(cid:20)Pk<n ⇔ f(cid:18)Pi<n ⇔ Pi<n bijf (Xi) = cPl<n bij(cid:18)Pk<n ⇔ Pi<n ⇔ Pk<n(cid:18)Pi<n ⇔ Pk<n(cid:18)Pi<n ⇔ Pi<n ⇔ Pi<n ckibij − cPl<n ckibij = cPl<n aklcln; ∀k, j = 1, . . . , n − 1 ∀k, j = 1, . . . , n − 1 ∀j = 1, . . . , n ∀j = 1, . . . , n ∀j = 1, . . . , n ∀k, j = 1, . . . , n − 1 ⇔ (CB)kj = c(AC)kj; ⇔ CB = cAC ⇔ cA = CBC −1. (⇐=) Conversely, if there exist a non-zero real number c and an invertible (n − 1)-square matrix C = (cij)i,j=1,n−1 such that cA = CBC −1. We will show that A, B define on G two Lie structures which are isomorphic. Indeed, we de- note by f : G → G the linear isomorphism of G which is defined, in the basis (X1, X2, . . . , Xn), by the following invertible n-square matrix c11 ... cn−1,1 0 M ′ =  · · · · · · · · · · · · ... c1,n−1 0 ... cn−1,n−1 0 c 0   =(cid:20)C 0 c(cid:21) . 0 It can be verified that f preserves the Lie brackets of (G, [·, ·]B ) and (G, [·, ·]A), i.e. f is also a Lie isomorphism. The proof is complete. (cid:3) 5. CONCLUSION We close the paper with some remarkable comments on the problem of the clas- sification of M D-class. 5.1. M D2k-class. We emphasize that the problem of classification of M D-algebras is still open up to now. There are at least three ways of proceeding in the classifi- cation of M D-class as follows: • The first way: By fixity of the dimension of M D-algebras. • The second way: By fixity of the maximal dimension of coadjoint orbits. • The third way: Combining form of the above ways. For example, the classifications of M D4-class, M D5-class and M D(∗, 1)-class and M D(∗, ∗ − 1)-class in the paper are the results belonging to the first way, but LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 23 the Arnal's list is the result of the second way. To classify M D-class by the second or third way, we give the following definitions. Definition 5.1. Each n-dimensional solvable Lie group G whose coadjoint orbits have dimension zero or 2k (0 < 2k < n) is called an M D2kn-group. The Lie algebra G = Lie(G) of G is called an M D2kn-algebra. When we do not pay attention to the dimension of the considered group or algebra, we will call G or G an M D2k-group or M D2k-algebra, respectively. Definition 5.2. The set of all M D2kn-algebras or M D2k-algebras will be denoted by M D2kn-class or M D2k-class, respectively. 5.2. Examples of M D4-algebras. It has long been known that the Lie algebra aff(C) of the group Aff(C) of the affine transformations of complex straight line is an M D44-algebra and the 5-dimensional Heisenberg Lie algebra h5 is an M D45- algebra. Now we introduce two examples of indecomposable M D4-algebras. The First Example: Let G2m = Span (X1, X2, . . . , X2m) be the 2m-dimensional real Lie algebra (2 6 m ∈ N) with Lie brackets as follows [X1, Xk] := Xk; [X2, X2j−1] := X2j; [X2, X2j] := −X2j−1; with 3 6 k 6 2m, 2 6 j 6 m. The Second Example: Let G2m+1 = Span (X1, X2, . . . , X2m+1) be the (2m + 1)-dimensional real Lie algebra (2 6 m ∈ N) with Lie brackets as follows [X1, Xk] := Xk; 3 6 k 6 2m; [X3, X4] = X2m+1; [X1, X2m+1] = 2X2m+1; [X2, X2j] := −X2j−1; [X2, X2j−1] := X2j; 2 6 j 6 m. Upon simple computation, taking Proposition 2.4 into account we get the fol- lowing proposition. Proposition 5.3. G2m is an indecomposable 2m-dimensional M D4-algebra and G2m+1 is an indecomposable (2m + 1)-dimensional M D4-algebra. 5.3. Some open problems. We have at least two open problems as follows. 5.3.1. Classify M D(n, m)-class, M D(n, n − m)-class; 2 6 m 6 n − 2, n > 6. 5.3.2. Classify M D2kn-class or M D2k-class; 2k > 4, n > 6. In the next papers, we will discuss the classification of M D(n, 2)-class, M D(n, n− 2)-class and M D4n-class for 6 6 n 6 8. Acknowledgments. The authors would like to take this opportunity to thank the scientific program of University of Economics and Law, Vietnam National Univer- sity -- Ho Chi Minh City for financial supports. 24 ANH VU LE ET AL. REFERENCES [ACL95] D. Arnal, M. Cahen, and J. Luwig, Lie Groups whose Coadjoint Orbits are of Dimension Smaller or Equal to Two, Letter in Mathematical Physics 33 (1995), 183 -- 186. ↑2, 3, 10 [BFNT13] L. Boza, E. M. Fedrian, J. Nunez, and A. F. Tenorio, A Historical Review of the Classi- fications of Lie Algebras, Revista De La Union Matematica Argentina 54 (2013), no. 2, 75 -- 99. ↑3 [Car94] E. Cartan, Sur la structure des groupes de transformations finis et continus, Nony, Paris, 1894. ↑1 [CS07] R. Campoamor-Stursberg, A note on the classification of nine-dimensional Lie algebras with nontrivial Levi decomposition, Int. Math. Forum 2 (2007), 25 -- 28. ↑2 [Che12] L. Chen, A Class of Solvable Lie Algebras with Triangular Decompositions, Communi- cations in Algebra 40 (2012), 2285 -- 2300. ↑2 [Do99] N. D. Do, Method of Noncommutative Geometry for Group C ∗-algebras, Chapman and Hall-CRC Press, Cambridge, 1999. ↑3, 4, 5 [DPU12] M. T. Duong, G. Pinczon, and R. Ushirobira, A new invariant of quadratic Lie algebras, Algebr. Represent. Theory 15 (2012), 1163 -- 1203. ↑2 [GT99] L. Yu. Galitski and D. A. Timashev, On classification of metabelian Lie algebras, Journal of Lie Theory 9 (1999), 125 -- 156. ↑2 [Gan39] F. R. Gantmakher, On the Classifiction of Real Simple Lie Group, Mat. Sb. 5 (1939), 217 -- 250. ↑1 [Gau73] M. A. Gauger, On the classification of metabelian Lie algebras, Trans. Amer. Math. Soc 179 (1973), 293 -- 329. ↑2 [Kat07] I. Kath, Nilpotent Metric Lie Algebras of Small Dimension, J. Lie Theory 17 (2007), no. 1, 41 -- 61. ↑2 [Kir76] A. A. Kirillov, Elements of the Theory of Representations, Springer-Verlag, New York, 1976. ↑2, 4, 5 [Le90a] A. V. Le, On the structure of the C ∗-Algebra of the Foliation formed by the K-Orbits of maximal dimendion of the Real Diamond Group, Journal of Operator Theory 24 (1990), 227 -- 238. ↑3, 7 [Le90b] , On the Foliations Formed by the Generic K-orbits of the MD4-Groups, Acta Mathematica Vietnamica 15 (1990), no. 2, 39 -- 55. ↑3, 12 [Le93] , Foliations Formed by Orbits of Maximal Dimension in the Co-adjoint Represen- tation of a Class of Solvable Lie Groups, Vest. Moscow Uni., Math. Bulletin 3 (1993), no. 48, 24 -- 27. ↑3, 7, 8, 12 [LS08] A. V. Le and K. P. Shum, Classification of 5-dimensional MD-algebra having commuta- tive derived ideal, Advances in Algebra and Combinatorics, Singapore: World Scientific 12 (2008), no. 46, 353 -- 371. ↑3, 7, 8, 9 [LHT11] A. V. Le, V. H. Ha, and T. H. N. Tran, Classification of 5-dimensional MD-algebras having non-commutative derived ideals, East-West Journal of Mathematics 13 (2011), no. 2, 115 -- 129. ↑3, 6, 8 [Mal45] A. I. Maltsev, On solvable Lie algebras, Bull. Acad. Sci. URSS. Ser. Math. 9 (1945), no. 1, 329 -- 356. ↑1 [Mub63] G. M. Mubarakzyanov, Classification of real structures of Lie algebras of fifth order, Izv. Vyssh. Uchebn. Zaved., Mat. 3 (1963), no. 34, 99 -- 106. ↑2 [PZ90] J. Patera and H. Zassenhaus, Solvable Lie algebras of dimension 6 4 over perfect fields, Linear Algebras Appl. 142 (1990), no. 1, 1 -- 17. ↑2 [Sno10] L. Snobl, On the structure of maximal solvable extensions and of Levi extensions of nilpotent Lie algebras, J. Phys. A43 (2010), no. 50, 1 -- 17. ↑2 [SK10] L. Snobl and D. Karasek, Classification of solvable Lie algebras with a given nilradical by means of solvable extensions of its subalgebras, Linear Algebra Appl. 432 (2010), no. 7, 1836 -- 1850. ↑ LIE GROUPS WHOSE COADJOINT ORBITS HAVE ONLY ZERO OR MAXIMAL DIMENSION 25 [SW05] L. Snobl and P. Winternitz, A class of solvable Lie algebras and their Casimir invariants, J.Phys. A: Math. Gen. 38 (2005), no. 12, 2687 -- 2700. ↑ [Tsa99] G. Tsagas, Classification of nilpotent Lie algebras of dimension 8, J. Inst. Math. Comput. Sci. Math. Ser. 12 (1999), no. 3, 179 -- 183. ↑2 [TKK00] G. Tsagas, A. Kobotis, and T. Koukouvinos, Classification of nilpotent Lie algebras of dimension nine whose maximum Abelian ideal is of the dimension seven, Int. J. Comput. Math. 74 (2000), no. 1, 5 -- 28. ↑ [Tur90] P. Turkowski, Solvable Lie algebras of dimension 6, J. Math. Phys. 31 (1990), no. 6, 1344 -- 1350. ↑2 [VH84] M. S. Vuong and H. V. Ho, Sur la structure des C ∗-algebres dune classe de groupes de Lie, J. Operator Theory 11 (1984), 77 -- 90. ↑3, 5 ANH VU LE, DEPARTMENT OF ECONOMIC MATHEMATICS, UNIVERSITY OF ECONOMICS AND LAW, VIETNAM NATIONAL UNIVERSITY - HO CHI MINH CITY, VIET NAM. E-mail address: [email protected] VAN HIEU HA, DEPARTMENT OF ECONOMIC MATHEMATICS, UNIVERSITY OF ECONOMICS AND LAW, VIETNAM NATIONAL UNIVERSITY - HO CHI MINH CITY, VIET NAM. E-mail address: [email protected] ANH TUAN NGUYEN, FACULTY OF POLITICAL SCIENCE AND PEDAGOGY, UNIVERSITY OF PHYSICAL EDUCATION AND SPORTS, HO CHI MINH CITY, VIET NAM. E-mail address: [email protected] TRAN TU HAI CAO, LE QUY DON HIGH SCHOOL FOR THE GIFTED, NINH THUAN PROVINCE, VIET NAM. E-mail address: [email protected] THI MONG TUYEN NGUYEN, FACULTY OF MATHEMATICS AND INFORMATION, DONG THAP UNIVERSITY, CAO LANH CITY, DONG THAP PROVINCE, VIET NAM. E-mail address: [email protected]
1206.4165	1	1206	2012-06-19T10:03:53	Rational matrix pseudodifferential operators	[ "math.RA", "math-ph", "math-ph", "nlin.SI" ]	The skewfield K(d) of rational pseudodifferential operators over a differential field K is the skewfield of fractions of the algebra of differential operators K[d]. In our previous paper we showed that any H from K(d) has a minimal fractional decomposition H=AB^(-1), where A,B are elements of K[d], B is non-zero, and any common right divisor of A and B is a non-zero element of K. Moreover, any right fractional decomposition of H is obtained by multiplying A and B on the right by the same non-zero element of K[d]. In the present paper we study the ring M_n(K(d)) of nxn matrices over the skewfield K(d). We show that similarly, any H from M_n(K(d)) has a minimal fractional decomposition H=AB^(-1), where A,B are elements of M_n(K[d]), B is non-degenerate, and any common right divisor of A and B is an invertible element of the ring M_n(K[d]). Moreover, any right fractional decomposition of H is obtained by multiplying A and B on the right by the same non-degenerate element of M_n(K [d]). We give several equivalent definitions of the minimal fractional decomposition. These results are applied to the study of maximal isotropicity property, used in the theory of Dirac structures.	math.RA	math	Rational matrix pseudodifferential operators Sylvain Carpentier ∗, Alberto De Sole †, Victor G. Kac ‡ November 20, 2018 Abstract The skewfield K(∂) of rational pseudodifferential operators over a dif- ferential field K is the skewfield of fractions of the algebra of dif- ferential operators K[∂]. In our previous paper we showed that any H ∈ K(∂) has a minimal fractional decomposition H = AB−1, where A, B ∈ K[∂], B 6= 0, and any common right divisor of A and B is a non-zero element of K. Moreover, any right fractional decomposi- tion of H is obtained by multiplying A and B on the right by the same non-zero element of K[∂]. In the present paper we study the ring Mn(K(∂)) of n × n matrices over the skewfield K(∂). We show that similarly, any H ∈ Mn(K(∂)) has a minimal fractional decom- position H = AB−1, where A, B ∈ Mn(K[∂]), B is non-degenerate, and any common right divisor of A and B is an invertible element of the ring Mn(K[∂]). Moreover, any right fractional decomposition of H is obtained by multiplying A and B on the right by the same non- degenerate element of Mn(K[∂]). We give several equivalent definitions of the minimal fractional decomposition. These results are applied to the study of maximal isotropicity property, used in the theory of Dirac structures. 1 Introduction Let K be a differential field with derivation ∂ and let K[∂] be the algebra of differential operators over K. The skewfield K(∂) of rational pseudodiffer- ential operators is, by definition, the subskewfield of the skewfield of pseu- dodifferential operators K((∂−1)), generated by the subalgebra K[∂]. In our ∗Ecole Normale Superieure, 75005 Paris, France, and M.I.T [email protected] †Dipartimento di Matematica, Universit`a di Roma "La Sapienza", 00185 Roma, Italy [email protected] Supported in part by Department of Mathematics, M.I.T. 02139, ‡Department Mathematics, Cambridge, of M.I.T., MA USA. [email protected] Supported in part by an NSF grant 1 paper [CDSK12] we showed that any rational pseudodifferential operator H has a unique right minimal decomposition H = AB−1, where A, B ∈ K[∂], B is a non-zero monic differential operator, and any other right fractional decomposition of H can be obtained by multiplying on the right both A and B by a non-zero differential operator D. In the present paper we establish a similar result for the ring Mn(K(∂)) of n×n matrix rational pseudodifferential operators. Namely we show that any H ∈ Mn(K(∂)) has a right minimal fractional decomposition H = AB−1, where B ∈ Mn(K[∂]) is non-degenerate (i.e. has a non-zero Dieudonn´e determinant det(B)), satisfying one of the following equivalent properties : (i) d(B) is minimal among all possible right fractional decompositions H = AB−1, where d(B) is the order of det(B) ; (ii) A and B are coprime, i.e. if A = A1D and B = B1D, with A1, B1, D ∈ Mn(K[∂]), then D is invertible in Mn(K[∂]); (iii) Ker A ∩ Ker B = 0 in any differential field extension of K. By (i), a right minimal fractional decomposition exists for any n × n matrix rational pseudodifferential operator H. We prove its uniqueness, namely that all right minimal fractional decompositions can be obtained from each other by multiplication on the right of the numerator and the denominator by an invertible n × n matrix differential operator D. Moreover, any right fractional decomposition of H can be obtained by multiplying on the right the numerator and the denominator of a minimal right fractional decompo- sition by the same non-degenerate matrix differential operator. We derive from these results the following maximal isotropicity property of the minimal fractional decomposition H = AB−1, which is important for the theory of Dirac structures [D93], [BDSK09], [DSK12]. Introduce the following bilinear form on the space Kn ⊕ Kn with values in K/∂K : (P1 ⊕ Q1P2 ⊕ Q2) =Z (P1.Q2 + P2.Q1), where R stands for the canonical map K → K/∂K and P.Q is the standard dot product. Let A and B be two n × n matrix differential operators. Define LA,B = {B(∂)P ⊕ A(∂)P P ∈ Kn}. It is easy to see that, assuming that det(B) 6= 0, the subspace LA,B of Kn ⊕ Kn is isotropic if and only if the matrix rational pseudodifferential operator 2 H = AB−1 is skewadjoint. We prove that LA,B is maximal isotropic if AB−1 is a right minimal fractional decomposition of H. Note that LA,B is independent of the choice of the minimal fractional decomposition due to its uniqueness, mentioned above. We wish to thank Pavel Etingof and Andrea Maffei for useful discussions, and Mike Artin, Michael Singer and Toby Stafford for useful correspondence. 2 Some preliminaries on rational pseudodifferen- tial operators Let K be a differential field of characteristic 0, with a derivation ∂, and let C = Ker ∂ be the subfield of constants. Consider the algebra K[∂] (over C) of differential operators. It is a subalgebra of the skewfield K((∂−1)) of pseudodifferential operators. The subskewfield K(∂) of K((∂−1)), generated by K[∂], is called the skewfield of rational pseudodifferential operators (see [CDSK12] for details). We have obvious inclusions : K ⊂ K[∂] ⊂ K(∂) ⊂ K((∂−1)). If the derivation acts trivially on K, so that C = K, letting ∂ = λ, an indeter- minate, commuting with elements of K, we obtain inclusions of commutative algebras C ⊂ C[λ] ⊂ C(λ) ⊂ C((λ−1)). It is well known that in many respects the non-commutative algebras K[∂] and K(∂) "behave" in a very similar way to that of C[λ] and C(λ). Namely, the ring K[∂] is right (resp. left) Euclidean, hence any right (resp. left) ideal is principal. Moreover, any two right ideals AK[∂] and BK[∂] have non-zero intersection M K[∂], where M 6= 0 is called the least right common multiple of A and B; also AK[∂] + BK[∂] = DK[∂], where D is the greatest right common divisor of A and B. Furthermore, any element H of K(∂) has a right fractional decomposition H = AB−1, where B 6= 0. A right fractional decomposition for which the differential operator B has minimal order is called the minimal fractional decomposition (equivalently, the greatest com- mon divisor of A and B is 1). It is unique up to multiplication of A and B on the right by the same non-zero element of K. Any other fractional decomposition of H is obtained from the minimal one by multiplication of A and B on the right by a non-zero element of K[∂]. See [CDSK12] for details. Of course all these facts still hold if we replace "right" by "left". 3 3 The Dieudonn´e determinant The Dieudonn´e determinant of an n × n matrix pseudodifferential operator A ∈ Mn(K((∂−1))) has the form det(A) = det1(A)λd(A) where det1(A) ∈ K, λ is an indeterminate, and d(A) ∈ Z. It exists and is uniquely defined by the following properties (see [Die43], [Art57]) : (i) det(AB) = det(A) det(B); (ii) If A is upper triangular with non-zero diagonal entries Aii ∈ K((∂−1)) of degree (or order) d(Aii) and leading coefficient ai ∈ K, then det1(A) = nYi=1 ai, d(A) = nXi=1 d(Aii). By definition, det(A) = 0 if one of the Aii is 0. Note that det1(AB) = det1(A) det1(B). A matrix A whose Dieudonn´e determinant is non-zero is called non-degenerate. In this case the integer d(A) is well defined. It is called the degree of det(A) and of A. Note that d(AB) = d(A) + d(B) if both A and B are non-degenerate. Lemma 3.1. (a)Any A ∈ Mn(K[∂]) can be written in the form A = U T (resp. T U ), where U is an invertible element of Mn(K[∂]) and T ∈ Mn(K[∂]) is upper triangular. (b) Any non-degenerate A ∈ Mn(K[∂]) can be written in the form A = U1DU2, where U1, U2 are invertible elements of Mn(K[∂]) and D is a diag- onal n × n matrix with non-zero entries from K[∂]. Proof. Recall that an elementary row (resp. column) operation of a matrix from Mn(K[∂]) is either a permutation of two of its rows (resp. column), or adding to one row (resp. column) another one, multiplied on the left (resp. right) by an element of K[∂]. Since the row (resp. column) operations are equivalent to multiplication on the left (reps. right) by the corresponding elementary matrix, the first operation only changes the sign of the determi- nant and the second does not change it. In the proof of (a) we may assume that A 6= 0, and let j be the minimimal index, for which the j-th column is non-zero. Among all matrices that can be obtained from A by elementary row operations choose the one for which the (1, j)-entry is non-zero and has the minimal order. Then, by elementary row operations, using the Euclidean property of K[∂], we obtain from A a matrix A1 such that all entries of the j-th column, except the first one, are 4 zero. Repeating this process for the (n − 1) × (n − 1) submatrix obtained from A1 by deleting the first row and column, we obtain the decomposition A = U T as in (a). For the decomposition A = T U , we use a similar argument, except that we start from largest j for which the j-th row is non-zero, we perform column operations to have the (j, n)-entry non-zero and of minimal possible order, and then we further make elementary column operations to obtain a matrix A1 such that all entries of the j-th row are zero, except the last one. The claim follows by induction, after deleting the last row and column. In order to obtain the decomposition in (b), we use the same argument, except that we choose among all matrices obtained from A by elementary row and column operations the one for which the (1, 1)-entry is non-zero and has the minimal order (it exists since det(A) 6= 0). Corollary 3.2. Let A ∈ Mn(K[∂]) be a non-degenerate matrix differential operator. Then (a) d(A) ∈ Z+. (b) A is an invertible element of the ring Mn(K[∂]) if and only if d(A) = 0. Remark 3.3. Let A ∈ Mn(K((∂−1))) and let A∗ be the adjoint matrix pseu- dodifferential operator. If det(A) = 0, then det(A∗) = 0. If det(A) 6= 0, then det(A∗) = (−1)d(A)det(A). This follows from the obvious fact that A can be brought by elementary row transformations over the skewfield K((∂−1)) to an upper triangular matrix, and in this case the statement becomes clear. 4 Rational matrix pseudodifferential operators A matrix H ∈ Mn(K(∂)) is called a rational matrix pseudodifferential op- erator. In other words, all the entries of such a matrix have the form −1, i, j = 1, ..., n, where aij, bij ∈ K[∂] and all bij 6= 0. Let hij = aijbij b(6= 0) be the least right common multiple of the bij's, so that bij.cij = b for some cij 6= 0. Multiplying aij and bij on the right by cij, we obtain H = A1b−1, where (A1)ij = aijcij. In other words H has the right frac- tional decomposition H = A1(b1In)−1. However, among all right fractional decompositions H = AB−1, where A, B ∈ Mn(K[∂]) and det B 6= 0, this might be not the "best" one. Definition 4.1. A right fractional decomposition H = AB−1, where A, B ∈ Mn(K[∂]) and detB 6= 0, is called minimal if d(B) ( ∈ Z+) is minimal among all right fractional decompositions of H. 5 Note that, if H = AB−1 is a minimal fractional decomposition, then 0 ≤ d(B) ≤ d(b), where b is the least right common multiple of all the entries of H. Proposition 4.2. Let A and B be two non-degenerate n × n matrix differ- ential operators. Then one can find non-degenerate n × n matrix differential operators C and D, such that AC = BD ( resp. CA = DB ) Proof. By induction on n. We know it is true in the scalar case, see e.g. [CDSK12]. By Lemma 3.1, multiplying on the right by invertible matrices, we may assume that both A and B are upper triangular matrices. Let A =  A1 0   , B =  U a B1 0   , V b where A1, B1 ∈ Mn−1(K[∂]) are upper triangular non-degenerate, U, V ∈ K[∂]n, and a, b ∈ K[∂]\{0}. By the inductive assumption, there exist C1, D1 ∈ Mn−1(K[∂]) non-degenerate, such that A1C1 = B1D1, and c, d ∈ K[∂]\{0} such that ac = bd. Hence, after multiplying on the right A by the block diagonal matrix with C1 and c on the diagonal, and B by the block diagonal matrix with D1 and d on the diagonal, we may assume that A1 = B1 and a = b . Consider the matrix M =  Viewed over the skewfield K(∂), it has a non-zero kernel (since M : K(∂)n 7→   ∈ Mn(K[∂]) . K(∂)n is not surjective), i.e. there exists a vector eX = (cid:18) X where X ∈ K(∂)n−1 and x ∈ K(∂), such that M eX = 0, i.e. Replacing eX by eXd, where d is a non-zero common multiple of all the denominators of the entries of eX, we may assume that eX ∈ K[∂]n. Note x (cid:19) ∈ K(∂)n, A1X + U x = V x . (4.1) A1 0 U − V 0 6 also that, since A1 is non-degenerate, it must be x 6= 0. To conclude the proof we just observe that, by (4.1), we have the identity AE = BF , where E =  1In−1 0   , F =  X x 1In−1 0   . 0 x Remark 4.3. Proposition 4.2 can be derived from Goldie theory (see [MR01, Theorem 2.1.12]), but we opted for a simple direct argument. Theorem 4.4. For every matrix differential operators A, B ∈ Mn(K[∂]) with det(B) 6= 0, there exist matrices A1, B1, D ∈ Mn(K[∂]), with det B1 6= 0, det D 6= 0, such that: (i) A = A1D, B = B1D, (ii) Ker A1 ∩ Ker B1 = 0. Proof. We will prove the statement by induction on d(B). If d(B) = 0, then B is invertible in Mn(K[∂]) by Corollary 3.2 (and Ker B = 0). In this case the claim holds trivially, taking D = 1In. Clearly, if P ∈ Mn(K[∂]) is invertible, then Ker A = Ker P A. Hence, if P and Q are invertible elements of Mn(K[∂]), then the statement holds for A and B if and only if it holds for P A and QB. Furthermore, if R ∈ Mn(K[∂]) is invertible, replacing D by R−1D we get that the statement holds for A and B if and only if it holds for AR and BR. Therefore, by Lemma 3.1, we may assume, without loss of generality, that A is upper triangular and B is diagonal. If Ker A∩Ker B = 0 i=1 be a non-zero element of Ker A ∩ Ker B, and let k ∈ {1, . . . , n} be such that fk 6= 0, fk+1 = · · · = fn = 0. The condition AF = 0 gives for i = 1, . . . , k, there is nothing to prove. Let then F = (cid:0)fi(cid:1)n Ai,1(∂)f1 + · · · + Ai,k−1(∂)fk−1 + Aik(∂)fk = 0 in K. This implies that there is some Li(∂) ∈ K[∂] such that (4.2) Ai,1(∂) ◦ f1 fk + · · · + Ai,k−1(∂) ◦ fk−1 fk + Aik(∂) = Li(∂) ◦(cid:16)∂ − f ′ k fk(cid:17) in K[∂] . Indeed, the LHS above is zero when applied to fk ∈ K, hence it must be divisible, on the right, by ∂ − f ′ . Similarly, from the condition BF = 0 we k fk 7 have that Bii(∂)fi = 0 in K for every i = 1, . . . , k, which implies that there is some Mi(∂) ∈ K[∂] such that (4.3) Bii(∂) ◦ fi fk = Mi(∂) ◦(cid:16)∂ − f ′ k fk(cid:17) in K[∂] . Let then A1, B1, D ∈ Mn(K[∂]) be the matrices defined as the matrices A, B, 1I with the k-th column replaced, respectively, by the following columns   L1 ... Lk−1 Lk 0 ... 0   ,   M1 ... Mk−1 Mk 0 ... 0   ,   −f1/fk ... −fk−1/fk ∂ − f ′ k/fk 0 ... 0   . It follows from equations (4.2) and (4.3) that A1D = A and B1D = B. Moreover, since det D = λ, we have d(B1) = d(B) − 1. The statement follows by the inductive assumption. 5 Linear closure of a differential field In this section we define a natural embedding of a differential field in a linearly closed one using the theory of Picard-Vessiot extensions. One may find all relevant definitions and constructions in Chapter 3 of [Mag94]. Recall [DSK11] that a differential field K is called linearly closed if every homogeneous linear differential equation of order n ≥ 1, (5.4) anu(n) + · · · + a1u′ + a0u = 0 , with a0, . . . , an in K, an 6= 0, has a non-zero solution u ∈ K. It is easy to show that the solutions of equation (5.4) in a differential field K form a vector space over the field of constant C of dimension less than or equal to n, and equal to n if K is linearly closed (see e.g. [DSK11]). Remark 5.1. In a linearly closed field, it is also true that every inhomo- geneous linear differential equation L(∂)u = b has a solution because the homogeneous differential equation ((1/b)L(∂)u)′ = 0 has a solution u such that L(∂)u 6= 0 (the solutions of ((1/b)L(∂)u)′ = 0 form a vector space over the subfield of constants C of dimension strictly bigger than the one of KerL). 8 More generally, if A ∈ Mn(K[∂]) is a non-degenerate matrix differential operator and b ∈ Kn, then the inhomogeneous system of linear differential equations in u =(cid:0)ui(cid:1)n i=1, (5.5) A(∂)u = b , admits the affine space (over C) of solutions of dimension less than or equal to d(A), and equal to d(A) if K is linearly closed. (This follows, for example, from Lemma 3.1(b).) Definition 5.2. Let K be a differential field with the subfield of constants C, and let L ∈ K[∂] be a differential operator over K of order n. A differential field extension K ⊂ L is called a Picard-Vessiot extension with respect to L if there are no new constants in L and if L = K(y1, ..., yn), where the yi are linearly independent solutions over C of the equation Ly = 0. Proofs of the following two propositions can be found in [Mag94]. Proposition 5.3. Let K be a differential field with algebraically closed sub- field of constants C and let L be a differential operator of order n over K. Then there exists a Picard-Vessiot extension of K with respect to L and it is unique up to isomorphism. Proposition 5.4. If K ⊂ L is an extension of differential fields and K ⊂ Ei ⊂ L, i = 1, 2, are two Picard-Vessiot subextensions of K, then the com- posite field E1E2 (i.e. the minimal subfield of L containing both E1 and E2) is a Picard-Vessiot extension of K as well. Definition 5.5. Let K be a differential field with algebraically closed sub- field of constants C. The unique minimal extension K ⊂ L such that (a) L is the union of its Picard-Vessiot subextensions of K; (b) L contains an isomorphic copy of every Picard-vessiot extension of K, is called the Picard-Vessiot compositum of K. It is proved in [Mag94] that the Picard-Vessiot compositum of K exists, and is unique up to isomorphism. Definition 5.6. Let K be a differential field with algebraically closed sub- field of constants. Let K0 = K and, for i ∈ Z+, let Ki+1 be the Picard-Vessiot compositum of Ki. We call L = ∪iKi the linear closure of K (it is called the successive Picard-Vessiot closure in [Mag94] ). 9 Remark 5.7. The linear closure is linearly closed. Remark 5.8. The linear closure of a differential field K with algebraically closed subfield of constants is the unique, up to isomorphism, minimal lin- early closed extension of K with no new constants. To see this, one needs to show that for any linearly closed extension L of K without new constants, one can extend the embedding K ֒→ L to an embedding of the Picard-Vessiot compositum of K, K1 ֒→ L. By Zorn's lemma one can find a maximal subex- differential operator L over K. As L is linearly closed, we can find a Picard- tension K ⊂ eK ⊂ K1 extending the embedding K ֒→ L. Denote by φ the embedding eK ֒→ L. Suppose that eK ( K1. This means that, by definition of K1, we have a non-trivial Picard-Vessiot extension eK ⊂ P ⊂ K1 for a Vessiot extension φ(eK) ⊂ P1 ⊂ L for the same differential operator. By can extend the embedding eK ֒→ L to an embedding P ֒→ L, which is a Proposition 5.3, these two Picard-Vessiot extension are isomorphic and one contradiction. Lemma 5.9. Let K be a differential field with algebraically closed subfield of constants, let L be its linear closure, and let X be a finite subset of L, not contained in K. Then there is an integer i and a Picard-Vessiot extension Ki ⊂ P ⊂ Ki+1 of Ki such that X ⊂ P but X 6⊂ Ki. Proof. Take the minimal i, such that X ⊂ Ki+1. Since Ki+1 is the Picard- Vessiot compositum of Ki, every element of X lies in a Picard-Vessiot exten- tion of Ki. The claim follows by the fact that the composite of two Picard- Vessiot extension is still a Picard-Vessiot extension (Proposition 5.4). Lemma 5.10. Let K ⊂ L be a differential field extension, and let C ⊂ D be the corresponding field extension of constants. If α ∈ L is algebraic over C, then α ∈ D and the minimal monic polynomial for α over K has coefficients in C. Proof. Let P (x) = xn + c1xn−1 + · · · + cn ∈ C[x] be the minimal monic polynomial with coefficients in C satisfied by α. Letting x = α and applying the derivative ∂ we get (cid:0)nαn−1 + (n − 1)c1αn−2 + · · · + cn(cid:1)α′ = 0. By minimality of P (x), it must be α′ = 0, i.e. α ∈ D. Similarly, for the second statement, let Q(x) = xm + f1xm−1 + · · · + fm ∈ K[x] be the minimal monic polynomial with coefficients in K satisfied by α. Letting x = α and applying the derivative ∂ we get f ′ m = 0, which, by minimality of Q(x), implies f1, . . . , fm ∈ C. 1αm−1 + · · · + f ′ 10 Lemma 5.11 (see e.g. [PS03]). Let K be a differential field with subfield of constants C. Then elements f1, . . . , fn ∈ K are linearly independent over any subfield of C if and only if their Wronskian is non-zero. Lemma 5.12. (a) Let K be a differential field with field of constants C, and let D be an algebraic extension of C. Then D ⊗C K is a differential field with field of constants D. (b) Let K be a differential field with field of constants C, and let L be a differential field extension of K with field of constants ¯C, the algebraic closure of C. Then, for every algebraic extension D of C, the differential field D ⊗C K is canonically isomorphic to a differential subfield of L. Proof. For part (a) we need to prove that every non-zero element f = Pi ci ⊗ fi ∈ D ⊗C K is invertible. Let C[α] be a finite extension of C in D containing all elements c1, . . . , cn, and let P (x) ∈ C[x] be the minimal monic polynomial for α over C. By Lemma 5.10, P (x) is an irreducible element of K[x]. Therefore K[x]/(P (x)) is a field, and f ∈ C[α] ⊗C K ≃ K[x]/(P (x)) is invertible. Next, we prove part (b). By the universal property of the tensor product, there is a canonical map ϕ : D ⊗C K → L given by ϕ(c ⊗ f ) = cf . This is a differential field embedding by part (a). Definition 5.13. Let K be a differential field with subfield of constants C. We know from Lemma 5.12(a) that ¯C ⊗C K is a d differential field with subfield of constants ¯C. We define the linear closure of K to be the one of ¯C ⊗C K. Recall that the differential Galois group Gal(L/K) of a differential field extension K ⊂ L is defined as the group of automorphisms of L commuting with ∂ and fixing K. One of the main properties of Picard-Vessiot extensions is the following Proposition 5.14 ([PS03]). Let K be a differential field with algebraically closed subfield of constants C, and let L be a Picard-Vessiot extension of K. Then, the set of fixed points of the differential Galois group Gal(L/K) is K. 6 Minimal fractional decomposition Given a matrix A ∈ Mn(K[∂]), we denote by ¯A the same matrix A considered as an endomorphism of ¯Kn, where ¯K is the linear closure of K. We have the following possible conditions for a "minimal" fractional decomposition H = AB−1 ∈ Mn(K(∂)), where A, B ∈ Mn(K[∂]) and B is non-degenerate: 11 (i) d(B) is minimal among all possible fractional decompositions of H; (ii) A and B are coprime, i.e. if A = A1D and B = B1D, with A1, B1, D ∈ Mn(K[∂]), then D is invertible in Mn(K[∂]); (iii) Ker ¯A ∩ Ker ¯B = 0. Obviously, condition (iii) implies: (iii′) Ker A ∩ Ker B = 0. Example 6.1. Condition (iii′) is weaker than condition (iii). Consider, for example, A = ∂(∂ − 1) and B = ∂ − 1. We have ex ∈ Ker ¯A ∩ ¯B, and Ker A ∩ Ker B = 0 unless the differential field K contains a solution to the equation u′ = u. Remark 6.2. Condition (iii) is equivalent to ask that A and B have no common eigenvector with eigenvalue 0 over any differential field extension of K. Proposition 6.3. In the "scalar" case n = 1, conditions (i), (ii) and (iii) are equivalent. Proof. It follows from [CDSK12] that conditions (i) and (ii) are equivalent. Moreover, condition (iii) implies condition (ii) since, if D ∈ K[∂] is not invertible, than it has some root in the linear closure ¯K. We are left to prove that condition (ii) implies condition (iii). Note that, by the Euclidean algorithm, the right greatest common divisor of A and B is independent of the differential field extension of K. Suppose, by contradiction, that 0 6= f ∈ Ker ¯A∩Ker ¯B, which means that A = A1(∂− f ′ f ), for some A1, B1 ∈ ¯K[∂], so that the right greatest common divisor of A and B is not invertible, contradicting assumption (ii). f ) and B = B1(∂− f ′ Theorem 6.4. (a) Every H ∈ Mn(K(∂)) can be represented as H = AB−1, with B non-degenerate, such that (iii) holds. (b) Conditions (i), (ii), and (iii) are equivalent. Any fraction which sat- isfies one of these equivalent conditions is called a minimal fractional decomposition. (c) If A0B0 −1 is a minimal fractional decomposition of the fraction H = AB−1, then one can find a matrix differential operator D such that A = A0D and B = B0D. Proposition 6.5. Theorem 6.4 holds if K is linearly closed. 12 Lemma 6.6. Assuming that Theorem 6.4(c) holds, let K = A1B−1 be a minimal fractional decomposition, with A1, B1 ∈ Mk(K[∂]). Let also V ∈ K[∂]k be such that AB−1V ∈ V[∂]ℓ. Then V = BZ for some Z ∈ K[∂]k. 1 Proof. After replacing, if necessary, A by AU1, B by U2BU1, and V by U2V , with U1 and U2 invertible elements of Mn(K[∂]), we can assume by Lemma 3.1 that B is diagonal. If V = 0 there is nothing to prove, so let the i-th fractional decomposition for H = AB−1. Hence, by Thorem 6.4(c), we have entry of V be non zero. Consider the matrix eV ∈ Mk(K[∂]) be the same as B, with the i-th column replaced by V . Clearly, eV is non-degenerate. By assumption AB−1eV = K lies in Mn(K[∂]), so that KeV −1 is another that eV = BeZ for some eZ ∈ Mn(K[∂]), so that V = BZ, where Z is the i-th column of eZ. Proof of Proposition 6.5. Part (a) holds by Theorem 4.4. In part (b), con- dition (iii) implies condition (ii) since, by assumption, K is linearly closed. Conversely, let A, B ∈ Mn(K[∂]) satisfy condition (ii). By Theorem 4.4 we have A = A1D, B = B1D with Ker A1 ∩ Ker B1 = 0, and by assumption (ii), D ∈ Mn(K[∂]) is invertible. Hence, Ker A ∩ Ker B = 0, proving (iii). Furthermore, it is clear that condition (i) implies condition (iii). Indeed if KerA ∩ KerB 6= 0, then by Theorem 4.4 one can find C, D, E such that A = CE, B = DE, KerC ∩ KerD = 0 and d(E) > 0. Then AB−1 = CD−1 and d(D) < d(B), contradicting assumption (i). To conclude, we are going to prove, by induction on n, that condition (iii) implies condition (i), and that part (c) holds. If n = 1 the statement holds by Proposition 6.3 and the results in [CDSK12]. Let then n > 1 and A, B ∈ Mn(K[∂]), with B non degenerate, be such that condition (iii) holds: Ker A ∩ Ker B = 0. Let also CD−1 = AB−1 be any other fractional decomposition of H = AB−1, with C, D ∈ Mn(K[∂]), D non degenerate. We need to prove that there exists T ∈ Mn(K[∂]) such that C = AT and D = BT . (In this case, d(D) = d(B) + d(T ) ≥ d(B), proving condition (i)). Firts, note that, if Ui, i = 1, . . . , 4, are invertible elements of Mn(K[∂]), then Ker(U1AU3) ∩ Ker(U2BU3) = 0, and we have (U1AU3)(U2BU3)−1 = (U1CU4)(U2DU4)−1. Hence, by Lemma 3.1 we can assume, without loss of generality, that B is diagonal, A, D are upper triangular, and hence C = 13 AB−1D is upper triangula as well. Let then A =  C =  A1 0 C1 0   , B =    , D =  B1 0 D1 0 U a V c 0 b   ,   , W d where B1 is diagonal and A1, C1, D1 are upper triangular n−1×n−1 matrices with entries in K[∂], with B1 and D1 non degenerate, U, V, W lie in K[∂]n−1, and a, b, c, d lie in K[∂], with b, d 6= 0. By assumption AB−1 = CD−1, meaning that (6.6) A1B−1 1 = C1D−1 1 , ab−1 = cd−1 , U b−1 = −C1D−1 1 W d−1 + V d−1 . Moreover, the assumption Ker A ∩ Ker B = 0 clearly implies that Ker A1 ∩ Ker B1 = 0 (if X ∈ Kn−1 is such that A1(X) = B1(X) = 0, then eX = (cid:18) X 0 (cid:19) ∈ Kn lies in Ker A ∩ Ker B). Hence, by the first identity in (6.6) and the inductive assumption, there exists T1 ∈ Mn−1(K[∂]) such that (6.7) C1 = A1T1 , D1 = B1T1 . The main problem is that we do not know that Ker a ∩ Ker b = 0 (it is false in general), hence we cannot conclude, yet, that c = at and d = bt for some t ∈ K[∂]. Let then ef −1 be a minimal fractional decomposition of ab−1 = cd−1. By the n = 1 case we know that there exist p, q ∈ K[∂] such that (6.8) a = ep , b = f p , c = eq , d = f q , and let k ∈ K[∂] be a right greatest common divisor of p ad q, i.e. there exist s, t, i, j ∈ K[∂] such that (6.9) p = ks , q = kt , si + tj = 1 . Eventually we will want to prove that we can choose k = p (i.e. s = 1, i = 1 and j = 0). Using the identities (6.7), (6.8) and (6.9), we can rewrite the third equation in (6.6) as follows U s−1 = −A1B−1 1 W t−1 + V t−1 , 14 and multiplying each side of the above equation by each side of the identity 1 − si = tj, we get (6.10) U + A1B−1 1 W js = (U i + V j)s . Since A1B−1 assumption and Lemma 6.6, that there exists Z ∈ K[∂]n−1 such that is a minimal fractional decomposition, we get, by the inductive 1 (6.11) W js = B1Z , U + A1Z = (U i + V j)s . Let x ∈ K be such that s(x) = 0. Letting X =(cid:18) Z(x) x (cid:19) ∈ Kn, we get A(X) =(cid:18) A1Z(x) + U (x) B(X) =(cid:18) B1Z(x) b(x) (cid:19) =(cid:18) W js(x) a(x) (cid:19) =(cid:18) (U i + V j)s(x) f ks(x) (cid:19) = 0 . eks(x) (cid:19) = 0 , Hence, since by assumption Ker A ∩ Ker B = 0, it follows that Ker s = 0. Namely, since K is linearly closed, s is a scalar, that we can shoose to be 1. In conclusion, we get, as we wanted, that k = p and q = pt, so that, by (6.8), (6.12) c = at , d = bt . Going back to the third equation in (6.6), we then get (6.13) U t = −A1B−1 1 W + V . Again, by the inductive assumption on the minimality of A1B−1 6.6, it follows that there exists Z ∈ K[∂]n−1 such that 1 and Lemma (6.14) W = B1Z , V = U t + A1Z . Hence, letting T =  T1 0   ∈ Mn(K[∂]) , Z t we get that C = AT and D = BT , completing the proof. Proposition 6.7. Part (a) of Theorem 6.4 holds, namely if A and B are two n × n matrix differential operators with B non-degenerate, then we can find n × n matrix differential operators C, D and E, such that A = CE, B = DE and Ker ¯C ∩ Ker ¯D = 0. 15 Proof. First, assume that the subfield of constants of K is algebraically closed. By Lemma 3.1, we may assume that is A upper triangular and B is diagonal. Consider a minimal fractional decomposition CD−1 of the fraction AB−1 in the linear closure of K. By Lemma 3.1(a), we can choose C and D to be upper triangular matrix differential operators. We may assume that all the diagonal entries of D are monic and, using elementary column transformations, that d(Dij) < d(Dii) for all i < j. Since the linear closure is the union of the iterate Picard-Vessiot compositum of K, all the coefficients of the entries of C and of D lie in some iterate Picard-Vessiot compositum of K. Take i minimal such that Ki satisfies this property. Assume i 6= 0. By Lemma 5.9, all the coefficients of the entries of C and D lie in some Picard- Vessiot subextension Ki−1 ⊂ P ⊂ Ki. Pick an automorphism φ of this exten- sion. By Theorem 6.4, the fractional decomposition φ(C)φ(D)−1 = CD−1 is still a minimal fractional decomposition because d(φ(D)) = d(D). So C (resp D) and φ(C) (resp φ(D)) are equal up to right multiplication by an invertible upper triangular matrix differential operator E. As all the diago- nal entries of φ(D) are monic and deg(φ(D)ij) < deg(φ(D)ii) for all i < j, E has to be the identity matrix. Hence C = φ(C) and D = φ(D) for all φ. It follows, by Proposition 5.14, that all the coefficients of the entries of C and D actually lie in Ki−1, which is a contradiction. So i = 0 and all the coefficients of C and D are differential operators over K. In the general case, one can find C, D and E satisfying the assumptions of the proposition, whose entries are differential operators a priori over K ⊗C ¯C. So all the coefficients of the entries of C and D lie in a Galois extension K ⊂ G. As the extension of the derivation to an algebraic extension is unique, all automorphisms commute with the derivation. Hence, using the same argument as above with the usual Galois theory, we obtain that the entries of C and D, hence those of E, are actually differential operators over K. Proof of Theorem 6.4. By Proposition 6.7, condition (i) implies condition (iii). Let AB−1 be a fractional decomposition, satisfying (iii). Then by Proposition 6.5 it satisfies (i) as a fraction of matrix differential operators over the linear closure of K, hence a fortiori over K. The implication (iii) ⇒ (ii) is clear by definition of a linearly closed field and (ii) ⇒ (iii) follows from Proposition 6.7. Hence part (b) of the theorem holds. If A0B−1 i s a minimal fractional decomposition of the fraction AB−1, then there is a matrix differential operator D over the linear closure of K such that A = −1B is actually a A0D and B = B0D. Since B is non-degenerate, D = B0 matrix differential operator over K. 0 16 Remark 6.8. We have the following two more equivalent definitions for a minimal fracitonal decomposition H = AB−1, with A, B ∈ Mn(K[∂]): (iv) the "Bezout identity" holds: CA+DB = I for some C, D ∈ Mn(K[∂]), (v) A and B have kernels intersecting trivially over any differential field extension of K. Condition (v) obviously implies condition (iii), and, also, condition (iv) implies condition (v) since the identity matrix has zero kernel over any field extension of K. To prove that (iii) implies (iv), we use the fact that any left ideal of Mn(K[∂]) is principal (cf. [MR01, Prop.4.10, p.82]). But if E ∈ Mn(K[∂]) is a generator of the left ideal generated by A and B, then by condition (iii) we have that Ker( ¯E) = 0, and therefore E must be invertible. 7 Maximal isotropicity of LA,B Let A, B ∈ Mn(K[∂]) with det B 6= 0. Recall that the subspace LA,B ⊂ Kn ⊕ Kn (defined in the Introduction) is isotropic if and only if AB−1 ∈ Mn(K(∂)) is skewadjoint, which in turn is equivalent to the following con- dition ([DSK12],Proposition 6.5): (7.15) A∗B + B∗A = 0 . Hence, LA,B ⊂ Kn ⊕ Kn is maximal isotropic if and only if (7.15) and the following condition hold: (vi) if G, H ∈ Kn are such that A∗H + B∗G = 0, then there exists F ∈ Kn such that G = AF and H = BF . Theorem 7.1. Suppose that A, B ∈ Mn(K[∂]) with det B 6= 0 satisfy equa- tion (7.15). If AB−1 is a minimal fractional decomposition, then LA,B ⊂ Kn ⊕ Kn is a maximal isotropic subspace. Namely, condition (iii) of Section 6 implies condition (vi). Proof. First, we prove the statement in the case when the differential field K is linearly closed. Due to equation (7.15), A maps Ker B to Ker B∗. Since, by assumption, Ker A ∩ Ker B = 0, this map is injective. Moreover, since Ker B and Ker B∗ have the same dimension (equal to d(B), by Lemma 3.1(b)), we conclude that we have a bijective map: (7.16) A : Ker B ∼ −→ Ker B∗ . 17 Let G, H ∈ Kn be such that A∗H + B∗G = 0. Since det B 6= 0, we have that B : Kn → Kn is surjective (by Lemma 3.1(b)). Hence we can choose F1 ∈ Kn such that G = BF1. Due to equation (7.15), we get B∗AF1 = −A∗BF1 = −A∗G = B∗H . Hence, H − AF1 ∈ Ker B∗, and by (7.16) there exists F2 ∈ Ker B such that AF2 = H − AF1. So, H = A(F1 + F2) and G = BF1 = B(F1 + F2), proving condition (vi). Next, we prove the claim for a differential field K with algebraically closed subfield of constants. Since, by assumption, Ker ¯A ∩ Ker ¯B = 0, we know by the previous result that there is a solution F ∈ Ln to the equations G = AF and H = BF , where L is the linear closure of K, and this solution is obviously unique (since two solutions differ by an element in Ker ¯A ∩ Ker ¯B). We will next use a standard differential Galois theory argument to conclude that this solution F must lie in Kn. By definition of the linear closure, all the entries of F lie in some iterate Picard-Vessiot compositum of K. Take i minimal such that Ki satisfies this property. Assume i 6= 0. By Lemma lem:5.9, all entries of F lie in some Picard-Vessiot subextension Ki−1 ⊂ P ⊂ Ki. As the solution F is unique in the linear closure, it is fixed by all the differential automorphisms of the extension Ki−1 ⊂ P, hence it lies in Kn i−1, which contradicts the minimality of i. In the general case, we know from the previous discussion that there is a unique solution F in (K ⊗C ¯C)n. Hence all the entries of F lie in a Galois extension G of K. We know that there is a unique way to extend a derivation to an algebraic extension, so all algebraic automorphisms of this Galois extension are also differential automorphisms. Hence F is fixed under the action of Gal(G/K) which means that it lies in Kn. Proposition 7.2. If a fraction AB−1 of matrix differential operators satis- fies A∗B +B∗A = 0, and AB−1 = (A0D)(B0D)−1 with Ker ¯A0∩Ker ¯B0 = 0, then LA,B is maximal isotropic if and only if D is surjective and KerD∗ ∩ (ImA0 ∗ + ImB0 ∗) = 0. ∗B0+B0 ∗A0 = 0, hence LA0,B0 is maximal isotropic, since A0B−1 ∗) = 0. We Proof. Assume that D is surjective and KerD∗ ∩(ImA0 have A0 is a minimal fractional decomposition. Let f, g ∈ Kn be such that A∗f +B∗g = 0. ∗g = 0. By Since KerD∗ ∩ (ImA0 maximal isotropicity of LA0,B0, we can find some h ∈ Kn such that f = B0h ∗) = 0, we get that A0 ∗ + ImB0 ∗ +ImB0 ∗f + B0 0 18 and g = A0h. Since D is surjective, there is k ∈ Kn such that h = Dk. So f = Ak and g = Bk, hence LA,B is maximal isotropic. ∗B0 + B0 ∗g + B0 ∗). Conversely, assume that LA,B is maximal isotropic. First, we prove that D is surjective. Take f ∈ Kn. Multiplying on the left by D∗ the equation ∗A0 = 0 and evaluating it at f , we get that A∗B0f + B∗A0f = 0, A0 hence by maximal isotropicity of LA,B, B0f = Bg and A0f = Ag for some g ∈ Kn. Therefore f − Dg ∈ KerA0 ∩ KerB0 = 0, hence f = Dg. So D is surjective. Next, take x ∈ KerD∗ ∩ (ImA0 In particular, ∗h for some g, h ∈ Kn and A∗g + B∗h = 0. By maximal x = A0 isotropicity of LA,B, we see that g = Bk and h = Ak for some k ∈ Kn. Multiplying the equation A∗ 0A0 = 0 by Dk on the right, we get that x = 0. ∗ + ImB0 0B0 + B∗ Remark 7.3. In the linearly closed case, a skewadjoint fraction AB−1 is a minimal fractional decomposition if and only if LA,B is maximal isotropic. Indeed, since KerD∗ ∩ (ImA∗ 0 ) = ±det(B0) 6= 0, we see that B∗ 0 is surjective, hence KerD∗ = 0. Therefore d(D∗) = 0 = d(D) and D is invertible. Here we used Corollary 3.2 and Remark 3.3. 0) = 0 and det(B∗ 0 + ImB∗ References [Art57] E. Artin, Geometric algebra, Interscience Publishers, Inc., New York-London, 1957. [BDSK09] A. Barakat, A. De Sole, and V.G. Kac, Poisson vertex algebras in the theory of Hamiltonian equations, Japan. J. Math. 4 (2009) 141-252. [CDSK12] S. Carpentier, A. De Sole, and V.G. Kac, Some algebraic proper- ties of matrix differential operators and their Deudonn`e determinant, J. Math. Phys. (2012) arXive:1201.1992 [DSK11] A. De Sole, and V.G. Kac, The variational Poisson cohomology, arXiv:1106.5882 [DSK12] A. De Sole, and V.G. Kac, Non-local Hamiltonian structures and applications to the theory of integrable systems, in preparation. [Die43] J. Dieudonn´e, Les d´eterminants sur un corps non commutatif, Bull. Soc. Math. France 71, (1943), 27 -- 45. [Dor93] I.Ya. Dorfman, Dirac structures and integrability of nonlinear evo- lution equations, Nonlinear Sci. Theory Appl., John Wiley & Sons, New York, 1993. 19 [MR01] J.C. McConnell and J.C. Robson, Non-commutative Noetherian rings, Graduate Studies in Mathematics, v.30, American Mathematical Society, Providence, RI, 2001. [Mag94] A. R. Magid, Lectures on differential Galois theory, University Lec- ture Ser., vol 7. AMS, 1994. [PS03] M. van der Put, and M. Singer, Galois theory of linear differen- tial equations, Grundlehren der Mathematischen Wissenschaften [Funda- mental Principles of Mathematical Sciences], 328. Springer-Verlag, Berlin, 2003. 20

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.